Risk Management and The Environment: Agriculture in Perspective

RISK MANAGEMENT
AND THE ENVIRONMENT:

AGRICULTURE IN PERSPECTIVE
Risk Management
and the Environment:
Agriculture in Perspective
Edited by
Bruce A. Babcock
Center for Agricultural and Rural Development (CARD),
Iowa State University, U.S.A.
Robert W. Fraser
Imperial College,
Wye, u.K.
and
Joseph N. Lekakis
Department of Economics,
University of Crete, Greece
"
~.
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-6158-4 ISBN 978-94-017-2915-4 (eBook)

DOI 10.1007/978-94-017-2915-4
Printed an acid-free paper
AII Rights Reserved

© 2003 Springer Science+Business Media Oordrecht
Originally published by Kluwer Academic Publishers in 2003
Softcover reprint of the hardcover 1st edition 2003
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming,
recording or otherwise, without written permission from the Publisher, with the
exception of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
Contents
Tables vii
Figures ix
Contributors x
Preface xii
Introduction
Risk Management and the Environment 1
in Agriculture: A Key Policy Theme
Bruce A. Babcock, Robert W. Fraser, Joseph N. Lekakis
Part I Theory
Chapter
The State-contingent Approach to

Modeling Environmental Risk Management
John Quiggin and Robert G. Chambers 11
2 The Precautionary Principle in Practice:

How to Write a Call Option
on the Environment
Greg Hertzler 29
3 Factors Determining Best Management

Practice Adoption Incentives and the Impact of Green
Insurance
Paul D. Mitchell and David A. Hennessy 52
4 Uncertainty and Adoption

of Sustainable Farming Systems
~~J~M~ ~
5 Incentive Design for Introducing

Genetically Modified Crops
Ross Kingwell 82
~ ConUn~
6 An Economic Risk Assessment of the

Impact on Producers of Removing
Quarantine Restrictions
Robert W. Fraser 96
Part II Case Studies
7 Risk Attitudes and Risk Perceptions

of Crop Producers in Western Australia
Amir K. Abadi Ghadim and David J. Pannell 113
8 Valuing Pest Control:

How Much is Due to Risk Aversion?
Terrance M. Hurley and Bruce A. Babcock 134
9 The Influence of Price Risk

on Set-aside Choice in the EU
Hild Rygnestad and Robert W. Fraser 145
10 Production Risks, Acreage Decisions, and

Implications for Revenue Insurance Programs
JunJie J. Wu and Richard M. Adams 161
11 The Effects of Crop Insurance and

Disaster Relief Programs on Soil Erosion:
The Case of Soybeans and Corn
Barry K. Goodwin and Vincent H. Smith 181
Conclusion
Bruce A. Babcock, Robert W. Fraser, Joseph N. Lekakis 196
Author Index 199

Subject Index 203
Tables
Table 1. 1: Problem Taxonomy 20
2. 1: Parameter Values 37
3. 1: Sequence of Events for Each Production System 58
5. 1: Parameter Values for the Numerical Analysis 89
5.2: Base Case Findings 90
5.3: Sensitivity Analysis of Problem of Illegal Production

of GM Crops 91
6. 1: Base Case Results of the Perception of the

Impact of Introducing Imports on Income Risk 103
6.2: Perception of the Impact of Introducing Imports on

Income Risk: More Elastic Demand 104
6.3: Perception of the Impact of Introducing Imports

On Income Risk: Greater Seasonal Variability 105
6.4: Perception of the Impact of Introducing Imports on

Income Risk: Disease Reduces Expected Yield 106

On Income Risk: Greater World Price Variability 107

On Income Risk: Higher Average World Price 107
7. 1: A Sample of an Elicited and a Contract Price Scheme

Distribution Used in Estimating the Risk Aversion
Coefficient of the Individuals in the Sample 122
7.2: Perceived Mean Yields for Chickpeas and Wheat 124
7.3: The Coefficient of Variation of Chickpea and Wheat

Yields Calculated from Individual Subjective
Probability Distributions. 126
viii Tables
7.4: The Range of Covariance Values Observed among

Farmers in the 1997 Sample 128
7.5: Proportion of Farmers with Different Levels

of Risk Aversion in the 1997 Sample 129
8. 1: Per-Acre Willingness to Pay for 8t Corn 141
9. 1: Prices and Risk, 1998/99 152
9.2: Parameter Values for Three Yield Response

Functions, Denmark 153
9.3: Cost-structure and Other Parameter Values for

Three Land Qualities, no Price Risk, Denmark 153
1998/99
9.4: Standard Leaching Rates for Three Land Qualities,

Denmark 154
9.5: Prices and Risk after Implementation of the Agenda

2000 Reform 154
9.6: Relative Expected Utility (Net Present Value) with

Price Risk, Heterogeneous Land Quality and Different
Levels of Risk Aversion 157
10. 1: Estimated Coefficients for the Logistic Regression

Model of Corn and Soybean Acreage Responses 171
10.2: Estimates of Acreage Elasticities 173
10.3: Changes in Crop Mix under Revenue Insurance 174
11. 1. Variable Definitions and Summary Statistics 186
11. 2: Parameter Estimates and Summary Statistics for

Models of Insurance Purchases 189
11. 3. Three-Stage Least Squares Parameter Estimates and

Summary Statistics for Structural Model of Insurance
Purchases, Fertilizer Use, and Soil Erosion 192
Figures
Figure 2. 1: Prices of a European Call Option on Futures 38
2.2: Values of an Option on Forward Contracts 40
2.3: Prices of a Put Option on Futures 41
2.4: Prices of a Call Option on Resource Stocks 42
2.5: Prices of a Call Option on Financial Stocks 43
2.6: Prices of a Call Option on the Environment 44
2.7: Prices of a Put Option on Futures 45
5. 1: An Illustration of the Farmer's Decision Problem 86
7. 1: A Sample Elicited Yield Distribution Table 119
8. 1: Effects of Bt Corn on Distribution of Corn Yields in

Boone County, Iowa 140
8.2: Effects of Bt Corn on Distribution of Corn Yields in

Hall County, Nebraska 140
9.1: Winsorising a Normal Price Distribution 147
9.2: Ratio of Expected Utility for Different Levels of

Poor land Distribution and Risk Aversion 155
9.3: Ratio of Expected Utility for Different Levels of

Poor land Distribution and Risk Aversion, with the
Agenda 2000 Reform 156
Contributors
Richard M. Adams
Professor, Department of Agricultural and Resource Economics,
Oregon State University, USA
Bruce A. Babcock
Professor of Agricultural Economics
and Director, Center for Agricultural and Rural Development (CARD),
Iowa State University, USA
Robert G. Chambers
Professor, Department of Agricultural and Resource Economics,
University of Maryland, USA
Robert W. Fraser
Professor of Agricultural Economics, Imperial College, Wye, UK,
and Adjunct Professor of Agricultural and Resource Economics,
University of Western Australia, AUSTRALIA.
Amir K. Abadi Ghadim,

Economist, Farm Business Development,
Department of Agriculture, Western Australia, AUSTRALIA
Barry K. Goodwin
Professor, Andersons Endowed Chair of Agricultural Policy,
Trade, and Marketing, Department of Agricultural, Environmental, and
Development Economics, Ohio State University, USA.
David A. Hennessy
Professor, Department of Economics, Iowa State University, USA
Greg Hertzler
Senior Lecturer in Agricultural and Resource Economics
and Associate Dean of Agriculture and Animal Science
Faculty of Natural and Agricultural Science
The University of Western Australia, AUSTRALIA
Terrance Hurley
Assistant Professor, Department of Applied Economics
University of Minnesota, USA
Contributors Xl
Ross Kingwell
Visiting senior lecturer, University of Western Australia
and Senior adviser, Western Australian Department of Agriculture,
AUSTRALIA
Joseph N. Lekakis
Professor of Resource & Environmental Economics,
Department of Economics, University of Crete, GREECE
Paul D. Mitchell
Assistant Professor, Department of Agricultural Economics,
Texas A&M University, USA.
David J. Pannell
Associate Professor and Principal Research Fellow
School of Agricultural and Resource Economics,
University of Western Australia, AUSTRALIA
John Quiggin
Senior Fellow, Australian Research Council
School of Economics, Faculty of Economics and Commerce
Australian National University, AUSTRALIA
HiJd Rygnestad
Managing Director, The Rygnestad Group,
Sevanna Park T-7C, Ithaca, NY 14850, USA
Vincent H. Smith
Professor of Economics, Department of Agricultural Economics
and Economics, and Co-Director, Agricultural Marketing Policy Center,
Montana State University, USA.
JunJie J. Wu
Associated Professor, Department of Agricultural and Resource
Economics, Oregon State University, USA.
Preface
Risk Management and the Environment: Agriculture in Perspective is a

modern academic work that seeks to bring out both to the private and the
policy sectors the importance of risk management in relation to the envi-
ronment in agriculture, as the world moves towards freer markets.
Many efforts were pooled together in making this book. Three years
ago, an attempt was made by one of the editors to get a project on 'Agri-
cultural Risk Management and Sustainabilty' (ARMAS) funded by the
European Commission. Probably deeming the proposal as prematurely
novel for Europe, the Commission's screening experts abandoned its
evaluation.
Following that experience it became apparent that the literature on the
theme ought to be strengthened and emphasized through a book by a
well known publishing house. The editorial team was formed relatively
quickly and an invitation to known experts in the field for contributions
was issued. Subsequently, Kluwer Academic Publishers, evaluated an
edited volume proposal package, and final revisions were made prior to
submitting the entire manuscript for publication.
We are gratefully acknowledging the moral support of several individu-
als as well as the patience of our publishers.
Rethimno, Crete, June 2002

Introduction
Risk Management and the Environment

in Agriculture: A Key Policy Theme
The profile of both risk management and environmental considerations in

agriculture has increased dramatically in the last ten years. Agricultural
policy changes over this period have increasingly exposed farmers to
market risk, while other sources of risk on the production side have also
become more apparent. As a consequence, producers' risk management
skills and the instruments used to manage risk have become a central
feature of the farm management literature and the general agricultural
policy debate. In addition, the awareness in society of the impact of agri-
culture on the environment has much increased, and farmers are more
and more being viewed as "stewards of the countryside" as well as pro-
ducers of food. Hence, the motivation for this book.
The US Model
During the last decade, US farmers have been using various price risk-
management methods including storage, forward cash contracts, selling
futures, buying futures, and private and public insurance programs.
Storage on the farm allows producers to sell their output at different
points throughout the year in an effort to secure an average price. The
risky part of this tool is that farmers may pass up good pricing opportuni-
ties. Forward cash contracts allow farmers to forward price with their lo-
cal elevator anytime during the growing season and deliver later. The
major problem with this method would be failure of the farmer to deliver
due to crop failure.
In addition to these options, the US Department of Agriculture and pri-
vate insurance companies have developed three major programs, In-
come Protection (IP), Crop Revenue Coverage (CRC), and Revenue As-
surance (RA), which are there to assist the farmer's own risk manage-
ment options (Harwood, Goble, and Perry, 1997). All three products offer
an annual revenue guarantee based on producer planted area, expected
2 Bruce A. Babcock, Robert W. Fraser, Joseph N. Lekakis
yield, and sign-up time futures price for harvest-time delivery. If the har-
vest-time price multiplied by the actual yield in that year falls below the
guarantee, the producer receives a compensatory amount of money.
CRC and RA both allow a farmer to replace yield shortfalls at current
market prices. If a farmer's yield falls short and the projected price is
lower that the actual harvest price, he is compensated for the yield loss
at the harvest price in order to allow him to purchase 'replacement' crop
quantities in the market. In addition, RA allows farmers flexibility by al-
lowing them to insure all theirn crops together in a "whole-farm" unit, or
to insure the crops separately.
The EU Experience
Risk management programs at the EU level are non-existent, although

interest in U.S.-style insurance programs seems to be growing. Even
disaster relief payments are usually done at the national level. There
have been though a few exceptional cases of extensive damages where
the European Commission also contributed (ECHO, food aid; and in-
vestment aid, structural funds). Communication with high-ranking officials
in the Commission revealed that data collection of national aids at the
EU level is currently rather weak.
To help stabilize farm revenues, the Commission has shifted focus
from market price support to direct payments for arable crops. The case
of Spain during the recent severe drought is an example of this policy. It
remains, however, the Commission's philosophy that market regulation is
needed to stabilize prices and revenues, and ensure food security. The
new approach developed in Agenda 2000 stresses that intervention
prices should be lowered to a safety net level, in order to avoid perverse
effects of intervention like oversupply, but it does not say that we should
abandon intervention. It has not been demonstrated up to now that risk
management policy would be more efficient than intervention. Neverthe-
less, options and futures markets could be further developed in Europe.
Such development will become more imperative if intervention itself
eventually is abandoned. Developments of options and futures markets
occurred relatively recently in Paris, London, and the Hanover.
Risk Management and the Environment

There are two problems associated with insurance programs, such as
those of the US, that are relevant to both the economic and environ-
mental sustainability of farming in the US and EU. First, participation by
Introduction: Risk Management in Agriculture 3
farmers involves paying premiums that increase production costs, so

small farmers may be excluded ipso facto. Second, the crops cultivated
by large farms that are covered by the programs are usually soil erosive
and chemical intensive, such as cotton and corn. The new revenue pro-
grams cover most major field crops that are produced using soil-intensive
and chemical-intensive methods.
The environmental effects of the insurance programs have recently be-
come a central issue in the literature related to risk management in agri-
culture. A number of researchers (e.g. Quiggin et al., 1993; Smith and
Goodwin, 1996; Babcock and Henessy, 1996) argue that insurance pro-
grams and agrichemicals are substitutes, that is, farmers who purchase
insurance are likely to reduce the application rates of fertilizers and pes-
ticides. If this is the case, the movement away from price supports and
towards state subsidized crop and revenue insurance will be beneficial to
the environment. Environmental activists on the other hand contend that
provision of insurance will actually increase agricultural pollution be-
cause insurance programs will induce farmers to increase output and, in
turn, the use of all production inputs including agrichemicals. The deter-
mination of the environmental effects of risk management becomes even
more complex if one takes into account moral hazard issues. The moral
hazard literature emphasizes the ability of the beneficiaries of an insur-
ance program to influence the likelihood of a loss to occur and, in turn,
the likelihood that the insurance indemnity will be collected. Certain
agrochemicals (especially fertilizers) raise both the expected crop yields
as the variance of yields. The moral hazard literature (e.g. Quiggin, 1991)
suggests that if the increase in variance of yields is large enough to off-
set the increase in the expected yields, farmers may raise the application
rates of the risk increasing inputs (such as fertilizers) in order to rise the
probability that the indemnity will be received. In other words, provision
of state subsidized insurance in the presence of moral hazard may actu-
ally work towards higher pollution levels.
The empirical evidence from studies carried out in the US is mixed.
Quiggin et al. (1993), Smith and Goodwin (1996), and Babcock and
Henessy (1996), all find evidence that yield insurance lowers agrochemi-
cals application rates. Goodwin and Smith (1997) find that federal crop
insurance and disaster relief programs have raised soil erosion in some
crops but decreased in for others. Horowitz and Lichtenberg (1993),
however, find that the purchase of crop insurance has resulted in higher
application rates for both fertilizers and pesticides.
The literature has not yet captured economic sustainability issues

stemming from insurance programs. As mentioned above, the rising op-
erating costs for small farmers due to insurance premiums may lead to
their exclusion. This may be especially true for farmers in southern EU.
Book Contents
The book is structured into two parts: a larger theoretical part containing
a selection of contributions to the general nature of the issues which
arise in agriculture in the context of risk management and the environ-
ment; and a smaller set of case studies which illustrates the main themes
of the theoretical papers in particular situations. We think that this struc-
ture facilitates an appreciation of the generic nature of many of the
problems associated with considering risk management and the envi-
ronment in agriculture, while at the same time showing the application of
theoretical approaches in a suitable range of cases.
More specifically, the theoretical section opens with an examination by
John Quiggin and Robert Chambers of "The state-contingent approach to
modeling environmental risk management". The aim of their chapter is to
describe the state-contingent approach to modeling uncertainty, and to
assess its implications for the analysis of problems involving agricultural
production and environmental risk. In so doing, they argue that there is a
very broad field of potential applications of the state-contingent frame-
work to problems involving environmental risk in agriculture, and that the
state-contingent approach is more amenable to the analysis of uncer-
tainty than other approaches. The second in this pair of chapters exam-
ining environmental risk is Greg Hertzler's study of "The precautionary
principle in practice: how to write a call option on the environment". This
study examines options and irreversibilities in environmental decision-
making. It aims to develop option-pricing formulas and show how much
society will invest in order to avoid undesirable outcomes. The approach
taken is to adapt dynamic investment theory from the finance literature -
an adaptation that requires considerable development as there are many
assumptions made in finance, which do not apply in an environmental
context. The chapter uses examples such as water rights to illustrate its
findings.
The second pair of chapters looks at theoretical issues associated with
the adoption of environmentally sensitive farming practices. Chapter 3,
written by Paul Mitchell and David Hennessy, studies "Factors determin-
ing best management practice adoption incentives and the impact of
green insurance". It develops a framework that allows the factors deter-
mining incentives for farmers to adopt a best management practice

