COMMODITY, FUTURES AND FINANCIAL MARKETS

Advanced Studies in Theoretical and Applied Econometrics

Volume 21

Managing Editors:
J.P. Ancot, Netherlands Economic Institute, Rotterdam, The Netherlands
A.J. Hughes Hallet, University of Strathclyde, Glasgow, United Kingdom

Editorial Board:
F.G. Adams, University of Pennsylvania, Philadelphia, U.S.A.,
P. Balestra, University of Geneva, Switzerland
M.G. Dagenais, University of Montreal, Canada
D. Kendrick, University of Texas, Austin, U.S.A.
J.H.P. Paelinck, Netherlands Economic Institute, Rotterdam, The Netherlands
R.S. Pindyck, Sloane School of Management, M.I. T., U.S.A.
H. Theil, University of Florida, Gainesville, U.S.A.
W. Welte, University of Lodz, Poland

The titles published in this series are listed at the end of this volume.
Commodity, Futures
and Financial Markets

edited by

Louis Phlips
European University Institute,
Florence, Italy

Springer-Science+Business Media, BV
Library of Congress Cataloging-in-Publication Data

Commodlty, futures, and f1nanC1al markets I edited by Lou1s Phlips.

p. CII'i. -- <Advanced studies in theoret1cal and appl1ed
econometr tes ; v. 21)
Includes bibliograph1cal references.
ISBN 978-94-010-5482-9 ISBN 978-94-011-3354-8 (eBook)
DOI 10.1007/978-94-011-3354-8
1. Financial futures. 2. Co~mDd1ty exchanges. 1. Phlips. Louls.
II. Serles.
HG6024. 3. C66 1990
332.64· 4--d c20 90-20549

ISBN 978-94-010-5482-9

Printed on acid-free papar

AII Rights Reserved

© 1991 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1991
Softcover reprint ofthe hardcover 1st edition 1991
No part of the material protected by thls copyright notice may ba reproduced or
utilized in any ferm or by any means, electronic or mechanical,
Includlng photocopying, recording or by any Information storage and
retrieval system, without written permission from the copyright owner.
In memory of Stefan
 TABLE OF CONTENTS

Contributors to this volume xiii

Introduction xv
Louis Phlips

PART I: EXPORT EARNINGS

1. Market Solutions to the Problem of Stabilizing Commodity Earnings

Andrew J. Hughes Hallett and Prathap Ramanujam
1. Introduction 1
2. Nonparametric Stabilization Rules 3
2.1. Operating on the Futures Markets: Optimal Hedging Strategies 4
2.2. Extensions: Private Stockholding and Speculation 6
2.3. Non-Market Interventions: Optimal Price Stabilization Rules 7
2.4. When is Price Stabilization Preferable to Hedging? 9
3. Conflicts between Market and Private Stabilization Schemes 9
3.1. Hedging 9
3.2. Price Stabilization 11
3.3. Strategic Errors 12
4. Intervention Costs: Stabilizing Net Revenues 14
4.1. Hedging Strategies with Costs 14
4.2. Buffer Stock Strategies with Costs 15
5. Empirical Analysis 17
5.1. Tests of Non-Normality 18
5.2. Hedging versus Price Stabilization: No Intervention Costs 20
5.3. Hedging versus Price Stabilization Allowing for 24
Intervention Costs
6. Private versus Aggregate Market Strategies 26
7. Conclusions 29
References 31
Appendix 32

vii
viii

2. Hedging Commodity Export Earnings with Futures and 35

Options Contracts
Joseph Ntamatungiro
1. Introduction 35
2. Hedging with Futures Markets 36
2.1. An Optimal Model of Hedging on Futures Markets 37
2.2. EstimatioN Results 39
2.3. Optimal Hedging and Uncertain Production 41
2.4. Estimating the Risk Gains 42
3. Hedging with Options Contracts 43
3.1. Introduction 43
3.2. Joint Hedge with Futures and Options Contracts 45
3.3. The Joint Hedging Performance 46
4. Conclusion 48
References 48
Appendices 50

3. Options to Alleviate the Costs of Uncertainty and Stability: 59

A Case Study of Zambia
Andrew Powell
1. Introduction 59
2. The Copper Price and Zambia 1964-1984 61
3. Instability and Uncertainty 67
4. Options to Stabilize or Insure Commodity Export Earnings 73
5. Conclusions 79
References 81
Appendix 83

PART IT: FINANCIAL MARKETS AND COMMODITY PRICES 85

4. The Response of Primary Commodity Prices to Exchange Rate Changes 87

1. Introduction 87
2. A Simple Static Model 90
3. Multi-Commodity Generalizations 93
4. Relation to Purchasing Power Parity 96
 ix

5. Stockholding and Intertemporal Price Adjustment 97

6. Futures Markets 103
7. The Exchange Rate Index 107
8. Choice of Weighting Schemes 108
9. Consequences of Exclusion of LDC Exchange Rates 112
10. Estimated Exchange Rate Elasticities 115
11. Conclusions 119
References 120
Appendix: Commodity Price Definitions 122

5. Exchange Rates and Storables Prices 125

Paul Kofman and lean-Marie Viaene
1. Introduction 125
1.1. Consumers of Primary Commodities 128
1.2. Producers of Primary Commodities 129
1.3. Traders in Primary Commodities 130
1.4. Speculators and Arbitragers 131
2. Specification of Agents' Optimizing Behaviour 132
2.1. Commodity Processor 132
2.2. Producer 134
2.3. Marketing Board 135
2.4. Speculator in Currency 137
3. Solving the Model in Equilibrium 138
3.1. Currency Spot Clearing 138
3.2. Commodity Spot Clearing 139
3.3. Currency Forward Clearing 141
3.4. Commodity Futures Clearing 143
4. The Link between Exchange Rates and Commodity Prices 143
4.1. The Correlation Coefficient 143
4.2. Exchange Rate Elasticity 146
4.3. An Expected Depreciation of the Currency 146
4.4. Decreasing Exchange Rate Volatility 147
5. Summary 149
References 151
x

6. An Evaluation of the Performance of Speculative Markets 153

Jerome L. Stein
1. Introduction 153
2. The Inadequacy of Conventional Tests of Market Performance 154
3. Welfare Measures 157
4. A Two-Period Model 162
4.1. The Structure of the Market Model 163
4.2. Bayesian Error 166
5. Application of the Theory 168
5.1. Research Design 168
5.2. Empirical Results 170
6. Conclusion 177
References 178

7. Dynamic Welfare Analysis and Commodity Futures 179

Markets Overshooting
Gordon C. Rausser and Nicholas Walraven
1. Introduction 179
2. Basic Specification 181
2.1. Basic Model Specification 181
2.2. Fixed Output and Overshooting 183
2.3. Endogenous Output and Empirical Results 184
3. A Dynamic Multicommodity Welfare Measure 187
4. Methodology, Data, and Empirical Results 191
5. Conclusion 203
References 203

P ART III: MONOPOLISTIC COMMODITY MARKETS 205

8. Futures Trading for Imperfect Cash Markets: A Survey 207

Ronald W. Anderson
1. Introduction 207
2. Motives for Trading Futures 210
3. Speculation and Hedging by Powerful Agents 212
4. Hedging by the Competitive Fringe 217
 xi

5. Strategic Futures: Cournot Oligopolists 221

6. Strategic Futures: Cartel Futures Policies 225
7. Strategic Futures: Storable Goods 229
8. Strategic Futures: Durable Goods 235
9. Exhaustible Resources 240
10. Futures Price Bias and Volatility 243
11. Conclusion 245
References 247

9. Duopoly, Inventories and Futures Markets 249

Blaise Allaz
1. Introduction 249
2. The Model with Futures Only 253
2.1. The Cash Market 254
2.2 The Futures Market 256
3. The Model with Inventories Only 259
3.1. Producers' Decisions at Time 2 260
3.2. Producers' Decisions at Time 1 261
4. The Model with Both Futures and Inventories 262
4.1. The Cash Market 263
4.2. Time 1 Decisions 263
5. Conclusion 268
References 268
Appendix 269

10. Monopsony Power and the Period of Commitment in 273

Nonrenewable Resource Markets
1. ID.troduction 273
2. Relations with Other Problems 278
3. The Markov Time-Consistent Equilibirum 283
4. A "Locally" Time-Consistent Equilibrium 291
5. A Comparison with Reproducible Goods 293
6. Conclusion 298
References 299
 Contributors to this volume
Blaise Allaz, Departement Finance & Economie, Centre HEC-ISA,
1, rue de la Liberation, 78350 Jouy-en-Josas, France.

Ronald W. Anderson, Department of Economics, City University of New York,

Graduate Center, 33 West 42nd Street, New York N.Y. 10036, USA.

Christopher L. Gilbert, Department of Economics, Queen Mary and Westfield

College, Mile End Road, London El 4NS, England.

Andrew J. Hughes Hallett, Department of Economics, University of Strathclyde,

100 Cathedral Street, Glasgow G4 OLN, Scotland.

Larry Karp, University of California, Berkeley and Department of Economics,

University of Southampton, Highfield, Southampton S09 5NH, England.

Paul Kofman, Erasmus Universiteit, Postbus 1738, 3000 DR Rotterdam,

The Netherlands.

Joseph Ntamatungiro, Banque Nationale du Rwanda, B.P. 531, Kigali, Rwanda.

Andrew Powell, Nuffield College, Oxford OXl lNF, England.

Prathap Ramanujam, Ministry of Policy Planning and Implementation, Sri Lanka.

Gordon C. Rausser, Department of Agricultural and Resource Economics, University

of California, 207 Giannini Hall, Berkeley, California 94720, USA.

Jerome L. Stein, Department of Economics, Brown University, Providence R.I.

02912, USA.

Jean-Marie Viaene, Erasmus Universiteit, Postbus 1738, 3000 DR Rotterdam,

The Netherlands.

Nicholas Walraven, Department of Agricultural and Resource Economics,

University of California, 207 Giannini Hall, Berkeley, California 94720, USA.

xiii
 INTRODUCTION

Louis Phlips

The stabilisation of primary commodity prices, and the related issue of the
stabilisation of export earnings of developing countries, have traditionally been
studied without reference to the futures markets (that exist or could exist) for
these commodities. These futures markets have in turn been s~udied in isolation.
The same is true for the new developments on financial markets.
Over the last few years, in particular sine the 1985 tin crisis and the
October 1987 stock exchange crisis, it has become evident that there are inter-
actions between commodity, futures, and financial markets and that these inter-
actions are very important. The more so as trade on futures and financial
markets has shown a spectacular increase. This volume brings together a number
of recent and unpublished papers on these interactions by leading specialists (and
their students).
A first set of papers examines how the use of futures markets could help
stabilising export earnings of developing countries and how this compares to the
rather unsuccessful UNCTAD type interventions via buffer stocks, pegged prices
and cartels.
A second set of papers faces the fact, largely ignored in the literature, that
commodity prices are determined in foreign currencies, with the result that
developing countries suffer from the volatility of exchange rates of these
currencies (even in cases where commodity prices are relatively stable). Financial
markets are thus explicitly linked to futures and commodity markets.
A third set of papers acknowledges the obvious fact that many commodity
markets are oligopolistic (if not cartelised). Market power in a commodity
market creates the possibility of manipulating the futures market for this
commodity. In the case of a natural resource such as crude oil, the presence of
OPEC must affect the working of the crude oil market.
I am particularly grateful to the contributors to this volume for accepting
to cross-referee each other and for their efforts to meet the deadlines set.
The first page of each of the three parts of this book gives a brief summary
of the papers enclosed.
xv
 PART I

EXPORT EARNINGS

A number of recent papers have considered optimal hedging strategies in

commodity markets, comparing them to buffer stock interventions designed to
reduce the risks associated with price and output uncertainty and/or to stabilise
the earnings of producers. However those studies have been conducted under
some rather severe restrictions: that price and production disturbances are jointly
normally distributed, that there is no private stockholding, that markets are
unbiased, and that stabilising activities are not constrained by insufficient
financing. These restrictions are clearly unwarranted in practice. Chapter 1
considers the implications of relaxing each one when hedging operations or buffer
stock are compared for stabilising producers' earnings.
Moreover, previous work (Gemmel, 1985, excepted) have failed to point out
that hedging by one producer does not affect the price distribution of other
producers, whereas buffer stocks operated by one producer will affect the price
distribution of all other producers in the same market. Hence there will be no
conflicts between the hedging operations of different producers in any market.
But different producers will want different buffer stocks intervention rules (unless
all supply distributions are all identical) - hence what is optimal for one
producer will be suboptimal for the others, and what is optimal for the market is
suboptimal for the individual producers. Chapter 1 examines these conflicts for
the major producers of five internationally traded commodities. Andrew Hughes
Hallett and Pratham Ramanujam find that producers may occasionally disagree
on which strategy should be implemented. Most of the disagreements would be in
what strength of intervention is needed. They also estimate the losses suffered by
each producer (compared to his privately optimal outcome) when either another
producer's rule is imposed or an optimal rule is adopted. Here they find that
averaged intervention rules lead to small losses, and that those losses are never
large enough to upset the ranking of buffer stock stabilisation versus hedging.
xvii
xviii

But optimal market stabilisation would produce longer losses and upset that
ranking. And dominance by one producer would imply even greater losses.
Thirdly, allowing for operating costs (i.e. attempting to stabilise net rather
than gross revenues) does not make quite a difference. Hedging, being cheaper to
undertake at a possible cost of reducing average revenues, becomes superior in
several cases where buffer stock interventions would have been better for
stabilising gross revenues - principally in the case of high unit value commodities
such as tin and copper. The rule of thumb is that hedging is more effective for
stabilising earnings from high unit value commodities (e.g. metals), and buffer
stocks are better for the low unit value ones (e.g. agricultural goods).
In Chapter 2, Joseph Ntamatungiro proposes futures and options contracts
as potential hedging instruments. Their separate hedging effectiveness is
evaluated in a portfolio model integrating basis risk and production variability. A
methodological "pre-harvest" hedging strategy for a sample of some representa-
tive exporters of cocoa, coffee, cotton and rubber gives an insight in the potential
gain LDCs may derive from using futures markets. Production uncertainty does
not explain why LDCs do not intensively operate on futures markets for it does
not necessarily reduce the optimal position on futures markets. The joint hedging
performance on futures and options markets is also evaluated.
Powell focusses on two debates concerning the effect of volatile primary
commodity prices on developing countries. First of all the distinction between the
costs of instability versus the costs of uncertainty in primary goods is discussed. A
theoretical model is developed that is capable of separating these costs. Using
historical forecast data on copper, a measure of uncertainty is obtained and using
the theoretical framework a study on Zambia shows that the standard expected
utility framework (which does not distinguish between instability and
uncertainty) may seriously underestimate the gains to a stabilisation programme
for a country such as Zambia. Secondly, Chapter 3 addresses the debate on the
alternatives to international commodity prices stabilisation schemes and in
particular the use of options contracts for Zambia to stabilise copper income. The
choice between futures and options is discussed and a simulation program is
conducted illustrating the returns that might have been available to Zambia by
using certain option stabilisation strategies.
 CHAPTER 1

MARKET SOLUTIONS TO THE PROBLEM OF

STABILIZING COMMODITY EARNINGS

Andrew Hughes Hallett and Prathap Ramanujam

1. INTRODUCTION
Both hedging and stockpiling provide producers with an opportunity to reduce
the risks associated with price and production uncertainties. Most of the
literature on this topic has been concerned with the welfare justification for, and
the consequences of, market stabilization. But there is the equally important
problem of how to stabilize earnings most effectively in a market where prices are
volatile and production uncertain, once the welfare case for doing so has been
established. That makes the welfare analysis operational.
Optimal hedging rules for a range of different objective functions, including
minimizing the variance of earnings, have been derived by McKinnon (1967),
Anderson and Danthine (1983), Newbery (1988) and Karp (1988). Stockpiling
rules have been analyzed by Nguyen (1980), Newbery and Stiglitz (1981), and
Ghosh et al. (1987). All these studies incorporate some crucial but implausible
assumptions; typically that prices and production are jointly normally
distributed, that there is no private stockholding, that markets are unbiased and
risks diversifiable, and that resources for stabilizing activities are not constrained.
Many authors! have pointed out that stockholding does in fact occur, and that
that, coupled with financial constraints, would imply the kind of asymmetries
and kurtosis in the price and quantity distributions which we observe in practice.
Similarly commodity producers do face significant risks but they operate
predominantly in economies where there are few opportunities to diversify those

! Wright and Williams (1982), Hughes Hallett (1986).

L. Phlips (ed.), Commodity, Futures and Financial Markets. 1-34.

© 1991 Kluwer Academic Publishers.
2

risks onto ot.her assets.

To deal with these criticisms we have recently developed hedging and
stockpiling rules which minimize the variance of producers' earnings in a
distribution free and model free set up (Hughes Hallett and Ramanujam, 1990).
In principle those rules could be applied in any market, whatever the price
distribution and without relying on econometric estimates of the associated
supply and demand elasticities. They are, however, still subject to two important
restrictions. First the stockpiling rules have been developed and tested only for
producers taken as a group, whereas under hedging each producer can operate his
own rule. The problem is that buffer stock interventions ,designed to minimize the
variance of earnings for the market as a whole will depend on a single inter-
vention parameter so, unless every producer faces exactly the same supply
distribution, what is optimal for the market as a whole will not be optimal for
individual producers (a point made by Gemmill, 1985). Our analysis therefore
needs to be extended to the case where individual producers who face different
supply conditions. That is the first purpose of this paper.
The second restriction is that these hedging and stockpiling rules do not
account for their operating costs. They are therefore designed to minimize the
variability of gross revenues. That may be reasonable if the stabilization
activities are being funded by international agencies, such as UNCTAD for
example, on behalf of the poorer and indebted LDC commodity producers. But it
is clearly unreasonable at the level of individual producers. The second purpose of
this paper is therefore to examine and compare hedging and stockpiling strategies
for stabilizing revenues net of storage and transaction costs.
Three restrictions remain in our analysis, however. We continue to assume
market invariance; i.e. that the act of stabilizing revenues does not change the
behaviour of market agents and traders. However, this assumption only matters
in as much as those changes in behaviour might affect the first four moments of
the bivariate price-quantity distributions. It does not matter if the parameters of
the underlying demand, supply or stockholding rules are changed (as happens in
Turnovsky's (1983) model where elasticities depend on the variance of prices) so
long as the net impact on those moments is negligible. Secondly, we continue to
take the mean pre-stabilization price as the price level which stockpiling rules
attempt to maintain. Formally that requires linear supply and demand functions
but, as Gilbert (1986) points out, a relaxation of that specification would only
lead to a redistribution of revenues between producers and consumers rather than
 3

to alterations in the "stabilizability" of the market. Since distributional questions

are not our concern here, this restriction is not important to our results. Thirdly,
all our conclusions are dependent on earnings stabilization as an objective; they
might not survive in a fuller welfare analysis. However earnings are quadratic in
prices and quantities, and Levy and Markowitz (1979) have demonstrated that
the mean and variance of a quadratic objective is a very robust approximation to
the general expected utility criterion of risk averse decision makers. So that
restriction is also unlikely to be important. That said, we make no attempt to
establish the welfare case for stabilization in this paper. It is simply assumed that
the necessary welfare evaluations have been made and that producers now wish
to determine the best way to achieve the earnings stability that they need.
Finally, earnings stability is defined in terms of the variance of random
fluctuations about mean or trend values. Hence our hedging or stabilization rules
are designed to "buy" predictability, rather than to induce stability in the sense
of modifying the dynamics of the earnings stream. Similarly the question of
whether the underlying prices are statistically stationary, stationary with discrete
jumps, or trending, plays no role here since we are concerned with random
fluctuations about some known trend or cycle.

2. NONPARAMETRIC STABILIZATION RULES

Markets may be stabilized using financial instruments or with buffer stock
interventions. The welfare objection to using buffer stock interventions is that
agents can achieve the same results more easily by taking appropriate positions
in competitive futures markets. The cost of maintaining the buffer stock itself
would also be saved. The difficulty with that argument is that a complete set of
contingency market.s does not exist for most commodity and financial markets
and, in some cases, no such markets exist. In addition, many producers or
producing countries are either too small or too indebted to secure the credit
needed to adopt the necessary positions.
For these reasons it is important to determine which strategy is most
effective for reducing earnings risk. If it is hedging, then policy needs to be
directed at providing access to contingency markets. If that is not possible,
compensatory finance, income support or buffer stock schemes aimed at
replicating the hedging outcomes might be introduced - although that will be
successful only if output fluctuations are fairly small (Hughes Hallett and
Ramanujam, 1990). If however price stabilization is the better strategy we must
4

evaluate (and cost) the buffer stocks needed. Producers therefore need to develop
and compare optimal hedging and buffer stock strategies in order to determine
their best revenue stabilization strategies. 2

2.1 Operating on the Futures Markets: Optimal Hedging Strategies

Consider the j-th producer, j=l...n, who can produce a quantity qj of some
commodity which he can sell at a spot price p. However it takes some time to
complete production and the final supply is subject to various shocks. In period 0
he plans to produce q., but the actual output q. is a random variable with mean
J J,
qj and variance o} Similarly the, as yet unknown, spot price p is a random
variable with mean p and variance (J2. Nothing else is assumed about the joint
p
probability distribution of (p,q/
As an alternative to receiving an uncertain revenue of Yj=pqj' the producer
can hedge on the futures market by selling (in period 0) a fixed quantity hj for
delivery in period 1 at a known price Pf If he does that he will be obliged to sell
his unhedged output qrhj (including the supply shock qr-q) at the spot price
(including the price surprise p-p). His hedged revenue is therefore

y~ = p(qrh) + Pfh j = pqj + (prp)h j (1)

If, in addition, the market is efficient so that forward prices are an unbiased
predictor of future spot prices (i.e. E(Pf)=P), then

E(Y~) = E(pq) = E(yj) (2)

where y~ denotes producer j's unhedged income. Hence hedging on an unbiased
J
market leaves average revenues unchanged. Direct calculation now yields
( h) -Epqj
VYj _ (2 2) - [( j
Epqj )] 2 -2h j (_PJ.lll+ 2 j) +hj(Jp
q_j(Jp+J.l21 2 2 (3)

2 Some comparisons have been attempted before but only for aggregated
producers where interventions are costless, where prices and quantities are jointly
normally distributed (so there is no private stockholding), and where risks are
diversifiable and prices unbiased. On this basis McKinnon (1967), Newbery and
Stiglitz (1981), Gemmill (1985) and Gilbert (1985) prefer a market approach
although the empirical results are mixed and show no clear superiority for
hedging. Later work (Gilbert, 1988) is more favourable to buffer stocks inter-
ventions. It is not at all clear that any of these results would extend to our more
general framework, or to the level of individual producers.
 5

where Itlk=E[(p-p)i(qr-qj)k], for i,k~1, are the bivariate moments from the joint
density functions of (p,q.) for each j. We can write V(y<?) for E(p2q~)-[E(pq.)12,
J J J J
and for convenience we shall also write p.(J (J. for ItJl·1 where p. is the correlation
J p J J
coefficient between p and qj" It is usually assumed that P{o on the argument
that a production shock which affects one producer will affect them all. A
negative shock would then lead to a fall in aggregate supply and higher prices
(and vice versa if the supply shock is positive). But there is no guarantee of this
since a producer who is differentiated either geographically or in production
structure may not suffer the same shocks as others and, depending on the demand
elasticities, may even be able to profit from a larger market share during high
price periods.
From (3) the optimal hedging rule is

hj = Pl(J/ (Jp + qj + It~l/ (J~ (4)

which is the result obtained by McKinnon (1967). If P{O, the greater is the
output uncertainty relative to price uncertainty, the less should be hedged.
Similarly the stronger the negative correlation, the smaller is the quantity that
should be hedged, but the greater the (positive) asymmetry the larger the hedge.
Notice also that if output variability is large enough, hj may be negative and the

producer should buy forward. Conversely if P/O and It~l is small, hj>qj. So
short or long positions can result from (4). Finally, if p and qj are symmetrically

and normally distributed, then It~l =1t~2=O. If they are independent then Pj=O as
well. Those conditions would imply

(5)

Under any hedging rule, the variance of revenue can be written as

V(yh) = V(y~) + [h~ - 2h hh>l:Ju2
J J J J P

using (3) and (4) again. Hence under the optimal hedging rule we get
Vy. (0) -h *2.(J2
( h) =Vy. (6)
J J J P

That shows an unambiguous gain, in terms of risk reduction, over doing nothing.
This risk reduction increases with increasing uncertainty in prices and quantities
6

(i.e. up and uj ), with greater skewness(IL~1>0), but lower association between

prices and quantity (assuming P(O). Thus, for a given average market size <lj'
"smaller" or more differentiated producers may benefit more from trading on the
futures markets than would large producers who can influence spot prices.

2.2 Extensions: Private Stockholding and Speculation

Almost all commodity markets display positively skewed price distributions, and
this is likely to be the result of stockholding activities. If the market exhibits
excess demand, that can only be met by running down ~tocks. Supply is inelastic
in the short term and once stocks are exhausted prices will rise very rapidly. If,
on the other hand, there is excess supply, agents can accumulate stocks so long as
they have sufficient finance. They may not wish to accumulate stocks
indefinitely, but they can certainly accumulate more easily than decumulate and
they can often hold stocks "in the ground" or fail to harvest. Hence prices fall by
considerably less than they rise in the excess demand case (Hughes Hallett,
1986).
The same arguments imply that stockholding will generate large bivariate
third moments. For example, large negative supply shocks are likely to push
market prices high above their mean value because stocks and short run supply
responses are inadequate to cope with the excess demand. But a positive supply
shock would push market prices below their mean by a smaller amount since
stocks can absorb the excess supply. Hence stockholding will imply

1L12 = E[(p-p)(q-qi] > ° and 1L21 = E[(P_p)2(q_<O] < °

What if the market is biased or inefficient: E(Pf)=p+k where kfO? Gilbert
(1988) argues that a large number of speculators in a market where price risks are
diversifiable will ensure that the futures price is an unbiased predictor of the
corresponding spot price. On the other hand, in a world where spot and futures
price are jOintly determined and where there is little or no speculative activity
and risks are not easily diversified, the futures price will generally be biased
(Turnovsky, 1983; Kawai, 1983). However retracing steps (3) to (6) for the biased
market case yields exactly the same hedging rule and variance reduction as in the
unbiased case; equations (4) and (6) can be applied just as before (Hughes Hallett
and Ramanujam, 1990). Average earnings, on the other hand, are affected since
h
E(y j) = E(pq) + kh j (7)
 7

follows from (1). This deals with the case where there are few speculators in the
market and price risks are not easily diversifiable. Hedging will produce a
systematic gain (loss) in expected earnings if k>O (k<O). But the same hedge and
gains in stability apply whether or not markets are inefficient or biased.

2.3 Non-Market Interventions: Optimal Price Stabilization Rules

Direct price stabilization requires producers to set up a buffer stock which will
buy some or all of the market's excess supply at some target price level (to
prevent prices falling further) and sell when there is excess demand at that price
(to prevent them rising further). Naturally the amounts to be bought or sold in
that operation will depend on the elasticities of demand and supply. But, given a
homogenous commodity, all producers face the same demand schedule so that the
interventions which each producer needs to stabilize his revenue will vary
according to his own supply conditions. Everyone will therefore wish to run his
own buffer stock operations and, unless everyone faces the same supply function,
private buffer stocks will not stabilize market revenue and interventions which
stabilize the market as a whole will be suboptimal for individual producers.
Let producer j's net buffer stock purchase be
BSj == AP-D) (8)
where S == Eq~ is aggregate supply, D is total demand, and A. is the proportion of
J J
excess demand/supply which producer j sells or buys at the target price level in
order to stabilize his revenues. Producer j will pick Aj to minimize the variance of
his own earnings.3 Suppose that he faces demand and supply schedules given by
D == a - bp + u a, b > 0 (9)

qS == a. + f3.p + v· a J., f3J. > 0

J J J J
Suppose also that the target price level is the expected equilibrium price, p. This
assumption allows us to abstract from any transfers between market participants
which would follow if p (or linear supply and demand functions) were not chosen
and thus ensures comparability with the hedging strategy.
Post-stabilization prices, p, satisfy BS+D==S where BS==EBS j is evaluated
at p. In other words

3 Aj==O implies no stabilization; Aj ==l implies perfect stabilization (although

the individual producer would probably find that hard to achieve). We are at
present ignoring the costs and financial commitment involved in maintaining the
buffer stock.
8

p = p + (l->}(u-v)/(,8+b) (10)
where P = (a=a)/(/J+b), v = Ev j , ,8 = E,8j and a = Eaj . Hence producer j's
stabilized revenue will be
(11)
Thus
E(y~)-E(y<?)=-A,p.ll ll· (12)
J J JJPJ
so that price stabilization will raise expected earnings above the pre-stabilization
or hedging level if p.<o (and it will reduce average earnings if p.>O). This is
J , J
because producers can sell all their expected output at stabilized prices, whereas
they can hedge only their expected output at forward prices - the remaining
supply shock will just fetch the spot price. Hence, in terms of average earnings,
price stabilization is the better strategy the more negatively correlated are prices
with supply shocks. That will depend on demand elasticities and the market
(production) structure.
Notice that there will generally be a conflict of objectives under price
stabilization since V(l) will be minimized at a finite value A40, while (12)
J J
shows E(yj) increases with Aj if P{O (and vice versa if PlO). In fact (11) and
(12) imply

V(yj) = (l-A/V(yj) + A2p211~ + 2AP-A j )P Cov(yj,qj) (13)

and hence that the optimal buffer stock intervention is defined by (8) with 4
V(yo) - p Cov(y<?,q.) A.
A* - J J J -.-1 (14)
j - V(yj) + p211~ - 2p Cov(yj,qj) B j

h
were j - p.ll
B . -- J122 j + q.ll
2 2 ll·2 + 2-q·J121 _2 2 an d A . -- B . + PJ112
- j +--
pq.p.ll ll·.
J JPJ J JP J J JJPJ

4 The second equality in (14) is obtained by rewriting (13) as

V(yj)= V(yj)+ A~[(V(yj)+ p211~_ 2PCOv(yj,qj)]- 2A}V(yj)- pCov(yj,qj)]
and then substituting in the direct evaJuations of V(yj) and Cov(yj,q} i.e.
V( y.0) -J122-P·llll.
_ j 2 2 2 + 2-PJ112j + 2-qJ121
j + _2
pll.2 + q.ll
_2 2 + 2--
pq.p.llil.
J JPJ J JP JJPJ
an d Cov (0
yj,qj ) -- J112
j + P_2 llj2 - 2-- °
pqlfpllj' t Yle
' ld

V(l) = V(y<?) + A~B. - 2A.A. and hence (14) as stated.

J J JJ JJ
 9

Finally, reorganizing (13) as indicated in footnote 4, and inserting ).j from (14)
gives
v(l)
J
= V(l)
J
- A~/B.
J J
(15)

which shows an unambiguous gain in earnings stability since we can rewrite Bj as

E[yO_E(Yo)_p( q.-q .)]2>0. However the conventional result that partial stabiliz-
J J J J
ation is optimal, O<).j<l, follows only if O<A{Bj' That will happen if "~2
and/or p. are weakly negative so that ,.J1·2+il .p.O' 0'.<0. A simple sufficient
J JJPJ '
condition for partial stabilization to be appropriate is weakly negative second and
third joint moments. But if those moments are positive ). j> 1, and if they are
strongly negative ),>:<0. Thus, as in hedging, positive p and q. correlations will
J J
require "over-stabilization", while very strong negative correlations call for
pro-cyclical "stabilization". Only in the special case where p and qj are
symmetrically distributed and P{O, is the conventional result O<).j<1 generally
valid. If, in addition, p and qj are independent then perfect stabilization ().j=l)
is best.

2.4 When is Price Stabilization Preferable to Hedging?

When do buffer stock stabilization programmes produce greater earnings stability
than hedging on the futures market? That turns on whether A]/B/hj 2 or 0';
not. This may be tested numerically. But these expressions take exactly the same
algebraic form, in terms of the (p,qj) distribution faced by producer j, as they did
for stabilizing aggregate earnings given the market's (p,q) distribution; compare
Hughes Hallett and Ramanujam (1990). Therefore we can use the same decision
tree test, applied to each (p,qj) distribution in turn. This test is given in Figure
1.

3. CONFLICTS BETWEEN MARKET AND PRIVATE STABILIZATION

SCHEMES
3.1 Hedging
Hedging changes the price distribution which an individual producer faces, but
not the price distribution faced by other market participants, since matching
demand and supply on the futures markets will leave the spot market's excess
10

Compute Ah=q1/ii using (1); AS=A* using (2); plus d=~V(yh)_V(ys)l/ aA=
J.t22-p2u;u!+iiJ.t2l+PJ.t12 and A+ =(d+~1)/(d-pJ.t12). Then test as follows:

Figure 1. Decision tree

 11

demand or supply position unchanged. 5 As a result, producer j will find no conflict

between the rule which minimizes rus own earnings variance and the rules wruch
other producers may be using to minimize their earnings variance; the rules are
independent in the sense that changes in producer k's rule will not affect
producer j's optimal rule. Moreover, because each producer can determine rus
own received price distribution without influencing the price distribution of
others, each rule is controlled by its own intervention parameter (hj or
A~=h>l; /q .). Nevertheless it is easy to check that the earnings of different
J J J
producers remain correlated. Hence producer j's hedging will affect producer k's
earnings distribution, and V(Yk) in particular. Consequently privately optimal
hedging rules will not minimize the variance of market revenues as a whole unless
all the supply distributions are the same. Similarly, a rule wruch is optimal for
the market as a whole will not minimize earnings variances for individual
producers. Inspection of (4) shows Eh~l'h*, where h* is the optimal hedge for the
J
market as a whole and is defined by (4) with joint moments J111 and 11:21 taken

from the aggregate (p,q) distribution 6 unless J111=El~1 and J121=El~I' A simple
sufficient condition for that is J1l1 =J111/n and J1~1 =J121/n for all j.

3.2 Price Stabilization

Any buffer stock intervention will change the price distribution for every
producer - including those who choose not to participate in the stabilization
programme. As a result the only thing that matters in a market of homogenous
goods is the impact of the aggregate buffer stock; producers cannot determine
their own price distributions even if they are operating private buffer stocks, and
the degree of stabilization is actually a function of the aggregated interventions
and a single intervention parameter.
The boot is on the other foot here. Under hedging every producer was free
to minimize his own earnings variance and will not care that market revenues are
not stabilized at the same time since there are no Pareto improvements to be had
by doing so (either in terms of trading between variances, or in terms of trading

5 This follows from the market invariance assumption. The spot market's
turnover will have fallen, however.
6 See Hughes Hallett and Ramanujam (1990) for this result.
12

smaller variances for higher average earnings). That remains the case even if
official policy is directed at opening up access to the futures markets. But under
price stabilization, producers will care that interventions which minimize the
earnings variance of one producer do not so so for others or for the market as a
whole. They will also be concerned that the interventions which stabilize the
market revenues do not do so for individual producers unless all the joint (p,qj)
distributions are identical; (14) shows AjfAk and AjfA*, where kfj and A*
minimizes V(y) for 7 y=};yj" In this case the conflicts between private and
aggregate stabilization rules do matter and it is not clear how policy should
attempt to resolve those conflicts. For this reason we focus on analyzing the
differences between buffer stock schemes and their policy implications in what
follows.

3.3 Strategic Errors

The possibility of conflicts between aggregate and private stabilization schemes
raises three important questions:
a) When does Aj=Ak or A*? Those are the circumstances in which producer j's
private actions will have the effect of stabilizing producer k's earnings .at the
same time, and also the circumstances in which individual producers will find
that a buffer stock used to stabilize market revenues will maximize their own
earnings stability.

Sufficient conditions 8 for Aj=Ak are Ji~FJifi for i,i=0,1,2; or Ji~2=Ji~2=0 and
prPk =0. Similarly sufficient conditions for Aj=A * are either JiiFJijf for
i,i=0,1,2; or Ji~2=Ji12=0 and p.=p=O. The former condition is nearly the case
. J
where producers have identical supply distributions (but not quite since it
refers only to joint moments up to fourth order). The latter condition

7 A* is given by (14) with joint/marginal moments from the aggregate (p,q)

distribution; Hughes Hallett and Ramanujam (1990). The non-linearity of (14)
means that no simple conditions on the moments of (p,qj) and (p,q) will ensure
the aggregate and private intervention rules all coincide.

8 The necessary and sufficient conditions, Jii2 + (j/{pO"j = Ji~2 + <lkPkO"pO"k'

are too particular to be of any interest here. The corresponding conditions for
. j k j k
hedgmg are Jill = Jill and Ji21 = Ji21·
 13

requires symmetric distributions with un correlated prices and quantities.

Both are rather more stringent than requiring all producers to have the same
price elasticity of supply, as suggested by Gemmill (1985). Beyond these two
special cases, interventions which are optimal for one producer will be
suboptimal for another. Thus, unless those special cases hold, stabilizing
market revenues will not minimize the earnings variability of market
participants.
b) When would producer j choose the same strategy as an agency charged with
stabilizing the market as a whole? Alternatively, how much would be lost
because such an agency would choose either the wrong strategy from the
individual producer's point of view, or suboptimal interventions within a
given strategy? Given that public/private conflicts do not arise under
hedging, this amounts to testing when the authority would select price
stabilization when producer j would do better with a hedging strategy that
he can control himself. Using the test in Figure 1, this happens if
)..sA > h*2(T2 in aggregate (16)

but for producer j.

Now if O=A 2/(Bh*2) but OJ=A 2/( Bjhj 2), (16) can only hold when O{ (T~ d.
The corresponding losses can be evaluated by computing V{yj) at ).. * or )..1:
and comparing that to the optimized values of V{y~) or V{yj).
c) Under what circumstances could a market stabilization programme actually
end up destabilizing producer j's income, or when would producer k's
privately optimal interventions have the effect of destabilizing producer j's
earnings? Taking the former case, that could only happen if V{i» V{yo)
J J
which follows from the definition of V{i) in footnote 4 if
J
A/B{AB/B - 2A j} > O.
Such an inequality can hold only when )"*>2)..j if A>O, or )"*<2)..j if A<O.
Similarly producer k's actions would destabilize producer j's earnings if
)..1:>2)..j when Ak>O, or if )..k<2)..j when Ak <0. These inequalities 9 are in

9 The corresponding condition for a~gregate hedging to destabilize producer

j's earnings is h*>2hj if h~O (but h*<2h j if h<O).
14

fact quite likely to hold when supply conditions vary significantly across
different producers. They are more likely to hold for certain highly
differentiated producers than between any producer and a market authority
whose .\* represents some kind of average of the .\j values.

4. INTERVENTION COSTS: STABILIZING NET REVENUES

Stabilization operations, whether undertaken on the futures markets or through
buffer stock interventions, have to be financed. Producers will need to take these
operating costs - principally storage, transactions and financing costs - into
account when deciding on their strategy and level of s'tabilizing activity. With
the exception of the case where an external agency finances these operations as
part of its aid or debt relief programmes, producers should aim to stabilize their
earnings net of operating costs.
To a first approximation, operating costs will vary with either the physical
volume or the value of the hedge on the buffer stock. The former represents any
warehousing or storage costs. The latter represents the transaction costs
consisting of the financing (or opportunity) costs of maintaining. a buffer stock or
of making the required margin payments on the forward contracts. These costs
will be roughly proportional to the value of the buffer stock or forward contract.
Note that the value of a forward contract is known in advance, but the actual
buffer stock costs will depend on the realized (post-stabilization) market price
which cannot be known beforehand. However the intended buffer stock is that
which returns prices to their target level, p. So producers will have to base all
their decisions on a buffer stock evaluated at the same (planned) price.

4.1 Hedging Strategies with Costs

If there are storage costs, producer j's revenue will be reduced by rhy where r is
the cost per unit hedged. We replace (1) with an expression for net revenue:

y~ = pqj + (PIP-r )hj (17)

which, in contrast to (2), implies E(y~)<E(yj) by an amount which depends on r
and hj . On the other hand, (17) shows that all our previous results carryover if
the forward prices, Pf> are interpreted as net of costs. In particular, since hj is
independent of Pf, (4) and (6) continue to apply without modification.
It is hard to think of plausible reasons why hedging costs should depend on
the volume rather than the value of the futures contract. The transactions costs
 15

(margin payments) case implies net revenues are given by

y~ = pqj + (ptp)h j - rpfh j (18)

where r can be interpreted as the real market rate of interest. Once again (18)

implies E(y~)<E(y?). At the same time, direct calculation shows that V(y~) is
J J J
unchanged:

V(y~) = E[pqj + (ptp)hj - rPfht - [E(pqj) - rPfht

(19)

which is identical to (3). So, in this case as well, transactions costs do not affect
the optimal hedging rule or the reduction in earnings variances which can be
achieved; (4) and (6) continue to apply. In fact, the only consequence of including
operating costs in the analysis is to introduce a new trade.<Jff, controlled by
choice of h., between maintaining average earnings and reducing their variance.
J

4.2 Buffer Stock Strategies with Costs

Warehousing costs are clearly important in this case. Those costs will be
r.>..(S-D) where r is the unit storage cost. Buffer stock managers are likely to
J
have a lot of difficulty in measuring excess demand and supply accurately, so it is
convenient to reformulate the buffer stock rule (8) in terms of deviations of
actual prices from their target level. That is easy to measure (indeed it requires
no special model or analysis of the market to do so) and it puts the intervention
rule into the form in which it would be used in practice.

Evaluating (8) at p, using the supply and demand functions at (9), and
then substituting again for v and u, yields

BS. = 'Y'(p-p) where 'Y' = )..(iJ+b) (20)

J J J J
where v=Evj and iJ=EiJj" Hence, if the storage costs are rA/iJ+b )(p-p), producer
j's net revenue will be

l J = pq.J - A.(p-p)q. - rA .(!3+b)(p-p)

J J J
(21)

which implies E(l)=E(y?)-..\.p.(T (T., just as in (12). But

J J JJPJ
16

Vy. (0) +A.p

( s) = ( I-A. )2 Vy. 2_2 u.+rA.,B+bu
2 2 2( ) 2 (22)
J J J JJ J P

and

In contrast to the hedging case, it is now the earnings variance and optimal
buffer stock interventions - and not the average earnings - which have been
changed by allowing for operating costs. For calculation purposes we can rewrite
(23) as A*=A./B., and the variance reduction as A2/B., where A. was given after
J J J J J J
- 2 2
(14) and where Bj=B/r (,B+b)u p+2r(,B+b)p/puj . The buffer stock inter-
ventions are now clearly smaller if either P/O, or if P(O and the uncertainty is
more in prices and demand than on the supply side (i.e. up~u/ But large supply
shocks and P(O could lead to larger interventions (and gains) than before.
The transactions costs case is very similar. The buffer stock needed to hold
prices at their target level would cost rpA/S-D)=rpA/,B+b )(p-p) at those
prices. If r is once again the market rate of interest, this is the financing or
opportunity cost of maintaining the buffer stock. Repeating steps (21) to (23) we
find average earnings are still the same as in (12), but (,B+b) is replaced by
p(,B+b) throughout (22) and (23). Hence
A~=A./f3. and V(l)=V(y~)-A~/B. (24)
J JJ J J JJ

where A. is given below (14) and B.=B.+r2p2(,B+b)2u2+2rp(,B+b)p.u U" The

J J J P JPJ
same remarks about reduced interventions apply, but with added force since the
presence of p will tend to increase Bj above Bj"
The upshot of all this is that operating costs will make little difference to
producers' ability to use the futures markets to stabilize their earnings, although
their average net revenue will be reduced. But those costs will, in a market with
volatile prices, reduce their ability to use buffer stocks to stabilize prices/
earnings although their average net revenue would be no lower. This is consistent
with the empirical evidence that the main difficulty with buffer stock stabiliz-
 17

at ion schemes is their financing and operating costs. Those costs have the effect
of sharpening the inherent conflict between maintaining average earnings and
increasing their stability. 10

5. EMPIRICAL ANALYSIS
We have examined the price and earnings distributions for the major producers in
five international commodity markets. From the 29 most important commodities
reported in the World Bank's "Commodity Trade and Price Trends", which
covers UNCTAD's list of 15 "core" commodities,11 we selected Wheat, Coffee,
Rubber, Copper and Tin to illustrate the scope which pr~ducers have for
stabilizing their earnings. Three of these commodities (Coffee, Rubber, Tin) have
or have had stabilization agreements; four are predominantly produced in
developing economies (Coffee 90%, Rubber 99%, Tin 86%, Copper 60%, while
Wheat is 97% produced in industrialized countries ).12 Our list contains two metals,
two perennials, and one cash crop, to provide a roughly representative "sample"
of commodity types.
We use estimates of the first four bivariate moments (i.e. p, <i j , (T~, (T~, and
J.Lip J.Li2' J.L~1 and J.L~2) to construct optimal hedging and stabilization strategies
for the sample period 1973-87.13 We can then determine which strategy would be

10 See Hughes Hallett (1986).

11 These "core" commodities were those picked out by the policy makers as
prime candidates for stabilization. They were Wheat, Rice, Bananas, Sugar,
Coffee, Cocoa, Tea, Rubber, Cotton, Jute, Wool, Iron Ore, Copper, Bauxite and
Tin (see "Yearbook of International Trade Statistics", United Nations, New
York, 1982, vol. 2).
12 The coefficients of price variation over 1973-87 were 0.26 for Wheat, 0.34
for Coffee, 0.38 for Rubber, 0.44 for Copper, and 0.27 for Tin, with a maximum
of 0.45 across all 29 commodities. For comparison, electrical machinery had a
coefficient of price variation of 0.06 and clothing 0.08 for the same period. Price
variability is measured here by the coefficient of variation of the US dollar price
per unit deflated by the UN's index of export unit values for manufactured goods.
13 The data used for the calculations reported here consists of monthly time
series for January 1973-December 1987. It was obtained from: Prices: the UN's
'Monthly Commodity Price Bulletins' and the IMF's 'International Financial
Statistics'. Quantities: For Coffee and Rubber: FAO's 'Monthly Bulletins on
Agricultural Statistics'. For Copper and Tin: 'Metal Statistics' (Metallgesell-
schaft A.G., Frankfurt am Main). The quantity data was available ouly as
quarterly data. By considering the monthly export volumes (UN Commodity
Year Book, 1986) of countries which among them hold more than 60-70% of the
export markets of the respective commodities, the quarterly output data was
18

best for stabilizing each producer's earnings, both net and gross earnings, in each
market. These results allow us to make a rough assessment of the actual scope for
earnings stabilization using the futures markets or buffer stocks, and of the
potential conflicts between aggregate and private stabilization rules. Our results
apply to individual producers who account for the majority of the supply in their
markets: for Wheat it is 85% of the volume traded, for Coffee it is 44%, for
Rubber 92%, for Copper 54% and for Tin 51%.

5.1 Tests of Non-Normality

The appendix contains a summary of the estimated moments of the spot price,
production and earnings distributions for each producer in each market. In every
case, the third moment shows strongly positive skews. This reflects the stock-
holding behaviour mentioned in section 2.2.
The appendix also reports the estimated joint moments from the bivariate
(p,q) distribution for each commodity. One item of interest here is the signs of
the correlation coefficients: only 8 out of 17 producers face the negative (p,qj)
correlation which is usually assumed in this sort of analysis. The others show
weakly positive correlations. Nevertheless, this is sufficient to show that price
stabilization would produce lower average earnings for the US in the wheat
market; for the Ivory Coast in the Coffee market; for Malaysia, Indonesia and
Thailand in the Rubber market; for Zambia and Zaire in the Copper market; and
for Thailand in the Tin market. A second point is that 19 out of 34 bivariate
third moments have the signs predicted for them as a result of stockholding
behaviour. On that evidence it would be hard to argue that any of these
distributions are symmetric and normal.
The usual test of whether a variable's distribution departs significantly
from a normal distribution is either the Kolmogorov-Smirnov test (or one of its
variants) or the X2 goodness of fit test. But these tests are not very powerful and
do not reveal whether the normality in the distribution is due to "skewness" or
due to "kurtosis" (Miller, 1986). As a result Pearson (1963, 1965) recommends
the following test statistics be used to detect the non-normality:

converted into monthly data. For Wheat: the Australian Bureau of Agricultural
Economics. Futures Prices: 'The Financial Times' - Commodities Reviews of the
Week on the first Saturday of the Month, January 1973 to December 1987.
 19

Skewness: 1\ = 1/nE(Yi-y)3/[1/nE(Yi-y)2]3/2 = /13/ (13

Kurtosis: 62 = [1/nE(Yi-y)4/[1/nE(Yi-y2 ]2} - 3 = [/14/ (14]- 3

The calculated values of 61 and 62 for the price and quantity distributions
in our five markets are given in Table 1. Under a null hypothesis of normality,
the critical values at a 1% significance level for 61, with a sample size of 180 is
0.424 and for 62 it is 2.49 (Pearson and Hartley, 1966). Three out of the five price
distributions and 12 out of the 17 quantity distributions show asymmetries which
are statistically significant. There is less evidence of significant ,kurtosis - none of
the price distributions and only 3 out of 17 quantity distributions show it. But,
overall, all five markets and hence all producers face significant departures from
normality in their price-quantity distributions.

Table 1. Tests of Non-Normality by Commodity Market

Market Spot Prices Producer Quantity Traded

61 62 61 62

Wheat .024 -.730 USA .802* .291

Canada 1.071 * .794
France 1.143* 2.349
Australia 1.067* 1.736
Argentina 1.534* 1.838
Coffee .613* 1.249 Ivory Coast .344** -.765
Columbia .110 .294
Brazil .275 1.261

Rubber .724* .899 Malaysia .292 .617

Indonesia 1.149* 3.049*
Sri Lanka 1.615* 3.498*
Thailand 1.197* 2.999*
Copper 1.419* 2.055 Zaire .513* .139
Chile .052 .242
Zambia .609* .642
Tin .011 -.506 Malaysia .020 .181
Thailand .781* .488

* denotes significance at the 1% level; ** at the 5% level

20

5.2 Hedging versus Price Stabilization: No Intervention Costs

Table 2, parts (a)-(e), contains the result of applying optimal hedging and buffer
stock stabilization strategies to minimize the variability of each producer's gross
revenue. That is revenue stabilization without counting the storage or trans-
actions costs involved, as might be appropriate if the exercise were to be part of
some development project or aid programme. Net revenue stabilization, where
producers must count the cost of their interventions, is dealt with in Table 3 and
section 5.3.

a) Hedging: columns 2 and 3 of Table 2 show that the' optimal hedge varies a
certain amount between markets, but more so between producers within any
given market. In the Wheat market, the largest producer (the US) should hedge
significantly more than its expected output while Canada, France, Australia and
Argentina should only sell between a half and three-quarters of their expected
production forward. This difference is the result of a positive p,qj correlation for

the US, coupled with strong asymmetries in the form of /L~l <0 and /Li2>0, which
makes p<p=E(Pf) the more likely outcome. 14 In that case hedging more than qj'
and then buying any shortfall on the spot market, will yield extra profits on
average. (This effect would be offset to some extent if /L~2>0 since that would
reduce the expected fall in p for a given output disturbance. But, by the same
argument, /L~l <0 would increase the optimal hedge.) That explains the US result;
the negative correlations faced by the other producers produce the opposite
effect, but modified now by both /L~l <0 and /L~2<0 (Argentina excepted).
Essentially the same things happen in the Coffee market, with P/O leading
the Ivory Coast to go short and Pj<O leading Columbia and Brazil to hedge 80%
of their expected output. These positions are less extreme because the /L~l and
/L{2 asymmetries are stronger and serve to modify the hedging rule towards full

hedging. That happens again in Rubber (Malaysia and Indonesia go slightly

short, Sri Lanka is three-quarters hedged); in Copper (Zaire and Zambia go

14 See again the Appendix for these figures.

 Hedging Price Stabilization
% Reduction in % Reduction in
I Dominant
Producer V(Yj) h* ). ~ Earnings Variance t:.Eys(%) ).~ Earnings Variance Strategy
J J J J

(a) Wheat
USA 5.1 x 1010 4161.6 1.522 26.0 -2.1 1.550 33.2 S
Canada 1.2 x 10 10 902.3 .694 5.1 +0.6 .702 6.2 S
France 5.0 x 10 9 749.7 .884 8.7 -{l.0 .737 6.3 H
Australia 6.0 x 10 9 442.4 .481 2.5 +0.6 .343 1.5 H
Argentina 4.1 x 10 10 281.3 .717 1.5 +0.8 1.707 12.0 S
(b) Coffee
Ivory Coast 1.7 x 10 9 25.6 1.247 43.9 -2.3 1.131 51.6 S
Columbia 4.5 x 10 9 33.8 .759 29.7 +2.0 1.455 90.9 S
Brazil 1.2 x 10 10 54.1 .828 29.1 +2.5 .982 47.7 S
(c) Rubber
Malaysia 1.4 x 10 9 138.2 1.065 69.1 -{l.4 1.064 71.7 S
Indonesia 5.9 x 10 8 86.9 1.187 63.9 -1.3 1.231 72.9 S
Sri Lanka 2.7 x 10 7 8.3 .760 12.7 +1.3 .818 19.9 S
Thailand 3.9 x 10 8 43.0 .997 23.9 -{l.7 1.077 25.0 S
(d) Copper
Zaire 1.5 x 10 8 16.5 1.125 24.9 -{l.4 1.024 26.0 S
Chile 9.4 x 108 49.7 .860 36.9 +0.4 .967 43.8 S
Zambia 6.7 x 108 55.7 1.115 65.3 -{l.1 1.078 67.1 S
(e) Tin
Malaysia 6.3 x 10 8 5.2 .964 56.4 +1.0 .855 53.5 H .....
'"
Thailand 2.1 x 10 8 3.1 1.618 59.7 -9.5 1.617 78.9 S
Units: y. in 1000 $; h":: in 1000 MTonnes; % or ratios elsewhere
J J
22

slightly short, Chile is 86% hedged); and in the Tin market (Thailand goes short,
Malaysia is fully hedged). None of these results are changed if the forward
markets are biased or inefficient - and in fact the biases would lead to changes of
1% or less in average earnings of 1% or less.

b) Price Stabilization: Under this strategy, the US and Argentinean market

authorities would want to intervene very strongly in the Wheat market, while
Canada, France and especially Australia would prefer relatively weak buffer stock
interventions: see columns 5 and 6 in Table 2. Indeed bot,h the US and Argentina
need "over-stabilizing" interventions if they wish to minimize earnings
insta.bility. Canada and France, in contrast, require a 70% stabilization rule and
Australia a 35% rule. There is a clear conflict of interest here between those who
prefer vigorous interventions and those who would prefer only partial
stabilization.
In fact the conflict may be worse than that because although the reductions
in earnings variances are not very different from those achievable under hedging,
it is clear that the US, Canada and Argentina. would want to choose buffer stock
stabilization schemes, while France and Australia would prefer to hedge on the
futures market. It is possible that the relative gains are small enough for these
differences on the choice of strategy to be overcome,15 but agreement on how best
to operate the chosen strategy might be more difficult to achieve. On the other
hand, with the exception of the US's 33% reduction in income stability, none of
these stabilization gains are large enough to cause a great deal of enthusiasm for
either strategy.
The picture for the Tin market is somewhat similar. Thailand would want
"over-stabilizing" interventions while Malaysia would only want partial
stabilization. Thailand would prefer a hedging strategy, Malaysia price
stabilization. This time the potential gains are larger: a 56% reduction in
earnings variance in Malaysia and a 79% reduction in Thailand. Once again
disagreement on strategies might not cause serious problems for Malaysia who
would not lose much stability but would gain 1% in average revenue with a
switch to price stabilization. Nevertheless Malaysia and Thailand are quite likely
to disagree on the strength of the intervention rules.

15 Especially as France and Australia stand to make marginal gains in

average earnings, to offset the smaller variance reductions, if they switch to price
stabilization (col. 7).
 23

The Coffee, Rubber and Copper market results suggest no disagreements

between producers; they would all go for price stabilization schemes, by a wide
margin in the Coffee case and by rather narrower margins in the Rubber and
Copper markets. However there would be some differences of opinion about the
form of the decision rules. One producer in each market would want only partial
stabilization (albeit with intervention parameters of .82 to .98) while the
majority would prefer mild "over-stabilization".

c) The Dominant Strategies: The gains in earnings stability are, with the
exception of the Wheat market, quite large. For most produc~rs they are in the
range of reductions of 50% to 80% in earnings variability, although two producers
in the Rubber market and one in the Copper market could only achieve 25%
reductions. By and large, the bigger gains appear where the worst instability
problems have arisen (Wheat again excepted). So these are strong results even if
subject to a margin of error.
Second, assuming some flexibility on behalf of the two Wheat producers
and one Tin producer who would marginally prefer a hedging strategy, price
stabilization would be chosen in each of the five markets. However, except in the
Coffee market and for the Thai Tin producers, the margins are pretty small
- very seldom more than 5% points - and they are compensated by a rise in
average earnings in 8 out of 17 cases. So the losses from choosing the "wrong"
strategy are not large; it is significantly better for all producers to do something
than to do nothing. This is a comforting conclusion in a world where estimated
parameters inevitably introduce elements of approximation and error. Moreover,
given some flexibility by those who would marginally prefer to hedge, producers
individually always choose the same strategy as would minimize market
revenues, even if they disagree on the precise form of decision rule. But the costs
of those disagreements could be significant. Argentina's buffer stock activities in
the Wheat market would actually destabilize Canadian, French and Australian
earnings (.Aj>2)'~) and Thailand's activities in the Tin market come close to
being destabilizing.
More important perhaps is the distribution of the gains between producers.
That varies quite a bit in all five markets, and the average gain for the large
producers individually can be quite different from the potential market gain (e.g.
24

for Copper the average in 45% against a potential market gain 16 of 80%) which
suggests the private-market rule conflict may be serious and may affect the fringe
producers disproportionately.

5.3 Hedging versus Price Stabilization Allowing for Intervention Costs

So far our results have shown which strategy is most effective, but not most cost-
effective, for stabilizing earnings. How much difference would it make if
producers had to take the costs of their interventions into account? Would they
want to choose different stabilization strategies, and is the degree of revenue
stability seriously reduced?
Table 3 reports the results of stabilizing net earnings after allowing for
transactions and storage costs. For illustrative purposes we have assumed an
interest rate of 10% p.a. and, since our data refers to monthly production figures,
the costs have been assessed on a monthly basis (r=0.008).
As we saw in section 4, the hedging strategy and its earnings variance
reduction are unaffected by the intervention costs. Average earnings are however
reduced. It turns out that those reductions are small, typically about .75% and
rising to 1.2% in two cases. Hence, except for the Wheat market, the earnings
variance reductions by hedging are obtained relatively cheaply and producers are
unlikely to change their views on the scope for stabilization by hedging.
The results of price stabilization are much less clear cut. Nothing much
changes in the Wheat and Rubber markets (compare Table 2). Producers would
still prefer price stabilization for net revenues - except for French and Australian
Wheat farmers who still prefer hedging. The earnings variance reductions and
intervention strategies are not noticeably changed. The degree of earnings
stability achieved in the Coffee market is reduced a certain amount - but not by
enough to change the dominance of buffer stocks as a stabilization instrument. So
once again producers are unlikely to revise their views in the Wheat, Coffee or
Rubber markets.
However, things are quite different in the Copper and Tin markets. In the
Copper case, two producers (Zambia and Zaire) would want to switch to a
hedging strategy if they have to pay their own costs. The third (Chile) would
wish to stick with a price stabilization strategy. That produces yet another

16 See Hughes Hallett and Ramanujam (1990).

 Table 3. Optimal Hedging and Price Stabilization Strategies Including Transaction and Storage Costs (r=.008)
Hedging Price Stabilization
% Reduction in I % Reduction in Dominant
Producer h'l' b.EyJ (%) Earnings Variance b.EyJ{%) ASJ Earnings Variance Strategy
J

(aJ Wheat
UA 4161.6 -1.2 26.0 -2.1 1.550 33.2 S
Canada 902.3 -0.6 5.1 +0.6 .702 6.2 S
France 749.7 -0.7 8.7 -0.6 .737 6.3 H
Australia 442.4 -0.4 2.5 +0.6 .343 1.5 H
Argentina 281.3 -0.6 1.5 +0.8 1.707 12.0 S

(b) Coffee
Ivory Coast 25.6 -0.9 43.9 -2.2 1.039 47.4 S
Columbia 33.8 -0.5 29.7 +1.9 1.409 88.1 S
Brazil 54.1 -0.6 29.1 +2.4 .972 47.2 S

(c) Rubber
Malaysia 138.2 -0.8 69.1 -0.4 1.064 71.7 S
Indonesia 86.9 -0.9 63.9 -1.3 1.230 72.9 S
Sri Lanka 8.3 -0.6 12.7 +1.3 .816 19.9 S
Thailand 43.0 -0.8 23.9 -0.7 1.076 25.0 S

(d) Copper
Zaire 16.5 -0.9 24.9 -0.3 .673 17.1 H (Change)
Chile 49.7 -0.7 36.9 +0.3 .925 41.9 S
Zambia 55.7 -0.1 65.3 -0.0 1.025 63.8 H (Change)

(e) Tin
Malaysia 5.2 -0.8 56.4 +0.0 .007 0.41 H
Thailand 3.1 -1.2 59.7 -0.0 .002 0.08 H (Change)

Units: as Table 2. on
Aggregate market price coefficients (i3,b) are (0.167, 0.096); (0.25, 0.07); (0.065, 0.023); (0.71, 0.22); and (0.727, 0.064) for '"
Wheat, Coffee, Rubber, Copper and Tin respectively.
26

market where producers would find it hard to agree on a stabilization strategy.

Although Zambia would not lose much by joining a price stabilization scheme, it
would cost Zaire and Chile quite a lot to change to price stabilization or hedging
respectively.
By contrast, any previous disagreements in the Tin market are now
removed; Thailand joins Malaysia in preferring a hedging strategy. Another
major change appears to be that the stabilizing power of buffer stock inter-
ventions is almost completely wiped out by their cost. 17 No variance reductions in
net earnings are possible. That is the result of a high mean price in a market
which has reasonably strong price responses and a high price volatility compared
to the other markets. It therefore appears that the net versus gross earnings
distinction is only crucial for commodities which have a high unit value and price
sensitive demand/supply responses.

6. PRIVATE VERSUS AGGREGATE MARKET STRATEGIES

We have emphasized that every buffer stock transaction will affect every
producer's price distribution and that the degree of stability achieved depends on
the aggregate buffer stock movements. Hence the price stabilization results
reported in Table 2 are unlikely to emerge in a market of a homogenous good,
since no producer will succeed in imposing his privately optimal rule on all the
other producers in the market.1 8 However, Table 2's calculations are useful because
they allow us to assess how much the privately optimal intervention rules differ
from one another and how much they differ from the aggregate or public ally
optimal rule. We can see where the public/private conflicts are most likely to
occur, and how much imposing public or compromise rules would cost individual
producers in terms of lost stability. The latter gives us an idea of each producer's
incentive to participate in a price stabilization scheme.

17 It is interesting that Tin is the only market to show this characteristic

because it was the Tin buffer stock that became insolvent in 1985 and caused the
collapse of that market. According to these results, the Tin producers would have
done better to instruct their buffer stock manager to follow a hedging strategy.
18 The case of non-homogenous products is a good deal more complicated,
but allows for non-cooperative behaviour and for producers to try to manipulate
their own price distributions in order to stabilize their own earnings stream. That
case is ruled out in our analysis by the assumption of a single price and the fact
that our commodity markets all have a single reference price (any price
differentials can be fully accounted for by quality/grade differences). Production
from different sources is therefore taken to be perfectly substitutable.
 27

The same conflict does not arise under hedging because each producer can
affect the price distribution which he receives without changing that for other
producers. Nevertheless, as we pointed out in section 3, the earnings distributions
are still correlated, so it is not clear what impact a market hedging authority
might have on individual earnings streams.
Table 4 considers three cases. The first two columns indicate the maximum
gain which each producer could make via price stabilization rather than hedging.
That merely establishes the relative advantage of price stabilization, ignoring
intervention costs, were the market to adopt each producer's privately optimal
decision rule in turn. These relative gains are variable but mostly quite small;
less than 10% (Columbian coffee, Indonesian rubber, Thai tin, and the three
small relative losses excepted). On that basis producers might not be too
concerned about choOSing the wrong type of strategy.
But these comparisons are rather artificial since no producer will in fact
succeed in imposing his own optimal price stabilization rule on the whole market.
These are two other possibilities: producers agree on a compromise intervention
rule based on a weighted average of the privately optimal rules, or producers
agree to use a rule which is optimal for the market as a whole. Columns 3 to 6 of
Table 4 show the results of using compromise rules, where interventions are
controlled by a single price stabilization parameter which is an average of the
private parameters weighted by each producer's share in total production. For
comparison, results are also included for the case where a compromise hedging
rule is used in each market - i.e. where some market authority undertakes all
hedging operations on behalf of the producers. These compromise rules imply
rather small private losses; less than 7% for all but one producer who would lose
13% of his maximum stability. The gains over doing nothing are still substantial,
except for the four smaller wheat producers who could have gained rather little
anyway.
Those hedging results are also rather artificial because there is no
compelling reason why policy makers should not just aim to provide access to the
futures markets. The more interesting case is therefore the compromise price
stabilization rule where the results depend on aggregate buffer stock movements.
Nevertheless the story is remarkably similar in that case; small private losses
(less than 7%) emerge for all but two producers. However those two producers
(Columbian coffee, Thai tin) now lose about 40% of their very substantial
privately optimal gains. Consequently the overall picture is that compromise
 Table 4. The Losses in Earnings Stability When Market Rather Than Privately Optimal Stabilization Rules are Adopted
(in %) - No Intervention Costs
Private Rules Compromise Rules Aggregate Market Rules
Hedging Stabilization Hedging Stabilization 00
'"
Producer >. Gain (a) Gain (b) Gain (c) Gain (b) Gain (c) Gain (b) Gain (e) Gain (b) Gain (e)
(a) Wheat
USA 1.55 +9.7 -3.4 +22.6 -4.5 +28.7 -7.1 +18.9 -55.1 -21.9*
Canada .70 +1.2 -1.4 +3.7 -2.0 +4.2 -26.4 -21.3* -8.0 -1.8*
France .88 -2.6 --0.4 +8.3 -1.5 +4.8 -4.0 +4.7 -17.6 -11.3*
Australia .48 -1.0 -3.6 --0.9* -7.2 -5.7* -3.5 -1.0* --6.4 -4.9*
Argentina 1.71 +10.7 --0.3 +1.2 -1.8 +10.2 -22.1 -20.6* -15.0 -3.0*
>.h = 1.052 >. s = 1.086 >.h = 1.052 >. s = 1.048
(b) Coffee
Ivory Coast 1.13 +13.7 -7.0 +36.9 --0.1 +51.5 -4.7 +39.2 --0.5 +51.1
Columbia 1.46 +87.1 -1.0 +28.7 -38.6 +52.3 -2.4 +27.3 -78.4 +12.5
Brazil .98 +26.2 --0.1 +29.0 -3.1 +44.6 --0.7 +28.4 --0.3 +47.4
>.h = 1.167 >. s = 1.167 Ah = 0.94 AS = 1.04
(c) Rubber
Malaysia 1.06 +8.4 --0.2 +68.9 --0.5 +71.2 --0.1 +69.0 --0.1 +71.6
Indonesia 1.23 +24.9 -1.9 +61.0 -2.6 +70.3 -1.4 +62.5 -1.4 +71.5
Sri Lanka .82 +8.2 -2.4 +10.3 -3.0 +16.9 -14.6 -1.9* -10.2 +9.7
Thailand 1.08 +1.4 --0.2 +23.7 --0.0 +25.0 --0.3 +23.6 --0.0 +25.0
Ah = 1.077 >.S = 1.105 Ah = 1.053 AS = 1.082
(d) Copper
Zaire 1.02 +1.5 --0.4 +24.5 --0.0 +26.0 --0.2 - +24.7 --0.0 +26.0
Chile .97 +10.9 -1.5 +35.4 --0.3 +43.5 -2.0 +34.9 --0.3 +43.5
Zambia 1.08 +5.2 -2.1 +63.2 --0.5 +66.6 -1.3 +64.0 --0.4 +66.8
>.h = 0.996 >.S = 1.019 Ah = 0.89 >.S = 1.03
(e) Tin
Malaysia .96 -6.7 -4.0 +52.4 -6.0 +47.5 -3.0 +53.4 -8.6 +44.9
Thailand
I 1.62 +47.6 -13.3 +46.2 -45.3 +33.6 -9.1 +50.6 -10.0 +68.9
Ah = 1.134 AS = 1.053 Ah = 1.132 AS = 1.107
Key: Gain fa} = % gain of price stabilization over hedging in terms of reduction of the variance of earnings.
Gain b = % loss when compromise or aggregate market rule is adopted in place of the privately optimal rule.
Gain c = % gain/loss when compromise or aggregate market rule is adopted rather than doing nothing.
 29

rules would not cost individual producers very much in terms of lost earnings
stability - except for one or two particular cases. The private versus market rule
conflict is therefore not very strong and the incentives to join a collective
stabilization scheme are quite large unless: (a) one's private gains are very much
larger than those of other producers in the same market (e.g. Columbia in the
Coffee market, Thailand in the Tin market), or (b) the private gains are so small
that the compromise rule becomes net destabilizing (e.g. Australia in the Wheat
market).
Things are not so favourable for rules which are optimal for the market as a
whole (columns 7 to 10 of Table 4). Again this includes the results of an
aggregate hedging rule although there is no compelling reason why policy makers
should want to impose it. The private losses are (mostly) larger than in the
compromise rule case, confirming our earlier conjecture that average private rules
will perform better, given diversity between producers, than rules which stabilize
the market as a whole. Price stabilization seems to imply losses which are slightly
smaller than hedging but with a few dramatic exceptions. The US in the Wheat
market and Columbia in the Coffee market both stand to lose more than 50% of
their potential gains, while there are no such "outliers" under aggregate hedging
rules. Moreover these larger losses imply that a number of producers will find
aggregate market rules net destabilizing - all the wheat producers and Sri
Lankan rubber producers (i.e. all those whose maximum gains were 25% or less)
would be destabilized. Consequently the conflicts between private and aggregate
intervention rules are likely to prove a much greater obstacle when policy makers
attempt to introduce market-wide stabilization policies. Those problems will
emerge in markets where (a) the maximum potential stabilization gains are
anyway small, or (b) there is great diversity in the potential private gains within
the market, or (c) there is a considerable gain in using hedging over price
stabilization (or vice versa).

