Professional Documents
Culture Documents
(Advanced Studies in Theoretical and Applied Econometrics 21) Andrew Hughes Hallett, Prathap Ramanujam
(Advanced Studies in Theoretical and Applied Econometrics 21) Andrew Hughes Hallett, Prathap Ramanujam
(Advanced Studies in Theoretical and Applied Econometrics 21) Andrew Hughes Hallett, Prathap Ramanujam
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
ISBN 978-94-010-5482-9
vii
viii
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of recent and unpublished papers on these interactions by leading specialists (and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
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
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
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 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
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)
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
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.
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
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
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)
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:
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.
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.
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
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
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
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.
implies E(y~)<E(y?). At the same time, direct calculation shows that V(y~) is
J J J
unchanged:
(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
Evaluating (8) at p, using the supply and demand functions at (9), and
then substituting again for v and u, yields
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
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
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%.
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
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.
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
(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.
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.
(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
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
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
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
Rubber
Mean Variance 3rd Moment 4th Moment
Price: (US$/MT) 919.16 50002.3 8.106 x 10 6 9.748 x 10 9
Quantity: (I OOOMT)
Malaysia 129.72 307.88 1577.53 3.429 x 10 5
Indonesia 73.26 156.90 2257.18 1.489 x 10 5
Sri Lanka 10.89 24.65 197.67 3948.3
Thailand 43.18 326.92 7072.87 6.412 x 10 5
Copper
Mean Variance 3rd Moment 4th Moment
Price: (US$jMT) 1600.1 1.409 x 10 5 7.514 x 10 7 1.005 x 1011
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
Malaysia Thailand
P -0.114 0.455
CHAPTER 2
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
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).
or
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:
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
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.
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(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.
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 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
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).
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
APPENDICES
March May
8 is the hedge ratio ~/y where y is the mean of exports during the period and J1.f the
hedging effectiveness.
51
For these countries t was fixed at the beginning of April; delivery months are July
and September
March/July July
Country
~ 0 /-If D-W ~ 0 /-If D-W
= 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
+ E(x)f xz (z - E(z))f(x,z)dxdz]f(y)dy
= Cov(x,z)/ y yf(y)dy
= Cov(x,z )E(y)
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
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
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
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)
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
ud
As 'Y1 = [-'Y21-1, and that I'Y19, then /Lfo reaches its minimum at r with
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
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
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
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
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
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
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
"'"~ 90
80
70
60
50
40
30
20
10
0
19731974197519761977 1978 1979 1980 19811982 198319841985 198619871988 1989 1990
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.
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.
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,
the 'instability matters' point of view l3 . In the n period case equation (9) then
becomes,
(12)
13 See Attanasio and Weber (1989) for a discussion and related analysis.
72
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
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
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
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.
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
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.
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:
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.
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
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
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
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.
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
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
! 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
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
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:
(25)
and
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
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
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
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.
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.
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
= - (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
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
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
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
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
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
(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)
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.
R A-c (19)
Hst =~--:2.
asu e as~
where A=fCrtet/i t is the deviation from covered interest parity. As expected,
o
45
o
45
lt 45
o
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
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
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.
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
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
where
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,
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'
In Figure 4, which portrays what happens when volatility of the exchange rate
148
~~----------~~----~.-
---t-=~~Le
q
f 4b. excess supply on commodity market
~~----~~~----~~
-------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
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?
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
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. 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.
t=O t=l
Endowment x*
Consumption 1-tl x*+as
Present value utility U(l-tl) U(x*+as)/(1+r)
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)
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.
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
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
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.
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.
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 ,
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.
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
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
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
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.
(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
(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)
where
ao
b =-
o 1~'
(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.
(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:
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
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
cotton 45.3
40 wheat 41.9
corn 33.2
Absolute 30
Deviation
cents/unit
20
5 10 15 20
speed (days)
Accuracy in terms of total absolute deviations and speed 1981
Absolute
Adjustment Deviotion
50% 75%
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)
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
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
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:
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.
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
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
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
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
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.
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.
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
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.
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
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
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
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:
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.
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,
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
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
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,
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
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,
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,
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.
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
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
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.
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
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.
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
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
(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.
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,
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.
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
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
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
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.
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.
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.
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.
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
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.
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
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.
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
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.
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:
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
Provided that the inverse demand and cost functions are linear.
30
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
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).
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
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
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
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
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
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.
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
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
.03 .06
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.
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)
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
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
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
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
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