(BMP) using imperfect information to be identified. In so doing, it shows
that these factors include: the cost of the information, the variability and
accuracy of the information, as well as its impact on optimal input use.
They go on to evaluate green insurance for losses due to BMP failure as
a policy instrument to encourage BMP adoption. This evaluation shows
that the factors determining the impact of this policy on adoption incen-
tives include insurance signal variance and correlation with actual
losses, the level of coverage, and farmers' risk aversion. The second
chapter in this pair is David Pannell's study of "Uncertainty and the
adoption of sustainable farming systems". David Pannell argues that un-
certainty has been under-recognized as an impediment to the adoption of
innovative land conservation practices. In particular, uncertainty inhibits
adoption because: farmers are risk averse; uncertainty allows misunder-
standing of the innovation; and in some cases of uncertainty farmers are
better off by waiting to adopt. He develops a framework in which adop-
tion is presented as a process involving the collection, integration and
evaluation of new information, which over time reduces uncertainty. He
goes on to discuss the range of factors that contribute to uncertainty
about conservation innovations and on this basis identifies some clear
implications for policy approaches to land degradation.
The final pair of chapters in the theoretical section of the book contin-
ues in the vein of looking at farmer behavior, albeit in the slightly differ-
ent context of evaluating the impact of environmentally sensitive policy
changes on this behavior in conditions of uncertainty. In Chapter 5, Ross
Kingwell examines "Incentive design for introducing genetically modified
crops". He argues that a policy change allowing the introduction of gen-
erically modified (GM) crops raises several issues. In particular, he looks
at the incentives required to reduce problems of illegal and improper use
of GM propriety technology used in growing GM crops. In so doing, he
develops a model of producer behavior, which captures the key influ-
ences on a farmer's response to GM crops, and illustrates the use of the
model in the context of cotton growing. The key findings of this study are
that the legitimate adoption of a GM crop by a farmer depends on their
attitude to risk, the relative profitability of growing the GM crop, the prob-
ability of detection of illegal or improper use of the GM crop and the se-
verity of the fines successfully imposed for fraud or contract breaches.
Then in Chapter 6, Robert Fraser undertakes "An economic risk assess-
ment of the impact on producers of removing quarantine restrictions". In
this study he develops a model, which can be used to examine the im-
pact of removing quarantine restrictions on the income risk of domestic

agricultural producers. In so doing he identifies four effects of the intro-
duction of imports on the perceived income risk of producers. In addition,
he identifies a range of empirical factors which influence the magnitude
of these effects, and uses a numerical analysis to evaluate the role of
each of these factors in determining the producer's perception of the
level of income risk following the introduction of imports. He argues that
this evaluation provides a set of conclusions relevant to quarantine policy
for agricultural products.
Turning now to the second part of the book, it contains five case stud-
ies that seek to apply theoretical themes developed in the first part to
specific situations. Three of these studies have a US context, while the
other two separately use the EU and Australia as a context. Again the
four studies are arranged in two sets, each of which illustrates a key
theoretical theme drawn from the first part of the book. The first theoreti-
cal theme concerns the adoption by farmers of new farming practices,
and the main factors influencing this adoption process. In this context,
Amir Abadi and David Pannell examine in Chapter 7 "Risk attitudes and
risk perceptions of crop producers in Western Australia". This study
notes both that risk and uncertainty have often been cited as causes of
poor adoption of rural innovations, and that empirical evidence relating to
this view is limited. To address this shortcoming, they present empirical
evidence from a study of the adoption of chickpeas in the wheat belt of
Western Australia. The findings of this study indicate that most of the
farmers surveyed were risk averse, that they considered chickpeas to be
riskier than conventional farm enterprises, and that they perceive a posi-
tive covariance between wheat and chickpea yields. They conclude by
discussing the implications of their findings for the adoption of chickpeas
in Western Australia. Chapter 8 is the second study in this set focusing
on the adoption of new farming practices and is Terry Hurley and Bruce
Babcock's study "Valuing pest control: how much is due to risk aver-
sion?" They show how to empirically model the economic benefits of pest
control and show how to measure the proportion of the benefits that ac-
crue because of risk-averse attitudes. They show that over-reliance on
the standard expected-utility theory to model firm behavior under uncer-
tainty can lead to erroneous conclusions about the nature of risk and the
affects of risk on behavior. They use Bt corn as their case study.
The second theoretical theme addressed by the case studies concerns
the impact of environmentally sensitive policy changes on producer be-
havior in uncertain conditions. Chapter 9 is the first of the three case
studies addressing this theme, and is Hild Rygnestad and Robert

Fraser's study of "The Influence of Price Risk on Set-aside Choice in the
European Union". Taking the EU's set-aside policy as a focus, the aim of
this study is to analyze the effect of price risk and risk aversion on a
farmer's choice of set-aside management in the presence of land hetero-
geneity. In the study emphasis is put on analyzing the choice between
rotational and non-rotational set-aside by taking account of output price
risk and farmer risk aversion, as well as farm level factors such as land
heterogeneity. This choice is explored using a simulation model, which
also features an assessment of the subsequent output and nitrate
reaching levels associated with each choice. The analysis indicates that
land heterogeneity has a smaller impact on set-aside decisions as price
risk or the farmer's risk aversion increases. They suggest the policy sig-
nificance of these findings is substantial given recent changes to CAP
price support, which have resulted in farmers facing much greater price
risk. Chapter 10 is the second of the three case studies looking at the
impact of environmentally sensitive policies on producer behavior under
uncertainty, and is Junjie Wu and Richard Adam's "Production Risks,
Acreage Decisions, and Implications for Revenue Insurance Programs".
Wu and Adams look at revenue insurance programs that are an increas-
ingly popular alternative to direct price supports or federal farm income
support programs in the U.S. Such insurance programs are likely to have
effects on cropping patterns, particularly if coverage is not universal.
These effects on cropping patterns may, in turn, have unintended envi-
ronmental consequences. Wu and Adams confirm that revenue insurance
will alter cropping patterns, although the predicted changes are small
relative to total acreage of each crop in the absence of such programs.
The effects of these acreage changes are likely to involve environmental
consequences, as the counties most prone to acreage shifts are also
those with higher potential for environmental damage. Finally, chapter 11
is Barry Goodwin and Vince Smith's study of "The effects of crop insur-
ance and disaster relief programs on soil erosion: the case of soybeans
and corn". This contribution is an empirical study of the relationship be-
tween participation rates in the U.S. crop insurance program and aggre-
gate data. Goodwin and Smith want to know if the subsidies on U.S. crop
insurance premiums result in greater soil erosion by inducing farmers
alter crop choice and production methods. They find that the participation
rates in the crop insurance program tend to be somewhat negatively cor-
related with the inherent erodibility of soils in a region.
In a broader context, we have seen that in the course of its develop-

ment and presentation, this collection of studies has raised several gen-
eral issues for the future of agri-environmental policy and the role of risk
management in an increasingly uncertain agricultural sector. These gen-
eral issues will be discussed further in the book's conclusion.
References
Babcock, B. A. and D. A. Hennessy (1996) 'Input Demand under Yield and Reve-
nue I nsurance' American Journal of Agricultural Economics, 78: 416-427.
Goodwin, B. K. and V. H. Smith (1997) Crop Insurance, Disaster Relief, and En-
dogenous Soil Erosion' North Carolina State University (unpublished paper).
Harwood, J., D. Heifner, K. Goble, and J. Perry (1997) 'Alternatives for Producer
Risk Management' in USDA, Agricultural Outlook Forum '97 Proceedings,
Washington, DC, February.
Horowitz, J. K. and E. Lichtenberg (1993) 'Insurance, Moral Hazard, and Chemi-
cal Use in Ahriculture, American Journal of Agricultural Economics, 75: 926-
935.
Quiggin, J. (1991) 'Comparative Statics of Rank Dependent Expected Utility The-
ory,' Journal of Risk and Uncertainty, 4: 339-350.
Quiggin, J. G. Karagiannis, and J. Stanton (1993) 'Crop Insurance and Crop
Production: An Empirical Study of Moral Hazard and Adverse Selection,' Aus-
tralian Journal of Agricultural Economics, 37: 95-113.
Smith, V. H. and B. K. Goodwin (1996) 'Crop Insurance, Moral Hazard, and Agri-
cultural Chemical Use' American Journal of Agricultural Economics, 78: 428-
438.
Part I
Theory
1
The State-contingent Approach to
Modeling Environmental Risk Management
John Quiggin and Robert G. Chambers
Risk and uncertainty are crucial features of agricultural production. The

particular concern of agricultural economists with risk may be explained
in part by the dependence of agricultural production on the natural envi-
ronment, taken broadly to include climate as well as soil conditions,
beneficial and harmful animal populations and so on. Conversely, the
activities of agricultural producers have a substantial impact on the natu-
ral environment and this impact is subject to both risk and uncertainty.
Unfortunately, the analysis of production under uncertainty has raised
numerous unresolved difficulties. Whereas modern non-stochastic pro-
duction theory is characterized by its elegance, generality, and clear
axiomatic basis, stochastic production theory hasn't advanced past the
most basic representation of a technology the production function. Ulti-
mately, however, even the primitive stochastic production function ap-
proach proves unworkable in most instances. In its place, one finds a
'parameterized distribution formulation' where decisions, such as the
choice of a scalar level of effort, are represented as the choice of one
member of a family of distributions parameterized by the effort level.
Here, instead of affecting the outcomes of productive activity, changes in
inputs shift probability distributions specified over a fixed set of possible
outcomes. In the process, the recognition of the physical roles, that in-
puts and outputs play in production is lost, and with it the symmetry be-
tween producer and consumer problems.
Chambers and Quiggin (2000) argue that the difficulties of analyzing
production under uncertainty may be overcome through a return to the
approach adopted by Debreu (1952) and Arrow (1953), based on pro-
duction sets in which uncertainty is represented using state-contingent
commodities. This approach has already proved useful in modeling non-
point and point source pollution problems (Chambers and Quiggin 1996;
12 John Quiggin and Robert C. Chambers
Quiggin and Chambers 1998). The object of this chapter is to describe

the state-contingent approach and its implications for the analysis of
problems involving agricultural production and environmental risk.
The chapter is organized as follows: the first section of the chapter is a
description of the modeling framework associated with the state-
contingent framework. The first, and most distinctive step is the specifi-
cation of a state space consisting of exogenously occurring states of
nature relevant to some environmental risk. Given this specification, a
state-contingent production technology is shown to be analogous to a
multi-output non-stochastic technology. The tools of duality theory such
as cost and distance functions may therefore be applied. Moreover, pref-
erences may be represented in a symmetrical fashion. Hence, concepts
of constant absolute and relative risk aversion give rise to corresponding
concepts of constant absolute and relative riskiness in production. In a
development more specifically related to environmental risk, the idea that
some inputs are associated with higher levels of risk is formalized using
the concepts of risk-substituting and risk-complementary inputs.
In the second section of the chapter these concepts are applied to the
modeling of policy issues. First, the problem of private optimization under
uncertainty and the implications of private optimization in the presence of
externalities are discussed. Then the problem of policy design is consid-
ered in a principal-agent framework. A wide range of policy problems is
represented using different assumptions regarding the information set
available to a planner, considered as the principal dealing with an agent
who undertakes production that gives rise to pollution. In the third sec-
tion, the state-contingent production framework is contrasted with alter-
native assumptions about technology and the representation of uncer-
tainty. Finally, some prospects for future development of the approach
are discussed.
1.1. The Modeling Framework

1. 1. 1. The state space
The basis of the state-contingent approach is the claim that environ-
mental problems involving uncertainty are best modeled in terms of an
explicit description of an exogenously given set of possible states of na-
ture. Actions may then be represented, as in Savage (1954), as map-
pings from the state space to some space of possible outcomes.
Ideally, a comprehensive set of states of nature is a mutually exclusive
and exhaustive set of possible descriptions of the state of the world. A
The State-contingent Approach 13
complete description is impossibly complex, so, in any practical analyti-

cal representation, we must abstract from those features, which are ir-
relevant to the problem at hand and include only those features that are
relevant. In a problem involving irrigation-related salinity, for example,
the relevant state variables might be those determining precipitation in a
catchment area. The production decisions of farmers and policy deci-
sions regarding irrigation would determine the river flows and salinity
levels in any given state of nature.
States of nature such as precipitation levels can be graded arbitrarily
finely, so that the choice between discrete and continuous state spaces
is primarily one of analytical convenience. The crucial analogy between
state-contingent production under uncertainty and multi-output produc-
tion under certainty may be developed most simply in the case of a finite,
discrete state space. Hence, we focus on the case of a state space n =
(1 ... S}. Thus, uncertainty is modeled by 'Nature' making a choice from a
finite set of mutually exclusive alternatives.
1.1.2. The technology
This technology transforms vectors of inputs, xEm~, committed by the
producer prior to Nature makinQ a draw from n. into a matrix of private
state-contingent outputs, =Em~xs and a matrix of environmental goods
eEm~xs . The typical element of z, zms corresponds to the amount of
output m, that would be produced if state s occurs. z, therefore, is a ma-
trix of ex ante or potential outputs. Ex post, only one column of Z actually
occurs - the one with the same index as Nature's draw from n. The s
column of z, corresponding to state s, is denoted by =s Em~ . Its domain
is m~ to allow for multiple outputs. Similarly, the typical element of e,
eks, corresponds to the amount of environmental attribute k that would
be produced if state s occurs
The technology is modeled by the input correspondence, which con-
verts a matrix of state-contingent outputs into sets of inputs that can pro-
duce that state-contingent output matrix.
Formally, it is defined by
X(z,e)={xEm~ : x can produce (z,e)Em~+K)XS}.
Intuitively, X (z.e), which we typically refer to, as the input set, is eve-
rything on or above an isoquant for the state-contingent technology.
1.1.3. Preferences
mN
In general, the preferences of producers depend on the !nputs xE,,~ +
that they commit and the state-contingent income y E D1: that they re-
ceive. Thus. the most general representation of preferences is a mapping
V: D1: x D1~ -- D1. Two special cases are of particular interest. The
separable effort specification
V (y,x) = W(y)-g(x)
where g(x) is a disutility of effort function is appropriate where the input

vector includes the producer's labour. The net returns specification
V (y,x) = W(y-w· x)
where w is an input price vector, is appropriate where inputs are pur-

chased in competitive markets. In either separable effort or net returns
specification, we may use a variety of functional forms for W. The most
usual is the expected utility representation,
s
W (y) = ~ Jrsu(ys)
The net returns objective function may also be represented by the cer-
tainty equivalent, that is, the smallest certain income which leaves the
individual as well off as y. More formally,
e (y - w· x) = inf{~ Em: W (c1 S ) ;::: W (y-w· x)}.
1.1.4. Risk premiums

The concept of a risk premium, which has proved valuable in analysis
using the expected-utility model, may be generalized using Luenberger's
(1992) benefit function for the preference structure. The benefit function,
B: D1 x yS -- D1, is defined for g E D1 s by:
B(w,Y) = max{ 13 Em: W(y - 131) ~ w}
if Wry - /31);::: w) for some {3, and - 00 otherwise. Similarly, concepts

of relative risk aversion may be analvzed usina the Shephard (1953)
Malmquist (1953) distance function 0: D1: x D1 -- D1 + defined by:
D (y,w) = Sup{A >OW(yl A) ~ w}
For an individual risk-averse with respect to 7r, we now define the ab-
solute risk premium:
r 7r (y)= max{c:W(( E7r [yj-c) 1) ~ Wry)}
= 8(W(y), E 7r [yj1),
and the relative risk premium:
V 7r (y) =sup{ A. > 0: W(EAy]1! A.) ~ W(y)}
Chambers and Quiggin (2000) use concepts of translation homotheticity

and radial homotheticity to define:
Definition 1: W displays constant absolute risk aversion (CARA) if, for all
y, t, r(y + t1) = r(y).
Definition 2: W displays constant relative risk aversion (CRRA) if. for all
y, t, u(ty) = u(y) .
Extension of these definitions to more general concepts, such as de-
clining absolute risk-aversion, is discussed by Quiggin and Chambers
(2001 ).
1.1.5. Cost functions and duality

A crucial observation, first made by Debreu (1952) and Arrow (1953), is
that the specification of production under uncertainty in terms of state-
contingent outputs is formally identical to the specification of a non-
stochastic technology with multiple inputs and outputs. It follows that the
theory of duality developed for multi-output non-stochastic technologies,
along with the associated tools such as cost and distance functions, is
applicable in full to the case of production under uncertainty. This obser-
vation contradicts a widely held belief that 'duality does not apply under
uncertainty'. As will be argued below, the prevalence of this belief arises
from a misinterpretation of the properties of the stochastic production
functions commonly used in the analysis of production under uncertainty.
Two indirect representations of the state-contingent input correspon-
dence prove useful. The first is the producer's cost function for a given
16 John Quiggin and Robert G. Chambers
vector of state contingent output and environmental goods. It is defined

by
c (z. e) = min {g (x) : x EX (z,e) }
if there exists an x EX(z,e) and 00 otherwise. In the case of linear input

pricing, we may write,
c{w, z, e) = min {w·x: xEX (z,e)}.
Under appropriate conditions. c(w,z,e) is also dual to X(z,e) in the

sense that:
X*(Z, e) = {x:w· x~ c(w,z,e), wEm:+}
= nw>o{x:w·x<!:c(w,z,e)}=X(z).
In other words X(z) can be resurrected from c(w, z) by looking at the

intersection of all the half spaces in input space, m~, containing input
bundles that are at least as costly as the cost-minimizing input choice for
all positive input price vectors. So in McFadden's (1978) terminology,
c(w,z) is a 'sufficient statistic' for X(z) . The second cost function that we
consider is the one that is relevant when the farmer privately chooses x
to minimize the cost associated with producing a given vector of state-
contingent outputs, without regard to the environmental effects. Denote
the farmer's private cost function by,
We define the (cost) certainty-equivalent output, denoted by

eC(z)Em+, as the maximum non-stochastic revenue that can be pro-
duced at cost (C(z), that is,
eC(z)=sup{e:C(el S sC(z)}.
Just as a risk-averse individual will pay a premium in each state to en-

sure the certainty outcome, achieving the certainty-equivalent output may
prove costly. That is, typically it should cost more to remove production
uncertainty and produce the same non-stochastic output in each state
than to allow for stochastic production.
Hence, we may define, as in the analysis of preferences, absolute and

relative production risk premiums and corresponding concepts of con-
stant absolute and relative riskiness.
1.1.6. Risk complements and risk substitutes
Considerable attention has been devoted to the notions of risk-reducing
and risk-increasing inputs. Intuitively, these notions seem clear: risk-
reducing inputs reduce the riskiness of state-contingent outputs, and
risk-increasing inputs increase the riskiness of state-contingent outputs.
Standard examples are pesticides and chemical fertilizers, respectively,
in crop production. Pesticides are not applied to increase yield in the ab-
sence of a pest outbreak so much as they are applied to control losses
when outbreaks occur. Hence, pesticides dampen the dispersion of the
state-contingent outputs, and thus are risk-reducing. Chemical fertilizers,
on the other hand, can significantly increase yield if correct weather con-
ditions prevail, but can also significantly decrease yield when these con-
ditions don't prevail. Hence, they often amplify the dispersion of state-
contingent outputs, and are risk-increasing. Having said this, the litera-
ture on production under uncertainty has struggled with formalizing a
definition of risk-increasing and risk-reducing inputs that matches this
simple intuition and which accords with general notions of increases and
decreases in risk.
The state-contingent approach leads to a rather different perspective.
Rather than thinking of input choices, in combination with random varia-
tion, determining a stochastic output, inputs and state-contingent outputs
are considered to be chosen jointly, in a preference-maximizing fashion.
Hence, it is natural to think in terms of complementary between input
choices and more or loss risky state-contingent output patterns rather
than in terms of simple causal relationships between input choices and
risk. This idea can be formalized in a natural way for the case of scalar
output.
Suppose that M = 1 and that the effort-cost function is differentiable in
input prices so that the cost-minimizing demand for the input s, by
Shephard's lemma, is:
Now suppose, that we were comparing two distinct state-contingent

output arrays, ZO and z', where z' is riskier than ZOo Then, heuristically,
xn(w,=) is a risk complement if xn(w,=') <?: xn(w,zO) and a risk sUbstitute if
XnCw,z') S xnCw,ZO). That is, an input is a risk complement if more of it is

used in producing riskier state-contingent output bundles than in pro-
ducing less risky production arrays. Conversely, an input is a risk sub-
stitute if moving from a less risky situation to a more risky situation is
associated with a decrease in that input's use. Quiggin and Chambers
(2000) show how this concept may be made more precise in different
ways, using different interpretations of the statement 'z' is riskier than
, z ° '. This idea was first put forward in Chambers and Quiggin (1996) where
the terms 'risk-increasing input' and 'risk-reducing input' were used in place
of 'risk-complement' and 'risk-substitute'.
1.2. Policy Modeling