7. CONCLUSIONS
This paper has derived and compared optimal hed.ging and price stabilization
strategies for individual producers in markets with price and production
uncertainties. These strategies are distribution free and model free, although the
price stabilization case implicitly assumes linear supply and demand functions.
We have also assumed the moments of the joint price-{)utput distributions, but
30

not necessarily the price elasticities or other model parameters, to be invariant to

any new interventions.
The second contribution was to extend those strategies for stabilizing net
revenues; that is to allow for transactions and storage costs. We have then
applied our decision rules to the problem of stabilizing the revenues of the major
producers in five important commodity markets. While two markets contained
producers who would disagree on the type of strategy which should be adopted,
the main differences were in the different strength of interventions which
different producers would prefer. That would of course be no problem had they
all decided to hedge on the futures markets. But, in fact, all but 3 out of 17
producers would select price stabilization. That makes agreement on the inter-
actions a necessity.
Producers could alternatively agree to a compromise rule or to stabilize
market revenues as a whole. We therefore examined the losses suffered by each
producer when either a compromise intervention rule or one which stabilizes
market revenues is imposed, taking that producer's privately optimal outcome as
the point of comparison. It is clear that compromise rules lead only to small
losses, and those losses are never large enough to disturb the superiority of price
stabilization as a strategy (Table 4). The aggregate market rules perform rather
worse because they would actually destabilize the earnings of producers whose
potential gains from stabilization are small and because the stabilization gains
are rather unevenly distributed.
Allowing for operating costs did make considerable changes in two of the
five markets. There are small losses in the amount of earnings stability achieved
(less than 5% in all but 3 out of 34 cases) and small changes in the average
earnings levels (again less than 2.5% in all cases). However that is enough to
make hedging superior for three more producers. The main change is that a buffer
stock's ability to stabilize earnings is wiped out for high unit value commodities
because the cost of financing the necessary stocks seriously reduces net earnings.
Among our examples, this happens in the Tin market and, to a smaller extent, in
the Copper market. Transactions and financing costs on the futures markets, on
the other hand, imply relatively small falls in average earnings and no losses in
stabilizing power. Hence, as a rule of thumb, hedging is likely to be more
effective for stabilizing the earnings from high value commodities and price
stabilization is likely to be superior for commodities with low unit values.
One final consideration is that, in practice, many buffer stock stabilization
 31

schemes have run into difficulties because they have inadequate stocks to restrain
price rises in boom periods and inadequate finance to purchase stocks when prices
are low. Naturally that has caused those schemes to collapse or become
ineffective. One might therefore suspect that our results are too favourable to
price stabilization since our buffer stock interventions appear to be imperious to
the threat of stock-out or bankruptcy. However, as soon as the transactions cost
terms are included, the dangers of stock-out or bankruptcy do influence the
amount of stabilization undertaken since changes in the size and costs of holding
stock contribute directly to the level and variability of net earnings. The larger
the stock, the greater its financing costs. That would penhlize any further
increases and hence reduce the chances of financial collapse. On the other hand,
the smaller the stock, the higher are net earnings and the more will interventions
move away from possible stock-out. As a result we find smaller interventions
when transactions costs are introduced, plus a tendency for producers to switch
to hedging in order to reduce the chances of stock-out or bankruptcy. Tables 2
and 3, for example, show the intervention parameters are uniformly smaller when
costs are incorporated, and a lot smaller in Copper and Tin where the danger of
collapse is large. Only in Wheat, where the scale of intervention is so small as to
rule out any significant danger of collapse, do we observe no appreciable changes.

REFERENCES
Anderson, R Wand J Danthine (1983), "Hedger Diversity in Futures Markets",
Economic Journal, 93, 370-89.
Gemmill, G (1985), "Forward Contracts or International Buffer Stocks? A Study
of Their Relative Efficiencies in Stabilizing Commodity Export Earnings",
Economic Journal, 95, 400-17.
Ghosh, S, C L Gilbert and A J Hughes Hallett (1987), Stabilizing Speculative
Commodity Markets, Oxford University Press, Oxford.
Gilbert, C L (1985), "Commodity Price Stabilization: the Massell Model and
Multiplicative Disturbances", Quarterly Journal of Economics, 100,
635-40.
Gilbert, C L (1988), "Buffer Stocks, Hedging and Risk Reduction", Bulletin of
Economic Research, 40, 271-86.
Hughes Hallett, A J (1986), "Commodity Market Stabilization and 'North-South'
Income Transfers", Journal of Development Economics, 24, 293-316.
Hughes Hallett, A J and P Ramanujam (1990), "The Role of Futures Markets as
Stabilizers On Commodity Earnings" in L A Winters and D Sapsford
(eds.), Primary Commodity Prices: Economic Models and Economic
Policy, Cambridge University Press, Cambridge and New York.
Karp, L. (1988), "Dynamic Hedging with Uncertain Production", International
Economic Review, 29, 621-37.
Kawai, M (1983), "Spot and Futures Prices of Nonstorable Commodities under
32

Rational Expectations", Quarterly Journal of Economics, 98, 235-54.

Levy, Hand H M Markowitz (1979), "Approximating Expected Utility by a
Function of Mean and Variance", American Economic Review, 69,
308-17.
McKinnon, R I (1967), "Futures Markets, Buffer Stocks, and Income Stability for
Primary Producers", Journal of Political Economy, 75, 844-61.
Miller Jr, R G (1986), Beyond ANOVA, Basics for Applied Statistics, John Wiley
& Sons, New York.
Newbery, D M G (1988), "On the Accuracy of the Mean-Variance Approxi-
mation for Futures Markets", Economic Letters, 28, 63-8.
Newbery, D M G and J E Stiglitz (1981), The Theory of Commodity Price
Stabilization, Oxford University Press, Oxford.
Nguyen, D T (1980), "Partial Price Stabilization and Export Earnings
Instability", Oxford Economic Papers, 32,340-52.
Pearson, E Sand H 0 Hartley (1966), Biometrika TableS for Statisticians, Vol. 1,
Cambridge University Press, Cambridge.
Turnovsky, S J (1983), "The Determination of Spot and Futures Prices with
Storable Commodities", Econometrica, 51, 1363-87.
Wright, B C and J C Williams (1982), "The Economic Role of Commodity
Storage", Economic Journal, 92,596-614.

APPENDIX: Estimated Bivariate Moments of the Price-Quantity Distributions

in 5 Markets 1973-1987
Wheat
Mean Variance 3rd Moment 4th Moment
Price: (US$/MT) 146.6 770.8 507.6 1.34 x 10 6

Quantity: (' OOOMT)

USA 2735.1 1.699 x 10 6 1.777 x 10 9 9.499 x 10 12
Canada 1301.0 5.426 x 10 5 4.282 x 10 8 1.117 x 10 12
France 848.5 2.201 x 10 5 1.180 x 10 8 2.589 x 10 11
Australia 919.4 2.912 x 10 5 1.677 x 10 8 4.016 x 10 11
Argentina 392.2 1.709 x 10 5 1.084 x 10 8 1.413 x 10 11

USA Canada France Australia Argentina

JL21 -144560.3 -61457.7 -76309.8 -54980.2 -44776.5

JL12 5.910x10 6 -9.591x10 5 -1.057x10 6 -1.472x10 6 9.186x10 5

~2 1.473x10 9 4.043x10 8 1.569x10 8 2.329x10 8 8.460x10 7

P 0.210 -0.082 0.0001 -0.147 -0.076

 33

Coffee
Mean Variance 3rd Moment 4th Moment
Price: (US$/MT) 2852.3 1.158 x 10 6 7.639 x 10 8 5.698 x 10 12

Quantity: (I OOOMT)
Ivory Coast 20.54 104.23 366.09 2.428 x 10 4
Columbia 44.51 223.27 366.00 1.642 x 10 5
Brazil 65.42 740.82 5551.9 2.339 x 10 6

Ivory Coast Columbia Brazil

2.336 x 10 6 -7.414 x 10 6 -3.482 x 10 4
il:21
tt12 6481.6 3.833 x 10 5 2.632 x 10 5

tt22 1.152 x 10 8 2.818 x 10 8 8.362 x 10 8

P 0.113 -D.1094 -D.1589

Rubber
Mean Variance 3rd Moment 4th Moment
Price: (US$/MT) 919.16 50002.3 8.106 x 10 6 9.748 x 10 9

Quantity: (I OOOMT)
Malaysia 129.72 307.88 1577.53 3.429 x 10 5
Indonesia 73.26 156.90 2257.18 1.489 x 10 5
Sri Lanka 10.89 24.65 197.67 3948.3
Thailand 43.18 326.92 7072.87 6.412 x 10 5

Malaysia Indonesia Sri Lanka Thailand

tt21 27310.7 27094.1 11117.37 -2.561 x 1011

tt12 5255.08 19241.8 96.62 -4652.14

tt22 2.654 x 10 7 1.341 x 10 7 1.859 x 10 6 1.259 x 10 7

P 0.110 0.256 -D.139 0.067

34

Copper
Mean Variance 3rd Moment 4th Moment
Price: (US$jMT) 1600.1 1.409 x 10 5 7.514 x 10 7 1.005 x 1011

Quantity: (' OOOMT)

Zaire 14.69 44.56 152.99 6250.6
Chile 57.74 206.97 152.40 1.150 x 10 5
Zambia 49.98 92.87 545.15 31447.06

Zaire Chile Zambia

/1-21 9.595 -5.96 x 10 5 2.00 x 10 5

/1-12 -932.46 1.039 x 10 4 -51.63

/1-22 5.132 x 10 6 4.0435 x 10 7 1.455 x 10 7

P 0.0407 -0.0624 0.105

Tin
Mean Variance 3rd Moment 4th Moment
Price (US$jMT) 10694.4 1.319 x 10 7 5.398 x 10 8 3.290 x 10 14

Quantity (' OOOMT)

Malaysia 5.392 2.645 0.0867 22.255
Thailand 1.907 0.5982 0.3613 1.2482

Malaysia Thailand

/1-21 4.671 x 10 6 1.883 x 10 6

/1-12 -2620.61 1229.87

/1-22 2.935 x 10 7 1.002 x 10 7

P -0.114 0.455
 CHAPTER 2

HEDGING COMMODITY EXPORT EARNINGS WITH

FUTURES AND OPTION CONTRACTS 1

Joseph Ntamatungiro

L INTRODUCTION
Less Developed Countries (LDCs) are subject to variability of the income they
derive from exportation of primary commodities. The associated risk arises from
the variability of both prices and production. Fluctuating export earnings curtail
LDCs' economic growth especially when a limited number of goods is exported.
This paper proposes futures and option contracts as efficient hedging
instruments.
Complete price stabilization schemes at an international level have proven
to be unsuccessful because they are very costly, ineffective in reducing income
risk and theoretically unrealistic. Policy makers in fact choose a level of desired
buffer stock by determining the limits within which the price may fluctuate
around the long run price according to their budget constraint. Economists
specify the band width by choosing the optimal degree of intervention
(analytically or by simulation). In a loglinear model, Nguyen (1980) shows that
partial stabilization may stabilize earnings even in situations in which complete
stabilization may destabilize it. However, he does not consider the fact that the
stabilization authority has incomplete information on the long run equilibrium
price, nor does he integrate in his model the expectations that economic agents
form on prices. Finally, since producers do not constitute an homogenous group,
partial stabilization may destabilize the revenue of some producers. For these

1 This paper is based on Parts III and IV of Ntamatungiro (1988). I am

indebted to professors J.P. Danthine who directed this research and A. Holly for
his help and encouragement.
35
L. Phlips (ed.). Commodity, Futures and Financial Markets. 35-58.
© 1991 Kll.fWer Academic Publishers.
36

reasons, stabilization gains are likely to be over estimated (Ntamatungiro, 1988).

The extensive literature on price stabilization by intervention hardly
mentions that the producer has the possibility of hedging his revenue on futures
and options markets. however these markets contribute to improving the
completeness of commodity markets (Breeden, 1984). The farmer may be short
on the futures markets by selling a portion of his crop for future delivery at a
fixed price, protecting himself from the vagaries of the spot price. Unfortunately,
with such a futures position he cannot benefit from a rise in spot prices. The
option is a more flexible hedging tool since it protects him from the price fall
while allowing him to benefit by selling his output at high spot prices. However
the option is not a "free lunch", for the producer has to pay the option premium.
Futures and option markets prove to be efficient in reducing the producer's
revenue fluctuations. When the hedge is not complete on one market, the
producer may increase his revenue insurance by jointly using futures and options
contracts.
Section 2 tests the hedging efficiency of futures contracts as proposed by
McKinnon (1967), Gemmill (1980) and the United Conference on Commerce and
Trade: UNCTAD (1983). Following Anderson and Danthine (1981), optimal
hedge ratios are derived for a naive pre-harvest hedging strategy. Full hedge
hypothesis is generally rejected for the majority of exporters of the sample,
mainly because of basic risk and production variability.
In Section 3, hedging opportunity by option markets is explored. As options
on soft commodities are written on futures contracts (no spot options), the hedge
will be "indirect". The optimal position on the options markets is derived
according to Wolf (1983). The agent may also use options and futures jointly.
Integration gains are computed.

2. HEDGING WITH FUTURES MARKETS

Traditional theory defines hedging as taking a position on the futures market
equal in magnitude, but of opposite sign to the position on the spot market
(Hieronymus, 1970). A producer, normally long on the spot market should be
short on the futures market. But traditional theory is misleading for it neglects
basis risk (Working, 1953), production variability (Rolfo, 1980) or multiple
production (Anderson and Danthine, 1981). The producer's optimal position is a
complex result of the joint probability distribution of his Qutput, spot prices and
futures prices. It also depends on the producer's risk preferences, for the hedger
 37

uses futures markets not only for insurance, but also for speculation. In what
follows, I derive the producer's optimal position on futures markets according to
Rolfo (1980) and Anderson and Danthine (1981).

2.1 An Optimal Model of Hedging on Futures Markets

Two dates are to be considered: t and T(t~T). In t, the producer decides to sell n
goods Y at time T at a vector of prices P. But he knows neither the quantity
which will be available in his stocks for sale nor which prices will prevail in
period T. He is subject to price and quantity uncertainty. Besides the n physical
markets, there exist n futures markets on the n goods. A futures contract is an
agreement to buy or to sell a stated quantity of an asset of a given quality for
delivery (in a specified place) at a future date at a specified price. The delivery
date for the traded futures contracts is 1 (T~I). In periods t and T, futures prices
are, respectively, F! and F~. For commercial reasons, the producer may choose a
vector (f) of positions on the n futures markets. Let fi (i=l,n) be a short position
when positive and a long position when negative. Positions f will be closed in T
by an offsetting trade (financial transaction if T<I, and delivery if T=l). Usually
the offsetting operation is made by a financial transaction in T consisting of
purchasing (or selling) f at the random vector F ~ if the producer is short (or
long) on the futures market in period t. The producer has a subjective probability
distribution on prices and quantities. Let ~ be the variance/covariance matrix of
(Y, P, F~):

~11 ~12 ~13

~ = ~21 ~22 ~23

~31 ~32 ~33

where ~33 = V(F~) is a symmetric and positive definite matrix. In T, the

producer's random revenue is
Rf-- ply + (FI_Fl)/f-
t T -
R c + (Fl_FI)/f
t T . (1)
The hedger uses futures contracts as a hedging and speculation management tool
by choosing f in order to maximize his certainty equivalent EC:
Max EC(f) = E(Rf) - tA V(Rf) (2)
where
38

E(Rf) = E(Rc) + f' E[F!-F11

V(Rf) = V(Rc) + f/E33f - 2f' Cov(Rc, F 1) . (3)
A is the Arrow-Pratt absolute risk aversion coefficient
Necessary (and sufficient because E33 is positive definite) conditions for a
maximum are given by:
1 1 1
E[F t -F T1 - A[E33f - Cov(R c' F T)l = 0
and the optimal positions are

f* = (1/A)E3~ E[F!-F11 + E3~[Cov(Rc' F1)1 (4)

The optimal hedge consists of two parts:
a) The pure speculation position: (1/A)E3~ E[F!-F11 is the optimal position of
an operator whose physical position equals zero (Danthine, 1978) or, more
generally, whose revenue Rc is non-random. This position depends on the
subjective futures price probability distribution and on risk aversion. It tends to
zero as the absolute risk aversion coefficient A tends to infinity. It also vanishes
when the vector of expected gains on futures II=E[F! -F11 equals zero. When
n=1, it has the same sign as II. In what follows, this part is neglected for two
reasons. On the one hand, LDCs are assumed to be highly risk averse. On the
other hand, futures market's efficiency assures 2 that E[Fs-Fs+i1 = 0 where E is
the expectation conditional on information in s (0~i9-s).
b) The pure hedge position: ~ = E3~[Cov(Rc' F1)1 is a function of the subjective
probability distribution of spot revenue and futures prices. 3 It does not depend on
the risk aversion parameter. It is verified that:
~ = lim f* and ~ = f* when II=O .
A-tw
With no loss of generality we assume that the pure hedge position is a fair
proxy of the optimal position. It may be shown that ~ minimizes the variance of
the producer's revenue given by (3). So the hedger is seen as a risk-avoider. he
enters futures markets in order to minimize the risk usually represented by the
variance of his revenue.

2 See Dusak (1973), Kofi (1973) or Bigman et al (1983). This hypothesis is

confirmed when correct expectations and log-normality are both assumed.
3 Johnson (1960) and Ederington (1979) derive the same result when n=1
and Y non-random. McKinnon (1967) finds a similar result for n=1 when P and
Yare normally distributed.
 39

The effectiveness of such a strategy is derived by comparing the risk of an

unhedged income V(Rc) with the minimum variance V*(Rf ), where V*(Rf ) is
V(Rf ) of (3) when f equals ~. The relative reduction in variance is

or

[COV(RC,Fiw E3~ [Cov(Rc,Fi)]

M- (5)
f - V(Rc)
As E33 is a positive definite matrix and V(Rc) positive, Mf is a positive
quadratic form (MFO). SO the pure hedge is effective for it reduces the variance of
spot revenue. The greater is Mf the more efficient is the pure hedge. Variance is
completely eliminated (pI) when production is not random and when there
exists no basis risk. In this case the pure hedge corresponds to a routine hedge.
The parameter Mf is the population coefficient of determination between Rc and
1
FT"
Pure hedge positions may be derived by estimating the coefficients of F i in
a theoretical multiple regression of Rc on F i:
n 1
R=a+Eb.FT+u (6)
c i=1 1 1
where the usual conditions of linearity are assumed to hold. Hedging effectiveness
is estimated by the multiple regression coefficient R2 of model (6).

2.2 Estimation Results

Optimal positions were estimated for a sample of 23 cocoa, coffee, cotton and
rubber exporting LCDs for the period 1968-1985 (1973-1985 for rubber).4 A fixed
pre-harvest hedging strategy is proposed. Positions are taken at the beginning of
the official harvesting period (t). Two delivery months reflecting supply and
demand conditions were considered. The hedging period is either one month (T 1)
or the position is offset in the middle of the delivery month (T 2).
Estimation results are grouped in Tables 2a-d in appendix A.I. Optimal

4 All my thanks go to the CSFM (Center for Studies of Futures Markets) of

the University of Columbia which provided me with futures price series.
40

positions are generally short (exceptions are in cotton: India and Brazil). The
estimated pure hedging position's magnitude varies substantially in different
countries. The highest hedge ratioS is 123% (coffee) and the lowest is (16%) cocoa.
The routine hedge hypothesis is rejected for the majority of the countries
considered. The theoretical result of hedging effectiveness (Jhf~O) is confirmed
empirically. The pre-harvest strategy could conSiderably reduce the variance of
the revenue of the countries under investigation. The variance reduction Jhf ranges
from a minimum of 4% (cotton) to a maximum of 87% (cocoa). Optimal
strategies are reproduced in Table 1 for different countries:

Table 1. Pre-harvest hedging strategy (1968-1985)

Commodity Country t(entry) Delivery Hedge Variance
month ratio' reduction
(offsetting period)

Brazil October March T2 70% 87%

Cameroon October May T2 26% 57%
Cocoa Ivory Coast October March Tl 29% 11%
Ghana October March Tl 34% 38%
Nigeria October March Tl 51% 49%
Togo October March Tl 59% 41%
Burundi April September T2 51% 51%
Brazil April September T2 24% 32%
Colombia October March Tl 60% 58%
Coffee Ivory Coast October May T2 71% 81%
Kenya October March T2 56% 74%
Uganda October March T2 27% 75%
Rwanda April September T2 46% 49%

Brazil October May 85%L 52%

Egypt October March 44% 38%
T'l
Tl
Cotton India October March Tl 67%L 11%
Sudan October May T2 114% 45%
Chad October May T2 56% 46%
Indonesia April April Tl 30% 36%
Liberia April April Tl 40% 53%
Rubber Malaya April April Tl 63% 61%
Thailand April April Tl 37% 50%
Zaire April April Tl 94% 63%

* 1973-1985 for rubber; L: Long position;

• Hedge ratio 5=~/y where y is the mean of exports during the period

S Hedge ratios are relatively stable. A stability test revealed no significant

changes of hedge ratios through time.
 41

2.3 Optimal Hedging and Production Uncertainty

Production uncertainty is considered an impediment,by the commodity exporting
LDCs, for using futures markets. Rolfo (1980) thinks that production variability
is responsible for the limited usage of futures markets by LDCs.6 In this section, I
identify the production variability's effect on the optimal hedge. Proposition 1
states conditions under which production variability reduces the optimal position,
while proposition 2 shows that production uncertainty does not necessarily reduce
the optimal hedging position.

Proposition 1: When the producer maximizes his expected utility, production

variability will reduce his optimal hedging position if it reduces his pure hedging
position, i.e. when

d = E3~E23E(Y) - E3~Cov(Rc' F J,) > 0 (7)

The proof follows result (4). Production variability does not affect the pure
speculation part of f* which is independent of Y. However it influences the pure
hedge position since this part depends on production Y. So production
uncertainty reduces the optimal hedge (d>O) when the pure hedge position of
non-random production (first part of (7)) is greater than the pure hedge position
of uncertain production (second part of (7))."

Proposition 2: When production Y is variable but independent of futures and

spot prices, its variability has no effect on the optimal position (d=O).
The proof (see Appendix A.2) is derived from the statistical statement that:
Cov(xy, z) = E(y)Cov(x, z) when y is independent ofx and z."
The variability effect may be estimated by regressing DREV = (y-Y)P on
FJ, according to the model (for n=l)
DREV = c + FJ,d + e . (8)
Estimation was conducted for cocoa and coffee. Estimated d (not re-
produced here) are positive for the majority of countries, but d is negative for
Brazil and Togo for cocoa, and Kenya for coffee. This confirms the idea that
production variability may decrease the optimal hedging position but it may

6 See also IBRD (1986). The authors assume routine hedge. But it is well
known that routine hedge is not optimal even when production is non-random
except when there is no basis risk.
42

increase it as well. Generally, estimated d's are statistically insignificant. Thus

production uncertainty should not necessarily influence the choice of optimal
position by LDCs.

2.4 Estimating the llisk Gains

A hedger will enter futures markets if such a strategy dominates his marketing on
spot markets. The gain from the futures market is estimated by G{
Gf = EC(f*) - EC(O)

A[V(Rc) - V*(Rf )]
C(f*)
2

C(f*) (9)

where EC(f*) and EC(O) are, respectively, the certainty equivalent when f=f*
and when f=O; a' is the relative risk-aversion coefficient; C(f*) represents the
estimated costs due to position f*. Transaction costs are financial costs relative to
initial margins (1,750$ per contract for coffee, 750$ for cotton, 500$ for cocoa,
and 6% of the face value of a rubber contract), and commission fees (30$ to 100$
by contract). The interest rate is fixed at 9.97% and commission fees at 65$ per
contract. The relative risk-aversion coefficient is unitary (a' =1),1
Table 3 gives the maximum degree of risk reduction with the pre-harvest
strategy /lp and its absolute and relative gains. The absolute gain Gf is positive
for the majority of the countries in the sample. Gains are considerable for cotton
(71% for Brazil), cocoa (7.6% for Brazil) and coffee (9.6% for Kenya). They are
relatively low for rubber, ranging from 0.0% (Indonesia) to 2.5% (Zaire). But
because /If is positive, those gains are greater when a' >1. When the commission
fee is fixed at 100$ per contract, gains decrease but remain largely positive for
cocoa, coffee and cotton.

7 Additional margins due to marking-to-market operations should be

considered especially when the operator is constrained on the credit market. It is
assumed here that Gf covers potential additional margins.
 43

Table 3. Gains of the pre-harvest hedging strategy (1968-1985)*

Commission=65$ Commission=100$
/1-f Gf ~ Gf ~
(in %) (in %) (in %)

Cocoa Brazil 87 661 7.6 644 7.4

Cameroon 57 54 1.1 49 1.0
Ivory Coast 11 182 1.5 169 1.4
Ghana 38 401 2.4 383 2.3
Nigeria 51 670 5.7 653 5.6
Togo 59 53 2.3 51 2.2
Coffee Burundi 51 52 2.8 51 2.7
Brazil 33 775 1.1 748 1.1
Colombia 58 849 1.9 819 1.8
Ivory Coast 81 791 5.0 774 4.9
Kenya 74 690 9.6 686 9.5
Uganda 75 351 3.0 346 3.0
Rwanda 49 68 4.7 67 4.6
Rubber Indonesia 36 -2 -<l.0 -70 -<l.3
Liberia 53 5 0.2 -3 -<l.1
Malaya 61 319 0.6 62 0.1
Thailand 50 46 0.3 -<l.01 -<l.0
Zaire 63 14 2.5 9 1.6
Cotton Brazil 52 1871 71.0 1862 70.0
Egypt 38 881 4.5 873 4.5
India 11 31 2.2 28 2.0
Sudan 45 528 5.4 513 5.2
Chad 46 35 2.1 33 2.0
Arithmetic mean 405 6.0 380 5.8
Standard deviation 97 0.031 98 0.031
* 1973-1985 for rubber.
Gains are estimated for a' =1, lfGf/M where M is the geometric mean of
revenue. Gf is in 10.000 US dollars.

3. HEDGING WITH OPTION CONTRACTS

3.1 Introduction
The recent two decades have been characterized by both a growing literature on
options and a huge volume of transactions on option markets. This is partly
because of the CFTC's 3-year pilot programme launched in 1981 and the regular
trading that followed then. Another reason is the growing interest in the
44

academic circle since the general equilibrim option's valuation by Black and
Scholes in 1973. An option is a contract that gives the buyer the right (not the
obligation) to buy or purchase a given quantity of an asset at a fixed price called
the exercise price (or the striking price) K. The option that gives the right to buy
is called a call while the option that gives the right to sell is called a put. The
seller of an option is a grantor or a writer. The option's price, paid by the buyer
to the writer, is the premium W. Some options are written on physical goods
while others are written of futures contracts, e.g. for agricultural commodity
options. The call (put) is usually exercised if the price of the underlying asset is
greater (smaller) than the exercise price. The option is' a more flexible hedging
instrument than the futures contract. It enables the buyer not only to benefit
from a favourable evolution of prices but also to limit losses in unfavourable
situations. Contrary to a futures contract, an option contract demands no
additional margins except for the writer of a futures option. In what follows it is
shown how the producer may hedge the revenue of one good with option
contracts.
Suppose that besides the spot market the producer has the opportunity to
buy 0 put options (results may be derived similarly for call options) which expire
in T at the prime W. When 0 is positive the producer is a buyer, while he is a
writer when 0 is negative. Options are written on futures contracts, so the hedge
is 'indirect'. The option's subjective value is w = Max(O, K-F j,). Risk neutrality
on the option market imposes E(w)=W. When futures prices are generated by a
log-normal process it is shown in appendix A.3 that
(10)

E(F j,) S2]

b1 = [ In - K - + ;- /s

E(Fj,) S2]
b2 = b 1 - s = [ In - K - + ;- /s

where T = T-t, K the exercise price, and NO the cumulative standard normal
distribution. The futures put's value corresponds to the Black evaluation formula
when s=(J2 T and E(Fj,)=F!. The producer's revenue in T then is
 45

Ro = Rc + O(w - Werr)
As in section 2, it may be shown that the pure hedge position on the options
market is
Cov(Rc'w)
0* = - -----=-- (11)
c V(w)
and

(12)

where
V*(Ro) = V(Ro) I .
O=O~
Then the pure hedging strategy by option contracts is effective since it reduces
the variance of revenue. If there is no basis risk, futures options correspond to
spot options.

Proposition 3: When production is non-random and when there exists no basis

risk, then the pure hedge position is an approximation of the hedge ratio 0 of
Black and Scholes (1973). Furthermore, the hedge is direct (see the proof in
appendix A.4).

3.2 Joint Hedge with Futures and Option Contracts

Suppose now that the producer takes a position f on the futures market and buys
a quantity of 0 puts on the option market. His revenue (n=l) in T then is
R = Rc + (F! - F4)f + (w - WlT)O (13)
Optimal positions on the futures and the options markets are derived under
the hypothesis of risk neutrality. They minimize the variance of the producer's
revenue and are equal to

where
46

2 1
2 Cov (w,F T )
, = 1 .
V(w)V(F T )
The optimal position on one market is a linear combination of its pure
hedge position with a part from the existence of the other market, the weights
being 1/(1-,2) and -l/(I-l). The pure hedge positions (~ and O~) are those
which separately minimize the variance of revenue on each market.

Result 1: When the correlation between F ~ and w tends to zero, optimal

strategies on futures and options markets are separable. Indeed r=0 implies
f*=f*
o c and 0*=0*. c

Result 2: When a market is redundant, it does not influence the optimal position
on the other market. Indeed when f*=0, then O*=O~ and when 0*=0, then
f*=f*.
o c

Result 3: When production y is non-random and when there exist no basis risk
then the options market is redundant for 0*=0. The optimal position on the
futures market is a routine hedge. Redundance follows the fact that routine hedge
on the futures market eliminates all variance. 8

3.3 The Joint Hedging Performance

The maximum degree of variance reduction depends on the correlation between

futures prices F~ and the value of the option w. In proposition 4 joint hedging
effectiveness is stated. Proposition 5 gives sufficient conditions under which a
market is redundant (see proof in appendix A.5).

Proposition 4: Joint hedging is effective. It reduces the variance of the producer's

revenue. It is more effective than separately hedging on futures or options
markets, for its minimum /J corresponds to the maximum variance-reduction
degree of each separate strategy. The minimum /J corresponds to I1f (110 ) if the
correlation between Rc and F ~ is greater (less) than the correlation between Rc
and w.

8 See also theorem 3 of Wolf (1983).

 47

Proposition 5: At the minimum of I'fo' one of the two markets is redundant.

When 'F'h the futures market is redundant (~=O, O*=Oc and I'fo=l'o) whereas
when 'F'Y2 the option market is redundant (0*=0, ~=~ and I'fo=l'f)'

Figure 1 describes I'fo according to whether ~ and 0* are either of the same sign
or of opposite signs. The joint hedge eliminates all the income's variance for
1
'Y= p(Rc,FT)p(Rc'w) ± v'(1-1'0) (1-l'f)' If 'FO, then I'f= 1'0 + I'f The para-
meter fJ gives the gain of the joint hedge with regard to hedging on one market
(fJ = I'fo -I'f or fJ = I'fo -1'0)' I'fo reaches its minimum II when one market is
red undan t (result 2).

~Of
Positions have Positions have
opposed signs the same sign

------------------j------- -------,------------------ ~o + ~f

~ I I ~

v
J.
• v = Min.

-1 o
Figure 1. Joint hedging effectiveness
48

4. CONCLUSION
This paper aimed to evaluate the income stabilizing role of futures and option
contracts for LDCs subject to high risk of the revenue they derive from
exportation of primary commodities. It showed how futures contracts may
constitute an effective hedging tool for producers. As an illustration, a
"pre-harvest" hedging strategy was proposed for cocoa, coffee, cotton and rubber
and proved quite effective. On the whole a routine hedge is not optimal and
countries should generally go short (some exceptions for cotton). Hedge ratios
and the maximum variance reduction degree vary according to the intrinsic
economic features of each country. Some authors think that the lack of
participation of LDCs in futures markets is due to variability in production. It
has been proved that this is not necessarily the case. Estimated net hedging gains
of the pre-harvest strategy are positive for the majority of commodities and
countries of the sample. This result is, however, only indicative since margin calls
were not considered.
Finally an evaluation of a joint hedge on futures and options markets was
made under the risk-neutrality hypothesis. Optimal positions on both markets
were derived. The effectiveness of the joint hedge depends on the correlation
between the value of the option and the futures price of the underlying
commodity. It was shown that the joint hedge improves the effectiveness of
hedging on one market for at least its effectiveness is identical to that on a single
market. However, does the marginal gain of the increase in risk-reduction cover
the additional cost due to the joint hedge? How do transaction costs modify the
optimal positions especially when the producer is rationed on credit markets? It
is also interesting to assess the long run effect of risk reduction on production and
on the equilibrium spot price.

References
Anderson, R Wand J-P Danthine (1981), "Cross Hedging", Journal of Political
Economy, 89, 1182-96.
Bigman, D, D Goldfarb and E Schlechtman (1983), "Futures Market Efficiency
and the Time Content of the Information Sets", The Journal of Futures
Markets, 3, 321-34.
Black, F (1976), "The Pricing of Commodity Contracts", Journal of Financial
Economics, 3, 167-79.
Black, F and M Scholes (1973), "The Pricing of Options and Corporate
Liabilities", Journal of Political Economy, 81, 637-54.
Breeden, D T (1984), "Futures Markets and Commodity Options: Hedging and
 49

Optimality in Incomplete Markets", Journal of Economic Theory, 32,

275-300.
Danthine, J-P (1978), "Information, Futures Prices and Stabilizing Speculation:,
Journal of Economic Theory, 17,79-98.
Dusak, K (1973), "Futures Trading and Investor Returns: An Investigation of
Commodity Market Risk Premiums", Journal of Political Economy, 81,
1387-1405.
Ederington, L H (1979), "The Hedging Performance of the New Futures
Markets", The Journal of Finance, 34, 164-70.
Gemmill, G (1980), "Commodity Stabilization: The Case for Establishing Inter-
Governmental Futures-Trading in Distant Contracts", Commodity Paper,
No.2.
Hieronymus, T A (1970), Economics of Futures Trading for Commercial and
Personal Profit, Commodity Research Bureau, New York.
IBRD (1986), Rapport sur Ie developpement dans Ie monde, IBRD, Washington
D.C.
Johnson, L L ~1960), "The Theory of Hedging and Speculation in Commodity
Futures' , Review of Economic Studies, 74, 139-5l.
Kofi, T A (1973), "A Framework for Comparing the Efficiency of Futures
Markets", American Journal of Agricultural Economics, 55, 584-94.
McKinnon, R I (1967), "Futures Markets, Buffer Stocks, and Income Instability
from Primary Producers", Journal of Political Economy, 75, 844-6l.
Nguyen, D T (1980), "Partial Price Stabilization and Export Earning Instability"
Oxford Economic Papers, 32, 340-52.
Ntamatungiro, J (1985), "On Using Futures Markets by Commodity Exporting
LDCs: The Case of the Ivory Coast". Paper presented at the Conference
organized by the Columbia Futures Center (CSFM) and the International
Futures and Commodity Institute (IFCI), Geneva, June 1985.
Ntamatungiro, J (1988), Stabilisation des recettes d'exportation: stock regulateur,
contrats a terme et options, Economica, Paris.
Priovolos, T (1981), Coffee and the Ivory Coast, D.C. Heath and Co., Lexington
MA.
Rolfo, J (1980), "Optimal Hedging under Price and Quantity Uncertainty: The
Case of a Cocoa Producer", Journal of Political Economy,88, 100-16.
Tomek, W G and R W Gray (1970), "Temporal Relationships Among Prices on
Commodity Futures Markets: Their Allocative and Stabilizing Roles",
American Journal of Agricultural Economics, 52, 372-80.
UNCTAD (1983), Commodity Exchanges and Their Impact on the Trade of
Developing Countries, Report from the UNCTAD secretariat, TD/B/C.
1/248/Crr. 1, 22 August 1983.
Wolf, A (1983), "An Agent Hedging with Commodity Options: Optimal Choice
of Options, and Portfolio Analysis", working paper series, #CSFM-70.
Working, H (1953), "Futures Trading and Hedging", American Economic
Review, 43, 315-43.
50

APPENDICES

Appendix A.I: Estimation of Hedge Ratios and Effectiveness of the

Pre-harvest Hedging Strategy for Cocoa, Coffee, Cotton and Rubber

Table2a. Estimation Results for Cocoa (1968-1985)

March May

Country 8 D-W 8 D-W

~ J1.f ~ J1.f

Brazil Tl 8,967 71% 63% 2.198 9,141 72%+ 61%+ 2.217

y=12,651 T2 8,863 70% 87% 1.780 8,i47 64% 74% 1.691

Cameroon Tl 2,186 29% 49% 1.845 2,222 29% 52% 1.417

y=7,538 T2 1,834 24% 52% 1.494 1,944 26% 57% 1.523

Iv-Coast Tl 6,756 29%' 11% 0.814 6,703 29%' 10% 0.848

y=23,364 T2 5,999 26% 11% 0.854 4,317 18%' 6% 0.766

Ghana Tl 8,982 34% 38% 2.484 8,864 33% 36% 2.503

y=26,582 T2 4,970 19%* 27% 2.428 4,307 16%- 22% 2.381

Nigeria TI 9,081 51% 49% 1.886 9,013 50% 43% 1.827

y=17,964 T2 3,731 21%* 23% 1.609 4,040 22%* 24% 1.546

Togo TI 1,042 59% 41% 1.628 1,040 59% 37% 1.742

y=1,765 T2 707 40% 25% 1,522 666 38% 22% 1.487

8 is the hedge ratio ~/y where y is the mean of exports during the period and J1.f the
hedging effectiveness.
 51

Table2b. Estimation Results for Coffee (1968-1985)

March/July May /September

Country ~ 8 iLf D-W ~ 8 iLf D-W

Brazilt Tl 14,866 17%* 27%0 1.910 14,484 17%* 27%0 1.908

y=87,167 T2 25,574 29% 32% 1.950 21,493 24% 32% 1.964

Burundit Tl 968 42% 45% 1.559 848 37% 44% 1.541

y=2,318 T2 1,282 55% 46% 1.895 1,178 51% 51% 1.934

Colombia Tl 28,405 60% 58% 1.608 27,934 59% 54% 1.549

y=47,699 T2 11,360 24% 32% 1.413 27,473 58% 56% 1.457

Iv-Coast Tl 13,949 61% 58% 1.512 14,162 62% 58% 1.521

y=22,923 T2 7,877 34% 65% 1.689 16,233 71% 81% 1.614

Kenya Tl 8,376 109%+ 59% 1.789 8,487 110%+ 62% 1.763

y=7,681 T2 4,312 56% 74% 1.892 9,421 123%+ 66% 1.878

Uganda Tl 7,583 47% 50% 2.045 7,820 48% 53% 2.043

y=16,246 T2 4,462 27% 75% 2.016 8,543 53% 60% 2.207

Rwandat Tl 745 32% 33% 1.692 720 31% 31% 1.668

y=2,301 T2 1,124 49% 38% 1.461 1,063 46% 49% 1.513

For these countries t was fixed at the beginning of April; delivery months are July
and September

Table2c. Estimation Results for Rubber (1973-1985

May July
Country
~ 8 iLf D-W ~ 8 iLf D-W

Indonesia Tl 26,215 30%* 36%0 1.677 22,826 26%* 32% 1.730

y=88,D43 T2 16,722 19%* 19% 1.789

Liberia Tl 3,081 40%* 53% 1.570 2,498 32%* 48% 1.635

y=7,729 T2 1,458 19%* 31%0 1.689

Malaya Tl 97,763 63%+ 61% 1.40 82,122 53% 53% 1.440

y=156,074 T2 73,384 47%* 46% 1.446

Thailand Tl 17,426 37%* 50% 1.405 10,738 23%* 41% 1.538

y=47,214 T2 7,996 17%* 35% 1.538

Zaire Tl 1,887 94%+ 63% 1.532 1,388 69%+53% 1.897

y=2,009 T2 1,000 50%* 35% 1.861
52

Table 2d. Estimation Results for Cotton (1968-1985)

March/July July
Country
~ 0 /-If D-W ~ 0 /-If D-W

Brazil Tl -11,766 84% 51% 1.676 -11,790 85% 52% 1.678

y=13,938 T2 -10,999 79% 44% 1.563 -11,122 80% 41% 1.506

Egypt Tl 9,207 44%* 38% 1.384 9,019 43%* 37% 1.39

y=21,056 T2 8,618 41%* 37% 1.572 5,947 28%* 31% 1.612

India Tl -3,055 67%* 11%0 1.918 -3,,088 68%* 11%° 1.702

y=4,550 T2 -1,967 43%* 4%°1.570 -1,990 44%* 4%° 1.506

Sudan Tl 6,091 36%* 24%0 1.540 6,490 38%* 24%0 1.543

y=16,923 T2 6,907 41%* 24%0 1.643 19,210 114%+ 45% 1.532

Chad Tl 1,467 35%* 34% 1.643 1,561 37%* 35% 1.658

y=4,199 T2 1,533 37%* 37% 1.580 2,339 56%+ 46% 1.571

We cannot reject the hypothesis that 0=1 at a 5% level

We cannot reject the hypothesis that 0=0 at a 5% level
We cannot reject the hypothesis that /-IFO at a 5% level

Note: y and ~ are in 10,000 MT

Appendix A.2: Proof of Proposition 2

We will prove proposition 2 for n=1 by replacing P, Y and F~ respectively by x,

y and z. Variable y is independent of x and z iff f(x,y,z) = f(x,z) of(y) where fO is
the joint density function of x, y and z in a continuous context
(P(x,y,z) = P(x,z)P(y) in the discrete case).
We have to prove that Cov(xy,z) = E(y)Cov(x,z).

Cov(xy,z) = I xyz[xy - E(xy)][z - E(z)]f(x,z)f(y)dxdydz

= Iy yU xz x(z - E(z))f(x,z)dxdz]f(y)dy

- E(xy)1 y [f xz (z - E(z))f(x,z)dxdz]f(y)dy

with Ixz (z -E(z))f(x,z)dxdz = 0.

 53

Cov(xy,z) =1 y y[l xz (x - E(x) + E(x))(z - E(z))f(x,z)dxdz]f(y)dy

= 1y y[l xz (x - E(x))(z - E(z))f(x,z)dxdz

+ E(x)f xz (z - E(z))f(x,z)dxdz]f(y)dy

with 1xz (z - E(z))£(x,z)dxdz = O.

Then
Cov(xy,z) = 1y y[Cov(x,z)]f(y)dy

= Cov(x,z)/ y yf(y)dy

= Cov(x,z )E(y)

and finally Cov(xy,z) = E(y)Cov(x,z) for y independent of x and z. So

d = Cov(PE(Y), F~) - Cov(PY, F~) = 0 for Y independent of P and F~ .•

Appendix A.3

The value of the put in t is given by e-rTMax(O, K-F~). Its expected value is
W = e-rTE[Max(O, K-F~)] or
K
W = e-rTI f(F~)(K-F~)dF~.
o
where f(.) is the density function of futures prices. When F~ N A(/-t,s) then we
have

By changing the variable F ~ in x, we obtain

54

InF~ - E(1nF~)
x = ----::s,------
with F~ = exp(sx + InE(F~) _s2/2) for InE(F~) = E(lnF~) + s2/2.

112
dF T = s exp(sx + InE(F T) - s /2)dx

The borns of the integral also change:

K -+ [InK -lnE(F~) + s2/2l/s = -b 2
0-+-00

-rT -b 2 1 2 -rT 1 -b 2-8 1 2

= e KJ - - exp -(¥ )dx - e E(F T)J - - exp-(!( )d(
-ro ..j[2i) -ro ..j[2i)

where ( = x - s, d( = dx)

W=e
-rT[
KJ
-b2 1 I 2 1 -b2-8 1
--exp-(¥ )dx-E(FT)J --exp-(!~ )de
..2]
-ro ..j[2i) -ro ..j[2i)

and finally

W = e-rT [KN(-b 2) -E(F~)N(-bd

with
 55

Appendix A.4: Proof of Proposition 3

When production is non-random we have Cov(Rc'w) = yCov(P,w). The non-
existence of basis risk assures that Cov(F~,w) = Cov(P,w) and V(F~) = V(P).
So the hedge ratio is
0* Cov(P, w) Cov(P,w) 1 1
_c=_ ~ _ _ _ = ___ = __
Y V(FJ) =- V(P) Bw/aP WI 8
where -1~8~O.·
Then the pure hedge position in discrete trading is a good approximation
of Black's continuous hedging strategy.

Appendix A.5: The effectiveness of the joint hedging strategy

(Proof of Propositions 4 and 5)
The variance is equal to
V(R) = u' Vu

1] V(Rc) Cov(Rc' w)
with u = [-~ and V = Cov(Rc,F~ Cov(w,F T1 )

Cov(Rc'w) V(w)

Cov(Rc'w)
(1-,2 )V(w)
56

2 1 2 2
1 [COY (R ,FT ) COY (Re,w),
V(R) - V(R ) = e + _ _""""'"----_
e ~()
1-1 VF( T1) Vw( )

2 1 2
COY (R ,F T ),
+ c
V(Fi)

2
2 COY (R ,w)
2(1-1 ) e
V(w)

Cov(R ,w )Cov (R ,FT1),4 COY 2 (Re' W),2]

+2 e c -2--~-
COV(W,Fi) V(w)

Arranging the terms, we obtain:

1 [COV 2 (R ,FT1) Cov 2 (R ,w)

V(R ) - V(R) - c + c
e - ~(
1-1 ) VF (1)
T Vw ( )
 57

Then

(14)

2 1 2 1
COY (Rc,F T ) COY (Rc'w) Cov(RC ,F T)Cov(RC'w)'y
+ -2
V(F~) V(w) j [V(F~)V(w)l

The magnitude of j.£fo depends on values taken by the correlation coefficient 'Y-
The minimum 9 of j.£fo is given by the solution of equation (15):

-2B·P + 2A'Y- 2B 2
2 2 =0 when B'Y - A'Y + B =0 (15)
(1-'Y ) V(Rc)
with

COV2(RC,F~) Cov 2 (Rc 'w)

A= + ud
V(F~) V(w)

Equation (15) has two real roots:

ud

9 V is a minimum for it may be shown that d2j.£fo( 'Y)/d'Y2 > O.

58

As 'Y1 = [-'Y21-1, and that I'Y19, then /Lfo reaches its minimum at r with

By substituting 'Y by r in (14), we find

and

As /Lfo has a positive minimum (both /Lf and /Lo are positive), then /Lfo is always
positive. It is also clear that /Lo' /Lf $ /Lfo.·
 CHAPTER 3

OPTIONS TO ALLEVIATE THE COSTS OF UNCERTAINTY

AND INSTABILITY: A CASE STUDY OF ZAMBIA

Andrew Powell

1. INTRODUCTION
The motivation for this Chapter stems from two observations. Firstly, the
majority of the theoretical and empirical literature on the effect of volatile
commodity prices on developing countries does not separate the notion of the
instability of those prices from the notion of uncertainty attached to the prices of
primary goods!. Secondly, although there has been a certain amount of work on
the role of futures markets to reduce uncertainty or instability there has been
little work on the use of option contracts in this fashion 2.
Two examples illustrate that uncertainty and instability are logically
distinct in this context. First of all, imagine that it is known with a high degree
of certainty that the price of a certain commodity will rise in the next year but
then fall back to a longer term mean. Secondly, consider a commodity where the
expected price for next year is this year's price but that there is considerable
uncertainty attached to this expectation. Ex post, it may turn out that the
instability of prices (measured by the standard deviation around the mean) over
the period is high for both commodities even though there may be little
uncertainty attached to the price history of the first commodity. Further, a high

! See for instance the review by Behrman (1987). Newbery and Stiglitz
(1981) does contain a discussion, see Chapter 20.
2 On futures see for instance McKinnon (1967), Newbery and Stiglitz (1981),
Gemmill (1985) and Gilbert (1985). On futures and options see Ntamatungiro
(1990) in this volume.
59
L. Phlips (ed.). Commodity, Futures and Financial Markets. 59-84.
© 1991 Kluwer Academic Publishers.
60

degree of instability can only be assured with some confidence for the second
commodity ex post if there is a large sample. In a small sample it could turn out
that the ex post instability is rather low and thus the degree of uncertainty
attached to each of the commodities is actually the reverse of the magnitude of
instability.
In general, results on the costs of instability have been mixed to say the
least and many authors have found the costs to be rather small (see Newbery and
Stiglitz, 1981 - henceforth NS - Table 20.8, page 296 and also see Kanbur 1984
reviewing NS). In these analyses the conventional measure of instability of the
price series is the standard deviation (or coefficient of variation) around the mean
of the historical series. Either, instability is considered as a cost per se or this
measure of instability is employed as a proxy for uncertainty. In section 3 of this
Chapter I employ a different measure of uncertainty exploiting historical
forecasts of actual commodity prices. The estimates suggest that in the case of
copper the use of instability underestimates the uncertainty attached to copper
prices.
The costs of uncertainty and instability may, however, be strongly
associated. Both Kanbur (1984) and NS suggest that there may be additional
macroeconomic effects not captured by a microeconomic orientation employed in
these calculations. In particular, and pertinent to the subject of this publication,
the relationship between commodity markets and credit markets is crucial. If
credit markets are perfect there may be no reason to expect high costs to
instability. NS and Kanbur (1984) both present models in which there is an
assumed foreign exchange constraint and conclude that instability may then lead
to much more significant costs. Note that a common justification for
imperfections in credit markets is uncertainty leading to the possibility of
default.
Another potential cost of uncertainty centres around policy responses and
in particular in relation to a commodity 'boom' or 'bust'3. An important
distinction in these 'trade shock' models is between a 'boom' or 'bust' that is
perceived as temporary versus one which is perceived as permanent. If the boom
is temporary and markets operate smoothly then rational agents would only
marginally adjust consumption and investment programs commensurate with the

3 See Corden (1984), Neary and van Wijnbergen (1986), Bevan et al (1988)
and Bevan et al (1989).
 61

change in permanent income. However, in practice there may be great

uncertainty attached to the prediction of the extent and of the period of the
boom or bust. As discussed in section 2 below there may have been considerable
costs in the case of Zambia due to this type of uncertainty.
At the same time there has been considerable practical difficulty in
stabilising commodity prices through the medium of international interventions 4•
Many difficulties stem from the disagreements over the objectives of the schemes
between producers and consumers and also between producers who may wish
different degrees of stabilisation. This suggests the use of techniques that allow
the design of programmes for individual countries and there' has been growing
interest in the use of futures and options markets for this purpose. In section 4 of
this Chapter I consider a simulation option stabilisation programme for Zambia.
Thus, the organisation of the Chapter is as follows. In section 2, I present a
comparison of the actual copper price versus World Bank predictions of the
copper price over the period 1964-1986 and briefly consider the costs to Zambia
of copper price instability and uncertainty. In section 3 I develop a simple
theoretical model to distinguish between instability and uncertainty and use this
to estimate the benefits to Zambia of the stabilisation of the copper price. In
section 4, I consider the choice between futures and options and then using a
simulation exercise illustrate the characteristics of returns that would have
accrued to Zambia had she followed an options type stabilisation program over
the period 1980-84 using daily price data over the period 1976-1985. Finally, in
section 5, I present some conclusions to the analysis together with suggestions for
further research.

2. THE COPPER PRICE AND ZAMBIA 1964-1984

Figure 1 and Figure 2 illustrate nominal and real copper prices over the period of
interest. Independence in Zambia in 1964 was accompanied by strong and rising
copper prices in both nominal and real terms. Although the price remained
volatile the price also remained high for roughly 10 years until the crash in 1975.

4 See Gilbert (1987) on UNCTAD's International Programme for Commodi-

ties (IPC) and Brown (1980) for a discussion of the political and practical
difficulties of commodity control. On recent failures see Anderson and Gilbert
(1988) on the case of tin. The failure to reach a new coffee agreement (see FT
12/1/90, 'coffee prices hit 14 year lows') leaves only the cocoa and rubber
agreements in tact.
 62

2.2
2.1
2.0
1.9
1.8
1.7
1.5
1.5
-;;-

·
tl 1.4
c
,
0

0
1.3
.<: 1.2
C 1.1
1.0
0.9
0.8
0.7
0.5
0.5
0.4
1950 1955 1960 1955 1970 1975 1980

Figure 1. Copper prices 1950-1984 (nominal $ price/tonne)

5.0

4.5

4.0

·,
3.5
~

·
tl
C
0
3.0
0
.<:
C
2.5

2.0

1.5

1.0
1950 1955 1950 1965 1970 1975 1980

Figure 2. Copper prices 1950-1984 (1983 $ price/tonne)

 63

Examining data over the whole century, Takeuchi et al (1987) conclude

that the period 1964-1974 was a period of high average real copper prices or a
period of a copper price 'boom'. In real terms the crash of 1975 was not reversed
during the period although in nominal terms the rises in 1979 and 1980 did
reverse the fall in 1975.
A full account of the political economy of Zambia over this period is well
beyond the scope of this Chapter and it is assumed that the reader has at least a
passing knowledge of the Zambian cases. However it is important to bring out a
number of aspects. First of all, in the period preceding independence there was a
general perception of neglect in investment in the area kn()wn as Northern
Rhodesia with the copper revenues largely leaving the region. Fry (1980) reports
up to 1/6 of GDP in 1954 falling to 1/8 of GDP in 1963 remitted out of the area.
Secondly, the new Government was highly dependent on mining for revenues at
the time of independence with mining accounting for 53% of Government revenue
in 1964 rising to 64% in 1966 6. The first years of independence were ones of
optimism with rising copper prices that were predicted to rise further.
It is therefore not surprising that the Government embarked on an
ambitious First National Development Plan (FNDP 1966) that contained many
social investment programs and gave a significant role to Government. Indeed
the Government took increasing control of mining, industry and agriculture for
reasons that in part stemmed from ideology but also for pragmatic reasons. This
strategy was detailed in both the FNDP and the Second National Development
Plan (SNDP 1972).
Bell (1981) argues that these rather rigid development plans provoked a
type of ratchet mechanism in Government expenditures. Government
expenditure rose quickly in both nominal and real terms during times of rising
copper prices but were not reduced sufficiently quickly during falls in the copper
price to stop deficits occurring. Indeed deficits did occur in 1971 to 1973

S There are many discussions on the J>0litical economy of Zambia over the
period. The excellent readings in Tordoff (1980) provide a good introduction and
Kydd (1988) discusses recent economic aspects very succinctly. Gulhati (1987)
gives an interesting political economy account and Burdette (1988) a more
comprehensive recent review. Jaycox et al (1986) and Ndulo and Sakala (1987)
are recommended reading on Zambia'S debt position.
6 Contribution of Mining to Government revenue quoted in Obidegwu and
Nzaramasanga (1981) Chapter 1.
64

inclusive but this was a period when Zambia had good access to international
credit and indeed Zambia borrowed heavily from commercial creditors 7. The
crash in copper prices in 1975 was a watershed for the Zambian economy. From
then on Government and trade deficits became the norm, Zambia was in constant
arrears with her external creditors and the economy was rarely out of a state of
near crisis. Bell (1981) maintains that despite the Government acknowledging
the dependence on copper very little attempt was made to devise means of
lessening the effect of volatile copper prices on the economy and in particular on
Government expenditure (Bell 1981, p22/p23).
Furthermore the FNDP and SNDP included strikihg errors in copper price
predictions. In 1966 the FNDP under-estimated the 1970 real copper price by
22% and in the SNDP the 1976 real copper price was over-estimated by 50%8! In
summary the 'boom' of 1964 to 1974 was not predicted accurately either in terms
of the extent of the copper price rises nor the dramatic fall in copper prices in
1975 that signaled its conclusion.
These types of mis-prediction errors could have quite asymmetrical costs.
Given a ratchet mechanism in expenditures and credit rationing, over-estimation
errors may be much more serious than under-prediction errors. This is a strong
argument for the use of a simple insurance type option strategy rather than a
simple futures strategy. As discussed later a simple put option strategy gives
price protection during low price periods whereas a simple futures strategy
reduces the variance of returns. Furthermore the costs of uncertainty are likely
to be greater given this type of ratchet mechanism in expenditures compared to a
case where expenditures are more flexible. An increase in uncertainty would be
likely to decrease investment further than would otherwise be expected.
An interesting feature of the Zambian case was that borrowing, especially
from commercial banks, was largely pro-cyclical with nominal copper prices.
Zambia borrowed heavily in the period 1970-1975 when nominal copper prices,
although volatile, were high. During the copper price crash of 1976 through 1977
the change in Zambian debt outstanding was small and commercial banks acted

7 Zambia borrowed $240m from commercials (and a further $300m from

bilaterals and multilaterals) over the period 1971 to 1973. World Bank Debt
Tables (various issues).
8 Authors own estimates from data from the FNDP (1966) and SNDP (1972)
and World Bank copper price data from World Bank (1975/1988).
 65

to reduce their exposures. During the nominal copper price rises of 1978-1980
Zambia borrowed significantly again and although commercials increased their
exposure, bilateral lending accounted for the majority of the positive change in
debt outstanding. In the period 1981-1986 commercials once again reduced their
exposure although both bilaterals and the multilateral agencies lent significantly9.
Two arguments follow from these 'stylized' facts that deserve further
attention. Firstly commercial banks may have acted in a myopic fashion simply
lending at times of high nominal copper prices when Zambia appeared a 'good'
credit risk. Secondly, this suggests a model of lending that is supply rather than
demand driven. Zambia borrowed heavily at times of rising copper prices
(1970-1975 and 1978-1980) and was clearly rationed during the period
1976-1977. The theoretical debt literature can explain both 'over lending' and
credit rationing in models that are labelled either 'ability' or 'willingness' to pay.
However the above suggests a third label namely, 'ability to borrow'10. It is
interesting to speculate if this pattern corresponds to that of Latin American
'over lending' in the late 1970's and rationing in the post 1982 era.
The Zambian Government was not alone in making poor predictions of
future copper prices. In the next section of this Chapter I employ World Bank
predictions to estimate the uncertainty inherent in future copper prices. A
selection of these figures are graphed in Figures 3 and 4 and the figures
themselves are detailed in an Appendix. Figure 3 plots the actual nominal
copper price and forecasts made in a number of years. Figure 4 plots the real
copper price and forecasts in similar fashion. The nominal forecasts are quite
dramatically different from the actual although the real forecasts are perhaps the
more revealing. Clearly, the view taken by the World Bank was that the crash of
1975 was largely temporary in real terms and that prices would recover to near

9 The change in debt outstanding to commercial creditors was $490m in

1970-1975, $--8.7m in 1976-1977, $140m in 1978-1980 and $-180m in 1981-1986
inclusive. From bilaterals and multilaterals, Zambia increased its borrowing by
$lbn in 1970-1975 and $1.2bn in 1978-1980 only gaining access to a further
$210m in the period 1976-1977. Figures from World Bank Debt Tables (various
issues ).
10 See Kletzer (1984) for a model in which overlending can occur and Eaton
et al (1986) for a review of the debt literature and examples of models with credit
rationing. It should be recognised that the categories labelled here are not
necessarily exclusive and I am indebted to Christopher Gilbert for suggesting the
third, alternative label.
66

210
200
190
180
170
160
150
140
130
120
£l
'-
.'!l
110
c 100
()
90
80
70
60
50
40
30
20
10
0
19731974197519761977 1978 1979 1980 1981198219831984198519861987198819891990

o Act 75 o 76 77 82 84

Figure 3. Nominal copper prices (actual and World Bank predictions)

200
190
180
170
160
150
140
130
a 120
2 110
100

"'"~ 90
80
70
60
50
40
30
20
10
0
19731974197519761977 1978 1979 1980 19811982 198319841985 198619871988 1989 1990

o Act + 75 0 76 /::,. 77 x 82 'f:J 84

Figure 4. Real copper prices (actual and World Bank predictions)

 67

pre-1975 levels. There was only a slow learning process that this was not going
to occur and that real copper prices would fall even further than the 1975 and
even the 1978 level l1.
I do not wish to suggest that the entire set of ills that amicted the Zambian
economy over this period stemmed from the development of copper prices and the
poor copper price prediction record. Many other policy mistakes were made. Fry
(1980) discusses poor planning in agriculture and Jaycox et al (1986) argues that
industrialization was inward looking and highly protected and that investments
were often poorly conceived and inefficiently managed. Gulhati (1989) argues
that after the copper price falls Zambia cut back on investment;and development
expenditures while consumption actually rose in real terms. Indeed, Jaycox et al
(1986) suggests that there was a universal inability (or unwillingness) to cut back
consumption levels reached in high commodity price periods throughout Eastern
and Southern Africa. In other words they argue that the ratchet mechanism
worked on consumption rather than investment type expenditures. However,
Jaycox et al (1986) also admits that the copper price decline was not predicted
by the Zambian Government nor the multinational institutions and the graphs
above indicate that the crash was also predicted as largely temporary. It remains
an open question to what extent many of the policy mistakes made were due to
the uncertainty of future copper prices and the disappointing prediction record.

3. INSTABILITY AND UNCERTAINTY

In this section I wish to present a method of distinguishing between commodity
price instability and uncertainty. Using a theoretical approach developed by
Epstein (1988), I develop a simple way of estimating the value of commodity
price stabilisation that includes both the benefit of reducing uncertainty and the
benefit of reducing instability. The results suggest that conventional measures of
instability may be poor proxies for uncertainty and therefore conventional
methods may seriously underestimate the value of commodity price stabilisation.
As a first step consider the World Bank copper prices and forecasts graphed
above 12. A conventional measure of the instability of the actual copper price is

11 The source of these figures is the Price Prospects for Major Commodities,
Report Number 814, World Bank (1975), (1976), (1977), (1982), (1984) and
(1986).
12 The actual data used in the analysis is detailed in an Appendix.
68

the standard deviation of the series about the mean. A measure of uncertainty,
however, is the standard deviation of the actual series about the forecasts. The
final row of Table 1 details the standard deviation and coefficient of variation of
annual averages of the actual copper price as measured conventionally. The
other rows of the Table present standard deviations and coefficients of variation
of the copper price around the forecasts. The first four rows consider the
uncertainty attached to 1,2,3 and Long Term (forecasts greater than 3 years)
respectively. The fifth row uses all of the forecasts available.
Compare these measures of uncertainty with the conventional measure of
instability at the foot of the Table. If one just takes J. year forecasts then the
forecast standard deviation is less in magnitude than the actual instability of the
series for both nominal and real prices. However, if one considers forecast
standard deviations over two years or more or taking all of the forecasts available
and computing the forecast standard deviation then the forecast standard
deviations are significantly greater than the standard deviation of the actual
copper price series. Similar results follow by considering the relevant coefficients
of variation.

Table 1. Analysis of World Bank Copper Price Forecasts

Nominal Prices Real Prices

Std CVN Std CVN
c/lb % index %
1 Year 5.3 7.5 6.6 7.2
2 Years 18.0 25.6 15.1 16.5
3 Years 29.5 41.9 21.0 22.9
Long Term 109.0 154.7 34.7 37.9
All 34.2 48.6 28.2 30.8
Actual 12.4 17.6 14.0 15.3

Note that these calculations do not necessarily represent the most natural
method of judging whether these forecasts are 'good' or not. In the above the
standard deviation and coefficient of variation are taken with respect to the
expected level of the future copper price. For short term forecasts it may be
more natural to consider the standard deviation and coefficient of variation of
actual changes in the copper price with respect to the forecast changes in the
copper price. For the purposes of this study I am assuming that what is
important to a government planner is the certainty of the future level of income
 69

(dependent on the level of the copper price) and not certainty attached to
changes in future income (dependent on changes in the copper price). In brief the
analysis shows that the conventionally measured instability of the series over-
states the uncertainty in one year predictions of the copper price but under-states
uncertainty with respect to predictions of greater than two years. Considering all
of the predictions employed the instability of the actual series significantly under-
states the uncertainty attached to the forecasts.
Newbery and Stiglitz (1981) propose a simple method for analysing the
benefits to price stabilisation. If y is a risky income stream and q is, say, a
rather less risky income stream then the benefit to a risk; averse agent of
swapping from y into the less risky stream, q, can be measured by B, where B is
defined by the following expression,
E[U( q-B)J = E[(U(y)J (1)
where E[·J is the expectations operator and U(·) is a concave utility function.
However, the usual expected utility framework does not distinguish between
instability and uncertainty in the sense that in a multi-period world an unstable
but purely deterministic income path and an uncertain income path with exactly
the same measure of dispersion would result in a similar utility. If the coefficient
of risk aversion is extremely high then the elasticity of inter-temporal
substitution must be correspondingly low. Furthermore, the framework cannot
distinguish between the case of low uncertainty, high instability and high
uncertainty, low instabili ty as discussed in the introduction.
Epstein (1988) develops a method that separates out these aspects. To
motivate the approach consider the following recursive utility formulation,
Vt = W(c t ,E[Vt+l]) (2)
Where ct is known consumption in period t and E[.J is the expectations operator.
The function W(.) is termed the aggregator function. Now define JL(.) as a
certainty equivalent function and consider the case when the aggregator has the
Constant Elasticity of Substitution (CES) form. Then 2 becomes,

V~ = (c~ + ,B/L[Vt+1 JP ) (3)

where ,B is a time preference parameter (O~,B~I). A simple certainty equivalent
function to make the above operational is the Constant Relative Risk Aversion
(CRRA) form as follows:
/L[Vt+1 J = [E t (Vt+l)l-ap/(l-a) O<a<1 (4)
70

tt[Vt+l 1= exp {Etln(Vt+l)} a=1 (5)

The above specification, (4) and (5), has two parameters, p and a, and the
separation of the cost of uncertainty from the cost of instability in the above is
reflected by the effect of a and p respectively. For instance consider a
deterministic consumption program. Then the above reduces to an inter-temporal
CES utility function,

V~ = Ei =0 tfc~+i ' (6)

where p, determines the elasticity of substitution, Es = (l_p)-I. Thus p reflects
the cost of an unstable but known consumption progralll relative to one that is
stable and known. When p = l-a<l, then (4) reduces to the common homo-
thetic expected utility specification,

V~ = Et(E~+i f1tC~ ) (7)

Finally consider the case when p .... O. At P = 0, the CES form reduces to the
simpler Cobb-Douglas formulation (see for instance Heathfield, 1971) and we can
write,
v = c [E (V )1-aJ f1/(1-a) a<1 (8)
t t t t+l

InV t = lnc t + f1EtlnVt+l a=1 (9)

Now consider the value of price stabilisation and the reduction of price
uncertainty. Let the vector y, y = (y t' Yt+ l' Yy+2 ... ), represent one risky
income stream and the vector q, q = (qt' qt+l' qy+2 ... ), represent a second
risky income stream. A multi-period version of the Newbery-Stiglitz approach is
to let B equal the monetary value at time t of substituting y by q and thus B
may be defined from the following equation,
Vt(y) = Vt(qt-B , qt+l' qy+2 ... ) (10)
A technique to obtain an expression for B is to expand both the left hand side
and right hand side of (10) by Taylor expansions. The CES formulation is rather
difficult to manipulate in an n period context. As a simplification consider the
Cobb-Douglas formulation over n periods and for the case where a = 1. Thus the
coefficient of relative risk aversion is assumed to be unity (as in Newbery and
Stiglitz, 1981, Table 20.8, p296). Note that the Cobb-Douglas form imposes an
elasticity of substitution of unity which may be rather high given recent
estimates in the literature and acts to predicate the results of the analysis against
 71

the 'instability matters' point of view l3 . In the n period case equation (9) then
becomes,

InVt = Et};f=otlnCt+i (11)

and an expression for B/qt can be derived using the Taylor expansion method as
follows,

(12)

where N is the coefficient of variation of Yt+' with a simllar definition for

~+i 1
N The log linearity of the Cobb-Douglas form implies independence
qt+i
through time and thus there are no covariance terms between income in different
periods. The multiplicative term in the expected values of future income can be
interpreted as a 'transfer' effect and the exponential term can be thought of as a
'risk' effect. The transfer effect is related to benefits in the 1st moment and the
risk effect, in this second order Taylor expansion, to effects in the second
moment. It is important to stress that the coefficients of variation in the above
formulation refer to measures of the uncertainty in future income and not to the
instability of income.
Consider the case of Zambian copper revenues. Abstracting from quantity
uncertainty, equate the coefficient of variation of revenue uncertainty with that
of price uncertainty (this is a simplification of the Newbery and Stiglitz 1981
method who did take quantity uncertainty into account employing an assumption
on the elasticity of demand). Now estimates of the percentage benefits to
uncertainty reduction can be calculated using equation (12) and employing the
estimates of the coefficient of variations of the forecast errors detailed in Table 1.
In the first instance consider only the benefit to uncertainty reduction (such that
the actual income stream is unaffected) and consider a complete reduction in
uncertainty (from the estimates given in Table 1 to the case of complete
certainty). Table 2 presents the results for different values of the time preference
parameter fl.