1.2.1. Individual choices
The natural starting pOint for an analysis of policies designed to modify
the environmental impact of producers' decisions is a consideration of
individual production choices in the absence of intervention. Consider a
producer's problem of choosing z to maximize
Wry) = W (pz-C (z))
In this problem, the comparative static behavior of output is the same

as that in the absence of externalities, since the producer disregards
externality effects. The first-order conditions for an optimum are given by
W, (y) p, - cs(w,z) ~ W, (y) sO (1 .1 )

161
where, Zs ~ 0,5 = 1,2, .... ,8 with equality at an interior optimum. Expression

(1.1) means that the marginal cost of producing the 5 state contingent
output is always greater than or equal to the marginal utility of increasing
the 5 state-contingent output divided by the expected marginal utility of
income. Adding the first-order conditions yields an arbitrage condition:
Lemma 1.1: Producer equilibrium must satisfy:
L (cs (w,z)j Ps) ~ 1 (1.2)

.IEQ
with the inequality replaced by an equality in the case of an interior equi-

librium.
Intuitively speaking, the left hand side of (1.2) is the marginal cost of in-
creasing all state-contingent revenues by the same small amount, and
(1.2) simply requires that this cost he at least as large as the uniform in-
crease in returns. If it were not, the decision maker could increase profits in
every state of the world simply by increasing each state-contingent reve-
nue. For an interior solution, (1.2) must hold as an equality.
Chambers and Quiggin (2000) refer to the set of state-contingent out-
put vectors z satisfying the arbitrage condition as the efficient frontier
and prove:
Proposition 1.1 : If prices are normalised to unity and the cost function
exhibits constant absolute riskiness, all elements of the efficient frontier
are equally cost/yo
Since all elements of the efficient frontier arc equally costly, the choice
of an optimal output vector will depend solely on the risk and return
characteristics of the efficient set of output vectors Z. Intuitively, we
would expect that highly risk-averse producers will choose output vectors
with lower mean returns and lower risk than risk-neutral or highly risk-
averse producers. With respect to risk-averse producers, this intuition is
clearly correct, since a risk-neutral producer will choose the output vector
z yielding the highest mean return. Comparisons between more and less
risk-averse individuals are more complex, since the concepts of "more
risk-averse" and 'more risky' may be defined in various ways. Quiggin
and Chambers (2000) consider how these concepts may be made pre-
cise in a manner consistent with the intuition that more risk-averse pro-
ducers will choose output vectors with lower mean returns and lower risk
If risk-averse producers seek lower risk, they will increase their use of
risk-substituting inputs and reduce their use of risk-complementary in-
puts. Hence, if polluting inputs are predominantly risk-substitutes, risk
aversion will exacerbate the severity of the externality problem. Con-
versely, if polluting inputs are predominantly risk-complements, risk
aversion will mitigate the externality problem. From a. policy viewpoint,
the problem of responding to the externality is complicated by the need
to take appropriate account of risk-aversion.
1.3. Principal-agent Theory and Policy Design

A large class of policy design problems may be modeled in terms of the
decisions of a planner who makes payments y to the producer and re-
ceives the output z. Hence z - y may be viewed as a state-contingent net

tax payment, which may be either positive or negative. The planner's
objective function is of the form
v = E [z - yJ + e (y - w· x) - m (e)
that is, the sum of the expected net tax payment by the producer, the
producer's certainty-equivalent income and m(e) , a monetary measure
of the environmental damage e. The planner may be required to satisfy
some reservation utility constraint yielding the producer at least some
minimum level of certainty-equivalent income.
The producer is assumed to have full information regarding the state-
contingent private and environmental output vector (z. e) and to be able
to observe the state of nature, s that is actually realized. By contract, the
planner is assumed to have access only to some subset of this informa-
tion. With varying assumptions about the planner information set, a range
of different policy problems can be formulated. Table 1.1 that follows was
adapted from Chambers and Quiggin (2001) illustrates a range of possi-
ble cases. The possibilities include not only hidden-information problems
but also moral hazard problems (commonly seen as a separate analytical
category involving 'hidden action'). The various cases have been ana-
lyzed by Chambers and Quiggin (1996, 2000, 2001) and Quiggin and
Chambers (1998, 2001).
Table 1. 1: Problem Taxonomy
Observables
State of Private Environmental Class of Problem
nature output output
yes yes yes full information
yes yes no externality with insurance
yes no yes equivalent to full information
yes no no nonpoint pollution
no yes yes point-source pollution with moral hazard
no yes no externality with moral hazard
no no yes externality
no no no private optimum
In all of these problems, the crucial interest lies in the relationship be-
tween the environmental externality, reflected in the exclusion of e from
the producer's private cost function, and the possibility for insurance
arising from interactions between the risk-averse producer and the risk-
neutral planner.
The full-information problem, listed in the first row of Table 1.1, is a
natural benchmark for the analysis. In this case, the producer chooses
the socially optimal output vector (z,e) which maximizes
V =E [z]-c (z,e) -m (e)
and receives a non-stochastic payment y, chosen so that the resulting

net income y - C(=,e) satisfies the reservation utility constraint. Thus, the
externality is fully internalized and the producer is fully insured.
The second problem is one where the planner has sl!fficient informa-
tion to fully insure the producer, but may choose not to do so. If pollution
is a risk-substitute, the planner may prefer the producer to adopt an out-
put less risky than that, which would be chosen under full insurance, with
the result that the private-cost-minimizing level of pollution is also lower.
The third problem illustrates the important point that, if the state of na-
ture is observable, the first-best outcome can be achieved even though
output is not observable. For example, if rainfall is the only relevant state
variable, an appropriately designed rainfall insurance scheme (combined
with Pigouvian taxes to internalize the externality) can achieve the first-
best, just as if there were no uncertainty. Hence, the commonly posed
dichotomy between hidden information and hidden action is misleading .
. Given knowledge of the relevant cost and objective functions, as is nor-
mally assumed, full information about the state of nature is equivalent to
full knowledge of the state-contingent output vector z.
The fourth problem is analyzed by Chambers and Quiggin (1996;
2000), drawing on the observation that different production inputs affect
the riskiness of output (to the farmer) in different ways. For example,
chemical pesticides are usually seen as reducing the riskiness of output
because they preserve output even in the worst states (severe pest in-
festations). Chemical fertilizers, on the other hand, are often seen as in-
creaSing the riskiness of output because they can actually decrease pro-
duction in the event of severe moisture shortfalls. Because chemicals are
the original source of much pollution runoff, the degree of pollution will
be related to the risk characteristics of the ex post output profile. As
noted above, these ideas may be made more precise using the concepts
of risk-complementary and risk-substituting inputs.
If pollution decreases the private marginal cost of the lowest state-

contingent output, an expansion of that lowest state contingent output
will lead to an increase in pollution activity. If it is also a risk-substitute,
that is, if reductions in the riskiness of the output distribution are associ-
ated with higher levels of pollution, then the largest state-contingent out-
puts will have larger probability-adjusted effects on pollution than the
smallest state-contingent outputs. Consequently, pollution must vary
positively with all state-contingent outputs. One can think here in terms
of a risk-substitute which partially generates the pollution runoff, but
which also enhances crop growth in all states of nature. (Pesticides are
some-times claimed to have this characteristic of both enhancing output
and limiting damage.) If pollution varies directly with all state-contingent
outputs, in equilibrium one expects incentives for the farmer to expand
output to be blunted when compared to the full-information problem. The
planner achieves this blunting effect by leaving the risk-averse farmer
with higher ex post returns than in the full-information problem.
In the fifth problem, the environmental externality may be internalized
through Pigouvian taxes, since e is observable to the planner. However,
it is not possible to fully insure the agent because the state of nature is
not observable. Under full insurance, the agent would have no incentive
to produce more than a minimum output level in all states of nature. The
problem is therefore a particular case of the moral hazard problem, ana-
lyzed by Quiggin and Chambers (1998), with an application to point
source pollution. The problem may be separated into two parts, following
the approach of Grossman and Hart (1983). The first step is to determine
the minimum agency-cost of a payment vector y that will elicit any given
(z,e). The second-stage is to choose the optimal (z,e). Quiggin and
Chambers (1998) derived a closed-form solution for the first-stage prob-
lem. After analyzing the properties of the second-stage, they conclude
that the optimal policy will be to focus public abatement activity on re-
ducing pollution in the least favorable states of nature, thereby enhanc-
ing the effectiveness of mechanisms designed to reduce pollution gener-
ated by firms while insuring them against risk.
In the sixth problem, the moral hazard problem is complicated by the
absence of direct information on the environmental output e. This implies
tha.t the producer will always choose e to achieve the minimum cost c(z)
of producing a given z. Hence, the planners problem may be divided, as
in Quiggin and Chambers (1998) into two stages, and the solution to the
agency-cost problem will be the same as that derived by Quiggin and
Chambers (1998). However, the optimal second-stage choice must re-

flect the externality associated with the unobservability of e.
The seventh problem is the obverse of the nonpoint source pollution
cases. If the risk-aversion of the producer is disregarded, the optimal
policy for the planner is simply to impose a Pigovian tax equal, at the
margin, to ml(eJ. However, this policy may be modified to avoid in-
creasing the risk faced by the producer.
In the final problem, the planner has no information and can therefore
take no action, apart from lump-sum transfers. Hence, the outcome is
that of the private optimization problem discussed in the previous sec-
tion.
1. 4. Comparison with Alternative Approaches
1.4. 1. Stochastic production functions

Most studies of production under uncertainty have been based on the
concept of a stochastic production function. Suppose we let x E m~
denote a vector of non-stochastic inputs, which are committed prior to
the resolution of uncertainty, sE{1,2, ...S} a random input, and
f : 9t~ x {1,2, .. .s} -+!n+ the production function. For example, one might
envision an agricultural technology where x represents the inputs that the
producer can control and s represents the randomly occurring rainfall
level. The stochastic production function model is based on the pre-
sumption that random output zE9t+ is determined by
z = f(x,s) ,
a non-stochastic multi-output technology characterized by fixed coeffi-

cients or zero substitute ability between outputs, that is a technology that
behaves as though it were Leontief in outputs. Because the fundamental
role that inputs play remains the same regardless of which s occurs, the
stochastic production function approach does not allow for the possibility
of substituting one state-contingent output for another.
The stochastic production function representation of the problem radi-
cally simplifies the options available to people faced with environmental
hazards. The individual's choice of action x can determine the outcome
for anyone state of nature. However, with the outcome in one state of
nature fixed, the outcome in every other state of nature is uniquely de-
termined by the mapping from the state-space to the outcome space. In
reality, people have available to them a wide range of self-protection ac-
tivities. Some will be beneficial in dealing with low-hazard states of na-

ture but useless in high-hazard states. Others will be most beneficial in
high-hazard states. It is therefore necessary to consider the problem of
how a given self-protection effort will be allocated between high-hazard
and low-hazard states.
The critical insight captured in the general state-contingent model is
that self-protection is not merely a matter of picking a scalar effort level
and waiting for nature to make a random draw. Rather, individuals must
allocate resources across a range of activities with the object of prepar-
ing for a range of adverse or favorable contingencies. Moreover, this ac-
tivity is subject to the usual logic of cost minimization, and can therefore
be analyzed in terms of the standard tools of duality theory such as the
cost function.
As Chambers and Qniggin (1998) observed, even the stochastic pro-
duction function technology can be represented by a cost function of the
form
c(w,z) = in! (wx,' f(x,s) <? Zs Vs}
However this cost function will not, in general, be differentiable or

strictly monotonic. As a result, the solutions to optimization problems
involving a stochastic production function technology will not in general,
be well behaved.
The problems arising from the use of the stochastic production function
technology are most acute in relation to principal-agent problems.
Mirrlees (1974) discovered that if uncertainty is represented by a sto-
chastic production function not exhibiting weak disposability of output
and the number of states of nature is finite, the principal can achieve the
first-best outcome. simply by specifying an arbitrarily large penalty if out-
put falls below f (x' ,s]) , where S1 is the least favorable state of nature,
and offering the fixed payment from the first-best contract otherwise. The
only way the agent can avoid the penalty is to commit at least effort x' ,
and since there is no payoff for any higher effort level, the agent will
choose exactly x' . In short, so long as one relies upon a stochastic pro-
duction function formulation, there is no moral-hazard problem to resolve.
Intuitively, it is easy to see that this arises directly from the use of the
stochastic production function approach without allowing for inefficiency.
Under this approach, output in every state is degenerately determined by
the effort level. Hence, once output in one state is known, output in every
other state is known.
This does not seem like a plausible description of most situations in

which incentive schemes are offered. If the agent is told that he will be
severely punished for falling below some minimum target, but will receive
no reward for performance above the target level, it is natural to suppose
that he will devote all his efforts to meeting the minimum target. In the
general state-contingent production framework, this requires reallocating
resources towards the least favorable state of nature and away from all
of the others.
Thus, the difficulties encountered by Mirrlees should have been re-
solved by the use of a more general state-contingent production technol-
ogy. This approach was not adopted, however. Instead, the problems
arising from the technological specification were papered over through
the use of an alternative representation of uncertainty-the parametrized
distribution framework.
1.4.2. Parametrized distribution framework

Any given choice generates a state-contingent vector of outcomes, which
for decision-makers with well-defined subjective probabilities, may be
described by a random variable. Many researchers have, therefore, cho-
sen to disregard the underlying state space, and analyze problems
purely in terms of choices over random variables. That is, each action is
identified with the cumulative distribution function it generates. Chambers
and Quiggin (2000) refer to this as the parametrized distribution ap-
proach to modeling uncertainty.
The difficulties of comparing the state-contingent and parametrized
distribution are exacerbated by the fact that discussions of the latter ap-
proach frequently use the term 'state' to refer to outcomes. Commonly,
the set of "states' is finite and the set of acts may parametrized by a
positive scalar, usually interpreted as 'effort', with the probability distribu-
tion of the 'states' being determined by a continuous mapping from effort
space (or input space) to the unit simplex whose dimension equals the
number of 'states'. This special case of the parametrized distribution for-
mulation is referred to by Chambers and Quiggin (2000) as the 'outcome-
state' representation.
The relationship between the parametrized distribution formulation and
the state-space representation of the problem is analogous to that be-
tween reduced forms and structural models in econometrics. The state-
space representation, which contains all the relevant information about
the possible states of nature, the input choices of the producers, and the
possible range of outcomes, corresponds to the structural form. The
parametrised distribution formulation, which confounds all these relation-

ships into a simple relationship between possible outcomes and inputs,
corresponds to the reduced form. As with the identification problem in
econometrics, it is always possible, in general, to derive a parametrized
distribution formulation from any state-space representation, but the re-
verse does not apply. In particular, as a general rule, most parametrized
distribution formulations may correspond to infinitely many different
state-space representations. This certainly achieves some economy in
representation and analysis just as reduced-form estimation achieves
some economy in econometric estimation. Unfortunately, the economy
generally is a false one because it is purchased at the cost of confound-
ing causal factors for any economic phenomena that may emerge in such
models. As a result, only limited comparative-static analysis can be un-
dertaken in the parametrized distribution formulation.
Shogren and Crocker (1999) argue that the parametrized distribution
formulation is consistent with an 'endogenous' representation of risk, in
which the capacity of individuals to choose between self-protection ac-
tions is emphasized. However, as Quiggin (2002) observes, the crucial
issue here is the nature of the technology of production under uncer-
tainty. In standard applications of the parametrized distribution formula-
tion, which implicitly incorporate a stochastic production function, the
individual's action is represented by a scalar variable, with high values of
the variable indicating a large amount of resources allocated to a single
self-protection activity. The problem here does not lie in the choice be-
tween representations but in the substantive restriction of the choice set
to a range of scalar effort levels. The solution, therefore, is not to aban-
don the state-contingent approach, but to integrate it with the modern
theory of production.
1.5. Concluding Comments

Work thus far has only scratched the surface of the field of potential ap-
plications of the state-contingent framework to problems involving envi-
ronmental risk in agriculture. Further generalization of the homotheticity
concepts underlying the definitions of constant absolute and relative
riskiness and risk-aversion will enhance the capacity to model individual
production decisions. Consideration of principal-agent problems may be
extend to encompass more richly defined state spaces and information
sets, allowing for multiple sources of environmental risk and for a variety
of monitoring and risk-sharing mechanisms. State-contingent analysis of
financial instruments such as futures and forward contracts (Chambers
and Quiggin 1997) may also be extended to cover instruments dealing

with environmental risks, including rainfall insurance and tradable emis-
sions permits. Despite its greater generality, the state-contingent tech-
nology is more tractable than the special cases previously analyzed.
Moreover, whereas previous analyses of self-protection under uncer-
tainty have been based on the implicit assumption that the range of ac-
tions available to individuals is highly restricted, the state-contingent ap-
proach leads to greater emphasis on the crucial role of individual and
social choice.
References
Arrow. K. (1953), Le role des valeurs boursiers pour la, repartition la meillure des
risques, Cahiers du Seminair d'Economie CNRS, Paris.
Chambers, R. G. and J. Quiggin (1996), 'Nonpoint pollution control as a multi-
task principal-agent problem', Journal of Public Economics 59(1): 95 116.
Chambers, R. G. and J. Quiggin (1997), 'Separation and hedging results with
state-contingent production', Economica 64(254): 187 209.
Chambers, R. G. and J. Quiggin (1998), 'Cost functions and duality for stochastic
technologies' American Journal of Agricultural Economics 80(2): 28895.
Chambers, R. G. and J. Quiggin (2000), Uncertainty, Production. Choice and
Agency: The State-Contingent Approach, Cambridge University Press, New
York.
Chambers, R. G. and J. Quiggin (2001), 'Production externalities and multitask
principal-agent problems', Working Paper, Australian National University. Can-
berra.
Debreu, G. (1952). 'A social equilibrium existence theorem', Proceedings of the
National Academy of Sciences 38: 886 93.
Grossman. S. and O. Hart (1983), 'An analysis of principal-agent problems',
Econometrica 51: 7-46.
Luenberger, D. G. (1992), 'Benefit functions and duality'. Journal of Mathematical
Economics 21 , 461-481.
Malmquist. S. (1953), 'Index numbers and indifference surfaces', Trabajos de
Estotistica 4: 209-242.
McFadden, D. (1978), 'Cost, revenue, and profit functions', in Fuss, M. and
McFadden, D. (eds.), Production Economics: A Dual Approach to Theory and
Applications, North Holland, Amsterdam.
Mirrlees, J. (1974), 'Notes on welfare economics, information and uncertainty', in
Balch. M., McFadden, D. and Wu. S. (ed.), Essays on Economic Behaviour un-
der Uncertainty, North-Holland. Amsterdam, pp. 243-61.
Quiggin, J. (2002), 'Risk and self-protection: a state-contingent view ',Journal of Risk
and Uncertainty, forthcoming.
Quiggin, J. and R. G. Chambers (1998), 'A state-contingent production approach
to principal-agent problems with an application to point-source pollution con-
trol', Journal of Public Economics 70: 441-472.
Quiggin, J. and R.G. Chambers (2000), 'Increasing and decreasing risk aversion
for generalized preferences'. Working Paper, Australian National University.
Canberra .
Quiggin, J. and R.G. Chambers (2001), The firm under uncertainty with general risk-
averse preferences: a state-contingent approach, Journal of Risk and Uncertainty,
22(1): 5-20.
Savage. L. J. (1954), Foundations of statistics, Wiley, New York.
Shephard, R. W. (1953). Cost and Production Functions, Princeton University
Press, Princeton, New Jersey.
Shogren, J. and T. Crocker (1999), 'Risk and its consequences', Journal of Envi-
ronmental Economics and Management, 37(1): 44-51.
2
The Precautionary Principle in Practice:
How to Write a Call Option
on the Environment
Greg Hertzler
Over the years, there have been three major approaches to making deci-
sions under risk. The first, static expected utility theory, is perhaps the
most popular, used by many applied economists. The second is dynamic
investment theory, used by financial economists. The third is difficult to
name. It is the approach we as natural resource and environmental
economists use. It seems to be a mixture of philosophy, common sense,
and confusion. We still haven't sorted out what we mean by the term
"option value." To avoid confusion in this chapter, I will try to be precise
in defining how risk affects decisions. This will, hopefully, lead to more
precise results about how to value our options for preserving the envi-
ronment.
We often talk about the "risk preferences" of decision makers. This
leads to confusion and misinterpretation because, strictly speaking, pref-
erences about risk aren't in our models. We include preferences about
consumption, wealth and time, but not about risk. Risk affects decisions
whenever the world is non linear. One source of non-linearity is the pref-
erence or utility functions in our models. There are many other sources,
however. Behavior under risk can be explained by the non-linearity that
is affected by risk. For the purposes of this discussion, non-linearities will
fall into the following categories:
• Functions: consumption preferences, wealth preferences, time
preferences, production and cost functions, endogenous prices;
• Probability distributions: non-normal distributions, co-variances;
• Asymmetries: infeasibilities, options, irreversibilities.
30 Greg Hertzler
Usually, preferences are singled out as the source of "risk aversion".

However, any non-linear function will alter behavior under risk. Prefer-
ence, production and cost functions are almost always non linear. En-
dogenous prices are multiplied by quantities and introduce non-linearity.
Only normal probability distributions are defined by linear differential
equations. Any other distribution is non linear. In finance, distributions
are often assumed to be log normal. As a consequence, risk exposure
increases with the size of an investment. Co-variances are a source of
non-linearity because they alter means and variances in a way similar to
endogenous prices.
Asymmetries are non-linearities that alter the probabilities of events.
Sometimes events simply can't happen and we use inequality constraints
in our models as an impermeable barrier. All decisions must be on the
feasible side of the barrier. In many cases, other approaches may be
more realistic. Options are more like drawing a line in the sand. Events
can push us to either side of the line and we invest in an option to com-
pensate us if we end up on the on the wrong side. Perhaps most of the
big environmental questions involve irreversibilities. An irreversibility is
like a turnstile at the train station. Being on either side is feasible, but if
you are not careful, events in an unpredictable crowd will push you
through and you can't get back. If going through is undesirable, you must
alter your decisions well before an unpredictable event pushes you
through.
This chapter examines options and irreversibilities in environmental
decision-making. The aim is to develop option pricing formulas and de-
termine how much society will invest to avoid undesirable outcomes. As
examples, stocks of exhaustible resources and water rights are used.
With a bit of imagination, the same results might apply to anyone's favor-
ite resource and environmental problem.
The approach will be to adapt dynamic investment theory from finance.
There are many assumptions made in finance, which do not apply to
natural resources and the environment, including complete markets of
many investors, continuous trading, and no profitable arbitrage. For natu-
ral resources and the environment, there often are no markets. Many
decisions are taken only once by a single agent, let alone continuously
by many investors. Environmental variables probably don't have prices
and profit is rarely the objective of environmental management.
So studying option pricing for the environment is more like a career
than a book chapter. Nevertheless, this chapter will show that many of
the assumptions of finance can be relaxed. However in some cases, the
The Precautionary Principle in Practice 31
formulas for pricing options are different and give different answers. The
examples in this chapter are small ones and there are bigger problems
yet to study. This chapter concludes with suggestions about how this
might be done.
2.1. Stochastic Dynamic Programming

Before presenting the results, there are a few mathematical details to
attend. Option pricing is an application of Ito stochastic calculus. Sto-
chastic calculus is a quick way to calculate expected values and vari-
ances (Hertzler, Harman and Lindner, 1997). Extending the theory of op-
tion pricing to environmental decisions requires stochastic dynamic pro-
gramming (Dixit and Pindyck, 1995; Hertzler, 1991).
In a general formulation of a stochastic dynamic program, decision-
makers in society are assumed to behave as if they maximize their ex-
pected utility subject to a budget constraint for the change in wealth.
J (0, Wo )" maxE 1£ e- P'u (Q )d' + J (T,WTl]

subject to "
dW = d (W,Q,D )dt + s(Q,D )dZ; Wo is given.
A society's satisfaction is summarized by an expected social welfare

function, J. Satisfaction is derived from the utility of consumption, U(Q).
integrated over all years, t, and discounted at the rate of time preference,
p. Satisfaction also includes expected utility of wealth at the end of the
planning horizon, T. Starting from time zero, initial wealth, WO, increases
with changes in wealth, dW. A change in wealth has an expected change
Odt, where 15 is the instantaneous mean, and an error term adZ, where ( j
is a vector of instantaneous standard deviations and dZ is a vector of
Weiner increments. The mean and standard deviations are functions of
wealth, W, consumption, Q, and a set of decision variables, D, chosen at
time t to apply over a decision interval of length dt.
This model assumes that wealth is a continuous-time Markov process
with rapid uncertainty. The Markov property says that, conditional upon
current wealth, future wealth doesn't depend upon the past. It allows the
stochastic change in wealth to be decomposed into its expected change
and its error (Grimmett and Stirzaker, 1992, p. 447). Rapid uncertainty
has many random events occurring within each decision interval. These
random events may be drawn from different probability distributions but
they must have the same mean and variance. According to the central
32 Greg Hertzler
limit theorem, their sum converges to a normally distributed process. In

continuous time, this process is white noise, e, which has mean E{e} = 0
and covariance E{ ee'} = Qldt, where Q is a correlation matrix. Weiner
increments are normally distributed white noise over time, dZ = edt, and
have mean E{dZ} = 0 and covariance E{ dZdZ'} = Qdt. The change in
wealth, dW = &::It + adZ, is a transformation of Weiner increments and
has mean E{dW} = &::It and covariance E{ (dW - &::It)(dW - &::It)'} =
aQatit.
A continuous-time Markov process with rapid uncertainty has a prob-
ability density that is the solution of two partial differential equations
called the forward and backward equations (Grimmett and Stirzaker,
1992, p. 494). These equations depend upon the functions 0 and (J. If the
functions are constant, the forward and backward equations are linear
and integrate to become the normal probability density. Over a short de-
cision interval of length dt, functions 0 and (J are approximately constant
and the change in wealth is normally distributed. Over longer intervals,
however, functions 0 and (J are not constant. The forward and backward
equations are non linear and do not integrate to become the normal den-
sity. Although the probability density of wealth exists, its functional form
may be unknown. This is an advantage, however, because it allows sto-
chastic dynamic programming to model probability distributions in a gen-
eral way, simply by specifying the functions 0 and (J.
Maximizing expected utility over society's time horizon is equivalent to
maximizing an equation, called the Hamilton-Jacobi-Bellman equation or
the Feynman-Kac formula, in each decision interval. This equation is a
partial differential equation in time and wealth, subject to a boundary
condition at time T.
aJ+ max {aJ

-
at
U + --0 +
aW
li--
a2 J (JQa ') -
aW 2
pJ = 0;
(2.1 )
J (T, W T ) = Jv (w T - W ); WT > W
1 0; wT s w.
The boundary condition introduces irreversibility into the model. ~x

pected utility at the terminal time equals ~he utility of wealth, V(Wr - W) ,
if wealth is above a subsistence level, W , and equals zero if wealth is
below the subsistence level. The boundary condition is not a hard con-
straint. Extinction is feasible, just not pleasant.
The expression to maximize in brackets is the current utility of con-

sumption U, plus the marginal utility of wealth, al/aN, multiplied by the
mean, 8, plus a risk premium equal to one-half the derivative of the mar-
ginal utility of wealth, l'2iJ2J / aN 2 , multiplied by the covariance, oQ(T'.
Optimality conditions are the derivatives set equal to zero.
au /0 Q + ~ _ 1/ R 0' (0"0 0"') _ O.
A. O'Q /2 O'Q '
0'0 _ 1/R 0'(0"00"') = o. (2.2)
O'D /2 O'D '
where, A.(W)_~'R(W)=_a2J/aw2.
oW' oj/oW
The first optimality condition is for consumption and the second is for
production and investment decisions. The marginal utility of consumption
is normalized by the expected marginal utility of wealth, A. The terms
containing R are marginal risk premiums. To simplify notation, R is de-
fined as the coefficient of absolute risk aversion. It measures the curva-
ture of expected social welfare with respect to current wealth. It is distin-
guished from an Arrow-Pratt coefficient of risk aversion that would meas-
ure the curvature with respect to terminal wealth. Marginal utility of
wealth and the risk aversion coefficient encapsulate all of the information
about the future. If their current values can be measured, optimal deci-
sions in a single period are also dynamically optimal.
2.2. A Model of Options Including the Environment

The instantaneous mean and covariance required in equations (2.1) and
(2.2) are determined by the stochastic differential equation for wealth.
This differential equation is derived in the Appendix. From Appendix
equation (2.12), the instantaneous mean is:
o(W,Q,D) = 0wW +m (om - Ow) M +n (On - ow)N +e (Oe - Ow) E
-qQ + (Y (1 + Oy) + O"yO"yWyy ) Y - c (Y) + (Y (Oy - Ow - Os) - S) S
+(X - X)K + (il - Y(1 + Oy))H -fofF;
(2.3)
where,
34 Greg Hertzler
On the left hand side, the instantaneous mean is a function of wealth,

W, consumption, 0, and decision variables, D. On the right-hand side,
the first line includes investment in options; the second line includes ex-
penditures on consumption, returns from production and returns from
resource stocks; and the third line includes returns from production con-
tracts, forward contracts and futures contracts.
Upper case letters denote quantities. The expected change in wealth
depends upon wealth itself, W, consumption, 0, and the vector of deci-
sion variables, D. This vector includes investments in options on forward
contracts, M, options on futures, N, and options on the environment, E. It
includes production, Y, and resource stocks, S. It also includes hedging
with production contracts, K, forward contracts, H, and futures contracts,
F. An environmental variable, X, is also included, but it is not a decision
variable and cannot be controlled.
Lower case letters denote prices. Prices include the value of options on
forward contracts, m, the value of options on futures, n, the value of op-
tions on the environment, e, the consumption price, q, the commodity
price, y, and the futures price, f. Prices expectations are modeled as log
normal distributions with percentage changes denoted by 8, standard
deviations by (J" and correlation coefficients by OJ.
In equation (2.3), investments include options on forward contracts,
options on futures and options on the environment. Wealth, W, could be
invested at a risk-free rate of 8W, even if the risk-free rate equals zero.
Options on forward contracts, M, options on futures, N, and options on
the environment, E, are valued at prices m, nand e and attract returns
above the risk free rate of m(8m - 8W)' n(8n - 8W) and e(8e - 8W). The
expected returns on investments, m8m , n8n and e8e , are defined by the
partial differential equations shown in equation (2.3) and are functions of
the commodity price, the futures price and the environmental variable.
The environmental variable could be almost anything about which data is
available and a prediction can be made. Examples include sunspots, the
Southern Oscillation Index, the size of the ozone hole, changes in world
temperature, stream flow in a catchment, rainfall on a farm and kanga-
roos in the back paddock.
Decisions are made about physical quantities, including consumption,
production and resource stocks. Consumption, 0, is purchased at price
q. Production, Y, has expected returns equal to the expected price at the
end of the season, y(1 + 8y ), as modified by the covariance between the
commodity price and production, (J"yo"yOJyY. A negative covariance will
reduce expected returns. Production costs, c(Y). In this simple model,
resource stocks, S, are valued at the commodity price for the economy
with a return on investment above the risk-free rate of Y(O'y - O'W). Stocks
will depreciate at the rate O'S, with a depreciation cost of yO'S. Exploring
for resource stocks and maintaining the stocks will cost 5 dollars per unit.
Some risk management decisions are made once in a year. These in-
clude production contracts and forward contracts. Production contracts,
K, are not for production directly but, rather, for the environmental vari-
able, X. If the environmental variable is less than a contract level, X,
production contracts will be profitable. Production contracts are an artifi-
cial construct that may, but probably won't exist. They are included in the
model as a way to design options on the environment. Forward contracts,
H, are profitable if the expected price at the end of the decision interval,
y(1 + O'y), is less than the contract price, h, at which commodities must
be sold. Commodity futures are traded continuously and there is no fixed
contract price that remains in force throughout the life of the contract.
Trading in commodity futures will be profitable if the change in the fu-
tures price, 0" is negative.
Also from equation (2.12) in the Appendix, the error terms for the
change in wealth are:
O"(D)dZ = y(Y + S - H + (am/aY)M)O" ydz y - J(F -(an/aJ)N)O" fdz f

+yYO"ydZy -(K -(ae/aX)E)XO"xdZx.
The first line of this equation includes price risks and the second line
includes quantity risks. Variances and co variances are found by squar-
ing the errors.
0"(0) Qo'(O) = y2(y +S -H +( iJrn/ay) Mt a; +f2(F -(an/at)N)2 if,

+y¥dy +(K -(ae/ax)Et ~a; (2.4)
-2y(Y +S -H +( iJrn/ay) M) f(F -( an/at) N)O"pfwyf
+2y(Y +S -H +( iJrn/aY)M)YYO"yO"ylUyy -2yY(K -(ae/ax)E)XO"yO"xOJ.,x.
The first line includes variances for price risks; the second line includes
variances for quantity risks; and the third and fourth lines contain co
variances.
2.3. Optimal DeciSions
More specific versions of the optimality conditions in equation (2.2) are

derived using the mean in equation (2.3) and the variances in equation
(2.4). Decisions about consumption and production will not be analyzed
36 Greg Hertzler
here. Optimal decisions about resource stocks, production contracts,

forward contracts, futures contracts, options on forward contracts, op-
tions on futures and options on the environment are determined by:
Resource Stocks
y (Oy - Ow - os) -s -R [y2(y +S -H + (amjay) M)O':
(2.5)
-yf (F - (anjaf)N)O'yO'twyt + y2y O'yO'yWyy ] = o.
Production Contracts
X -x -R[(K -(aejaX)E)X 2 0'; -ylXO'YO'XCVr.¥] =0. (2.6)
Forward Contracts
y (1 + Oy) - h - R [y2 (y + S - H + (amjay) M)O':
(2.7)
-yf(F - (anjat)N)O'YO'twyt + y2yO'py{Uyy] =0.
Futures Contracts
tOf -R [yf(Y +S-H+(am/ay)M)crYO'fWyf -f2(F -(an/at)N)o?] =0. (2.8)
Options on Forward Contracts

m(om - ow) - R [y2(y +S- H+ (amjay)M)cr:
(2.9)
-yf (F -(anj at) N) O'yO'fWyt + yZYO'yO'y{Uyy ] (amj ay) = o.
Options on Futures
n( on -4v) - R [yf(Y +S - H+( ami ay) M) O'yo;Wyf -t2(F - (en/ at)N) o?]( en/ at) =0. (2.10)
Options on the Environment

e( 0e -ow) + R [(K -(aejaX)E)X 2 0'; -ylX O'yO'xCVr.¥ ]( aejax) = o. (2.11)
Although the optimality conditions seem complex, they are easy to in-
terpret. Terms beginning with R are marginal risk premiums. If prefer-
ences are linear, agents are risk neutral in the traditional sense, the mar-
ginal risk premiums are zero and the optimality conditions collapse to
profit maximizing behavior. If agents are averse to risk they will choose
an optimal portfolio that balances profit and risk.
In the finance literature, lack of profitable arbitrage and continuous
trading are assumed. The portfolio becomes riskless and the prices of
options are independent of risk preferences (Jarrow, 1999; Merton, 1990,
p. 281, Hull, 1997, p. 539). For a society making decisions about re-
sources and the environment there is no arbitrage, indeed, there may be

no trading at all, the portfolio is risky and agents may be risk-averse.
Nevertheless, option pricing formulas can be derived from optimal deci-
sions. These formulas will be illustrated for the parameters in Table 2.1.
Table 2.1: Parameter Values
y 160 oW 0.05 O'y 0.20

f 155 Oy 0.10 O'f 0.30
s 10 ot 0.11 oy 0.30
X oS 0.03 O'S 0.15
X 0.90 ax 0.30
h 156
2.3.1. Options on futures

Options can be priced without the usual assumptions about how markets
work. To demonstrate this, consider how a single agent would price op-
tions on futures. If the agent takes out a futures contract there will be a
loss on the contract if the futures price rises. An option on futures lets the
agent avoid losses from a rising futures price, but gain from a falling fu-
tures price. A call option on futures gives the agent the right but not the
obligation to buy a futures contract at an agreed exercise price.
Combining the optimality conditions for futures contracts and options
on futures, equations (2.8) and (2.10), shows that futures and options on
futures are equivalent hedges against price risk.
The marginal risk premiums are eliminated and the expected loss or
gain from holding a futures contract equals the expected capital gains
from owning options on futures. Options incorporate the same informa-
tion as futures prices and hedge against the same risks.
Substituting for the expected change in the price of an option on fu-
tures, n8n , from equation (2.3) gives a partial differential equation defin-
ing how the price of an option evolves beginning from the current time.
This is combined with boundary conditions that an option must satisfy at
the time of maturity.
38 Greg Hertzler
a? 11 iJ 2n 2 2
-+ 12-2 f at -now =0;
a if
n (r, f) -If
- - i; f >• i
0; f sf.
The boundary conditions are for a European call option that can be exer-
cised only at the time of !!laturity, r, but not before. If the futures price, f,
exceeds exercise price, f, the option is "in the money" and will be exer-
cised; otherwise it is worthless. With the boundary conditions, the solution
becomes a cumulative probability, in this case the Black-Scholes formula
for pricing options on futures (Merton, 1990, p. 347; Hull, 1997 p. 277).
Figure 2.1 shows the prices of a call option on futures as a function of
the futures price and time. The exercise price is set to equal the agent's
expected futures price, f(1 + 0,), or $172/ton and the options prices are
the amounts the agent would pay to avoid the upside risk of a higher
than expected futures price. In Table 2.1, the current futures price is
$155/ton. Suppose, at time zero, the option is 52 weeks from maturity. In
this case, the agent would pay $11.55/ton for the option. The price
changes as the option matures. If the futures price at maturity of the op-
tion is less than $172/ton, the option will be worthless. The agent will
throwaway the option and buy futures contracts for less than the exer-
cise price. Otherwise, the option at maturity is valued as the difference
between the futures price and the exercise price. The agent will exercise
the option to buy futures contracts at the exercise price.
Options Price
($/unit)
0 ....
~'" 8 8
Time
(weeks)
"'~ttg]N
'" :!l 0
8 a 8 ~ ol :;;
N fD ~ ....
8 8 ....... ~
v :; ~ 8 . Iii ~ ~ ~ Underlying Price
~ @ ~ .... .... ($/unit)
~
Figure 2. 1: Prices of a European Call Option on Futures.