13 See Attanasio and Weber (1989) for a discussion and related analysis.
72

Table 2. Estimated benefit to complete uncertainty reduction

Using World Bank Prediction Errors

(3 Nominal Prices Real Prices
1.0 68.2% 10.8%
0.9 45.4% 6.14%
0.5 7.68% 1.75%
Expressed as a percentage of income in the first year of the analysis, 1975

The estimates depend, in a fairly sensitive fashion, on the value of the time
preference parameter used. As (3 approaches unity more weight is given to the
long term predictions which have a greater degree of uncertainty attached to
them and the estimates in real price terms are large indeed. Further, if planners
are concerned with nominal prices rather than real prices the costs of uncertainty
are even more dramatic.
Now I turn to estimates of the benefit of price stabilisation including both
uncertainty and instability reduction. Consider the case of complete stabilisation
from the actual real price series to the mean of the actual real price over the
years of the forecasts. Thus there are benefits to stabilisation from uncertainty
reduction and from instability reduction. Using the above estimates in real price
terms it turns out that for (3=1, the benefit in 1975 to the stabilisation of prices
over the period of the forecasts in 1975 would have been 28.9% of income in 1975.
The figure in 1976 is 14.85% of income in 1976 and in 1977, 14.76% of income in
year 1977.
These figures can be compared to estimates using the simplified Newbery
and Stiglitz (1981) method, employing the actual instability of the price series.
Using this technique an estimate of only 4.57% of the income in 1975 is obtained
for the complete stabilisation of prices to the mean using real prices. Hence,
conditional on the assumption that the time preference parameter is reasonably
high then, employing a conventional measure of instability, the usual expected
utility method may seriously underestimate the benefits of commodity price
stabilisation for a country like Zambia.
To summarise this section it is possible to separate the costs of instability
from the costs of uncertainty in a theoretical analysis. Relative to this approach,
given our assumptions on the coefficient of relative risk aversion and the
elasticity of inter-temporal substitution, the conventional expected utility
analysis may seriously underestimate the benefits of price stabilisation. Further,
 73

the results were obtained using a rather high elasticity of inter-temporal

substitution (employed to simplify the analysis) which would act to predicate the
results against an 'instability matters' point of view.

4. OPTIONS TO STABILISE OR INSURE COMMODITY EXPORT

EARNINGS
There have been a number of theoretical and empirical papers on the use of
commodity futures for the stabilisation of commodity export earnings but the
work on commodity options is much more limited. However, options type
contracts have a number of significant advantages for the developing country
producer. I wish to draw out out three pointsl4. First of all the maturity of liquid
traded futures contracts tends to be fairly short (in many commodities trading 9
months to 12 months out may force the hedger into rather illiquid contracts).
Although short maturity futures may allow a farmer to hedge a particular crop or
possibly such contracts can be 'rolled over' to hedge a further crop year in general
these contracts may not be of sufficient duration to hedge the price risk inherent
in say a mining project or a substantial investment in an agricultural product.
Secondly, the use of traded futures contracts requires good access to credit
in that initial and variation margin payments are required. Margin payments are
designed to control the 'credit' or 'performance' risk associated with the futures
contract and may be very large indeed especially if producers are hedging a
number of years production. Two important observations stem from this. First
of all, hedging may then reduce the uncertainty associated with the producer's
revenues but may actually increase the instability of the producer's cash-flow.
This increased variability may not be important given that the producer has
adequate access to credit at market interest rates. Secondly, a reasonably perfect
credit market is a pre-requisite for this type of insurance market to operate
smoothly. If, for some reason, the producer's credit standing is not good, default
is a serious possibility or there is the lack of effective collateral, then the
producer may not be able to hedge using traded futures contracts which in turn
may increase the risk associated with the producers income and hence further
affect the producer's credit rating. It seems reasonable to conjecture at the
present time that many developing country commodity producers are heavily

14 For a wider discussion on the merits of futures versus options contracts in

this context refer to Powell and Gilbert (1988).
74

credit constrained. Thus, traded futures contracts may have only limited
usefulness as a stabilization tool.
Finally, options give a much greater degree of flexibility than futures in
moulding return distributions to that preferred by the producer. The simplest
option strategy to hedge returns is that of purchasing a put option l5 . This
resembles a kind of insurance arrangement in that an up-front payment is made
for the guarantee of a minimum price for the production. If the spot price turns
out to be below the exercise price of the put then the producer exercises the
option obtaining the exercise price. If the spot price turns out to be greater than
the exercise price then the producer will let the option 'expire without exercising
the option and simply sell the production at the higher spot price. Thus the
exercise price is the minimum price that would be obtained for the amount of
production covered by the option. As argued above, if the costs of uncertainty
are asymmetric in that a negative price shock is much more costly than a positive
price shock is beneficial (due, say, to credit rationing), then the put purchase
strategy may be appropriate.
The disadvantage of this simple strategy is that an up-front premium is
required. This problem may be circumvented by the simultaneous sale of a call
option to finance the put option. If the put and the call option are written on
identical quantities but at different exercise prices such that the value of the put
and the value of the call are equal then the resultant combination option is
'zero-cost' with complete insurance at the exercise price of the put but with a cap
at the exercise price of the call This is often referred to as a Max-Min strategy.
The total cap implied by Max-Min is not always attractive to a producer
who may feel that there is a danger of losing out to competitors during a high
commodity price period. An alternative zero-cost option has been termed a
ZECRO 16. Here, the cost of the put is again offset by the sale of a call but the two
options have a common exercise price. However, the quantities on which the two
options are written will be adjusted such that their values match exactly. For
instance, if the common exercise price is below the current spot price (such

15 An excellent introduction to options is provided by Cox and Rubinstein

(1986).
16 ZECRO (Zero Cost Ratio Option) is a registered trademark and service-
mark of J Aron &; Company; J Aron is a member of the Goldman Sachs group.
 75

that an 'out of the money' put is financed by an 'in the money' call) then the call
option will be written on only a fraction of the quantity of the amount covered
by the put.
The Max-Min and ZECRO strategies represent two extremes and it follows
that there are a wide range of zero-cost options that fall between these two
strategies. These examples, however, give some measure of the potential
flexibility of option strategies and how return distributions can be moulded to fit
the preferences of a particular commodity producer.
Note, however that both of these strategies require writing calls as well as
the purchase of put options and thus if they were performed using traded
contracts margin payments on the call options would in general be required.
Typically, therefore, these strategies are performed directly with dealers (who
may offset their positions on exchanges) and thus if the dealer does not ask for
margin payments or other forms of collateral then the dealer faces a performance
risk on the call option component of the option. The lack of credit worthiness
coupled with the lack of effective collateral may limit the development in long
term contracts of this nature for a developing country producer. In what follows
I abstract from such performance risks but it is worth noting that this is a well
known type of externality affecting the functioning of insurance and credit
markets and that there may be a case for public intervention to promote such
contracts developing (see Anderson et al, 1989).
To illustrate the use of options contracts to reduce instability and
uncertainty, I consider the impact of a set of option schemes on Zambian export
revenues over the period October 1980 - January 1984 using daily price data over
the period 1976-1985. The first stage is to design and price options that might
have been available over this period. The second stage is to design a Zambian
hedging strategy given the set of these instruments. Assuming Zambia is small in
World markets allows the separation of these two stages!7. In this analysis I
abstract from the problem of currency risk. It is possible that Zambia would be

17 Although Zambia has an extremely high concentration of commodities in

exports, 92% in the period 1981-1983, it remains relatively small as a world
copper producer, 12.2% of world exports in the period 1981-1983 (figures from
World Bank (1986). The assumption of a lack of market power must also extend
to the weakness of Zambia's position in CIPEC (the copper producer's
organisation - Conseil Intergouvernemental des Pays Exportateurs de Cuivre) or
the weakness of that body in influencing the course of copper prices.
76

concerned with the stabilisation of Kwacha export earnings whereas I consider

the stabilisation of earnings in sterling. I leave this complication for further
study.
Following the arguments above, that zero cost put call strategies might be
the most attractive to LDC producers, I restrict attention to zero-cost, put-call
combinations. For ease of computation I consider only European option
contracts with three maturities 1,3, and 5 years. This corresponds to the pricing
of 1982 production, for example, in 1981, 1979 and in 1977 respectively. The
option exercise price is defined in two ways: (i) a constant exercise price set at a
specified percentage of the mean price over the entire period 1976-1985, and (ii)
an exercise price specified as a percentage of the current average monthly price.
All prices are expressed in pounds (sterling) per tonne and correspond to
the LME standard grade spot contract. Option prices are imputed using the
Black-Scholes (1973) formula l8 . I investigated three strategies for each maturity
defined in the following way,
(i) The put exercise price was defined at 80%, 90% and 100% of the
reference price (fixed or variable as discussed above) for the 1 year strategy,
100%, 110% and 120% for the three year strategy and 110%, 120% and 130% for
the three strategy.
(ii) For each put exercise price I found the highest call price that would
allow the put to be financed (to the nearest 10% of the reference price).
(iii) The call was then written on that proportion of the put tonnage that
would equate the Black-Scholes value of the put and the call exactly.
Each estimated put-call combination price series defines an implicit price
series that would have been obtained if one tonne of copper had been hedged
using that strategy. Table 3 and Table 4 contrast the distributional
characteristics of these implicit series with the spot price distribution for the
fixed exercise price and variable exercise price strategy respectively.
Consider first the fixed exercise price strategies (Table 3). Copper prices
fell sharply in the early eighties, so the average prices obtained by these hedging
strategies are considerably higher than historical prices. Ex post it would have

18 This formula is used as a convenient approximation to the 'true' option

values, in particular the Black Scholes (1973) random walk assumption is
questionable for many commodity prices - see Hoag (1978), Brennan (1988) and
Powell (1989a).
 77

Table 3. Estimated Option Characteristics - Constant Exercise Price

Exercise Prices Mean Std Cvn Skew Max Min

Duration Put Call
(Years) (%) (%) Pnds/Tonne (%) (%) Pnds/Tonne
Spot 915 104 11.4 9.7 1123 740

80 130 913 105 11.5 10.0 1122 740

90 110 911 101 11.1 9.9 1117 754
100 100 917 95 10.4 10.2 1119 838

3 100 130 920 96 10.5 10.2 1121 838

110 130 962 67 6.9 7.6 , 1120 922
120 130 1022 34 3.4 4.1 1118 1006

5 110 130 962 67 6.9 6.9 1120 922

120 130 1022 35 3.4 4.1 1120 1005
130 130 1091 6 0.5 0.5 1119 1089

Table 4. Estimated Option Characteristics - Variable Exercise Price

Exercise Prices Mean Std Cvn Skew Max Min
Duration Put Call
(Years) (%) (%) Pnds/Tonne (%) (%) Pnds/Tonne
Spot 915 104 11.4 9.7 1123 740
1 80 140 924 99 10.8 8.8 1123 740
90 120 932 89 9.6 7.6 1142 740
100 110 950 83 8.7 9.9 1268 836
3 100 150 920 113 12.2 10.2 1142 710
100 140 949 113 11.9 10.2 1268 780
120 120 981 135 13.8 12.0 1395 780

5 110 150 913 102 11.2 9.4 1115 740

120 150 916 99 10.8 9.1 1112 763
130 130 951 89 9.4 -{l.9 1107 777

been advantageous to have hedged over this period, but there is no reason to
suppose that this favourable movement in the mean return would be expected for
future hedging strategies. The standard deviation and coefficient of variation are
the more interesting statistics. These variability measures fall sharply as the
horizon increases and as the tightness of the band is reduced. Indeed, in the final
78

130/130 5 year strategy, the implicit realised price is held near constant.
The variable exercise price strategies (Table 4) give rather little reduction
in variability. This is because the exercise prices follow the spot distribution.
From the standpoint of variability reduction, the fixed exercise price contracts
are likely to be much more attractive to LDC producers. But both contracts
reduce uncertainty significantly. Indeed, abstracting from quantity variability
complete insurance is obtained at the put exercise price and the uncertainty that
does exist is then only 'upward' in the sense that a higher price may give higher
rewards but producers are insured against very poor outcomes.
I now consider the case of Zambia hedging actual output. Over the period,
monthly output varied dramatically with a coefficient of variation of output of
20.5% (considerably greater than that of price which is 11.5% over the period).
It is therefore not surprising that revenue variability is only marginally reduced
as indicated in the results presented in Tables 5 and 6. However, the final
column of the Tables computes quarterly variability figures and this does reveal
significant variability reduction in the case of the fixed price strategies once again
at long maturities and with narrower bands. Again, the variable price exercise
strategy does not show a marked decrease in variability. However, as argued
above both strategies result in a reduction of price uncertainty giving complete
insurance for expected production at the exercise price of the put option.

Table 5. Zambian Revenue Characteristics - Constant Exercise Price

Exercise Prices Mean Std Cvn

l
Skew Max Min Cvn
Duration Put Call
(Years) (%) (%) Pounds(m) (%) (%) Pounds(m) ( Qt1y
(%
Spot 43.8 10.0 22.9 19.2 75.6 22.3 9.3

80 130 43.9 10.1 22.9 22.9 75.6 22.3 14.7

90 110 43.9 10.0 22.8 19.3 75.4 23.1 14.6
100 100 44.1 9.6 21.8 19.9 75.4 27.4 13.8

3 100 130 44.3 9.6 21.7 19.7 775.6 27.4 13.8

110 130 46.5 9.2 19.7 17.8 75.6 29.2 11.1
120 130 49.6 8.8 17.8 15.8 75.6 32.6 9.0

5 110 130 46.5 9.2 19.8 17.9 75.6 29.2 0.4

120 130 49.5 8.9 17.9 15.9 75.6 32.5 8.4
130 130 52.9 8.9 16.9 14.9 78.0 36.8 7.6
 79

Table 6 Zambian Revenue Characteristics - Variable Exercise Price

Exercise Prices Mean Std Cvn Skew Max Min Cvn
Duration Put Call (Qtly~
(Years) (%) (%) Pounds(m) (%) (%) Pnds(m) (%
Spot 43.8 10.0 22.9 19.2 75.6 22.3 19.3
80 140 45.1 11.5 25.4 22.9 79.1 25.8 18.6
90 120 45.6 11.3 24.7 20.7 77.8 25.8 17.5
100 110 46.7 10.9 23.3 17.4 74.8 25.8 15.9
3 100 150 44.3 10.1 22.8 20.0 77.8 25.3 13.6
100 140 45.9 10.4 22.8 19.8 77.8 29.2 13.9
120 120 47.5 11.2 23.6 19.9 75.4 ' 29.2 15.6
5 110 150 43.8 10.1 23.0 19.4 75.6 22.3 13.5
120 150 44.0 10.2 23.2 20.6 75.6 24.6 13.5
130 130 47.6 12.6 26.4 25.2 81.7 29.2 13.7

5. CONCLUSIONS
In this paper I have presented a theoretical framework capable of separating the
costs of the instability of commodity prices from the uncertainty attached to
those prices. With reference to Zambia I have argued that uncertainty
surrounding future copper prices has been an important source of costs. Using
historical forecasts a measure of uncertainty can be obtained and with standard
assumptions on risk aversion it was shown that the conventional method of
valuing commodity price stabilisation may seriously underestimate the benefits of
price stabilisation for a country like Zambia.
Over periods of rising nominal copper prices, Zambia has had fairly good
access to credit markets and it follows that instability may then not represent a
significant cost at all. However, poor predictions of the copper price (by the
Zambian Government but also by the multilateral agencies) may have been very
costly to the Zambian economy. In the period of high copper prices, 1964-1975,
a large amount of debt was built up with both commercial creditors and
bilaterals and multilaterals on a ruixture of fixed and floating interest rate
contracts. Large social investment programmes were embarked on and
consumption levels grew. The rigid national development plans did not lend
themselves to flexibility in funding requirements and many have argued that a
ratchet mechanism developed in Government expenditure. After the copper price
crash of 1975 Zambia was in constant debt servicing difficulties. In turn this
80

meant that further credit was withdrawn and especially through the period
1976-1977 Zambia was highly credit constrained. This, at a time when Zambia
was highly dependent on imported goods for investment related purposes. The
nominal copper price rises of 1978-1980 provoked a second round of lending and
given further optimistic price forecasts Zambia was also keen to borrow. In real
terms, however, the price did not rise to the pre 1975 levels and then in the
1980's nominal copper prices fell once more deepening Zambia'S predicament.
These poor predictions may have been extremely costly in the case of Zambia.
Secondly in this paper I have considered the use of options as a tool to
manage commodity price uncertainty and instability. L have argued that zero-
cost options (puts financed by calls) might be appropriate tools to manage the
associated risks and to illustrate these tools in practice I have presented a
simulation study for Zambia. Two types of hedging strategies were considered
where the exercise price of the put option was (i) fixed over the whole period of
the simulation and (ii) expressed as a percentage of the fluctuating spot price.
The latter did not result in significant reduction in the coefficient of variation of
returns but the former did reduce the instability of returns considerably.
However, perhaps the most important attribute of the long term options
hedging programme is that uncertainty in returns is reduced in a particular
fashion. For instance with put-call strategies the put option provides complete
price insurance at the exercise price of the put. Thus, if a minimum return is
important (due perhaps to the necessity of avoiding default) a put options
strategy may be an appropriate policy option. In general, given any asymmetry
in the costs of uncertainty an options strategy will tend to dominate a simple
futures hedging strategy that acts only on the variance of returns.
There is much further work to be done in connection with this research.
First of all, it would be interesting to discover if the results on uncertainty and
instability are only restricted to Zambia or whether there are many other
country/commodity combinations where commodity price uncertainty (as
opposed to instability) has been as dramatic. Secondly, the above work takes a
simple approach that does not include agents' decisions conditional on
uncertainty. A model of investment and consumption is the obvious extension of
the above to illustrate the costs of uncertainty more fully. Finally, there is
further work to be done in characterising how options contracts may be the
appropriate tools for managing price uncertainties when there are strong
asymmetries in costs due to, say, the possibility of default and subsequent
 81

penalties. The link to credit markets is particularly important. The fall in

commodity prices in the early 1980's was an important contribution to
developing countries' repayment problems and measures to manage commodity
price risks should be considered as important in designing a solution to payment
problems 19.

REFERENCES
Anderson, R W, C L Gilbert (1988), "Commodity Markets and Commodity
Agreements: Lessons from Tin", Economic Journal, 98, 1-15.
Anderson, R W, C L Gilbert and A Powell (1989~, "Securitization and
Commodity Contingency in International Lending' , American Journal of
Agricultural Economics, May, 553-565.
Attanasio, 0 P and G Weber (1989), "Intertemporal Substitution, Risk
Aversion and the Euler Equation for Consumption", Economic Journal,
99,59-73.
Behrman, J R (1987), "Commodity Price Instability and Economic Goal
Attainment in Developing Countries", World Development, 15, 559-573.
Bell, M (1981), "The Decline and Fall of Planning in Zambia", University of
Aston Management Centre Workin~ Paper Series, no 213.
Bevan, D L, P Collier and J W Gunning l1988), "Consequences of a Commodity
Boom in a Controlled Economy: Accumulation and Redistribution in
Kenya 1975-83", World Bank Economic Review, 1, 489-513.
Bevan, D L, P Collier and J W Gunning with A Bigston and P Horsnell (1989),
Controlled Open Economies: A Neoclassical Approach to Structuralism,
Oxford University Press, Oxford.
Black, F and M Scholes (1973), "The Pricing of Options and Corporate
Liabilities", Journal of Political Economy, 81, 637-54.
Black, F (1976), "The Pricing of Commodity Contracts", The Journal of
Financial Economics, No3, 167-79.
Brennan, M J (1988), "The Cost of Convenience and the Pricing of Commodity
Contin~ent Claims", University of British Colombia Working Paper.
Brown, C P l1980), The Political and Social Economy of Commodity Control,
MacMillan, London.
Burdette, M M (1988), Zambia, Between Two Worlds, Westview Press, Boulder,
Colorado and Avebury, England.
Coppock, J D (1962), International Economic Instability, McGraw Hill, New
York.
Corden (1984), "Booming Sector and Dutch Disease Economics: Survey and
Consolidation", Oxford Economics Papers, 36, 359-80.
Cox, J and M Rubinstein (1986), Options Markets, Prentice-Hall, Englewood
Cliffs N.J.
Eaton, J, M Gersovitz and J E Stiglitz (1986), "The Pure Theory of Country
Risk" European Economic Review, 30, 481-513.
Epstein, L G (1988), "Risk Aversion and Asset Prices", Journal of Monetary
Economics, 22, 179-192.

19 See Powell (1989b) for a discussion on this point.

82

FNDP (1966), First National Development Plan, Office of National Development

and Planning, Lusaka.
Fry, J (1980), "The Zambian Economy", in W Tordoff (ed), Administration in
Zambia, Manchester University Press, Manchester.
Gemmill, G (1985), "Forward Contracts or International Buffer Stocks? A Study
of Their Relative Efficiency in Stabilizing Commodity Export Earnings",
Economic Journal, 95, 400-417.
Gilbert, C L (1985), "Futures Trading and the Welfare Economics of Commodity
Price Stabilization", Economic Journal, 95, 637-66l.
Gilbert, C L (1987), "International Commodity Agreements : Design and
Performance", World Development, 15, 591-616.
Gulhati, R (1987), "Boxed in Zambia: Economic Recovery or Downward Spiral"
unpublished draft, World Bank, Washington, D.C.
Heathfield, D F (1971), Production Functions, The Anchor Press, Tiptree, Essex,
UK. '
Hoag, J W (1978), "The Valuation of Commodity Options", mimeo, University
of California, Berkeley.
Jaycox, E V K, R Gulhati, S Lall, and S Yalamanchili (1986), "The Nature of
the Debt Problem in Eastern and Southern Africa", World Bank Reprint
Series, 378, World Bank, Washington, D.C.
Kanbur, S M R (1984), "How to Analyse Commodity Price Stabilisation: A
Review Article", Oxford Economic Papers, 36, 336-58.
Kletzer, K M (1984), "Asymmetries of Information and LDC Borrowing with
Soverei~n Risk", Economic Journal, 94,287-307.
Kydd, J (1988), "Coffee after Copper? Structural Adjustment, Liberalisation and
Agriculture in Zambia", Journal of Modern African Studies, 26, 227-5l.
McKinnon, R I (1967), "Futures Markets, Buffer Stocks and Income Stability for
Primary Producers", Journal of Political Economy, 75, 844-6l.
MacBean, A (1966), Export Instability and Economic Development, Harvard
University Press, Cambridge, Mass.
Ndulo, M and M Sakala (1987), "Stabilisation Policies in Zambia 1976-85",
World Employment Programme Research Working Paper, 13, Inter-
national Labour Organisation, Geneva.
Neary, J P and S van Wijnbergen (eds.) (1986), Natural Resources and the
Macroeconomy, Blackwell, Oxford.
Newbery, D M G and J E Stiglitz (1981), The Theory of Commodity Price
Stabilization, Oxford Universitx Press, Oxford.
Obidegwu, C F and M Nziramsanga (1981), Copper and Zambia: An Econometric
Analpsis, Lexington Books, Mass.
Powell, A (1989a), "A General Method of Moments for Estimating the Para-
meters of Stochastic Processes for Asset Prices: An Application to the
Jump-Diffusion Process for Oil-Futures", Applied Economic Discussion
Paper, Institute of Economics and Statistics, Oxford.
Powell, A (1989b), "The Management of Risk in Developing Country Finance",
Oxford Review of Economic Policy, 5, 4, 69-87.
Powell, A and C L Gilbert (1988), "The Use of Commodity Contracts for the
Management of Developing Country Commodity Risks" in D Currie and
D Vines (eds.) Macroeconomic Interactions Between North and South,
Cambrid?,e University Press, Cambridge.
SNDP (1972), 'The Second National Development Plan", Office of National
Development and Planning, Lusaka.
Takeuchi, K, J Strongman, S Maeda and C STan (1987),"The World Copper
Industry", World Bank Commodity Staff Working Papers, no 15.
Tordoff, W (ed.) (1980), Administration in Zambia, Manchester University Press,
 83

Manchester.
World Bank (1975/88), Price Prospects for Major Commodities, Commodities
and Export Projections Division, Economic Analysis and Projections
Department; Economic and Research Staff, Report Nos.814/75, 814/77,
814/79, 814 82, 814/84, 814/86, The World Bank.
World Bank (1986), Commodity Trade and Price Trends, Johns Hopkins
University Press, Baltimore and London.
World Bank (1989), World Debt Tables 1988/9, Washington, The World Bank.

APPENDIX
Actual Nominal Copper Price and World Bank Forecasts
Made in the Following Years:

Actual 1975 1976 1977 1982 1984 1986

67-69 Avg 58
70-72 Avg 54
1973 81
1974 93
1975 56 57
1976 64 81 67
1977 59 98 80 70
1978 62 109 104 80
1979 90 117 112 95
1980 99 126 120 120
1981 79
1982 67 75
1983 72 100
1984 62 65
1985 64 192 180 180 134 70
1986 62 80 65
1987 81 68
1988 72
1989 75
1990 200 118 78
84

Actual Real Copper Price and World Bank Forecasts (1975=100)

Made in the Following Years:
Actual 1975 1976 1977 1982 1984 1986
70-72 Avg 170
1973 200
1974 190
1975 100 104
1976 112 136 112
1977 96 154 124 114
1978 85 158 149 121
1979 111 158 149 134
1980 113 158 149 158
1981 94
1982 81 83
1983 90 104
1984 80 77
1885 81 172 160 169 121 79
1986 67 82 73
1987 79 74
1988 77
1989 79
1990 135 90 81
 PART II

FINANCIAL MARKETS AND COMMODITY PRICES

There is now a significant literature on the question of how dollar primary

commodity prices respond to changes in the value of the dollar against other
currencies. Gilbert reviews the simple partial comparative static model first
proposed by Ridler and Yandle which predicts that the dollar commodity price
will fall (rise) in proportion to a rise (fall) in the value of the dollar with a less
than unit elasticity. He concludes that, despite claims to the contrary in the
literature, this conclusion is robust to generalisation of the model to a multi-
commodity world, to allow stockholding or to take into account futures trading.
It is therefore a puzzle that a number of authors report greater than unit
estimated elasticities. Gilbert locates this problem in the choice of exchange rate
index and gives new estimates of the elasticity which are consistent with the
theory both for commodity indices and for individual commodities.
Kofman and Viaene deal with the effects of exchange rate depreciation on
prices and quantities in a model with storable primary commodities. The agents
are confronted with uncertainties in commodity price, exchange rate, capital
controls, and output. Most of these uncertain and/or stochastic factors reflect the
international trade context within which most of primary commodities trade take
place. Whereas previous literature focused on a single agent approach (stressing
terms-Df-trade and welfare effects) this model considers all relevant agents with a
complete market structure based on a micro approach.
In Chapter 6, Jerome Stein provides a quantitative welfare measure of the
performance of speculative markets. It is applied to the two most actively traded
contracts on the Sydney Futures exchange: bank accepted bills and the share
price index. The bank accepted bills market is shown to function three times
more effectively, in a welfare sense, than does the share price index. The welfare
losses are due to Bayesian errors, whereby the participants do not know the true
distributions of the stochastic variables.
85
86

Finally, Rausser and Walraven construct dynamic welfare measures for a

system of futures markets that express the allocative efficiency of a particular
market as a function of its accuracy and speed of adjustment following a shock to
the system. The system comprises futures prices for t-bills, exchange rates
(German mark, British pound, Canadian dollar, and Japanese yen), and agri-
cultural commodities (corn, wheat, and cotton) for delivery in 1981 and 1982.
The results suggest that, although agricultural, exchange and financial markets
all overreact to a disturbance, agricultural markets do so to a much greater
degree. Owing to their much greater size, however, the welfare loss arising from
the overshooting is likely to be much larger for interest rate and exchange rate
markets.
 CHAPTER 4

THE RESPONSE OF PRIMARY COMMODITY PRICES TO EXCHANGE

RATE CHANGES

Christopher L. Gilbert

1. INTRODUCTION
Consider a single commodity produced and consumed in a number of different
countries. How does the US dollar price of the commodity respond to a change in
the US dollar exchange rates of these countries? This question was first posed in
this form by Ridler and Yandle (1972) in the Vietnam war period during which
the US dollar declined sharply in value. Their result was extended by Gilbert
(1973) and then subsequently rediscovered by a number of writers during the
early eighties when the dollar rose to very high levels - see in particular
Dornbusch (1985), Fleisig and van Wijnbergen (1985) and Sachs (1985). Recent
treatments are provided by Gilbert (1989a) and Radetzki (1990, pp.62-65).
It is invariably supposed that primary commodities are more or less
homogeneous, so that commodities sell at the same price irrespective of origin,
and that they are competitively marketed, so that the f.o.b. price is the same in
all countries. These assumptions may not be entirely correct for all commodities
but I shall follow standard practice and take it that they hold as a good
approximation. This amounts to the assertion that the Law of One Price holds
for primary commodities. Consequently, the problem of 'exchange rate pass
through' is much more straightforward in primary commodity markets than in
markets for less homogeneous industrial goods produced under oligopolistic or
imperfectly competitive conditions (Dornbusch, 1987; Giovannini; 1988, Fisher,
1989).
The current exercise is to examine the implications of exchange rate
changes within this context. For specificity, consider a single commodity and a
87
L. Phlips (ed.), Commodity, Futures and Financial Markets. 87-124.
© 1991 Kluwer Academic Publishers.
88

general rise in the value of the dollar against all other currencies. Since the
purchasing power of the dollar is enhanced, this change must be expected to
reduce the dollar price of the commodity, or at least not to increase it.
Conversely, other currencies are now worth less, and so the local currency prices
of the commodity should rise, or at least not fall. Taking both implications
together, it follows that the dollar price of the commodity should fall but by a
smaller proportion than the general rise in the dollar, and the local currency
prices should rise but by a smaller proportion than the decline in their values.
The lower dollar price of the commodity will result in an expansion of demand
and a contraction of supply in the United States, and t'he higher local currency
prices outside the United States will have the converse effect on supply and
demand there. This will result in a new equilibrium.
The model developed by Ridler and Yandle (1972), which embodies this
intuition and forms the basis for the discussion in section 2 of this chapter,
implies that dollar commodity prices will fall (rise) by a proportion, strictly
between zero and one, of the rise (fall) in the value of the dollar. This must be
expected, at least in a two country model, since if dollar commodity prices were
to fall by more than the rise in the dollar the implication would be that prices
would have fallen in both dollar and non-dollar terms. This would be inconsistent
with market clearing. It is therefore remarkable that a large number of studies
have found that dollar commodity prices have a greater than unit response to
changes in the value of the dollar - see Table 1 where only the estimate reported
by Dornbusch (1985) is within the unit interval. A major purpose of this chapter
is to assess whether there is in fact a greater than unit response, and if there is
not, to enquire why so many studies have reported an over-large response.

Table 1. Long Run Exchange Rate Elasticity Estimates

Author Long Run Price Exchange Rate Index
Elasticity Index

Beenstock (1988) -1.42 UN Effective LDC exchange rate

Cote (1987) -1.67 Economist IMF relative normalized unit
-1.05 IMF labour costs
-3.26 J. Commerce
Dornbusch (1985) -{).82 IMF MERM
Gilbert (1989a) -1.14 IBRD GDP weighted, PPP-corrected
-1.27 IBRD MERM, PPP-corrected
 89

The chapter is structured as follows. In section 2, I develop the simple

Ridler and Yandle (1982) model and demonstrate that it implies that dollar
commodity prices will exhibit a less than unit response to changes in the value of
the dollar. It has been suggested (Chambers and Just, 1979) that this result need
not hold in a multi-commodity model. This might provide a theoretical justifi-
cation for some of the excess response estimates, although it is difficult to see
that it could explain excess response of primary commodities in aggregate.
However, in Gilbert (1989a) I demonstrated that under reasonable assumptions
the response of individual commodity prices will remain within the unit interval.
I discuss this issue in section 3. In section 4 I note that the adjustment should
relate to real exchange rate indices, i.e. indices corrected for departures from
Purchasing Power Parity (PPP).
Primary commodity prices are flex-prices and in a world in which some
other prices adjust sluggishly one frequently finds that flexible prices can over-
shoot. This suggests that even if in equilibrium primary commodities have a less
than unit price response to exchange rate changes, the short run response might
exceed unity. This is another confusion. It is true that if exchange rates over-
shoot, in response say to a monetary shock, and if a commodity price has a near
unit response to exchange rates, the commodity price will itself overshoot its
equilibrium value. However, the question at issue is not the response of exchange
rates and commodity prices to a monetary shock but the response of commodity
prices to an exchange rate shock. In section 5, I examine the dynamics of
commodity price adjustment and conclude that there is no possibility of over-
shooting in response to an exchange rate shock, although if commodity risk is not
completely diversifiable one may find that commodity prices respond with a lag
to exchange rate changes.
In section 6, I investigate how the fact that many primary commodities are
actively traded on futures markets affects these conclusions. This issue has been
discussed in a number of recent papers (Viaene, 1989; and Kofman, Viaene and
de Vries, 1989; Kofman and Viaene, 1990). Nevertheless, I conclude that futures
markets are almost completely irrelevant to the question of the exchange rate
response of spot commodity prices.
At the level of pure theory the simple Ridler and Yandle result is robust.
The remaining sections of the chapter are empirical. In section 7, I look at the
definition of the exchange rate index to which the commodity prices should be
related. There are two issues here - the currency coverage of the index and the
90

choice of weighting scheme. In section 8, I report results which demonstrate that

inappropriate choice of weights can result in estimated elasticities which lie
outside the theoretically predicted range, and in particular I note that the
frequently used MERM index will tend to give over high estimates.
On the question of coverage, almost all indices exclude Less Developed
Country (LDC) currencies. This is potentially important in analyzing the
response of primary commodity prices to exchange rate changes since many
commodities are substantially produced in LDCs. In section 9, I show that if
LDC exchange rates move more than proportionately to other developed country
rates with respect to changes in the value of the dollar then again one could find
commodity price responses which lie outside the theoretically predicted range. I
report some evidence from Latin American primary producing countries that
suggest this has indeed been the case. This relates to the findings reported in
Gilbert (1989a) that commodity price responses to exchange rate changes have
been exacerbated by LDC indebtedness.
Section 10 gives new estimates of the exchange rate response of the
commodities which make up the World Bank 33 Commodity Index. The results
are broadly compatible with the theory, but a significant group of commodities,
including many metals, are estimated as exhibiting greater than unit exchange
rate responses, and a further group of commodities appear to exhibit much lower
responses than the theory requires. Section 11 contains brief conclusions.

2. A SIMPLE STATIC MODEL

Ridler and Yandle (1982) were, so far as I am aware, the first authors to provide
a formula which enables prediction of the quantitative size of the response of
primary commodity prices to exchange rate changes under the Law of One Price.
Suppose there are n countries. Write the local currency price of the commodity in
country i as Pi' and country i's US dollar exchange rate (local currency units per
dollar) as Xi. Write consumption of the commodity in country i as ci=d(Pi) and
production as Qi=Qi(Pi)' allowing for the possibility that either or both of Ci
and Qi may be zero. Let total world consumption be C=ECi and total production
be Q=EQi. Write country i's consumption and production shares as wi=C/C and
wi=Q/Q. I take the USA to be country 1 so that Xl =1. Write the US dollar
price PI =p for simplicity. Then Pi=xiP, ci=d(xiP) and Qi=Qi(xiP).
Market clearing in a primary commodity market requires
 91

Qt + St_l = Ct + St (1)
where St is the level of inventories at the end of period t. In a long term static
equilibrium stocks will be constant and so (1) implies (dropping the time
subscript)
n. n.
~ Ql(x.p) = ~ C1(x.p). (2)
i=l I i=l I
The comparative statics of a change in exchange rates may be analyzed by total
differentiation of (2) to obtain
n n
~ Q~(pdx. + x.dp) = E C~(pdx. + x.dp) . (3)
i=l I I I i=l I I I

Noting that Qi'xiP=fiQi=fiwiQ, where fi is the elasticity of supply of the

commodity in country i, and d'x.p=-e.C.=-e.w.C where e· is the uncompen-
I I I I I I
sated own elasticity of demand of the commodity in country i (defined as
positive) we may write equation (3) as

n f·W· [dXi
E - + -dP] = - i
n e·w· [dx
E + -dP] . (4)
i=l I I xi P i=l I I xi P

Approximating derivatives as first differences and solving, we obtain

n
b..lnp = - E v.b..lnx. (5)
i=l I I

wi fi + wiei
where vi = --:n::-------
E (W.f. + wJ.eJ.)
j=l J J

The weight of the change in each country's exchange rate therefore depends on its
importance in both the production and consumption of the commodity and on the
equilibrium supply and demand elasticities of the commodity.
Provided that the commodity is not a Giffen good in any economy and that
there are no perverse supply responses (I take it that both these assumptions are
reasonable), the weight attached to each country's currency change is non-
negative (vi~O). Furthermore, and without assumption, EVi=l. Now, the
summation in (5) runs through all n countries, but by definition the US dollar
exchange rate against US dollars Xl =1 and hence b..x l =0. Consequently we may
change the limits of summation in equation (5) to exclude the USA:
92

n
llinp = - E v.llinx. . (6)
i=2 1 1
Now suppose the appreciation of the dollar is uniform across all countries so that
lllnxi=O, i=2, ... ,n. Then
n
llinp = - ( E v.)O = - (I-vI)O. (7)
i=2 1
We find therefore that, as anticipated, the commodity price falls by a fraction of
the appreciation of the dollar.
It will generally be natural to assume that demand elasticities are uniform
across all consuming countries and that supply elasticities are uniform across all
producing countries, if for no other reason than the lack of reliable country-
specific estimates. Setting fi=f and ei=e, i=I,2, ... ,n, we obtain
f e (8)
Vi =""ET"e Wi +""ET"e Wi
and, in the case of a uniform 10011% appreciation of the dollar

llinp = - [1 _ _f_ WI _ _e_ WI] 0 (9)

f+e f+e

he fall in the dollar commodity price is therefore a weighted average of the shares
of countries other than the USA in production and consumption of the
commodity with weights proportional to the supply and demand elasticities
respectively. Since this is an equilibrium model, the relevant elasticities are those
relating to the long run. It is generally believed that although short term supply
and demand elasticities are low for primary commodities, the long run elasticities
are often much higher. There is however no a priori basis for supposing that
either demand elasticities are higher than supply elasticities in the long run or
vice versa and in general one would need to take into account the exchange rate
movements of both producing and consuming countries.
Table 2 tabulates the Ridler-Yandle dollar commodity price elasticities for
major primary commodities using 1986 share data assuming in the first column
that the demand elasticities are uniform and dominant (elf very high) and in the
second column that supply elasticities are uniform and dominant (el f near zero).
The final column takes those demand and long run supply elasticities given by
Bond (1988) and this allows computation of the exchange response using equation
(9). The notable feature of this table is that in no case are the US production or
consumption shares so low as to give exchange rate elasticities which are
 93

substantially less than unity, the lowest figure being for aluminium (on the
consumption side) at 0.73. All the elasticities computed in the final column are in
the range 0.75 to 1.0. If the simple Ridler-Yandle model provides a reasonable
description of the world, this is the range of responses for which we should look.

Table 2. Ridler-Yandle Commodity Price Elasticities (1986 share data)

Consumption Production Bond

weights weights weights
Cocoa 0.80 1.00 0.93
Coffee 0.79 1.00 0.91
Cotton 0.90 0.86 0.88
Jute 1.00 1.00 1.00
Natural rubber 0.84 1.00 0.89
Sisal 0.99 1.00 0.99
Sugar, raw 0.96 0.96 0.96
Tea 0.96 1.00 0.97
Tobacco 0.91 0.91 n.a.
Wheat and wheat flour 0.94 0.89 0.91
Wool 0.94 0.95 0.95

Petroleum, crude 0.78 0.85 n.a.

Bauxite/aluminium 0.73 0.99 0.79

Copper 0.79 0.84 0.81
Lead 0.80 0.91 0.86
Nickel 0.82 1.00 n.a.
Tin 0.85 0.94 0.85
Zinc 0.85 0.95 0.92
Notes: Consumption of agricultural commodities is apparent; production of
non-ferrous metals includes secondary production but excludes direct
use of scrap; secondary production figures for CMEA countries are
taken as proportional to total consumption. The 'Bond weights' take
demand and supply elasticities from Bond (1988). In cases where she
gives only one elasticity, the other is taken as the sector average.
Sources: FAO: Production Yearbook, 41, 1987; Trade Yearbook, 41, 1987,
Rome; Metallgesellschaft AG: Metallstatistik, 74, 1987, Frankfurt am
Main; UN: Energy Statistics Yearbook, 1987, UN, New York.

3. MULTI-COMMODITY GENERALIZATIONS
There are two ways that the Ridler and Yandle model discussed in section 2 may
be interpreted. On the first interpretation, the price of a particular commodity
varies in response to exchange rate changes but the prices of all other
commodities (primary, intermediate, final) remain fixed. On the second
94

interpretation, all other prices are allowed to vary and the elasticities fi and ei
are interpreted as total elasticities. The first conceptual experiment is obviously
flawed since other commodity prices will clearly respond in the same way as the
price of the commodity of interest - it would be inconsistent to use the Ridler-
Yandle model to evaluate the response of the price of natural rubber to a change
in exchange rates holding the price of synthetic rubber constant and then to
repeat the exercise for synthetic rubber holding the price of natural rubber
constant. The difficulty with the total elasticity approach, however, is to find a
reasonable basis for assigning magnitudes to the total elasticities.
These problems are illuminated by multi-commodiiy generalizations of the
Ridler-Yandle result discussed in Chambers and Just (1979) and Gilbert (1989a).
Chambers and Just argued that in a multi-commodity world, the dollar price of
an individual commodity might respond more than proportionately to a change
in the value of the dollar. In Gilbert (1989a) I demonstrated that, although this
contention is not actually incorrect, the Chambers and Just result will only hold
exceptionally, and under the reasonable assumption of gross substitutability it
may be ruled out. Suppose there are g traded commodities in the world economy
and an unspecified number of non-traded commodities. The dollar price of good j
is Pj and the g-vector of traded good prices is p. Qij=Qij(xiP) and Cij=dj(xiP)
are respectively the production and consumption of commodity j in country i.
Write aggregate production of commodity j as Qj =l:8ij and aggregate
consumption of j as Cj=l:i Cij" Market clearing requires
n.. n··
l: QIJ(X.p) = l: CIJ(x.p) (j=1, ... ,g). (10)
i=1 I i=1 I
Write Eijk=81.nQi/81nPk and eijk=-81nCi/81nPk. Also define the production
and consumption shares as Wij=Qi/Qj and wij=Ci/Cj" Total differentiation of
(10) gives

~ ~ f.. kw.. [dXi

i=1 k=1 IJ IJ Xi
+ dPk]
Pk
= ~ ~ [dx
e.. kw .. i
i=l k=1 IJ IJ Xi
+ dPk]
Pk
(j=l, ... ,g). (11)

Write aijk = WijEijk + wijeijk (1=l, ... ,n; j,k=1, ... ,g).
Then, approximating derivatives by first differences, (11) becomes

~ [~ a. ·k]
k=l i=1 IJ
~lnpk = - i=2
~ k=1
~ a.IJ·k~lnx.I (12)
 95

Now write Ai=(aijk ) and A=EiAi . Then (12) may be written as

-1 n
Lllnp = -A E A.lD.Elnx. (13)
i=2 1 1
where l' =(1, ...,1), the g-vector of units. Equation (13) indicates that the change
in each commodity price depends on the entire set of own and cross supply and
demand elasticities, and the entire set of production and consumption shares.
This unhelpful result simply makes explicit the observation that the Ridler-
Yandle elasticities in equation (5) must be interpreted as total elasticities.
In the case of a uniform 10011% appreciation of the dollar, equation (13)
simplifies to
Lllnp = - (I-A-1 A 1)lO (14)
In Gilbert (1989a) I quoted a proof, due to J.A. Mirrlees, that gross substitut-
ability in production and consumption is sufficient to ensure -Ol~Lllnp~O so that
one should again expect that the long run elasticity of dollar commodity prices
with respect to a change in the value of the dollar should lie in the unit interval.
Although equation (14) is too general to be useful, a specific example may
illustrate the implications of considering interactions between commodities.
Consider again the case of natural (n) and synthetic (s) rubber which are close
substitutes in consumption but not in production. Suppose that the supply
elasticity fn of natural rubber is 0.5 and assume that the supply elasticity fS of
synthetic rubber is 1.0. Take both the own elasticity of demand for natural
rubber (defined as positive) enn and that for synthetic rubber ess to be equal to
0.5 and the aggregate cross elasticities to be equal to -2.0. (The normalization
Pn=ps=l ensures that this is compatible with Slutsky symmetry.) The USA is
not a producer of natural rubber (wln=O). For illustrative purposes, I take its
share of world synthetic production w1s to be 0.2. On the demand side I take its
share of natural rubber consumption wIn to be equal to 0.1 and that of synthetic
rubber consumption w1s to be equal to 0.2. Now consider a general 10%
appreciation of the dollar. The Ridler-Yandle result (9) would imply that the
dollar natural rubber price would fall by 9.17% and the dollar synthetic rubber
price would fall by 8.00%. The multi-commodity generalization (14) revises these
figures to 8.81% and 8.46% respectively. Substitution from synthetic rubber
moderates the fall in the dollar natural rubber price and enhances the fall in the
synthetic price. By taking into account the high degree of substitutability
between natural and synthetic rubber we find that the prices move in a much
96

more uniform manner than predicted by the Ridler-Yandle formula.

The illustrative calculations in Table 2 suggested that dollar exchange rate
elasticities should be in the range 0.75-1.0, although I noted that there might be
some commodities for which the response would be lower. The multi-commodity
generalization of the Ridler-Yandle results suggest that these low responses are
likely to be averaged up towards the others thereby reinforcing the earlier
conclusion.

4. RELATION TO PURCHASING POWER PARITY

If a currency is inflating relative to other currencies, that currency will need to
depreciate over time to keep domestic prices in line with those obtaining else-
where. This underlies the notion of Purchasing Power Parity (PPP). The PPP
theory is well set out by Dornbusch (1988) and the empirical evidence relating to
PPP is surveyed in Kravis and Lipsey (1978). Both the strong hypotheses that
PPP holds precisely in all countries at all times, and the weaker hypothesis that
exchange rate movements may be explained in terms of adjustments towards
PPP, are incompatible with the major movements in developed country exchange
rates over the past two decades. However, it is reasonable to suppose that there
does remain a weak tendency towards PPP within the exchange rate system; and
that the currency of any country where inflation is particularly fast must be
expected to depreciate to offset this relative inflation.!
These observations are important in the current context since use of the
Ridler-Yandle formula in the presence of rapidly depreciating currencies will tend
to overestimate the rise in the value of the dollar and hence the likely fall in
dollar commodity prices. If, for example, there is rapid inflation in Brazil, it
would be wrong to see sharp depreciation of the Brazilian cruzeiro as likely to
result in a fall in the dollar price of coffee. What matters is the real and not the
nominal devaluation of the cruzeiro. If there is a real devaluation, then the
Brazilian supply curve for coffee will shift down in dollar terms, and there will be
a fall in the dollar coffee price; but if the cruzeiro devaluation only matches
Brazilian inflation, the supply curve will not shift and there will be no change in
dollar coffee prices.

! PPP requires the Law of One Price to hold for all commodities, traded and
non-traded. Hence PPP implies the Law of One Price for primary commodities,
but the converse is not true.
 97

Gilbert (1973) extended the Ridler-Yandle result to take into account

different inflation rates, and in Adams and Vial (1988) and Gilbert (1989a) the
connection with PPP is made explicit. The most simple approach in the context
of the Ridler-Yandle model is to make both consumption and production depend
on real (i.e. deflated) prices as ci =d(xi P/7ri ) and Qi=Qi(xi P/7ri ) where 7ri is a
suitable price index for country i. The algebra proceeds exactly as in section 2
and one obtains in place of (6)

Ev·1 [~lnx.-~ln7r.].
~lnp = - i=1 1 1
(15)
The weights Vi are exactly the same as those derived in section 2. Noting that
~lnxl =0, we may rearrange (15) as

Ev·1 [~lnx.-~ln(7r.f7rl)]
~lnp = ~ln7rl - i=2 1 1
(16)
or
n
~ln(p/7rl) = - E v.~lnx'r (17)
i=2 1 1

with ~ln(7r/7rl)=~ln7ri-~ln7rl' the rate of inflation in country i relative to that

in the USA and lnxt=lnxi-ln( 7ri / 7r1)' Equation (17) indicates that the deflated
dollar commodity price (note that deflation is using the US deflator) responds to
weighted changes in real exchange rates. If PPP held at all times, then one would
have ~lnxi=~ln(7r/7rl) for all i and dollar commodity prices would increase
uniformly with US inflation (equally, we could explain DM or Yen commodity
prices entirely in terms of German or Japanese inflation). There would be no
need to consider the question of the impact of exchange rates on commodity
prices since there would be no possibility of independent shocks arising from
exchange rate changes. In fact PPP does not hold precisely so the question
addressed in this chapter does have content. But the implication is that one
needs to use an exchange rate index which takes explicit account of changes in
the degree of departure from PPP.

5. STOCKHOLDING AND INTERTEMPORAL PRICE ADJUSTMENT

The Ridler-Yandle model introduced in section 2 and developed in sections 3 and
4 compares prices in long run static equilibria associated with different sets of
exchange rates. We now have to ask how relevant is this model in considering the
98

dynamic adjustment of a commodity price over time to a change in exchange

rates. In particular, is it possible that a shock to the exchange rate may result in
different short and long run responses? Since the publication of Dornbusch
(1976), we have become familiar with the idea that, in a world in which some
prices are sticky, the flexible prices may overshoot their equilibrium values. Van
Duyne (1979) and Frankel (1986) proposed models in this spirit in which a
commodity supply shock (Van Duyne) or a monetary shock (Frankel) could give
rise to short run overshooting by both commodity prices and exchange rates.
However, this does not imply short run failure of the Law of One Price since for
that, one needs to establish a discrepancy in the' relative movements of
commodity prices and exchange rates. To examine this question we need to
consider the impact of an exchange rate shock within a model which allows
stockholding.
For a non-storable commodity, the short and long run effects of an
exchange rate adjustment may be quite different. For example, consider an
agricultural crop for which there is no carryover from one harvest year to the
next and where the only price response within the harvest year is in consumption.
In using the Ridler-Yandle formula to compute the effects of changes in exchange
rates within the harvest year, the supply side would be irrelevant and, if demand
elasticities are uniform across countries, the dollar price would change in
proportion to one minus the US consumption share. However, if supply is very
elastic from one year to the next, the long run effect of the commodity price
change will depend substantially on the supply side weights, and if supply
elasticities are high and uniform across countries, the dollar price change will be
proportional to one minus the US production share. If the USA is a net exporter
of the commodity, the dollar price will overshoot its long run level. For a storable
commodity, stockholding tends to smooth out these movements.
I first outline a simple rational expectations market model, and then
generalize this model to allow multiplicity of currencies. Long run equilibrium in
a market for a storable commodity has production equal to consumption while in
short run equilibrium, supply, equal to production plus carryover (lagged stocks)
is equal to demand, consumption plus stock demand. This is equation (1). The
model employed in the commodities literature makes stock demand a linear
function of the expected capital gain from holding stocks. We may approximate
that capital gain as
EtlnPH 1 -lnPt - rt
 99

where rt is the interest rate over the intended stockholding period. Stock demand
is then
St = a(EtlnPH 1 -lnPt - rt )· (18)
Equation (18) may be rationalized as the outcome of agents' decisions to
maximize expected utility in a two asset portfolio selection model involving
commodity stockholding and a risk free asset with return r if agents' utility
functions are negative exponential - see for example Gilbert (1989b). In that case
a=l/ Au; where A is the representative agent's coefficient of absolute risk and
U;=Et(lnPHCEtlnPHl)2. This is the model employed by ~uth (1961) in his
pioneering work on rational expectations, and more recently by Pesaran (1987)
and Gilbert and Palaskas(1990). A crucial feature of this specification is that it
ignores the non-negativity constraint on stockholdings. 2
The most simple model in which the stock demand function (18) may be
embedded is one in which production and consumption are both linear in the
logarithm of the current price:

Qt = Q[1 + f(lnPClnp)] + elt (19)

and
(20)
where Inp=Elnp, and f and e may be interpreted respectively as the supply and
demand elasticities at p. The disturbance terms elt and e2t are stationary
additive shift terms which incorporate all the observed and unobserved variables
affecting production and consumption. Since market clearing requires Q=C, we
may normalize such that both are equal to unity with the implication that
quantities are measured relative to 'normal' production or consumption. The
market clearing identity (1) now becomes
St = St-l + et + (e+f)(lnPClnp ) (21)
where et =e lt -e2t" Elimination of Inp between this equation and the stock
demand equation (17) gives a second order difference equation in St' the solution
of which may be equivalently be expressed in terms of InPt - see Pesaran (1987)
and Gilbert and Palaskas (1990). The resulting price equation is

2 See Deaton and Laroque (1989) and Gilbert (1989b).

100

InPt -lnp = ), [lnPt_Clnp+rt_1] - !-:;:-~ i!O),i [EteHi-),Et_1 eHi]

- ),i!O),i [Etrt+i-),Et_1rt+i] (22)
where), is the stable root of the equation
y2_[2+e~f]Y+1=0. (23)
Equation (22) shows the commodity price adjusting to ),-quasi differences in the
expected value of current and future supply and demand shift variables e and of
interest rates r.
The model defined by equations (18-20) together with the market clearing
identity (1) is excessively simply, since all the dynamics arise from stockholding.
A straightforward generalization is to make either production or consumption
depend on the lagged expectation of the current price, Et_1InPt, in place or
instead of the current price itself. This complicates the resulting solved price
equation without essentially changing its form - see Muth (1961), Pesaran (1987)
and Gilbert and Palaskas (1990). However, introduction of longer price lags or of
partial adjustment terms into the production and consumption equations results
in the characteristic equation (23) becoming of higher and possibly infinite order,
and this prevents analytic representation of the solution. For the purpose of
examining the effect of exchange rate changes on commodity prices, it is therefore
preferable to remain with the simple model and to use this as a guide as to the
likely effect in realistically complicated contexts.
I now extend the model to accommodate exchange rate movements. I again
suppose that production, consumption and now stockholding takes place in each
of n countries, allowing for the possibility that in many cases one or more of these
activities may be absent. Equations (18-20) are replaced by

Sit = (¥iGi [EtlnPi,HClnPit-rit) (24)

(25)
and

Cit = Gi [1 - ei(lnPiClnPi)] + ei2t (i=l, ... ,n). (26)

As in section 2, we have
InPit = InPt + lnxit (27)
where Xi is country i's dollar exchange rate.
 101

In principle, this model might give much more complicated dynamics than
that implied by equations (18-20). However, Gilbert and Palaskas (1990) show
that the assumption of Uncovered Interest Parity (UIP) makes this model
isomorphic with the former (see also Gilbert, 1985a). To see this, note that DIP
implies
rit = r t + (E t lnxi ,t+1-lnxit) (28)
where rt=r lt , the interest rate in the USA. Equation (28) states that the interest
rate in country i exceeds or falls short of that in the United States by the extent
of the expected depreciation or appreciation of country i's currency against the
US dollar. Equations (27) and (28) allow us to simplify the stock demand
equation (24) to obtain

Sit = a/\[EtlnPH clnPt -r t ] (29)

so that stock demand in each country is proportional to the expected capital gain
from holding stocks expressed in US dollars (or, by symmetry, in any other
currency). Aggregating equation (29) across countries, we obtain the original
stock demand equation (18) where a is now given by
n
a= E w.a.
i=l I I
where wi=C)E/~i as in section 2.
The market clearing identity (1) now becomes

St = St_1 + et + (e+E) [(lnPClnp) + i!2 vi (lnxiClnj)] (30)

in place of (21), where et=Eiwieilt-Eiwiei2t and the exchange rate weights Vi are
exactly those given in equation (5) in the context of the Ridler-Yandle model.
Write
n
Inxt = E v.lnx· t . (31)
i=2 I I
We may define an exchange rate adjusted price pi as
~=~+~. (~
Equation (30) now reduces to (21) with p* replacing p. Making the same
transformation in the stock demand function (18) one obtains
St = a(EtlnPi+ClnPi-rt) (33)
where
102

The implication is that generalization of the model defined by equations

(18-20) gives a model which is isomorphic with the original but defined in terms
of an exchange rate adjusted price p* and with an appropriately defined exchange
rate adjusted interest rate r*. Substituting back in terms of the dollar price p and
the dollar rate of interest r, equation (22) becomes

InPt -lnp = >.[lnPt_ClnHrt_l] - !~; i!/[Etet+C>.Et-leHi]

- >. i!O >.i [EtrHi->.Et_lrt+i] (34)

- (1->') i!O >.i [Et(lnxHi-Inx)->.Et_l (InxHClnX )]

The dollar commodity price is therefore seen to adjust to a discounted forward
sum of weighted anticipated future exchange rate quasi-differences. To aid
interpretation of equation (34), consider the most simple case in which all
exchange rates follow martingale processes; i.e. all changes in exchange rates are
unforecastable. This assumption implies that
Etlnxt + i = lnxt , all i>O.
On this assumption, the final term in equation (34) simplifies to

- [(lnxClnX)->. (lnxt _ C lnx)]

and equation (34) reduces to equation (22) of the single country model with the
exchange-rate adjusted price p* replacing p. Writing this equation in terms of the
dollar price p, one obtains

InPt -lnp = >. [lnPt_ClnHrt_l] - !~; i ! / [EteHi->.Et-l eHi]

- >'i!O >.i[Etrt+i->.Et_lrt+i] - [lnxt-InX] + >.[lnxt_Clnx] (35)

The dollar price adjusts smoothly to the new information on production,

consumption and interest rates, but also to changes in exchange rates. There is
no possibility of overshooting in this model and indeed, the model suggests a
degree of lagged response.
The eventual conclusion of this analysis is therefore that the introduction of
stockholding does not in any way invalidate or even significantly modify the
results we obtained the the static Ridler-Yandle equilibrium model. The dollar
commodity price adjusts in relation to the same weighted basket of currencies
 103

where the weights reflect the production and consumption elasticities and shares.
On the assumption of UIP it is irrelevant where stockholding takes place and
there is therefore no requirement to modify the Ridler-Yandle weights to take the
location of stockholding into account. The only additional insight obtained from
this model is that, under risk aversion, the commodity price will follow a partial
adjustment towards its new equilibrium value rather than jump immediately to
the new value. However, that implication is the consequence of the risk aversion
assumption and should not hold in models which assume risk neutrality, or,
equivalently, complete diversifiability of commodity price risk.

6. FUTURES MARKETS
The most important world currencies and very many major primary commodities
are traded on organized futures markets. To what extent do these markets force
modification of the results of the preceding sections?
First consider the situation in which there is commodity futures trading but
there are no currency futures. Consider a non-storable agricultural commodity in
which production decisions are made on the basis of an expected price and where
there is no possibility of responding to the current price, but consumption
responds to the current price. In this situation, producers have an incentive to
hedge their output by selling their expected production forward. Note that
consumers do not benefit from hedging in this model since they are not required
to make advance commitments. Producers can hedge by selling futures. If either
there is no production uncertainty or if production risk is additive, and if there is
no basis risk, producers will wish to hedge their entire expected output and the
standard separation theorem (Anderson and Danthine, 1983) will imply that
production decisions depend only on the futures price and no longer on the
expected price.
To see this, consider a producer maximizing the expected utility derived
from profits IT where
IT = pq + (f-p)h - wz . (36)
q is the random variable denoting his commodity output (realization q), p is the
random commodity price and z is his input of effort which costs him w per unit.
He sells h units of output forward at fixed price f. The production function is
q = g(z,'E) (37)
104

where f is the stochastic disturbance. The first order condition for maximization
with respect to z is the same as in the absence of hedging:
(38)
where gz is the stochastic marginal product of effort, 8g(z,f)/fk. The first order
condition with respect to the hedge h requires that marginal utility U' be
orthogonal to the spot price innovation p-f - see Nermuth (1982) and Gilbert
(1985b ):
EU'(f-p) = o. (39)
To obtain the Anderson and Danthine separation theorem we need to assume
that the marginal product gz is non-stochastic. This will be the case if the
disturbance term e in (37) enters additively. We may then substitute from (39)
into (38) to obtain
fgz=w (40)
which simply states that the marginal product of effort valued at the commodity
forward price is equal to the cost of effort. However, this result will fail to hold if
either production disturbances are multiplicative, as is realistic (Newbery and
Stiglitz, 1979), or if the price obtained through futures hedging is itself stochastic
(i.e. there is basis risk, see Gilbert, 1989b).
In general, with risk averse producers, the fact that output is sold forward
will shift the supply curve and may affect its elasticity (Newbery and Stiglitz,
1979). For example, suppose that the producer's utility function is negative
exponential giving the standard mean-variance portfolio selection model. For
simplici ty, I also suppose that there is no correlation between his additive output
disturbance and the commodity price, which is realistic for a small producer
producing under independent conditions. His expected utility is then
EU = (pg(z)-wz) _-1ap2g(z)20-; (41)
where 0-; is the coefficient of variation of the commodity price. The first order
condition for maximization of (41) is

l
w
g' (z) =
p 1-apg(z)o-p
2J . (42)

Hence the higher the variance of the price, the higher the marginal product of
effort and the lower the commodity output for any expected price p. If output is
fully hedged, the corresponding coefficient of variation in the denominator of (42)
 105

disappears and commodity output is at the level corresponding to risk neutrality.

Thls is sufficient to make clear that the existence of a commodity futures
market alters the way that producers behave. In the present context, however,
the crucial point is that conditional upon producers' supply functions spot market
clearing is still given by equality of consumption demand with actual output,
exactly as in the absence of hedging. For the non-storable agricultural
commodity, the spot price is simply given by
n n
E C.(x.p) = E Q. (43)
i=l I I i=l I
where Qi is aggregate realized output in country i and Ci(xiP) is aggregate
consumption demand in country i whlch depends on the local currency
commodity price Pi=xiP. We are therefore back in the Ridler-Yandle framework
of section 2, and the analysis of the effects of changes in exchange rates is exactly
as previously. Furthermore, since in thls case the short run price response to
exchange rate changes only reflects demand side responses, the fact that
producers can hedge their output has absolutely no effect on this response,
although the long run response will reflect the altered supply elasticities and
production shares.
The same argument applies if the commodity is storable although we now
have to worry about the effects of futures trading on commodity storage. I
addressed thls question explicitly in Gilbert (1989b). In the absence of basis risk
there is again a separation theorem, and stock holdings depend only on the
futures price. By selling their stocks forward, i.e. going short futures, stock-
holders can reduce or even eliminate the risk associated with the stockholding
activity and this implies that they will be willing to hold larger quantities of
stock for any anticipated capital gain. In terms of equation (18), the coefficient a
becomes higher and thls implies that the adjustment parameter ). is closer to
unity. One has the same stockholding equation (35) but with faster adjustment
to the new equilibrium value. The effects of commodity futures trading is there-
fore seen in the values of the production elasticities and price adjustment para-
meters, but conditional on these elasticities the Ridler-Yandle formulae still
apply.
The analysis of the hedging decision becomes more complicated once
currency futures are introduced. If a producer's costs are in terms of local
currency and revenues are received in dollars, then the producer may also wish to
hedge the currency risk associated with his export revenues. Thus if we take
106

equation (36), the producer's profits in local currency become

II = x{P<i + (f-p)h) - wz. (44)
In principle, however, the commodity price risk associated with p and the
exchange rate risk associated with x will not be independent. A world shortfall in
production of a commodity will result in a rise in the dollar commodity price and
therefore either a fall or a rise in aggregate world receipts of dollars by exporters
of the commodity depending as whether demand is elastic or inelastic. Suppose
that the USA is a net importer of the commodity. A fall in the value of US
imports will in equilibrium require a rise in the valu~ of the dollar, and vice
versa. The correlation between currency values and commodity prices will
determine the form of the optimal hedge.
This topic forms the subject matter of Kofman, Viaene and de Vries (1989),
Viaene (1989) and Kofman and Viaene (1990), all of which employ a two country
two period model of equilibrium in spot and futures markets which generalize the
single country models of Kawai (1983) and Turnovsky (1983). There are two
major difficulties with their approach. First, it is unclear that the extent of
interaction between commodity prices and exchange rates which obtains in a two
country model in which commodity production risk is a major or the only source
of uncertainty is representative of the likely interactions in a multi-country
multi-commodity world in which production risk is only one of many sources of
uncertainty. By focussing on these interactions, 'North-South' models (see also
Currie et al., 1988) tend to exaggerate their quantitative significance. But second,
however complicated the form of the producer's (or marketing board's) optimal
hedge, it will remain the case that spot market clearing is given by equation (1)
ill. which consumption and stock demand for the physical commodity is equated
to available supply. Furthermore, the spot commodity price will adjust to
exchange rate changes in the manner described in sections 2 to 5. Futures
trading, whether on commodity or currency markets, is only relevant in so far as
it alters the supply and demand elasticities, production and consumption shares
and price adjustment parameters. If we condition our analysis on a specific set of
parameter values, it is irrelevant whether or not futures markets are present.
In order for futures trading to affect the analysis of spot market reactions to
commodity prices, the existence of the futures market must induce changes in the
pattern of spot market trades. In Gilbert (1989b) I emphasized the role of futures
markets in facilitating storage. It may also be true that futures markets induce
 107

higher levels of intertemporal substitution in consumption since the opportunity

cost of delaying consumption in periods of shortage is made explicit. However, in
the end futures markets remain derivative markets and in long run equilibrium
the spot price is determined by the intersection of supply and demand curves in
the presence of futures trading as in its absence. The focus of interest should
therefore be on how, if at all, futures markets affect the elasticities of these
curves and not how they affect spot market prices, since in the long run it is only
through altering the shape and location of these curves that futures trading can
affect spot prices.

7. THE EXCHANGE RATE INDEX

The Ridler-Yandle result (section 2) demonstrated that we should expect dollar
primary commodity prices to respond with an elasticity of approaching unity to
an appropriately weighted average of dollar exchange rates. We have already
defined an exchange rate index x in equation (31). If we now reintroduce the PPP
corrections we obtain a real exchange rate index X as

InX t =
i=2 I
E
v· [lnx·t-ln( 1I"'t l 11" It)] =
I I
E v.lnx'!'t
i=2 I I
(45)
and it follows from equation (5) that the long run elasticity of the commodity
price with respect to this index should be (negative) unity provided the 'correct'
weights vi are used. Note however that the weights in this index sum to one
minus the US share, with the implication that a 1000% real appreciation of the
dollar against all other currencies will result in only a 100(I-v 1)0% rise in this
index. Essentially, therefore, the predicted long run exchange rate elasticity has
been absorbed into the exchange rate index. It is more usual to consider indices
with weights summing to unity, and this suggests a modified index

InXi = i!2 vi [lnXit -In( 1I"it l 11" It)] = i!2 vilnxit (46)
v.
h
were
*_
vi - I-v
I
(i=2, ... ,n).
1
The equilibrium real dollar commodity price response to this index is predicted to
be - (I-vI)' and this is in line with the calculations in Table 2.
There are two important practical difficulties in calculating the real
exchange rate index X* defined in equation (46). First, the multi-commodity
model developed in section 3 indicates that the 'appropriate' weights vi depend in
principle on the production and consumption shares and on the own and cross
108

supply and demand elasticities of consumption and production of all traded

commodities. It is clearly not feasible to calculate these general equilibrium
weights. We have seen that the effect of these inter-commodity substitution
effects is to even out the exchange rate elasticities across commodities. This
raises the question of whether or not it is better to use commodity-specific
exchange rate weights, as implied by the Ridler-Yandle formulae of section 2, or
to use some non-specific (e.g. GDP) weights which may to a greater or lesser
extent approximate the general equilibrium weights. I discuss this question in
section 8.
The second difficulty relates to the PPP corrections. Many primary
commodities are substantially produced in developing countries. It may be
difficult to obtain good time series for exchange rates for some of these countries
where foreign exchange transactions have been controlled, but it will often be
impossible to compute the appropriate PPP corrections. On the other hand,
rapid inflation in a significant number of these countries implies that failure to
make the PPP corrections will result in the overall exchange rate index becoming
swamped by massive depreciation in the inflating countries. The result is that in
most empirical work investigators have employed exchange rate indices which
cover only the developed countries. I examine the consequences of less than full
coverage in section 9.