2.3.2. Options on forward contracts

In much of the literature on hedging, forward contracts and futures con-
tracts are treated as equivalent. However, there is a distinction. Forward
contracts have a fixed contract price and the agent gains or loses de-
pending upon whether the commodity price is below or above the con-
tract price at the time of maturity. Although futures contracts "lock in" a
price, there is no fixed contract price and the agent gains or loses de-
pending upon whether the futures price falls or rises. This distinction may
not be important for choosing between forward contracts or futures con-
tracts, themselves, but options on forward contracts are less valuable
than options on futures because forward contracts are less flexible.
To show this, the optimality conditions for forward contracts and op-
tions on commodity prices in equations (2.7) and (2.9) are combined.
h - y(1 + 0 ) = m(om - ow) .

Y - amlay
The expected return from holding a forward contract equals the ex-
pected capital gains from owning an option on forward contracts. Sub-
stituting for the expected change in the value of an option on forward
contracts, mOm, from equation (2.3) gives a partial differential equation
that is solved subject to boundary conditions.
an an (L" )
a+~V1-y +12 0/ 2 y
11 8 2m 2 2
O"Y -mow =0;
m(r,y)= {Y - y; ~ > y
0; ysy.
Options on forward contracts have an additional term, (am/ay'lii - y), in
the option pricing formula. This term compares the fixed contract price to
the changing commodity price. This differs from options on futures for
which the "contract price" is Simply the current futures price. The
equivalent term would be (ani af 'It -
f) which is always zero.
The boundary conditions are for a European put option. At the time of
maturity, if the exercise price, y, exceeds the commodity price, y, the
option is "in the money" and will be exercised. This option priCing formula
is a non-linear differential equation and no analytical solution is known.
Therefore, the solution was calculated numerically using the Crank-
Nicholson method for finite differences (Hull, 1997, p. 378; Burden and
Faires, 1997, p. 692).
40 Greg Hertzler
Figure 2.2 shows the values of an option on forward contracts as a

function of the commodity price and time. The exercise price is set to
$156/ton, which equals to the contract price for a forward contract in Ta-
ble 2.1. Therefore, an option on forward contracts avoids the downside of
a commodity price that is lower than the contract price but retains the
upside of a commodity price that is higher. The value of an option on
forward contracts is the amount the agent would pay to avoid the down-
side and retain the upside. From Table 2.1, the current commodity price
is $160/ton. For this commodity price a year from maturity, an option on
forward contracts has a value of $7.15/ton. The agent's expected com-
modity price, y (1 + 0y), is $ 176/ton. Even though this is $20/ton above
the exercise price, the agent is still willing to pay to avoid the downside.
Option Price
($/unit)
8 '
;;i ~ 8 8 Time
~ '" ~ . 8 8
~ ... "'. 8 (weeks)
Underlying Price ~ ;! § ~ § ~ ~
($/unit) ~ ~ :e
Figure 2.2: Values of an Option on Forward Contracts.
Petzel (1984) first proposed and Bardsley and Cashin (1990) first ap-
plied the Black-Scholes formula for options on futures to the evaluation
of the benefits from government programs. The Black-Scholes formula or
its equivalent is now used to value many things from crop insurance
(Just, Calvin and Quiggin, 1999; Mahul, 1999; Stokes, Nayda and Eng-
lish, 1997) to old growth forests (Conrad, 1997). Valuing a government
program as if it were an option on futures is equivalent to assuming that

the support guaranteed by the government varies continuously. Valuing
an old growth forest as if it were an option on futures is equivalent to as-
suming that the government continuously changes its mind about how
much forest to save. Figure 2.3 shows the prices of a put option on fu-
tures for an exercise price of $156/ton. At a futures price of $160/ton one
year from maturity, the price of a put option on futures is $1 O.16/ton. This
overvalues the benefits to the agent by about $3/ton. Comparing Figure
2.3 with Figure 2.2 shows that options on futures are priced higher than
options on forward contracts. Because insurance and government guar-
antees are like options on forward contracts rather than options of fu-
tures, they are less valuable than is typically thought.
30.00
25.00~111
Option Price
($/unit) 15,00
10,00
5,00
., .. 0
88 '<o~
t;i.,; ~ 8 co~?a"'" Time
., ....
- '"
...... - w,8.
..
v 00 0
~ ~ C'I (weeks)
Underlying Price - ;! ~ ~ §~ §
o
($/unit) - ~ le ~
a;
..
Figure 2.3: Prices of a Put Option on Futures.
2.3.3. Options on resource stocks
Exploration for resource stocks creates a call option. These stocks can
be extracted and sold if the commodity price is high enough. Originally,
the Black-Scholes formula was derived for options on financial stocks
(Black and Scholes, 1973). This can be compared with an option on re-
source stocks by combining equations (2.5) and (2.9).
42 Greg Hertzler
(s:
Y\U -8w -8s
)
-s = m{8m - 8w ) .
Y am/ay
The return above costs for finding and holding resource stocks equals
the capital gains from an option on forward contracts. Substituting for the
expected change in the value of the option on forward contracts from
equation (2.3) gives the option pricing formula.
an
-+ -
an (y V'w
( ) ) 1/ 0 2m 2 2
+ Os +5 + /2-2- Y u y -mow = 0;
Of 0' 0'
m (r, y ) = {Y - y; y > ~
0; y:s; y.
The Black-Scholes formula for options on stocks also has a term such
as (ami ay )y{8w + 8s ) for the opportunity cost of holding stocks. Resource
stocks must also be found and maintained with an additional term
{amjay)s . The prices of a European call option on resource stocks for an
exercise price of $156/ton are shown in Figure 2.4.
8 Ei ~
8 8. ..::e!!!
8 8. ...: ~ ~ "..
~ , 8 ..: ~ ~ ~ ~ - Underlying Price
~ ~
~ - ;! - ($/unit)
Figure 2.4: Prices of a Call Option on Resource Stocks.
For comparison, the prices of a European call option on financial

stocks are shown in Figure 2.5.
8 8.,.; Ie
Time 8~~
~
(weeks) 8 8.,.;:g ~ -
!i! :; ~ 8 ~ ~ § ~ ~ - Underlying Price
~ :Ii ~ - (S/unit)
:! -
Figure 2.5: Prices of a Call Option on Financial Stocks.
In this case, less flexibility makes call options on resource stocks more
valuable than call options on financial stocks. This can be explained by
the cost of finding and maintaining the stocks. The owner of the stocks
has already invested in exploration and maintenance. If anyone else
wants to own a call option on resource stocks, they must also invest or
purchase the existing stocks at the option price.
2.3.4. Options on the Environment
Options on the environment are not a new idea. Rainfall insurance and
hail insurance are examples (Bardsley, Abbey and Davenport, 1990;
Quiggin, 1994). Weather derivatives are now available for energy com-
panies, concert promoters and farmers to insure against unfavorable
temperatures as well as rainfall. Or, consider a water right. A water right
is a call option on the flow of a stream or river. Combining equations
(2.6) and (2.11) shows that options can be written on almost any envi-
ronmental variable.
The expected return from a production contract equals the expected

capital gains from investing in an option. Whether or not production con-
tracts actually exist does not matter for designing an option pricing for-
mula. Production contracts and options on the environment are equiva-
lent methods of managing environmental risk and so long as one or the
other exists, the risk can be hedged. Substituting for the expected
44 Greg Hertzler
change, eOe, from equation (2.3) gives a partial differential equation for
the price of options on the environment.
oe oe ( -
-+- X -x ) + 72--X
1/ 02e 2 2
ax -eow =0;
01 oX oX2
i - X; i
e (r y ) =
> X
{
, 0; i ~X.
Like options on forward contracts, options on the environment are de-

fined by a non-linear differential equation that must be solved numerically.
However in this case, the boundary conditions are like those used previ-
ously for put options. This is because a call option on the environment may
be exercised at low values of the environmental variable. At the time of
~aturity, if the environmental variable, X, is less than the exercise level,
X , the option on the environment is "in the money" and will be exercised.
Suppose the option is a water right to divert stream-flow from a river
with multiple users. In a wet year, the option is "out of the money" be-
cause stream flow is high. In a dry year, the option can be exercised to
maintain the agent's diversions while others must cut back. In Table 2.1,
the environmental variable is scaled until it has a mean of 1 for an aver-
age year. The contract amount of 0.9 can be interpreted as 90% of aver-
age. The environmental variable is also multiplied by $1 to give units of
dollars. Figure 2.6 gives the prices of rights to divert water in a year with
less than 90% of average stream flow. If, at the beginning of the year, an
0.250000
0.200000
O.loUUU'J~-
Option Price
(S/unit)
0.1 nnnrln -<.-.• -
0.050000
.. 0
N'"
co(t~~"" Time
Environmental Variable
($/unit)
.
., ~ ..
.. !iI
ill !:j N (weeks)
Figure 2.6: Prices of a Call Option on the Environment.

average year is expected, the price of the water right is $0.054. If only
70% of average is expected, the price of a water right is $0.104.
For comparison, Figure 2.7 shows the prices that would be predicted if
water rights were valued as if they were put options on futures. If an av-
erage year is expected, the predicted value for the water right would be
$0.067. However, if only 70% of average is expected, the predicted value
would be $0.216, twice the value predicted in Figure 2.6.
As with all options, prices of water rights are independent of the
agent's risk aversion. In addition prices are independent of the correla-
tion between stream flow and production. However, risk aversion and
correlations affect the quantity of water rights an agent will buy. This
quantity can be found by solving the optimality conditions in equations
(2.5) through (2.11) as a system. Solving the system several times for all
users of the river will give the demands and supplies of water rights.
0.1
Option Price
($/unit)
0.1
... 0
",'"
~ r! co
"'-
ocid~~o ",<1i~- Time
o c:i ~ ;J; co o ~ ~ '" (weeks)
Environmental Variable
($/unit)
0 ci ~ ~
- ......
..,",:;j:'"
Figure 2.7: Prices of a Put Option on Futures.

These could be aggregated into a model of a market for water rights
which is a market for trading in options on the environment.
2.4. Concluding Remarks

Options can be written on any environmental stock or flow that is con-
tinuously measured and has enough historical data to estimate a sto-
chastic differential equation. However, the Black-Scholes option pricing
46 Greg Hertzler
formula does not apply. A new option pricing formula is derived which
considers the provisions of the contract that underlies the option. A sim-
ple example is a forward contract with a fixed contract price over the life
of the contract. This inflexibility lowers the value of options on forward
contracts compared with options on futures. Another example is the
value of a water right. The contract level must be agreed when a water
right is purchased. Because of this inflexibility, a water right is much less
valuable than would be predicted by the Black-Scholes formula or any of
the usual methods of calculating cumulative probabilities.
However, this new option pricing formula is far from a complete model
of options on the environment. For example, exploration for resources
and creating a call option on resource stocks is a complex production
problem and not a simple investment as modeled here. Further, the price
of minerals is an endogenous variable in the system. Perhaps a major
area for future work is options on environmental variables that are en-
dogenous. Society may want to invest in options on old growth forests,
but the value of those same forests depends upon how much forest is
preserved. The benefits of an option that is exercised should be meas-
ured by consumer's willingness to pay, producer's benefits and scarcity
rent. Most of the large environmental questions, such as greenhouse gas
emissions, overpopulation, and available farmland, are in this category.
Pricing these options is not an impossible task. Notice that equation
(2.1) for the formulation of a stochastic dynamic program is, itself, an
option pricing formula, complete with boundary conditions. Expected so-
cial welfare is the value of a call option on wealth as affected by deci-
sions over the planning horizon. As one example, the precautionary prin-
ciple for greenhouse gases can be put into practice by formulating a
suitable model of society's wealth during global warming and solving
equation (2.1). The result will show how much society is willing to invest
to avoid the risks of global warming.
Appendix: Stochastic Wealth
To derive the stochastic differential equation for wealth, begin with soci-
ety's wealth as the sum of all assets and liabilities.
W =bB+aA.
Wealth, W, consists of risk-free bonds, B, with price b, and a vector of

risky assets, A, with prices a. Negative quantities of B and A are liabili-
ties. The change in wealth is found by Ito stochastic differentiation.
dW = dbB + daA + (b + db) dB + (a + da) dA.
The first and second terms on the right hand side are capital gains or
losses on beginning inventories of bonds and risky assets. The third and
fourth terms are acquisitions and depreciation valued at ending prices.
Assume that the quantity of bonds can change over time by acquisitions
and the quantities of risky assets can change by acquisitions and physi-
cal degradation.
dB = Bdt;
dA = (A -A8A)dt -AaAdZA.
Acquisition of bonds is B and acquisition of risky assets is A. Risky
assets are expected to degrade by A8A , with error AaAdZA . Substituting
these changes in bonds and risky assets gives another expression for
the change in wealth.
dW = dbB + daA + (b + db)Bdt + (a + da)((A - 8AA) dt -AaAdZA).
By definition, the acquisition of bonds and risky assets must be fi-

nanced by profits generated by the economy.
1rdt + d 1r = (b + db) Bdt + (a + da) Adt.
The left hand side of this equation is the stochastic profit at the end of
each decision interval. Profit substitutes for acquisitions in the change in
wealth.
dW = db[W -aA]/b +daA -a8AAdt -aAaAdZA +JZCit +d1r.
Bonds have been eliminated by rearranging the equation for wealth to

solve for B and substituting into the change in wealth. The terms
da8AAdt and daA8AdZA have been eliminated as well because dadt
equals zero by the rules of stochastic differentiation and dadZA equals
zero because the covariance between the prices of assets and their
physical degradation is assumed to be zero.
The vector of risky assets can include both natural resource and finan-
cial assets. Assume that risky assets include stocks of exhaustible re-
sources, options on forward contracts, options on futures and options on
48 Greg Hertzler
an environmental variable. Of these, only resource stocks will depreciate.

Assume this depreciation is not stochastic.
dW = db [W - yS - mM - nN - eE ]/ b
+ dyS - yosSdt + dmM + dnN + deE + Jrdt + dJr.
Resource stocks, S, are valued at the price for all commodities in the
economy, y. Options on forward contracts, M, options on futures, N, and
options on the environment, E, have market prices m, nand e.
Society benefits from production, plus any gains from forward and fu-
tures contracts. Subtracting consumption expenditures, costs of produc-
tion and costs of finding and maintaining resource stocks gives profit.
Consumption goods, Q, are purchased at price q. Production, Y, will

sell for price y, and will cost c(Y) to produce. Resource stocks, S, will
cost s per unit to find and maintain. K is a forward contract written on an
environmental variable, X, and will add to profit if the environmental vari-
able is less than a contract level, X. Forward contracts, H, add to profit
if the contract price, h, exceeds the price society would receive for seil-
ing the commodity. Commodity futures contracts, F, add to profit if the
contract prices, i exceeds the futures prices, f.
Profit is stochastic because the commodity price, production, the envi-
ronmental variable and the commodity futures price are stochastic.
d Jr = dyY + (y + dy) dY - dX
Substituting in profit and its stochastic derivative gives another expres-

sion for the change in wealth.
dW = db[W - yS -mM -nN -eE]/b +dyS - yosSdt +dmM +dnN + deE
+ [ -qQ + yY - c (Y) - sS + (X - X) K + (h - y) H] dt
+dyY + (y + dy )dY - dX
For commodity futures, the contract price, f, is simply the futures

price, f, at the time the contract was taken out. Hence, the contract price
and the beginning futures prices cancel from the change in wealth.
Decision-makers in society must form expectations about prices, pro-
duction and the environmental variable. Assume log-normal distributions.
db = bowdt;
dy = YOydt + YCJydzy ;
df = fOfdt + fCJfdz f
dY = YCJydZy ;
dX =XCJxdZx .
Substituting in the differentil equations for expectations and rear-

ranging gives yet another expression for the change in wealth.
dW = Ow [W -mM -nN -eE]dt +dmM +dnN + deE
+[-qQ + (Y(1 + Oy) + CJyCJyWyy ) Y -c(Y) + (Y(Oy - Ow - Os) -8)S
+(x -X)K +(h -y(1+0y ))H-fOtF]dt

+y(Y +S -H)CJydzy -fFCJtdzt + yYCJydZy -JO(CJxdZx '
For society, the commodity price and production will be negatively cor-
related with covariance dydY equal to CJyCJYWyY, where WyY is the cor-
relation coefficient.
Options on forward contracts, options on futures and options on the
environment are assets that must be purchased at prices m, nand e. The
value of an option on forward contracts is a function of the commodity
price, y, and the time to maturity, r - t, where r is a maturity date in the
future. In other words, the value of an option on forward contracts is the
function m(t,y). Its differential equation is found by stochastic differentia-
tion.
Substituting in the differential equation for the commodity price gives

the final result.
Because m is nonlinear, its expected change, in brackets, depends

upon the expected change in the price, YOy, as well as the variance of
the price, y2 CJy2. Its error term is a transformation of the error term for
the price.
Similarly, options on futures are defined by a function of the futures
price, n(t,f), which evolves according to a stochastic differential equation.
50 Greg Hertzler
1/ 8 2n 2 2]
en en + /2-2-f en
dn = [-+-fo,
a a a a, dt+-fa,dz,.
if
Options on the environment are an asset with a value, e(t,X) , which

depends upon the environmental variable and also has a differential
equation.
ie
de= [ -+/2--2 1/ ie 2 2]
8 2e X ax dt+-Xaxdzx·
a ox ox
Finally, substituting in these expectations for options on forward con-
tracts, options on futures and options on the environment gives the sto-
chastic differential equation for wealth.
dW = [O'wW +m(om -ow}M +n(O'n -ow}N +e(O'e -O'w}E

-qQ + (y (1 + O'y) + ayaywyy ) Y - c (Y) + (y (O'y - O'w - O's) - s )8
+(X -X)K +(ii - y(1 +Oy))H -fO',F]dt
+y(Y +8 -H + (am/ay)M)aydzy
(2.12)
-f(F - (an/af) N)a,dz, + yYaydZy -(K - (ae/ax)E) XaxdZx ;
am am a2m 2 2.
where: mO'm =-+-yO'y +>;;-2 yay,
at ay ay
" =an
nUn - +an
- f O'f +/2-2
v a2n f2 a 2.,
f
at af af
ae a2e x2 2
eO'e =-+>;;-2 ax'
at aX
To make the presentation simpler, expected changes in the values of

options on forward contracts, options on futures and options on the envi-
ronment have been abbreviated as mOm, nOn and eOe.
References
Bardsley, P., A. Abbey and S. Davenport (1984), "The Economics of Insuring Crops
against Drought." Australian Journal of Agricultural Economics, 28:1-14.
Bardsley, P., and P. Cashin (1990), "Underwriting Assistance to the Australian Wheat
Industry: An Application of Option Pricing Theory." Australian Journal of Agricultural
Economics, 34:212-222.
Black, F., and M. Scholes (1973), "The Pricing of Options and Corporate Liabilities."
Journal of Political Economy, 81 :637-654.
Burden, R. L., and J. D. Faires (1997), Numerical Methods. Sixth Edition. Pacific
Grove, California: Brooks/Cole Publishing.
Conrad, J. M. (1997), "On the Option Value of Old-Growth Forest." Ecological Eco-
nomics, 22:97-102.
Dixit, A. K, and R. S. Pindyck (1995), Investment Under Uncertainty. Princeton:
Princeton University Press.
Grimmett, G. R., and D. R. Stirzaker (1992), Probability and Random Processes. Sec-
ond Edition. Oxford: Oxford University Press.
Hertzler, G. (1991), "Dynamic Decisions Under Risk: Application of Ito Stochastic
Control in Agriculture." American Journal of Agricultural Economics, 73:1126-1137.
Hertzler, G., J. Harman and R. K Lindner (1997), "Estimating a Stochastic Model of
Population Dynamics with an Application to Kangaroos." Natural Resource Model-
ing, 10:303-343.
Hull, J. C. (1997), Options, Futures and Other Derivatives. Third Edition. Upper Saddle
River, New Jersey: Prentice Hall.
Jarrow, R. A. (1999), "In Honor ofthe Nobel Laureates Robert C. Merton and Myron S.
Scholes: A Partial Differential Equation That Changed the World." Journal of Eco-
nomic Perspectives, 13:229-248.
Just, R. E., L. Calvin and J. Quiggen (1999), "Actuarial and Asymmetric Information
Incentives." American Journal of Agricultural Economics, 81 :834-849.
Mahul, O. (1999), "Optimal Area Yield Crop Insurance." American Journal of Agricul-
tural Economics, 81 :75-82.
Merton, R. C. (1990), Continuous-Time Finance. Cambridge Massachusetts: Black-
well.
Petzel, T. E. (1984), "Alternatives for Managing Agricultural Price Risk: Futures, Op-
tions and Government Programs." AEI Occasional Paper. Washington DC: Ameri-
can Enterprise Institute for Public Policy Research.
Quiggin, J. (1994), 'The Optimal Design of Crop Insurance." Economics of Agricultural
Crop Insurance: Theory and Evidence. D. L. Hueth and W. H. Furtan, Editors, pp
115-134. Boston: Kluwer Academic Publishers.
Stokes, J. R., W. I. Nayda, and B. C. English (1997), "The Pricing of Revenue Assur-
ance." American Journal of Agricultural Economics, 79:439-451.
3
Factors Determining Best Management
Practice Adoption Incentives and the Impact
of Green Insurance
Paul D. Mitchell and David A. Hennessy
Purchased crop inputs such as fertilizer and pesticides contribute to

widespread non-point source pollution problems for surface and ground
water. This pollution can generally be reduced or even eliminated by re-
ducing input use through improved efficiency and by using non-polluting
or less polluting substitutes such as information and insurance. Best
management practices (BMPs) relying on increased information and im-
proved timing already exist, for example split nitrogen applications, soil
nutrient testing, precision agriculture, and integrated pest management.
Analyses often show that these and similar BMPs generate savings by
reduced input use that more than cover BMP costs, yet producers do not
readily adopt such practices.
To encourage adoption of profit-enhancing BMPs such as these, the
Federal Agriculture Improvement and Reform Act of 1996 created the
Environmental Quality Incentives Program (EQIP) to provide financial
incentives to farmers and ranchers with the mandate to maximize envi-
ronmental benefits per dollar expended (U.S. Department of Agriculture,
Natural Resource Conservation Service, 1997a). Specifically, EQIP pro-
vides incentive payments " ... to encourage a producer to perform land
management practices such as nutrient management, manure manage-
ment, integrated pest management, irrigation water management, and
wildlife habitat management" (U.S. Department of Agriculture, Natural
Resource Conservation Service, 1997b, p. 2).
EQIP follows traditional federal legislation by using subsidies to en-
courage voluntary adoption. Such programs can be appropriate for BMPs
such as terraces, riparian buffer strips and conversion of environmentally
sensitive land to non-crop uses, since cost is often the primary impedi-
Best Management Practice Adoption Incentives 53
ment to adoption (Wu and Babcock, 1996). However, for profit-enhancing

BMPs such as those targeted by EQIP, subsidy based incentive provi-
sion may not be appropriate, since in addition to cost, other factors such
as profit uncertainty and risk aversion influence BMP adoption decisions.
Nevertheless EQIP focuses on cost and ignores policy instruments that
take advantage of these other pertinent factors.
One policy instrument currently receiving interest is "green insurance"
for producers adopting an input reducing BMP. Green insurance is a sin-
gle peril insurance that covers losses due to BMP failure. The risks as-
sociated with adoption and use of some BMPs are often cited as reasons
for the observed low adoption rates (Greene et al. 1985; Feather and
Cooper, 1995; Hrubovcak, Vasavada, and Aldy, 1999; Nowak, 1992;
Westra and Olson, 1997). Green insurance directly addresses these risks
to reduce their effect as impediments to BMP adoption. Like other types
of single peril insurance, some types of green insurance potentially can
be privately provided without premium subsidies, and so attain the effi-
ciency of market-based provision of BMP adoption incentives.
Multiple peril crop or revenue insurance fails as green insurance be-
cause it pays indemnities based on observed yield and thus is subject to
moral hazard and adverse selection. BMP failures generally result in
small yield losses that do not trigger crop insurance indemnities, but can
be sufficient to prevent adoption. Raising crop insurance coverage so
that indemnities are paid for BMP failures results in prohibitively high
premiums due to moral hazard and adverse selection. Green insurance
avoids this problem by using an insurance Signal other than crop yield to
separate yield losses from BMP failure from losses due to other factors.
Green insurance products are currently being marketed and others are
under development. IGF Insurance has marketed a corn root-worm inte-
grated pest management (IPM) insurance for producers following a certi-
fied crop consultant's recommendation to not treat with insecticides. If
the IPM recommendation fails, indemnities are paid based on the ob-
served root rating and lodging. Similarly, potato IPM insurance has been
developed for producers adopting IPM for potato late blight. American
Agrisurance sells rainfall insurance for producers using split nitrogen ap-
plications and post-emergent weed control, since excessive rain during
the late spring can delay these activities and cause yield losses. For
producers relying on no-till, American Agrisurance sells soil temperature
insurance, since a cool spring can delay planting of no-till fields and
cause yield reductions.
54 Paul D. Mitchell and David A. Hennessy
Previous studies of the impact of insurance and information on input

use and BMP adoption incentives focus solely on crop insurance and do
not consider insuring the BMP. Miranowski, Ernst and Cummings (1974)
concluded that subsidizing information collection costs is superior to crop
insurance for encouraging adoption of IPM among cotton farmers.
Chambers and Quiggin (1996) use a principal-agent model to derive the
optimal level of net insurance and crop output for producers creating
pollution as a joint product, but do not address technology adoption or
use of information. Babcock and Hennessy (1996) demonstrate that ni-
trogen fertilizer and crop insurance are substitutes for Iowa corn produc-
ers, while Smith and Goodwin (1996) show that chemical inputs (fertil-
izer, herbicides, pestiCides) and crop insurance are substitutes for dry-
land Kansas wheat producers. Hennessy and Babcock (1998) derive
conditions under which information and inputs are substitutes, but do not
address imperfect information, insurance or risk aversion.
The theoretical analysis that follOWS identifies other pertinent factors
besides cost that determine adoption incentives for a BMP that utilizes
imperfect information to increase input use efficiency. In order to evalu-
ate green insurance as an alternative policy instrument for encouraging
BMP adoption, the analysis also identifies factors determining the impact
of green insurance on incentives to adopt this BMP. Unlike other studies,
this analysis of producer incentives to adopt information-based technolo-
gies includes the effects of imperfect information, risk aversion, and in-
surance. Discussion of policy implications focuses on the efficient allo-
cation of government expenditures for provision of BMP adoption incen-
tives.
3.1. Model Description

The representative producer modeled here manages a homogeneous
unit of land normalized to one acre, all devoted to the production of a
single crop. All profit is converted to physical output by normalizing the
crop price to one and all other income and wealth is ignored. Per acre
profit is denoted ;r and the producer derives utility from this profit ac-
cording to a strictly increasing, strictly concave and thrice differentiable
utility function u(;r). The risk averse producer maximizes the expected
utility of profit, with the profit specification depending on the production
system used. However, for all production systems, the same crop growth
function f(x, (), 6") transforms a purchased input x and stochastic inputs ()
and 6" into per acre crop yield, where fx > 0, fxx < o. The producer
chooses one of three production systems - the status quo system, the
BMP, or the BMP with green insurance.
The input B can be a naturally occurring substitute for x so that to > 0
and txo < 0, as when x is nitrogen fertilizer and B is nitrogen from natural
sources. Alternatively, B can measure the level of a naturally occurring
"bad" so that to < 0 and tXiJ > 0, as when x is a pesticide and B measures
the pest population. Because the natural processes and factors deter-
mining B are many and complex, the producer treats B as a stochastic
input with a distribution depending on available information.
Random production shocks independent of B are aggregated into the
random variable E: with density h(E:). Assume many local factors that are
only observed with difficulty determine E: so that it is non-contractible.
Realized yield could be used to determine E: ex post, but because yield is
subject to manipulation by the producer, accurate determination requires
prohibitively high monitoring costs and is not feasible.
3. 1. 1. Status quo
For the status quo, B follows the unconditional density v(B). The pro-
ducer's objective and first order condition are:
Maxf,.foU(Jr)v (B)h (E:) dB dE: (3.1 a)
fefou'(Jr)(fx(x,B,E:) -r)v(B)h(E:) dB dE: = 0 (3.1 b)
where, Jr = t (x,B,E:) - rx and r is the per unit price for x when normalized
by the output price. Single and double primes indicate first and second
derivatives respectively and subscripts indicate partial derivatives with
respect to the subscripted variable(s). Use x
> 0 to denote the optimal
value of x implicitly defined by the first order condition and use EUSQ to
denote the value of the objective function when evaluated at this opti-
mum. Since txx < 0 and u"< 0, the second order condition is satisfied
and EUSQ is a maximum.
3.1.2.8MP
For the BMP, the producer first collects imperfect information concerning
B, the-n chooses x. Denote this ex ante signal of the true level of B as a
and the resulting conditional density of B as q(B I a). The producer
chooses the optimal decision rule x * (a) to maximize expected utility
given a. The objective and first condition are:
Maxf,}oU(Jr)q(B I a)h(E:) dB dE: (3.2a)
(3.2b)
where 7T = '(x, f), c:) - rx. For any a the first order condition implicitly de-
fines an x * (a) > 0. Expected utility for this system is the expected value
of the objective at x*(a): EU SMP =
faJ..foU(7T(X * (a)))q(f) I a)h(c:)g(a) df) dc: da , where g(a) is the density of a.
Since 'xx < ° and u" < 0, the second order condition is satisfied and
EUSMP is a maximum.
Statistically the relationship between a and f) is described by the jOint
probability density b(f),a) , such that a and f) have a positive covariance.
Note that b(f),a) can be factored into q(f) I a) and g(a), the conditional
density of f) and the unconditional density of a respectively. The uncon-
ditional density v(O) observed when using the status quo production
system is obtained by integrating b(f),a) over a.
The information a provides is imperfect for reasons specific to each
BMP, but generic examples include measurement error inherent in the
information technology, a time lag occurring between the observation of
a and the realization of 0, or spatial variation in the realization of f). Sig-
nals of this sort include the late spring soil nitrogen test, which imper-
fectly estimates soil nitrogen available later that summer, and scouting
for adult insects as part of corn rootworm IPM, which imperfectly esti-
mates the corn rootworm larval population the following spring.
3.1.3. BMP with green insurance
Green insurance covers losses due to BMP failure. Two types of BMP
failure can occur - either too little or too much of the input x is applied,
resulting in profit loss due to reduced yields or wasted expenditures. The
imperfect nature of a can mislead the producer into applying too little or
too much x, or waiting to collect and process a can result in factors such
as weather preventing or delaying application of x. Also errors can occur
because the producer does not completely or correctly understand the
BMP, or the BMP as recommended is not quite correct for the producer's
land or operation.
Producers have been primarily concerned with input deficiencies and
associated yield losses. Input surpluses are usually less costly to pro-
ducers, since yield loss is minor (if it occurs at all) and typically difficult
to detect. Agronomic research has also focused primarily on detecting
input deficiencies and measuring associated yield losses to develop rec-
ommended application rates. BMP failures that cause input deficiencies
are a concern to policy makers as well, since such failures are often easy
to observe and create a bad reputation for the BMP and so reduce adop-
tion. Thus in the analysis here, BMP failures that result in over applica-
tion of the input x are ignored and the focus is on failures that result in an
input deficiency.
Indemnities are based on a signal 5 from which expected yield loss due
to input deficiency as a result of BMP failure can be inferred. As a result
of measurement error, spatial variation, time lags, and other such fac-
tors, the signal 5 is an imperfect signal of input deficiency and depends
on x and 0 since they determine this deficiency. Denote its conditional
density as W(5 I x,O) . Define 5 such that high values of 5 indicate a small
input deficiency and vice versa, so that 5 is positively correlated with O.
This specification does not explicitly include a, but because x depends
on the observed a, the specification depends indirectly on a and as-
sumes that all dependence of 5 on a is mediated by x. Signals of this sort
already exist for important BMPs such as the late spring soil nitrogen test
or corn rootworm IPM. The end of season stalk test and the SPAD chlo-
rophyll meter reading measure nitrogen deficiency and the root rating
and percent of stand lodged measure corn rootworm damage (Varvel,
Schepers, and Francis, 1997; Spike and Tollefson, 1991).
Denote the indemnity received as 1(5, fl) , where fl is an index of the
level of coverage such that I~ > O. Because the policy insures against
BMP failure from input deficiency and 5 and 0 are positively correlated, te
and 15 should have opposite signs. Because 5 and 0 are positively cor-
related, high realizations of 5 are associated with high realizations of 0
and vice versa. If te > 0, then high realizations of 0 and 5 are associated
with high yields and low losses, so that indemnities should decrease in 5,
but if te < 0, then high realizations of 0 and 5 are associated with low
yields and high losses, so that indemnities should increase in 5. In either
case, te and 15 have opposite signs.
Denote the premium paid as M(fl) , which can be actuarially fair or
greater to allow the insurance provider to cover costs and earn a normal
profit. Lastly, it is assumed that the producer does not choose the level
of coverage fl, but that green insurance is a take-it-or-Ieave-it offer. Sub-
sequent research can derive and characterize the optimal fl.
The producer purchasing insurance chooses the optimal decision rule
x * (a, fl) by maximizing expected utility conditional on the observed a
and offered fl. The objective and first order conditions are:
MaxI..fofs u (n)W(5 I x,O)q(O I a)h(e) d5 dO de (3.3a)
fEfoIsu'(n)(fx(x,O,e) -r)w(5 I x,O)q(O I a)h(e) d5 dO de +
J.·fofsu(n")w x (s I x,B)q(B I a)h(e) ds dB de = 0 (3.3b)
where, 1C = f(x,B,e) -rx +/(s,f3) -M(f3). For any a and f3, (3.3b) implicitly
defines x * (a,f3) > 0, which may differ from x * (a) due to the insurance.
Expected utility is the expected value of the objective function at
x * (a, f3): faf&fofsu(1C(X * (a,f3)))w(s I x,B)q(B I a)h(e)g(a)ds dB de da. Note
that fxx < 0 and u" < 0 are no longer sufficient to satisfy the second order
condition. Rather the second order condition is satisfied by assumption
to ensure that EU G1 is a maximum.
Table 3.1 summarizes the sequence of events and clarifies the infor-
mation set available for each production system. In each system, pro-
ducers choose optimal input use as a function of their information. For
the status quo system, input use is constant at X. For the BMP, produc-
ers choose x* as a function of the observed a. With insurance, producers
choose x* as a function of a, but adjust their choices to respond to the
insurance coverage f3. Note that for all three systems, B and a are jointly
distributed according to b(B,a). For the status quo, a is ignored so that it
seems B - v(B), where v(B) = fab(a,e)da as in (3.1), while for (3.2) and
(3.3), b(e,a) is factored into g(a)q(e I a).
Table 3.1. Sequence of Events for Each Production System
Status Quo BMP Green Insurance

Pay premium M(f3)
Draw 0 and a Draw 0 and a - b(e,a) Draw Oand a -b(e,a)

- b(B,a)
Observe a Observe a
Apply x Apply x* (a) Apply x*(a,[J)
Draw e - h(e) Draw e - h(e) Draw e - h(e)
Draw S - w(s I x,e) and

determine l(s,f3)
Profit = f(X,O,E)-rX Profit = f(x*(a),O,&) Profit = f(x*(a,fJ),O,&)