8. CHOICE OF WEIGHTING SCHEMES

The Ridler-Yandle model implies that it may be useful to use commodity specific
weights in constructing exchange rate indices to explain and forecast commodity
price movements. The multi-commodity generalization of that model, discussed
in section 3, implies however, that there will be averaging of responses across
commodities so that the use of commodity specific weights may be undesirable.
Since construction of commodity specific indices is also a time consuming activity
it is important to know whether or not it is useful. I examined this question for
the seven non-ferrous metals quoted on the London Metal Exchange (LME) in
Gilbert (1987a).
In that paper I estimated a system of equations of the form
Vlnp jt = .BjO + .BjlVlnpj,t_l -.Bj2(1nfj,t_Clnpj,t_l) -.Bj3VlnXjt (47)
 109

where Pj is the LME spot ('prompt') price for metal j converted to cllb, fj is the
three month forward quotation converted using the three month sterling-dollar
exchange rate for the same day, Xj is the real exchange rate index defined in
(44), possibly using commodity specific weights, r is the US 90 day Treasury Bill
rate, 7r is the US Producer Price Index (all items), and OEIP is the OECD
industrial production index. The operator 'V ' denotes the innovation in the
variable in question so that Vzt=zCEt_1Zt. The innovations for the interest rate
r, US price level 'Ir and the activity variable InOEIP were obtained as the
residuals from first order autoregressions (dependent variables r, .6.1n'lr and
.6.1nOEIP respectively). The commodity price innovations Vlnpj were defined as
the change from the lagged forward price
Vlnpjt = Inpjt -lnfj ,t_1 (j=l, ... ,7)
and the innovation in the exchange rate indices were defined as the weighted sum
of the similarly defined innovations (interpolated where no forward quotation was
available) which, together will the simplifying assumption that changes in the
PPP correction term are all unanticipated, gave
. f]
VlnX't = nE v·· [ Inx·t-lnx. t-1
J i=2 IJ 1 I,
where xf is the forward exchange rate for currency i. The summation is over all
OECD countries. Note that the exchange rate index is defined using the same
convention as (45) so that the predicted dollar commodity price elasticity with
respect to it is unity. The seven metals are silver, aluminium, copper, nickel,
lead, tin and zinc. 3
The motivation of this specification was to investigate both the weak form
efficiency of the LME price formation process, which requires .BjO=.Bj 1=.Bj2 =O,
j=l, ... ,7, and its strong form efficiency, which requires .Bj4 =O, .Bj3' .Bj5' .Bj6 >O,
j=l, ... ,7. The activity innovation VlnOEIP was included to increase the power of
the tests by reducing residual variance. The results of these efficiency tests are
reported in Gilbert (1987a). Here I report some additional results obtained by
estimating equation (47), simplified by setting some small, poorly defined and
insignificant coefficients to zero, using alternative weighting schemes for the
exchange rate indices.

3 The sample consisted of 32 non-overiapping end month price quotations

(end January, April, July and October to avoid end-year complications) - see
Gilbert (1987a). I am grateful to Sompheap Sem for data preparation.
110

I investigated five different weighting scheme, three commodity-specific and

two non-specific. The three commodity-specific schemes were
i) consumption weights wi' appropriate if demand elasticities are much
greater than supply elasticities;
ii) production weights wi' appropriate if supply elasticities are much greater
than demand elasticities;
iii) shares in production plus consumption t(wi+wi ), appropriate if demand
and supply elasticities are equal.
The two non-specific weighting schemes were
iv) 1980 GNP weights; and
v) the IMF MERM exchange rate index, scaled by the factor (I-VI) where
VI is the GNP weight in (iv) so that the predicted elasticity is again
unity. (The MERM index uses US trade-based weights.)

Table 3. Choice of Exchange Rate Index

Long Run Exchange Rate Elasticities and Log-likelihoods
Commodity Specific Indices Non-specific Indices
Consumption Production Average GDP MERM
Shares Shares Shares Weights Weights
Silver -2.79 -3.20 -2.75 -2.52 -3.26
Aluminium -0.30 -0.04 0.00 -0.02 -0.24
Copper -1.05 -1.24 -0.06 -1.04 -1.26
Nickel -0.32 1.12 -1.08 -0.20 -0.14
Lead -1.20 -0.72 0.05 -1.35 -1.37
Tin -1.66 -1.75 -1.03 -1.51 -1.79
Zinc -0.21 -0.10 -1.87 -0.19 -0.21

Average -1.08 -0.85 -0.96 -0.98 -1.18

Log-likelihood -320.1 -321.6 -320.9 -321.1 -321.9
Notes: Sample 1978ql-1985q4. Estimation by modified SUR - see text and
Gilbert (1987a).

The resulting estimated long run elasticities (estimated using a version of the
SUR procedure modified to take into account different numbers of observations
for each metal) are reported in Table 3. The initial question is which choice of
index gives the best fit. Since the five models are non-nested no formal test is
available, but in any case it is doubtful whether any of the available non-nested
tests would be able to discriminate between alternative weighting patterns in a
 111

model such as this in which the overall fit is relatively poor. However, an
impressionistic test is provided by comparison of the model log-likelihoods listed
in the final rows of the table and here it is notable that the three commodity-
specific indices give higher likelihood values than the two non-specific indices.
This indicates that, despite the worries arising from the multi-commodity
analysis of section 3, there is some merit in constructing commodity-specific
exchange rate indices. Overall, the highest likelihood results from use of
consumption weights, and it is apparent from inspection of the estimated
elasticities that some anomalous figures result from the use of production weights
(see in particular the estimate for nickel, where Canada is a,major producer).
Turning to the two non-specific indices, the use of GNP weights gives a higher
likelihood than the model which employs the MERM index.
A second comparison relates to the size of the estimated long run exchange
rate elasticities, predicted to be unity using this choice of index if multi-
commodity considerations can be ignored. The long run elasticities are
distributed fairly widely around this theoretical value: aluminium, nickel and
zinc all show much lower elasticities which is consistent with producer pricing on
a dollar basis (these three markets either are or have been dominated by powerful
producers), while silver, lead and tin are seen as being much more sensitive to
exchange rate changes than predicted by the model. Only copper has an elasticity
close to the theoretical value. Averaging over all seven metals does however give
a value which appears approximately correct.
Comparison of the estimated elasticities with respect to the GNP-weighted
and MERM indices shows that in each case, use of the MERM index gives a
higher response. This is due to the fact that the trade based MERM index gives a
particularly high weight to the Canadian dollar whose value moves more closely
with the US dollar than do the values of other OECD currencies. (The MERM
weight is 12.9% compared with 3.6% in the GNP-weighted index.) The result is
that the MERM index shows a lower variability than the GNP-weighted index
(its coefficient of variation is 10.3% over the sample in this study, compared with
11.6% for the GNP-weighted index), and as a consequence, it is estimated to
have a larger coefficient value. Whatever the merits of the MERM dollar index in
explaining the competitiveness of the United States in international trade, its use
does appear to be misplaced in the analysis of the changes in the prices of
internationally traded goods.
This final conclusion is supported by the estimates of long run exchange
112

rate responses of the World Bank commodity price indices reported in Gilbert
(1989a) where, using a longer sample of quarterly data, it is found that use of the
MERM dollar index gives long run elasticities which are generally outside the
unit interval, whereas the GNP-weighted index gives lower elasticities which are
broadly within the predicted range. I report further evidence on this question in
section 10.

9. CONSEQUENCES OF EXCLUSION OF LDC EXCHANGE RATES

As noted, it is general to relate primary commodity prices to indices of the value
of the dollar relative to other developed country exchange rates. This decision is
the result partly of convenience, but also reflects the fact that PPP corrections
are both more important and more difficult in relation to LDC exchange rates.
However, since LDCs are important producers of many primary commodities,
and indeed some commodities are produced almost entirely in LDCs, exclusion of
this set of exchange rates may bias the estimated commodity price response to
exchange rate changes.
To see this, suppose that countries 1 to m are developed countries (the USA
remaining country 1), and countries m+1 to n are LDCs. We may write equation
(17) as
m n
b.ln(p/7r1) = -}; v.b.lnx'!' - }; v.b.lnx'!'
i=2 1 1 i=m+1 1 1

and writing vi=v/va (i=2, ... ,m) and vi=v/vb (i=m+1, ... ,n) where
a m n
v = }; v. and }; v.
j=2 J j=m+1 J
we obtain
am b n
b.ln(p / 7r1) = - v }; v'!' b.lnx'!' - v }; v'!' b.lnx'!' (48)
i=2 1 1 i=m+1 1 1

= _ va b.lnxa _ vblnxb

where xa is the observed real dollar exchange rate against developed country
currencies and x b is the unobserved counterpart for the LDCs. Note that
va +vb =1-v 1. Now suppose that the investigator follows standard practice and
regresses b.ln(p/ 7r1) simply against the developed country index b.lnx a , perhaps
also including other variables (activity etc.) as controls. In the case of a simple
 113

regression, the estimated coefficient b will be given by

b = _ va _ vb Cov(Lllnxa ,Lllnxb) = _ va _/\ vb (49)
Var (Lllnxa) a

= - (I-vI) - (l\a-1)vb
where I\a is the simple regression coefficient of Lllnxb on Lllnxa. (If additional
control variables are included the same formula will hold but in terms of the
simple regression of the residuals of the regressions of Lllnxb and Lllnxa on the
control variables.)
The implication of equation (49) is that exclusion of LDC exchange ra~es
from the exchange rate index will increase the (negative) commodity price
elasticity with respect to the developed country index if LDC exchange rates
move proportionately more than developed country rates, and decrease the
elasticity if the converse is the case, where the magnitude of the relative
movements is measured by the beta coefficient of the unobserved LDC exchange
rate index. It is obvious that omission of LDC exchange rates has no effect if they
move exactly in line with developed country rates. But the finding, documented
in Table 1, that commodity prices appear to show excess volatility with respect
to changes in the value of the dollar may be explained in terms of the Ridler-
Yandle model of section 2 if LDC exchange rates tend to move in an even more
exaggerated manner than developed country rates.
In Gilbert (1987b) I provided some evidence on this question by considering
real wage rates and effective dollar exchange rates for a group of major Latin
American primary commodity producers. The results are su=arized in Table 4.
The currencies of all five of the countries considered showed a real depreciation
against the dollar in the first half of the eighties and for three of these countries
(Brazil, Chile, Dominican Republic) this depreciation clearly exceeds the
appreciation of the dollar against other developed countries (the MERM dollar
index averaged 143.7 in 1984 on a 1980=100 basis). Data on real wages is less
comprehensively available, but in Colombia and Peru (not Chile) there is
evidence of a sharp fall in real wages accompanying this real depreciation, and
this will have lead to a downward shift in commodity supply curves. This
evidence is therefore consistent with the view that exclusion of LDC exchange
rates from the exchange rate indices used in econometric study of primary
commodity price changes may result in the appearance of excess sensitivity to
114

Table 4. Real Wages and Effective Dollar Exchange Rates for

Major Latin American Primary Commodity Producers
Brazil Chile Colombia Dominican Peru
Republic
Real dollar exchange rates
1972 116.5 92.3 141.1 n.a. 72.1
1976 92.3 129.6 129.5 n.a. 64.4
1980 100.0 100.0 100.0 100.0 100.0
1984 150.8 173.1 119.8 187.7 120.2
Real wages
1972 n.a. 39.3 n.a. n.a. 131.4
1976 n.a. 46.2 80.7 n.a. 126.8
1980 n.a. 100.0 100.0 n.a. 100.0
1984 n.a. 128.1 87.2 n.a. 66.2
Notes: real exchange rates calculated as nominal exchange rates over ratio
of GDP deflator to US GNP deflator; real wage rate indices
calculated as nominal wage rates over consumer price index.
Sources: exchange rates, consumer price indices and GNP /GDP deflators:
IMF, International Financial Statistics (various issues); wage rates:
Chile (wage earners) and Peru (non-agricultural activities): ILO,
Yearbook of Labour Statistics (various issues), Colombia (average of
hot and cold regions): Departament Administrativo Nacional de
Estadistica, Boletin de Estadistica (various issues). I am grateful to
Carmen Dolado for assembling these data.
exchange rates.
In Gilbert (1987b, 1989a) I speculated that these real depreciations against
the dollar may have been related to the LDC debt problem. The major part of
LDC debt, particularly in Latin America, is dollar denominated. Consequently,
the rise in the value of the dollar in the first half of the eighties resulted in a
negative wealth effect on indebted LDCs. This was met in part by partial
reneging on debt service obligations, but to the extent that debt was serviced
there was an increased requirement for foreign exchange which could be met by
increased exports. Currency depreciation may be seen as a mechanism both for
accommodating the adverse wealth effects and for generating the required
additional export earnings. However, as LDCs as a group attempted to increase
export revenues, the result was to induce a fall in real commodity prices in
addition to that already resulting from the recession in the industrialized
countries. I estimated this debt-induced fall to be around 7% over the period
1981-85 (Gilbert, 1987b).
 115

10. ESTIMATED EXCHANGE RATE ELASTICITIES

Hitherto, most empirical analyses which have considered the response of primary
commodity prices to exchange rate changes have either used aggregate exchange
rate indices or have looked only at individual commodities or at a small group of
commodities. Here I attempt to redress this balance by computing exchange rate
elasticities for the entire set of commodities which go up to make the World
Bank 33 Commodity Index.
There are various ways in which this might be done. The procedure I have
adopted here is to specify an unrestricted distributed lag model linking the
commodity price Pi (i= 1, ... ,33) deflated by the US Producer Price Index (all
items) USPP, the PPP-corrected GDP-weighted exchange rate index X
considered in section 7, and (to control for demand effects) the index of OECD
industrial production OEIP. The model was estimated on quarterly data over the
sample 1962ql-1989q3. It appeared generally acceptable to set the length of the
distributed lag to four quarters.
4
In[Pit/UDPP t ] = i.l:0 + i.l:1t + j~l i.l:2}n[Pi,t_/USPPt_j] (50)
4 4
+ E i.l:3 ·lnX t _· + E i.l:4 .lnOEIP t-' (i=1, ... ,33).
j=O J J j=O J J
This equation was estimated for the 33 commodities of the World Bank index
(sugar occurs twice). 4 The long run exchange rate elasticity fl is given by
a3 4 4
(3 =~ wherea2 = E i.l:2 · and a 3 = E i.l:3 .. (51)
a2 j=l J j=O J
The asymptotic variance of the estimated value of fl may be computed as
(Bardsen, 1989)
_ a j?Var(a2) - 2,8Cov(a 2,a3) + Var(a 3)
Var (fl) = -2 . (52)
1 - a2
The major advantage of adopting this non-structural approach is that
although the estimated long run exchange rate elasticities will undoubtedly be
less efficient than those which could be obtained from structural models, the

4 The GDP-weighted exchange rate index was constructed by Sompheap Sem

in 1985 and I have subsequently updated it. I am happy to provide this index to
anyone submitting a floppy disk. I am grateful to Betty Dow for providing me
with the commodity price data, and to Michael Chui who performed the
computations.
116

reduced form approach reduces the possibility of specification error and also of
pre-test bias problems. However, the estimated elasticities may still be in error if
important variables have been excluded from the specification.
The estimated elasticities are given in the first column of Table 5 and the
asymptotic standard errors, computed using (52) in the second column. The third
column gives the t test against the null hypothesis that jJ=O and the final column
the t test that jJ=-1. Discarding the estimate for the ISA sugar price, where the
high elasticity may reflect the residual nature of the free market in sugar, the
average estimated elasticity is -0.89 which is clearly in line with the Ridler-
Yandle model set out in section 2. However, the asymptotic standard errors
indicate that these long run responses are in general poorly determined so that it
is only possible to draw weak inferences. Only seven of the estimated elasticities
are in the range -0.75 to -1, although only seven of the remainder do not differ
significantly from -1. Thirteen of the commodities have negative values in excess
of unity, but in none of these cases is difference significant. Despite this lack of
significance, it is notable in particular that the metals feature particularly
prominently in this group of commodities exhibiting excess sensitivity to
exchange rate changes. These numbers therefore provide weak support for the
view that exchange prices tend to over-react to exchange rate changes.
A further group of commodities exhibit lower elasticities than is compatible
with the Ridler-Yandle model, and in some cases these are even positive. This is
in line with results reported in Jabara and Schwartz (1987) who found systematic
violations of the Law of One Price for a number of agricultural commodities
between the United States and Japan despite near perfect arbitrage for other
commodities. They suggest that controls on trade together with domestic support
schemes may be responsible for these apparent anomalies. In that spirit one
should also note that a number of the commodities for which estimates are given
in Table 5 have been the subject of international commodity agreements (cocoa,
coffee, natural rubber, sugar, tin - see Gilbert, 1987c, for a discussion). In
addition, many of the metals and minerals have been the subject of cartelised or
other quasi-monopolistic producer pricing which has taken the form either of
producers 'stabilizing' exchange prices (copper, lead), of producer pricing
sometimes resulting in a two-tier market with producer prices influencing and
being influenced by exchange prices (aluminium, copper, zinc - see McNicol,
1975, Ghosh et al., 1987, and Slade, 1989) and in some cases outright cartel-
ization (phosphate rock - see Radetzki, 1990). Not all of these actions or arrange-
 117

Table 5. Estimated Exchange Rate Elasticities (sample 1962q2-1989q3)

long run asymptotic

elasticity standard t test t test
(beta) error {J=O {J=-1

1 Cocoa -0.491 1.589 0.309 0.320

2 Coffee -0.117 0.936 0.125 0.943
3 Tea -0.123 0.382 0.322 2.296
4a Sugar -ISA -3.158 *** *** ***
4b Sugar - IBRD index -1.414 0.562 2.516 0.737
5 Oranges 0.193 0.490 0.394 2.435
6 Bananas -0.876 0.424 2.067 0.316
7 Beef -0.180 0.282 0.008 2.908
8 Wheat -0.993 0.640 1.552 0.011
9 Rice -1.491 0.657 2.269 0.747
10 Maize -0.907 0.554 1.637 0.168
11 Sorghum -0.925 0.497 1.861 0.151
12 Coconut oil -0.199 0.354 0.562 2.263
13 Copra -0.599 0.381 1.572 1.052
14 Groundnut oil -0.401 0.608 0.660 0.985
15 Groundnut meal -0.877 0.325 2.698 0.378
16 Palm oil -1.092 0.630 1.733 0.130
17 Soybeans -1.158 0.495 2.339 0.319
18 Soybean meal -1.066 0.389 2.740 0.170
19 Cotton -1.076 0.394 2.731 0.193
20 Jute 0.611 0.713 0.857 2.259
21 Wool -1.507 0.460 3.276 1.192
22 Natural rubber -1.172 0.421 2.784 0.409
23 Tobacco -0.071 0.184 0.386 5.049
24 Logs -0.968 0.261 3.709 0.123
25 Copper -2.331 1.450 1.608 0.918
26 Lead -1.811 0.594 3.049 1.365
27 Tin -3.263 1.769 1.845 1.279
28 Zinc -1.663 0.960 1.732 0.691
29 Bauxite 0.744 0.687 1.083 2.539
30 Iron ore -0.531 0.192 2.766 2.442
31 Phosphate rock -1.246 0.817 1.525 0.301
32 Aluminium ingot -0.855 0.336 2.545 0.432
33 Nickel -1.633 0.546 2.991 1.159
Notes: The wool sample is 1974ql-1989q3. *** indicates that the formula for
the asymptotic variance gave a negative value. Commodity price data
were provided by the World Bank. For definitions, see appendix.

ments have been successful or successful for very long, but while they were in
operation it is far from obvious that a model based on equality of competitive
supply with demand could be expected to apply.
The overall conclusion from these figures must therefore be that although
commodity price behaviour appears broadly consistent with the Ridler-Yandle
118

model, there is evidence both that the prices of a significant group of

commodities appear to respond to exchange rate changes with greater sensitivity
than the theory predicts, while another group exhibits lower elasticities than is to
be expected.
Finally, in Table 6 I report the results of estimating the equation (50) over
the same period using the World Bank 33 Commodity Price Index (133), and also
its three major component subindices: agricultural foods (IAF), agricultural
non-foods (INF) and metals and minerals (IMM). The first panel of the table
uses the PPP -corrected exchange rate adjusted index discussed in section 8 and it
is notable that the aggregate elasticities with respect to this index are all close to
and insignificantly different from unity, with that for the foods index being
slightly in excess of unity. Use of the PPP-corrected MERM index 5 (second panel)
gives very similar results and it is less apparent here that the estimated
elasticities are higher than when the GDP-weighted index is used. On the other
hand, this contrast is clearly evident when the PPP corrections are omitted
(third and fourth panels).

Table 6. Estimated Long Run Exchange Rate Elasticities

(sample 1962ql-1989q3, asymptotic standard errors in parentheses)
Exchange Rate 33 All Non-Food Metals
&
Index Commodities Foods agricultural Minerals
GDP-weighted PPP-corrected --{l.941 -1.028 --{l.891 --{l.879
(0.403) (0.767) (0.352) (0.359)
MERM PPP--corrected --{l.966 --{l.951 --{l.880 --{l.984
(0.538) (0.959) (0.445) (0.369)
GDP-weighted --{l.842 --{l.956 --{l.865 0.797
(0.349) (0.638) (0.280) (0.358)
MERM --{l.956 -1.052 --{l.986 --{l.851
(0.441) (0.805) (0.352) 0.399)

5 I am grateful to Kaija-Leena Rikkonen for extending the MERM index

back to 1960.
 119

11. CONCLUSIONS
It is standard to take primary commodities to be near homogeneous products sold
in near competitive markets. The Law of One Price may therefore be taken to
hold to a high level of approximation. The implication is that adjustment of
dollar commodity prices to changes in dollar exchange rates will be straight-
forward. A simple static partial model, due originally to Ridler and Yandle
(1972), gives a formula which shows that dollar prices will fall (rise) by nearly
the same amount that the dollar rises (falls) in value against other currencies.
The actual elasticity, which must be within the unit interval, will depend on the
US share of production and consumption of the commodity and on supply and
demand elasticities, being closer to unity the less important and the less elastic
are US production and consumption in relation to the rest of the world.
It is possible to propose more complicated models which take into account
substitution between commodities in production and consumption, stockholding
and futures trading. The main contribution of this paper is to argue that these
complications neither change the theoretical results in any significant way nor are
necessary to account for actual historical experience. Despite the simplicity of the
llidler-Yandle model, it does appear to be sufficient to give an adequate account
of the effects of exchange rate changes on primary prices. In particular, there is
no possibility of commodity prices overshooting in response to an exchange rate
shock, although if commodity price risk is imperfectly diversifiable, a lagged
response is likely.
Nevertheless, there has been a tendency for authors to report elasticities of
commodity prices with respect to changes in exchange rates which lie outside the
unit interval. I have suggested that this is likely to be largely a problem of
economic statistics rather than economic theory. There is some weak evidence
that prices from futures exchanges may over-react to exchange rate changes, and
this suggests use of a commodity price index based on physical transactions.
Furthermore, it is essential to take care in the choice or construction of the
exchange rate index. There is some merit in using commodity-specific weights,
but if, as is standard, a common index is adopted, it is important to use weights
which relate to activity rather than trade. In particular, the commonly used
MERM index is inappropriate for this purpose. Furthermore, whichever exchange
rate index is used should be corrected for the effects of differential inflation. But
it is still possible to obtain anomalous results if LDC exchange rates are not
covered in the index and if these move more sharply than other developed
120

country rates against the dollar. There is some evidence that indebtedness
problems have induced exactly this effect.

REFERENCES
Adams, F G and J Vial (1988), "Explaining Recent Metals Price Swings",
Resources Policy, 14, 85-96.
Anderson, R Wand J-P Danthine (1983), "Hedger Diversity in Futures
Markets", Economic Journal, 93, 370-89.
Bardsen, G (1989), "Estimation of Long Run Coefficients in Error Correction
Models", Oxford Bulletin of Economics and Statistics, 51, 345-50.
Beenstock, M (1989), "An Econometric Investi&ation of North-South
Independence", in D Currie and D Vines (eds.), Macroeconomic Inter-
actions Between North and South, Cambridge University Press,
Cambridge.
Bond, M E (1988), "An Econometric Study of Primary Commodity Exports from
Developing Country Regions of the World", IMF Staff Papers, 34,
191-227.
Chambers, R G and R E Just (1979), "A Critique of Exchange Rate Treatment
in Agricultural Trade Models", American Journal of Agricultural
Economics, 61, 249-57.
Cote, A (1987), "The Link Between the US Dollar Real Exchange Rate, Real
Primary Commodity Prices, and the LDCs' Terms of Trade", Review of
Economics and Statistics, 49, 547-51.
Currie, D, D Vines, T Moutos, A Muscatelli and N Vidlis (1988), "North-South
Interactions: A General Equilibrium Framework for the Study of Strategic
Issues", in D Currie and D Vines (eds.), Macroeconomic Interactions
Between North and South, Cambridge University Press, Cambridge.
Deaton, A Sand G Laroque (1989), "On the Behaviour of Commodity Prices",
Princeton University, Research Program In Development Studies,
processed.
Dornbusch, R (1976), "Expectations and Exchange Rate Dynamics", Journal of
Political Economy, 1984, 1161-76.
Dornbusch, R (1985), "Policy and Performance Links Between LDC Debtors and
Industrial Nations", Brookings Papers on Economic Activity, 1985.2,
303-68.
Dornbusch, R (1987), "Exchange Rates and Prices", American Economic Review,
77, 93-106.
Dornbusch, R (1988), "Purchasing Power Parity", in J Eatwell, M Milgate and P
Newman, The New Palgrave Dictionary of Economics, 3, 1075-85,
Macmillan, London.
Fisher, E (1989), "A Model of Exchange Rate Pass Through", Journal of
International Economics, 26, 119-37.
Fleisig, Hand S van Wijnbergen (1985), "Primary Commodity Prices, the
Business Cycle and the Real Exchange Rate of the Dollar", World Bank,
Global Analysis and Projections Division, Discussion Paper 1985-19.
Frankel, J A (1986), "Expectations and Commodity Price Dynamics: the Over-
shooting Model", American Journal of Agricultural Economics, 68,
344-48.
 121

Ghosh, S, C L Gilbert and A J Hughes Hallett (1987), Stabilizing Speculative

Commodity Markets, Oxford University Press, Oxford.
Gilbert, C L (1973), "Buffer Stocks, Exchange Rate Changes and Inflation",
paper presented to World Bank Commodity Conference, Merton College,
processed.
Gilbert, C L (1985a), "Efficient Market Commodity Price Dynamics", World
Bank, Commodity Studies and Projections Division, Working Paper
1985-4.
Gilbert, C L (1985b), "Futures Trading and the Welfare Evaluation of
Commodity Price Stabilisation", Economic Journal, 95, 637-6l.
Gilbert, C L (1987a), "Metals Market Efficiency in Relation to Foreign Exchange
and Financial Markets", World Bank, International Commodity Markets
Division, Working Paper 1987-9.
Gilbert, C L (1987b), "The Impact of Exchange Rates and Developing Country
Debt on Commodity Prices", Oxford University, Institute of Economics
and Statistics, Applied Economics Discussion Paper, 30.
Gilbert, C L (1987c), "International Commodity Agreements: Design and
Performance", World Development, 15, 591-16.
Gilbert, C L (1989a), "The Impact of Exchange Rates and Developing Country
Debt on Commodity Prices", Economic Journal, 99, 773-84.
Gilbert, C L (1989b), "Futures Trading, Storage and Price Stabilization", Review
of Futures Markets, forthcoming.
Gilbert, C Land T B Palaskas (1990), "Modelling Expectations Formation in
Primary Commodity Markets", in D Sapsford and L A Winters (eds.),
Primary Commodity Prices: Economic Models and Policy, Cambridge
University Press, Cambridge.
Giovannini, A (1988), "Exchange Rates and Traded Goods Prices", Journal of
International Economics, 24, 45-68.
Jabara, C Land N E Schwartz (1987), "Flexible Exchange Rates and Commodity
Price Changes: the Case of Japan", American Journal of Agricultural
Economics, 69, 580-90.
Kawai, M (1983), "Price Volatility of Storable Commodities Under Rational
Expectations in Spot and Futures Markets", International Economic
Review, 24, 435-59.
Kofman, P, J-M Viaene and C G de Vries (1989), "Primary Commodity Prices
and Exchange Rate Volatility", Erasmus University Rotterdam, Institute
for Economic Research, Discussion Paper 8912/G.
Kofman, P and J-M Viaene (1990), "Exchange Rates and Storable Prices",
Erasmus University Rotterdam, this volume.
Kravis, I Band R E Lipsey (1987), "Price Behaviour in the Light of Balance of
Payments Theories", Journal of International Economics, 8, 193-246.
McNicol, D L (1975), "The Two-price System in the Copper Industry", BeU
Journal of Economics, 6, 50-73.
Muth, J F (1961), "Rational Expectations and the Theory of Price Movements",
Econometrica, 29, 315-35.
Nermuth, M (1982), Information Structures in Economics, Lecture Notes in
Economics and Mathematical Systems, 196, Springer Verlag, Berlin.
Newbery, D M G and J E Stiglitz (1979), The Theory of Commodity Price
Stabilisation, Oxford university Press, Oxford.
122

Pesaran, M H (1987), The Limits to Rational Expectations, Blackwell, Oxford.

Radetzki, M (1990), A Guide to Primary Commodities in the World Economy,
Blackwell, Oxford.
llidler, D and C A Yandle (1972), "A Simplified Method for Analyzing the
Effects of Exchange Rate Changes on Exports of a Primary Commodity",
IMP Staff Papers, 19, 550-78.
Sachs, J D (1985), "The Dollar and the Policy Mix: 1985", Brookings Papers on
Economic Activity, 1985.1, 117-97.
Slade, ME (1989), "The Two Pricing Systems for Non-ferrous Metals",
Resources Policy, 15, 209-20.
Turnovsky, S J (1983), "The Determination of Spot and Futures Prices with
Storable Commodities", Econometrica, 51, 1363-87.
Van Duyne, C (1979), "The Macroeconomic Effects of Commodity Price
Disruptions in Open Economies", Journal of International Economics, 9,
559-82.
Viaene, J-M (1989), "The Equilibrium Approach to Non-storable Commodity
Prices", Erasmus University Rotterdam, Institute for Economic Research,
Discussion Paper 8913/G.

APPENDIX: COMMODITY PRICE DEFINITIONS

1. Cocoa: ICCO daily price, average New York and London, nearest future
trading months.
2. Coffee: (other mild arabica); 1960-79, average of Salvadoran central standard,
Guatemalan and Mexican prime washed, ex-dock New York for prompt delivery;
1980-1989, ICO indicator price, average New York and Bremen/Hamburg
markets, ex-dock.
3. Tea: London auctions, average price received for all tea.
4a. Sugar: (world, ISA daily price), fob and stowed at greater Caribbean ports, in
bulk.
4b. Sugar (index): IBRD composite price, weighted price for the free market,
EEC, and US domestic contract, using LDCs' export value of sugar to the
respective markets as weights (LDCs exclude Cuba).
5. Oranges: (Mediterranean exporters), naval, trovitas hamlins, EEC indicative
import price, cif Paris.
6. Bananas: (central and south American), first-class quality tropical pack,
importer's price to jobber or processor, fob US ports; 1987-1989 prices estimated
based on average wholesale prices at New York and Chicago.
7. Beef: (US), imported, frozen boneless, 85% visible lean cow meat, fob port of
entry.
8. Wheat: (Canadian); 1960-1972q2, no. 1, northern, basis in store Ft. William-
 123

Port Arthur, export ii; 1970q3-1972q2, Thunder Bay; 1972q3-1973q3, no. 1

Canadian western red spring; 1973q4-1985q1, 13.5%; 1985q2-1989 St. Lawrence
export.
9. Rice: (Thai), 5% broken, white, milled, government standard, fob Bangkok,
export price.
10. Maize: (US) no. 2 yellow, export bid prompt 30 day shipment, barge
delivered to port elevator at Louisiana Gulf.
11. Sorghum: (US), no. 2, milo yellow, export bid prompt or 30 day shipment,
rail delivered to port elevator at Texas Gulf.
12. Coconut oil: (Philippines/Indonesian), bulk, cif Rotterdam.;
13. Copra: (Philippines/Indonesian), bulk, cif n.w. European ports.
14. Groundnut oil: (Nigerian/west African), bulk, cif UK; from 1977q1, any
origin, cif Rotterdam.
15. Groundnut meal: any origin, cif Hamburg, 50%; 1977q4-1984q4, 48% Indian,
cif Rotterdam; 1985ql-1989 Argentine, 48/50%, cif Rotterdam.
16. Palm oil: (Malaysian), 5% bulk, cif Rotterdam.
17. Soybeans: (US), no. 2, bulk, cif Rotterdam.
18. Soybean meal: (US), 44% extraction, cif Rotterdam.
19. Cotton: (Cotton Outlook A Index); 1969-1973, Mexican st. middling 1-1/16
inch (sinaloa/sonora), cif Liverpool; 1974, cif Europe; 1981q3-1989 middling
1-3/32 inch.
20. Jute: (Bangladesh), white D, fob Chittagong/Chalna.
21. Wool: (raw Dominion), crossbred, 56's, clean, cif UK.
22. Natural rubber: (US) RSS no. 1, in bales, spot, New York.
23. Tobacco: (Indian), flue-cured, average export unit value.
24. Logs: (Philippines), Lauan, 3.6 m or more x 60 cm, Tokyo delivered; 1977ql-
1989 (Malaysia), Sabah sq. best quality, Meranti, sale price charged by
importers, Japan.
25. Copper: 1960-1980, LME, electrolytic wire bars, settlement price;
1982-1986q2, high grade; 1986q3-1989 Grade A.
26. Lead: (LME), 99.97% purity, refined, settlement price.
27. Tin: (Malaysian), Straits, minimum 99.85% purity, Penang, ex-smelter,
official settlement price; 1984q3-1989 Kuala Lumpur tin market (KLTM).
28. Zinc: (LME), good ordinary brand, settlement price, UK; 1984q3-1989 high
grade.
29. Bauxite: (US), import reference price, based on imports from Jamaica.
124

30. Iron ore: (Brazilian), 68% purity, cif North Sea ports; 1975-1989, 65% purity;
61.5% Fe content of the iron in moist ore, standard sinter feed, granular 100%
under 5 mm.
31. Phosphate rock: (Moroccan), 75% tpl, fas Casablanca; 1976-1977q1, 73% tpl;
1977q2-1980, 72% tpl; 1981-198970% bpI, contract price.
32. Aluminium: ingots minimum 99.7% purity, transaction price (representative
free market price), EEC duty paid, cif Europe.
33. Nickel: (Canadian) electrolytic cathodes, contract price, fob US duty
included; from 1980 (LME), spot, minimum 99.8% purity, official morning
session price.
 CHAPTER 5

EXCHANGE RATES AND STORABLES PRICES

Paul Kofman and Jean-Marie Viaene

1. INTRODUCTION
Current interest in the effects of devaluation on trade in primary commodities is
dated far back in history. Taussig (1896) already considered the problem of
international bimetallism in raw materials trade, i.e. Indian wheat exports.
According to Taussig " ... the uncertainty and irregularity in the conduct of the
exchanges between gold-standard and silver-standard countries ... " is one of the
presumed evils of flexible exchange rates. Even though Taussig mentions the
aspect of gamble introduced into exporter's decisions, he neglects these micro-
aspects and focuses mainly on the terms-of-trade effects. The discussion sort of
dissipated for some decades until the Great Depression reintroduced the topic.
Many countries depreciated their currencies to counteract hyper-inflation. Where
Taussig more or less coincidentally used an example of raw materials trade, Li
(1935) explicitly acknowledged the importance of the composition of foreign trade
in evaluating the effects of these depreciations. Raw material exporters, according
to Li, were the first to suffer from these actions. Unfortunately, Li gives us
neither analytical nor theoretical underpinning of his observation. However, he
does mention the demoralizing effect of depreciating exchange on importers and
exporters: " ... except when [transactions are] hedged by forward contracts, credit
transactions will be greatly reduced ... ".
After Li some more decades passed until Dominguez (1972) revived the
discussion at a time when the Bretton Woods system was being cracked.
Dominguez highlights the special interaction between time structure,
expectations, and forward/futures markets in primary commodities. Here, the
financial topic of arbitraging the different futures markets based upon exchange
125
L. Phlips (ed.), Commodity, Futures and Financial Markets. 125-152.
© 1991 Kluwer Academic Publishers.
126

rate movements gets all attention. Besides this inter-commodity effect where
only arbitragers are being influenced by flexibility of exchange rates, the intra-
commodity effect is not mentioned. The notion that all agents involved in
commodity trade might be affected by flexible exchange rates first appears in
Baron (1976). Kawai and Zilcha (1986) consequently make use of Baron's results
who does not specify the commodity in question. They observed the special
features of primary commodity trade to make it especially vulnerable to exchange
rate volatility. First of all, producers are often confronted with a price quoted in
foreign currency. Second, even more frequently forward exchange markets do not
exist for the producers' currencies. Finally, futures comrilOdity markets do exist
but are usually situated in a foreign, often the consuming, country.
Unfortunately, Kawai and Zilcha stress the central role of the producer in this
exchange rate versus commodity trade setting, thereby completely neglecting the
interactions with other agents involved in the commodity trade. A similar
approach, adopted by Thompson and Bond (1987), investigated the importance
of forward markets for producers dealing with exchange rate volatility. Jabara
and Schwartz (1987) looked into the importers' problems with flexible exchange
rates. Not the producer but the consumer is their point of interest. Cote (1987)
and Khan and Montiel (1987) more recently developed models stressing the
importance of exchange rate depreciations as a policy instrument to influence
terms-Df-trade of primary commodities exporting countries. Unlike the former
authors they consider the flexibility of exchange rates as an endogenous variable
influencing trade. These studies seem to have completed the circle.
In this chapter, which is a continuation of earlier work by Viaene (1989)
and Kofman et al (1990), we continue to investigate the problem in line with the
approach followed by Kawai and Zilcha. Likewise, we are of the opinion that
primary commodities and exchange rates are strongly interrelated and deserve a
proper evaluation.
Empirical work on the subject has become quite voluminous since
commodity price instability increased dramatically with the collapse of the
Bretton Woods system. We have chosen a few recent papers that have drawn
some attention. Chu and Morrison (1986, p.167) conclude that one of the
dominant sources of commodity price variability over the period 1969-82 has
been the volatility of exchange rates. Especially during periods of excessive
 127

exchange rate volatility!, they state that exchange rate volatility is the one main
source of heavily fluctuating commodity prices. Bond (1987) lends support to the
policy-oriented approach as discussed by Khan and Montiel (1987) by observing
the effectiveness of depreciated exchange as an improved price incentive for
producers. Jabara and Schwartz (1987) find that the exchange rate shocks are
seldom completely passed-through in commodity prices, thereby questioning the
supposed relationship. These authors, however, only tested for trade between the
US and Japan. There is an efficiently working forward exchange market for the
Yen-Dollat trade and this may bias their results. Furthermor~, both countries
have easy access to extensive commodity futures markets. The same goes for the
Australian-US trade in wheat as investigated by Thompson and Bond (1987).
Our chapter is based on some empirical facts that are more or less agreed
upon. We assume that primary commodities are quoted in prices that are foreign
to the producers. Few primary commodities are consumed directly, such that we
choose processing firms as consumers. Forward currency markets do only exist for
a few currencies and primary producers rarely have access to these markets.
Commodity futures markets exist for a number of primary commodities but are
situated outside the producing countries. Producers are therefore unable to
operate effectively on these markets. These stylized facts can be observed for
primary commodities that are produced in developing countries. They hardly
apply to the situation encountered in countries like the US or Australia where
primary commodities are being produced and exported as well. There is, however,
an important difference hiding in the fact that the former group of countries
relies on primary commodities as the main source of export earnings unlike the
latter group. As noted in Kofman et al (1990), the absence of forward and futures
markets prevents these countries from hedging themselves against exchange rate
risk. This easily leads to volatile export earnings, worsening terms-of-trade effects
(c.f. Ghosh et ai, 1987, for a thorough investigation of the importance of primary
commodities in LDCs) and, consequently, a depreciating and volatile currency.
The latter effect makes the currency unfit for forward trading. For these
countries, where volatile exchange rates are assumed to have the most impact,
our model characteristics apply. Since we believe it to be important to under-
stand the role of each participant we will characterize each agent in the following

! The authors remain rather vague about the definition of excessive

volatility.
128

discussion. In each case we make references to recent literature.

1.1 Consumers of Primary Commodities

Jabara and Schwartz (1987), who look at primary commodities trade from a
consumer point of view, are exemplary for the treatment of primaries consumers.
The commodities in question are assumed to be consumed directly since their
empirical research is entirely based upon wholesale prices. Theoretical models like
Turnovsky (1983) and Kawai (1983) regard the consuming part as an ad hoc
simple demand relationship of the form D=a-bP without further refinement or
explanatory behavioural relationship whatsoever. Ueda (1983), although focussing
on the manufacturing industries, discusses the use of primary commodities as
intermediate inputs in these industries for Japan. In his paper he uses inventory-
consumption ratios for raw materials. Exchange rate effects are, despite Japan's
scarcity in primary commodities, not very significant due to a high substitut-
ability of intermediate goods imports. This option of Switching among
commodities is not possible in Gersovitz (1986) who considers agro-industrial
processing industries. These are confined to a specific input. Unlike the
manufacturing intermediates, processing leaves most of the original product
intact. Exchange rate pass-through is therefore not as easily dissipated as a small
proportion of final input prices. More thorough representations of intermediate
processing are to be found in Williams (1987) and Hirshleifer (1988) who
construct an industry-wide model with futures markets. Williams uses the
processor as a tool for questioning the usual assumption of risk aversion in futures
markets. 2 Hirshleifer uses an equilibrium approach following the model by
Anderson and Danthine (1983). These authors explicitly consider the trans-
formation process of primary commodities into final goods where Hirshleifer
actually considers two types of consumers, final and intermediate. Pityfully,
these are one country models and demand for the processed commodity is treated
as constant. 3 In our model we have chosen the latter type of intermediate
consumer, situated in a country foreign to the producer. Our consumer is
fortunate to have a commodity futures market at his disposal. He also has the
possibility to store the primary commodity. The model neglects direct final

His approach is similar to a single agent model.

2
Anderson and Danthine investigate random final output prices, but intro-
3
duce a final good futures price next to an intermediate good futures price. In our
opiuion, the former usually does not exist.
 129

consumption since we assume that most primary commodities have to be

processed even if only marginally in some cases.

1.2 Producers of Primary Commodities

Producers of primary commodities are assumed to be faced with foreign prices of
their goods. Since production decisions usually take time to realize and input
costs are incurred in domestic currency, a producer runs exchange rate risk.
When the outcome of the production process itself is uncertain, producers also
face output risk. Optimally, a full set of markets (spot and future) enables
producers to hedge against uncertain export earnings. Such an optimal situation
has been described in Kawai and Zilcha (1986) and Kofman et al (1990). Previous
studies neglected the international setting and considered producers only in a
framework where they could completely hedge their positions on a futures
market. A typical result from both studies as emphasized by Danthine (1978) and
Feder et al (1980) is the separation theorem. With the option of futures trading
(and/or forward trading), the production decision is solely determined by the
prevailing futures price and current input cost. Aspects of the subjective
distribution of spot prices and producer's risk preference are completely shifted to
the futures and/or forward position taken. As long as there is no output
uncertainty, these results are correct. Anderson and Danthine (1983) and Marcus
and Modest (1984) consider the stochastic output case which severely complicates
the outcomes. A further complication is introduced by Karp (1988). Here,
dynamic hedging is considered as an optimal way to deal with stochastic output.
In reality it hinges upon the nature of a commodity futures market. 4 Since more
information, as to the outcome of the production process, becomes available over
contract maturity, producers might want to change their positions to avoid
excessive open positions at delivery. The resulting optimal dynamic hedges,
however, are based on diffusion processes and can hardly be integrated in a
market model like ours. We make use of the fact that producers are, most of the
time, not their own merchants. Instead they are being offered a guaranteed price
by a board of traders. This isolates producers from foreign exchange risk.

4 Veljanovski (1986) gives a good overview of the essential differences

between futures and forward markets. One of these is the possibility of
continuous hedging.
130

1.3 Traders in Primary Commodities

The Nigerian Cocoa Board is a typical example of a trading organization
gathering, stockpiling and merchandising production while offering fixed prices to
producers. These so-called marketing boards usually aim at several stabilization
targets, be it producer price, producer income, regional income or export
stabilization. Helleiner (1967) dealt with these domestic stabilization targets for
several Nigerian marketing boards and concluded that some preconditions like
orderly marketing and reduction in speculative activities have to be fulfilled to
achieve these targets. Dealing with storage in terms of buffer stocks has been
discussed several times. Keynes (1930) refers to the' benefits of valorization
schemes for price stabilization but also mentions possible dangers in the case of
countries that are completely dependent on one primary commodity. McKinnon
(1967) is far less enthusiastic and concludes that intervening boards should better
rape the benefits of futures trading than fix on spot price stabilization. Similar
doubts (c.f. Newbery and Stiglitz, 1981) about consumer buffer stocks have been
ventured in the past few years and the disappearance of the Tin Board is a good
example of such a failure. Ghosh et al (1987) introduce policy rules when
analyzing these consumer boards. They conclude that these rules are only
beneficial for producers if supply restrictions drive up average prices. Another
market intervention is the price support rule. Together with the supply
restrictions, these measures intervene in the production process whereas buffer
funds only intervene in the trading process. Shonkwiler and Maddala (1985) step
in the expectations formation process by introducing an effective lower bound on
prices. Just (1990) considers a whole set of policy measures creating an artificial
institutional framework. The expectation approach is more fully discussed in
Gilbert and Palaskas (1990). They argue that storage plays an essential role in
explaining the expectations process. Whereas they concentrate on producer
storage, we have shifted the storage process to the marketing board. In exchange
for offering a guaranteed price to producers, commodities are being transferred to
the board just after 'harvest'. This stockpiling takes into regard the possibility of
futures trading. The latter is important in view of the above mentioned critique
that countries producing primary commodities are usually isolated from futures
markets. The costs incurred when holding inventories play an essential role in
futures price formation according to Working (1942). This is somewhat in contra-
diction with the view that futures prices reflect expected spot prices (inclusive a
risk premium hedgers have to pay to speculators). We are of the opinion that a
 131

mixture of these views might be most appropriate. 5 Concluding, we opted for a

marketing board offering a guaranteed price while covering itself against
exchange rate and price risk taken over from producers. Our marketing board has
access to a forward 6 and a futures market and also has the possibility to store.

1.4 Speculators and Arbitragers

These agents have been most extensively discussed in the past. Keynes (1930),
being an active speculator himself, considered aspects of risk premia in
commodity markets. Another author, having had a significant influence on
further treatment of speculators, is Working (1977). More recently Newbery and
Stiglitz (1981) and Stein (1986) paid extensive attention to the speculators'
impact on price stability. The already discussed risk premium (as a compensation
for taking over risk) is crucial in these studies. The expectational part of this
premium has also been considered by Peck (1976), assuming adaptive
expectations, Bray (1981), assuming rational expectations, and Stein (1986)
assuming Bayesian expectations. The latter takes due notice of the fact that
there are informed and uninformed traders on futures markets. In this chapter we
will not further discuss these topics of efficient markets since they contribute
little to our analysis. The same applies with respect to the already mentioned
volatility discussion.? Important for our chapter is the remark by Kamara (1982)
that hedger's positions are motivated partially by stability desires and partially
by desires to increase expected profits. This leads necessarily to a mixture of
hedging and speculation. Our model makes use of this notion. We do not
introduce separate commodity speculators since there are already two sides of the
market, namely the marketing board on the one side and the processing consumer
on the other. Both have a mixture of hedging and speculation. There is no need
to complicate matters by adding more agents than strictly necessary. On the
foreign exchange market, we do use speculators. These speculators are treated

5 An interesting discussion confirming our view is Wright and Williams

(1989).
6 This is a violation of our earlier assumption of absence of a forward
currency contract for primary commodity producers. Nevertheless, when the
marketing board is a government institution it has better negotiating power on
the currency exchange market than plain producers.
7 A review of these topics is given in Labys (1980) and Kanbur (1984).
132

according to a portfolio model. This is an improvement, since we are now able to

take into account the important property of covered interest parity. A similar
arbitrage condition will be shown to exist on the commodity futures markets.
This is necessary to complete the market since there are no natural counter-
parties as in the commodity case.
The references made in the discussion of agents are far from exhaustive and
should only be considered as an indication of main-stream thinking in our kind of
models.
Our model, which considers two dates, four markets, one primary
commodity and four agents neglects the factor markets: There is, however, scope
to elaborate on these markets since we introduce explicitly factor inputs in the
producer's production function. An interesting topic for further research would be
the feedback effect of volatile exchange rates in case these inputs are being
imported from the foreign processing country (c.f. the agricultural chemical
inputs and high yielding varieties).
The remainder of the chapter is organized as follows. Section 2 considers
the behaviour of each agent separately. It is shown that the values for their
decision variables, obtained under optimization, depend on a set of (expected)
prices. Their mutual behaviour, as outlined in section 3, clears each market and
gives equilibrium values for this set. A graphical exposition explores the causality
of our model. Section 4 elaborates on the ways the model is influenced by the
international trade context. Analytically as well as graphically a number of
random as well as policy shocks are thus investigated.

2. SPECIFICATION OF AGENTS' OPTIMIZING BEHAVIOUR

Four agents, as introduced in section 1, operate in our model. Each agent
maximizes expected utility. Decision making is assumed to take place at time t,
one period in advance of realization time t+ 1. In the following sections, a time
index will only be added to variables that are contemporaneous to decision time
t. A tilde above a variable indicates randomness.

2.1 Commodity Processor

There are nc identical foreign consumers processing primary commodity X. They
produce subject to a quadratic cost function and face a competitive output price
p. At time t, a representative processor decides upon his production level for time
t+l but is confronted with an uncertain commodity input price as well as a
 133

stochastic demand component. The processor uses primary commodity inputs in

a fixed proportion m to the planned output level. This primary commodity is
quoted in (the processor's) foreign currency. Moreover, since it is storable, the
processor can order its input in two ways. To smooth processing he can either
take the input into storage, one period in advance, or purchase the input forward
on the commodity futures market. The cost function for storage assumes as in
Kawai (1983) that inventory holding cost can be expressed as the cost of
deviating from a desired inventory level. The processor's profit function is
specified as

Vc = pX -!X2 - mqX + Kct(q-f£) + Ict(q-rqt) - ~1~Clci (1)

where
p = perfectly competitive price of output
X = output
m = input propensity of primary commodity to output
q = spot price of primary commodity at time HI (in foreign currency)
Kct = forward purchases of primary commodity
f£ = futures price at time t for delivery of primary commodity at t+l
I ct = inventory position
r = one plus the foreign interest rate (opportunity cost)
h = storage cost coefficient
Processors maximize the expected utility of profits
EU c = J~Uc(Vc(q»~'(q)dq, (2)
where the utility function Uc is characterized by the measure of absolute risk
aversion (-U ' I lU' )=oc and Ie is the subjective probability density function of q.
Since its first two moments are assumed to exist, we can write E(q)=q and
Var(q)=CT~. By a second order Taylor expansion of Uc at q, we obtain the
following results for the processors' decision variables
X = p - mn1. = (p--mq) + mRq + mAq , (3)

Ict = I ct + -il ' (4)

(5)
134

where Rq=q-f( is the futures market risk premium and Aq=f}-rqt is the
convenience yield of holding the commodity in storage. 8 9 The production decision
in equation (3) is made separately of the inventory and futures trading decisions
according to the separation theorem. Equations (4) and (5) show that storage and
futures operations depend on hedging as well as on speculative purposes.
Nevertheless, the optimal futures decision in equation (5) shows that the
production hedge is shifted entirely to the inventory decision.

2.2 Producer
Our model assumes np identical domestic producers of primary commodity Q. A
representative producer makes his production plans one period in advance and
faces at decision time a guaranteed price in domestic currency, q*. Therefore the
producer has to cope only with output risk. His linear production function, which
includes an additive production risk (e.g., due to weather influences, diseases,
labour strikes, etc.) reads
Q= y + 71, (6)
where
Q = primary commodity output per producer
y = production inputs
71 = random risk component (regardless of production scale)
Thus, the producer's profit function at time t+ 1 is
Vp = q*Q _wY_~Q2, (7)
where w indicates the input cost next to a quadratic production cost. Assuming
maximization of expected utility of profits

8 Under the assumption of rational expectations (i.e. q=f() this term is

equivalent to the arbitrage condition in Wright and Williams (1982, p.597). We

believe our Aq to give a better interpretation of arbitrage since it consists of two
contemporaneous prices. The arbitraging processor can make a comparison
between selling spot and investing the proceeds at rate r or selling forward.
9 An alternative specification is to make use of Ict=mX or desired inventory

level equals input requirements. Though this would have no qualitative

implications for the equilibrium prices, expressions for the latter would become
cumbersome.
 135

(8)

where the utility function Up is characterized by absolute risk aversion a p and ~

as the subjective probability density function of 0, we can write E(O)=O=O and
Yar(0)=u2 on the assumption that its first two moments exist. Using the second
order Taylor expansion of Up at 0 and maximizing with respect to the input
decision variable leads to
Q=y=q*-~. (9)
l+a p u
Whereas production plans react positively to prices and variance of production
risk,10 it is negatively related to input prices.

2.3 Marketing Board

When considering price supports, one usually assumes that dealers or a ministry
act as a marketing board for the commodity in question. They propose a
guaranteed price which corresponds to their best expectation of the future price
divided by the exchange rate. This currency conversion takes place since the
guaranteed price is quoted in domestic currency while the world trade price is in
foreign currency. Thus, the marketing board faces two types of risk: the exchange
rate risk and the commodity price risk since it expects a production equal to npQ
and a random price in foreign currency, ii. To lessen the riskiness ofits position,
the board has several possibilities to cover itself against the different risks. It can
enter into
• a forward contract to sell a specified quantity of commodity futures for
delivery at time HI,
• a forward contract to cover its foreign currency transaction at the future
date, and
• a possibility to store the commodity.
The hedges need not necessarily be complete, since there is also room for
speculative activity by the board. The marketing board's profit function at the
future date reads as follows:

10 This positive relationship is due to our assumption of independence

between production risk and production scale.
136

n
Vm = (q--q*e)Q~ + (e-ft)H mt - Kmt(q-i£) + Imt(q-rqt) - ~Imt-Iml,
m
(10)
where
np,nm = number of producers and marketing boards respectively
e = spot exchange rate at time t+ 1 expressed as the foreign price of
domestic currency (English definition)
ft = forward exchange rate at time t for delivery of currency at t+1
Hmt = forward sales of foreign currency at time t
Kmt = forward sales of primary commodity
Imt = inventory position.
Note that the board is interested in its profits in foreign currency since it incurs
no cost in domestic currency. We chose to use the British definition of the
exchange rate (the amount of foreign currency one gets for one unit of domestic
currency) to avoid the product of two random variables. Assume that the board
maximizes its expected utility of profits, i.e.,
(11)
where Um is the utility function characterized by a measure am of absolute risk
aversion and" is the joint subjective probability density function of q and e.
Assuming the existence of the first two moments of the latter variables, we can
write E(q)=q, E(e)=e. Var(q)=CT~, Var(e)=CT~ and E(q-q)(e-e)=Cov(q,e)=
PCTqCTe as in the processor's case. After approximating Um by a second order
Taylor expansion at (q,e), the board maximizes with respect to its decision
variables. This results in
n ] pR
Kmt = [ Q~ + Imt + ( 1 2) (12)
m amCTeCTq -P

(13)

(14)
 137

where R=e-ft is the forward market risk premium. From these equations, it can
easily be seen that the board's position is a mixture of hedges and speCUlative
positions. Whereas the latter are incorporated in the Rand Rq terms, the former
are reflected in the Q and I terms. On the futures market, the board covers its
expected primary commodity holdings. On the forward exchange market these
holdings are covered in terms of currency.

2.4 Speculator in Currency

There are ns speculators on the foreign exchange markets. !'- representative
speculator in the producing country has control of some initial wealth Vst which
he likes to allocate in a portfolio with domestic and foreign assets. Given that he
has no cost in domestic currency, his wealth is expressed in foreign currency. His
problem is to decide on the optimal holding of domestic assets with return i-I,
foreign assets with return r-l and the optimal amount of forward sales of foreign
currency in order to maximize his one-period expected utility of his end-of-period
wealth Vs. Next to exchange rate uncertainty the speculator also faces a specific
country risk for the homeland. This country risk derives mainly from the
possibility of capital controls in the domestic country, the possibility of political
risk at home and of a domestic private default risk. The budget constraints faced
by this specific speculator are:
Vst = F st + etD st ' (15)

Vs = rtF st + (e-c)itD st + (e-ft)Hst ' (16)

where
Fst = foreign assets
Dst = domestic assets
Hst = forward purchase of domestic currency
it = one plus the domestic interest rate.
In equation (16) the specific country risk at home results in a negative premium
c on domestic investments. Assume that speculators maximize their expected
utility of future wealth
EU s = J~J~ Us(Vs(e,c»,{(e,c)de de, (17)
where Us is characterized by the measure as of absolute risk aversion and ,{ is
the subjective joint probability density function of e and e. Assume that the first
moments exist. Thus, E(e)=e, E(e)=c and Var(e)=u; respectively Var(e)=.p2
138

and E(e-e)(c-c)=Cov(e,c)=p*u;~2. To simplify matters, assume that Cov(e,c)

=0. 11 Furthermore, it is assumed that speculators and marketing boards have the
same marginal probability density function for the exchange rate. Approximate
EU s ' as before, by a second order Taylor expansion around (e,c). Maximizing
with respect to the portfolio decision variables results in
D _ A-c (18)
st - ---:;:2j
as'l' t

R A-c (19)
Hst =~--:2.
asu e as~
where A=fCrtet/i t is the deviation from covered interest parity. As expected,

domestic asset demand is an increasing function of A and a decreasing function of

c. Since his monetary preference is in foreign currency, the speculator will hedge
an amount iDs on the forward market. The remaining part is of speculative
character.

3. SOLVING THE MODEL IN EQUILIBRIUM

In solving the model as outlined in section 2, we made use of the approaches by
Viaene (1989) and Kofman et al (1990). A graphical approach will link the
markets in a general equilibrium framework after a market-by-market analysis of
the equilibrium conditions. Four markets are distinguished, spot and forward
currency and spot and futures commodity.

3.1 Currency Spot Clearing

Market clearing on the spot currency market can be expressed as follows:
n sSt _ 1 = nsDst + npqi_l (It (20)
where St-l =(Hs +rD s)t_l is the spot sale of domestic currency, induced by the
former period's speculative decision. Since the speculator has foreign currency

11 We chose to focus on the relationship between the exchange rate and

commodity price and thus on Cov(e,1i.). Nevertheless, a Cov(e,c)<O seems
plausible since when the currency is depreciating (el) capital controls are levied
such that premiums increase (cn.
 139

preferences, he will reimburse his receipts in domestic currency into foreign

currency. Spot demand for domestic currency consists of the domestic asset term
plus a production uncertainty term. At spot time (=t), production uncertainty
disappears and the true value of IIt =Qt-Qt-l becomes known. Since we value
this excess/shortage term at the price guarantee (at production decision time),
the latter term shows net domestic currency demand for IIt>O and net supply for
lit <0. Making use of equation (18) leads to:

A = c + [St-l - qt_ll1tnp/nsha;¢ . (21)

In equation (21) we only have to know ft to solve for et , the equilibrium spot
exchange rate, without any further interaction with the rest of the model. This
equation has a simple graphical representation which is given in Figure 1. The
45 0 line mirrors the solution for ft on the horizontal axis. The second 45 0 line is
used to get a solution for rtet/i t such that the difference between the first and
the second projection represents (fr-rtet/it)=At. The second 450 line has an
intercept which indicates the deviation from covered interest parity, formally
represented by the right hand side of (21). Here, covered interest parity fails to
apply due, partly, to the country risk factor c and to its conditional variance ¢2.
An unanticipated positive supply shock (IIt>O) at the current period contributes
to a lowering of At, ft given, and hence, to an appreciation of the spot exchange
rate.

3.2 Commodity Spot Clearing

For the commodity markets a similar approach can be followed. The market
clearing condition for the spot period is given:
n
nc(K c-mX+I c )t_l -nm(Km-Q~-Im)t_l + npllt = ncIct + nmImt'
m
(22)
where the first two terms represent the carry-over of last periods' speculative
components of processors and marketing boards, respectively supply of and
demand for the commodity. The third term on the left hand side shows the
volume of production being supplied on the spot market after the output
uncertainty is resolved. The two inventory positions represent spot commodity
demand. Substituting equations (4) and (14) into equation (22) leads to the
following expression:
140

o
45
o
45

Figure 1. Spot currency equilibrium

lt 45
o

Figure 2. Spot commodity equilibrium

 141

where the S's replace the speculative carry-over expressions. In equation (23) we
only have to know if
to solve for qt. Like the spot exchange market, the
equilibrium commodity spot price is determined using the futures price as input
and its solution is not instrumental in solving for the other variables of the
model. A graphical representation can be constructed similar to the one depicted
if
in the previous section. In Figure 2 the 45 0 line mirrors the solution for on the
horizontal axis and the second 45 0 line provides a solution in rqt: The intercept is
proportional to the state of excess supply on the market as indicated by the right
hand side of (23). Without loss of generality, the graph makes use of a positive
intercept. An unanticipated positive supply shock (Ot>O) at the current period
drives the basis risk Aqt upward and the spot price of the commodity declines
(for if given).
3.3 Currency Forward Clearing
Having discussed the two spot markets we are now left with the solutions on the
forward/futures markets. The latter are causal in this model. The currency
forward market is cleared according to:
nmHmt + nsH st = 0 . (24)
Taking the optimal forward currency decisions of the marketing board in
equation (13) and the speculator in equation (19) leads to the forward market
risk premium
(25)
where

Il2 = pnmosO'e/O'q[nmos + ns om (1-p2)] ~ 0, dependent on sign of p.

For e given, forward rate ft is known. With pjO this risk premium is determined
by a net supply of forward currency and the futures commodity risk premium R q.
The sign of p determines whether R and Rq are positively or negatively related.
When there is no correlation between the exchange rate and commodity price
(i.e. p=O), the forward risk premium only depends on variance of the exchange
142

rate and net supply of forward currency. The e-curve in Figure 3 (related to 01)
has been constructed with this information: it represents all combinations of f
and fq which sustain an equilibrium on the forward market. The curve is drawn
under the assumption of a negative p and of a positive intercept, or net supply of
forward domestic currency. Having Q explicit in (25) is useful in assessing the
effect of a change in the supply of storables on the forward prices ft. What this
expression indicates is that for Rq=O, and increase in Q and, as such, an increase
in future returns in domestic currency always leads to an unexpected appreciation
in forward rate ft so as to maintain the exchange marke,t in equilibrium. This is a
recomforting outcome.

q q

Figure 3. Forward/futures market equilibrium

 143

3.4 Commodity Futures Clearing

The market clearing condition for the commodity futures market consists of the
aggregated positions of processing firms and marketing boards:
ncKct - nmKmt = °.
Making use of equations (3), (5) and (12), we obtain
(26)

R = 01[nCm(p-mq)-npQ-ncSc(t-1)-npOt+nmSm(t-1)] + 02Rq' (27)

where
dependent on the sign of p,

02 = [(ncam(1-p2)+nmac)/cVl~ + ncm2am(1-i)]o-eO"q/nmP ~ 0,
dependent on the sign of p.
For q given, futures rate ifis known. Equation (27) is quite similar to equation
(25) in connecting the forward premium to the futures premium. The crux, once
more, is embodied in the sign of p. The q-curve in Figure 3 (related to 01) is
drawn under the assumption of a negative p. It represents all combinations of ft
and if which sustain an equilibrium on the commodity futures market. Its slope,
02' is always steeper than the slope of the e-curve, II2 for all values of p, except
for Ipi =1 in which case the curves run parallel and the model fails to provide a
solution. As a matter of fact, 1°21> I II21 is the stability condition of the model.

4. THE LINK BETWEEN EXCHANGE RATES AND COMMODITY PRICES

In the following sections, several important topics will be discussed, all relating
to the interaction between exchange rates and commodity prices. The crucial
importance of the correlation coefficient p between exchange rates and
commodity prices will first be covered since it is the main determinant of the sign
of the effects caused by shocks and changing variance. The latter disturbances
stress the exchange rate framework within which primary commodities are being
traded.

4.1 The Correlation Coefficient

Figure 3 does not lend itself to the immediate interpretation of comparative
statics since the slopes of the e- and q-curves are functions of p whose sign is
analytically unknown. When the correlation coefficient is zero, the two markets
are independent of each other and the e-curve runs horizontal and the q-curve
144

runs vertical. When a positive p is assumed, both curves are positively sloped,
when a negative p is assumed both curves are ne~atively sloped. Ex post, as
indicated by Table 1, neither a positive p nor a negative p can be ruled out. Table
1 splits our sample into three sub-periods: a period of fixed exchange rates and of

Table 1. Ex Post Correlation Coefficients

between Exchange Rates and Primary Commodity Prices

Primary 1957.1-1971.1V 1972.I-1979.IV 1980.l-1988.II

Commodity
Bauxite 0.509 a -0.892 0.845 b
(Guyana)
Beef n.a. -0.589 0.105 e
(Argentina)
Butter 0.357 -0.577 0.518
(New Zealand)
Coal 0.660d -0.657 0.700
SAustralia)
ocoa Beans n.a. -0.563 0.412 e
~razil)
otton 0.343 -0.126 n.a
(Egypt)
Iron Ore n.a. -0.184 0.838
(Brazil)
Manganese 0.767£ -0.712 0.755
(India)
Newsprint 0.057g 0.139 0.822h
(Finland)
Petroleum -O.184i 0.535 0.781i
(Venezuela)
Plywood -O.579 k -0.288 -0.308
(Philippines)
Rice 0.482 0.660 0.818
(Thailand)
Rubber -0.130 0.796 0.567 c
(Malaysia)
Tea n.a. -O.657!,? -O.10lJ
(Australia)
Zinc n.a. n.a. -O.238 m
(Bolivia)

Data Source: LF.S.

a1969.1 b1986.IV c1987.II dI966.II1 e1987.1V £1960.1 g1965.1
h1987.1 i1960.l i1988.I k1963.1 11973.1 m1982.1V
n.a. = not available
 145

convertible dollars 1957.1-1971.1V, a period of flexible exchange rates with a

depreciating dollar 1972.1-1979.1V, and a period of tight US monetary policy
with first appreciated and then depreciating dollar. The table indicates that a
single correlation coefficient can change sign and magnitude over time: see, for
example, bauxite, coal and manganese. Moreover, there is some tendency for
correlation signs to be negative across commodities in the second period and
positive across commodities in the third period. Rather than analyzing the factors
determining the sign of the correlation coefficients we focus here on its
implication for the price of a single primary commodity. The technique adopted
is to evaluate a change in either the commodity supply ot the commodity
demand. This indicates the direction of the shift in either the world supply curve
or the world demand curve and the direction of the commodity price changes
needed to reestablish equilibrium. From (25) and (27) the reduced form in Rq
and the following comparative results are readily obtained (Viaene, 1989):

8(q-i{) n (fleIl I q*)

--- = p 50 dependent on the sign of p (29a)
8Q fl 2-Il 2 '

8(q-i{) = -nc mfl l < 0 . (29b)

&p fl 2-Il 2

We know that Il 1>0, sign Il2= sign fll = sign fl2= sign p, and 1fl21 > 1Il21·
Hence, the effects on i{ resulting from either a demand (29b) or a supply (29a)
shock are not symmetric. In (29b) a rise in the demand for the primary
commodity resulting from a rise in the price (and the supply) of final goods
definitely leads to a higher futures price whatever the value of p. In contrast
(29a) predicts an outcome of the supply shock which depends upon p. A necessary
zero, the condition is always satisfied but for high positive values of p condition
(30) becomes binding and the likelihood of a perverse price response to a supply
shock increases. Note that the highest positive correlation coefficient has been
found for tin for the sub-period 1980.l-1988.II, a market which collapsed in 1985!
Note that Ghosh et al (1987, p.180) mention the sharp rise in Brazilian tin
production as a fundamental cause for this collapse. 12

12 The measures in Table 1 give a good diagnostic for the market in question
but their power in predicting collapses is somewhat limited in that they are ex
146

4.2 Exchange Rate Elasticity

Equation (27) can be presented in a slightly different way by not substituting (3)
for X:

where

112 = [(ncam(l-i)+nmac)CTeCTq/nmacCT~pl ~ 0, dependent on the sign of p,

and where III has already been defined. It is an equation which lends itself to
estimation by specifying a relationship between the cO,mmodity price, and the
exchange rate, the various expectation terms, the demand for and the supply of
the primary commodity (the positive and negative signs in the square brackets,
respectively). The output effect is in conformity with what we might have
expected, in that, for R given, an increase in commodity supply leads, ceteris
paribus, to a drop in futures prices. From (27') we can immediately read off the
response of the commodity price with respect to a change in the exchange rate as
given by,

1/11 2= ~ /[~~ (I-i) + 1] .

CTe nm am
(28')

After suitable normalization of the quantities such that q=e=l, 1/11 2 can be
interpreted as an elasticity. It quantifies the popular observation that " ... when
the dollar appreciates, the price of gold rises ... ", or in terms of our model, a
negative elasticity. Where this elasticity is negative with a negative correlation,
it will become positive when the correlation coefficient is positive. The absolute
value of (28') is positively related to Ip I and CTq/ CTe and negatively related to
nc/nm and a c/ am'

4.3 An Expected Depreciation of the Currency

The initial position of our model at 01 in Figure 3 is one of no excess supply on
either market. This is sufficient to guarantee that the expected spot prices that
will prevail at contract execution equal the forward rates: R=Rq=O. Because the
marketing boards precommitted themselves by offering a contract price q* to

post correlation coefficients. In the case of tin, we recomputed the correlation

coefficient for the period 1980.I-1985.III, just prior to collapse, and found a value
for p=0.81O which is quite high.
 147

producers, an expected depreciation in the exchange rate e which occurs in the

time interval between the production decision and the crop leads to a differen-
tiated response between the short and the long run. The short run response is
obtained by the partial derivatives in (25) and (27):

:Ie :Iq = -1, =-1.

The outcome in the short run (0 2 in Figure 3) is an equiproportional decrease in

the forward rate which results from equiproportional shifts of the e- and q-curves.
The futures price is unaffected since production plans are already determined. In
the long run, the next production decision period (t+I), the marketing boards
who previously offered q* will pass the exchange rate depreciation on to the
producers by revising their price guarantee upward to the commodity producers.
The extent of this revision depends on the competitive nature of marketing board
activities, and hence, on their willingness to, partially or totally, get rid of the
windfall profits on their operations. As a result, world commodity supply is
expected to be higher while demanding firms, facing no exchange risk, will
maintain their orders constant. The shifts of the e- and q-curves according to (25)
and (27) produce the new long run equilibrium point 03, This point is
characterized by a decrease in the new futures price declared at period (t+I), a
subsequent decrease in the spot commodity price at (HI) and by a rise in the
spot and forward rates relative to O2,

4.4 Decreasing Exchange Rate Volatility

Consider the joint announcement at period (t) by central banks that the inter-
national monetary system will revert to fixed exchange rates at period (t+I).
Following this new information, agents will adjust their conditional variance of
the exchange rate formed at period (t) towards zero. Taking the limit ((1e-+0) is
thus necessary to assess this effect. When this is done in the forward currency
market we get a very simple expression:
lim lim (. *-) lim
R = (12-+0 III ItnsDs-npq Q + (12-+0 II 2Rq ,
(12-+0
e e e
leads to

In Figure 4, which portrays what happens when volatility of the exchange rate
148

f 4a. excess supply on currency market

~~----------~~----~.-
---t-=~~Le

q
f 4b. excess supply on commodity market

4c. excess supply on both markets

f

~~----~~~----~~
-------t--~~~ __ e

q
Figure 4. Fixed versus £lexi ble exchange rates
 149

declines to zero, the e-curve turns horizontal. When we take the same limit of the
futures market equilibrium condition, we get
(iH'{) = -03[n cm(p-mq) - npQ - n cSc(t-l) - npOt + n mSm (t-l)),
where

Note that 03>01/02' The q-curve becomes vertical in Figure 4 and the new
equilibrium futures price deviates from its expected price by a fraction 03 of the
state of excess demand. The flexible exchange rate equilibrium,is given by point
01 in Figure 4 compared with point 02 which indicates what would happen if the
exchange rate were fixed instead. No general proposition can be derived because
it all depends on the starting conditions, that is, whether the state of the markets
are such that the forward/futures prices deviate initially from their respective
expectations. For the purpose of illustration Panel 4a is characterized by an
excess supply on the forward currency market only, Panel 4b by an excess supply
of futures only, and Panel 4c by excess supply on both markets. Panels 4a and 4c
commonly predict a rise in ft and a drop in f{; Panel 4b predicts a drop in ft and

a f{ effect which is not clear. These are three examples out of 18 possible
combinations, 9 cases per sign of p. However, it is easy to verify that effects of
excess demands are symmetrical to the excess supply ones.

5. SUMMARY
Sections 3 and 4 give an extensive analysis of a primary commodity model taking
into account the fact that the primary commodity is storable, quoted in the
demanding country's currency and being traded internationally. Section 3 gives
the clearing conditions for currency and commodity markets, spot as well as
forward. Several important propositions can be made.
First, on the currency spot market we notice that covered interest parity is
not being satisfied, partly due to the inclusion of country risk.
Second, the currency spot market is only related to the forward exchange
rate and its solution has no feedback with the rest of the model. A similar
conclusion can be drawn for the spot commodity market where the only link with
the rest of the model consists of the feeded commodity futures price.
Third, the model stresses the purely speculative nature of futures trading
150

by the demanding firm, see equation (5). The hedging motive is completely
shifted towards the inventory decision.
Fourth, the forward/futures markets clearing conditions prove the
importance of the correlation between exchange rates and commodity prices.
Section 4.1 gives a full discussion of the correlation coefficient and its impact on
futures prices when there is a commodity supply or a commodity demand shock.
Whereas the rise in demand gives the usual result (Le., a rise in the futures
price), the rise in supply gives a similar result (Le., a drop in the futures price)
only for non-positive values of the correlation coefficient. It seems interesting to
investigate whether the breach of this condition (i'.e, a high positive p)
contributes to the inefficiency of a commodity futures market.
Fifth, the effect of expected currency depreciation on prices is treated as a
stepwise process. Since commodity contracts are settled in advance (Le., the
marketing board price guarantee), it is the forward currency market adjusting
first to the new expectation. One period later, the realized profits/losses will be
passed through to the commodity futures market. The size of this pass-through
depends in part on the market power of the marketing board.
Sixth, a change in the conditional variance of the exchange rate (with fixed
versus flexible exchange rates as the limiting case) has been analyzed for several
initial states. No general proposition can be stated since combinations of excess
supply and/or excess demand equilibria on both currency and commodity
markets determine whether forward and futures prices will rise or decline when
moving from a flexible to a fixed exchange rate regime. Once the preconditions
are known, the results are straightforward. Hypothesis testing is reduced merely
to observation of particular countries and commodities.
In this chapter, as well as in the above propositions, we explicitly avoid any
statement about the desirability of fixed versus flexible exchange rates. Social
welfare is not taken into consideration. It must be understood that prices alone
do not determine what is good or bad. Nevertheless, we are of the opinion that it
is worthwhile to investigate the potential effects the exchange rate exerts on the
primary commodity market. In section 4.4 the simplicity of a fixed exchange rate
system is shown as far as it concerns our analysis. Unfortunately, it is the flexible
exchange rate case that increases potential outcomes considerably. The next step
should be the empirical specification, measurement and testing of relations like
(27') to indicate which possibility is most appropriate.
 151

REFERENCES
Anderson, R Wand J-P Danthine (1983), "Hedger Diversity in Futures
Markets", Economic Journa~ 93, 370-89.
Baron, D P (1976), "Flexible Exchange Rates, Forward Markets and the Level of
Trade", American Economic Review, 66, 253--66.
Bond, M E (1987), "An Econometric Study of Primary Commodity Exports from
Developing Country Regions to the World", IMF Staff Papers, 34,
191-227.
Bray, M (1981), "Futures Trading, Rational Expectations, and the Efficient
Market Hypothesis", Econometrica, 49, 575-96.
Cote, A (1987), "The Link between the U.S. Dollar Real Exchange Rate, Real
Primary Commodity Prices, and LDCs' Terms of Trade", The Review of
Economics and Statistics, 69, 547-51.
Chu, K Y and T K Morrison (1986), "World Non-Oil Primary Commodity
Markets: A Medium-Term Framework of Analysis", IMF Staff Papers, 33,
139-84.
Danthine, J-P (1978), "Information, Futures Prices and Stabilizing Speculation",
Journal of Economic Theory, 17, 79-98.
Dominguez, J R (1972), Devaluation and Futures Markets, Lexington Books,
Massachusetts.
Feder, G, R E Just and A Schmitz (1980~, "Futures Markets and the Theory of
the Firm under Price Uncertainty', Quarterly Journal of Economics, 94,
317-28.
Gersovitz, M (1986), "Agro-Industrial Processing and Agricultural Pricing
under Uncertainty", Review of Economic Studies, 53, 153--69.
Ghosh, S, C L Gilbert and A J Hughes Hallett (1987), Stabilizing Speculative
Commodity Markets, Clarendon Press, Oxford.
Gilbert, C L and T B Palaskas (1990), "Modelling Expectations Formation in
Primary Commodity Markets" in L A Winters and D Sapsford (eds.),
Primary Commodity Prices: Economic Models and Policy, Cambridge
University Press, London.
Helleiner, G K (1967), "Marketing Boards and Domestic Stabilization in
Nigeria", Review of Economics and Statistics, 48,69-78.
Hirshleiter, D (1988), "Risk, Futures Pricing, and the Organization of
Production in Commodity Markets", Journal of Political Economy, 96,
1206-22.
Jabara, C Land N E Schwartz (1987), "Flexible Exchange Rates and Commodity
Price Changes: The Case of Japan", American Journal of Agricultural
Economics, 69, 580-90.
Just, R E (1990), "Modelling the Interactive Effect of Alternative Sets of Policies
on Agricultural Prices" in L A Winters and D Sapsford (eds.), Primary
Commodity Prices: Economic Models and Policy, Cambridge University
Press, London.
Kamara, A (1982), "Issues in Futures Markets" A Survey", Journal of Futures
Markets, 2, 261-94.
Kanbur, S M R (1984), "How to Analyse Commodity Price Stabilisation? A
Review Article", Oxford Economic Papers, 36, 336-58.
Karp, L S (1988), "Dynamic Hedging with Uncertain Production", International
Economic Review, 29, 621-37.
Kawai, M (1983), "Price Volatility of Storable Commodities under Rational
Expectations in Spot and Futures Markets", International Economic
Review, 24, 435-59.
Kawai, M and Y Zilcha (1986), "International Trade with Forward-Futures
152

Markets under Exchange Rate and Price Uncertainty", Journal of Inter-

national Economics, 20, 83-98.
Keynes, J M (1930), A Treatise on Money, Vo1.2, Macmillan, London.
Khan, M Sand P J Montiel (1987), "Real Exchange Rate Dynamics in a Small,
Primary-Exporting Country", IMF Staff Papers, 34, 681-710.
Kofman, P, J-M Viaene and C G de Vries (1990), "Primary Commodity Prices
and Exchange Rate Volatility", in L A Winters and D Sapsford (eds.),
Primary Commodity Prices: Economic Models and Policy, Cambridge
University Press, London.
Labys, W C (1980), "Commodity Price Stabilization Models: A Review and
Appraisal", Journal of Policy Modeling, 2, 121-36.
Li, C-M (1935), "The Effect of Depreciated Exchange upon Merchandise
Movements", Quarterly Journal of Economics, 49, 495-502.
Marcus, A J and D M Modest (1984), "Futures l\:Iarkets and Production
Decisions", Journal of Political Economy, 92, 409-26.
McKinnon, R I (1967), "Futures Markets, Buffer Stocks and Income Stability for
Primary Producers", Journal of Political Economy, 75, 844--61.
Newbery, D M G and J E Stiglitz (1981), The Theory of Commodity Price
Stabilization, Oxford University Press, Oxford.
Peck, A E (1976), "Futures Market, Supply Response and Price Stability",
Quarterly Journal of Economics, 90, 407-23.
Shonkwiler, J Sand G S Maddala (1985), "Modelinl7 Expectations of Bounded
Prices: An Application to the Market for Corn' , Review of Economics and
Statistics, 67, 697-702.
Stein, J L (1986), The Economics of Futures Markets, Basil Blackwell, London.
Taussig, F W (1896), "The International Silver Situation", Quarterly Journal of
Economics, 11, 1-35.
Thompson, S Rand G E Bond (1987), "Offshore Commodity Hedging under
Floating Exchange Rates", American Journal of Agricultural Economics,
69,46-55.
Turnovsky, S J (1983), "The Determination of Spot and Futures Prices with
Storable Commodities", Econometrica, 51, 1363--87.
Ueda, K (1983), "Trade Balance Adjustment with Imported Intermediate Goods:
The Japanese Case", Review of Economics and Statistics, 65, 618-25.
Veljanovski, C G (1986), "An Institutional Analysis of Futures Contracting" in B
Goss (ed.), Futures Markets: Their Establishment and Performance,
Croom Helm Sydney.
Viaene, J-M (1989), The Equilibrium Approach to International Commodity
Prices, discussion paper, 8913/G Erasmus University, Rotterdam.
Williams, J (1987), "Futures Markets: A Consequence of Risk Aversion or
Transaction Costs?", Journal of Political Economy, 95, 1000-23.
Working, H (1942), "Quotations on Commodity Futures as Price Forecasts",
Econometrica, 10,39-52.
Working, H (1977), Selected Writings of Holbrook Working, Chicago Board of
Trade, Chicago.
Wright, B D and J C Williams (1982), "The Economic Role of Commodity
Storage", Economic Journal 92, 596--614.
Wright, B D and J C Williams (1989), "A Theory of Negative Prices for
Storage", Journal of Futures Markets, 9, 1-13.
 CHAPTER 6

AN EVALUATION OF THE PERFORMANCE

OF SPECULATIVE MARKETS 1

Jerome L. Stein

1. INTRODUCTION
Speculative markets are those where the price expected to prevail in the future
has a profound effect upon the current price. In these markets, the relevant stock
is large relative to the relevant flow. Examples of these markets are: foreign
exchange, where the ratio of short term foreign currency claims to the value of
trade is large; commodity markets where the ratio of inventory to current
consumption is large; the government securities market where the ratio of the
existing stock of government securities to the current budget deficit is large; the
stock market where the outstanding stock is large relative to new issues.
There has been considerable price volatility in these markets, since they
have become deregulated. The deregulation is international. Interest rates have
become more "volatile" since 1979, when the Federal Reserve changed its
operating procedure. As a result of the greater asset price volatility, futures
markets have developed, particularly in Treasury bonds and stock market
indexes.
Deregulation and futures markets have their critics and defenders. 2 The
critics claim that there is too much price volatility. The defenders claim that the

1 I thank Akio Yasuhara for his criticisms of an earlier draft and his
suggestions for revision, which I have followed, and to Robert Brooks who
supplied the Australian data. This paper also benefitted from comments received
at seminars at Monash University and at the Handelshogskolan in Stockholm.
2 In some cases (such as Australia and New Zealand) financial market
deregulation has been introduced by Labour governments.
153
L. Phlips (ed.), Commodity, Futures and Financial Markets. 153-178.
© 1991 Kluwer Academic Publishers.
154

deregulation and futures markets serve important economic functions. The price
volatility simply reflects the variability of the fundamentals of the economy; and
the price of the asset reflects all available information about future events, i.e.,
the fundamentals.
In his survey of research in futures markets, Barry A. Goss (1986) wrote:
"The literature contains no comprehensive framework for analyzing the
performance of futures markets; nevertheless there is probably a reasonable
degree of agreement on the functions of these markets ... [risk management and
price discovery] ... The real difficulty is that there exist no clearly agreed criteria
for evaluating the performance of futures markets in these functions."
The aim of this paper is to provide a quantitative welfare measure of
speculative market performance, whether it be for commodities or financial
instruments. One can then answer the following questions. In quantitative terms,
which speculative markets are functioning well and which are functioning badly?
What are the causes of poor market performance? How effectively does the
speculative market process information? How can we measure the welfare
contribution of futures markets?

2. THE INADEQUACY OF CONVENTIONAL TEST OF MARKET

PERFORMANCE
The main criteria of speculative market performance, which characterizes the
state of the art, are whether: (1) the current price is as variable as dictated by
the "fundamentals"; (2) the current spot and futures price contain all relevant
information about the subsequent price; (3) there are unexploited profit
opportunities.
The standard method of evaluating the functioning of speculative markets
has been the tests of the Efficient Market Hypothesis EMH (see Samuelson,
1965), which may be stated as follows. The 'forecast error' FE(h)=p(t)-q(t;t-h),
defined as the spot price p( t) less the futures price q( tit-h) at time t-h on a
contract maturing at time t, is equal to /1, where /1 is i.i.d. with a zero
expectation.
FE(h) = pet) - q(t;t-h) = /1 (1)
The null hypothesis is that (1) FE=/1 is Li.d. with a zero mean. The futures price
should be the best unbiased estimate of the subsequent spot price. When account
is taken of an. objectively determined risk premium based upon an economic
theory: e.g., the CAPM or consumption based CAPM, the expectation of /1 need
 155

not be zero.
The researcher generally regresses the FE upon a vector V(t-l) of
information known at time t-1. This hypothesis is often tested by an equation of
the form (2), where p'(t) and q'(t;t-l) are logarithms. Included in V(t-l) are
lagged prices and all other known economic variables. The EMH is that the
futures price contains all relevant information about the subsequent price. We
denote logarithms by primes.
p'(t) = a + bq'(t:t-l) + cp'(t-l) + J.L (2)
The null hypotheses (2) are that: a=risk premium; coefficients b=l, c=O. If
the null hypotheses are not rejected, the researcher claims that the market is
efficient. The futures price contains all relevant available information about the
subsequent price. Insofar as the spot or futures price is volatile, the volatility is
due to the fundamental factors in the economy.
There are several limitations of the tests of the EMH which can be seen in
Table 1. It refers to hypotheses (1) and (2), applied to two contracts traded on
the Sydney Futures Exchange: 90 day bank accepted bills (BAB), and share price
index (SPI). These data are contained in a paper by Robert Brooks.

Table 1. Tests of EMH (1) and (2): Bank Accepted Bills, Share Price Index,
Sydney Futures Exchange, 1983.2-1989.2, 1984.1-1989.2
Maturity Bank accepted bills Share price index
t(b-l) t(FE') t(b-l) t(FE')
h=1 1.94 -1.23 1.03 .538
h=2 .34 -1.6 -1.17 -.052

h=3 -.79 -1.0 -2.06 .973

Coefficient of variation
Spot price: 3.2% 33.4%
Source: The empirical work in this paper is based upon data in Brooks (1989).
For t(FE'), N=25; for t(b-l), N=22.

The column labelled t(FE') is the t-value of the mean forecast error. It is
the mean divided by the standard error of the mean. The null hypothesis (1) is
that t(FE') is not significantly different from zero. The column labelled t(b-l) is
the t-value of the difference between regression coefficient b and unity, from
regression equation (2). Null hypothesis (2) is that t(b-l) is not significantly
different from zero. For each contract, three distances from maturities were used:
156

1,2,3 months.
No forecast error t(FE') is significantly different from zero. Four of the
t(b-1) values are not significantly different from zero, and in two cases they are
almost significant at the 5% level.
The EMH tests are "pregnancy tests": markets are either efficient or they
are not. The basic question is which markets are operating well, and which ones
are operating poorly? A major objection to the EMH literature is that this
elementary question cannot be answered within the framework of the EMH, as is
seen in Table 1. The last row of the table contains the coefficient of variation, the
standard deviation as a percent of the mean, of the spot price of bank accepted
bills and of the share price index, respectively. The variability of the share price
index is approximately ten times that of bank accepted bills. Does this mean that
the market in share price indexes is "worse" than that for bank accepted bills?
Several points emerge from this discussion. First: one cannot rank market
performance on the basis of the EMH. An economic model must be used to
answer the question: What determines the difference between the price p(t) and
the futures price q(t;t-h), referred to as the "forecast error" FE(h)=
p(t)-q(t;t-h)? Second: the EMH tests are incapable of answering the questions:
what is the social benefit of futures markets; which markets are performing well
and which are performing badly?3 We need a standard to compare market
performance with welfare measures. The remainder of the paper is devoted to
these issues.
The theory is developed in sections 3 and 4 and is applied in section 5 to
markets in financial instruments, stock price indexes and traditional
commodities.

3 Another objection to the EMH is that there may be "rational bubbles".

Equation (1) can be a self-fulfilling prophesy. The price expected to prevail in
the next period may be generated by irrelevant factors, and the price this period
is tied to the expected price by the cost of carry. Thus (1) may always be
satisfied, but the price movement is not tied to fundamentals. The critics of this
point of view claim that there are economic forces which prevent such bubbles
from being long lived, when the fundamentals are given. That is, there must be
an underlying model of which equation (1) is only one part. All bubbles are
ephemeral, when the conditions of production and consumption are taken into
account.
 157

3. WELFARE MEASURES
The evaluation of the performance of speculative markets is based exclusively
upon the social welfare generated by different intertemporal allocations of
resources. Welfare is evaluated in terms of the present value of the utility of per
capita consumption. A very simple model (see Samuelson, 1972; Stein, 1986:ch.6)
is sufficiently general to encompass commodities and financial instruments.
Let there be two periods t=(O,l), with a given population. The amount
saved is denoted by "s". Quantity "s" can be negative, if there is net borrowing
from abroad. Let the initial amount of goods available for consumption be
defined as unity. Therefore (l-tl) is the amount of consumption at time t=O. Let
U(l-tl) be the utility of current consumption, where function U has the usual
properties.
The physical return on the saving is m, the marginal product of the
investment less the storage and spoilage costs. Define a=l+m, the amount of
goods in period t+1 obtained from the investment of one unit in period t. The
discount factor r reflects the marginal rate of time preference.
In period t=l there will be x* of goods available from sources other than
from past savings or investment. Variable x* is the fundamental source of the
uncertainty. When the saving decision is made at time t=O the value of x* is
unknown. The savings made at t=O have accumulated to as. There will be a total
of (x*+as) of goods available for consumption in t=1. See Table 2 for a summary
of the problem.

Table 2. Welfare Implications of the Two Period Model

t=O t=l

Endowment x*
Consumption 1-tl x*+as
Present value utility U(l-tl) U(x*+as)/(1+r)

The present value of the utility in the second period is U(x*+as)/(1+r).

The total welfare or utility W over the two periods is equation (3). We shall
suppose that the utility function U is that of the representative consumer.
W(s) = U(l-s) + U(x*+as)/(1+r). (3)
158

Regardless of the type of economic organization or market structure, every

society must decide upon the optimal amount of saving s. The basic source of
uncertainty is x the quantities of goods which will be available during the next
period. The expected value of x is denoted by Ex, which is the objective mean of
the distribution of x.
This model can be applied to commodities or to financial instruments. In
commodities a unit of the good could be sold to consumers at price p(t) which is
equal to the marginal utility of consumption U'(c(t)) = U'(I-6) divided by the
marginal utility of income. (During a relatively short time span, the marginal
utility of income will be treated as a constant and ignored. Over a longer
interval, the use of deflated prices will take variations of the marginal utility of
nominal income into account.) If a unit of the commodity is stored, then a units
of the good will be available in the next period. At time t+l the consumers value
the good at price p(t+1) which is equal to the marginal utility of consumption
U'(C(t+1)) = U'(x*+as).
In the case of financial instruments, there is also a storage decision.4 A
corporation issues securities with a face value of $1 to finance capital formation.
These securities are offered to financial intermediaries who bid price p(t). The
corporation purchases capital goods valued p(t) dollars with the proceeds. The
value of the capital goods is equal to the marginal utility of consumption
sacrificed to produce these goods. Thus the bid price p( t) = U 1 (c( t)).
The financial intermediary does not know at what price it will be able to
sell these securities to institutional investors at a later date. Thus the inter-
mediary is storing the securities. At time t+ 1, the financial intermediary sells
these securities in the capital market to institutional investors/savers at price
p(t+l). The price p(t+l) represents the marginal utility of consumption sacri-
ficed to purchase these securities U 1 (c( t+ 1)). In each case, there is the same
intertemporal optimization problem. Total welfare is (3) and marginal welfare is
(4).
W'(S) = aU'(x+as)/(1+r) - U'(I-6) = a p(t+1)/(1+r) - p(t) (4)
If U"<O is constant, then the equation for total welfare is (3a), and
marginal welfare is (4a). A second order approximation is made of function
U(I-6) around the point 8=0; and the function U(x+as) is expanded around the

4 For an analySiS of the role of financial intermediaries and futures markets

in capital formation, see Stein (1986, ch.7).
 159

point Ex.
W(s) = {U(Ex) + U'(Ex)[(x-Ex) + as) + U"(Ex)/2 [(x-Ex) + as)2}/(Hr)
+ [U(l) - U'(l)s + U"(1)s2/2) (3a)

W'(s) = {aU'(Ex) + aU"(Ex)[x - Ex + as]}/(Hr) - [U'(l) - U"(l)s) (4a)

The components of equation (4) or (4a) are drawn in figure 1. Curve P(D) is
U'(l-s) the marginal utility of consumption sacrificed. it is a positive linear
function of saving. Curve pel) is aU'(x+as)/(Hr) the pre~ent value of the
marginal utility of the saving; and is a negative linear function of saving. Curve
EP(l) is the expected present value of the marginal utility of saving. Two crucial
assumptions are made in this paper. [AI) The "endowment" x* in period t+l is
independent of the amount saved in period t. See footnote 5 below. [A2] The
marginal net return m=a-l is deterministic.
Three values of welfare W(s) and saving are considered. The first is W(s**)
that results when saving s** is chosen by an omniscient social planner. The
second is W(s*) that results when the saving s* is chosen if there were rational
expectations, to be defined below. The third is W(s) that results when the saving
s is actually chosen.
If the savings-investment decision were directed by an omniscient social
planner whose utility function were of the representative consumer, he would
select that amount of saving s=s** to maximize quantity W(s) in equation (3) or
(3a). The omniscience assumption implies that at time t=D the planner knows
the subsequent value of x. Marginal welfare W'(s**) is equal to zero. This is
where curve P(D), the marginal utility of saving in period 0, is equal to the
present value of the marginal utility of savings in period 1, curve P(l).
Next, suppose that the planner is not omniscient, but has rational
expectations in the following sense. He only knows the objective expectation Ex
of the distribution of stochastic variable x. The rational expectations RE
optimization is to select a value of s=s* to maximize the expected value of
welfare, where the expectation is taken over the distribution of x. This is where
curve P(D) intersects curve EP(l), the present value of expected marginal utility
of savings in period 1. In figure 1, it is assumed that (Ex-x) is positive so that
the saving s* with RE is less than s** that would occur under omniscience.
The welfare loss is defined as the difference between the maximum welfare
160

W(s**) and the welfare W(s) that results from saving s. Call the welfare loss
1(s). This is equation (5). The welfare loss is the integral of the difference
between curve P(l) and P(O) between the optimal s** and actual saving s.
1(s) = W(s**) - W(s) = J W'(v)dv. (5)
The welfare loss is expressible as equation (6) or (6a). Term K is the sum of
the slopes of curve P(l) and P(O).
1(s) = [W'(s)]2/2K (6)
1(s) = (s**-s)2K/ 2 = [(s**-s*) + (s*-s)]2K/ 2 (6a)
Figure 1 illustrates these equations very simply: Suppose saving s were
actually selected in period (t=O). The marginal welfare loss is distance
W' (s)= bc. The welfare loss is the area of triangle abc. But the area of the
triangle is precisely [W' (s )]2 /2K=(bc)2 /2K. That is, the welfare loss or area of
the triangle is proportional to the square of the vertical distance between curves
P(l) and P(O).
There are two sources of loss. This is seen in (6a) when saving is s in figure
1. The first is (s*-s) where the saving differs from the RE value s*. The second is
(s*-s**) where the RE saving s* differs from the perfect foresight value s**.
Therefore, the area of the triangle or welfare loss is proportional to the square of
the sum of two errors. The first error is the deviation from the RE solution and
the second is the deviation of the RE saving from what would occur under
omniscience.
The difference between the perfect foresight and the RE values of saving
arises because the value of the stochastic variable x differs from its true mean
value Ex. The standard of comparison that I use is in evaluating social losses is
the expected welfare loss E1(s*) that would result when saving s=s* the RE
value. This is equation (7).
E1(s*) = E[W'(s*)]2/2K (7)
Welfare loss E1(s*), the expectation of triangle aed, is unavoidable and
results because x deviates from its expected value. This loss occurs despite the
fact that the agents are using all available information concerning the
distribution of x efficiently.
The Social 10ss statistic S1 that I use to evaluate the functioning of
speculative markets is equation (8). It is the expectation of: the ratio of the
actual welfare loss to the unavoidable welfare loss. The expectation in the
numerator is taken over values of x and s; whereas the expectation in the
 161

I'

/
/
I'

----~--~------~.~ *
.Q
,Il l u··. ·. .
l
I
I I ....
. . -. .
/1 II . . . ..
.....

.-
162

denominator is taken over the values of x, since s is fixed at the RE value s*.
SL = EL(s)/EL(s*) = E[W'(s)]2/E[W'(s*)]2 (8)
The constant K which appears in both the numerator and denominator cancels.
Hence, I do not need to know the slopes of the curves P(I) and P(O) in making
welfare comparisons.
Graphically, the SL statistic is the ratio of the expectation of triangle abc
to the expectation of triangle aed. The lower the value of the SL statistic the
better is the allocation of resources and the functioning of the market.

4. A TWO PERIOD MODEL

A market model is now developed to explain what determines the actual amount
of storage s produced in the market, and thereby explain the factors which
determine the welfare loss L(s) in different markets. In this part, I show three
things. First, the marginal welfare is proportional to the difference between the
spot price at time t=l, denoted p(I), and the futures price at time t=O for
delivery at time t=l, denoted q(l;O). This is equation (A). The value of
[p(l) - q(I;O)] is often referred to as the "forecast error" FE(t).
W'(s) = ap(I)/(l+r) - p(O) = [p(l) - q(I;O)]a/(l+r) (A)
Second, the welfare loss L( s) is proportional to the square of the "forecast
error". This is equation (B), where the factor of proportionality
K' = (a/(l+r))2/2K.
L(s) = (1/2K)[W'(s)]2 = K'[p(l) _q(I;O)]2 = K'[FE]2 (B)
Third, the "forecast error" FE is the sum of three elements: an i.i.d. term
with a zero expectation f, a Bayesian error y and a positive, zero, or negative risk
premium c. Thus the welfare loss is proportional to the square of the sum of these
three elements.
The model below shows what determines the components of the welfare
loss. We consider a storable "commodity" such as traditional commodities or
financial instruments. In each case, the storage decision involves the sacrifice of
present consumption for future consumption. The basic model underlying our
analysis concerns the amount of investment or storage that will be carried from
period t=O to t=l. We briefly present the two period model.
 163

4.1 The Structure of the Market Model

Equations (9a) and (9b) are inverse demand curves for a commodity, which
relates price pet) and consumption C(t). The intercept of demand is U(t) and the
slope is bet). It is assumed that parameter U(t) is exogenous. 5
pel) = U(l) - b(l)C(l) (9a)
p(O) = U(O) - b(O)C(O) (9b)
Consumption in period t=O is production Z(O) less the quantity S of
inventory investment: equation (lOa). Consumption in period t=l is production
Z(l) plus the inventory disinvestment: equation (lOb). There are no stocks left
over at the end of period t=l, and the transversality condition is met. 6
C(O) = Z(O) - S (lOa)
C(l) = Z(l) + S (lOb)
Equation (11) results when (10) is substituted into (9). The term Pn(t) =
U(t) - b(t)Z(t) is the price in period t that would have occurred if there were no
carryover, i.e., if S=O.
p(O) = Pn(O) + b(O)S (lla)
pel) = Pn(l) - b(l)S (lIb)
Curve P(O) in figure 1 corresponds to p(O) in equation (lla); and curve
pel) in figure 1 corresponds to ap(l)/(1+r), where pel) is given by (lIb).
Refer to the agent who carries inventories (storage) as the dealer. The
dealer has three options. He can sell the commodity at time t=O at the spot price
p(O). He can hold the inventory unhedged, and sell it at uncertain price p(l). If
there is a futures market, he can hedge his inventory s by selling a quantity of
futures at price q(1;O). Assume that the futures price at maturity q(l;l) is the
prevailing spot price pel), so that q(l;l)=p(l).
When the dealer optimizes, the quantity stored will be given by equation
(12). The present value of the marginal product of storage, evaluated at the
futures price aq(l;O)/(1+r), is equal to the spot price p(O). The futures price is

5 Foreign exchange would be a valid example in the short period, a quarter,

but not in a longer period. This means that U is independent of p: i.e., if the
repercussions of the exchange rate upon the rest of the economy could be ignored.
We are dubious that such an assumption is justified over a longer period. The
reason is that the foreign exchange rate has macroeconomic implications for
aggregate demand. Hence the "endowment" is not necessarily independent of the
price of foreign exchange. It is the latter point that underlies the "Optimum
Foreign Exchange Market" argument in Stein (1963).
6 This transversality condition rules out speculative bubbles.
164

used as a certainty equivalent price. This is the same as saying that the futures
price is equal to the marginal cost of storing a commodity.
aq(l;O)/(Hr) = p(O) (12)
The futures market consists of many participants with different subjective
expectations of the price that will prevail at t=l. There are commercial firms,
professional speculators and amateur speculators in the futures market. Some
participants are more knowledgeable than others.
The futures price will equate the supply of futures with the demand.
Futures will be supplied by short hedgers, and by those who think that p(l) will
be less than q(l;O). Futures will be demanded by long hedgers, and by those who
think that p(l) will exceed q(ljO). The market equilibrium futures price will be a
weighted average of the prices subjectively expected by the market participants
to prevail at t=l. The weight of the i-th group is wi' and its subjective
expectation is EiP(l;O). Denote the weighted average of subjectively expected
prices by Ep(l;O). The subjective expectation is taken at time t=O. (See Stein,
1986; ch.2,3 and Varian (1989) for a discussion of these issues.)
There may, or may not, be a risk premium between the futures price and
the market expected price. The sign will depend upon the balance between short
and long hedging, opportunities to diversify risk with other assets and risk
aversion. The existence of a systematic risk premium with a given sign is
questionable (see Peck, 1985). Nevertheless, let the risk premium be RP, of
indeterminate sign in general.
q(l;O) = Ep(l;O) - RP = Ewi EiP(ljO) - RP, EWi = 1 (13)
Define the Bayesian error, denoted by y, as the difference between the
subjective expectation held by the market Ep(l;O) and the objective (rational)
expectation Ep(I). The characteristics of the Bayesian error are discussed below.
y = Ep(I;O) - Ep(l) (14)
The Bayesian error refers to the situation where the market participants do
not know the objective mean denoted by Ep(1). When the market has Muth
Rational Expectations (MRE), then there is no Bayesian error, the subjective
and objective expectations are the same, y=O. The convergence to MRE is the
situation where the variance of the Bayesian error converges to zero.
Using (12)-(14), the quantity of storage will be such as to satisfy equation
(15).
aEp(l)/(Hr) + (y-RP)a/(Hr) = p(O). (15)
Equation (15) is the general case. There are three relevant cases. The first
 165

case is perfect foresight, so that the expected price and actual price are equal,
i.e., £=0 in (16). In this case, there will obviously be no risk premium.
p(l) - Ep(l) = £ (16)
Then the quantity of storage satisfies equation (17). The left hand side is
the present value of the marginal product of storage evaluated at the actual price
p(I). It is described by curve P(I) in figure 1. The quantity of storage will be s**,
where curve P(I) intersects curve P(O).
ap(I)/(l+r) = p(O) ~ s=s**. (17)
Using equation (12), the futures price will equal the subsequently realized
price. There will be no forecast error. Equation (17a) implies that storage is s**.
p(l) = q(ljO). (17a)
The second case is where there is no risk premium and there is rational
expectations (Bayesian error y=O). Then the quantity of storage s=s* satisfies
equation (18).
aEp(I)/(l+r) = p(O) (18)
The left hand side is the rationally expected present value of the marginal
product of storage, and is described by curve EP(I) in figure I, and the right
hand side by curve P(O).
Using (12) and (18), when the futures price is equal to the rationally
expected price, equation (18a), the quantity stored will be s*.
Ep(l) = q(ljO) (18a)
In the third case, when there is a Bayesian error and risk premium,
quantity (y-RP) in equation (15) is not zero. This is repeated as (19). If (y-RP)
is negative, then the left hand side of (19) is described by curve E*P(1) in figure
1. The quantity of storage is s<s*<s**, where curve E*P(1) and P(O) intersect.
aEp(I)/(l+r) + (y-RP)a/(l+r) = p(o). (19)
When there is a Bayesian error and risk premium, the "forecast error" is
given by (20).
p(l) - q(ljO) = [p(l) - Ep(I)] + [Ep(l) - Ep(ljO)] + [Ep(ljO) - q(ljO)] (20a)
FE = £-y + RP. (20b)
The first term is the Li.d. term £ which has a zero expectation. It produces the
difference between the perfect foresight curve P(I) and the curve EP(I), when
there are RE and no risk premium. The second term is the Bayesian error,
whereby the market's subjective expectation Ep(ljO) differs from the true or
objective expectation Ep(I). The sum of the last two terms, Ep(l) - q(ljO)
generates the curve E*P(1) which differs from the curve EP(I).
166

In figure 1, distance fc=y-RP<O. That is curve EP*(l) is below curve

EP(l). The subjectively expected price less risk premium is less than the
rationally expected price, at any quantity of storage. That is why the quantity of
storage s is less than the rational expectation value s* in figure 1.

4.2 Bayesian Error

The Bayesian error y(T;t) is the difference between the subjective and objective
expectation taken at time t of the price which will prevail at time T>t. The
market participants are assumed to act in a Bayesian manner. The start with a
prior concept of what will be the expected price. They' take a sample from the
information set; and on that basis have an estimate of the expected price. The
posterior expectation of the price is a weighted average of the prior and the
sample. In each period prior to date T, the participants take a sample from the
information set and form a new prior estimate of the population mean. Thereby,
a sequence is generated of subjective estimates of the population mean.
Corresponding to each subjective estimate, there is a Bayesian error
between the unknown population mean and the subjective estimate. A sequence
of Bayesian errors y(T;t) is generated.
It has been shown (see Stein, 1986, 69-76, 92-93) that: (i) the Bayesian
errors are serially correlated; (ii) when the distributions of the fundamental
economic variables are constant, the expectation and variance of the Bayesian
error var y(T;t) decline asymptotically to zero as the cumulative number of
samples grows steadily.
The expectation of the Bayesian error that will be made at time t,
conditional upon information known at time 0, is Ey(T;t) as described by (21).
The precision of the initial prior is 8(0), n is the sample size that is taken at time
t from the information set and nt is the cumulative items sampled up through
time t. The precision (reciprocal of the variance) of the underlying distribution is
k.
Ey(T;t) = [8(0)j(8(0)+nkt)] y(O) (21a)
Figure 2 describes the expectation of the Bayesian error, as a function of time.
The initial value is y(O) or y' (0). These errors will be serially correlated, and will
converge asymptotically to zero. The variance of the Bayesian error (21) also
converges asymptotically to zero. That is why I have called this process
Asymptotically Rational Expectations (op. cit.).
 167

III

...
E
.-
VI
Ill::
=
)
0::

\
0::
~

l Z
l I
I .a::
•I
~ Io-ot

/ \.. VI
LU
I' ~
:.-
.....
.1
..' \ .a:
Q:a
..... /~ \
. ., ~/
0/'
........
I- N
IeJ
:rJ
.....
.
lSI ~

=
~
LU ., Ill::
IS'
. .,
.... I-
IeJ
., .....
L!I
~
LL.
:n :ra
168

var y(t) = l/knt . (21b)

These variables will be of considerable importance in the subsequent
analysis.

5. APPLICATION OF THE THEORY

5.1 Research Design
We showed that the marginal welfare is the difference between ap(l)/(l+r), the
present value of the marginal product of storage at time t=l, and p(O), the spot
price at time t=O, equation (4). Moreover, equation (12) shows that the futures
price q(l.;O) is linked to the spot price p(O) by the cost of carry. It can also be
viewed as a no arbitrage equation, between the spot and futures markets.
W'(s) = ap(l)/(Hr) - p(O) (4)
aq(l;O)/(Hr) = p(O) (12)
It follows that marginal welfare is proportional to the forecast error
FE=p(l)-q(l;O), equation (22a).
W'(s) = FE a/(Hr) (22a)
Using the decomposition of the FE in (20), the marginal welfare is (22b).
W'(s) = (t-y+RP)a/(Hr) (22b)
The welfare loss is (23); and the expectation of the welfare loss over the
space of t-y+RP is (24). The factor of proportionality K' =(1/2K)(a/(Hr))2.
L(s) = K'[p(t+1) - q(t+1;t)]2 = K'.[FEf (23)
K' = (1/2K)(a/(Hr))2 (23a)
The expectation of the square of the FE is simply the mean square forecast
error M5E=E[FE]2=E[p( t+ 1)-q( t+ l;t )]2.
EL(s) = K'E[p(t+1) - q(t+1;t)]2 = K'·M5E (24)
It was shown in equation (20), that the forecast error consists of three
parts: an unavoidable error t, a Bayesian error y, and a risk premium. Therefore,
the expected loss in general is (25).
EL[i]=K'[t-y+RP]2. (25)
The above expression is the expectation of triangle abc in figure 1. Our
social loss statistic compares this area with what would have occurred if there
were rational expectations, triangle aed. The latter is K· var t, where var t is the
variance of the unavoidable error. Therefore, our 5L statistic is (26).
5L = M5E/var t. (26)
This equation can be used to compare the welfare implications of
alternative markets. The lower the 5L the better is the market functioning. The
 169

social utility of a futures market is measured by the reduction in social loss that
occurs. The social loss produced by the Bayesian error and risk premium is
trapezoid cbed in figure 1.
Using equations (26) and (20), the social loss is equation (27). The co-
variance terms between the unavoidable error, Bayesian error, and risk premium
should be zero.
S1 = 1 + (var y/var E) + (RP)2/var E (27)
The risk premium has a clear theoretical meaning, although in many
empirical studies it is a euphemism for the forecast error. Consider the case of
financial futures. The risk premium is the percentage difference between the price
that the financial intermediary expects to sell the securities to the institutional
investors and the bid price to the corporation. When there is a futures market,
the financial intermediary bids p(t) for (say) an Aaa bond and simultaneously
sells an optimal quantity of Treasury bonds in the futures market. The
intermediary knows that there is a relation between the expected spot price of
the Aaa bond and the expected spot price of a Treasury bond. Moreover, he
knows that the spot price of the Treasury is tied to its futures price by the cost of
carry. When the intermediary markets the Aaa bonds, he purchases the Treasury
bond futures, generally prior to maturity.
1et R2 be the square of the correlation between the price of the Aaa at time
t+1 and the futures price of the Treasury bond. These relations permit the
intermediary to reduce risk. The risk premium RP has been shown (Stein, 1990)
to be equation (28). Parameter a is the coefficient of absolute risk aversion of the
intermediary.
RP = a(1-R 2)var p. (28)
The percentage risk reduction, when an optimal position is taken in the
futures markets, is R2. If there are no possibilities to hedge, then R2=0. If the
R2=1, (such that the Aaa interest rate is perfectly correlated with the Treasury
bond interest rate and the spot/future relationship in Treasury bonds were
constant), then there would be no risk premium charged.
For example, during the period 1971.02-1987.12, the variance of the
monthly Aaa bond yield var p was 5.2% p.a. The square of the correlation
coefficient between the monthly change in the Aaa bond yield and in the monthly
change in the Treasury bond yield was .87. If the Treasury spot and future were
perfectly correlated, the percentage risk reduction by an intermediary who
hedged optimally would be 87%. The risk premium would decline from (a var p)
170

to .13a var p.
The problem is how to measure the variance of the unavoidable error var E.
We believe, subject to some tests below, that when we are one month from
maturity practically all of the MSE is due to unavoidable error. At that distance,
people will know as much about the expected price as they ever will. Hence, there
will be a minimum of Bayesian error, or Muth Rational Expectations will be as
valid as it ever will be. Similarly, the risk premium will be as low as it can be,
whatever the opportunities for diversification. Let MSE(j;l) denote the MSE in
the j-th commodity where the futures is one month dist,ant from maturity. Then
MSE(j;l) will be our measure of K I . var E, the social loss from the unavoidable
error. Hence the SL(j;h)=MSE(j;h)/MSE(j;l) measure refers to the social loss for
the j-th commodity where the futures price has a maturity h-months in the
future.
SL(j;h) = MSE(j;h)/MSE(j;l). (29)
We can make three types of comparisons, although only the last two will be
done in this paper. The first is to compare the SL statistics for a given
commodity and distance from maturity in the case where there is and where
there is no futures market. Second, compare the MSE in a given (j-th)
commodity over several maturities: for example, MSE(j;l), MSE(j;2), MSE(j;3).
Third, compare the SL for several commodities for a given distance from
maturity. Thus we compare futures markets in: bank accepted bills, share price
index, corn, cattle, pork, soybeans, wheat and soybeans for h=3 months. The
lower the SL statistic, the better is the market performance.

5.2 Empirical Results

Robert Brooks of Monash University (Australia) calculated and tested the
statistical significance of SL statistics for Bank Accepted Bills (BAB) and Share
Price Index (SPI) futures, traded on the Sydney Futures Exchange during the
period 1983.2, 1989.2. These are the two most actively traded contracts. His aim
was to use some recently developed statistical tests to determine whether SL
differed from unity. I use his data set to examine several aspects of the theory
presented above.
First: I show that the risk premium is not an important component of the
SL. Second: the SL is due to the Bayesian errors. Third: the Bayesian errors are
quite noticeable in the data and are consistent with equation (21a) and figure 2
above. They are serially correlated. Fourth: I show that the SL increases with
 171

distance to maturity and, for any given distance from maturity, is greater for the
Share Price Index than for Bank Accepted Bills. Fifth: I compare the S1 in these
financial instruments with data drawn from the US commodity futures markets.
Here, I draw upon Beom Hong's calculations of S1 in cattle, pork, corn, soybeans
and wheat over a thirty year period.

1. The Unimportance o/the Risk Premium

The forecast error FE in (20) is the sum of an unavoidable error, a Bayesian error
and a risk premium. The expectation of the unavoidable error is zero. If it is
assumed that the expectation of the Bayesian error is zero Ey(O)'=O, the expected
forecast error E(FE) should be the risk premium. This is a number which may be
positive, zero, or negative, depending upon the balance of hedging and the
adequacy of speculation (Stein, 1986,52-54; Peck, 1985).
Denote the spot and futures prices as p(t) and q(t;t-h), and their
logarithms as p'(t) and q'(t;t-h).
The first test of the existence of a risk premium is the test whether the
mean forecast error FE' =p' (t )--q' (t;t-h) is significantly different from zero.
This is described by the t-statistic in Table I, in the column labelled t(FE'). For
both BAB and SPI, and for each distance to maturity h=I,2,3, one cannot reject
the null hypothesis that t(FE' )=0.
The second test concerns the regression of the price p' (t) upon the futures
price q'(t;t-h) and the lagged price p'(t-h), during the period 1984.1, 1989.2. If
there is a risk premium, the intercept term should be significantly different from
zero. But, for BAB and SPI, and for h=I,2,3 we cannot reject the null hypothesis
that the intercept is zero.
The conclusion is that there is no evidence that there is a significant risk
premium which produces a forecast error. It is likely that the existence of futures
markets produced a high R2; and hence the risk premium, described by (28), is
not significantly different from zero. This is a welfare gain.

2. The Bayesian Errors are Present in the Forecast Errors

We now turn to the Bayesian error between the subjective expectation and the
objective estimate of the population mean. The theory, equation (21) or figure 2,
claims that the Bayesian errors are serially correlated but go asymptotically to
zero if the distribution of the fundamentals is stationary. 1et FE(h)=
p(t)--q(t;t-h) be the forecast error on a h-month futures contract. The serial
172

correlation aspect of the Bayesian error states that FE(h) is positively related to
FE(l). If the 1 month future was an overestimate of the price, then the earlier
h=2,3 contract also to overestimate the price. 7
Table 3 describes the regression FE(h)=a(h)+b(h)FE(l) for the bank
accepted bills and share price index for h=2,3 months and FE(l) refers to the
I-month contract. The intercept is never significant, so we only report the slope
b(h). The standard errors are in parentheses. The ** denotes significance at the
5% level; the AR(l) indicates that the AR(1) transformation was used. The
period is 1983.2-1989.2.
The coefficients are highly Significant, which is con'sistent with the theory of
Asymptotically Rational Expectations. The Bayesian errors are important
components of the forecast error, and they are serially correlated.
Table 3. The Relation, FE(h)=a(h) + b(h)FE(I), between the One Month
Forecast Error and the Two, Three 'Month Forecast Errors, Bank Accepted Bills
and Share Price Index, 1983.2-1989.2

Contract b(h);(se) R2 DW

LorT
Bank Accepted Bills
h=2 .66 1.22
h=2, AR(I) 1.28 .15 ** .79 1.8
h=3 1.11 .21 ** .55 1.56
h=3, AR(I) 1.57 .21 ** .74 2.2
Share Price Index
h=2 1.34~ .496t* .23 1.92
h=3 1.57 .66) * .198 1.88

Figure 3 describes the relation between the forecast errors for bank accepted
bills: BFE(3) and BFE(I). Figure 4 does the same for the share price index:
SPI(3) and SFE(l). Note that the forecast error for the one month SPI future
concerning the crash was quite small relative to the three month SPI future. The
figures are consistent with the theory in equation (20) and figure 2 above.
On a given contract, the serial correlation of the Bayesian errors produces
serial correlation of forecast errors. This is true both before, during and after the
stock market crash.

7 We have argued that the expectation of the variance of the Bayesian error
at one month from maturity is less than the values more distant from maturity,
so that MSE(I) is as good a proxy for the unavoidable error as we are likely to
obtain.
FIGURE 3. BANH BILL FORECAST ERRORS, 1 AND 3 MONTHS, 83.2 89.2
4.
3....
2
1
0Jj
-1 .:
-2
-3 ' ..
-. :
-4
-5 f,..~..,....,...,..."I"""'"'I"..,....~..,....,-,-.,....,...,...r-r-.,...,.-,-,...-p-..,..,...,.
1983 1984 1985 1986 1987 1988 1989
I-'
-;r
_BFE1 ....... BFE3 c.>
FIGURE 4 SHARE PRICE INDEX FORECAST ERRORS 1 AND 3 MONTHS, 83.2 89.2. ~
500··- - - - - - - - - - - - -......

0~~

-500

-1000

-1500~ i ,

1983 1984 1985 1986 1987 1988 1989

_SFE1 ....... SFE3

 175

9. The Social Loss Statistics

The ultimate object is to evaluate errors MSE(h)=E[p(t)--q(t;t-h)]2 and for
logarithms MSE'(h)= E[p'(t)--q'(t;t-h)]2, for commodities and maturity of
futures contracts.
Table 4 presents the social loss SL statistics. In each case, the SL rises with
distance from maturity; and the SL is greater for the share price index than for
bank accepted bills. Brooks (1989) tested the significance of SL(2) and SL(3): are
they different from unity, by using the King-Smith additive t-test.

Table 4. Social Loss Statistics: Bank Accepted Bills and Share Price Index,
1982.2-1989.2
h=(I) h=(2) h=(3)
SL(h)=MSE(h)/MSE(1 )
Share Price Index 1 7.2** 12.2**
Bank Accepted Bills 1 1.66* 2.2**
* denotes significance at the 5% level and ** denotes significance at the 1% level
Source: Brooks (1989)

Brooks concluded that for h=3, there were significant social losses in both
markets, but for h=2, there a significant social loss only in the SPI market.
We shall approach the issue in a somewhat different way in Table 5. If we
calculate the mean square errors of the logarithms MSE', then the units for
various commodities are comparable. We know that the logarithmic forecast
error FE' is not significantly different from zero. Therefore, the variance of FE'
will be used as the MSE'. In Table 5, we present the mean square errors FE' and
social loss statistics S1' in two ways. The first column refers to the share price
index (SPI), and the second to the bank accepted bills (BAB). The two rows
refer to the 1,2,3 month futures contracts.
Reading down the column are the social loss statistics for a given
commodity over h=2 and h=3. The SL, equation (30), is a variance ratio which is
F distributed.
SL = 1 + var ylvar Eo (30)
In each case S1' rises with distance from maturity. This again is consistent
with equation (21), the variance of the Bayesian error is positively related to the
distance from maturity (see Stein, 1986, 73-74).
Column 3 compares the S1' between bank accepted bills and the share
176

price index, for each contract maturity h=I,2,3. The ratio of the SL' for two
commodities j and k, for a given distance h to maturity, is (31). This ratio of
variances reflects the ratio of social losses, and enables us to compare markets.
S1'(j;h)/S1'(k;h) = [1 + var y/var f](j;h)/[1 + var y/var f](k;h). (31)

Table 5. Logarithmic Loss Statistics, Share Price Index and Bank Accepted Bills:
Comparison Between Markets as well as Across Maturities
h=distance S1' (SPI;h) S1'(BAB;h) SL' (SPI)/S1' (BAB)
1 1 1 1
2 4.77** 1.61 2.96
3 7.29** 2.27** 3.21
* denotes significance at the 5% level and ** denotes significance at the 1% level

The S1' for the share price index is larger than that for bank accepted bills.
In both cases the ratio is 3: 1. It seems that the futures market in bank accepted
bills operates better than does the market in share price index. It is quite possible
that these differences in the two S1' are just due to sampling error. I am not
aware of a statistical test of the hypothesis that ratio S1' (j;h) is drawn from the
same distribution as ratio S1' (k;h). Recall that each S1' is a ratio of variances.
Nevertheless, I am dubious that the ratio 3:1 in column 3 is due to sampling
errors. I believe that the bank accepted bill market is indeed informationally
more efficient than the share price index market.
Beom Gyo Hong (1989) calculated the social loss statistics SL' (h) =
E[p' (t )--q' (t;t-h)]2/E[p' (t )--q' (t;t-1)]2 for five agricultural commodities over a
thirty year period. The prices were deflated by the producer price index. The S1'
for a three month futures contract are given in Table 6.

Table 6. Social Loss Statistics for Traditional Commodities

S1'(3)=MSE' (3)/MSE' (1)

Cattle 3.28
Pork 3.32
Corn 2.29
Soybeans 3.13
Wheat 4.49
 177

The social loss statistics for commodities are close to those for bank
accepted bills which are less than that for the share price index. Since the
forecast error does not result from the risk premium, the differences result from
the Bayesian errors. This means that the price discovery process is worst for the
share price index.

6. CONCLUSION
The main conclusions are as follows. We showed that the conventional EMH tests
have very little economic significance, and are not capable of comparing market
performance. The SL statistic and its composition, particularly the Bayesian
error, provide an economically significant way to evaluate market performance
and the efficacy of the price discovery process.
On the basis of our analysis, the social losses are produced by Bayesian
errors. The latter are much more significant in the share price index than in the
market for bank accepted bills or traditional commodities. The price discovery
process for interest rates is significantly superior to that for share prices. The
social losses rise with distance to maturity. When there are futures markets, the
risk premia do not contribute significantly to social losses.
The fact that the price discovery process is much poorer for the share price
index than for interest rates does not imply that share price futures are not
socially useful. We propose that the welfare analysis in this paper be used to
evaluate the social contribution of all futures markets. One could measure the SL
and characteristics of the Bayesian errors on a given commodity before and after
the introduction of futures markets. The marginal welfare loss is given by
equation (4) and the total loss is given by equation (6), when there are no futures
markets.
Although we have not compared SL statistics pre and post futures in this
paper, this is feasible. Thereby one can estimate the social utility of futures
markets.! An operational difficulty is that when there are no futures markets, it is
often difficult to: obtain data for p(t) and p(t+l), due to the heterogeneity of the
spot market. An example would be traditional commodities, which differ by
location and characteristics of delivery. When there are futures markets, one

! The economic functions implications of financial futures are discussed in

Stein (1986, ch.7). The papers by Cox (1976) and Figlewski (1982) are valuable
in understanding and testing the price discovery process.
178

generally uses the price of the expiring future as the spot price, and there is
perfect homogeneity in the spot market. No such problems exist in the use of
financial data: the spot market is fairly homogeneous and actual spot prices can
be used. The technique developed in this paper should therefore be applied to the
financial markets pre and post futures. As a result, we would be able to see just
how useful these markets have been socially.

REFERENCES
Brooks, R D (1989), "A Social Loss Approach to Testing the Efficiency of
Australian Financial Futures", Monash University, Dept. of Economics
Working Paper.
Cox, C (1976), "Futures Trading and Market Information", Journal of Political
Economy, 84, 121-37.
Figlewski, S (1982), "Information Diversity and Market Behavior", Journal of
Finance, 37, 87-102.
Goss, B A (ed.) (1986), Futures Markets: Their Establishment and Performance,
Croom Helm, London and Sydney.
Hong, B G (1989), Speculation and Market Performance, Ph.D. dissertation,
Brown University, Providence R.I.
King, M and M D Smith (1986), "Joint One-Sided Tests of Linear Regression
Coefficients", Journal of Econometrics, 32,367-83.
Peck, A E (1985), "The Economic Role of Traditional Commodity Futures
Markets", in A E Peck (ed.), Futures Markets: Their Economic Role,
American Enterprise Institute, Washington, D.C.
Samuelson, P A (1965), "Proof that Properly Anticipated Prices Fluctuate
Randomly", Industrial Management Review, 6, 41-49
Samuelson, P A (1972), "Proof that Unsuccessful Speculators Confer Less Benefit
to Society than their Losses", Proc. National Academy of Sciences, 69,
1230-33.
Stein, J L (1963), "The Optimum Foreign Exchange Market", American
Economic Review, 53, 384-402.
Stein, J L (1986), The Economics of Futures Markets, Basil Blackwell, Oxford.
Stein, J L (1990), "Rational Expectations and Welfare in Financial Futures
Markets" in B A Goss (ed.), Rational Expectations and Efficiency in
Futures Markets, Routledge, London.
Varian, H (1989), "Differences in Opinion in Financial Markets" in C Stone (ed.),
Financial Risk: Theory, Evidence and Implications, Kluwer AcademIC
Publishers, Dordrecht.
 CHAPTER 7

DYNAMIC WELFARE ANALYSIS AND COMMODITY

FUTURES MARKETS OVERSHOOTING

Gordon C. Rausser and Nicholas Walraven

1. INTRODUCTION
Much of the analysis conducted on commodity futures markets focuses on partial
equilibrium frameworks (Stein, 1981). However, linkages among markets implied
by general equilibrium representations show that such analyses can suffer from
serious limitations. In particular, studies of futures market efficiency which
search for single series martingale or random walk processes cannot be expected
to classify markets correctly (Rausser and Carter, 1983).
Linkages among markets mean that inefficiencies in one market may be
transmitted to related markets. Nowhere is this more likely to be evident than in
commodity futures markets. Since these markets reflect price expectations,
differential information flows in the various markets will generally result in
varying speeds of adjustment to causal forces.
Varying speeds of market adjustment have been used by Dornbusch (1976)
and others to show that exchange rates can overshoot as a result of such market
behaviour. In this work, and the subsequent work by Frankel (1979), exchange
rates overreact to a monetary shock in order to compensate for the disequilibrium
arising in a more slowly adjusting goods market. In the Dornbusch formulation,
the long-run steady state remains unchanged while the exchange rate equates
(temporarily) demand and supply in both the exchange and goods markets. For
an expansionary monetary shock, the exchange rate moves to a level higher than
that implied by the new long-run equilibrium and falls gradually as the sticky
goods market adjusts. Prices in the efficient market overshoot the eventual
equilibrium levels in order to clear the relatively inefficient goods market.
179
L. Phlips (ed.). Commodity. Futures and Financial Markets. 179-203.
© 1991 Kluwer Academic Publishers.
180

In addition to disequilibrium conditions arising in all sectors from monetary

shocks, similar behaviour can arise from other types of shocks. In the case of
commodity markets, market-specific shocks frequently can result from droughts
and other weather-related phenomena. In the case of exchange rate markets,
political instability is one frequent source of unexpected shocks. Furthermore,
attempts on the part of central banks to manage the value of their country's
currency can lead to disequilibria in exchange rate markets, which, in turn, spill
over to related markets. In fact, in many agricultural commodity markets,
interest rate markets, and exchange rate markets, u~anticipated government
policy is a likely source for shocks that arise in a specific market.
The basic Dornbusch model will be extended in this paper to examine the
linkages among three groups of markets: interest rates, exchange rates, and
commodity markets. Price expectations, as represented by futures markets, will
be emphasized. Unlike the Dornbusch model, in which all goods prices except
exchange rates are presumed to be sticky, interest rates and agricultural prices
will be allowed to be "flexible". Futures contracts for agricultural commodities
are homogenous, frequently traded, and (mostly) storable and, thus, are
presumed to be flexible and governed by instantaneous arbitrage.
Allowing flexibility in some markets with less flexibility in other markets
admits the possibility of overshooting in spot markets for interest rates, exchange
rates, and commodity markets. To the extent that this overshooting exists, does
it carryover to the formation of expectations as reflected by futures markets? If
resource allocation decisions are based on these expectations, what are the
welfare implications of overshooting? These questions are investigated by
quantifying the dynamic linkages among US Treasury bills (T-bill); the British
pound; the Canadian dollar; the German mark; the Japanese yen; and three
agricultural commodity markets - corn, cotton, and wheat.
A vector autoregressive moving average (ARMA) model is empirically
estimated for a specific period of tight monetary policy, viz., 1980 through the
spring of 1982. Based on the dynamic adjustment paths, pricing effiCiency is
examined, accuracy and speed of convergence measures are calculated, and
dynamic welfare measures are computed. The accuracy measure is the total
absolute deviation from final equilibrium levels of each price series during the
adjustment period. The speed measure is the number of trading days required for
some percentage of the total deviation to occur.
 181

2. BASIC SPECIFICATION
The existence of differential responses to monetary shocks among markets can
lead to "overshooting". More specifically, price stickiness in some markets is a
necessary but not a sufficient condition for overshooting. Hence, whether pricing
inefficiencies in one market leads to overshooting and allocative inefficiencies in
another market is an empirical question. To demonstrate this result, a basic
model specification linking interest rates, exchange rates, sticky price, and
flexible price markets is advanced followed by two alternative formulations. Both
formulations are based on an open economy, one of which im~oses fixed output
and the other allows the possibility for endogenous output responses to dynamic
price adjustment paths.

2.1 Basic Model Specification

For the case of fixed output, the manufactured good sector is represented by two
basic equations. The first specifies output supply and the second excess demand;
yS_yP=O (1)
m m

y! - Y! = ao(P c - Pm) - a 1(i - u - r) + a3(p~ + e - Pm) (2)

where y~ denotes output supply, Y! defines potential output, y! denotes
demand for manufactured goods, Pc denotes the price of commodities, Pm denotes
the price of manufactured goods, i denotes the nominal rate of interest, u denotes
rate of inflation, j' denotes the long-run equilibrium real rate of interest, and e
denotes the exchange rate. All variables are defined in logs. The rational
expectations for commodity prices are generated by an arbitrage condition,
p~=i+sc, (3)
where sc represents the storage costs associated with withholding the flexible
commodity stock from one period to another. For the money market, a standard
money demand equation is assumed with the following equilibrium condition;
m-p=¢y->..i (4)
where m is the logarithm of the money supply, p is the logarithm of the general
price index, y is the logarithm of income, and>" is the interest rate semi-elasticity
of demand for real balances. This specification presumes that the money market
clears at each moment in time.
Under the above specifications, manufactured goods are produced and
182

imported; agricultural commodities are exported not imported, with agricultural

exports normalized to be zero; domestic and foreign manufactured goods are
perfect substitutes; and, for the agricultural commodity, representing the
flex-price good, purchasing power parity is assumed to hold in the long run.
However, deviations from purchasing power parity occur in the short run due to
differential adjustment speeds in exchange rates and manufactured goods prices.
Note also that the small country assumption is presumed to hold.
For the price index (p), the underlying utility functions are assumed to be
Cobb-Douglas so that the prices are weighted by their expenditure shares, Le.,
p= £r1Pm + ~Pc + (1- £r1 - ~)(p~ + e). (5)
In the short run, uncovered interest parity is presumed to hold
i-i*=ee (6)
where i* is the foreign short-term nominal interest rate and ee is the expected
rate of depreciation or appreciation of domestic currency. Since the small
country's assumption is imposed, the nominal interest rate adjusted for expected
appreciation is always equal to the (given) foreign rate. Implicitly, the
specification assumes perfect substitutability between domestic and foreign
interest-bearing instruments (a one-bond world), absence of risk premia, and
perfect capital mobility.
Since the manufactured goods market is presumed to be sticky in the short
run, the price adjustment process is specified to be some distributed lag of excess
demand, Le.,
Pm = rr(y! -y!) + u. (7)
For the flexible price commodity market, expectations are presumed to be
rational,
.e .
Pc = Pc' (8)
Finally, for the case of exchange rate expectations, two alternative specifications
are evaluated. One is based on rational expectations,
ee = e, (9a)
while the other is based on regressive expectations,
ee = -5(e - e). (9b)
For the second specification (9b), the expected rate of appreciation or
depreciation is assumed to be proportional to the gap between the exchange rate
and its long-run equilibrium value (e). Hence, if the spot rate exceeds its long-run
value, these investors expect the rate to gradually appreciate at a speed of
 183

adjustment equal to o. Finally, the normalization imposed is Pm=Pc'Yc=O where

yc is the level of commodity output.

2.2 Fixed Output and Overshooting

The excess demand equation (2) does not admit output responses to alternative
dynamic price paths. Combining equations in the basic model specification, the
following price dynamic equations are obtained for the case of rational
expectations (9a)
a ~ 1-a-~
Pc = ~Pm-Pm) + -4Pc-pc) + 1 (e-e) + u + r + sc -j.f (10)
A A A

(11)

Pm = - n [
a
0
+ -~a1] (p -P ) + n [~a1]
A mm
a - -
0
(p -P )
A cc
(12)

These dynamic equations (or equivalently those resulting from the regressive
expectation specifications on exchange rates) can be solved to obtain the
following commodity price determination equations:
Pc = III - r/J Y + A(r+U) -
1
b
i- u- r) (13)

(14)

where 01 is the negative root of the solutions which were obtained by solving
equations (10), (11), and (12).1 The results for regressive and rational

1 Note that ~sc = ~r = ~ym( = ~Y = ~P~ = 0) and m = III are assumed

in these derivations.
184

expectations of exchange rates collapse to the same outcome when 5={}1.

It is transparent from equation (14) that overshooting of agricultural
commodity prices follows immediately from the assumption of no output
response. Since all parameters in the basic model specification are positive, the
coefficient on a change in money supply (t.m) is greater than one. This result
shows that the transmission of price disequilibrium in the manufactured good
market to deviations of commodity market prices are higher the larger is u1 , for
a given rate of adjustment and interest semi-elasticity of money demand. Note
also that the degree of overshooting is decreasing in the responsiveness of money
demand. This is simply because more of a shock to the money supply will be
absorbed within the money market the larger is the value of A. Finally, the
length of the time the overshooting will last is a negative function, {}1' or
equivalently when regressive expectations are rational, 5.
The driving force behind the overshooting result of equation (14) is the
sticky price market represented by equations (1) and (7) in combination with the
fixed output specification, equation (2). The arbitrage, equation (3), combined
with short-run disequilibrium between nominal interest rates and the long-run
real rate of interest generates short-run disequilibrium in agricultural commodity
markets. For example, to compensate the holders of grain inventories for
foregOing present consumption, the grain price must rise at the nominal interest
rate between harvests once convenience yields, storage costs, and risk premia are
properly incorporated. If money growth occurs so that a liquidity effect causes a
fall in the nominal interest rates and in the short run the real interest rate falls
due to sticky manufactured goods prices, a better return is available for storing
grain than dollars and investors compete to hold grain inventories. This causes an
immediate jump in the price of grain so that the condition for asset market
equilibrium of equal rates of return is restored. All commodity prices jump and
then are expected to rise at the new, lower interest rates.

2.3 Endogenous Output, Overshooting, and Undershooting

The fixed output specification can be relaxed by replacing equations (1) and (2)
by the following three equations:
{} d
Ym = Ym' (la)

(2a)
 185

_ ao _ _ a1 _
yP =_(p _p ) - - r (2b)
m 1-a c m I-a
2 2
where 0<~<1.
Replacing equations (1) and (2) by (la), (2a), and .(2b) and leaving
equations (3) through (9b) unchanged leads to the following price dynamic
equations in place of equations (10), (11), and (12):
. aCbo~b31P _ a 2 + bo lP _
Pc = (Pm-Pm) + - - (pc-pc) (lOa)
A+b 11P A+b 11P

(l1a)

I-a -a...+b IP
+ _.* - - -1 E
1 " 3 (e-e_) + u + r-I
A+ blIP A+b 11P

(12a)

_ II [b + b + b1(a1-b0 tP-b 3IP)] (p _p )

o 3 A+b 11P m m

where
ao
b =-
o 1~'

From the above price dynamic equations, commodity price determination

equations can be analytically derived. Equation (13) for endogenous output
remains the same but equation (14) becomes
186

(14a)

where -02 is the negative root of the solution obtained by solving equations
(IDa), (Ua), and (12a). Note that the results for the regressive and rational
expectations of exchange rates collapse to the same outcopie when 0=°2 ,
For equation (14a), overshooting is only one of three possible outcomes.
This can be seen from

~p aC¢(bo+b 3 )
__c = 1 + ___--=-_~_-=- __ (15)
~m l-a l +0 2 >.+¢(b o+b 3 +02b l )

Hence, given that the denominator of the second term is positive, whether
overshooting, undershooting, or neutrality results from a particular monetary
shock depends on the numerator of the second term in equation (15). More
precisely,
> overshooting
[ < undershooting (16)
= neutral
In contrast to the result obtained for fixed output (see equation (14)), the
mixed results of (16) reflect the possibility that the output and demand responses
to prices (ao+a3), appropriately modified for the effect of output on money
demand (¢) and the dynamic output adjustment (l-a2), can swamp the initial
moves in flexible price commodity markets that would otherwise result. Hence, it
depends empirically on the relative size of the expenditure share of the sticky
price markets (a l ) and the responsiveness of output to prices and, in turn, the
responsiveness of demand for real money balances to changes in output levels.
If the overshooting result occurs, its degree will depend upon four key
parameters: the relative share of the sticky price markets in the domestic
economy (al ); the responsiveness of money demand (A); the effect of output
 187

response on money demand; and the speed of adjustment (U). The length of time
overshooting will last is a negative function of U. This parameter plays a major
role in the speed of convergence to any new long-run equilibrium resulting from a
monetary shock. The remaining parameters dictate the magnitude of tJ.Pc during
any particular period. The accuracy of any particular level of price relative to the
long-run equilibrium level (tJ.Pc) and the speed of convergence (U) are the major
components of the appropriate dynamic welfare measure.