-rx*(a) -rx * (a,fJ) - M(fJ) + l(s,fJ)
3.2. BMP Adoption Incentives

Due to the complexity resulting from the various sources of uncertainty
and the composite functions, intuitive analytic results using the model as
specified are difficult to obtain. Therefore, the analysis uses second or-
der approximations of expected utility in a manner similar to Hennessy
and Babcock (1998). For notation, a bar over a variable denotes its
mean, o} is the variance of j, and 0jk is the covariance between j and k.
Use R(Jr) = -u" (Jr) / u' (Jr) to denote the Arrow-Pratt risk aversion function
and define i = f(x * (a),jj,e) - rx * (a). For convenience omit the argu-
ments of the crop growth and utility functions when obvious.
PROPOSITION 1. For a producer adopting the BMP, expected utility in-
creases by approximately T1 + T2 + T3, where T1
=-O.S[x-x*(aWu'(i)fxx T2 = O.50;u'(i)fxxx;, and T3 = oauu'(i)fxuxa .
Proof. See appendix.
The incentive to adopt the BMP depends on four key factors interacting
through the crop growth function: (i) the effect adoption has on optimal
input use, (ii) the noise and variability of the BMP signal, (iii) the quality
of the BMP signal as a measure of true input availability, and (4) pro-
ducer preferences. The terms T1, T2, and T3 capture the first three fac-
tors respectively, while the utility function captures the last factor. The
sum of the terms is ambiguous, but the signs of each term can be deter-
mined.
The difference [x - x * (a)] in T1 measures the impact of BMP adoption
on optimal input use. Because T1 > 0, as this impact increases, the BMP
becomes more valuable and adoption incentives increase. Since fxx < 0,
T2 < 0, implying that a more noisy and variable BMP signal decreases
adoption incentives. To show T3 > 0 requires further discussion, but indi-
cates that a high quality BMP signal increases adoption incentives.
°
Assume a(! > o. If 'xu < 0 (x and () are substitutes), when a is high, ()
is likely to be high and thus fx low, so that optimal input use decreases
in a. The reverse is true if fxo> 0 (x and () are complements), since in this
case when a is high, fx is likely to be low so that optimal input use in-
creases in a. In either case, fxo andxa have the same sign so that T3 is
positive. Similar logic implies that if 0 aU < 0, then fxo and xa have oppo-
site signs so that T3 is again positive.
The model specification does not include a cost for information collec-
tion. Cost matters, and the higher the cost for information, the lower the
BMP adoption incentive. However, assuming the BMP has no cost
serves as a useful benchmark and allows the analysis to focus on these

other pertinent factors.
Risk preferences do not affect BMP adoption incentives in Proportion
1. This occurs because of the assumption that fx - reO at the point
used for the second order expansions, i.e. that uncertainty does not
cause input use to deviate significantly from the standard deterministic
optimality condition fx - r = O. The greater the deviation of fx - r from
zero, the greater the approximation error of Proposition 1.
Proposition 1 supports the common perception that risk is an important
factor in the BMP adoption decision. However, the effect of risk does not
arise from the curvature of utility in stochastic profit (risk preferences),
but from the curvature of profit in the stochastic variables. Indeed, under
expected profit maximization Proposition 1 hardly changes - u' =1 and
drops out of the terms, so that the same factors still determine BMP
adoption incentives.
In general, voluntary programs using cost share subsidies focus exclu-
sively on cost and ignore these other factors identified by PropOSition 1.
Green insurance and other policy instruments can exploit these other
factors and can be used with, or instead of, cost share subsidies to en-
courage BMP adoption. An exploration of these other factors and how
they impact BMP adoption incentives follows.
BMP adoption changes optimal input use. Intuitively, if BMP adoption
causes a relatively large change in optimal input use, this indicates a
relatively greater increase in input use efficiency and so in adoption in-
centives too. T1 = - 0.5[x - x * {aWu' {i)fxx captures this intuition by us-
ing x - x * (a) to approximate x - E[x * {all. The greater the change in
optimal input use, the greater the BMP adoption incentive.
Because T1 > 0, the greater the change in optimal input use, the less
the need for government intervention to encourage BMP adoption. Also,
assuming that adoption reduces optimal input use and that pollution in-
creases with input use, the greater the reduction of pollution with BMP
adoption. A difficult policy problem occurs when status quo input use
creates a significant pollution problem, but the best BMP does not sub-
stantially reduce optimal input use. An example of such a BMP is IPM for
perSistent insect and weed pests - despite IPM adoption, pesticide use
remains high. Relatively greater subsidies are required to encourage
adoption and relatively small reductions in pollution result. In such cases,
research expenditures to develop new BM Ps or alternative strategies
such as area wide management or pest eradication may be more effec-
tive.
Intuitively, as the signal a becomes noisier and more variable, the in-
centive to adopt the BMP should decrease. Similarly, as the information
content of the signal a improves so that it more accurately reflects true
input availability, adoption incentives should increase. In Proposition 1,
T2 = O.50';u'(,r)fxxx; and T3 = O'aou'(,r)fxoxa capture these effects. As pre-
viously discussed, T2 < 0 and T3 > O. For producers using a, these two
terms describe the necessary tradeoff between the welfare gain from the
information in a resulting from its covariance with e and the welfare loss
due to the noise and variability in the same signal.
Since the variance of a depends positively on both the variance of e
and signal noise, research that reduces either factor increases adoption
incentives. Reducing signal noise also increases the correlation with e,
further increasing adoption incentives. Improved information technology
reduces signal noise and alternative practices can reduce the variance of
e. For example, research can indicate how to manage soils to reduce
nutrient variability by using manure applications, cover crops, and differ-
ent crop rotations, while boll weevil eradication and area wide manage-
ment of corn rootworm have successfully reduced the variance of pest
pressure.
3.3. Green Insurance

PROPOSITION 2. Green insurance increases expected utility for a producer
adopting the 8MP by approximately T4 + Ts + T6 + T7 where T4 =
I
O.50';u'(lss-RI;), TS=O'SlJu'RI/ol T6=j)J'(Ip-Mp) and T7=

O.5fi2U'[fxxx~ +Ipp -Mpp -R(lp _Mp)l!].
Proof. See appendix.
Proposition 2 specifies the additional BMP adoption incentives pro-
vided by private provision of green insurance in a competitive market,
extra incentives obtained without government expenditures on subsidies.
Four key factors interacting through the indemnity and premium sched-
ules determine these additional incentives: (1) the variance of the insur-
ance Signal, (2) the correlation between the insurance signal and actual
losses, (3) the level of coverage, and (4) producer risk aversion. Explo-
ration of these four factors and how they can potentially be used to en-
courage BMP adoption follows.
Since the insurance signal is not a perfect measure of losses due to
BMP failure, the curvature of expected utility in the insurance Signal de-
termines the effect of the insurance signal's variability on adoption in-
centives. T4 = O.50';u' (Iss - AI;) determines this effect and its sign de-
pends on Iss (the curvature of profit in 5) and risk aversion (the curva-
ture of utility in profit). Because yield is concave in 0, damage or yield
loss is convex in o. Then because 5 and 0 are positively correlated, a
reasonable assumption is that expected damage or yield loss is also
convex in 5. Given that indemnities should be positively correlated with
losses, the indemnity schedule should also be convex in 5 (Iss > 0),
though this need not be the case, since the insurer chooses the sched-
ule. If Iss > 0, producers prefer increasing the variance of 5 since it in-
creases expected profit, but due to risk aversion, producers prefer to
avoid the increased income variability. As a result, the impact of the vari-
ance of 5 is analytically ambiguous.
Intuitively, producers prefer indemnities that are tightly correlated with
actual losses, so that insurance decreases income variability by paying
the largest indemnities when most needed and vice versa. However,
since the insurance signal is imperfect, some "insurance risk" remains
because the insurance can pay indemnities that are in excess of losses
or too low and thus destabilize income. The positive correlation between
5 and 0 captures this correlation between the insurance signal and actual
losses-if 0 is low,s tends to be low so that indemnities are high, and
vice versa. In Proposition 2, T5 = O"sou'RI.t1J supports this intuition since it
is positive. As the covariance between 5 and 0 increases, the insurance
further reduces income variability, and the greater the risk aversion, the
greater the increase of adoption incentives.
Interestingly, if Iss> 0, developing new or alternative insurance signals
that are less noisy and more accurately reflect true yield losses has an
ambiguous impact on the adoption incentives of producers purchasing
the insurance. If the correlation between the signal 5 and true input
availability 0 increases, the insurance risk is reduced and adoption in-
centives increase with insurance coverage. However, if the increased
correlation is associated with a reduction in the variance of 5, then adop-
tion incentives decrease because of the convexity of the indemnity in 5. If
this convexity effect is ignored, the anticipated returns to research that
improve the insurance signal in this manner will be less than actually
realized.
The impact of the offered level of insurance coverage on adoption in-
centives depends on the how the indemnity and premium schedules are
designed. In Proposition 2, insurance coverage has a first order and a
second order effect on adoption incentives, captured by T6 and T7 re-
spectively. Intuitively the first order effect should be positive - insurance
coverage should increase adoption incentives, which requires (1/3 - M /3) >
O. However, the second order effect is negative, assuming (I pp - Mpp) < 0,

or if positive, its effect is negligible, so that as coverage increases,
eventually producers prefer not to purchase insurance.
3.4. Conclusion
This chapter identified various factors determining adoption incentives for

a BMP that utilizes imperfect information, then discussed the potential to
use these factors to encourage BMP adoption. Traditional incentive pro-
grams focus on cost and ignore these other factors. Besides reducing the
cost of information collection, reducing signal variance and increasing its
correlation with true input availability also encourages BMP adoption. In
addition, BMPs with greater impacts on optimal input use are more read-
ilyadopted.
This chapter also identified factors determining whether green insur-
ance increases BMP adoption incentives. In general, any factor that in-
creases the correlation between the indemnity and actual losses in-
creases the positive impact of green insurance on adoption incentives.
Thus reducing insurance signal variance increases the positive impact of
green insurance, though the likely convexity of the indemnity in the in-
surance signal counteracts this. Increasing the correlation between the
insurance signal and true input availability also increases the positive
impact of green insurance. The signs of the first and second order effects
of the level of insurance coverage depend on the design of the indemnity
and premium schedules. For risk aversion to increase the positive impact
of green insurance requires that the positive correlation of the indemnity
with actual losses dominate the negative effect of indemnity uncertainty
due to insurance signal noise. The analysis indicates the care needed in
designing the indemnity and premium schedules to create the desired
incentives.
This chapter provides a conceptual exploration of BMP adoption incen-
tives and the impact of green insurance. The analysis demonstrates the
potential value of green insurance and introduces it as another policy
instrument to mitigate some types of non-point source pollution. The
analysis also raises interesting questions that warrant further research.
For example, the analysis ignores the impact of information and insur-
ance on optimal input use, yet the policy motivation for encouraging BMP
adoption is the resulting environmental benefits due to reduced input
use. Indeed, some information-based BMPs actually increase input use
(Harper et al., 1990; Osborn et al., 1994). What conditions ensure that a
BMP reduces input use and pollution? Also, what conditions make green
insurance a more efficient incentive mechanism than other voluntary and

mandatory mechanisms for attaining socially optimal pollution? Addi-
tional analysis and a more comprehensive model are required to address
these and related questions.
Appendix
Proof of Proposition 1.
Status quo utility is a function of x , B, and e. Take a second order Tay-
lor series expansion at ( x(a) , B , e), where x(a) denotes the status quo
optimum x when a = a:
u(K(x,B,e)) == u(f - rX(a)) + (x - x(a))(u'(fx - r)) +
(B - B)u' to + (e - e)u' ~ + O.S(x - x(aW[u" (fx - r)2 + u' fxx] +
O.S(B - B)2[U" f,} + u' f(I(J] + O.S(e - e)2[u" ~2 + u' faJ +
(x - x(a))(B - B)[U" f,,(fx - r) + u' fx"l + (x - x(a))(e - e)[u" fe(fx - r) + u' fxel +
(B - B)(e - e)[u" f,,~ + u' feel. The utility function and its derivatives are
evaluated at f - rX(a) and the crop growth function and its partial deriva-
tives are evaluated at (x(a) , B, e). Noting that when a = a,
v(B) = q(B I a) in (3.1 a), take the expected value of both sides:
(A 1) Ie I" U(K(X, B, e))q(B Ia)h(e)dB de == u(f - rX(a)) +

O.Su:(u" fi + u' f(I(J) + O.Su;(u" f; + u' fa:) + O.S(x - x(a))2u' fxx
because of independence assumptions and because, from the first order

condition, fx - r == 0 at the point of expansion.
Utility for the BMP is a function of a, B, and e. Take a second order
Taylor series expansion at (x * (a), B, e):
u(K(x(a), B, e)) == u(f - rx* (a)) +

(B - B)u' f" + (e - e)u' fe + (a - a)u' (fx - r)xa + O.S(B - B)2(U" f,,2 + u' f(I(J) +
O.S(e -e)2(u" ~2 + u' fa:) + O.S(a-a)2(u" (fx - r)2 x! + u' fxxx! + u' (fx - r)xaa) +
+ (a - a)(B - B)(u" (fx - r)f"xa + u' fx"xa) + (B - B)(e - e)(u" f,,~ + u' fee). The
utility function and its derivatives are evaluated at f - rx * (a) and the
crop growth function and its partial derivatives are evaluated at
(x * (a) , 1i , 8). Noting that when a = a, q(B I a) = q(B I a) in (3.2a), take

the expected value:
(A2) faf&fou(7r(B,E:,a))q(B I a)h(E:)g(a)dBdE: da 5!!

u(f - rx * (a)) + O.5a;(u"fo2 + u'foo) + O.5a~(u"f.2 + u'f«) +
0.5a~u'fxxx~ + aaOu'fxoXa
because of independence assumptions and because, from the first order
condition, fx - reO at the pOint of expansion.
Because v(B) = q(B I a) in (1 a) and q(B I a) = q(B I a) in (3.2a) when
a =a, both first order conditions are identical so that x(a) = x * (ii). To
obtain the reported expression, substitute x * (a) for x(a) in (A 1) and
subtract it from (A2).
Proof of Proposition 2
With insurance, utility is a function of B, E:, a,S, and p. Following the

method of Proposition 1, take the expected value of a second order Tay-
lor series expansion at (1i, '&, a,S, Po)' where Po = 0, and use
u"= -Ru':
(A3) fafEfoIsu(7r)w(s I x * (a,p),1i)q(B I a)h(E:)g(a)ds dB dE: da 5!!
u(7r) + 0.5a;u' (f... - Rf}) + 0.5a;u' (fOIl - Rfo2) + 0.5a~u'fxxx! +

a aoU' fxoXa - asou' RI/o + fJu' (I fJ - MfJ) +
0.5PV[f~~ + IfJfJ - MfJfJ - R(lfJ - MfJ)2].
The utility and risk aversion functions are evaluated at

7r = f - rx + 1(5, Po) - M(Po) and the crop growth function at (x * (a, Po), 1i,
8). Since Po = 0, x * (a) = x * (ii,Po) and I(P) = M(P) = 0 and

7r = f -rx. Lastly, integrate (A2) across 5, which does not change its
value, and subtract it from (A3).
References
Babcock, B.A., and D.A. Hennessy (1996), "Input Demand under Yield and Revenue
Insurance." American Journal of Agricultural Economics, 78: 416-427.
Chambers, R.G., and J. Quiggin (1996), "Non-poi nt-Source Pollution Regulation as a
Multi-task Principal-Agent Problem." Journal of Public Economics, 39: 95-116.
Feather, P., and J. Cooper (1995), "Voluntary Incentives for Reducing Agricultural
Nonpoint Source Water Pollution." U.S. Department of Agriculture, Economic Re-
search Service, Agricultural Information Bulletin No. 716. Washington, DC, May.
Greene, C.R., RA Kramer, G'w. Norton, EG. Rajotte, and R.M. McPherson (1985),
"An Economic Analysis of Soybean Integrated Pest Management." American Jour-
nal of Agricultural Economics, 67: 567-572.
Harper, J.K., M.E Rister, J,W. Mjelde, B.M. Drees, and M.O. Way (1990), "Factors
Influencing the Adoption of Insect Management Technology." American Journal of
Agricultural Economics 72: 997-1005.
Hennessy, D.A., and BA Babcock (1998), "Information, Flexibility, and Value Added.
Information Economics and Policy, 10: 431-449.
Hrubovcak, J., U. Vasavada, and J.E Aldy (1999), "Green Technologies for a More
Sustainable Agriculture." U.S. Department of Agriculture, Economic Research
Service, Agricultural Information Bulletin No. 752. Washington, DC, July.
Miranowski, JA, U.F,W. Ernst and F.H. Cummings (1974), "Crop Insurance and In-
formation Services to Control Use of Pesticides." U.S. Environmental Protection
Agency Research Report EPA-600/5-74-018.
Nowak, P. (1992), "Why Farmers Adopt Production Technology." Journal of Soil and
Water Conservation, 47: 14-16.
Osborn, C., D. Hellerstein, C. Rendleman, M. Ribaudo, and R. Keim (1994), "A Pre-
liminary Assessment of the Integrated Crop Management Practice." U.S. Depart-
ment of Agriculture, Economic Research Service, Staff Report AGES 9402, Febru-
ary.
Smith, V.H., and B.K. Goodwin (1996), "Crop Insurance, Moral Hazard, and Agricul-
tural Chemical Use." American Journal of Agricultural Economics, 78: 428-438.
Spike, B.P., and J.J. Tollefson (1991), "Yield Response of Corn Subjected to Western
Corn Rootworm (Coleoptera: Chrysomelidae) Infestation and Lodging." Journal of
Economic Entomology 84: 1585-1590.
U.S. Department of Agriculture, Natural Resource Conservation Service. Environ-
mental Quality Incentives Program (EQIP) Fact Sheet (1977a).
http://www.nhq.nrcs.usda.gov/OPNFB960PNeqipfact.html.
U.S. Department of Agriculture, Natural Resource Conservation Service. Environ-
mental Quality Incentives Program (EQIP) Questions and Answers (1977b).
http://www.nhq.nrcs.usda.gov/OPNFB960PNeqipQ%26A.html.
Varvel, G. E., J. S. Schepers, and D. D. Francis (1997), "Chlorophyll Meter and Stalk
Nitrate Techniques as Complementary Indices for Residual Nitrogen." Journal of
Production Agriculture, 10: 147-151.
Westra, J., and K. Olson (1997), "Farmers' Decision Processes and Adoption of Con-
servation Tillage." Department of Applied Economics Staff Paper P97-9, University
of Minnesota.
Wu, J.J., and BA Babcock (1996), "Contract Design for the Purchase of Environ-
mental Goods from Agriculture." American Journal of Agricultural Economics, 78:
935-945.
4
Uncertainty and Adoption
of Sustainable Farming Systems
David J. Pannell
There is wide interest among agricultural policy and research institutions

in the process of adoption of innovations that promote land conservation,
impediments to that adoption, and possible measures to promote adop-
tion. Implicit in this interest is a perception that, despite the existence of
government programs, adoption by farmers of "sustainable"1 farming
practices has been lower and slower than would be socially optimal (e.g.
Lockie and Vanclay, 1997; Rae and Gruen, 1997). Many factors have
been suggested as contributing to this (e.g. Pannell, 1999; Vanclay,
1997; Cary and Wilkinson, 1997; Sinden and King, 1990), including: high
implementation costs, lack of direct payoff from implementation, lack of
physical and human capital, lack of a sufficient "stewardship" ethic
among farmers, farming subcultures and social pressures, lack of a suit-
able regulatory framework, and risk and uncertainty.
This chapter focuses on the last of these factors. It is argued that un-
certainty has been under-recognized as one of the key factors inhibiting
uptake of land conservation practices. In part, this under-recognition may
be because the majority of the enormous volume of research conducted
on adoption of agricultural innovations has focused on innovations with
short-term productivity-oriented benefits. It will be argued here that the
problems of uncertainty about "sustainable" innovations are much more
profound and intractable than for most farming innovations.
In addition, it seems that uncertainty has been under-recognized as an
impediment to adoption even for productivity-oriented innovations. Risk
and uncertainty have often been considered as factors reducing the rate
of adoption of rural innovations (Lindner et al., 1982; Tsur et al., 1990;
Leathers and Smale, 1992; Shapiro et al., 1992; Smale and Heisey,1993;
Feder and Umali 1993). However, this has largely been assumed, rather
68 David 1. Pannell
than known, as they have rarely been addressed adequately in empirical

studies of adoption (Lindner, 1987). The lack of empirical research may
largely be attributable to the great difficulty of accurately measuring the
relevant uncertainty-related variables.
However, in a recent study, Abadi Ghadim and Pannell (1998) and
Abadi Ghadim et al. (1999) have shown that uncertainty plays a clear,
measurable and substantial role in the adoption of a new type of crop.
Their conceptual framework (based on Bayesian decision theory) and
empirical findings have profound implications for adoption of
"sustainable" farming innovations, and it is these implications that are the
focus of this chapter.
To introduce important background, the next section is an informal out-
line of the conditions for adoption of an agricultural innovation. Then the
more formal framework of Abadi Ghadim and Pannell (1999) for consid-
ering the role of uncertainty in adoption decisions is presented briefly.
Thereafter, the various rol~s of uncertainty in the adoption process are
expanded on, drawing on available evidence and numerical examples.
Finally, implications for extension and policy are discussed.
4.1. The Conditions for Adoption of an Agricultural Innovation