3. A DYNAMIC MULTICOMMODITYWELFARE MEASURE

The allocative efficiency loss during the joint adjustment of 'all prices to some
new stationary state following a shock requires the development of general
equilibrium welfare measures (Rausser and Just, 1981). Conventional dynamic
and simultaneous supply and demand equations, where futures prices (pf)
representing expectations enter the supply equations, will be presumed. As shown
in numerous places (Zellner and Palm, 1974; Rausser and Carter, 1983), if a set
of endogenous variables is generated by a dynamic simultaneous equation model,
then it is often possible to solve for the transfer function of individual endogenous
variables (such as exchange rates, interest rates, etc.) through algebraic
manipulation. In essence, each endogenous variable in a structural form model
has associated with it an explicit and unique transfer function equation which
expresses the endogenous variable as a linear combination of current and past
values of the exogenous variables and an ARMA error term. Similarly, given that
each exogenous variable can be expressed in terms of an ARMA process, it is
possible to respecify the transfer function equation as an ARMA process for each
endogenous variable. Accordingly, it is possible to represent the basic model
presented in section 2 as a multivariate time series model as long as the relevant
error terms in each exogenous variable can be represented as an ARMA process.
If the vector ARMA process represents adequately the endogenous price
variables, then this framework approximates a rational expectations formulation,
with the error iIi the approximation caused by transactions costs, risk aversion of
agents, etc. (Rausser and Carter, 1983). The resulting price path is a function of
structural supply (S) and demand (D) parameters and expected (futures) prices.
More specifically, assume that an initial steady-state level of prices (f» exists and
there is a shock (Zo) at time o. Since f> represents the net effect of all past
adjustments, the future expected prices (p!) for any time after time 0 are:
188

(17)
where
Q.Z = i-I
E [ A-1(L)B(L)) AO; (18)
1 0 j=O j
A(L) and B(L) represent polynomials in the lag operator that reflect the dynamic
interactions of the endogenous prices. These lag polynomials are assumed to be
stationary and invertible. If there are no subsequent shocks, the effect of Zo at
any time may be expressed as the initial steady state (f» plus the net effect of Zo
to that time. The deviation of prices from the event~al long-run equilibrium
changes over time; therefore, the amount of welfare loss also changes over time.
To determine the welfare loss in a market at any time during the adjust-
ment period, consider the static welfare analysis depicted in Figure 1. The
long-run steady state is p following the shock, and y is the corresponding
quantity. Given a stable and invertible model, A(L)p", = B(L)Z", -I p; that is,

y Yt
Figure 1. Static welfare analySiS
 189

given no other shocks, the price path converges to its new equilibrium level. The
expected level of prices for t periods following the shock is EPt = (QtZo +p).
It is clear that the welfare loss at Pt' relative to p, is given by triangle abc.

This area is given algebraically by halving the product of the base and height of
abc:
(19)

Adding and subtracting S(p) = D(p) and taking a first-order Taylor series
expansion about p, the welfare loss at any time t is
WL t = ~ ~B + A)(pf _ p)2 (20)
where A represents the slopes of the supply equations and B the slopes of the
demand equations. Note that this measure depends on the squared deviation of
prices from the eventual equilibrium level and slope parameters of the supply and
demand functions. This result is similar to the partial equilibrium measure of
Stein for welfare losses due to futures market price inaccuracies. Because A and B
represent ffIj/fJp and -(aD/fJp), respectively, the welfare loss expressed in terms
of the supply and demand elasticities at (pS) is:

WLt=~~ T/ s [T/d-T/s][pf_p]2 (21)

P T/ s
here T/ s' T/d define the supply and demand elasticities, respectively. The total
consumer and producer surplus at (pS) is given by the area of triangle deb in
Figure 1 which, when expressed in terms of elasticities, is
1 __ Al (p/y) + Bl (p/y) 1 __ [T/d - T/ s]
TS = "2" py = "2" py - - - . (22)
B 1 (p/y)A 1PY T/sT/d

Dividing equation (21) by equation (22) gives the percentage of total surplus lost
as a result of the deviation of futures prices, i.e.,
nWL = T/2(1/-)2(
s p P -p .
f _)2 (23)
Representation (23) has several advantages. First, the expression is solely in
terms of the elasticity of supply, the new steady-state price, and the squared
deviation of prices during the adjustment period. The last two variables are
known for each market; therefore, the percentage welfare loss may be expressed
in terms of one parameter, the elasticity of supply. Another advantage is the lack
190

of scale for this loss measure. This allows various markets to be compared
regardless of their size.
The form of this welfare measure depends upon several assumptions. First,
the results hold only to the extent that the linear representation approximates
the supply and demand relationships. Note that, in this measure, futures prices
affect welfare through expectations, and the critical parameters are those
appearing in the spot market supply and demand relationships. The dependence
of (23) on the supply elasticity rather than both suppll and demand elasticities
occurs because of the linear structure and because p in the supply equation
determines the quantity in any period. Once one component of the supply and
demand relationships determining the surplus measure is fixed (i.e., the supply
elasticity at (p,y)), any change in the other component has offsetting effects on
total surplus and welfare loss. That is, rotating the demand curve clockwise
around (p,y) in Figure 1 proportionately increases both the welfare loss (21) and
the total surplus (22), leaving their ratio unchanged. The supply elasticity
becomes the scaling factor in (23) because the expected price determines the
quantity via supply.
Of course, to obtain the total welfare loss for the adjustment period, one
should discount losses at future dates by some discount rate. Specifying the
number of periods for an arbitrary amount of the total adjustment to occur, the
total welfare loss due to the deviation of prices is a function of the discount rate,
elasticities of the supply and demand functions, the number of periods for the
adjustment to occur, the new steady-state level of prices and quantities, and the
squared deviations of futures prices from the new steady state following a shock.
The first three parameters are assumed to be constant over the adjustment
period, so welfare loss may be viewed as a function of the dynamic adjustment
path of prices. In other words, the welfare loss depends on the accuracy (squared
deviations) and the speed of convergence of the price series.
A crucial assumption for this view of the efficiency of observed price series
is the nature of the long-run equilibrium in an environment of slowly adjusting
prices. Futures markets clear each day, and allocative decisions involving futures
prices may be made during the adjustment of prices to their new equilibrium
levels. If the final equilibrium is affected by the series of temporary equilibria of
all related markets, any measure of welfare loss based upon the final equilibrium
which it obtains must be incorrect. The long-run equilibrium which would exist
in the absence of lagged adjustment of prices is unobservable; so the new
 191

efficiency measures developed in this paper, while correcting for some of the
possible misspecification of previous studies, may remain only partial evaluations
of total efficiency.
Although the loss measure in equation (23) provides a convenient
comparison of the relative efficiency of various markets, the absolute levels of
welfare loss remain informative. The measurement problem that arises for the
absolute welfare loss (22) is the absence of observations of y, the equilibrium
quantity. Although some approximations of yare used in the empirical section,
the approximations may be crude. A loss measure consisting of the forecast error
weighted by the particular market's importance to the economy is both easily
determined from available data and useful in assessing the total welfare loss in
each market.
Dividing both sides of equation (21) by Cp,y) yields

VVL ~ [pf _ 1>]2

-=-=- = ~ ...! (~d - ~s) -_- (24)
yP ~d P
This value expresses the welfare loss scaled by the total sales in the market or, in
other words, the welfare loss per unit of revenue. An attraction of this quantity is
the absence of unobserved equilibrium quantities from the right-hand side of the
expression. The scaled welfare measure depends only on the elasticities of supply
and demand, the steady-state price level, and the squared deviation of prices.

4. METHODOLOGY, DATA, AND EMPIRICAL RESULTS

To capture empirically the dynamic price linked paths, a multivariate time series
model is specified for an eight-market system. This model incorporates the
relationships of T -bills and exchange rates (British pound, Canadian dollar,
German mark, and Japanese yen - all in cents per unit of foreign currency) with
corn, cotton, and wheat prices. As mentioned earlier, the dynamic interactions of
the price series depend on the structural relationships of the underlying supply
and demand functions. The existence of these interactions among agricultural
commodities, as well as the relationship of agricultural commodities to interest
and exchange rates, has been documented in numerous studies (Rausser, 1985).
The accuracy of the estimated effiCiency for each market and for the system
depends critically upon expressing the dynamic interactions adequately by
identifying a suitable time-series representation. VVe chose vector ARMA
representations because of their parsimony relative to the more widely used
192

vector autoregressive models. Nevertheless, large-vector ARMA models still fall

into the general class of overparameterized models, implying that some sort of
restrictions other than identifying the order of the autoregressive and moving
average polynomials may become necessary (Sims, 1980).2
The data used to estimate the vector ARM A consists of 205 observations
for the March, 1981, delivery contracts of the eight variables mentioned above
and 195 observations for March, 1982. These data span the period beginning in
the spring of 1980 through the spring of 1982, a period during which financial
markets adjusted to the new Federal Reserve (Fed) policy of targeting the money
supply rather than interest rates. The two sets of observations provide estimates
of pricing and allocative efficiency immediately following the Fed policy change
(1981 delivery contracts) and much later (1982 delivery contracts).
The choice of the steady-state vector of prices used in the dynamic analysis
influences all of the subsequent results. The welfare measures developed in
section 3 depend on the steady-state level of prices. In addition, since the models
all have autoregressive terms, the new equilibrium price levels, as well as the
dynamic adjustment paths, depend upon the initial steady state chosen.
The initial equilibrium of each series is obtained by forecasting from the
end of the time series until no further change in the variables is observed. This
approach, of course, provides only one of many possible steady-state levels for
the vector of series. At any point in the sample, one could assume that there are
no further shocks and find a different steady-state level. Anyone of these steady
states is preferable to some ad hoc level, such as the mean for each series, because
the simultaneous observation of all series at their mean level may be highly
unlikely. Since there is no definite trend in the price series, the choice of the last
observation, rather than another simultaneously observed set of prices, will not
affect the results.

2 One option is to use t-tests to set individual elements of parameter

matrices to zero (Tiao and Box, 1981), reducing the degree of
overparameterization. The undesirable decrease in statistical power due to the
extra coefficients, therefore, may be reduced by constraining particular values to
zero. The increase in power is achieved, of course, at the risk of biasing the
remaining parameter estimates. Since there is little ,Prior information concerning
parameter values used in the vector ARMA models (in particular, whether or not
to include variables in certain equations), the possibility of biased parameter
estimates is high. One should, thus, avoid selecting extremely low significance
levels for any tests. The major concern in this study is not hypothesis testing but
in reflecting as much of the dynamic adjustment as possible.
 193

The degree of correlation among the series in this study may be highly
positive or negative, so a simultaneously observed set of prices was chosen - the
last observation vector. Then, assuming no further disturbance, the estimated
parameters are used to compute successive forecasts until there is no change in
the forecast price. The estimated models are stable and invertible, so the
forecasts converged to the equilibrium level, ii - usually in about 25 periods.
The proper type of shock to consider when calculating the multipliers is an
interesting question relating to the selection of the initial steady state. Most
authors who construct vector autoregressive models analyze the dynamic
properties of their models by using one standard deviation of a' single series as a
shock. In other words, one element of the shock vector is the standard deviation
of the corresponding error series, and all other elements are zero. The probability
of observing this particular shock may be extremely low, especially when the
residual error terms are correlated. For example, given positive expected
correlation of interest rate and the costs of storage, one might not expect to see
large positive shocks both in T-bills and all commodity prices. Even if all the
series are poSitively correlated, adjustment by the correlation matrix will reflect
the relationships among errors.
The plausibility of the shock is very important since the resulting dynamic
patterns are used to construct empirical measures. As mentioned earlier, positive
shocks in one market or set of markets might be associated with a particular type
of shock in a related market. An arbitrary choice of a particular shock might
obscure this empirical relationship in the efficiency measure.
Consideration of a large positive shock in one leading market and none in
another should yield a different adjustment pattern than a simultaneous shock in
several markets. If some particular type of shock rarely occurred and, therefore,
hardly affected the estimated relationships among the price series, then one
should not use it to calculate the efficiency measure for the entire sample. In
other words, the most likely shock during the sample period should be used to
summarize the relative efficiency of the markets. Accordingly, a multimarket
shock is employed here. It is generated by a one-standard deviation vector of
errors from the fitted ARMA model multiplied by the empirical correlation
matrix of the errors. This procedure yields the best estimate of the signs and
relative magnitudes of the elements of the shock vector given the observed data.
Multiplication by the correlation matrix adjusts each standard deviation by its
correlation with all other series times the standard deviation of that particular
194

series. This perturbation helps to demonstrate the dynamic interactions of the

series and allows calculation of the empirical efficiency measures resulting from
the single shock.
Fitting an (AR(1)MA(6,13,18)) model to the differenced 1981 data and an
(AR(1)MA(5,6,8,14,15)) model to the differenced 1982 data yield the parameter
estimates necessary for constructing the efficiency measures. Graphs of eight
markets in terms of these efficiency measures, accuracy and speed, are given in
Figures 2 and 3. Figure 2 effectively summarizes the previous comparison of the
dynamic behavior of each market from the perspective of the absolute deviation
measure of accuracy. Figure 3 displays a similar summary for a total deviation
measure. In both graphs, the furthermost point on the left for each series
represents the number of periods following a shock required for 50 percent of the
total response to occur. Similarly, the furthermost point on the right represents
the number of periods required for 75 percent of the total fluctuations to occur.
The 1981 Canadian dollar series in Figure 2, for example, achieves 50 percent of
the total adjustment by period 7 and 75 percent by period 17. The German mark,
in contrast, achieves both 50 percent and 75 percent of its adjustment at
point 18. When the deviation measures for two series overlap, parentheses
indicate differing values of the speed measure. To illustrate, in Figure 3, the
British pound and Canadian dollar have total deviation values of -1.28 and
-1.05, respectively, which are too close to distinguish on the graph. The
parentheses surrounding the symbol for the pound indicate that its speeds for 50
percent and 75 percent adjustment were 9 and 15 days, respectively. The vertical
axes for Figures 2 and 3 measure the total absolute deviation and the total
deviation, respectively. The values of these deviations are printed along the right
side of each group.
These results indicate that agricultural markets for 1981 delivery tended to
adjust more quickly and to deviate more than either interest rate or exchange
rate markets. As shown in Figure 2, the agricultural markets achieved 50 percent
of the total absolute deviation by period 7 in 1981. Furthermore, 75 percent of
the total absolute adjustment occurred by period 7 for cotton, period 10 for corn,
and period 14 for wheat. In contrast, both interest rates and exchange rates
generally took much longer to reach either 50 percent or 75 percent adjustment.
The total deviation values in Figure 3 show that the agricultural series
dropped to their final levels while the other markets oscillated about their initial
levels. The combination of large negative total deviations and large positive
 195
Absolute
Adjustment Deviation
50% 75%

cotton 45.3

40 wheat 41.9

corn 33.2
Absolute 30
Deviation
cents/unit
20

........ ••••• ••••• Japanese yen 14.7

•• ••••••••• British pound 12.4
10 ••••••••••••••••• T-bill 10.0

German mark 3.4

Canadian dollar 1.4

5 10 15 20
speed (days)
Accuracy in terms of total absolute deviations and speed 1981

Absolute
Adjustment Deviotion

50% 75%

15 .............. wheot 15.0

Absolute corn 14.1
Deviation cotton 12.8
cents/unit
10

Joponese yen 6.4

............ T-bilis 6.0
5

Canadian dollar 1.5

Brltl sh pound 1.2
German mark 0.7
5 10 15 20
speed (days)
Accuracy in terms of total absolute deviations and speed 1982
196

Absolute
Adjustment Deviation
50% 75...

]
T-bill -0.05
"opon..e yen -0.41
Canadian dollar -1.05
8tltllh pound -1.28
Glrman mark -2.35

Total -'5I
Deviation
wheat -16.61

1
cents/unit carn -18.61

-20
-25

, cotton -29.83
-30~~1----~.----~----•.' ------------------
5 '0 15 20
speed (days)

Accuracy in terms of total deviation and speed 1981

Adjustment Deviation
50... 75...
5
T-blll 4.2
corn 3.8

t.r:::.t::: 3:r.or
8rltllh pound
2.1
.6
.2
o German mark -.05
Total
Deviation Catton -2.0
cents/unit
-5

-10
wIIlat -11.2

5 20
speed (days)
Accuracy in terms of total deviation and speed 1982
 197

absolute deviations of the agricultural markets suggests that they generally fell
after the period 0 reaction to the shock. The agricultural markets overshoot to
the greatest degree followed by exchange rates. The empirical results indicate
that agricultural markets exhibited both greater net overshooting and a faster
speed of convergence in 1980-8l.
For March, 1982, delivery contracts, the German mark continued to show
little deviation and to adjust slowly. The agricultural markets continued to
deviate substantially more in absolute value than exchange rates or interest
rates, but the speed was much more similar to the financial markets than
previously. Indeed, the British pound achieved both 50 percent ~nd 75 percent of
its total adjustment faster than any other series, while the speed for the
agricultural series was similar to that of the Japanese yen and T-bills.
The total deviation measures in 1982 are markedly different from 1981. In
contrast to the --{l.5 to -30 range for 1981, the total deviation varied from 4.2 for
T-bills to -11.2 for wheat; and the distinct difference between agricultural and
financial contracts observed in 1981 is less clear.
The empirical dynamic adjustment paths can be used to compute the
welfare measures developed in section 3. Figures 4 and 5 express the percentage
of total welfare lost as a result of the dynamic path of prices for 1981 and 1982,
respectively. The percentage of total welfare lost as a function of squared
elasticity of supply are represented along the vertical axis in these figures.
The figures show clearly that a much greater percentage of total welfare in
the cotton and Japanese yen markets is lost in both 1981 and 1982. The other
series have values markedly smaller, suggesting that, relative to total trade of the
commodities, the cotton and Japanese yen markets were inefficient allocatively.
Of course, this observation depends upon an assumption that the elasticity of
supply for yen and cotton is nonzero and not drastically different from the other
exchange rates and agricultural commodities, respectively.
The elasticity of supply for both yen and cotton in 1981 must be
approximately .[25, while the elasticity of supply for the other commodities must
be approximately fi for there to be a roughly comparable percentage welfare loss
in all the markets. Although there are no empirical estimates of daily
supply-and-demand elasticities, general assumptions regarding relative elasticities
may be made. It seems reasonable to assume that there is some degree of price
responsiveness, even on a daily basis. The difference in the loss measure becomes
more pronounced if all series have roughly similar elasticities of supply. The
198

Figure 4
1981 Percentage Welfare Lass

Percent
Welfare Loss
xlOOO

80
o cotton
+ Japanese yen
• T-bills
60 v corn
x wheat
o German mark
• Canadian dollar
40 ll. British pound

Squared Supply Elasticity

 199

Figure 5
1982 Percentage Welfare Loss

Percent
Welfare Loss
x 1000
o cotton
+ Japanese yen

60 • T-bills
v corn
x wheat
[J German mark
• Canadian dollar
40 I:;. British pound

20

Squared Supply Elasticity

200

similarity of the relative values in 1981 and 1982 suggests that greater welfare
loss may be endemic to the yen and cotton markets.
Daily quantities supplied or demanded are necessary to calculate the
welfare loss. Quantities supplied are readily available for agricultural markets,
and the Federal Reserve Bulletin reports average daily trade in T -bills. The only
source of data regarding the volume of spot market currency transactions seems
to be a sampling done by the New York Federal Reserve Bank every three years.
The last available data, sampled for the month of April, 1983, are given in Table
1. These monthly trade volume numbers provide an estimate of the daily volume
which may be used to estimate the total welfare loss in the exchange markets. 3
The welfare measure can be rearranged to be an expression involving two
multiplicands, specifically:

WL, ~ [:; ('d - ,,)] [[;,][~n (25)

The first term, involving elasticities of supply and demand, shall be denoted
subsequently as the elasticity multiplicand. The second term, hereinafter called
the deviation multiplicand, consists of the squared forecast error, constants,
market revenues, and the squared equilibrium price.
The deviation multiplicand and its components are given in Table 1. The
deviation multiplicand indicates that the loss due to deviations in the
agricultural markets is minuscule compared to the losses in the T -bill, German
mark, and Japanese yen markets if the elasticity multiplicands are of similar
magnitude across markets. That is, if the elasticity multiplicand, Tls/ Tld( TId-TIs)'
is roughly comparable, the welfare loss is much less in the agricultural markets.
The greater deviation of prices for agricultural series are more than
counterbalanced by the large volume of trade in the financial markets.
The deviation multiplicand indicates that the Japanese yen, followed by
T -bills and the German mark, should exhibit the greatest daily welfare loss due
to slowly adjusting prices. The relatively small squared deviations of T -bill prices
are offset by its enormous size, causing any deviation of prices to cause a great
welfare loss. The Japanese yen exhibits the highest squared deviations among the
financial markets, and its relatively large volume gives it a large welfare loss.

3 The exchange rate volumes in Table 1 may overstate actual volume by up

to 25 percent due to double counting.
 Table 1. Welfare Loss Measure

Squared Yearly Steady Yearly Deviation

Series deviation volume* state value multiplicand

1981 1982 1981 1982 1981 1982 1981 1982 1981 1982

billion million
dollars dollars dollars dollars

US Treasury bills .27 .56 3707.0 4618.0 .851 .869 3155.0 4013.0 .515 .634

British pound .64 .04 t 2.21 1.84 8611.0 861.0 .060 .003

Canadian dollar .02 .04 .838 .810 318.0 318.0 .004 .006

German mark .04 .02 .486 .416 1929.0 1929.0 .147 .078
Japanese yen 1.22 .93 .469 .421 1006.0 1006.0 3.14 1.56

Corn 2.71 1.85 8.3 9.2 3.50 2.54 29.0 23.5 .002 .001

Cotton 4.11 4.52 10.7 15.2 .847 .628 4.3 4.6 .010 .008

Wheat 6.61 2.80 3.8 4.0 4.40 3.51 16.7 14.0 .001 .0005

* Dollars for Treasury bills; million bushels for corn and wheat; and thousand bales for cotton.
t Blanks indicate no data available.
0
'"
......
202

Agricultural markets have squared deviations about ten times greater than the
financial markets, but their trade value makes them have a relatively low welfare
loss.
Figure 6 depicts the trade-{)ff between elasticities of supply and demand for
the elasticity multiplicand. For welfare losses in agricultural markets to be as
large as in the T -bill market, the elasticity multiplicand (k in Figure 6) must be
roughly 100 times greater for agricultural markets than for T -bills.

20

15

10

5 k= 10

k= I
2 3 4 5

Figure 6. Elasticity multiplicand trade-{)ffs

 203

5. CONCLUSION
Allowing for varying flexibility among exchange rates, interest rates, and
commodity markets and dynamic linkages among these various markets,
overshooting is revealed as a common empirical phenomenon. For the eight
futures markets investigated (T-bill, the British pound, the Canadian dollar, the
German mark, the Japanese yen, corn, cotton, and wheat), overshooting occurs in
formation of expectations for each market.
Although interest rate, exchange rate, and commodity markets are all
shown by the estimated vector autoregressive, moving, average model to over-
react to an initial shock, commodity markets (corn, cotton, and wheat) do so to a
much greater degree than either exchange rate or short-term interest rate
markets. However, the period length of this overreaction, for a major portion of
the degree of disequilibrium, is much shorter for the agricultural commodity
markets. In the context of resource allocation decisions, the dynamic welfare
measures reported suggest that the cotton and yen markets have the greatest loss
as a proportion of the total consumer and producer surplus in each. For
comparable elasticities of supply and demand, the total welfare losses are found
to be the largest in the short-term interest and Japanese yen exchange-rate
markets.

REFERENCES
Dornbusch, R (1976), 'Expections and Exchange Rate Dynamics', Journal of
Political Economy, 84, 1161-76.
Frankel, J (1979), 'On the Mark: A Theory of Floating Exchange Rates Based on
Real Interest Differentials', American Economic Review, 69, 610-22.
Rausser, G C (1985), 'Macroeconomics and U.s. Agricultural Policy' in B L
Gardner (ed.), u.s. Agricultural Policy: 1985 Farm Legislation, American
Enterprise Institute for Public Policy Research, 207-52, Washington D.C.
Rausser, G C and C Carter (1983), 'Futures Markets Efficiency in the Soybean
Complex', The Review of Economics and Statistics, 65, 469-78.
Rausser, G C and R E Just (1981), 'Principles of Policy Modeling in Agriculture'
in Modeling Agriculture for Policy Analysis in the 1980s, Federal Reserve
Bank of Kansas City (September 1981), 139-74, Kansas City.
Sims, C (1980), 'Macroeconomics and Reality' Econometrica, 48, 1-48.
Stein J (1981), 'Speculative Price: Economic Welfare and the Idiot of Chance'
Review of Economics and Statistics, 63, 223-32.
Tiao, G C and G E P Box (1981), 'Modelling Multiple Time Series with
Applications', Journal of the American Statistical Association, 76, 802-16.
Zellner, A F and F Palm (1974), 'Time Series Analysis and Simultaneous
Equation Econometric Modeis', Journal of Econometrics, 2, 17-54.
 PART III

MONOPOLISTIC COMMODITY MARKETS

Ronald Anderson surveys the literature on futures trading when the underlying
cash market is imperfectly competitive. The issues addressed include: the
viability of trading futures contracts for goods produced by agents who exert
some control over cash prices, whether futures trading increases or decreases the
market power of dominant agents, and the implications of cash market power for
the bias and volatility of futures prices. The existing studies are largely
theoretical and have examined a variety of market structures including:
monopoly, Cournot-type oligopoly, cartels, and dominant-firm and competitive
fringe. Goods types range from perishable (eg. live animals) to storable (eg.
grains) to durable (eg. precious metals) to exhaustible (eg. crude oil).
The first general finding of this literature is that when the underlying good
is produced under imperfect competition there may be a strategic motive for
trading futures that is distinct from the traditional hedging and speculative
motives. The strategic motives for futures trading vary depending upon the
market structure and the nature of the good. Other conclusions are that for most
imperfectly competitive cash markets futures trading is viable. The possible
exceptions to this are when a powerful agent faces a significant competitive fringe
which is collectively very risk averse or when the desired futures trades of power-
ful agents would enable them to routinely manipulate the futures contracts at
maturity. When futures markets are present and when agents do not cooperate,
they tend to reduce the profitability of powerful agents and to improve the
allocation of resources for three reasons: (1) powerful agents hedge by selling
futures which gives them an incentive to increase their production, (2) the risk
averse fringe hedges and increases its production, or (3) powerful agents seek a
strategic advantage over their rivals by selling futures which serve as a pre-
commitment to produce. There are two cases where these forces do not operate
and where futures trading can increase the profits of powerful agents. First,
futures contracts can aid in the enforcement of production quotas for a cartel
205
206

that is an imperfect monitor of cash markets. Second, when the good in question
is a durable good that is sold rather than rented, the purchase of futures
contracts by powerful agents can commit the agents to maintaining the future
value of the good and thus enhance the initial sale price. Finally, there is a loose
association between the likely competitive benefit of futures trading and the
direction of bias of futures prices: a downward (upward) bias tends to be
associated with large agents being net short (long) futures which tends to
increase (decrease) production.
In Chapter 9, Blaise Allaz makes an interesting contribution to this
literature in the case of a commodity that is storable and sold by duopolists. He
first shows that Cournot duopolists producing a good for which there exists a
futures market will use the opportunity offered by the futures market to sell
forward part of their production in order to modify the equilibrium outcome on
the spot market. This confirms that producers can indeed trade on the futures
market for a strategic motive when the cash market is not perfectly competitive.
He then goes on to show that, when the good is storable, inventories can be used
strategically, just like futures. However, futures dominate inventories as a
strategic tool in the hands of noncompetitive producers. Throughout the chapter,
the analysis is conducted under the assumption of perfect foresight. This strong
and unrealistic assumption is made in order to focus on strategic behaviour and
to ignore the risk hedging rationale for futures trading. It is discussed and
justified at the end of the chapter.
Finally, in Chapter 10, Larry Karp considers a monopsonistic buyer who
imports a nonrenewable resource and uses tariffs to extract rent from competitive
producers. A Markov equilibrium, in which all agents' expectations of future
actions are conditioned on the current level of the stock of the resource, is
studied. Monopsony power decreases as the monopsonist's period of commitment
decreases, but a substantial degree of market power remains even in the limit as
the period of commitment becomes infinitesimal. This equilibrium is contrasted
to a non-Markov time-consistent equilibrium for a nonrenewable resource. Karp
also contrasts the equilibrium for this problem to Markov equilibrium in other
industrial organisation models. In particular, there is an interesting similarity
between the durable goods monopolist discussed by Anderson and the
monopsonist who imports a reproducible good. Karp concludes that the existence
of futures markets may enhance the market power of an agent who is otherwise
unable to make commitments regarding its behaviour in the future.
 CHAPTER 8

FUTURES TRADING FOR IMPERFECT CASH MARKETS: A SURVEY

Ronald W. A nderson l

L INTRODUCTION
The purpose of this essay is to survey the literature concerning the role of trading
futures contracts when those contracts call for the delivery of a good which is
produced under imperfectly competitive conditions. This literature makes up only
a small part of the overall literature on futures markets where the normal form of
the market is typically assumed to be perfect competition. Indeed, for some, the
vigorous rivalry of open outcry trading in a futures market pit seems to epitomize
a perfectly competitive market. Nevertheless, there has been an increasing
awareness that a number of interesting issues concerning futures markets require
a departure from the competitive assumption at least for certain aspects of the
analysis. In the present essay the studies covered share the assumption that in
some respect the underlying market which will determine the price of the futures
contract at its maturity is imperfectly competitive. 2 By and large these markets
take the futures market to be one in which there is free entry and exit so that
there is no natural market power endowed in any futures trader. Thus any
element of imperfect competition in the futures market would be inherited from

I We are grateful to Christopher Gilbert, Larry Karp, Louis Phlips, and

Margaret Slade for their helpful comments. All views expressed and all errors are
the responsibility of the author.
2 The market underlying the futures contract is conventionally called the
"cash" market. The link between the two markets is either because the futures
contract calls for the delivery of the good exchanged in the cash market or
because the terms of the futures contract calls for a cash settlement in an amount
which depends upon the price prevailing in the cash market at a particular time.
207
L. Phlips (ed.). Commodity, Futures and Financial Markets. 207-248.
© 1991 Kluwer Academic Publishers.
208

the underlying cash market.

What are the basic questions that motivate these studies? The first set of
issues concerns the viability of futures trading when the underlying market is
imperfectly competitive. Practitioners sometimes state that an adequate degree
of competition is a necessary condition for the successful launching of a new
futures contract. The intuition for this view is the idea that small traders would
be reluctant to trade a contract when they were aware that some other trader has
the power to affect the price of the contract. While this intuition has some
appeal, it is clear that some departure from atomistic competition is consistent
with future trading. For there is or has been active futures trading for crude oil,
tin, aluminum, copper, coffee, and cocoa. In each of these markets there is either
a large producer or a producers' cartel which possesses sufficient power to
influence the price of the underlying good. Thus we see that the presence of some
powerful producers does not necessarily undermine futures trading.
There is a need then for further analysis to determine if there is some
partial incompatibility between futures trading and market power in the sense
that marginal increases in the degree of market power will tend to depress futures
trading. Alternatively, are there some circumstances in which market power and
futures trading will be compatible and other circumstances in which they will
not?
One possible impediment to futures trading for goods produced under
imperfect competition is that participation of small traders may be discouraged.
An alternative possibility is that the powerful agents would like to discourage
futures trading and may furthermore have the power to suppress it. There is at
least some casual evidence to suggest this might be the case. For example, the
copper futures markets were long boycotted by the large suppliers of copper.
Furthermore, the fact that a diamond futures contract has never been
successfully introduced is sometimes attributed to the resistance by DeBeers.
While these cases suggest there are circumstances when powerful producers may
seek to suppress futures trading, there are other cases, such as tin and coffee,
when the opposite has been true - powerful producers have been active in the
futures market. Again this suggests the need to take a closer look at the issues
through the device of an economic model.
Beyond the question of the feasibility of trading futures for goods produced
by powerful agents is the question of its desirability. In particular, is there reason
to think that futures trading can be used by powerful producers in a way that
 209

enhances their market power with the result that there is a net loss of social
welfare? If so are there public policy actions available which would assure that
futures trading would not be used in such a way? Alternatively, is there reason to
believe that futures trading actively tends to counteract the tendency of powerful
producers to misallocate resources? Were this the case, a conceivable public
policy would be to promote futures trading when cash markets are imperfect.
Our focus is on the case when the cash price is noncompetitive but not
necessarily when the futures price is distorted from competitive levels. When the
latter occurs we have a manipulation of the futures market. Futures market
manipulations, sometimes called corners or squeezes, are generally viewed as
distinct from the general exercise of market power in the cash market. They tend
to be treated in completely separate bodies of law and typically are addressed
with different remedies. 3 Nevertheless, there would appear to be some link
between the two. Often those accused of manipulating the futures market are
powerful hedgers who are in a position to influence prices in the cash market, at
least temporarily. Furthermore, at times the futures trading by large producers
has been taken as prima facie evidence of intended futures manipulations. In
part, the literature we survey has attempted to clarify the relation between these
two forms of market power.
One of the traditional sets of questions addressed in the literature on
futures markets concerns the price performance of these markets. Since most
previous treatments have assumed perfect competition, there is reason to
reconsider the answers obtained when we come to markets with imperfect
competitive producers. In particular, do the actions of powerful producers tend to
give rise the greater biases in futures price than those for competitively produced
products? Beyond the questions of first moments of futures prices are those
concerning volatility. In particular, is there any incentive for powerful producers
either to increase or decrease the volatility of futures prices compared to what
would obtain otherwise?
The literature which addresses these issues is generally fairly recent, in
most cases having appeared within the last ten years. Perhaps as a reflection of
the methods of most contemporary economics, most of the studies are theoretical.

3 American futures markets are covered by extensive laws aimed at

preventing and punishing manipulations. See Johnson (1982) and Edwards and
Edwards (1986). A comparison of American and British treatments of
manipulation is given by Anderson (1986).
210

It will be seen that these models yield a variety of answers to the questions
above. In part, this survey can be read as a compilation of alternative necessary
and sufficient conditions for certain results to hold. Hopefully, this will clarify the
appropriate hypotheses for empirical testing, a task which generally lies ahead.

2. MOTIVES FOR TRADING FUTURES

According to a now classic, if somewhat simplistic, nomenclature most of the
activities of participants in futures markets are either hedging or speculation. 4
Corresponding to these two types of traders are the two accepted functions of
futures markets: risk shifting and price discovery. In the classic view agents who
face unwanted cash market price risk seek to transfer that risk by trading
forward through futures contracts. Any imbalance between short or long hedging
must be met by speculators. However, before speculators would be willing to
accept these risks they must have the expectation of a price change that would be
favorable to their assumed positions. Thus speculators naturally would seek to
accumulate information as to the probable level of prices of the good at some
future date. 5
As we have already noted, this classic view is founded on an assumption of
perfect competition, that is, that neither hedgers nor speculators are in a position
to influence the prices at which they trade. Consequently, if we relax this
assumption to allow for agents who can influence prices, we open up the
possibility that the agents may speculate and hedge in a different manner than
their competitive counterparts or that there may be different motives for trading
futures. If so the market may perform different functions.
In this regard recent developments in the theory of imperfect competition
are suggestive of a new dimension for the use of futures contracts. A theme that
runs throughout much of this literature is that actions in the present which alter
an agent's freedom of action in the future can impart a strategic advantage to
that agent. For example, an oligopolist who undertakes an investment which
expands his productive capacity beyond an amount that is required for

4 A hedger is a futures market participant who has a natural position in the

cash market. A speculator is a futures market participant who has no underlying
cash position.
5 This view of futures trading has been widespread at least since Keynes and
Hicks. For a modern version see Anderson and Danthine (1983).
 211

anticipated demand can deter the entry of potential rivals and thereby maintain
its advantage within the market. 6
By its nature, a futures contract is an unconditional commitment to buy or
sell something at a future point in time. Of course, agents generally do not hold
contracts until maturity; however, in closing out their positions the agents must
make off-setting trades which mean that they fully incur the monetary gain or
loss due to any change in price. Thus futures contracts are effective commitments
to transact at future times. This simple and obvious fact means that, for agents
who behave strategically, futures contracts may have an important additional
function beyond risk shifting and speculation. It is this strategic use of futures
contracts that has been discovered and explored in the literature that concerns
us.
In order to see more concretely how futures trading can have a strategic
role, consider an agent whose profits at time 1 will depend upon the cash price,
P, and who at that time will take some action, x. Let II(P,x) be these profits. If
this agent possesses market power his actions will influence the cash price, which
we represent as p(x). If this is the only source of profits his optimal actions at
time 1 will satisfy the necessary condition, 7
II p' + II = o. (1)
P x
Now suppose that the agent, at some prior time 0, has sold a number of futures
contracts, f, (where £<0 if the agent has purchased futures) at a futures price, P~.
If this position is held until time 1 the profit from the futures position will be

(P~ - P~)f. (2)

The strategic role of futures trading can emerge if the futures price at time 1
depends upon the price the agent can influence. Let the relation of these prices be
given as P~=g(P). Then taking the combined cash and futures profits into
account, the optimal actions by the agent at time 1 must satisfy
II p p' + II x - g'p'f = O. (3)
Thus we see that the future position can change the value of x that the

6 One of the first models of this type was given by Spence (1977). For a
survey of this recent industrial organization literature see Tirole (1988).
7 A subscripted function indicates partial differentiation, and a prime
indicates differentiation.
212

agent will find most profitable. The larger the futures position, f, or the closer the
relation between the cash and futures price, g', the more will the action be
influenced by the futures position. If the futures contract is based on precisely the
good traded in the cash market we will have pi=p so that g' =l.8 The possibility
of strategic futures trading emerges because the futures position may be selected
at an earlier time not because of the possibility of making a profit on the futures
market (i.e. speculation) nor because the fluctuations of futures profits and losses
will balance those on the cash market (i.e. hedging) but because it will influence
the x that will solve equation (3). To pursue these issues further we require a
more detailed formulation of the cash market profit relation. This is done in a
number of models that we survey below.

3. SPECULATION AND HEDGING BY POWERFUL PRODUCERS

The first set of issues confronted in the theory of futures markets for imperfectly
competitive cash markets is to examine how the conventional activities of
hedging and speculation may be transformed by the presence of powerful traders
in the cash market. Perhaps the most widely accepted stylized notion of a futures
market is one where producers facing price risk seek to hedge this risk by trading
futures contracts for their product. Thus a logical starting point for the analysis
of the effect of imperfectly competitive cash markets is to allow for a model of
futures trading where imperfectly competitive producers are able to hedge and
speculate in futures in a way that has no particular strategic significance. This
was studied first for the case of monopoly by Anderson and Sundaresan (1984)
and later for the case of oligopoly by Eldor and Zilcha (1986).
Following Anderson and Sundaresan consider a monopolistic producer of
the cash good; then its cash market profit function can be written as
II(P,x) = p(x)x - c(x) (4)
where x is now interpreted as the monopolist's production and c(x) is the total
cost of production. If the monopolist's good is precisely that called for in the
futures contract and if the contract matures at the time the cash market meets,
then the optimal output level for the monopolist satisfies

8 In futures markets the difference between the futures price and the cash
price is called the basis. When the relevant cash good is precisely that deliverable
on the futures contract then at contract maturity the cash and future price must
coincide (i.e., g(P)=P). In this case we say there is no basis risk.
 213

p(x) + p' (x) (x-f) - c' (x) = O. (5)

In fact, this is the necessary condition for an interior solution. There is also the
possibility that there can be a corner solution at x=O. By a coincidence of
terminology, this corresponds to what is known as "cornering the futures
market".
Notice that equation (5) can be interpreted as saying the monopolist should
equate marginal revenue to marginal cost. The monopolist's existing futures
position affects his production by shifting the marginal revenue by the amount
_p' (x)f. When the monopolist has sold futures (bO) since p' <0, this means that
marginal revenue is increased. If marginal costs are non-decreAsing the effect of
the futures sales is to increase production above the level that would prevail in
the absence of futures. Also it follows that the greater the futures sales, the
greater the cash output, other things equal. Notice also that if the monopolist has
purchased futures contracts (f<O) the output would be depressed.
The intuition for this result is that the monopolist takes into account the
effect that his production will have on both his cash market profits and his
profits from futures trading. If he has previously sold futures, any action that
tends to depress the price will tend to increase his futures profits. Thus given a
short futures position the monopolist is induced to increase production beyond
the level that would maximize cash profits. The monotonically increasing
functional relation between the powerful agent's futures sales and his output was
obtained first by Anderson and Sundaresan for monopoly and has been extended
by a number of others for a variety of different market structures. It is the key
reason that the models of futures markets for imperfectly competitive cash
markets differ from the conventional competitive theory.
From the perspective of welfare analysis it is now clear for the case of static
monopoly that the key to whether the availability of futures trading improves or
worsens the allocation of resources is to know what determines the monopolist's
futures position. 9 If the monopolist sells futures, production is increased and the
allocation of resources may well be improved. 10 If the monopolist takes a long

9 The futures market taken by itself is a zero sum game so that any profits
and losses would represent a pure income redistribution. Obviously, if there are
such profits and losses they can affect social welfare. These distributional issues
have not been examined in the literature on futures for imperfect cash markets.
10 For f sufficiently large the monopolist would produce to a point where
marginal cost exceeds the market price which is conceivably worse than the
no-futures allocation.
214

position, there is clearly a reduction in production which represents a movement

away from the Pareto optimal point.
Thus the important issue becomes to model the futures market equilibrium
in order to determine the futures position of the powerful producers. In order to
model the determination of futures positions and price at time 1, it is necessary
to specify what are the objectives and information of the agents participating in
the futures market. Anderson and Sundaresan have characterized the rational
expectations equilibrium futures price under the assumptions that (i) the
monopolist faces a finite number of price-taking agents in the futures market and
(ii) there is full public information about all the relevant parameters at time 1. In
particular, an important assumption is that all agents can observe the
monopolist's future position and thereby can form expectations about the next
period's cash price according to condition (5). A number of cases are possible
depending upon whether the competitive traders are speculators or have a
hedging motive for trading futures and depending upon whether the monopolist is
risk-neutral or is risk-averse and thereby potentially has a hedging motive for
trading futures. The futures market analysis of Anderson and Sundaresan can be
summarized in the following.

Proposition 1 (Anderson and Sundaresan):

Suppose competitive futures traders are risk-averse speculators and there is
public information. (1) If the monopolist is risk-neutral, then in the unique
rational expectations equilibrium the futures price is unbiased and all agents hold
zero futures positions. (2) If the monopolist is risk-averse, then in equilibrium the
monopolist may either hold long or short futures positions depending upon the
nature of the demand and technological uncertainty.

The reason for part (1) of the proposition is that it can be shown that (a)
the risk-neutral monopolist will hold a position only if he expects to earn a profit
in the futures market and (b) given public information pure speculators would
always seek to hold long positions when the monopolist is short. Consequently,
the only time the market will clear is if all agents wish to hold no position which
occurs if the futures price is an unbiased predictor of the cash price. This result is
similar to other results for rational expectations equilibrium in the purely
 215

speculative markets. It appears likely that it would extend to the case of private
information if all agents had common priors. 11
Part (2) of the proposition shows that a producer will hedge short or long
depending on the correlation of his output with demand. Under the assumption
that either the costs are nonstochastic or that they are not positively correlated
with the demand uncertainty, the monopolist will sell futures. Since the
competitive agents are pure speculators this implies that the equilibrium price is
biased downward, i.e., the futures price must be expected to rise in order to
induce speculators to hold the price risk. Thus in this case we see that the futures
market functions in much the same way as in the classi'Cal Keynes/Hicks
paradigm. More importantly, given the results on the cash market, we see that
introducing a futures market when the monopolistic producer is risk-averse will
tend to improve the allocation in the cash market.
The proposition emphasizes the importance of the dominant producer's
attitudes toward risk. The large wealth of some dominant producers might
suggest relatively high risk tolerances. However, the often very large
concentration of that wealth in a single commodity points towards relatively low
risk tolerances. Thus the nature of dominant producer's risk preferences an
empirical question. As of yet this has not been investigated.
When the competitive agents are themselves hedgers the results depend on
the balance of hedging interest in much the same fashion as in competitive
markets (see Anderson and Danthine, 1983). For example, the competitive agents
may be consumers whose hedging leads them to take long futures positions. If the
balance of hedging were long, this would induce a positive bias to the futures
price (i.e., futures would tend to fall over time). This in turn would induce the
monopolist to reduce his short positions. It is conceivable that this incentive
would be strong enough to induce the monopolist to go long futures. In this case
the futures market could exacerbate the misallocation of resources in the cash
market.
The preceding results clarify some of the basic issues of the role of futures
markets when the underlying good is produced monopolistically. it is unclear how
much of this analysis also applies to an industry with a small number of firms.
The literature on oligopoly amply demonstrates how seemingly small differences

11 See Tirole (1982) and Milgrom and Stokey (1982).

216

in the structure of industries can give rise to very large differences in the
outcomes. 12 Futures trading which by its nature involves commitments for
intertemporal transacting potentially could affect cash market competition in
very complicated ways. The oligopolistic market structure which would likely
have the most straightforward interaction with futures trading would be a static
oligopoly where cash market decisions are taken simultaneously with futures
market decisions. This is the structure that was examined by Eldor and Zilcha
(1986).
The Eldor-Zilcha model is a conventional Cournot oligopoly extended to
allow for demand uncertainty and futures trading. 13 Specifically N firms producing
homogeneous products and facing identical cost functions simultaneously choose
output levels and futures positions. Then one period later the demand
uncertainty is resolved, the output is sold on the cash market, and the futures
contracts mature. The firms are assumed to be risk-averse. The behavior of the
firms is non-cooperative so that they examine a Nash equilibrium in the cash
market. Thus the structure differs from the one Anderson and Sundaresan
described above principally in that there are more than one producer and output
and initial futures positions are determined simultaneously. Eldor and Zilcha's
principal results can be summarized in the following propositions:

Proposition 2 (Eldor-Zilcha 1):

Let firm i's output be xi and its futures sales be fi . Then xi=fi if and only if
E(p)=l xi~fi if and only if E(p)~l xi~fi if and only if E(p)~l

Proposition :1 (Eldor-Zilcha 2):

If pf=E(p) then in equilibrium the output of each firm will produce more than it
would produce in the absence offutures.

The first proposition is similar to that for the case of competitive firms in the
absence of technological uncertainty and basis risk (see Anderson and Danthine,

12 The classic example of this is the difference in the Cournot and Bertrand
analyses of duopoly.
13 They refer to this as "forward trading". In this context forward and
futures trading are equivalent.
 217

1983). The (imperfectly competitive) producer has two motives for trading
futures: hedging and speculation. When futures are unbiased (E(p )=pf) only the
hedging motive is operative and the producer sells his output forward. When the
futures price is biased upward the speculative motive leads the producer to
overhedge. When the futures price is biased downward (so that a short futures
position is expected to lose money) the specnlative motive leads the producer to
underhedge. Depending upon the degree of bias and the producer's risk aversion
it is conceivable that the producer wonld buy futures.
The second proposition is also similar to others establish~d for competitive
firms. If the future price is unbiased the producer is able to eliminate all risk by
selling its output forward. (Recall there is no output uncertainty). In view of the
firms' risk aversion this means that they wonld be willing to produce more than
they would otherwise. The innovation on the part of Eldor and Zilcha is to show
that this will hold for a Nash equilibrium in the model they study. It should be
noticed that their two propositions fall short of demonstrating that in their model
futures markets necessarily improve the allocation of resources over that achieved
in the oligopoly without futures. The reason is that they do not study the
determination of the equilibrium futures price. They do not discuss this, but it is
clear that the futures price would necessarily be unbiased in a rational
expectations equilibrium if (a) there were public information and (b) there were
at least one risk-neutral speculator trading futures. In other cases, such as if the
others trading futures were a finite number of risk-averse speculators or if
consumers were hedgers, the effect of futures trading is ambiguous. However, to
the extent that the futures market is dominated by the pattern of risk-averse
oligopolists being met by a sufficient supply of speculation, the result that
"futures trading causes output to rise and price to fall" is likely to be a good
approximation.

4. HEDGING BY THE COMPETITIVE FRINGE

In the analysis of the previous section the futures trading can influence the cash
market through the supply decisions of the powerful (i.e. price-making)
producers. Specifically, the powerful producers face an exogenously given net
demand schedule which is uninfluenced by the introduction of futures trading or
by changes in the level of the futures price. This construction is general enough to
allow this demand schedule to be net of sales by cash-price-taking producers. In
this sense, the results above apply to cases when there are small competitive
218

producers as well as the powerful ones. However, a crucial implicit assumption is

that competitive participants in the cash market (either producers or consumers)
do not trade futures. For were they to trade futures there would be an additional
influence of futures trading onto competitive cash market supply or demand.
Newbery (1984) investigates the implications of allowing a feed-back of
futures trading onto competitive supply. In his model a perishable good is
produced by a single powerful producer and a competitive fringe consisting of a
large number of identical, price-taking firms. Output is uncertain and depends on
a single random variable that affects all firms multiplicatively (i.e. all firms will
receive the same yield relative to expected output). Th'ere is a futures market
which meets at the time the agents make their production decision and there is
no basis risk. The participants on the futures market are the dominant producer,
the competitive fringe producers, and a number of price-taking, risk-averse
speculators. Consumers of the cash good do not trade futures.
An important assumption for Newbery is that the typical competitive fringe
producer is risk-averse so that hedging will be an important motive for trading
futures. In contrast, it is assumed that the dominant producer is risk-neutral and
thus is not interested in futures as a hedging device. As discussed in section 3 the
comparative risk tolerances of powerful producers appear unclear on theoretical
grounds. Thus it should be borne in mind that this analysis is likely to apply to
markets where the predominance of hedging is done by small price-taking
producers of the good.
In order to establish the general tendency created by fringe producer
hedging Newbery examines the special case where the dominant producer has no
futures position. Under the particular parametric assumption of identical
constant absolute risk-aversion for all agents he shows the following proposition
holds.

Proposition 4 (Newbery 1):

Suppose that the competitive fringe and competitive speculators trade futures
but that the dominant producer does not. If all agents take the cash price
distribution as given, independent of the aggregate futures position of the fringe,
then in equilibrium (a) the fringe sells futures and (b) the expected supply by the
fringe is greater than that which would hold in the absence of futures.

This is similar to other propositions within the literature on competitive futures

 219

markets which show that hedging by risk averse producers tends to lead to
increased production other things equal. This is not a final prediction of the effect
of futures trading within Newbery's framework because it does not take into
account what the dominant producer's futures market decision. will be. Rather it
describes an aspect of the residual cash market demand faced by the dominant
producer. In effect, futures trading by the fringe tends to depress residual cash
market demand. Furthermore, actions which would tend to raise the futures price
(e.g., purchases of futures by a large trader) would tend to increase fringe
production and depress cash market net demand.
In considering the optimal futures market decision for the dominant
producer, the supply effect is only one consideration. There is also the speculative
aspect. Here we see from (a) in the proposition above that if the dominant
producer has no position the equilibrium future price will be biased downward.
(There must be a tendency for the future price to rise or else the speculators wold
not buy the futures contracts sold by the competitive fringe.) Thus fringe hedging
tends to create a profit opportunity for the dominant producer. Furthermore, to
profit from this, the dominant producer would buy futures. Thus we see the
dominant producer faces two, conflicting tendencies. To make an expected profit
on the futures market the powerful producer would tend to buy futures. However,
this action would tend to raise the futures price which would have the effect of
increasing the supply of fringe producers and thus depressing the dominant
producer's cash market profits.
It is worth emphasizing a technical property of this framework which may
appear surprising in light of results on purely competitive speculative markets.
The dominant producer is assumed to be risk neutral; however, its optimal
futures position is finite even if the futures price differs from the expected cash
price. The reason for this is that the dominant producer recognizes that its
futures position will influence the cash price. Consider the decision of the large
producer starting initially with no futures position. In light of the proposition
above the futures price would be biased downward and there would be a tendency
to purchase futures. As the dominant producer would purchase futures the agent
realizes that the futures price would tend to rise. This has two effects. It
decreases the expected profits on the inframarginal futures purchases holding the
expected cash price constant. Beyond this it would tend to increase fringe supply
and decrease the expected cash price thus depressing both the expected futures
profit and the expected cash market profit. This implies the dominant producer
220

would increase its long futures position only to the point where these influences
exactly equal the expected profit per unit of futures positions.
The diverse incentives faced by the dominant producer mean that their
implications for equilibrium are correspondingly complex. In general, it would
appear that a variety of qualitatively different results are possible including those
in which the dominant producer may be either long or short futures. Newbery
gives parametric examples in which the dominant producer is long futures and
extracts an expected profit from the futures market. Nevertheless, he
demonstrates that, if somehow the dominant producer were able to precommit
himself to not trading futures,14 in the resulting equilibrium he would be better- off
in the sense that his expected profit would be greater than the combined cash and
futures profits when he is active in trading futures.
Newbery emphasizes that the preceding example establishes the possibility
that a dominant producer has an incentive to suppress futures trading by the
competitive fringe and, even, to suppress the futures market itself. In fact, there
is some evidence that in the case of certain metals, large producers have
discouraged the creation and growth of futures markets (see, e.g., Slade, 1989). In
his initial study Newbery did not investigate the feasibility for the dominant
producer of suppressing futures trading. He did however provide a discussion of
how the dominant producer may choose to randomize its cash market supply in
an effort to increase the cash price uncertainty faced by the competitive fringe.
This could have the effect of depressing fringe supply and thus on average raising
the cash price.
In a comment on Newbery, Phlips (1984) argues that the most plausible
means of futures market suppression would be to do precisely the opposite, i.e.,
to stabilize the cash market by announcing in advance the cash market price so
that competitive producers and consumers would face no uncertainty and would
have no incentive to trade futures. As evidence for this case he cites the case of
Belgian fertilizer producers who cooperated by every year announcing the price
schedule that would hold for the next year. 15 It is clear that such a policy of price
stabilization is most readily achievable if the good is storable. Subsequently,

14Technically, if his strategy space was restricted to have zero futures.

15 Another case of an apparent attempt to suppress futures trading took place
in the 1960's when copper producers maintained constant prices with the result
that hedging interest in the futures suffered. See Prain (1975).
 221

Newbery has considered the question of the feasibility of futures market

suppression by a dominant producer within a model of a storable good. We will
discuss this below after we have introduced the possibility of storage into the
analysis.
The analysis of Newbery did not allow for the possibility that the users of
the cash good, who also face price uncertainty, might hedge their risk in the
futures market. In principle, this could be the case. In light of the analysis of
fringe producer hedging, it would appear that hedging by competitive users of the
cash good would increase the demand for the good. For example, if risk-averse
buyers of the good hedged, this could lead to an increase in deIhand since there is
less uncertainty about a source of inputs. This would benefit the dominant
producer. Furthermore, to the extent that long hedging dominates, there would
be a tendency for the futures price to be biased upward. Thus there would be a
speculative incentive for the dominant producer to sell futures. That would tend
to depress fringe supply of the case good thus to increase expected cash market
profits. The dominant producer is likely to find the establishment of the futures
market to be profit-enhancing. As of yet there has been no analysis of a market
with a dominant producer facing risk-averse consumers who hedge on the futures
market; consequently, the argument above is still a conjecture.

5. STRATEGIC FUTURES: COURNOT OLIGOPOLISTS

In the previous section we have summarized the implications of powerful
producers trading futures to reduce risk or to influence the production decisions
of competitive producers and consumers. We now investigate how they may
choose futures positions in order to shape their own future production decisions.
In doing so we focus squarely on the strategic role of futures trading first
discussed in section 2.
In a series of studies Allaz (1987, 1989) and Allaz and Vila (1986)
considered futures trading for a good that will be produced by Cournot
duopolists. The important difference of this framework and that of Eldor and
Zilcha (1986) is that futures positions are chosen before the production decisions
are made. We have already seen in the study of monopoly that a pre-existing
futures sale tends to increase the powerful producer's output. This continues to
be the case for Cournot oligopolists. However, unlike the monopoly case, the
prior futures sale can impart an advantage to the oligopolist in that it establishes
a pre-commitment to produce similar to that of a Stackelberg leader. Thus an
222

oligopolist has a strategic incentive to sell futures. Of course, its rivals have the
same incentive, so that in equilibrium they will both tend to sell futures. As a
consequence,

Proposition 5 (Allaz):
Suppose quantity-setting oligopolists can trade futures prior to choosing output.
Then if there is perfect foresight, in Nash equilibrium each oligopolist sells
futures resulting in greater industry output and a lower cash price than would
prevail in the Nash equilibrium without futures.

The proposition shows that, even if large producers are not likely to hedge
in the futures market, the availability of the futures market tends to improve the
allocation of resources. It is worth illustrating this insight through a parametric
example. Consider a Cournot duopoly with linear demand given by
P=a-bx 1-bx 2 and constant and equal marginal costs given by c. (Here xi is the
output of firm i.) If the firm i has made prior futures sales of fi and if there is no
basis risk, the necessary condition for an optimal output decision of firm 1 leads
to the reaction function,
A ~ f1
~=y-~+~ W
where A=(a-c)/b. A similar expression holds for firm 2. Notice that an increase
in the firm's futures sales f1 shifts out the firm's reaction function.
If neither firm has prior futures sales the resulting cash market eqUilibrium
is the familiar Cournot solution where xl =x2=A/3. This is shown in figure 1 as
the intersection of the no-futures reaction functions depicted by solid lines. If
instead, one firm were a Stackelberg leader, it could increase its profits by
increasing its output above the Cournot level thus causing its rival to reduce its
output. In the current parametric example, the Stackelberg solution where firm 1
is the leader is xl =A/2 and x2=A/4. Futures sales not matched by a rival's
futures sales can allow a firm to achieve the advantage of Stackelberg leadership
even if production choices will be made simultaneously ala Cournot. We saw in
equation (6) above that futures sales result in a parallel shift of the firm's
reaction function. If for some reason firm 2 did not sell futures its reaction curve
would remain the solid line R2 in figure 1. Thus by selling an amount of futures
~ the leader can shift its reaction curve to that gi ven by the dot ted line in the
figure, which intersects R2 at the Stackelberg amounts. Inserting these in the
 223

x,
A
\
R,
\
\
\
,
A

\
\
R,

X,

4 ¥ A

Figure 1. Effect of strategic futures sales by firm I

equation (6) we see that the appropriate level offutures sales is ~ =A/ 4.
The actual futures positions of the oligopolists will depend upon how the futures
market game at the prior period is played. In particular, it will depend upon the
payoffs of the firms and thus will be affected by any uncertainty that may exist
and the agent's attitudes toward risk. It will also depend upon the nature of the
solution that applies (e.g. whether they play non-cooperatively or cooperatively).
Allaz shows that non-cooperative equilibrium in the futures market tends to
improve the allocation of resources in the cash market. The simplest case is if
there is perfect foresight (or equivalently if the oligopolists are risk-neutral and
the futures price is unbiased). The the relevant payoff functions are obtained by
solving the reaction functions to give equilibrium production levels as functions of
prior futures choices, Xl (fl'f2), x2(fl ,f2), and inserting in the cash market profit
functions. The solution will be the Nash equilibrium in the game using these
payoffs. For the linear example introduced above the solution is, f~=f~A/5 and
Xl (f~,f~)=x2(f~,f~)=2A/5. Note that this implies an increase in industry over
that in Cournot equilibrium without futures of 2A/15.
224

A comment is in order on the use of perfect foresight models to represent

the futures market. In light of the traditional view of futures markets as a device
for hedgers and speculators to deal with uncertainty, this framework might seem
inappropriate. In fact, since the hedging and speculative motives are absent in a
perfect foresight world, it is a very useful device for highlighting the implications
of the strategic motive for trading futures. Of course, realistically there is
uncertainty about price changes in futures markets, so that a complete analysis
would allow also for this so that the combined impact of hedging, speculation and
strategic trading can be assessed.
If the futures choice is made when either the demand or cost on the cash
market are uncertain, potentially the decisions can be modified for risk-averse
powerful producers. Allaz (1989) examines this case under the assumption of (a)
linear demand with an uncertain intercept, (b) constant and known marginal
costs which are equal for all firms, (c) firms maximize expected utility which is a
linear function of the mean and variance of profits, and (d) there is at least one
risk-neutral speculator so that the futures price is unbiased. In this case the
optimal futures position of the producers is composed of two additive
components. The first, strategic, component is the same as that just discussed; in
Nash equilibrium it is a positive number (Le., requires no futures sale). The
second component is a hedging component similar to that found for competitive
futures (see, e.g., Anderson and Danthine, 1983). It can be positive or negative
depending upon the exact shape of the distribution of the prices; however, unless
the distribution is very skewed the hedging term tends to be positive.
Consequently, the hedging motive tends to reinforce the strategic motive. The
risk-averse oligopolist sells even more futures than would be required for strategic
reasons alone and the aggregate cash market output is increased accordingly.
All these results pertain to the case where the oligopolists meet to trade
futures at only one time prior to the cash market decision. Now since futures
markets meet repeatedly over a substantial period of time prior to the contract
maturity, it is conceivable that once they had established a position the agents
may revise their futures positions. This problem has been analyzed by Allaz and
Vila (1986) for the case of perfect foresight, linear demand, and constant
marginal cost. The intuition for their results can be seen by considering the
problem of a firm immediately after it and its rival have selected a futures
position non-cooperatively. As is typically true in Cournot-type equilibria output
levels are less than those at which there is a tangency of the rival's reaction
 225

function and the firm' iso-profit line. Thus, if the rival's reaction function is
unchanged, there is an incentive to increase production; this implies there is an
incentive to revise the firm's futures position by selling additional units. Again
both firms have the same incentives. Thus both will tend to sell more. The same
is true each time the futures market meets. Taking this argument to the limit we
have,

Proposition 6 (Allaz and Vila):

Assume perfect foresight. As the number of meetings of the futures market grows
indefinitely, the accumulated futures position immediatelY' prior to cash
production grows and in the limit approaches the amount that results in price
equaling marginal cost. In this sense futures trading tends to produce competitive
outcomes even for Cournot producers.

6. STRATEGIC FUTURES: CARTEL FUTURES POLICIES

The models that we have reviewed to this point differ in the assumptions that
they make about the objectives of the powerful producers or about the structure
of the choices faced by these producers. Nevertheless, the predictions of the
models are similar: there is a tendency for futures trading to result in greater
industry output than would prevail without futures. Consequently, futures
trading tends to diminish the large producers' market power and result in an
improved allocation of resources. The reasons for this result are roughly that
large producers may sell futures to hedge their risks (section 3), that futures
hedging reduces risk for competitive producers and thus increases their
production (section 4), and that oligopolists may sell futures to obtain a strategic
advantage over their rivals (section 5).
There is another strand of the literature on futures trading for imperfectly
competitive cash markets which recognizes that futures contracting need not
reduce the misallocation of resources on imperfect cash markets. We have already
seen that if powerful producers have purchased futures prior to making their
production decisions they will produce less than they would have otherwise. 16 In

16 The models discussed above did allow for the possibility of powerful
producers purchasing futures, at least implicitly, since if the futures price had a
strong downward bias the speculative incentive would be to go long futures.
However, if the futures market were open to a large group of speculators, it could
226

light of this, large producers might have an incentive as a group to somehow

commit themselves to not sell futures and, possibly, to purchase futures.
The analysis of Allaz (1989) made use of the coefficient of conjectural
variation and, thus, could allow for various types of behavior of the oligopolists
on the futures market. The conclusion given above in section 5 were based on the
Cournot assumption that increasing futures sales does not induce any change in a
rival's futures position. Allaz showed that, under the alternative conjecture that
each unit of futures sold leads the rival to sell one more unit, the futures position
of the oligopolists is long in the perfect foresight case. That is, cooperation by
oligopolists on the futures market tends to result in 'reduced production and
higher cash prices. Under uncertainty the sign of the oligopolist's futures position
is ambiguous since then the oligopolists tend to sell futures to satisfy their
hedging motive.
Allaz's analysis does not investigate in detail the cooperative use of futures
contracts by an oligopoly. A number of issues are of interest. For example, what
forms of cooperative agreements are achievable through the futures market? Why
should a producers' cartel be able to cooperate on a futures market more readily
than in making their production decisions directly? Does cooperation on the
futures market constitute a manipulation of the futures market and thus run
afoul of anti-manipulations laws that may apply to those market? Is cooperation
on a futures market susceptible to being undermined by cheating?
These questions have been addressed by Anderson and Brianza (1989) in
their analysis of policies for futures markets that might be adopted by a cartel as
a means of enforcing production quotas for its members. Their work was
motivated by the fact that in several cases of alleged futures market
manipulation the suspected manipulator was a group of large producers. These
producer groups argued that they were not manipulating the futures market but
were simply using futures as a tool for achieving certain price objectives in the
cash market.17 In investigating this claim, Anderson and Brianza point out the
difficulty of directly monitoring production and sales in a geographically

be argued that the bias would be small in which case the tendencies for futures to
induce increased production would seem to hold.
17 An episode of this type took place in the coffee market and has been
discussed by Greenstone (1981). Another case arose out of the collapse of the tin
market and is discussed by Anderson and Gilbert (1988).
 227

dispersed cash market. This suggests that a cartel would be interested in finding
some means of creating an incentive to respect quotas. As was observed above, if
a cartel were able to assure that members' futures positions were long then
individual members would have an incentive to restrict their production to some
extent. Now it is possible that, in a centralized futures market where there is
some official reporting of volume and open interest and where informal
information flows are often fairly effiCient, a cartel may be able to monitor
members' futures positions at least certain times. In this case, allocating futures
positions to members at a point in time may be a mechanism for achieving
cooperative outcomes noncooperatively on the cash market.
In considering which types of cooperative outcomes could be achieved
through futures allocations, Anderson and Brianza first observe that the range of
agreements that might be mad~ by a duopoly which could enforce production
quotas is given by the Pareto frontier defined as maximum profit achievable for
firm 2 given that firm I's equals a specified amount. (When side-payments are
feasible this is just the possible sharings of the maximum joint profits for the
industry.) They then establish the following,

Proposition 7 (Anderson-Brianza 1):

Suppose that there is perfect foresight. If the penalty to cornering the futures
market is sufficiently high, any pair of profits on the cash market Pareto frontier
can be achieved non-cooperatively given the appropriate initial allocation of
futures positions to the cartel members.

When side-payments are feasible any cooperative agreement involves

maximizing joint profits. In this case the appropriate cartel futures policies take a
simple form: each firm is allocated a long futures position equal to the proposed
output of the other firm. In effect, the firm is committed to purchasing its rival's
output. In the linear example introduced in section 5 above, it is well known that
to maximize joint profits, each firm would produce half of the monopoly output:
Xl =x2=A/4. If we set fl =-A/4 and x2=A/4 in the reaction function (6) we see
that x I =A/4.
In the proposition above the assumption that there is perfect foresight is
principally a convenience. Anderson and Brianza show that under uncertainty
with risk-averse producers there will exist a similar cartel futures policy;
however, if risk aversion is sufficiently strong, the policy may involve going short
228

futures. In effect, the desire to hedge risks may overwhelm the desire to use
futures to restrain production.
The caveat in the proposition that there be sufficient penalties for
manipulating the futures market is an important possible limitation to what can
be achieved by a cartel through cooperative agreements on futures markets. The
problem is that the long futures position for the cartel members may mean that
the optimal production policy for the members would be to produce nothing,
driving up the futures price at maturity to the highest attainable level, and thus
extracting the maximum profits on their futures positions. This is a simplified
description of what is involved in a classic futures market corner. While given the
long futures position this may be the optimal decision of the cartel members,
unfortunately for them this may be anticipated by the market so that the initial
futures price would be bid up so that no futures profit would be earned by the
cartel. For this reason, they would have an interest in finding some means of
making futures manipulations infeasible. Thus, somewhat surprisingly, public
policies aimed at punishing futures market manipulations have the effect of
encouraging cartel future policies to promote the (moderate) restriction of
output.
Finally, Anderson and Brianza consider how member cheating on the
futures market may undermine the effectiveness of a cartel futures policy.
Suppose that the cartel is able to verify at a point in time what the futures
positions of its members are and at that time allocates long futures positions to
them. If at a later time there is no cooperation on the futures market, each
member could sell some futures contracts, reducing its long position and thereby
reducing its incentive to restrict output.1 8 In fact, we have seen from Allaz and
Vila's (1986) analysis of multi-period Nash equilibrium that members would sell
futures each period in equilibrium. However, a cartel that recognizes this
possibility of cheating on the futures market (as well as on the cash market) can
anticipate it and adjust the futures allocations accordingly. In particular,
Anderson and Brianza show the following proposition is true,

Proposition 8 (Anderson and Brianza 2):

Suppose there is perfect foresight and there is noncooperative futures trading on

18 In this perfect foresight context "cheating" simply means acting

noncooperatively subsequent to some initial cooperation.
 229

T days prior to the maturity of the futures contract and the production decision
on the cash market. Then there exists an initial allocation of long futures
positions T+1 days prior to maturity which will result in any point on the cash
market Pareto frontier of the duopoly.

Thus the cartel simply increases its initial allocations by the amount of the
anticipated cheating. A technical complication is that the total cheating depends
on the initial allocations so that the two need to be solved simultaneously. More
importantly, the size of the required initial allocation rises linearly with T, the
number of occurrences of noncooperative trading between the, initial allocation
and production decisions. In principle, the futures markets meet continuously so
that one might argue infinitely large positions are required. The alternative
interpretation of the proposition is that there must exist some friction which at
least slows the pace of noncooperative futures trading if a cartel is to be
successful in implementing production quotas through a futures policy.

1. STRATEGIC FUTURES: STORABLE GOODS

The results that have been discussed so far have not explicitly taken into account
the activity of storage and thus have ignored the links between current cash
market conditions and futures prices. While this is convenient for the analysis,
most futures markets are based on storable goods so that explicit recognition of
storage in the analysis would be desirable. Beyond this, in a number of
commodity markets the most powerful agent is a producer's organization which is
responsible for achieving price objectives and for whom one of the principal tools
is storage of the good.
When powerful agents simultaneously have the possibility of storing their
goods and of taking positions on the futures market there are significant
interactions between the two activities. This raises several questions. First, is the
selection of inventory levels altered by the availability of futures? Second, are
futures trading and storage by powerful producers substitutes in the sense that
one tends to supplant the other as was suggested in section 4 above? Third, what
are the public policy actions which are most appropriate when powerful agents
can store and can they affect the futures market?
We have already seen a variety of models of futures trading for imperfect
cash markets where the goods are perishable. It is not surprising then that there
is a variety of possible market structures for storable goods. Allaz (1990)
230

considers the case of a quantity setting duopoly with simultaneous decisions. The
firms choose production levels in two periods. The powerful firms are the only
producers and are the only agents who store the good. There is perfect foresight.
At first the analysis is carried in the absence of futures. If a competitive
firm could produce now for later sales, its sole objective in storing would be to
reduce cost of the good next period. 19 This leads to the rule that inventory levels
should be set so that the marginal cost of producing now and storing the good
should equal the marginal cost of production next period. When firms are
oligopolists, it has been noted by Rotemberg and Saloner (1985) and
demonstrated in the current context by Allaz that inventory levels will exceed
competitive, or cost minimizing levels. The reason is that for quantity setting
oligopolists, a rival's next period output will be a decreasing function of the
oligopolist's inventories. (E.g., firm 2's period 2 output is a decreasing function of
firm l's inventories carried from period 1 to period 2.) Thus increasing
inventories promises a firm a strategic advantage similar to that of being a
Stackelberg leader. Of course, the same incentive is faced by all oligopolists, and,
in noncooperative equilibrium the effect is to increase aggregate production,
decrease price and profits.
If futures markets are added to the model, still maintaining the perfect
foresight assumption, Allaz demonstrates that in equilibrium the oligopolists hold
smaller inventories than in their absence. In fact, they hold precisely the level of
inventories which equates the marginal cost plus storage of goods from period 1
with the marginal cost of production for goods in period 2. The reason is that in
equilibrium the agents engage in sales of futures contracts for the purpose of
gaining strategic advantage. Carrying additional inventories serves no additional
strategic purpose. In this sense futures markets dominate inventory holding for
strategic purposes. They both serve the purpose of committing the firms to
selling in the future; however, the futures contract does so costlessly.
We have already remarked that assuming perfect foresight in a model of
futures trading can be useful in emphasizing the strategic role of futures as
distinct from the hedging and speculative aspects of futures. It is important to
realize however that results obtained in a perfect foresight model may not extend
directly to the case of uncertainty. In the present context, it would be interesting

19 This rule also applies to a monopolistic producer. It holds true even if

there are sales in the initial period and even if the prices in two periods differ.
 231

to examine simultaneous inventory choice under uncertainty and futures trading.

Uncertainty would appear to exert several influences, not all of them in the same
direction. First, uncertainty would create a hedging role for futures markets
which, as was shown by Allaz (1989) and discussed in section 5, would tend to
induce greater futures sales than would be most desirable for strategic purposes.
This may mean that inventory levels should be set lower in order to partially
counterbalance this effect. On the other hand, uncertainty would tend to reduce
current production and inventory holding in the absence of futures. However,
given futures can be used to hedge inventories, this may induce greater inventory
holding. In the absence of a careful analysis it is not clear how'these conflicting
tendencies would be resolved.
An interesting issue, not explicitly discussed by Allaz, is whether, given
that inventories would be used strategically by oligopolists if futures markets are
not available, consumers would benefit by the introduction of futures. The
answer would appear to be possibly yes. The reason is that futures trading frees
the oligopolists to choose inventories so as to minimize costs. As a result the total
cost of producing a given quantity at time 2 is reduced. It would appear that
these cost savings could be passed onto the consumers. Again this is not totally
straightforward and should be subjected to more careful analysis.
In the results we have just reviewed there was an important symmetry in
the strategic use of futures by all producers. Newbery (1990) examines some of
the same issues in a model where there is an asymmetry between a dominant
producer and a fringe of price-taking producers. This is an extension of his
previous analysis (1984) which raised but did not fully resolve the question of
whether it is in the interests of a dominant producer to suppress futures trading
in order to discourage the supply of the good by the fringe. One issue that he
addresses is the generality of proposition 4, that is, that a dominant producer
would benefit from the elimination of a futures market. He is able to relax the
assumption that all agents have constant absolute risk aversion by showing that
the net demand faced by the dominant producer would increase with the
elimination of a futures market if the fringe producers are risk-averse, sell futures
in the futures market, and either (1) marginal products are non-stochastic or (2)
the coefficients of variation of output and price are not large and the futures do
not hedge so that the comments made on this point in section 4 apply here as
well.
The major question investigated in Newbery (1990) is whether the
232

dominant producer could profitably use storage as a means of suppressing the

futures market. In this regard Newbery argues that a dominant firm has a
comparative advantage in storage. Specifically, in a model where a dominant firm
can choose to store at the same time it establishes its futures position, he shows

Proposition 9 (Newbery 2):

If storage costs are the same for the dominant firm as for competitive stock
holders, the dominant firm can profitably monopolize storage if (1) the dominant
firm is risk neutral and (2) expected spot price exceeds the futures price by a
constant positive risk premium.

The basic reason for this is first that if competitors do the storage they
equate the cost of the stored good with the futures price, which is less than the
expected spot price. If the dominant firm replaces the competitors in storing
some given amount, the expected cash price would be unchanged and by not
hedging the good (since they are risk-neutral) they would effectively earn the
whole of the risk premium on the stocks. Competitors could not profitably store
since in doing so they would depress the expected spot price and the futures price
(since the risk premium is assumed constant).
Newbery recognizes that his second assumption, which implies the normal
backwardation (risk premium) be insensitive to the amount of hedging, is both
important and open to question. Alternatively, we might expect that if the
amount of hedging dropped with the number of speculators held constant the risk
premium would drop implying an increase in the futures price if all the storage
were undertaken by the dominant firm. In this case there would be an incentive
for the competitive agents to store, which would greatly complicate Newbery's
line of argument and might compromise his subsequent results. His primary
defense of this assumption is that there is free entry of speculators and that
commodity risks are diversifiable. If this is the case, however, then we would
expect that there would be zero risk premium implying that the comparative
advantage of the dominant firm is storage would be negligible (i.e., monopolizing
storage would produce zero profits).
Using his proposition, Newbery proceeds to construct a parametric example
of a model in which a dominant producer could profitably suppress futures
trading. The assumptions of the framework include linear demand, linear fringe
supply, additive production shocks that are perfectly correlated for the dominant
 233

producer and the fringe, and a constant downward bias in the futures market.
The question of the profitability of futures market suppression reduces to a
comparison of the dominant producer's profits in two cases. In the first case,
storage is undertaken by the competitors who, along with the fringe, hedge in the
futures market. In the second case, the dominant producer monopolizes storage,
suppresses the futures market, and thus decreases fringe supply. Newbery argues
that in order to succeed in suppressing the futures market the dominant producer
would have to store more than would the competitors, thus increasing the supply
of the good from storage next period and increasing the dominant producer's cost
of goods in the second period. Furthermore in order to discourage private storage
the firm must assure that the expected next period price is not too high. These
factors tend to reduce the dominant firms profits. The issue is whether these
effects are more than balanced by the reduction in fringe supply brought about
by the suppression of the futures market.
Newbery shows that at least for one set of parameters in his model the
suppression of futures trading yields greater profits for the dominant firm than
does allowing futures trading. He concludes from this analysis that there is a risk
that futures trading may be stifled in markets where there is a high concentration
of supply and that suppression of these markets is likely to be socially harmful.
A critical part of this argument is that he assumes that the futures market
will disappear if (1) the cartel stores more than would competitive stockholders
and (2) the variability of prices falls because of increased storage. The sufficiency
of these conditions is not demonstrated within the model nor is it obvious. The
conditions seem to conflict somewhat with the argument employed elsewhere that
there is free entry of speculative services, implying an absence of significant,
avoidable fixed costs. Furthermore, the analysis reveals that the variability
referred to in (2) is the "expected value of the variance of prices over two
periods". In contrast, the standard analysis of the supply of speculation reveals
that the relevant variable is the variance of the futures price at maturity.20 Under
Newbery's assumption of additive supply uncertainty, the latter variance is a
constant unaffected by the level of storage.
A somewhat different perspective on the use of futures by the dominant
producer of a storable good is provided by Anderson's and Gilbert's analysis of

20 See, for example, Newbery and Stiglitz (1981), p.184.

234

the tin collapse. It is seen that a cartel engaged in price stabilization through a
buffer stock arrangement may be both helped and hindered by the existence of
the futures market. The typical buffer stock scheme aims at defending a price
floor and a price ceiling by adding to the stockpile when prices are at the floor
and selling from the stockpile when prices are at the ceiling. There are two
constraints on the operation. One is the stock-out constraint which says you
cannot hold negative stocks. The analysis of Salant (1983) shows that speculative
purchases of stocks may accelerate the time that this constraint is felt and thus
may succeed in attacking a price ceiling. Anderson's and Gilbert's study of the
tin buffer stock shows that the ability to defend' a price floor may be
compromised by a second, financial, constraint which says that the buffer stock's
borrowing of money cannot exceed some amount.
A futures market can aid the buffer stock manager's defense of a price floor
by giving him greater leverage than may be available for cash market purchases.
The buffer stock's long futures positions support the futures price which in turn
induces private stockholding and thus aids in supporting the cash price.
Consequently, increasing leverage through futures trading can mean that a given
credit line supports a greater total long position in the good (both physical stock
and futures). However, the futures market is a mixed blessing for the buffer
stock. The reason is that a price floor is subject to attack through speculative
selling of the good. When all the good is controlled by the buffer stock, it is
difficult to mount an attack by selling the physical good. A futures market
facilitates such an attack by making it possible for speculators to short sell.
Judging from the experience of the tin market, it appears that the buffer stock at
times can meet this threat again using a feature of the futures contract. In
particular, if it has a dominant long position in both the cash and futures market,
it is able to occasionally create a squeeze in the maturing contract. The risk of
such a squeeze helps to discourage short-selling and thus keep the attackers of a
price floor at bay.
To summarize, we see that Newbery emphasizes the interest of a dominant
producer in suppressing futures trading in order to discourage supply by fringe
producers. Allaz's results for storable and nonstorable goods also suggests that
dominant producers have an incentive to suppress futures. For if the futures
market were absent oligopolists would be prevented in engaging in strategic
futures sales with the result that the expected industry profits would rise. These
two views are in contrast with the study of Anderson and Gilbert which shows
 235

that a cartel may welcome futures trading as a means of increasing leverage and
thus relaxing the financial constraint on a price support operation. Finally, the
analysis of Anderson and Brianza discussed in section 6 shows that the interests
of a cartel may be served by a futures market if the cartel members are able to
cooperate in allocating initial futures positions to members.

8. STRATEGIC FUTURES: DURABLE GOODS

The underlying good of the models of the previous section was storable; however,
it was implicitly assumed that in use it would be perishable. For if the good in
question were a durable good which produced a stream of services over time, the
strategies of a dominant producer would be complicated by the possibility of
intertemporal price discrimination. Furthermore, if these strategies are
anticipated by purchasers, they will ultimately undermine the profitability of the
dominant producer. This phenomenon was first described by Coase (1972). It has
been interpreted as indicating the limits on the ability of a monopolistic producer
of a durable good (e.g. a specialized new computer) to extract the full consumer
surplus by charging a very high price initially and letting the price decline
sharply later. It is perhaps not so widely recognized that the analysis applies to
certain commodities as well. A commodity is subject to the analysis if it is
valuable for the services that it can give in a relatively untransformed state over
an extended period of time or if the costs of recovering the commodity from
finished goods are low relative to its original price. Perhaps the best examples are
the precious metals such as gold, platinum, and, to a lesser extent, silver.
However, the analysis may also apply to other metals with relatively efficient
scrap markets or to commodities where relatively large stocks are held for their
"convenience yield".
Anderson (1985) observes that if there is futures or forward trading for a
durable good produced by a dominant firm, this firm can trade forward in such a
way as to counteract the profit dissipating effects of buyer foresight. To see why,
suppose that by some means a firm had acquired a dominating position in gold
bullion in the economy and that the production of new bullion is insignificant
relative to the current stock. How would the dominant firm seek to profit from
its gold stock? Profits are increased to the extent the firm would be able to
extract the consumer surplus. One method to extract the consumer surplus is to
discriminate intertemporally. In principle, initially it sets the price very high and
would sell to the few consumers who valued to good most. Then some time later
236

it would reduce the price and sell to the consumers who valued the good
somewhat less and so on. The problem with this strategy is that the firm has
little incentive to restrain itself in the speed with which it works down the stock.
Thus prices would drop relatively quickly, and even consumers who value the
good highly will realize that be deferring their purchases they could buy at much
more attractive prices. The recognition of this prospect reduces the demand and
consequently the dominant firm's profitability. If the firm were to commit itself
to restraining its later sales it would enhance its ability to sell at a higher price
initially. Futures purchases would serve this purpose since its futures market
profits would be depressed by any increase in its later cash market sales.
These effects can be demonstrated in a simple two period model of a
durable good produced by a monopolist. At time t=1 the monopolist produces a
quantity ql of a good that lasts two periods without depreciation after which it
does out of existence. At t=2 the monopolist produces an amount q2 of the good
that is perfectly substitutable with the good produced at t=l. The total stock of
goods in existence at t=1 is Q l =ql and at t=2 is Q2=ql + q2. We assume that
the monopolist can produce any amount desired at a constant marginal cost
which without loss of generality we assume to be zero. To further simplify the
notation we assume that the discount rate is zero.
The critical aspect of the model is that the consumer recognizes that the
utility will derive from a flow of services over time and that the discounted
marginal utility of subsequent services will be equated with the price that
prevails later. Specifically, competitive consumers act to equate the value of the
marginal utility of the durable goods services with the effective cost of those
services. We represent this with an inverse demand relation that states the
consumers' marginal per period rental offer (rt ) is a decreasing function of the
stock of goods held by consumers; for simplicity we take this to be linear,
rt = a - bQtt = 1,2. (7)
We assume that at t=1 consumers can purchase the good from the
monopolist at a price PI and that at t=2 they can either purchase from the
monopolist or on a competitive resale market at price P2. 21 In this case the
consumers will acquire goods until the value of services equals the implicit rental
rate which is defined as the interest cost less the capital gain from holding the

21 The same phenomenon will hold in the absence of competitive resale if

consumers solve an intertemporal utility maximization problem and have rational
forecasts of the path of prices. See Tirole (1988), ch.l.
 237

good. Under our assumptions of static linear demand and zero interest rate this
implies,
pt=a-bQt+Pt+lt=I,2. (8)
where P3=0.
Now the objective of the monopolist is to maximize the present value of
durable goods sales,
w = Plql + P2 q2 (9)
subject to (8). This is a dynamic problem because the anticipated choice of q2
will affect profits at t=l through its effect on Pl' Solving the monopolist's
dynamic program yields the equilibrium results given in column I of Table 1.
Notice that even though we have assumed zero marginal cost, there are positive
sales in both periods. Furthermore, since the price drops from .9a to .3a, the
implicit rental in period 1 is .6a.

Table 1. Equilibrium with Durable Goods

Variable Units No Futures Futures

f alb -0.5
pf a 0.5
ql alb 0.4 0.5
PI a 0.9 1.0

q2 alb 0.3 o
P2 a 0.3 0.5
w a 2/b 0.45 0.5

The introduction of futures trading into the analysis causes two

modifications. First, in making his output decision the monopolist must take into
account not only the profits from cash market sales but also the profits derived
from his previously established futures position. Second, consumers will rationally
take into account the effect of the monopolist's futures position on subsequent
output decisions. To explore this, we now allow for the trading at t=l of a
futures contract calling for delivery of the monopolist's good at t=2. Thus the
monopolist's t=2 profits are given as,
238

f
11"2 = P2 q2 + (p -P2)f. (10)
With these modifications we can reconsider the monopolist's dynamic program.
Combining (8) and (10) and differentiating we obtain the first-order condition for
q2 which can be rearranged as

q2 = HE- q1 + f). (11)

As we have seen previously in other contexts prior futures sales tend to induce
the monopolist to produce more than he would have otherwise.
At t=l the monopolist chooses q1 and f to maximize,
w = P1 q 1 + 11"2 ,(12)
subject to (8) and (10). Since we assume perfect foresight both the monopolist
and the consumer recognize that price next period will be determined by the
production rule (11). Furthermore, an implication of perfect foresight is that
pf=P2 so that no profits or losses will be made on the futures market.
Nevertheless futures trading has an influence on the equilibrium because the
monopolist has the power to bring about a different realization of P2; in
equilibrium he simply will find that P2=pf is optimal.
Combining (8), (10), (11) and (12) yields the t=l objective function.
Differentiating with respect to f and q1 and solving the resulting first-order
condi tions yields
_Ia_ f
q1- (13)
1 1)--'
That is, in the perfect foresight equilibrium the monopolist goes long futures by
an amount exactly equal to his first period production. As a consequence, in view
of (11), he will produce less at t=2. The full characterization of the equilibrium is
obtained by inserting (13) in the appropriate expressions; the results are
summarized in Table 1 column 2.
The effect of introducing futures trading for the durable goods monopolist is
seen by comparing columns 1 and 2. With futures, the monopolist's wealth, W,
has increased so that the monopolist is able to overcome the dissipating effect of
the good's durability. This occurs because he produces more at a higher price at
t=1. The price is high at t=l because consumers foresee that the monopolist will
produce nothing at t=2. The reason he produces nothing is that he has purchased
futures in an amount that makes it optimal for him to do so.
lt has been recognized previously that a durable goods monopolist can
overcome the profit disSipating effects by renting his good rather than selling
 239

(see, e.g., Bulow, 1983). Anderson shows that the monopolist's production and
profits under the sale-pIus-futures regime are precisely those that would be
obtained in a rentals-only regime. Indeed, since the monopolist makes a futures
commitment to buy-back all of its first period sales (see equation (13)) there is a
formal equivalence between the futures regime and a leasing arrangement, which
is in turn equivalent to rentals.
There are important institutional differences between the sales-pIus-futures
regime and a leasing arrangement. First, unlike a lease, the counterparties to the
monopolist's futures purchases need not be the holders of the good. If this is the
case, the owners of the public stocks need to be concerned about the likely price
change and thus must forecast the monopolist's production next period. However,
if they are aware of the monopolist's futures position they rationally forecast that
production will be restrained in the future and that prices will be supported.
Second, public policies aimed at preventing dominant producers from engaging in
rentals only strategies may be unable to prevent the price-support strategy
executed through futures.
The above analysis showing that purchases of futures contracts effectively
restrains future output and thus supports current durable goods prices applies to
a simple two-period model. In a multiperiod model it is clear that the same
obtains if the monopolist is able at the outset to buy futures contracts for all
future periods. The problem with this is that futures contracts are typically
traded only for a limited number of relatively near maturity dates. Thus the
important issue is whether prices can be supported through short-term contracts
only. Anderson shows that in fact this is the case; specifically,

Proposition 10 (Anderson 1)
Assume the linear demand for durable goods services (equation (7)) holds for
t=I,2,3, .... If for every t there is trading for a futures contract maturing at t+1,
then in the unique perfect equilibrium ql=.5a/b, qt=O for t=2,3, ... , and ft =-ql.

The strategy works because it is rationally foreseen that in each period it

will be optimal for the monopolist to commit himself to buying back all the goods
outstanding in the next time. Again, despite the obvious institutional differences,
this strategy produces profits equivalent to a regime of rentals only for all
periods.
As with other results on strategic futures trading the effects are
240

demonstrated most clearly under the assumption that there is perfect foresight.
What is the effect of introducing uncertainty into the analysis? The principal
change emerges if the monopolist is risk-averse, for then in the face of demand
uncertainty there will be a tendency to hedge through futures. At times this
hedging motive will conflict with the strategic motive. Specifically, Anderson
shows that the monopolist's optimal futures position will consist of two parts.
The first is the strategic part and is a long position equal to the first period sales.
The second part is a hedging component which under plausible conditions is
short. That is, hedging leads the monopolist to reduce his long positions. As a
consequence his expected future production is increased j which reduces the price
that he can sell his durable goods for initially. Still, the monopolist's expected
utility is greater than under the sales regime without futures. However, the
expected wealth under the sale-plus-futures regime is shown to be less than under
a rentals-only regime. This points out a third significant way that rentals and
futures differ as means of solving the durable good monopolist's problem.

9. EXHAUSTIBLE RESOURCES
For a number of the commodities, the reason that there are powerful producers is
that they are resources which nature has allocated to a small number of
countries. Some of these commodities, crude oil for example, are traded on
futures exchanges or organized forward markets. The models reviewed above did
not explicitly allow for the good to be in exhaustible supply and thus did not
explore how futures trading may alter existing results on natural resources
produced by powerful producers. In particular, it is interesting to enquire
whether futures trading would lead to a departure from the Hotelling rule (1931)
for the extraction rate of an exhaustible resource.
This issue has recently been addressed by Brianza, Phlips and Richard
(1987) and Phlips and Harstad (1990). In both of these there is a simple model of
a resource that can be produced at zero cost, sold and consumed in either of two
periods with zero discounting between periods. Total production in the two
periods is constrained by the total stock available. In the case of duopoly (see
Phlips and Harstad, 1990) the basic relation is,
si = qil + qi2' (14)
where si is the total initial reserve of the good for producer i and qit is that
producer's output in period t. Note that by these constraints the choice of output
 241

in the first period will imply the output level for the second period so that the
firm's production decisions are essentially static.
Now suppose that prior to making production decisions there has been
trading in a futures contract calling for delivery of the duopolists' goods in period
2. Let fi be the i-th firm's futures sales. Finally, assume perfect foresight at the
time production is decided and that demand is linear and constant for the two
periods. Then the (Nash) equilibrium production plans for the firms are,
fi fi
~1 = si - 3 +6 (15)

fi fi
qi2 = si + 3-6 (16)
for i=a,b and j#. From equation (16) we see that as usual a firm's futures sales
tends to increase its production in the period in which the futures contracts
mature. Furthermore, from the same expression we note that a rival's future sales
tends to increase the rival's output in that period and that in turn depresses the
firm's output half as much. Now since anything increasing the output in the
second period must decrease first period output by an equal amount, we see by
equation (15) that the firm's futures sales tends to depress its initial output while
the rival's future sales tends to increase the firm's output.
The implications of these production rules for the path of natural resource
prices are seen in the follOwing relations,

PI = f> + 8(f a +fb) (17)

P2 = f> - O(fa +fb ), (18)
where 0 is a positive parameter. f> is the price which would hold in equilibrium in
the absence of futures trading. Thus without futures these equations reproduce
the Hotelling path which (in the absence of discounting) is for prices to be
constant. 22 Thus we see that if there have been net aggregate futures sales for
period t=2 the effect is to depress P2 below the Hotelling path by a certain
amount and to raise PI above the Hotelling path by the same amount. When
there are net futures purchases for period t=2 the opposite price distortion
results.

22 The results for monopoly yield similar qualitative conclusions.

242

Futures trading for natural resources thus generally leads to a departure

from the familiar Hotelling rules. The nature of the modification depends upon
whether the powerful producers buy or sell futures and in what quantities. This
raises the issue of what will be the result of the futures market in the initial
trading period. As we might expect, the results are sensitive to the assumptions
made about the participants in the futures market.
Brianza, Phlips and Richard (1987) consider a monopolistic producer of the
nat ural resource and assume the other participants in the futures market are
represented as a risk-neutral pure speculator. They conclude that if the
monopolist and the speculator have the same beliefs about demand that will
prevail at t=2 there will be no trade in equilibrium of the futures market.
However, if their beliefs diverge they find that the futures market will be active.
The monopolist goes short futures if he believes demand at t=2 will be less than
does the speculator. When the opposite is true the monopolist goes long futures.
To see the reason for their result note that the speculator will be willing to
take any position desired in the futures market so long as pf=P2' Then, if the
expectations of the speculator and the monopolist agree, the futures profits of the
monopolist drop out, and the total profits of the monopolist are the cash profits
which can be written as,
A f2
?fm = ps - Jt 2' (19)
where p is the Hotelling price, s is the monopolist's total resource and Jt is a
positive constant. This is strictly concave in f taking is maximum at f=O. Thus in
equilibrium purely speculative futures trading leads to prices along the Hotelling
path.
An issue not resolved by this analysis is how the monopolist's risk-aversion
would affect the results. It seems a plausible conjecture that demand uncertainty
would lead to futures sales and that this would shift production to later periods,
leading to prices rising less than in the absence of futures.
In their modelling of futures trading for the output of the duopolistic
natural resource market, Phlips and Harstad depart from the conventional
competitive view of a centralized futures market open to a large number of price-
taking traders. Instead futures contracts are negotiated among three participants:
the two duopolists and a single speculator. The producers are taken to be
risk-averse and thus have some hedging demand for futures. The agents are
 243

allowed to have differing beliefs. Trade takes place by each agent offering to buy
and/or sell a specified amount for futures contracts at a specified price. Then
according to an order randomly determined each agent is able to agree to any
offer still outstanding. There is no recontracting. In this framework, there are a
multiplicity of (perfect) equilibria; however, all result in the same net positions.
Numerical calculations reveal that the resulting positions are sensitive to
parameters, notably beliefs.
This treatment of a futures market is rather unconventional. It is an
interesting use of game theory to give a more explicit micro-foundation for
models of futures trading. The particular specification turn's heavily on the
assumption that there is a single speculator that meets the duopolists and that
there is no recontracting. As such the contracts described may have more in
common with an insurance contract negotiated with a single underwriter rather
than the typical futures markets open to public participation.

10. FUTURES PRICE BIAS AND VOLATILITY

Having completed our review of the theoretical contributions to the literature on
futures trading with imperfectly competitive cash markets, it is worth
summarizing their findings on the traditional questions of price performance in
futures markets. On the issue of bias (i.e., the firm moment of the futures price),
we have seen a variety of results. A number of studies made the assumption that
there is perfect foresight and required that the futures price must coincide with
the cash price that is anticipated for the maturity date of the futures contracts.
This can be interpreted as representing the market outcome when there are
homogeneous beliefs and there is at least one risk-neutral speculator (who is not
limited in the size of positions he can take) participating in the futures market.
For in this case, the futures price must equal the expected cash price of the
maturity date. From the analysis of Newbery (1984, 1990) we have seen that,
when only the dominant producer is risk-neutral and the other agents including
the fringe producers are risk-averse, then the futures price can be biased. For
Newbery, since he excludes the possibility of consumer hedging, hedging is likely
to be short and the futures price will be less than the expected cash price. More
generally, as Anderson and Sundaresan (1984) point out the futures price can be
greater than, equal to or less than the expected cash price depending upon the
relative importance of long and short hedging and the supply of speculative
services which in turn depends importantly on the attitudes towards risk of the
244

agents. This is entirely analogous to the results obtained for the literature on
futures for competitive cash markets (see e.g. Anderson and Danthine, 1983).
On the issue of market power and the volatility of futures prices, most of
the analyses are silent. The most important exception is Newbery (1990) who
argues that a dominant producer has an incentive to stabilize prices in order to
discourage a futures market and thus to prevent the fringe from hedging. The
specific prediction is that highly concentrated industries are less likely to have
futures markets. The predictions are less clear when a concentrated market has
futures trading nevertheless. One might argue that dominant producers in highly
concentrated industries are likely to be attempting to suppress futures trading
and thus may engage in an effort to stabilize cash prices. This might translate
into reduced futures market volatility. On the other hand, if there is no realistic
prospect of suppressing futures trading, the dominant, risk-neutral producer
might randomize its production decisions so as to increase cash and futures price
volatility as a means of curbing fringe supply. This was conjectured by Newbery
(1984).
As of now none of these predictions have been subjected to specific
empirical testing. However, an interesting study by Slade (1989) indirectly has
shed some light on the relationship of cash market concentration and futures
price volatility. She investigates the determinants of the variance of monthly
price changes for six metals: aluminum, copper, lead, nickel, silver, and zinc. She
includes forward prices on the London Metal Exchange and (with the exception
of silver) official producer prices for immediate delivery. She uses a variety of
statistical methods to investigate the effects of concentration and of futures
trading. In the case of regression analysis, the relation was of the form,
Vir = aC ir + PFi + rXir ' (20)
where Vir is the variance of monthly price changes in market i for period r, Cir
is the concentration for that observation, Fi is a dummy variable that equaled 1
if market i was a forward market and was zero otherwise, and Xir is a vector of
explanatory variables to control for other factors. It was found that Il was
significantly negative while {3 was significantly positive. Quantitatively, it was
found that whether the market was a futures market or not had a larger impact
on differences in volatility than did differences in concentration.
For our purposes the most important finding is that the results imply that
there is negative relation between cash market concentration and futures market
 245

volatility. 23 Thus, the evidence lends some support for the view that in
concentrated industries dominant producers may seek to stabilize futures prices.

11. CONCLUSION
We have seen that the subject of futures trading for imperfectly competitive cash
markets has been investigated in a large variety of models in recent years. In
light of this what is the state of our knowledge concerning the questions raised in
the introduction?
First, there is no fundamental, general incompatibility between futures
trading and the fact that the underlying good is produced under imperfect
competition. To a great extent, the traditional risk-shifting and price-discovery
functions of futures trading exist for such markets. Beyond this, powerful cash
market producers can at times be active futures market users for strategic
purposes. It is conceivable that under certain circumstances, powerful producers
have an interest in pursuing actions aimed at suppression the futures market. In
particular, this would be true if the powerful producer must coexist with a fringe
of highly risk-averse competitive producers. However, the difficulty and costliness
of suppressing futures trading makes this appear to be an exceptional
circumstance.
Second, there are a variety of reasons why futures trading for imperfectly
competitive cash markets would seem to be socially beneficial. In part, this is
simply a reflection of the fact that this promotes hedging and information
aggregation much as for competitive markets. Beyond these effects there are
several strong forces which tend to improve competition with the result of
increased production and lower cash prices. These forces are: (1) hedging by
powerful producers leads them to sell futures, (2) hedging by competitive fringe
producers leads them to increase production, and (3) strategic trading by

23 Our interpretation is somewhat tangential to Slade's purpose. Her primary

conclusion is that by facilitating speculation futures markets result in increased
volatility. In our own view, the finding that, other things equal, variances of
prices are higher in futures markets than in metals producer price systems does
not warrant this conclusion and might be attributable to other factors. The most
important is due to the fact that, as was acknowledged by Slade, rationing is
widely practiced in the metals producer price system. As a consequence, we never
observe the extremes of how prices would fluctuate if they were allowed to reach
an equilibrium of supply and demand. In contrast, futures prices are equilibrium
prices; they give a reading of the value to the marginal investor which would not
be observed if there were rationing.
246

oligopolists leads them to sell futures.

Third, there are identifiable circumstances when the futures market can be
used as a tool of the powerful producers to increase their market power. If
oligopolistic producers are able to cooperate in their trading of futures, they can
use these futures positions as a means of promoting cartel discipline. Specifically,
even if production quotas cannot be directly enforced, by appropriately allocating
long futures positions to cartel members they will be led to obey their quotas
voluntarily. The second circumstance when futures trading can increase market
power is if the producer or group of producers have a durable product. Then by
buying futures they can credibly commit themselves to'restricting output later
which has the effect of increasing the price that they can charge for goods in
earlier periods.
Fourth, futures markets for imperfect cash markets are not inherently more
prone to manipulation of the delivery terms of the futures contracts than are
futures with broadly competitive cash markets. However, the one circumstance
when manipulations are a particular concern is the case when powerful producers
adopt long positions in an attempt to increase their cash market power.
Fifth, the price bias in the futures market with an imperfectly competitive
cash market can be positive or negative depending upon the balance of hedging
and strategic trading in the market. If speculators are risk averse and long
hedging by consumers is inconsequential, we can say that there is a downward
bias (i.e., normal backwardation) if the producers' net positions are short futures.
There is an upward bias if producers' net positions are long futures. In light of
the fact that cash market supply is generally an increasing function of the
powerful producers' futures sales, we see that measuring the bias is potentially a
means of testing the hypothesis that futures trading improves competition in cash
markets.
In closing we suggest two avenues for future research that could
significantly increase our understanding of the function of futures trading for
imperfect cash markets. The theoretical development that would be most
interesting would be to reconsider some of the models described above under
conditions of asymmetric information. In particular, the models we have reviewed
have made the assumption (at least implicitly) that the futures positions of
powerful agents are observed so that forecasts of future cash prices can take this
into account. In practice, futures positions of agents are likely to be imperfectly
observable. An interesting extension of previous models would be to treat futures
 247

positions as private information.

With few exceptions there has been little empirical work on futures trading
for imperfect cash markets. The theoretical models we have reviewed can provide
useful frames of reference for case studies of particular markets. Beyond this the
question of the interaction between cash market power (e.g. as measured by
concentration) and futures price bias and volatility is an unresolved issue that
might be treated using cross-sectional information.

REFERENCES
Allaz, B (1987), Strategic Forward Transactions Under Imperfect Competition:
The Duopoly Case, Ph.D. dissertation, Princeton University.
Allaz, B (1989), "Oligopoly, Uncertainty, and Strategic Forward Transactions",
workin/i paper, HEC-ISA.
Allaz, B (1990), "Duopoly, Inventories, and Futures Market", this volume.
Allaz, Band J L Vila (1986),: "Futures Markets Improve Competition", working
paper, Princeton University.
Anderson, R W (1985), "Market Power and Futures Trading for Durable Goods",
working paper, City University of New York, Graduate Center.
Anderson, R W (1986), "Regulation of Futures Trading in the United States and
the United Kingdom", Oxford Review of Economic Policy, 2, 41-57.
Anderson, R W (ed.) (1984), The Industrial Organization of Futures Markets,
D.C. Heath, Lexington.
Anderson, R Wand J P Danthine (1983), "Hedger Diversity in Futures
Markets", Economic Journal, 93, 370--89.
Anderson, R Wand M Sundaresan (1984), "Futures Markets and Monopoly", in
R W Anderson (ed.) (1984).
Anderson, R Wand C L Gilbert (1988~, "Commodity Agreements and
Commodity Markets: Lessons from Tin', Economic Journal, 98, 1-15.
Anderson, R Wand T Brianza (1989), "Cartel Behavior and Futures Trading",
working paper, City University of New York, Graduate Center.
Brianza, T, L Phlips and J F Richard (1987), "Futures Markets, Inventories and
Monopoly", CORE D.P. 8725.
Bulow, J I (1982), "Durable-Goods Monopolists", Journal of Political Economy,
90, 314-32.
Coase, R H (1972), "Durability and Monopoly", Journal of Law and Economics,
15, 143-49.
Edwards, L Nand F R Edwards (1984), "A Legal and Economic Analysis of
Manipulation in Futures Markets", Journal of Futures Markets, 4, 333-66.
Eldor, R and I Zilcha (1986), "Oligopoly, Uncertain Demand and Forward
Markets", working paper, Tel-Aviv University.
Greenstone, W D (1981), "The Coffee Cartel: Manipulation in the Public
Interest", Journal of Futures Markets, 1, 3-16.
Hotelling, H (1931), "The Economics of Exhaustible Resources", Journal of
Political Economy, 9, 137-75.
Johnson, P M (1982), Commodity Regulation, Little Brown, Boston.
Milgrom, P and N Stokey (1982), "Information, Trade, and Common Know-
ledge", Journal of Economic Theory, 26,17-27.
248

Newbery, D M P984), "The Manipulation of Futures Markets by a Dominant

Producer' in R W Anderson (ed.) (1984).
Newbery, D M (1990), "Cartels, Storage, and the Suppression of Futures
Markets", European Economic Review, forthcoming.
Newbery, D M and J E Stiglitz (1981), The Theory of Commodity Price
Stabilization, Clarendon Press, Oxford.
Phlips, L (1984), "Comment on 'The Manipulation of Futures Markets by a
Dominant Producer'" in R W Anderson (ed.) (1984).
Phlips, L and R M Harstad (1990), "Interaction between Resource Extraction
and Futures Markets: A Game-Theoretic Analysis" in R. Selten (ed.)
Game Theory in the Behavioral Sciences, Springer Verlag, Heidelberg.
Prain, R (1975), Copper: An Anatomy of a Market, Mining Journal Books,
London.
Rotemberg, J and G Saloner (1985), "Strategic Inveptories and the Excess
Volatility of Production", working paper, MIT.
Salant, S (1983), "The Vulnerability of Price Stabilisation Schemes to
SpeCUlative Attack", Journal of Political Economy, 91, 1-38.
Slade, M E (1989), "Market Structure, Marketing Method, and Price
Instability", working paper, University of British Columbia.
Spence, A M (1977), "Entry, Capacity, Investment, and Oligopolistic Pricing",
Bell Journal of Economics, 8, 534-44.
Tirole, J (1982), "On the Possibility of Speculation Under Rational
EX,l>ectations", Econometrica, 50, 1163-82.
Tirole, J (1988), The Theory of Industrial Organization, Cambridge, MIT.
 CHAPTER 9

DUOPOLY, INVENTORIES AND FUTURES MARKETS

Blaise Allaz

1. INTRODUCTION
Most of the literature dedicated to futures or forward markets is built on the
assumption that goods traded on these markets are produced by perfectly
competitive firms, i.e. firms that act as price-takers. The results then focus on
optimal rules for hedging or speculating.
It is only recently (early to mid-eighties) that this assumption was
questioned. This is somewhat surpriSing since there exists a great number of
commodities that are traded on futures markets for which the underlying cash
(spot) markets are definitely not perfectly competitive. This fact was pointed out
by Anderson (1984) and Newbery (1984).1'2
The aim of this chapter is to study the relationship between cash market
power and futures/forward trading. We analyze how production decisions are
altered and how the spot market equilibrium is affected when a good produced by
nonperfectly competitive firms - in our case, a duopoly - is also traded on a
futures market or when the firms can make forward contracts. We try to answer
the following questions:
i) Is the cash market power of the duopolists increased or decreased when the
good they produce is also traded on a futures market?

1 Newbery cites a study of his "which lists eight commodities for which single
countries controlled more than 50 per cent of world trade, and another thirteen
for which single countries controlled between 25 and 50 per cent (averaged over
the 1977-79 period)". (Newbery, 1984, p.35).
2 Greenstone's 1981 paper can also be credited for drawing our attention on
the noncompetitive nature of the spot market for coffee.
249
L. Phlips (ed.). Commodity, Futures and Financial Markets. 249-271.
© 1991 Kluwer Academic Publishers.
250

ii) What are the producers' rationales (if any) for trading on the futures
market or for making forward contracts?
We start by showing that, in a framework of imperfect competition, there
can exist, besides hedging and speculation, a third rationale for trading on a
futures market and this new motive is one of strategy. Futures trading is
strategic if its direct purpose is not to make profits on the futures market but
rather to influence the outcome on the spot market. We show below that, in a
two-period model where trading on the futures market occurs before production
takes place, the outcome on the spot market will typically depend on the net
forward sales (or purchases) made by the producers on the futures market. 3
Since strategic playing only makes sense within an environment of
imperfect competition, this strategic rationale cold not have been pointed out in
the early literature.
The first section assumes that the good is not storable. This is relaxed later
when we analyze optimal inventory and futures decisions. We show that
inventories can also be used strategically and then pay special attention to how
the firms will combine futures and inventories for their strategic purposes. Under
perfect foresight, we obtain a separation result that says that although
inventories can also serve the same strategic aim as futures, all the firms'
strategic plays will be done through futures transactions.
However, before developing our own model, let us briefly review the
previous literature and show how our model relates to it.4 All models developed so
far can be more or less classified according to the following three criteria:
1) whether the cash market is characterized by monopoly (la) or oligopoly (lb);
2) whether the good is perishable (2a) or durable (and hence storable) (2b);
3) whether forward and spot (i.e. production) decisions are taken simultaneously
(3a) or sequentially (3b).
We can trace back the first attempts to analyze futures markets within the
context of an imperfectly competitive cash market to Anderson and Sundaresan

3 The precise conditions under which this result holds are developed below.
They obviously have to do with the behavior of the producers relative to one
another (i.e. conjectural variations) and on the timing of spot and futures
decisions.
4 We shall only very briefly review the previous literature since this is
already done in great detail in Ronald Andersons's survey that can be found in
this same volume. We apologize to authors we do not mention.
 251

(1984) and Newbery (1984).

Anderson and Sundaresan built a two-period model where the good traded
on the futures market at time 1 is produced by a monopolist at time 2. Since the
monopolist is, by definition, the only producer, he entirely controls the supply of
the good that is called for delivery. Therefore, the monopolist realizes that his
trading on the futures market will have an impact on his spot market profits and,
hence, trades on the futures market accordingly. The model built by Anderson
and Sundaresan is of the type (la, 2a, 3b).
Newbery (1984) considers the case of a risk neutral dominant producer
facing a fringe of risk averse competitive producers (farmers). Farmers take their
spot and forward decisions simultaneously. To simplify his analysis, Newbery
assumes specific functional forms and that preferences are of the mean-variance
type. His model is of the type (la, 2a, 3a).
In the previous two papers, the good was assumed to be perishable and,
hence, nonstorable. In his 1985 paper, Anderson relaxes this assumption by
considering the case of a monopolist that produces a durable good. The good is
sold to price-taking consumers. There are two production periods, time 1 and
time 2. There also exists a competitive secondary market for used units of the
good (which does not depreciate). Finally, at time 1, all agents can trade futures
contracts that call for the delivery of units of the good at time 2. Anderson
assumes full public information and rational expectations. The model is of the
type (la, 2b, 3b).
Brianza, Phlips and Richard (1987) also consider the case where the good
traded on the futures market is produced by a monopolist and is storable. They
pay special attention to the role played by the producer's and the representative
speculator's expectations. Their model is of the type (la, 2b, 3b).
Eldor and Zilcha (1986) consider the case of an oligopoly made up of
identical risk averse firms facing an uncertain demand. The firms simultaneously
act on the forward and spot markets. The model is of the type (lb, 2a, 3a).
Allaz and Vila (1989) consider the case of two Cournot duopolists. they
assume perfect foresight and linear demand and cost functions. In the first part of
their paper, they consider a two-period model. At time 1, producers and
speculators meet on the futures market. At time 2, actual production occurs and
agents trade on the spot market. In the second part of the paper, they extend the
model to N periods of trading. For N successive periods, the agents can take
positions on the futures market that call for the delivery of the duopolists' good
252

at time N+1. The model is of the type (1 b, 2a, 3b).

Phlips and Harstad's (1990) paper is a two-stage game-theoretic approach
to the interaction between resource extraction (the cash market) and futures
markets. Their model features two duopolists facing one representative
speculator. The analysis again focuses on the agents' expectations. The model is
of the type (lb, 2b, 3b).
Allaz (1988) is a two-period linear model of an oligopoly made up of n
firms. Speculators are assumed to be price-taking. Special emphasis is put on the
behavioral assumptions made upon the firms, i.e. the type of conjectural
variation that is assumed. The model is of the type (1 b, 2a, 3b).
Finally, Anderson and Brianza (1989) focus on the behavior of a cartel
whose good is traded on a futures market. The initial two-period analysis is
extended to several periods of futures trading but inventory decisions are not
considered. The model is of the type (lb, 2a, 3b).
Our own contribution develops a two-period model of a duopoly that can
trade at time 1 on the futures market and then produce and sell its production at
time 2 on the spot market. The good is assumed to be storable. The novel feature
is that inventories are shown to be able to serve a strategic purpose just as
futures can. The analysis then focuses on the firm's optimal inventory and futures
decisions at time 1. The model belongs to the class (lb, 2b, 3b).
The analysis proceeds in three steps: in section 2, we study the optimal
trading and production decisions of two duopolists that produce a good that is
traded on a futures market. We show that, under some assumptions, the
producers will trade on the futures market for a strategic reason by which we
mean that the first aim of the trading will not be to make profits on the futures
market but rather to influence the equilibrium on the spot market and hence
indirectly affect total profits.
In section 3, we modify the previous analysis by allowing the producers to
build up and carryover inventories of the good they produce. We study the
question to what extent inventories can playa strategic role analogous to futures
positions. Since most commodities that are traded on futures markets can be
stored, this seems to be a desirable assumption. To focus on the inventory
decision, we temporarily ignore the futures market. We show that inventories can
also serve a strategic purpose. 5

5 It has already been shown in the academic literature that producers can
 253

Section 4 puts together the previous two frameworks. We study the futures
trading, inventory and production decisions of the duopolists and pay special
attention to the relative effectiveness of futures and inventories as strategic tools
in the hands of noncompetitive producers.
Since the aim of the paper is to focus on the strategic behavior of
noncompetitive producers that have access to a futures market, we ignore the risk
hedging rationale for futures trading (even though it may well be dominant) and
conduct the analysis under the assumption of perfect foresight. The use of this
strong and unrealistic assumption will be discussed and justified at the end of
section 4 in the light of the results obtained.

2. THE MODEL WITH FUTURES ONLY

The framework is the following: there are two periods, two producers (a duopoly)
and n price-taking speculators. 6
- At time 1, the agents can sell (buy) forward/futures contracts that call
for delivery (purchase) of the duopolists' good at time 2.7
- At time 2, actual production takes place and the agents trade on the spot
market.
There is full public information, i.e. perfect foresight since there is no
uncertainty.
The producers' total profits are, respectively,
ill = px - b + [q-p]f
il2 = py - c + [q-p]g
where ili is firm i's profit, p is the spot price (time-2 price), q is the futures price
(time-1 price), x (resp. y) is firm 1 (resp. firm 2)'s output (at time 2), b (resp. c)
is firm 1 (resp. firm 2)'s total cost and f (resp. g) is the futures sale (£>0 (resp.
g>0)) or purchase (kO (resp. g<O)) made by firm 1 (resp. firm 2) at time 1.
Total profits are the sum of profits made on the spot market, i.e. px-b

hold inventories of their goods for strategic purposes. See, for example, Saloner
(1984) and Rotemberg and Saloner (1985).
6 The number of speculators is irrelevant as long as they behave as price-
takers.
7 Although futures contracts can differ significantly from forward contracts,
we shall use indifferently both terms in this paper. This seems justified since our
analysis is conducted under certainty and, hence, forward and futures prices must
be identical.
254

(resp. py--c) and profits made on the futures market, i.e. [q-p]f (resp. [q-p]g).
The speculators' profits are
Ilk = [q-p]hk , (k=I, ... ,n)
where hk is the futures sale (if >0) or purchase (if <0) made by speculator k at
time l.
The following assumptions are made on the cost and inverse demand
functions:
p = p(x+y) with p' (x+y) < 08
b = b(x) b' > 0, b" ~ 0
c = c(y) c' > 0, c" ~ o.

2.1 The Cash Market

We solve the model backwards and first consider the production decisions of the
duopolists. They are made at time 2 for given futures positions. Each firm selects
the level of output that maximizes its profits:
firm 1 chooses x to maximize III (x) = px - b + [q-p]f
firm 2 chooses y to maximize Il 2(y) = py - c + [q-p]g.
Under Coumot behavior, the necessary first-order conditions for a
maximum are, respectively for firm 1 and firm 2
p' [x-fj + p - b' ~ 0 (la)
p'[y-g] + p - c' ~ O. (lb)
Sufficient second-order conditions are given by
p" [x-fj + 2p' - b" < 0 (2a)
p"[y-g] + 2p' -c" < O. (2b)
In the case of an interior solution, the first-order conditions say that
production should be expanded until the marginal revenue from selling another
unit on the spot market be just equal to the marginal cost of this unit. Notice
that, with forward sales, the marginal revenue expression is given by p' (x-f)+p
for firm 1, and not simply by p'x+p, as is typical, because the decrease in price
necessary to sell another unit on the spot market does not affect the units already

8 To be traded on a futures exchange, a good must be homogenous. That is

why the inverse demand function takes the form p(x,y)=p(x+y). When
considering forward contracts, one could work with heterogeneous goods. There
also exist futures contracts, traded on different futures exchanges, for goods that
differ only slightly, e.g. wheat at Chicago and wheat at Minneapolis. We shall
nevertheless keep the assumption of homogeneity in this paper.
 255

sold forward at time l.

Notice also that if the firm has taken a very large long position on the
futures market (£<0, a forward purchase) at time 1, it may then find it optimal
not to produce at time 2. However, we show below that this case will never be
the outcome under perfect foresight and Cournot behavior.
The Cournot-Nash equilibrium on the spot market is given by the pairs
(x,y) that simultaneously satisfy the four equations above.
It can be seen that the two first-order conditions implicitly define x and y
as functions of f and g.
Some comparative statics of the equilibrium indicate that
8x/ lJf = p'D/!::.. 8x/8g = -p'B/!::..
fJy/lJf= -p'C/!::.. fJy/8g = p' A/!::..,
where
A = p"(x-f) + 2p' - b"
B = pll(x-f) + p'
C = pll(y-g) + p'
D = pll(y_g) + 2p' - C"
!::.. = AD - BC.
From the second-order conditions, A<O and D<O. Moreover, we can refer
to the stability conditions to ensure that !::..>0.9 hence
8x/ lJf > 0 and By/8g > O.
Using the concept of reaction functions, we can rewrite
8x/8g = RIP' A/!::.. = R 1[By/8g] and fJy/lJf= R2P'D/!::.. = R 2[8x/lJf]
where Rl (resp. R 2) is the slope of firm 1 (resp. firm 2)'s reaction function.
The impact of a change in the level of firm 2's futures position on firm 1's
equilibrium output depends on whether firm l's reaction function is downward or
upward sloping and, similarly, the impact of a change in the level of firm l's
futures position on firm 2's equilibrium output depends on whether firm 2's
reaction function is downward or upward sloping. In general, under Cournot
behavior, the reaction functions are downward sloping10 and thus
o< 8x/ lJf < 1 0 < By/ 8g < 1
-1 < 8x/8g < 0 -1 < fJy/{}f < 0
with 18x/ {}fl > 18x/ 8g I and I fJy / 8g I> I fJy / {}fl·

9 See Dixit (1986).

10 See however Bulow, Geanakoplos and Klemperer (1985).
256

We have just seen that the equilibrium output levels depend on the levels of
the futures positions assumed at time 1 by the two firms. The choices of spot and
futures transactions are thus not independent. Hence, when producers have to
decide how much to sell (or buy) forward on the futures market, they will take
into account the effect that their futures positions will have on the spot market
equilibrium. There is a strategic rationale for trading on the futures market.
Typically, agents trade on a futures market either for hedging or for
speculating. We have just shown here that in the duopoly (or oligopoly) case
there is a third rationale for futures trading, a strategic one. In fact, under
perfect foresight (no uncertainty), the only rationale for trading on a futures
market is the strategic one. In effect, since in equilibrium the futures price will
always be equal to the spot price, no profit will ever be realized on the futures
market. Hence, nobody will use the futures market for speculating. Moreover,
since there is no uncertainty, nobody will need to hedge. Therefore, the only
reason left for futures trading is the strategic one.
The speculators's actions at time 2 are limited to delivering (or taking
delivery) of the quantities they sold (bought) forward on the futures market at
time 1.

2.2 The Futures Market

Let us then move backwards to time 1 on the futures market and start first by
analyzing the producers' decisions.
At time 1, both producers must decide how much to buy or sell forward on
the futures market, taking into account the effect that their futures positions will
have on the spot market equilibrium.
Thus, at time 1, the producers' problem is
for firm 1, to max III (f) = p(x(f,g)+y(f,g))x(f,g) - b(x(f,g))
for firm 2, to max II 2(g) = p(x(f,g)+y(f,g))y(f,g) - c(y(f,g)).
Notice that, at time 1, the terms [q-p]f and [q-p]g are absent from the
firms' profit functions. This is because, under perfect foresight, all agents forecast
the futures price to be equal to the expected future spot price which, in this case,
is also the realized spot price. If q were different from p, there would exist
unbounded riskless arbitrage possibilities.
 257

The first-<>rder conditions for the producers' problems are l1, respectively,
p' {[8xjOf] + [ByjOfJ}x + p[8xjOf] - b / [8xjOfJ = 0 (3a)
p' {[8xj Bgj + [Byj Bg]}y + p[ Byj ogj- c ' [Byj Bgj = 0 (3b)
which, after substitution of the expressions for the partial derivatives and using
the time-2 first-<>rder conditions, can be rewritten
f[8xjOfJ + x[ByjOf] = 0 (4a)
g[ByjBgj + y[8xjBgj = O. (4b)
These two time 1 first-<>rder equations plus the two time 2 first-<>rder
equations form a 4-equation system in x,y,f,g which defines the solution to the
full producers' 2-period game.
At time 1, on the futures market, each speculator selects the level of
forward sale (or purchase) that maximizes his profit
IIk(hk ) = [q-pjh k ·
Under perfect foresight, this problem is degenerate because of the no-
arbitrage condition q=p. Hence, speculators are indifferent about the levels of
their hk's. Any set of values will do for the hk's.
For the futures market to be in equilibrium, it must be the case that the
aggregate long (or short) position be zero. Mathematically, the condition is that
f + g + ;hk = O.
Since the speculators are indifferent about the level of their forward
transactions, they may as well choose them to be such that ;hk=-[f+gj. Hence,
the futures market equilibrium is characterized by q=p and Ekhk=-[f+gj.
We are now able to prove the following results characterizing the producers'
optimal futures and spot decisions 12:

Proposition 1:
a) If the producers participate in the futures market at time 1, they will
never choose to buy forward so much that it then becomes optimal not to
produce at all at time 2.
b) Moreover, if p(Ym»b/(O) and p(xm»c/(O), where xm and Ym are the
monopoly output levels, then the equilibrium will be an interior solution, with
x>O and y>O.

11 We assume an interior solution and that the second-<>rder conditions are

satisfied.
12 Proofs are given in the appendix.
258

Lemma 1: If equations (1), (2) and (3) are satisfied,13 then x>f and y>g.

Lemma 2: Let x (resp. y) be firm 1 (resp. firm 2)'s optimal output level with
forward trading and! (resp. ;r) be firm 1 (resp. firm 2)'s optimal output level
without forward trading. then, if the spot profit functions are strictly concave, 14
x<x when f<O and y<;r when g<O.
x=x when f=O and y=;r when g=O.
x>x when £>0 and y>;r when g>O.

Proposition 2: If equations (1), (2) and (3) are satisfied,~5 then pi I ~O is sufficient
to ensure that
i) x>f>O and ;r>g>O
ii) x>! and y>;r.

Under perfect foresight no profits are ever realized on the futures market
because the forward price is equal to the spot price (q=p) in equilibrium.
Therefore, total profits correspond to spot profits. Hence, firms will never choose
to take such a large long position on the futures market that their optimal output
levels will be zero because they can do as well by not trading and not producing
at all. Moreover, as proposition 1 shows, they may be able to do strictly better by
producing at least a little.
Lemma 1 says that the strategic motive will not lead a producer to sell his
entire production forward.
Lemma 2 says that forward sales lead to higher outputs and forward
purchases to lower outputs than is the case without forward trading. This is
because the marginal revenue from producing and selling another unit on the spot
market depends on the actions taken at time 1. Forward sales (purchases) rotate
upwards (downwards) the marginal revenue function, which leads to higher
(lower) optimal output levels than without forward sales (purchases). Indeed,
when part of the production is sold forward, the decrease in price necessary to sell
a further unit on the spot market, at time 2, does not affect the units sold

13 Given proposition 1( we only consider the case of an interior solution and,

hence, equations (la) and Ib) must hold with equality.
14 The spot profit function are III (x) = px - band II 2(y) = py - c.
15 See footnote 13 above.
 259

forward at time 1.
Finally, proposition 2 says that when an equilibrium exists and occurs at an
interior point and when the inverse demand function is concave, then the
equilibrium will be characterized by forward sales and hence higher output levels
than without forward trading. In this sense, forward trading in the duopoly case
(with Cournot behavior) is welfare increasing for society but producers' profits go
down.
When the good they produce is traded on a futures market, two duopolists
behaving ala Cournot face a prisoner's dilemma. If only one of them traded on
the futures market, he would be made better off. 16 However, when both do so they
end up being worse off.
In the following section, we show that when the good produced is storable,
inventories may also serve a strategic purpose.

3. THE MODEL WITH INVENTORIES ONLY

The aim of this section is to show that, when the good produced by the two
duopolists is storable, inventories may be held not only in order to minimize costs
(when costs are strictly convex) but also for a strategic purpose, just as futures
can as we have just shown.
There are still two periods and two producers but we temporarily ignore the
futures market and the speculators.
At time 1, producers can now produce for storage. No sales occur at time 1.17
At time 2, producers can decide to produce again and then sell their total
production (time 1 + time 2 productions) on the spot market.
The producers' profits are

16 Being the only producer to trade on the futures market confers the same
benefits as those accruing to a Stackelberg leader because the producers' reaction
functions in the spot market depend on the positions the producers have taken at
time 1 on the futures market (i.e. on the levels of their forward sales). Therefore,
by changing the levels of their forward sales, the two producers can shift their
reaction functions. When producer 1 is the only producer to trade on the futures
market, he can use the possibility to sell part of his production forward to move
his reaction function in such a way that it will intersect producer 2's reaction
function at the point of Stackelberg equilibrium. That is how regular Cournot
competition, at time 2, can lead to the Stackelberg equilibrium when producer 1
is the only one to trade forward at time 1. (See Allaz, 1987, for a formal proof).
17 Introducing spot sales in the first period will not affect the results as long
as the inverse demand functions of time 1 and time 2 are unrelated.
260

for firm 1, III = px - b(x-X) - B(X)

for firm 2, II2 = py - c(y-Y) - C(Y)
where IIi is producer i's profit, p is the spot price, x (resp. y) is firm 1 (resp. firm
2)'s total production or sales, X (resp. Y) is firm 1 (resp. firm 2)'s inventories,
produced at time 1 and carried over to time 2, b (resp. c) is producer 1 (resp.
producer 2)'s cost function at time 2 and B (resp. C) is producer 1 (resp.
producer 2)'s cost function at time 1. 18 19
We assume: p=p(x+y), p'<O, b'>O, B'>O, c'>O, C'>O, b">O, B">O,
c">O, C">O.

3.1 Producers' Decisions at Time 2

Starting again with time 2 decisions, both producers must choose how much more
to produce, given the inventories they have carried over from period 1, or,
equivalently, how much to sell on the spot market, to maximize their profits.
Thus,
firm 1 max III (x) = px - b(x-X) - B(X)
firm 2 max II 2(y) = py - c(y-Y) - C(Y).
Profit maximizing sales must satisfy 20
p'x + p - b' = °
p'y + p - c' = 0.
These two first-order equations implicitly define the optimal total output
levels, x and y, as functions of the inventory levels X and Y. It is easy to
calculate that
&x/ax = -b"N/A &x/BY = c"L/A
By/OX = b"M/A By/BY = -c"K/A.
where
K = p"x + 2p' - b" < ° (from firm l's second-order condition)
L = p"x + p'
M = p"y + P

18 For producer 1, X is time 1 production carried over as inventories to time

2, x-X is time 2 production and X+(x-X)=x is total production or sales.
19 The cost functions at time 1 and time 2 may differ, e.g. to account for the
cost of storage.
20 In the case of an interior solution. We also assume that the second-order
conditions are satisfied.
 261

N = p"~y + 2p' - c"~ < 0 (from firm 2's second-order condition)

A = KN -LM.
p" <0 is then sufficient to ensure that K<L<O, N <M<O and A>O and,
hence,
8x/8X> 0 8x/BY < 0
By/8X < 0 By/BY> O.
In the same way as futures affected each producer's reaction function by
modifying the marginal revenue curves, inventories also affect the reaction
functions but now through their impact on the marginal cost curves.

3.2 Producers' Decisions at Time 1

Let us move backwards to time 1. At time 1, both producers must decide how
much to produce for storage, taking into account how the storage decisions will
affect their profits on the spot market at time 2. They choose levels of inventories
that
for firm 1 max III (X) = px - b(x-X) - B(X)
for firm 2 max II 2(Y) = py - c(y-Y) - C(Y)
where x = x(X,Y) and y=y(X,Y)
The optimal storage levels must satisfy21
p' {[8x/8X] + [By/8X]}x + p[8x/8X]- b ' {[8x/8X]-I} - B' = 0
p' {[8x/8Y] + [By/8Y]}y + p[By/BY]- c' {[By/BY]-I} - C' = 0
Substituting the time 2 first-order conditions and simplifying yields
p' [By/8X]x+ b ' -B' = 0 (5a)
p/[8x/8Y]Y+C / -C/=O (5b)
The terms b I -B I and c I -C I tell the producers how to divide production
between the two periods so as to minimize production costs. If inventories were
carried over for the only purpose of minimizing total production costs, the
first-order conditions would simply be
b ' -B' = 0
c' - C' = 0
and production would be divided between the two periods in such a way that
marginal costs be equal between periods.
However, firm 1 (resp. firm 2)'s first--order condition includes a

21 In the case of an interior solution. We also assume that the second-order

conditions are satisfied.
262

supplementary term, p'[ByI8X]x>O (resp. p'[axloy]y>O). This is because, in a

two-period framework, inventories can also be used by the producers for a
strategic purpose, and these terms represent the strategic effect.
If the cost functions are identical in the two periods,22 i.e., b=B and c=C and
strictly convex, then
X>x-X andY>y-Y.23
With identical cost functions (zero cost of storage) the producers should
divide production equally between the two periods to minimize costs. Therefore,
there must exist another reason for producing more than that in the first period:
the strategic rationale. Thus, for strategic reasons, each firm produces more in
the first period and carries more inventories than simple cost minimization
requires. The strategy is to produce more at time 1 because that diminishes
production costs at time 2 and, hence, enables the firms to playa more aggressive
sale policy on the spot market. 24
We have just shown that both futures contracts and inventories could serve
a strategic purpose when producers do not behave competitively. Since most
commodities that are traded on futures exchanges are storable, our next objective
is to find out how a firm's optimal policy will combine futures and inventories
when both of them are available to the firm. For that purpose, we put together
the two models we have just developed.

4. THE MODEL WITH BOTH FUTURES AND INVENTORIES

There are two time periods, two producers and n price-taking speculators.
At time 1, all agents can sell (buy) futures contracts that call for the
delivery (purchase) of the good produced by the duopolists. Producers can also

22 This would imply a zero cost of storage.

23 Proof: B'=b'+p'[By/8X]x. Since BO=bO (the functions are identical)
and the second term on the right-hand side is positive, the result follows
immediately from the strict convexity of the cost function. The proof for firm 2 is
similar.
24 Notice that, in the case of constant returns to scale (constant marginal
cost), if both producers have identical cost functions across periods, i.e., bO=BO
and cO=C(), with b'=B'=constant and c'=C'=constant, then, the second
period equilibrium will be a corner solution with x-X=O and y-Y=O (zero
production at time 2) and there can be multiple equilibria in the first period.
Proof: substitute b'=B' and c'=C' in the first-<lIder conditions for X and
Y. They reduce to p' [By I oX]x and p' [axl OY]y which are strictly positive since
p'<O, ByI8X<O and x>O, y>O when p"~O_ Therefore, X and Y should be
 263

produce for storage. However, no spot sales occur at time 1. 25 Quantities produced
at time 1 are carried over as inventories to time 2.
At time 2, producers can produce again and then sell their total production
on the spot market.

4.1 The Cash Market

At time 2, both producers must decide how much to sell on the spot market,
given their inventories carried over from time 1 and given the positions they have
assumed on the futures market at time 1. The optimal sales levels must solve
max II 1(x) = p(x+y)x - b(x) - B(x-X) + [q-p(x+y)]f;
Max II2(y) = p(x+y)y - c(y) - C(y-Y) + [q-p(x+y)]g
where the symbols carry their previous meanings.
Assuming an interior solution, the first- and second-order conditions for a
maximum are
p'[x-f] + p - b' = 0 p' [y-g] + p - c' = 0
and
p" [x-f] + 2p' - b" < 0 p"[y-g] + 2p' -e" < o.
The two first-order equations implicitly define x and y as functions of the
four time-l variables: f,g,X,Y. The following eight partial derivatives can be
calculated:
fJx/8f=p'D/l:1 fJx/8g=-p'B/l:1 fJx/IJX=-b"D/l:1 fJx/8Y=c"B/l:1
fJy/8f=-p'C/l:1 fJy/ 8g=p' A/l:1 fJy/8X=b"C/l:1 fJy/8Y=-e" A/l:1
where the letter A, B, C, D, and l:1 represent the same expressions as before. 26

4.2 Time 1 Decisions

Moving to time 1" the producers must now select how much to produce for
storage and what positions to take on the futures market. 27 They must solve

increased until x-X and y-Y, the second period production levels, are zero.
25 Introducing spot sales at time 1 will not affect the results provided that
the demand at time 1 and time 2 are unrelated.
26 See the producers' maximization problems at time 2 in the model with
futures only.
27 As before, in the model with futures only, the speculators' problems are
degenerate since there is no uncertainty and, hence, no profits can be realized on
the futures market. For Simplicity we assume that they are willing to take the
matching positions from the producers. The futures market equilibrium is again
described by a zero net aggregate long (or short) position and the equality
between q and p.
264

for firm 1, max III (f,X) = px - b(x) - B(x-X)

for firm 2, max II 2(g,Y) = py - c(y) - C(y-Y)
where p=p(x+y), x=x(f,g,X,Y) and y=(f,g,X,Y).
For firm 1,
fJIIl/fJf = p' {[fJx/&fJ+[fJy/&fJ}x+p[fJx/fJfj-b/[fJx/&fJ = 0 (6a)
fJIIl/fJX = p' {[fJxlfJX)+[fJylfJX]}x+p[fJxlfJX)-b ' {[fJxlfJX)-I}-B' = 0 (6b)
Collecting terms,
[fJx/&fJ{p/X + p - b / } + [fJy/&fJp/X = 0
[fJxlfJX){p/X + p - b / } + [fJy/fJX)p/X + b ' - B' = O.
Substituting the time-2 first-order conditions and simplifying yields
f[ fJxl &fJ + x[ fJy I &fJ = 0
p/f[fJxlfJX) + p/x[fJylfJX) + b ' -B' = O.
Using fJxl fJX=-[ fJxl &fJb I I Ip I and fJy I fJX=-[ fJy I fJfjb I I Ip I , firm 1's first-
order conditions for f and X reduce to
f[ fJxl &fJ + x[ fJy I &fJ = 0
b ' -B' = O.
Similarly, for firm 2,
g[ fJy I fJg) + y[ fJxl fJg) = 0
c' - C' = O.
It can be noticed that the first-order conditions for f and g are identical to
those obtained (for f and g) in the model with futures only, and that the strategic
terms have dropped out from the first-order conditions for X and Y.
These two remarks lead to the following separation result:

Proposition 3:
Under perfect foresight, when the duopolists can use both futures and inventories,
the firms' optimal policies are to use inventories only to minimize costs (when
costs are strictly convex) and then to rely exclusively on forward transactions to
carryon their strategic aims.

Let us summarize our findings. In section 2, we showed that two Cournot

duopolists producing a good that is traded on a futures market would sell part of
their production forward for a strategic purpose. Thus we showed that, besides
hedging and speculation, there could exist a third rationale for trading on a
futures market when the underlying cash market was not perfectly competitive.
In the third section we saw that, when the good produced was storable,
 265

inventories could also serve the same strategic purpose. Forward sales made at
time 1 affect the marginal revenue curve for spot sales at time 2. Inventories
carried over from time 1 affect the marginal cost curve at time 2. In both cases,
the producer is less price sensitive at time 2 on the spot market and, hence, tends
to increase his output.
In the case of storable goods, a producer can use either forward sales and/or
inventories to serve his strategic aim. However, we obtained, in the third section,
a separation result that said that although the producer could use both futures
and inventories for his strategic aim, when both of them were available, he would
use inventories only to minimize costs and then rely exclusively on futures for
strategy. In this sense, futures dominate inventories as a strategic tool in the
hands of imperfectly competitive producers.
What is the reason for that separation result? A first clue may be obtained
from Tinbergen's law in macroeconomics which says that a country needs to have
at its disposal as many independent instruments as it has policy objectives.
Applying this reasoning to our microeconomic model, we see that each producer
has two objectives: cost minimization (a necessary behavior to maximize profits)
and strategy. He also disposes of just two instruments: futures and inventories.
Since inventories can serve either objective whereas futures can only serve the
strategic one, it seems reasonable to rely on inventories to minimize costs and on
futures for strategy.
A second explanation has to do with the assumption of perfect foresight.
We have seen that, under perfect foresight, trading on the futures market is
costless in the sense that no profits, positive or negative, can be made on the
futures market because of the no arbitrage condition q=p.28 Hence, the cost of
establishing a strategiC position is nil. This is not the case with inventories. To
achieve its strategic aim through the use of inventories, a firm must produce
more at time 1 than at time 2 and, hence, increase total costs since the cost
function is strictly convex by assumption. Without strategic playing, production
would be equally divided between the two periods so as to minimize total
production costs over the two periods. Therefore, futures enjoy a comparative
cost advantage over inventories as a strategic tool under perfect foreSight.
This naturally leads us to the question of the robustness of our results. They
have been derived under three main assumptions:

28 We ignore transaction costs.

266

i) the sequential nature of the decision process, i.e. the assumption that
production occurs only after trading took place on the futures market;
ii) perfect foresight;
iii) Cournot-type behavior of the two duopolists.
The first assumption is fundamental because there cannot be any strategic
play in a one-period model where forward and spot decisions are taken
simultaneously.
Whether forward and spot decisions are taken simultaneously or
sequentially is an empirical question. In the ten papers (including this one)
mentioned in the introduction, eight assume a similar sequential decision process.
Let us simply say that results obtained under this assumption are compatible
with some observed cartel behavior, notably that of the coffee cartel Pancafe. 29
On the other hand, the assumption of perfect foresight is obviously
unrealistic. It is used in order to focus exclusively on the strategic rationale for
futures trading. It also greatly simplifies the analysis. When uncertainty is
introduced in the model, e.g. by supposing that the spot price is not known yet at
time 1 when trading takes place on the futures market, strategic purposes
interact with risk hedging if the producers are risk averse. However, if both
producers and at least one speculator are risk neutral, then the (expected values
of the) results will be identical to those derived under perfect foresight. 3D Hence,
perfect foresight can be seen, in some way, as the limiting case of uncertainty
where all agents are risk neutral.
When the producers are risk averse, they will trade on the futures market
not only for the strategic reason but also to hedge their profits against a possible
low spot price realization. We saw above that two duopolists behaving a la
Cournot would take a short position on the futures market so far as strategy was
concerned. Hence, under Cournot behavior, the strategic and risk hedging
motives both reinforce each other in leading the firms to sell forward at time 1. 31
Risk averse firms face three objectives - cost minimization, strategy and
risk hedging - but have only two instruments at their disposal, futures and

See Allaz (1988).

29

Provided that the inverse demand and cost functions are linear.
30

31 These conclusions were derived in a less general model without inventories

where it was assumed that the cost and inverse demand functions were linear. See
Allaz (1987).
 267

inventories. In that case, we no longer expect the separation result (proposition

3) to hold under uncertainty.32 Even in the case where both producers are risk
neutral, if all speculators are risk averse, the separation result cannot be expected
to hold since the speculators will now require a risk premium to trade on the
futures market and, therefore, trading on the futures market will no longer be
costless for the firms.
The assumption of Cournot behavior is often made, first because Cournot
behavior is seen as a useful benchmark case and, second, because of its technical
tractability. It is however far from being innocuous. Allaz (1987) has shown 33 that
if we have price-competition it la Bertrand at time 2 on the spot; market (instead
of Cournot-quantity competition), then the positions initiated at time 1 on the
futures market will have no impact on the spot market equilibrium outcome at
time 2.
Allaz (1988) and Anderson and Brianza (1989) have considered different
types of behavior (i.e. conjectural variations) at time 1 on the futures market
wrule still assuming Cournot behavior at time 2.34 They show that collusive
behavior at time 1 leads the producers to take a long position (a forward
purchase) on the futures market in order to restrain production at time 2. Buying
forward at time 1 reduces the marginal revenue from any unit produced and sold
on the spot market at time 2. This way, the firms refrain from overproducing at
time 2 and globally achieve monopolistic profits.
These results prove that the type of behavioral assumptions made in the
model plays a crucial role. More research effort should be devoted to this point.
Further refinements of the model would certainly have to introduce a more
soprusticated behavior for the speculators and emphasize the role that agents'
expectations can play along the lines of Brianza, Phlips and Richard (1987) and
Phlips and Harstad (1990). The generalization to an oligopoly rather than a
duopoly would offer harder technical work but should not modify our results.

32 Unless all agents are risk neutral and the inverse demand and cost
functions are linear.
33 In a model without inventories.
34 Anderson and Brianza (1989) explain why it may be more realistic to
assume non-Cournot behavior at time 1 rather than at time 2. Ideally, we would
have general conjectural variations both at time 1 and at time 2 but this is
technically very difficult.
268

5. CONCLUSION
We have shown in this paper that two duopolists producing a good that is traded
on a futures market will use the opportunity offered by the futures market to sell
forward part of their production in order to affect the equilibrium outcome on the
spot market. In the case of Cournot behavior, the producers end up worse off
because, when ignoring the impact of their actions on their competitors' actions,
they tend to increase output and, hence, globally overproduce which leads to
lower profits.
Inventories can also be used strategically, just like futures, but they suffer a
comparative cost disadvantage over futures under perfed foresight.
Whether the producers take short or long positions on the futures market
and, hence, increase or decrease production at time 2 depends on the type of
behavior that is assumed for the duopolists. 35 The result that forward transactions
can be used for a strategic purpose can however be considered to be fairly robust
provided that forward and spot decisions are taken sequentially.
Further research efforts could be devoted to improving the information
structure of the model and to modelling more sophisticated agents' expectations.
A complete model ought certainly to be derived in a framework where
uncertainty plays a crucial role.

REFERENCES
Allaz, B (1987), Strategic Forward Transactions Under Imperfect Competition-
The Duopoly Case, unpublished Ph.D. dissertation, Princeton University.
Allaz, B (1988), "Oligopoly, Uncertainty and Strategic Forward Transactions",
working paper, Centre HEC-ISA.
Allaz, B and J-L Vila (1989), "Cournot Competition, Forward Markets and
Efficiency", mimeo, Centre HEC-ISA and Stern School of Business, New
York University.
Anderson, R W (1984), "The Industrial Organization of Futures Markets: A
Survey" in R vi Anderson (ed.), The Industrial Organization of Futures
Markets, D.C. Heath, Lexington.
Anderson, R W (1985), "Market Power and Futures Trading for Durable Goods",
working paper, City University of New York, Graduate Center.
Anderson, R Wand T Brianza (1989), "Cartel Behavior and Futures Trading",
working paper, City University of New York, Graduate Center.
Anderson, R Wand M Sundaresan {l984), "Futures Markets and Monopoly" in
R W Anderson (ed.), The Industrial Organization of Futures Markets,

35 This point is not proved in this paper which has only considered Cournot
behavior.
 269

D.C. Heath, Lexington.

Brianza, T, L Phlips and J-F Richard (1987), :Futures Markets, Inventories and
Monopoly", Core Discussion Paper 8725.
Bulow, J I, J D Geanakoplos and P D Klemperer ~1985), "Multimarket
Oligopoly: Strategic Substitutes and Complements', Journal of Political
Econom1/> 93, 488-511.
Dixit, A (1986), "Comparative Statics for Oligopoly", International Economic
Review, 27, 107-22.
Eldor, R and I Zilcha (1986), "Oligopoly, Uncertain Demand and Forward
Markets", working paper, Department of Economics, Tel Aviv University.
Greenstone, W D (1981), "The Coffee Cartel: Manipulation in the Public
Interest", Journal of Futures Markets, 1, 3-16.
Newbery, D M G (1984), "The Manipulation of Futures Markets by a Dominant
Producer" in R W Anderson (ed.), The Industrial Organization of Futures
Markets, D.C. Heath, Lexington. '
Phlips, Land R M Harstad (1990), "Interaction Between Resource Extraction
and Futures Markets: A Game-Theoretic Analysis" in R Selten (ed.),
Game Theory in the Behavioural Sciences, Springer Verlag, Heidelberg.
Rotemberg, J J and G Saloner (1985), "Strategic Inventories and the Excess
Volatility of Production", working paper, Sloan School of Management
and Department of Economics, M.LT.
Saloner, G (1984), "The Role of Obsolescence and Inventory Costs in Providing
Commitment", working paper, Department of Economics, M.LT.

APPENDIX
Proof of proposition 1:
a) Since, under perfect foresight, q=p (forward price = spot price) in equilibrium,
no profits are ever realized on the futures market in equilibrium. Therefore, total
profits are equal to spot profits.
Now, let !<O and g<O be such that if f<! (g<g), then it is optimal not to
produce at time 2,36 Thus, if producers choose f<! and g<g, then total profits are
nonpositive (negative if there are sunk costs).
Since not participating in the futures market and not producing does not
yield a worse outcome (and may save the sunk costs), and participating in the
futures market with f>! and g>g and producing at least a little may yield a
better outcome (see (b) below), it is never optimal to choose f<! and g<g (Le.
this strategy is dominated).

36 ! is defined by II(O,!)=II(xo,D, Le. ! is the level of f for which firm 1 is

indifferent between producing or not (x o is the optimal output level for a given
level of forward sales). See Allaz (1987) for a proof that if f<! (resp. g<g), then it
is optimal not to produce at time 2. In this case, we obtain a corner solution for x
and y. Refer to the first-order conditions (la) and (lb) above.
270

b) Since, under perfect foresight, total profits are equal to spot profits, it is
enough to show that producing a little will yield strictly positive profits:
dII 1,spot(x=0,y=Ym)/dx = p(Ym) - b/(O)

which are strictly positive when p(ym»b/(O) and p(xm»c/(O), where Ym and
xm are the monopolistic output levels.

Proolollemma 1:
From equation (la): x - f = -(P-b/]/p/.
Substituting the expressions for the partial derivatives {}xl Of and fJy IOf into
equation (3a) yields
p/X[p/--c l l ] + {p" [y-g] + 2p' - c" }[p-b/] = 0,
which can be rewritten as
-[p-b / ] = {p/X[p/--c"]}/{pll[y-g] + 2p' - cl l } < O.
Hence, x-bOo
The proof that y-g>O is similar.

Prool 01 lemma 2:
Firm l's profits with forward trading is: II(x) = px - b + [q-p]f.
Firm l's profits without forward trading is: II(!> = P! - b.
dII/dx = pi (xo+y)xo + p(xo+y) - b/(xo) -p/(xo+y)f= 0 (A1)
dII/~ = pi (!o+Y)!O + p(!o+y) - b/(!o) = 0 (A2)
Rewrite (A1) as:
dII/dx = pi (xo+y)xo + p(xo+y) - b/(xo) = pi (xo+y)f. (A3)
If the spot profit function, px-b, is concave, then pl'(Xo+y)x o
+2p ' (xO+y)-b 'l (xo)<O for all nonnegative values of x and y.
When f=O, (A1)=(A2) and, hence, xo=!o.
When bO, the right-hand side of (A3) is negative (since pi <0) and,
therefore, x must go up for the left-hand side to decrease.
When £<0, the right-hand side of (A3) is positive (since pi >0) and,
therefore, x must decrease for the left-hand side to increase.

Proo101 proposition 2:
Substitute (la) into (3a) and (lb) into (3b) and rearrange the terms to get
 271

fIx = {pll[y-g] + p'}/{pll[y-g] + 2p' - ell} = C/D.

g/y = {p"[x-f] + p'}/{p"[x-f] + 2p' - b"} = B/A.
From lemma 1, y>g and x>f. hence p" <0 yields D<C<O and A<B<O.
Since these equations hold in the case of an interior solution, x>O and y>O and
result (i) follows.
Result (ii) is obtained by applying lemma 2 with £>0 and g>O.
 CHAPTER 10

MONOPSONY POWER AND THE PERIOD OF COMMITMENT

IN NONRENEWABLE RESOURCE MARKETS

Larry Karp!

1. INTRODUCTION
International markets for many commodities are imperfectly competitive. The
study of such markets is often complicated by their dynamic nature. It is useful
to divide commodities for which dynamic considerations are important into two
groups: "reproducible goods", such as coffee, and nonrenewable resources, such as
oil. The first group is characterized by the fact that investment decisions, e.g.
tree planting, are an essential part of planning. These decisions require making
predictions about future levels of prices and quantities. Although investment
decisions are usually important for commodities in the second group as well, their
distinguishing feature is the scarcity value of the resource: extracting one unit
today forecloses the option of extracting that unit at a later time. The current
extraction decision, much like the current investment decision, depends on
expectations of future prices. Despite these similarities, the situation of a pure
monopsonist (i.e., one who faces no competition) who uses tariffs to extract rent
from competitive producers, and who in unable to make binding commitments
about his future behavior, differs greatly depending on whether the monopsonist
imports a reproducible good or a nonrenewable resource. This chapter
concentrates on the case of a monopsonist who imports a nonrenewable resource;
we use the reproducible good as a point of contrast. We emphasize the behavior
of the monopsonist as his period of commitment becomes small. We discuss how

! This chapter benefitted from many conversations with David Newbery. The
usual disclaimer applies.
273
L. Phlips (ed.), Commodity, Futures and Financial Markets. 273-300.
© 1991 Kluwer Academic Publishers.
274

financial markets, such as futures and forward markets, may either enhance or
detract from the exercise of market power.
Agents who are unable to make binding commitments regarding their
future behavior may find the possession of (what appears to be) market power of
little value to themj indeed such market power may be disadvantageous. This
observation has been applied to a range of macro- and micro-economic problems.
The important features of these problems are: (i) there is a dominant agent who
cannot make commitments, but who recognizes his ability to affect his
environment, (ii) there are many small agents who have rational expectations
and who take the environment (e.g. prices) as given, and (iii) there is neither
objective randomness nor uncertainty about the abilities and goals of other
agents.2
These features describe elements of the problem of a pure monopsonist who
imports a nonrenewable resource and attempts to capture rents by means of a
tariff or some other trade restriction. The exporting firms, who behave
competitively, have rational expectations and their current level of sales depends
on prices in the current and future periods. We assume that there is no
uncertainty or intrinsic randomness, so that rational expectations correspond to
point expectations which are correct in equilibrium. The importer is unable to
make commitments about his behavior in the future.
After reviewing several papers which have studied this situation, we discuss
the extension contained in this paper. In section 2, we consider the relation
between the nonrenewable resource problem studied here and the broader class of
problems that share the three features listed above. The specific problem studied
here is of intrinsic interest since the analysis suggests what type of forces operate
to limit market power in some commodity markets, and the extent to which
institutions such as futures and forward markets oppose or encourage those
forces. The analysis is also useful for the light it sheds on the more general class
of problem. Sections 3 and 4 provide the formal analysis. Section 5 compares the
nonrenewable resource and the reproducible goods models. A conclusion
summarizes the results.

2 Not all of these assumptions are crucial, but they emphasize the important
features of the problem. For example, there may be several rather than one
dominant agent. The assumption of perfect information and no uncertainty is
made partly to simplify the exposition, but also because the Markov equilibrium,
which is defined and used below, is more plausible under those circumstances.
 275

Newbery (1976) and subsequently Kemp and Long (1980) pOinted out that
the optimal policy for the importer of a nonrenewable resource is time-
inconsistent. That is, if the importer were able to make binding commitments, he
would commit himself to a trajectory from which he would later want to deviate. 3
Therefore, the prohibition against making binding commitments decreases the
present discounted value of market power for the importer. The
time-inconsistency occurs because at the initial time the importer would (for
example) like to promise that future tariffs will be relatively high. If the
competitive producer believes this, he expects the market price to be low in the
future, and is consequently willing to sell in the current perio'd for a low price.
The importer's optimal time-inconsistent policy balances the current gains of
buying at a low price against the future opportunity cost associated with high
future tariffs (and low future consumption). The benefits, to the importer, of this
policy tend to be large near the beginning of the trajectory and the opportunity
costs of foregone consumption tend to be large later in the trajectory. In the
future, then, the importer faces low benefits and high costs of keeping to the
trajectory that was announced at the initial period: in the future the importer
would like to deviate from that trajectory. The optimal policy is time-
inconsistent.
Maskin and Newbery (1990) use a two period model to show that market
power can be disadvantageous. That is, the present discounted payoff to the
monopsonist may be higher in the competitive equilibrium than in the
equilibrium where the monopsonist recognizes his market power and is unable to
make binding commitments about his future behavior. This possibility arises
because, absent the ability to make binding commitments, the competitive
equilibrium is not feasible for the monopsonist. The existence of competing
buyers is necessary for market power to be disadvantageous.
Karp (1984) considers an infinite horizon problem in which the monopsonist
is the sole consumer of the good, so that the possibility of disadvantageous

3 In special circumstances the monopsonist's optimal policy is

time-consistent; this occurs if (i) the monopsonist is the sole consumer and
extraction costs are independent of remaining stock, or, (ii) extraction costs are
constant and both the monopsonist and the Rest-of-the-Wodd (competitive)
demand have the same constant elasticity. See Karp and Newbery (1989a) for a
review of this material. Throughout the introduction we assume that these
circumstances do not obtain. Therefore the optimal plan will be time-inconsistent
and the requirement of consistency strictly reduces the monopsonist's payoff.
276

market power does not arise. The monopsonist seeks to maximize the present
discounted value of the stream of consumer surplus and tariff revenue, and is able
to make a binding commitment for the tariff applied to a stock of size 5. The
motivation of this assumption is that the exporter exhausts a sequence of mines,
each of which contains 5. The assumption implies that it is possible to sign a
contract that governs the terms of sale of the contents of the current mine, but
not of subsequent mines. That is, there is limited ability to precommit. As 6-tO
the monopsonist's power vanishes: the limiting equilibrium is equivalent to
perfect competition.
The exporter's current supply depends on future 'prices (and thus future
tariffs) as well as on the current price. Given the assumption of an infinite
horizon and stationary demand and cost functions, the future tariff levels depend
(plausibly) only on the level of the stock remaining at the time the tariff becomes
effective. Therefore the choice of the "current tariffs" (the tariff trajectory
associated with the sale of 6) affects the timing but not the level of "future tariffs"
(the sequence of trajectories, each of which is associated with the sale of an
amount 5). The current tariff consequently effects the timing but not the level of
the trajectory of future prices. As 5 becomes small, the interval over which it will
be consumed also becomes small. (In equilibrium the rate of extraction, which by
assumption equals the rate of consumption, approaches zero only as the resource
is exhausted.) The benefit to the importer resulting from affecting the timing of
future policies is of a smaller order of magnitude than is the cost resulting from
delaying current consumption by means of a tariff. Therefore as &-+0 it is optimal
to set the tariff equal to zero.
It is not surprising that decreasing the monopsonist's ability to precommit
results in a lower present discounted value of market power; the importance of
the previous result is its demonstration that the limiting value of monopsony
power is in this case zero. However, it is worth enquiring how sensitive the result
is to the form that precommitment takes. Before taking up this question) which
is central to the paper, it is useful to describe the monopsonist's problem in a
slightly different manner. This simplifies matters by introducing terms which
would otherwise be ambiguous and also provides a convenient way of discussing
various equilibrium concepts.
We have described the monopsonist's problem as one of maximizing the
present discounted value of the sum of consumer surplus and tariff revenues
(hereafter "utility"), subject to the behavior of competitive sellers who have
 277

rational (point) expectations and who choose their sales trajectory to maximize
the present discounted value of the resource, and subject to a limited (perhaps
nonexistent) ability to make binding commitments about future behavior. An
alternative way of posing the problem is to break up the single maximization
problem, i.e. to replace it by a sequence of problems, each with a finite horizon.
We replace the monopsonist by a fictitious sequence of monopsonists, each of
whom makes decisions during one of the time intervals. We define the time
horizon of each element of the sequence of problems as a "period". At each period
the "current" monopsonist wants to maximize the present discounted value of
utility. That is, each monopsonist wants to maximize the sum of the payoff in his
own period and the discounted value of his "successor's" payoff. However, each
monopsonist is indifferent to the payoff of his "predecessor" except insofar as that
corresponds to his own payoff. This indifference incorporates the time-consistency
constraint. This construction replaces an optimization problem which involves a
time-consistency constraint with a noncooperative game amongst a succession of
agents, who make their moves sequentially. Note that the game is non-
~ooperative. The fact that the fictitious sequence of monopsonists represents the

single monopsonist who is constrained by the inability to make commitments is

embodied in the fact that the current monopsonist cares about his successors'
welfare.
With this reformulation of the problem we can now unambiguously speak of
the behavior of "current and succeeding monopsonists"; the length of the period
of precommitment is equivalent to the length of time during which the current
monopsonist holds tenure. A decrease in each monopsonist's tenure corresponds
to a decrease in the ability to precommit. The limiting case occurs when each
tenure is of infinitesimal length. the assumption in Karp 91984) implies that the
period of commitment is endogenous. This assumption may be reasonable in the
context of nonrenewable resource, where one can think of a succession of mines,
each of which is small relative to the total initial stock. However, perhaps a more
plausible alternative is to think of agents being able to make commitments for a
period of time which is exogenous, as, for example, is likely to be the case with
elected or appointed government officials. Hereafter we denote an exogenous
period of commitment as f.
A decrease in either period of commitment (f) or in the amount of stock for
which the monopsonist can make commitments (fi) are both likely to lead to a
reduction in market power. However, in the limiting case where f-lO, market
278

power does not necessarily vanish. We construct an equilibrium in which the

monopsonist retains some market power even as the period of commitment
becomes infinitesimal. This provides a lower bound on the benefit of monopsony
power, and it demonstrates the sensitivity of the lower bound to the nature of the
precommitment technology.

2. RELATIONS WITH OTHER PROBLEMS

Before turning to the formal model we discuss the more general issue of time-
consistency. There is a large literature on this subject; reference to this literature
helps motivate the interest in the problem at hand and also makes it easier to
understand the results we obtain. Although the issue of time-consistency plays an
important role in both macro- and micro-economic models, we restrict attention
to the latter. 4 Karp and Newbery (1989a) survey the literature on time-
inconsistency problems in nonrenewable resource models.
Time-consistent equilibria are generally considered more plausible, i.e.,
empirically relevant, than (optimal) time-inconsistent equilibria. The limiting
case where the ability to precommit becomes infinitesimal is seldom of intrinsic
empirical interest. However, this limiting case has been widely studied because it
is often analytically more tractable than the problem in which the period of
commitment is finite, and because the former provides a bound on the loss in
profits due to the limitation in the ability to precommit. 5
Perhaps the most widely studied industrial organization problem in which
time-consistency plays a central role concerns the monopolist who produces and
sells a durable good. Anderson (1989) discusses the equilibrium of a durable
goods monopolist and shows how the existence of futures markets can increase
the seller's market power. A durable good yields a flow of services, so the sales

4Kydland and Prescott (1977) were among the first to discuss the problem of
time-consistency in a macro-economic context. Since then a variety of models
have demonstrated that the rational expectations of private agents and the
inability of governments to make binding commitments may make macro-policy
ineffective. Chari, Kehoe and Prescott (1988) review the literature on time-
consistency in macro-policy. Similar problems arise in international macro models
where two or more governments interact.
5 Welfare under an infinitesimal period of commitment provides a lower
bound in the Markov equilibrium we use in this paper. This is not true for other,
more sophisticated types of equilibria. See, for example, Ausubel and Deneckere
(1989).
 279

price of the good at a point in time depends on the value of the future flow of
services; this stream of value depends on the amount of services, and thus on the
stock of the durable good in the future. The monopolist would like to raise the
current price by promising that future sales, and hence the future stock, will be
low. However, in the next period the monopolist takes previous sales as given and
maximizes profits by selling more than he had previously announced would be
sold. Buyers with rational expectations anticipate this in the first period; in the
absence of the ability to precommit to low future sales, the initial promise is not
credible and cannot be the basis for an equilibrium. This problem parallels very
closely that of the monopsonist importer of a nonrenewable resource described in
the Introduction. Recall that the monopsonist importer would like to deviate
from the optimal time-inconsistent policy by raising or lowering the tariff (thUS
lowering or raising the price received by the exporter). 6
Coase (1972) conjectured that if the durable goods monopolist produces at
constant cost, he looses all market power as the period of commitment, f,
approaches 0: monopoly power vanishes "in the twinkling of an eye". This
conjecture, if correct, has important welfare implications, since it suggests that in
certain situations monopoly and perfect competition are similar, and that in the
limiting case they are identical.
Models of the durable goods monopolist include papers by Stokey (1981),
Bulow (1982), Bond and Samuelson (1984), Gul Sonnenschein and Wilson (1986),
Kahn (1987), and Ausubel and Deneckere (1989). It is widely accepted that if
production costs are constant and there is no capacity constraint, and if in
addition buyers' expectations about future sales depend smoothly on only the
current stock of the good, then the Coase Conjecture is correct. Under more
general circumstances the Conjecture is incorrect. There are, then, two avenues
for overturning the Conjecture. The first is to modify the technical description of
the problem by, for example, introducing capacity constraints or convex
production costs. The second route is to endow the buyers with sophisticated
beliefs about the monopolist's behavior off the equilibrium trajectory. This leads
to a game-theoretic model of buyer and seller interaction. For example, if buyers
are convinced that the monopolist will sell at marginal cost should he ever

6 If the importer is a pure monopsonist he wants to deviate from the first-

best, inconsistent, policy by lowering the tariff; if he faces competing buyers, he
may want to deviate by raising the time-inconsistent tariff.
280

deviate from a "reference trajectory", then in the event that the monopolist does
deviate, he will be unable to charge a price higher than marginal cost; this
(credible) punishment may, in some circumstances, support a reference trajectory
arbitrarily close to the first best (precommitment) path. This equilibrium is
subgame perfect, and thus time-consistent. In this case, the inability to pre-
commit is virtually painless for the monopolist.
These conclusions from the durable goods monopolist model offer two
important insights for the problem of the monopsonist who consumes a non-
renewable resource. The first is that unless some restriction, in addition to
rational expectations, is imposed on the beliefs held by the competitive resource
sellers (who correspond to the buyers in the durable goods problem) there is
likely to exist a continuum of consistent (indeed, subgame perfect) equilibria,
some of which provide a level of utility for the monopsonist nearly equal to that
obtained under precommitment. A plausible restriction is that sellers' beliefs
about future prices depend only on information that has direct economic
relevance. In the present model, this is the current stock of the nonrenewable
resource. (This corresponds to the assumption, in the durable goods model, that
competitive agents' expectations of future sales depend only on the stock of the
durable good.) We refer to this as the Markov assumption, since it means that
the future depends on the current "state", but not on the manner in which the
current state was reached; the state is the level of the resource remaining in the
ground. This rules out the sort of reputational equilibria described above, in
which the competitive agents' beliefs change radically as the result of a small
deviation from equilibrium behavior on the part of the dominant agent.
The pragmatic defense of the Markov assumption is that it often implies a
unique stationary equilibrium; in this case the model yields welfare implications.
Perhaps more importantly, the Markov assumption seems plausible, in that it
excludes the possibility that a large agent interacts strategically with the
"competitive ocean".
The second insight provided by the literature on the durable goods
monopolist is that in cases where the competitive steady state is not reached
instantaneously, the Markov equilibrium with infinitesimal period of commit-
ment is not likely to duplicate the competitive equilibrium; that is, market power
does not vanish in the twinkling of an eye. In the types of models we are
considering, the Markov assumption implies that the equilibrium is characterized
by an endogenous function of the state (see equation (I), below). For both the
 281

durable goods monopolist and the nonrenewable resource monopsonist, this

function determines the price facing competitive agents. If the competitive steady
state is reached instantaneously, as is the case with durable goods produced
under constant costs, the competitive price function is simply a constantj the
same (trivial) function provides a Markov equilibrium when an agent has market
power but the ability to commit to future actions for only an infinitesimal period.
If, however, the competitive steady state is not reached instantaneously, as is the
case with the durable goods model when costs of production are convex or with
the nonrenewable resource model considered here (where the steady state occurs
when the resource is exhausted), the endogenous competitive price function does
not provide an equilibrium for the case where an agent has market power, even if
his period of commitment is infinitesimal.
To see why this is so, consider an abstract problem in which the "state" is a
variable y, and the endogenous price function in the competitive equilibrium is
P*(y)j y may be the stock of the durable good, or of the nonrenewable resource,
or something else, depending on the context. If P*(y):p*, a constant, demand is
infinitely elastic in every period, and there is no possibility of market power. If,
on the other hand, P*(y) is a non-trivial function, then an agent with market
power still takes the current price as given at every point in time, since he takes
the current value of the state and the endogenous function as givenj however, if
he is able to influence the rate of change of the state, he is able to affect the
future trajectory of prices. In this case the agent with market power who faces
the function P*(y) solves a different control problem than would his competitive
counterpart who takes the trajectory of the realizations (in the competitive
equilibrium) of P*(y) as given. Therefore the realizations of the state, y, will be
different in the competitive and the noncompetitive equilibria. Consequently,
market power does not vanish even as the period of commitment becomes
infinitesimal.
The important result is that determining whether the competitive steady
state is reached immediately, provides a test for whether market power vanishes
in a Markov equilibrium as the period of commitment approaches O. In the
nonrenewable resource model, the steady state, which implies exhaustion of the
resource, is not reached instantaneouslyj by analogy with the durable goods
monopoly, market power does not vanish in a Markov equilibrium as the period
of commitment vanishes.
Before presenting the formal model, one further issue should be clarified.
282

We have spoken of the equilibrium we are interested in as "time-consistent".

This term is generally understood to mean that the dominant agent has no
incentive to deviate from the equilibrium trajectory; that is, the continuation of
the initial equilibrium represents an equilibrium for any sub-problem whose
initial condition is a point on the equilibrium path. We briefly consider this
equilibrium in section 4, where we refer to it as "locally time-consistent". The
modifier "local" reminds the reader that only on-the-equilibrium-path behavior is
taken into account.
The consistent equilibrium we are chiefly interested in, however, imposes
stronger requirements; the continuation of this equilibrium represents an
equilibrium for any sub-problem regardless of whether or not its initial condition
is on the equilibrium path. In other words, even if over some interval the
monopsonist had deviated from equilibrium behavior, so that the initial condition
of the remaining problem is off the equilibrium path, the continuation of the
initial equilibrium remains an equilibrium. This requires that equilibrium
behavior depend on the state at which agents find themselves (e.g., the stock
level of the resource); such dependence is of course implied by (although it does
not imply) the Markov assumptionJ
The alternate formulation, discussed in the Introduction, in which the
optimization problem subject to a consistency constraint is replaced by a non-
cooperative game amongst a fictitious sequence of monopsonists, provides a
useful way of distinguishing the two types of dynamically consistent equilibria.
The Markov equilibrium can be obtained by solving the noncooperative game

7 One way to think of the difference between the locally time-consistent and
the Markov equilibrium is as follws: Take a candidate for the equilibrium
trajectory and suppose that in the past the monopsonist has not deviated from
this trajectory (that is, he has previously imposed the tariff indicated by the
candidate trajectory) and it is time for him to choose his current tariff; given the
expectation that in the future he will impose the tariff indicated by the candidate
trajectory, will he also want to adhere to this trajectory in the current period? If
the answer is "yes" then the candidate trajectory is (at least) locally time-
consistent. Now suppose instead that in the past the monopsonist had deviated
from the tariff indicated by the candidate trajectory. This means that the
extraction path will not have been the same as that indicated by the candidate
trajectory, since producers respond to price changes. Therefore in the current
period the remaining stock of the resource is different than the level indicated by
the candidate trajectory. This is what is meant by being "off the equilibrium
path". In the current period will the monopsonist want to use the strategy (i.e.,
the decision rule) indicated by the candidate trajectory? The answer must be
"yes" in a Markov equilibrium.
 283

amongst the monopsonists, where the strategy space of each monopsonist is the
set of functions which take the current state (e.g., the stock of the resource) into
the set of feasible actions (e.g., the current tariff). In this case the monopsonists
use feedback strategies; their decisions are state contingent. In the Nash
equilibrium to this game each monopsonist takes the current state and the
decision rules of all other monopsonists as given and chooses his own action to
maximize his objective. Of course, the decision rules of predecessors are of no
interest to the current monopsonist, given the current state. The locally time-
consistent equilibrium can be obtained by solving the noncooperative game
amongst the monopsonists, where the strategy space of each mcmopsonist consists
of the set of feasible actions, such as the current tariff. (That is, each agent acts
as if the decision of all other agents are not state contingent.) In the Nash
equilibrium to this game each monopsonist takes the current state and the
actions of all other monopsonists as given. Again, the current monopsonist has no
interest in the actions of his predecessors, given the current state.
Of the two, the Markov equilibrium is certainly more plausible. Given the
sequential nature of decisions it is reasonable that the current monopsonist knows
that his decisions affect those of his successors. He should understand how his
own departure from equilibrium would affect his successors' behavior. The locally
time-consistent equilibrium may therefore be of limited intrinsic interest.
However, it provides an instructive contrast to the Markov equilibrium; it helps
to explain why monopsony power does not vanish as the period of commitment
becomes small, and to explain why the opposite conclusion is reached under
different assumptions about what is meant by commitment.

3. THE MARKOV TIME-CONSISTENT EQUILIBRIUM

This section uses a linear example to demonstrate that monopsony power does
not vanish in the limit as f ....O, and to assess the extent to which market power
remains valuable in a Markov equilibrium. The linear example permits a (nearly)
closed form expression for the equilibrium trajectory for finite f and for f=O, and
it is possible to see quite clearly what happens as f becomes small. Since the
equilibrium rules can be explicitly calculated, it is not necessary to invoke further
assumptions about the smoothness of the endogenous functions: the existence
proof is constructive. Therefore the linear case provides a simple illustration.
We assume that extraction equals consumption at all points in time, and
we define extraction as a rate, x. The monopsonist, who is the only consumer of
284

the resource, obtains the flow of utility ax-x 2/2, so that the tariff-ridden
domestic price is pd=a-x. The stock of the resource is S* and the extraction
costs are (a*-bS*)x. Demand and cost are linear in consumption/extraction, and
cost is linear in the remaining stock. We assume that (l<*~ a. This restriction is
essential, since it ensures that the non-negativity constraint on the stock is never
binding; without this restriction there does not exist a linear equilibrium. 8 Given
that this restriction holds, it must be the case that b>O; otherwise it would never
pay to extract, no matter how great the potential supply. If the restriction holds
with strict inequality, it is never economic to extract literally every scrap of
resource. The requirement that average costs be less than the choke price implies
that the initial stock must be greater than (a*-a)/b. The economically
extractable stock is simply S*-{ a*-a)/b. We denote this quantity as S, which we
hereafter refer to as the stock. This means that cost can be written as (a-bS)x,
which simplifies notation.
Although we are interested in the limiting case as the period of commit-
ment, f, approaches 0, it is helpful to begin with the case where f is positive. We
assume that over a period of commitment the tariff, price, and the rate of
extraction are constant, and that there is no discounting within a period; these
assumptions do not affect the limiting case. If at the period beginning at time t
the extraction rate is set to xt ' the seller's total cost for that period is
(a-bS t +bXt f/2)x t f. (This expression is obtained by integrating the cost over
(t,t+f).) If the price paid to the seller is Pt' the buyer's utility is (a-xt /2-Pt)Xt f
over that period.
We can obtain a Markov equilibrium for this model by applying dynamic
programming arguments. We begin with the "last period", at which time the
monopsonist has a single period problem. It is easy to verify that the equilibrium
price in the last period is a linear function of the stock and the extraction rate,
and that the monopsouist's value function is quadratic in the stock. We can then
use an inductive proof to show that this is true at every stage. In a stationary
equilibrium (obtained as the "last period" recedes to infinity) these functions are
independent of the stage index. Straightforward but fairly tedious calculations
verify that the price function is of the form

8 Karp and Newbery (1989b) consider a variation of this model, in which

a*<a and b=O. There the equilibrium is highly noulinear, and cannot be
obtained in closed form.
 285

P(S,X;f) = a -.o( f)[S-Xf] (1)

We suppress the time index on S and x; .0 is independent of time in the stationary
equilibrium. The intercept of the price function must equal the buyer's choke
price. If the intercept of the induced price function were less .than a this would
imply that the buyer had made a commitment to set tariffs in such a way as to
refrain from consuming an amount of stock which is bounded away from O. Such
a commitment cannot be part of a Markov equilibrium. If the intercept of the
induced price function were greater than a, the buyer would be subsidizing
imports over the final part of the trajectory, which could not be part of an
equilibrium. The buyer takes .0 as given, since this is determined by future
behavior, to which he cannot commit himself in the present; .0 is positive from
the second order conditions to the seller's optimization problem; we emphasize
that .0 depends on f.
In equilibrium, x is a linear function of S, so the current price is a linear
function of the current stock in equilibrium. However, the current importer is at
liberty to choose a non-equilibrium tariff, and thus to induce a non-equilibrium
rate of extraction if he wishes. Equation (1) demonstrates that as long as f is
positive the importer will have some influence on the current price, but that as
f...o this influence vanishes. The total payments for a period are pXf=
(a-.BS)Xf+(Xf)2, so the rate of payments is (a-.BS)X+X2 f. As f....O the rate of
payments is proportional to a function of the current stock, which the importer
takes as given. The importer treats the function as exogenous, although of course
it is endogenous to the problem.
In a stationary equilibrium the importer's dynamic programming equation
is
J(S;f) = m~{[a-x/2 -P]Xf + pJ(S-Xf;f)} (2)
where J is the value function; we have used (1) to eliminate the current price and
p=e-u is the 1 period discount factor. This formulation shows the importer
choosing the rate of extraction rather than the tariff; given the price function and
the assumption that sellers are competitive, the two formulations are equivalent.
We remarked above that J is a quadratic function, the parameters of which
depend on f. Taking a first order expansion of both sides of (2), around £=0,
yields the continuous time dynamic programming equation
rJ(S) = m~[.BS - x/2 - JS(S)]x (3)
The function J(S) represents J(S;O) and the parameter .0 is .0(0); the limiting
286

form of equation (1) has been used to simplify the expression. The expression
((3S-x/2)x gives the difference between the rate of consumer utility and the rate
of payments to the seller; the term JSx is the opportunity cost of consuming x, or
the "rent" times consumption. Karp and Newbery (1989b) provide the details of
the derivation of an equation like (3) for general nonlinear demand. The
advantage of the special case considered here is that it admits a closed form
solution. The first order condition to (3) requires (3S-x=J S' The left side of this
equation is the difference between domestic and world price, or the unit tariff;
the right side is the rent. The equation states that the unit tariff is positive if and
only if the rent is positive. Whenever consumption is' positive the consumer
surplus must be positive, and consequently the rent must be positive. Therefore
the monopsonist charges a positive tariff along the equilibrium path.
Since we know that J is quadratic, it has the form 'Yo +'Y1 S+( 'YS2)/2. The
boundary condition requires J(O)=O (if the stock is exhausted, the buyer's future
payoff is 0) which implies that 'Yo=O. It is then easy to show that for (3) to hold,
it must be the case that 'Y1 =0. Therefore, in order to obtain the equilibrium, we
need only two parameters, (3, the slope of the induced price function, and 'Y, the
parameter of the buyer's value function.
Substituting J=( -yS2)/2 into (3) and performing the maximization implies
q = ({3-'Y)2 (4a)
x = ({3-'Y)S (4b)
Equation (4a) gives 'Y as a function of (3; there are two positive roots to this
equation, so we need additional information before proceeding. Equation (4b) is
the control rule; it gives the equilibrium extraction as a function of the stock.
We now consider the competitive seller's optimization problem. His
objective is to maximize the present value of the stream of profits, taking price as
given. This is a standard control problem, and the continuous time equilibrium
requires that price evolve according to
p = r(p-a+bS) (5)
where p=dp/dt. Using (1) to eliminate p (with E=O), and simplifying, gives
p = r(b-(3)S (5')
Differentiating P=a-(3S, using the fact that dS/dt=-x and (4c), implies
p = (3({3-'Y)S (6)
Equating the right sides of (5') and (6) implies
(3({3-'Y) = r(b-(3) (7)
 287

Equations (7) and (4a) give two equations in two unknowns, 'Yand {3. To solve
these, it is convenient to define 1]=~'Yj from (4b), x=1]S, so 1]>0. Equation (7)
implies that {3=rb/(1]+r). Using the definition of 1] in (4a) to eliminate 'Y implies
1]2 +rrr-r{3=O. Using the previous expression to eliminate {3, and mUltiplying both
sides of the resulting equation by (1]+r) gives the cubic
1]3 + 2r1]2 + r21] - r2b = o. (8)
By inspection there is a positive root to this equation, and by Descartes' Rule of
Signs, this root is unique. We have already established that 'Y is positive (from
inspection of (4a)). From the definition of 1], {3 must therefore be positive. This
establishes that the positive root of (8) gives the equilibrium rat'e of extraction in
the formula x=1]S.
Several characteristics of the equilibrium are apparent. We summarize and
comment on these in the following Remarks.
Remark 1 The equilibrium (unit) tariff is an increasing function of the
stock level and approaches 0 as the stock approaches o. Therefore the tariff
monotonically decreases over time. This monotonicity does not hold for more
general models, but holds here because the equilibrium is linear. The present
model makes sense only for S~a/b, so that extraction costs are non-negative.
Karp and Newbery (1989b) show that the monotonicity holds more generally
when the stock is low. However, if extraction costs are constrained to be non-
negative, the tariff is not necessarily a monotonic function of the stock, when the
stock exceeds a critical level. The reason is that as the stock becomes very large
the marginal cost savings due to added stock approaches 0, and the scarcity value
of the stock also approaches o. Therefore the buyer's rent (= tariff) approaches o.
Remark 2 The equilibrium rate of extraction, 1]* (the positive root of (8)),
is an increasing function of b, the marginal cost savings due to an increase in the
stock. This can be shown by totally differentiating (8). An increase in b implies
that extraction is cheaper, so it is not surprising that this increases the rate of
extraction.
Remark 3 The equilibrium rate of extraction is an increasing function of
the interest rate. Again, this is expected, since agents extract and consume more
quickly the more impatient they become. Nevertheless, the result is not a
foregone conclusion, because the buyer is strategic, and r affects the incentives for
both buyer and seller. This result is slightly more complicated to prove than was
Remark 2. Totally differentiating (8) gives d1]*/dr=-H(1]*,r,b)/G(·), where G is
a positive function and H(· )=21]2 +2rrr-2rb. As r.... O, 1]*....0 and for positive r, 1]* is
288

positive, so the derivative must be positive over some range. If it were ever
negative, it would have to be the case that H switches sign at some point; this
implies that for some values of rand b H=O. Suppose, contrary to our assertion,
that this equality holds for some parameter values. This implies that
rr? +r21)=r 2b. Substituting this equality into (8) implies that 1)3 +r1)2=0, which is
a contradiction. Therefore, H never switches sign, and d1)* /dr is positive for all
values of band r.
The simplicity of this model makes it possible to obtain welfare
comparisons in a straightforward manner. It is easy to calculate the buyer's and
seller's welfare in the two cases where the buyer imposes the Markov equilibrium
tariff and under free trade. The ratio of either agent's welfare in the two market
structures depends on the parameters band r, but not on the stock of the
resource S. This is because the value functions are all of the form of a constant
times the square of the stock. In taking the ratio of welfare, only the ratios of the
constants remain. We now derive the remaining formulae necessary for welfare
comparisons, and then present simulation results.
The parameter r, obtained above, is the only information we need
concerning the buyer's value function, in the Markov equilibrium. However, it is
useful to have a more general formulation in order to compare across market
structures. In order to obtain this, we begin with an arbitrary (linear) price
function, p=a-BS, and an arbitrary extraction rule, x=hS. The parameters B
and h should remind the reader of f3 and 1), obtained above. The buyer's welfare is
a quadratic function of the state, (gS2)/2 and solves
(rgS 2)/2 = (BS-x/2-gS)x (9)
This equation is obtained in the same manner as the standard dynamic
programming equation (compare (3) and (9)), although there is no maximization
here, since the extraction rule is taken as given. The parameter g should remind
the reader of r. Substituting x= hS into (9) and simplifying, implies
rg = [(2(B-g) -h]h (9')
As a quick check of the calculations, replace g, B and h in (9') by r, f3 and 1), and
use the definition 1)=~r; the result reproduces (3), as it should; if the arbitrary
parameters Band h just happen to be the "right" ones, the buyer receives his
maximum level of utility. Solving (9') for g implies
g = (2B-h)h/(r+2h) (10)
The seller's level of utility can be obtained in the same manner. His value
 289

function is (g*S2)/2. We omit the details, but it is easy to show that g* is given
by
g* = 2(b-B)h/(r+2h) (11)
We now need to find the equilibrium values of hand B under perfect
competition. We can solve the social planner's problem to obtain h. This is a
standard control problem and yields h as the positive root of
h2 + rh - rb = 0 (12)
Repeating the procedure used to obtain (7), yields
B = rb/(h+r). (13)
This provides the information needed to compare we!£are for the two
different market structures. For the monopsonist in the Markov equilibrium, we
obtain (l and TJ by solving (7) and (8); we identify these values with Band h, and
use (10) and (11) to obtain the parameters of the buyer's and the seller's value
functions. For perfect competition, we obtain Band h by solving (12) and (13),
and then use these values in (10) and (11).
Table 1 provides welfare comparisons under the two market structures for
two levels of the discount rate (r=.03,.06) and three levels of the cost parameter
(b=.5,1,2). Recall that the slope of the buyer's demand parameter was set equal
to 1 by choice of units, so the three values of b correspond to cases where the
sensitivity of cost to stock is less than, equal to or greater than the sensitivity of
domestic price to consumption. Each entry of the table has three elements. The
first element is the ratio of the buyer's welfare under perfect competition to his

Table 1. Welfare Comparisons of Perfect Competition and Monopsony*

.03 .06

.5 (.42 1.31 1.046) (.5 1.35 1.046)

(.37 1.27 1.046) (.44 1.3 1.052)
2 (.31 1.23 1.041 ) (.34 1.27 1.044)
* The first element in each entry is the ratio of the buyer's payoff under perfect
competition to his payoff under monopoly; the second element is the ratio of the
seller's payoff; the third element is the ratio of social welfare (consumer plus
producer surplus).
t b = cost parameter; r = interest rate
290

welfare under the Markov tariff. The second element gives the corresponding
ratio for the seller's welfare, and the third element gives the ratio of social
welfare (buyer + seller welfare). Market power confers substantial benefits on the
buyer; his welfare is two to three times greater when he is able to use a tariff,
even though he is prohibited from making binding commitments, and is restricted
to a Markov equilibrium. The benefits of market power are greater, the larger is
b (the smaller is the cost of extraction) and the smaller is the discount rate.
Facing a monopolist is relatively less costly for the seller when the extraction
costs or the discount rate are lower.
The insensitivity of relative social welfare, to the parameter values, is
striking. For all six cases the loss in social welfare due to imperfect competition
ranges between just over 4 to just over 5 percent. This is a small number, relative
to the changes in the buyer's and the seller's welfare; it suggests that the primary
effect of market power is on distribution, and that the efficiency effect is
secondary. This type of result is standard in welfare economics.
Table 2 compares extraction rates under the two different market
structures, for the six sets of parameter values. For both market structures the
rate of extraction is constant. Rather than reporting the rates, Table 2 gives the
time (in years) it takes to extract half of the available quantity of the resource
(the "half-life" of the resource). The qualitative results are not surprising. A
larger discount rate or lower costs of extraction imply faster extraction, and
extraction is faster under perfect competition than under the monopolist. For the
parameter values reported in the table, the half life is approximately twice as
long under the monopsonist as under perfect competition.

Table 2. Comparison of Half-life of Resource

Under Perfect Competition and Monopsony*

b\rt .03 .06

.5 (6.4 12 .54) (4.7 8.2 .58)

1 (4.4 8.9 .49 (3.2 5.9 .54)
2 (3 6.7 .45) (2.2 4.5 .5)
* The first element in each entry is the half-life (in years) of the resource under
perfect competition; the second element is the half-life under monopsony; the
third element is the ratio of the two half-lives.
t b = cost parameter, r = interest rate.
 291

4. A "LOCALLY" TIME-CONSISTENT EQUILIBRIUM

The previous section studied a Markov equilibrium. In that equilibrium the buyer
and seller know that future decisions will be conditioned upon the value of the
state in future periods. In the limit as the period of commitment shrinks to 0,
supply at any point in time becomes perfectly elastic from the standpoint of the
monopsonist at that point. It is nevertheless optimal for the monopsonist to
impose a tariff at each point in time. This section considers an equilibrium that is
time-consistent but not perfect. The buyer takes the actions (rather than the
decision rules) of his "predecessors and successors" as given and sets a policy to
maximize the present discounted value of utility; he is constrained by the
behavior of competitive profit maximizing sellers with rational expectations.
Hansen et a1. (1985) consider this type of equilibrium in a more general non-
renewable resource mode1. 9 The principal result in this section is that as the
period of commitment shrinks to zero, market power vanishes: the equilibrium
involves zero tariffs.
To demonstrate this we begin with the seller's maximization problem. We
use the linear model of the previous section, but set b=1. The seller's profits
(revenue - cost) in period tare 1ft=[PCa+(St+St+€)/2](SCSt+€). We have
written the cost term, derived in the previous section for the discrete period
problem, to eliminate Xt. The seller takes the price trajectory as given and wants
to maximize the present discounted value of profits. The first order condition for
an interior solution to this problem requires that the extraction at period t be
chosen to satisfy
(l-pF)(a-p t) = (l-p)St+€ (14)
where p is the one period discount factor, e-H , and F is the forward operator,
FPt=Pt+€. The assumption that average extraction costs depend on the stock but
not the rate of extraction makes this a very simple problem, since it means that
the current extraction decision (at an interior equilibrium) depends on price in
the current and subsequent periods, but not directly on price in the more distant
future; of course, the validity of the assumption that the equilibrium is interior
does require consideration of more distant prices.

9 The principal difference between the models is in the form of the cost
function. In our model average extraction costs depend only on the stock of the
resource, whereas in their's it depends on both the stock and the rate of
extraction.
292

We now turn to the buyer's optimization problem. to For the discrete stage
Markov equilibrium it does not matter whether we model buyers as choosing
consumption, world price, or a tariff. The present equilibrium, on the other hand,
is extremely sensitive to the variable that agents take as given. The most
plausible scenario has buyers taking the sequence of future world prices as given.
This means that buyers assume that future tariffs will be adjusted to maintain
world price on an equilibrium trajectory. II In this case it is possible to show, under
the assumption that the equilibrium is differentiable, that free trade is
approached as the period of commitment approaches o. To see this, define
xt£=SCSH£' as before, and use (14) to obtain
£dxt/dPt = (1+p)/(I-p) (15a)

£dxH£/dPt = -1/(I-p) (ISb)

and dxHi/dPt=O for i~2. At t the buyer chooses Pt to maximize
CD •
Lt ;: . E p'( a-xH j2 -PHi£)xHi /. (16)
1=0
The first order condition for the optimal choice of Pt requires
{-(I-p)xt + [(a-xCPt)(1+p) - p(a-Xt+£-Pt+ £)]/£} £(I-p) = 0 (17)
which uses (15). The term £/(I-p) .... I/r as £....0, so for (17) to hold in the liruit it
must be the case that the expression in {} equals 0 for all value of l. The first
term vanishes and under the assumption of differentiability the term in square
brackets approaches
x +:i> + lim(a-xt-pt)/£. (18)
£.... 0
By the assumption of differentiability, the first two terms in (18) are finite. The
third term in (18) remains finite if and only if the difference between the
domestic and world price is at most of the same order of magnitude as £. In the
limit, this requires that the tariff vanish. In that case, a-xt =Pt' so the first two
terms in (18) sum to O. Therefore, it must be the case that the difference between
the domestic and world price be of a smaller order of magnitude than £.

to Recall that we are constructing a (locally) time-consistent equilibrium to

the monopsonist's problem by obtaining the Nash equilibrium to the
noncooperative game amongst the succession of fictitious buyers.
II Commodity price stabilization agreements resemble this arrangement.
However, those agreements are (ostensibly) directed to a different purpose and
operate in a different institutional framework.
 293

This provides another example of a situation where market power vanishes

in the limit as the period of commitment approaches o. The example may not be
robust, since the assumption that extraction costs are independent of the rate of
extraction eliminates much of the inter-period dynamics. 12 Also, as we discussed
above, the equilibrium concept ("local time-consistency") is less plausible than
the Markov equilibrium. However, the example provides a useful contrast.
In the Markov equilibrium we saw that in the limiting case, buyers take the
current world price as given at every point in time. Nevertheless, market power
persists (i.e., it is optimal to use a positive tariff) because current actions affect
the sequence of future prices. The effect of current actions on the level of price at
any instant in the future is infinitesimal, but the cumulative effect on all such
prices is of the same order of magnitude as is the cost of foregoing current
consumption without decreasing current price. In the locally time-consistent
equilibrium obtained in this section, on the other hand, the buyer is able to
choose the current price at every point in time. In the limit, this does him no
good at all, because of his inability to commit to subsequent prices (by
assumption) and because of the seller's ability to arbitrage across time periods.

5. A COMPARISON WITH REPRODUCIBLE GOODS

The previous two sections examined the situation of a (pure) monopsonist who
imports a nonrenewable resource. The principal conclusion was that even as the
period of commitment shrinks to 0, it remains optimal to impose tariffs in a
Markov equilibrium, and market power confers substantial benefits on the
importer. This section considers how matters are changed if investment decisions
rather than depletability provide the source of the forward looking dynamics. We
conclude the section with a discussion of the role of forward and futures markets
in the two situations.
The production of many commodities requires quasi-fixed inputs. 13
Producers' investment in the quasi-fixed input depends on their expectations of

12 To see this, the reader should compare the necessary conditions for the
producer's optimization problem here and in the model studied by Hansen et al
(1985).
13 The stock of a quasi-fixed input is fixed in the short run, but can be
increased by investment over the medium to long run. The adjustment is not
instantaneous because of assumed non-linearities in the cost of investment.
Perennial crops, such as coffee or fruit, provide examples of agricultural goods for
which quasi-fixed inputs are important.
294

future prices, and thus of future tariffs. The monopsonist would like to promise a
low future tariff, in order to encourage investment and thus increase future
supply, and use a high current tariff in order to extract rent generated by the
quasi-fixed input. This program is exactly the reverse of the desited policy when
a nonrenewable resource is imported by a pure monopsonist; recall that in that
case the importer would like to use a low current tariff and threaten to use a high
tariff in the future, to cause the current supply schedule to shift out. For neither
type of commodity is the importer's desired policy credible when his period of
commitment is finite.
Karp (1987) studied the limiting case, as the, period of commitment
approaches 0, of the problem of the monopsonist who imports a reproducible good
(produced using a quasi-fixed input). In the limiting eqnilibrium the monopsonist
can do no better than to behave myopically. That is, at every point he sets a
tariff equal to the inverse of the short run elasticity of supply.14 Producers
anticipate this and accordingly invest less than they would have done under
perfect competition. Since the monopsonist cannot credibly commit to using a
zero tariff at every point in time,15 the free trade equilibrium is not feasible. In the
long run monopsony power is likely to be disadvantageous since the steady state
stock of the quasi-fixed input is small. This means that the steady state supply
curve facing the monopsonist is to the left of the steady state supply curve under
competitive conditions. The short run benefits may not be sufficient to
compensate the monopsonist for the long run costs, if adjustment toward the long
run equilibrium occurs fairly rapidly but not instantaneously. In this case,
market power is disadvantageous.
We are now in a position to discuss how the existence of futures or forward

14 Karp (1987) invokes the Markov assumption, but without calling it by

that name. The limiting results of that paper, described in the text, can be
obtained by applying the methods used in section 3 of this paper. Therefore, it is
valid to compare the effects of monopsony power for the two types of
commodities.
15 In a Markov equilibrium, the seller's investment decision depends only on
the current "state", which in this case is the level of the quasi-fixed input, and
for positive periods of commitment, possibly also on the current tariff. However,
the dependence of current investment on the current tariff is of a smaller order of
magnitude than f, the period of commitment. As f"'O investment decisions
depend only on the level of the quasi-fixed input, so the monopsonist has no
influence on that decision. He therefore can do no better than to maximize the
current rent by imposing the myopic tariff. Karp (1988) provides a formal
 295

markets affects commodity markets where buyers have power and competitive
sellers have rational expectations. The essence of futures (or forward) markets is
that they permit agents to make a commitment about some aspect of their future
behavior. The fact that in the absence of such markets (or a similar institutional
structure) agents cannot make commitments is what diminishes their market
power. Therefore, one might expect that in such circumstances futures markets
would benefit the agent with market power and harm competitive agents.
For reproducible commodities (produced with quasi-fixed inputs), the first
half of this conjecture is correct but the second half is not. The previous
discussion of such markets indicated that sellers as well as buyers are harmed by
the buyer's inability to make binding commitments. This inability makes it
impossible for the buyer to promise to use low future tariffs, but does nothing to
inhibit his use of a high current tariff. Futures markets confer upon the buyer a
greater ability to make an implicit commitment to use a relatively low future
tariff. This encourages current investment because it implies a high price for
sellers in the future. The buyer (a government agency) can make this implicit
commitment by taking a long position in the futures market by an amount
sufficient to drive the futures price up to a desired level. Given that the
government is then committed to taking delivery at a high price, in the future, it
has no incentive (at the future time) to charge a tariff higher than the level
which induces domestic consumers to purchase the contracted quantity. It is, of
course, not necessary for the government agency to take physical possession of
the contracted quantity. Both buyers and sellers benefit from the existence of
futures or forward markets.
The situation is different if the (pure) monopsonistic buyer imports a
nonrenewable resource. In that case, the buyer would like to depress the price
which the seller receives in the future in order to be able to consume at a low
price in the current period. If the government agency takes a long position, it
drives up the futures price. When the contract matures the government would
need to set the domestic price sufficiently low that domestic demand is strong
enough to enable the government to liquidate its contract. This gives the
government an incentive to choose a relatively low tariff at the time when the
contract matures; this is the opposite of what the government would like to
commit itself to, at the initial time. Therefore, if the government is to benefit

statement of this argument in a slightly different context.

296

from the existence of the futures markets its agency needs to take a short
position. If it does this, the agency puts itself in competition with domestic
consumers in the spot market, at the time the contract matures. This gives the
government the incentive to decrease domestic demand by using a high tariff,
which implies a low spot price (received by sellers). This outcome is in line with
what the government would like to commit to at the initial time.
In the two models of pure monopsony, based respectively on a reproducible
good produced with a quasi-fixed input, and on a nonrenewable resource, the
dominant importer is able to use futures markets to alleviate 16 the problems caused
by the inability to make future commitments. However, the government agency
may want to take positions on either side of the market, depending upon whether
it wants to raise or depress spot prices in the future. In both cases the
monopsonist is likely to benefit from the existence of the futures markets; in the
former, sellers also benefit, but in the latter they are made worse off.
There is another situation in which futures markets unambiguously affect
the equilibrium in markets for nonrenewable resources. This occurs where there
are competing buyers, so that monopsony power can be disadvantageous. This
circumstance is discussed, and its empirical importance assessed, in Maskin and
Newbery (1990). Market power can be disadvantageous in this case if the price
trajectory in the Markov equilibrium lies below the trajectory to which the
monopsonist would like to commit; that is, the monopsonist would like to
commit to a lower future tariff than it uses in equilibrium. This resembles the
situation with the reproducible good, described above; as was the case there, the
existence of futures markets enables the monopsonist implicitly to commit to a
low future tariff. It does so by taking a long position in the futures market. Here
the existence of futures markets benefits both the monopsonist and the seller and
tends to decrease the welfare of competing buyers.
Anderson (1989) shows how futures markets enable a monopolist producing
a durable good to overcome the commitment problem. As we discussed in section
2 above, the durable goods monopolist would like to commit to a higher price in
the future than is credible in an equilibrium without commitment. In this
respect, the durable goods monopolist's problem resembles that of the

16 In some cases futures markets serve as a perfect substitute for the ability
to make binding promises; more generally, however, they merely alleviate the
commitment problem.
 297

monopsonist who imports the reproducible good (which uses the quasi-fixed
input). In view of the previous comments, it is no surprise that the durable goods
monopolist takes a long position in the futures market. This alters the incentives
faced by the monopolist when the contract matures. Since he has a long position
he wants the price to be high and therefore restrains production at the time the
contract matures. Consumers are worse off under the higher price. Thus, the
existence of futures markets benefits the agent with market power and harms
agents without power.
As these various special cases indicate, the existence of futures markets is
likely to enhance the market power of an agent who is otherwise unable to make
commitments regarding its behavior in the future. The dominant agent may use
the futures market by taking either a long or a short position, even if the agent is
always a buyer in the spot market. The welfare effects for competitive agents can
be positive or negative; this depends on whether those agents benefit from an
increase in the dominant agent's ability to make commitments. For the cases we
have considered, competitive sellers benefit from the existence of futures markets
when it is optimal for the monopsonistic buyer to take a long position. Of course,
a buyer would be expected to take a long position for hedging, which is the
primary ostensible reason for futures markets. The previous comments suggest
that when both the hedging and the strategic motives for taking a position in the
market operate in the same direction, both buyers and sellers are likely to benefit
from the existence of futures markets.
The existence of futures markets may alter the equilibrium in nonrenewable
resource markets for other reasons. Futures markets are almost always associated
with conditions where there is substantial uncertainty. The discussion here,
however, has maintained the assumption of perfect certainty. This means that
producers solve a non-stochastic control problem, which leads to a fairly simple
intertemporal arbitrage relation. This relation summarizes the producers' forward
looking behavior, which, together with the inability to make commitments, limits
the monopsonist's power. If producers were not forward looking, the monopsonist
would have no problem of commitment. To the extent that futures markets
stabilize the future price trajectory, they tend to encourage forward looking
behavior by producers, and thus indirectly may actually inhibit the
monopsonist's power .17

17 This chain of reasoning seems plausible, but it is no more than that. We

298

6. CONCLUSION
Buyers who behave strategically in markets for nonrenewable resource find their
ability to exercise power limited by the forward looking behavior of sellers and
the difficulty of making commitments about their own future behavior. We have
shown that even if the period of commitment becomes arbitrarily small, it is
likely that a substantial amount of market power remains. This result holds in a
Markov equilibrium, which eliminates the use of punishment strategies and
therefore tends to bias the model against the conclusion that market power
persists in the limit. The ability to exercise market power, even under the
Markov restriction, may increase the payoff to importers by a factor of 2 to 3.
There is a corresponding loss in welfare to resource sellers. The net loss in social
welfare is small, so the distributional effects of market power swamp the
efficiency effects.
We compared the model of the Markov equilibrium in nonrenewable
resource markets with a number of different situations. Section 2 drew out the
relation between the nonrenewable resource problem and models of more general
interest in industrial organization, such as that of the durable goods monopolist.
We discussed how the use of punishment strategies in non-Markov equilibria
might enable a dominant agent to circumvent problems caused by the inability to
make commitments. Such equilibria typically require that competitive agents'
beliefs about about how the dominant agent will behave in the future are very
sensitive to the dominant agent's current behavior.
Section 4 contrasted the Markov equilibrium with a non-Markov but
(locally) time-consistent equilibrium. This material emphasizes the extent to
which the effect of a decrease in the period of commitment depends on
assumptions about the degree of rationality that agents possess.
Section 5 contrasted the Markov equilibria in models with nonrenewable
resources and with reproducible goods. Although the structure of those two
problems is very similar, the characteristics of the equilibrium differ greatly. We
used this contrast as a basis for comparing the effects of financial markets on
equilibria in markets for the two types of commodities. For reproducible goods
markets, the introduction of auxiliary financial markets results in a Pareto

have not shown that futures markets do in fact stabilize prices for nonrenewable
resources; neither have we specified the exact sense in which stability encourages
producers to be more forward looking, or what the direct effect of such stability
would be for the monopsonist.
 299

improvement. The effect is less certain in markets for nonrenewable resources.

Importers with market power are likely to benefit from the creation of these
financial markets, but competitive sellers may benefit or be made worse off. If, in
the Markov equilibrium without financial markets, market power is dis-
advantageous (due to the presence of competing consumers), then the intro-
duction of futures markets benefits both producers and the monopsonist, but
harms competing consumers. If, on the other hand, market power is advantageous
in the Markov equilibrium without financial markets, the introduction of those
markets is likely to harm producers. However, future markets might lessen the
effective power of the monopsonist by strengthening the forward' looking behavior
of the producer.

REFERENCES
Anderson, R W (1989), "Futures Trading for Imperfect Cash Markets", this
volume.
Ausubel, L M and R J Deneckere (1989), "Reputation in Bargaining and Durable
Goods Monopoly", Econometrica, 57, 511-32.
Bond, E and L Samuelson (1984), "Durable Good Monopolies with Rational
Expectations and Replacement Sales", Rand Journal of Economics, 15,
23-32.
Bulow, J (1982), "Durable Goods Monopolists", Journal of Political Economy.
90,314-32.
Chari, V V, P J Kehoe and E C Prescott (1988), "Time Consistency and Policy",
Federal Reserve Bank of Minneapolis Research Dept., Staff Report 115.
Coase, R (1972), "Durability and Monopoly, Journal of Law and Economics, 15,
143-49.
Gul, F, H Sonnenschein and R Wilson (1986), "Foundations of Dynamic
Monopoly and the Coase Conjecture", Journal of Economic Theory,
XXXIX, 155-90.
Hansen, L P, D Epple and W Roberds (1985), "Linear Quadratic Duopoly Models
of Resource Depletion" in T Sargent (ed.), Energy, Foresight and Strategy,
Resources for the Future.
Kahn, C (1986), "The Durable Goods Monopolist and Consistency with
Increasing Costs", Econometrica, LIV, 275-94.
Karp, L (1984), Optimality and Consistency in a Differential Game with Non-
renewable Resources", Journal of Economic Dynamics and Control, 8,
73-97.
Karp, L (1987), "Consistent Tariffs with Dynamic Supply Response", Journal of
International Economics, 23, 69-76.
Karp, L (1988), "Consistent Monopolists and the Benefits of Market Power",
working paper, Dept. of Agricultural and Resource Economics, University
of California, Berkeley.
Karp, Land D M G Newbery (1989a), "Intertemporal Consistency issues in
Depletable Resources, Handbook of Natural Resources, vol III, forth-
coming.
Karp, Land D M G Newbery (1989b), "Optimal Tariffs on Exhaustible
300

Resources", working paper, Dept. of Applied Economics, Cambridge

University.
Kemp, M and N V Long (1980), "Optimal Tariffs and Exhaustible Resources", in
M C Kemp and N'v Long teds.), Exhaustible Resources, Optimality and
Trade, North Holland.
Kydland, F and E Prescott (1977), "Rules Rather than Discretion: the
Inconsistency of Optimal Plans", Journal of Political Economy, 85,
473-9l.
Maskin, E and D M G Newbery (1990), "Disadvantageous Oil Tariffs and
Dynamic Consistency", American Economic Review, forthcoming.
Newbery, D M G (1976), "A Paradox in Tax Theory: Optimal Tariffs on
Exhaustible Resources", SEER technical paper, Dept. of Economics,
Stanford University.
Stokey, N (1981), "Rational Expectations and Dura1;>le Goods Pricing", Bell
Journal of Economics, 12, 125-42.
Advanced Studies in Theoretical and Applied Econometrics

1. Paelinck, J.H.P. (ed.): Qualitative and Quantitative Mathematical Economics. 1982

ISBN 90-247-2623-9
2. Ancot, J.P. (ed.): Analysing the Structure of Econometric Models. 1984
ISBN 90-247-2894-0
3. Hughes Hallet, A.J. (ed.): Applied Decision Analysis and Economic Behaviour.
1984 ISBN 90-247-2968-8
4. Sengupta, J.K.: Information and Efficiency in Economic Decision. 1985
ISBN 90-247-3072-4
5. Artus, P. and Guvenen, O. (eds.), in collaboration with Gagey, F.: International
Macroeconomic Modelling for Policy Decisions. 1986 ISBN 90-247-3201-8
6. Vilares, M.J.: 'Structural Change in Macroeconomic Models. Theory and Estima-
tion.1986 ISBN 90-247-3277-8
7. Carraro, C. and Sartore, D. (eds.): Development of Control Theory far Economic
Analysis. 1987 ISBN 90-247-3345-6
8. Broer, D.P.: Neoclassical Theory and Empirical Models of Aggregate Firm
Behaviour. 1987 ISBN 90-247-3412-6
9. Italianer, A.: Theory and Practice of International Trade Linkage Models. 1986
ISBN 90-247-3407-X
10. Kendrick, D.A.: Feedback. A New Framework for Macroeconomic Policy. 1988
ISBN Hb: 90-247-3593-9; Pb: 90-247-3650-1
11. Sengupta, J.K. and Kadekodi, G.K. (eds.): Econometrics of Planning and Ef-
ficiency.1988 ISBN 90-247-3602-1
12. Griffith, D.A.: Advanced Spatial Statistics. Special Topics in the Exploration of
Quantitative Spatial Data Series. 1988 ISBN 90-247-3627-7
13. Guvenen, O. (ed.): International Commodity Market Models and Policy Analysis.
1988 ISBN 90-247-3768-0
14. Arbia, G.: Spatial Data Configuration in Statistical Analysis of Regional Economic
and Related Problems. 1989 ISBN 0-7923-0284-2
15. Raj, B. (ed.): Advances in Econometrics and Modelling. 1989 ISBN 0-7923-0299-0
16. Aznar Grasa, A.: Econometric Model Selection. A New Approach. 1989
ISBN 0-7923-0321-0
17. Klein, L. R. and Marquez, J. (eds.): Economics in Theory and Practice. An Eclectic
Approach. Essays in Honor of F. G. Adams. 1989 ISBN 0-7923-0410-1
18. Kendrick, D. A.: Models for Analyzing Comparative Advantage. 1990
ISBN 0-7923-0528-0
19. Artus, P. and Barroux, Y. (eds.): Monetary Policy. A Theoretical and Econometric
Approach. 1990 ISBN 0-7923-0626-0
20. Duru, G. and Paelinck, J.H.P. (eds.): Econometrics of Health Care. 1990
ISBN 0-7923-0766-6
21. L. Phlips (ed.): Commodity, Futures and Financial Markets. 1990
ISBN 0-7923-1043-8

Kluwer Academic Publishers - Dordrecht / Boston / London