Pannell (1999) argues that farmers are likely to come to any radical inno-
vation with skepticism, uncertainty, prejudices and preconceptions. Un-
less they are new to farming, they will have trialled other innovations in
the past and concluded that at least some of them fell far short of the
claims made for them. They will be particularly wary of a system that is
radically different from that with which they are familiar and comfortable.
They will probably hold an attitude that the people advocating such a
radical system do not understand the realities of farming, or at least of
their farm.
In getting past this initial set of attitudes and beliefs, there are several
specific hurdles that must be overcome. The following sub-sections de-
scribe the states of farmer awareness or knowledge that must be
achieved.
4. 1. 1. Awareness of the innovation
In this context, "awareness" means not just awareness that an innovation
exists, but awareness that it is potentially of practical relevance to the
farmer. Reaching this point of awareness is a trigger that prompts the
farmer to open his or her ears and eyes - to begin noting and collecting
Uncertainty and Adoption of Sustainable Farming Systems 69
information about the innovation in order to inform their decision about

whether or not to go to the next step of trialling the innovation.
4.1.2. Perception that it is feasible and worthwhile to trial the innovation
There is strong evidence that, the world over, most farmers are "risk-
averse" (Antle, 1987; Bardsley and Harris, 1987; Myers, 1989; Pluske
and Fraser, 1996). This is evident from the observation that they will not
leap into large-scale adoption of a new innovation. Rather, they generally
employ small-scale trials, adjusting the scale either upwards towards full
adoption or downwards towards disadoption as they gain knowledge and
confidence in their perceptions about its performance.
Conducting a trial incurs costs of time, energy, finance and land that
could be used productively for other purposes. To be willing to trial an
innovation, the farmer's perceptions of it must be sufficiently positive to
believe that there is a reasonable chance of adopting the innovation in
the long run. It is not necessary for the innovation to be thought to be
better than current practice, because the farmer realizes that the results
of a trial may revise his or her perceptions upwards. However, it cannot
be too much worse or the chance of recovering the cost of the trial
through later productivity improvements will be too low.
This trial phase is very important. If small-scale trials are not possible
or not enlightening for some reason, the chances of widespread adoption
are greatly diminished. This is because farmers will be very unlikely to
leap to full-scale adoption due to the real risk that the innovation will
prove a full-scale failure.
4.1.3. Perception that the innovation promotes the farmer's objectives
Lindner (1987) in a wide-ranging review of the adoption and diffusion
literature concluded that the objectives of individual farmers figure cen-
trally in the adoption and diffusion process. He found that, "there is com-
pelling empirical support for this emerging consensus that the final deci-
sion to adopt or reject is consistent with the producer's self interest." (p.
148)
"Self interest" in this context is considerably broader than merely
"profit". It may, for example, include objectives related to risk, leisure
and environmental protection. Nevertheless, profit is a particularly im-
portant element of "self-interest". Indeed, the available evidence indi-
cates that, although the speed of uptake of innovations is influenced by a
range of factors (including social and demographic factors), the final
level of uptake seems to depend primarily on economic factors (e.g.
70 David J. Pannell
Marsh et al., 2000). There is also evidence that even for innovations ori-
ented towards resource conservation, economic considerations are the
most important determinants of actual adoption decisions (Cary and Wil-
kinson, 1997; Sinden and King, 1990).
4.1.4. Impacts of uncertainty on adoption
Within the adoption process, uncertainty has several negative influences.
The key ways in which uncertainty inhibits adoption are as follows.
The fact that the final result of adopting a particular practice is highly
uncertain is an intrinsic discouragement to adoption for most people. As
noted earlier, most farmers are averse to risk and uncertainty, meaning
that they place greater weight on potential negative outcomes than on
positive outcomes. This relates to the condition above of meeting the
farmer's objectives. For some farmers, avoidance of risk and uncertainty
is an important objective.
Even if farmers are not discouraged by uncertainty per se, they may
well be discouraged by the consequences of that uncertainty, particularly
if it results in inaccurate perceptions or misinformation. This also relates
to the condition regarding farmers' objectives. If a farmer perceives incor-
rectly that an innovation is not consistent with his or her objectives, this
misperception is an impediment to adoption. The condition relating to
trialling is also relevant here. If the farmer does not conduct trials, a
chance to correct the misinformation is missed. Indeed, if the farmer is
badly misinformed, this in itself may cause the farmer to believe that a
trial is not worthwhile, trapping him or her in a state of ignorance.
Irreversibility of environmental damage is often proposed as a reason
for action to enhance conservation. This is a different motivation than
aversion to uncertainty. It relates to the concept of "option value"
whereby keeping open the options for resource use has a positive value
due to the potential for unforeseen circumstances. To the extent that an
option value is relevant to the farmer's objectives, it may influence his or
her behavior. This appears to act in favor of adoption. However, if a con-
servation practice is itself irreversible to some extent (or expensive to
reverse), there is then an option value in not adopting it. For example,
this would apply to the strategies that involve planting of trees on a pro-
portion on cropland (as recommended in parts of Australia to prevent soil
salinization in non-irrigated regions).
4.2. Conceptual Framework

The adoption process consists, in large part, in the collection, integration
and evaluation of new information. In other words, it is a process in
which uncertainty is reduced steadily over time. Early in the process, un-
certainty is very high, and the quality of decision-making may be low. As
the process continues, if it proceeds at all, uncertainty falls and better
decisions can be made. Viewed in this light, it would be fair to say that
the adoption process is never completed, in the sense of reaching zero
uncertainty. All options are continuously open to question and review, as
new information is obtained and/or circumstances change. The concep-
tual framework presented below is included to reinforce and clarify these
ideas. The framework highlights the role of learning in the dynamics of
adoption, and clarifies the benefits of trialling.
The framework represents a farmer's decision problem regarding the
allocation of land to a new "sustainable" farming system and to traditional
methods. For simplicity it is assumed that the decision involves only a
single new system and a single traditional system. The sustainable sys-
tem is characterized by short-term costs and long term benefits. It is as-
sumed in this discussion that a single-year trial of the system gives use-
ful information about its performance. Potential flaws in this assumption
are considered later.
Let,
As =Area of sustainable farming system,
An = Area of traditional farming system,
AT = Total arable area on the farm = As + An,
gs = Gross margin of sustainable farming system, and
gn = Gross margin of traditional farming system.
Assume that the farm's land is heterogeneous (e.g. in soil structure,

chemical composition of the soil, weed species present) so that gs and
gn vary within the farm. For any given value of As it is possible to calcu-
late Gs and Gn , the mean gross margins of sustainable and traditional
farming across the areas on which they are grown. Assuming profit-
maximizing behavior, Gs will fall as As is increased, due to the heteroge-
neity of land with respect to the value of gs - gn . Profit (n) is:
n =Gs x As + Gn x An ( 4.1)
If the farmer maximizes profit for the current period, some area of the
sustainable farming system will be grown so long as the gross margin of
sustainable farming is greater than that of traditional farming on any part
72 David J. Pannell
of the farm. Of course, such a simplistic approach is inappropriate for the

assessment of sustainable farming systems. The framework below in-
cludes the key elements of time, risk, and learning. A quantitative imple-
mentation of the framework may also include spatial linkages between
the farming systems and, depending on the purpose of the analysis, off-
farm effects.
It is assumed that the farmer's objective is to maximize the expected
value of the net present value of profits2. Therefore the farmer is con-
cerned with the gross margins of the alternative farming systems in fu-
ture years beyond year 1. 3
Consider that the farmer is uncertain about the economic performance
of the sustainable farming system. There will be uncertainty about its
biological productivity and its capacity to prevent land degradation and
there may also be uncertainty about sale prices and input costs, espe-
cially if it involves production of a new product unfamiliar to the farmer. A
trial of the system will provide information about its yields, prices and
impacts on the resource base. This information is likely to reduce the
farmer's uncertainty in future years and allow better deCision-making.
Before conducting a trial of the sustainable system, the farmer is un-
certain about the value of Gs for any given As, but is able to subjectively
state a probability distribution for it. From the information generated by
the trial, the farmer revises his or her subjective beliefs about the profit-
ability of the system. Based on this revised (probably more accurate)
perception, the farmer decides whether or not to continue with the new
system and, if so, what area of the farm to devote to it. With each year of
trialling, this decision is refined and improved. A trial in year t provides
information that allows improved estimates of Gs for subsequent years.
This in turn allows improved selection of As for subsequent years.
If the farmer decides to trial the sustainable system, the dynamic profit
function can be expressed as:
n = Gs 1xAs1+Gn1 x(AT - A s 1) + NPVt=2.. N[Gst xAst+Gntx(AT-AstJJ (4.2)
where,
Jl = the net present value,

Ast = the area of the sustainable system in year t,
Gst = the average gross margin of sustainable farming in year t given
Ast. In this and subsequent equations, G represents the (unknown) ac-
tual gross margin, not the farmer's subjective estimate.
The gross margins have time subscripts in part because they are
changing due to land degradation, and in part because the sustainable
system is likely to have up-front costs and delayed payoffs.
If the farmer chooses not to trial the sustainable system in year 1, the
profit function is:
110 = Gn 1 x AT + NPVt=2.. N[Gsto x Asto + Gnto x (AT - Asto)} (4.3)
The 0 subscripts signify that these values may be different to those in

equation (4.2) due the absence of a trial in year 1. AstO is different to Ast
because information collected in the trial in year 1 affects subsequent
decision making about the area of the sustainable system. Gsto is differ-
ent to Gst because Gs depends on As (which has changed) and also be-
cause the absence of a trial in year 1 means that the impacts of the
sustainable system on resource conservation are delayed.
The difference between the two equations indicates whether the bene-
fits of the trial outweigh the opportunity costs.
11-110=GS 1 xA s 1- Gn1 xAs1 +/ (4.4)
where, / represents the benefits in later years of trialling in year 1.
/ = NPVt=2 .. N[Gst x Ast +Gnt x (AT - Ast)-Gsto xAsto-GntO x(A T-AstO)] (4.5)
Rearranging gives:
/ = NPVt=2.. N[(Gst - Gsto} x Asto + (Gst - GntJ x (Ast - Asto)] (4.6)
Thus, the benefits of trialling can be decomposed into two elements:

the gain in profitability for the area that would have been allocated to the
sustainable system in future years even without the trial in year one, (Gst
- GstO) . AstO, plus the gain in profit on the area converted from the tradi-
tional to the sustainable system in future years as a result of the trial,
(Gst - Gnt)· (Ast - Asto)·
The first element springs from actual biophysical changes set in place
directly by the trial. In cases where the trial is conducted on a small
scale, this element is likely to be small in magnitude. The second ele-
ment springs from changes in perceptions due to the trial, leading to
changes in subsequent management.
At the start of the next year, exactly the same decision problem is
faced again, with the exception that perceptions about the sustainable
74 David J. Pannell
system are likely to be different than they were in year 1, especially if a

trial has been conducted. When viewed in this light, the trial can be seen
as the first step in adoption. Indeed, it might be considered that trialling
is indistinguishable from adoption - that each production system is al-
ways and forever on trial, with different decisions made as perceptions
and expectations evolve.
4.3. Factors that Contribute to High Uncertainty

about Conservation Innovations
"Sustainable" farming systems are prone to high levels of uncertainty for a

range of reasons.
4.3.1. Lack of experience
Early in the process of any innovation, uncertainty is high. Indeed, the
remaining level of uncertainty may provide a useful, measurable index of
the extent to which the adoption process has progressed. In the case of
adoption, the problem of uncertainty due to lack of experience has a
"Catch 22" style mirror problem: lack of experience due to uncertainty. If
uncertainty is so high as to inhibit trialling, it is also inhibiting the key tool
available for reducing uncertainty. Information from observing other
farmers' experiences with the innovation provides a potential way out of
this vicious cycle, but in cases where adoption levels are perSistently
very low (as with some conservation measures), even this solution is un-
available. The social process of diffusion of innovations is very important
(e.g. Rogers 1995), but it depends on early adoption by a minority to
seed the process.
4.3.2. Partial relevance of off-farm information
Even if some farmers have adopted an innovation, the relevance to other
farmers of their experiences will vary. For an innovation such as a
slightly modified cropping input, the potential to extrapolate results to
other farmers is probably high. On the other hand, results from some
land conservation practices may be more region-specific. Consider soil
salinity in non-irrigated regions of southern Australia. The key strategy to
prevent such salinity is to attempt to use a greater proportion of the rain-
water, to prevent it draining deep into the soil and raiSing the naturally
saline water table that exists across much of Australia. To this end, prac-
tices such as establishment of high-water-using perennial plants are ad-
vocated. A problem is that the underground geology throughout much of
the region is very complex, so that farmers may have little precise idea
about which land is contributing to a raised water table in a particular

site. Thus, even if perennial plants successfully prevent a salinity prob-
lem at one site, they may fail to do so at another, depending on the un-
derground rock and soil formations and soil types.
4.3.3. Externalities
Some land degradation problems have important "external" impacts. For
example wind erosion on one farm may impose costs on another farm,
such as "sand-blasting" of crops, or burial of fences. Externalities can con-
tribute two different types of uncertainty about the consequences of adop-
tion of land conservation practices. Firstly, a farmer may be uncertain
about who will be the beneficiary if he or she does adopt. If there is a risk
that the benefits will flow mainly to farmers other than the adopter, the in-
centives to adopt are reduced. Secondly, a farmer may be uncertain about
whether their adoption will be ineffective if other farmers do not adopt. For
example, some hydrological catchments span more than one farm, such
that several farms in the catchment contribute to rises in a saline water
table. In such cases, adoption by anyone individual may make a relatively
small contribution to preventing rises in the water table, although further
rises would be prevented if all farmers adopted.
4.4. Factors that Reduce the Information Value of Trials
Given that farmer uncertainty about some land conservation practices is

high, the importance of conducting on-farm trials to reduce this uncertainty
is highlighted. Unfortunately, there is a range of reasons why trials of land
conservation practices may produce information of low quality, and so be
ineffective at reducing uncertainty.
4.4. 1. Long time scales
In the conceptual framework presented earlier, it was assumed that a trial
provides useful information in the first year. For many agricultural innova-
tions, this is realistic (e.g. a new crop variety). However, many land degra-
dation processes are slow relative to the time frames used for most man-
agement decision-making (e.g. soil salinization, soil acidification). In
evaluating a trial, one requires the degradation to be continued under the
old farming system for long enough for differences under the new farming
system to become apparent. Obviously, the slower the degradation proc-
ess, the longer it will take to be convinced about differences in degradation
rates. Unfortunately the great variability inherent in extensive agricultural
production further delays the confident recognition of any such difference.
76 David J. Pannell
Further, long time scales mean that uncertainty about other variables (e.g.
prices) over the relevant time scale is much greater than for a short-term
problem, further adding to the difficulty of decision making.
4.4.2. Heterogeneity of the land
In the last section, the spatial heterogeneity of land degradation problems
was recognised as an impediment to diffusion of innovations from farm to
farm. The same issue applies at the scale of a single farm. A large part of
the potential information value of a trial is derived from its relevance to
other parts of the farm. If a farmer perceives that the trial results are less
than fully transferable, the trial's benefits are reduced.
4.4.3. Minimum scale needed
For many agricultural innovations, it is possible to conduct trials on a small
scale without sacrificing much of the information content of the trial. For
example, new crop types are typically trialled on a scale that represents
just a few percent of the total area of crop on a farm. As knowledge of and
confidence in the crop increases, the scale of production increases. By
contrast, for innovations intended to prevent dryland salinity by increasing
water use, a small-scale trial may have no measurable impact. Especially
when combined with long time scales and geological heterogeneity, the
scale necessary to have an observable impact in a reasonable time may
be little smaller than full-scale adoption. Farmers would naturally be reti-
cent about leaping to such full-scale adoption given their state of high un-
certainty.
4.4.4. Observability
Related to the problem of minimum scale is the issue of observability.
Clearly, low observability of results reduces the information value of a trial.
Examples would include practices intended to reduce soil degradation in-
volving underground processes, such as soil compaction, soil acidification,
soil salinization or leaching of nutrients. Of course, some consequences of
adoption of the practice would be observable in their impacts on above-
ground plant growth. However, if the prime motivation for adoption is pre-
vention of below ground processes, above ground production may provide
a highly imperfect indicator.
4.4.5. Low covariance with traditional practices
Even if a conservation practice is easy to trial on a small scale, giving ob-
servable results quickly and providing information that is relevant to the
whole farm, the information value of the trial may be low relative to most
productivity-related innovations because of the problem of low covariance.

For example, when wheat farmers trial a new variety of wheat they expect
its yields and prices to be highly correlated with traditional varieties. It may
well differ in mean yields, but the farmer would assume that climatic condi-
tions that result in high yields of one variety would also result in relatively
high yields of another variety. This is an enormous benefit in the interpre-
tation of trial results. It makes it possible to extrapolate results with some
confidence to climatic conditions that have not been experienced in the
trial, on the basis that they have been experienced with traditional varie-
ties. This is commonly not the case for land conservation innovations. They
typically are radically and fundamentally different to any existing practices
on the farm. Each observation of the trial's impacts is an isolated observa-
tion, poorly correlated with other observations of events on the farm. This
problem appears to apply to many conservation innovations including, for
example, liming to reduce soil acidity, and tree planting to reduce salinity.
4.4.6. Poor implementation
If an innovation is not implemented properly in a trial, the results of the trial
are clearly compromised. Unfortunately, this outcome is more likely with
land conservation practices than for productivity-oriented innovations be-
cause (a) they are commonly less familiar to the farmer and less similar to
existing farm practices, and (b) they can be more complicated, with more
scope for errors. For example, implementing trials of an agroforestry sys-
tem integrating trees with cropping or livestock would clearly be more
prone to poor implementation than trials of a new crop or a new type of a
traditional crop input.
4.5. Factors that Increase the Cost of Trials
Compounding the problems outlined above is a set of factors that contrib-

ute to trials of conservation practices being highly costly.
4.5.1. Time and effort needed
Poor implementation noted in the last section flagged the greater-than-
average complexity of some land conservation innovations. This is likely to
mean that the amount of time and effort needed to prepare for and conduct
a trial is higher than for simpler innovations.
4.5.2. Minimum scale needed
If the minimum scale for a trial is large (see pOint 3 in the last section),
this further increases the time and effort required. Probably even more
78 David J. Pannell
importantly, it also increases the opportunity cost of land devoted to the

trial.
4.5.3. Irreversibility
The concept of "option values" was outlined earlier. If a practice is irre-
versible or expensive to reverse, the resulting inflexibility imposes a cost
on the farmer due to lost option value. Establishment of trees provide a
good example. Suppose that a farmer establishes a large area of trees to
reduce land degradation, but subsequently a highly effective conserva-
tion technology becomes available that allows traditional farming to con-
tinue without dedicating large areas of land to trees. Because it is expen-
sive to remove the trees, the farmer may be worse of than if he or she
had never established the trees. If a farmer considers such an outcome
to be realistically possible, it would provide a disincentive to adoption.
This is, at heart, a problem of uncertainty. If the farmer knew in advance
whether an improved technology would become available, there would be
no risk of mistaken non-adoption.
4.6. Implications
Based on this discussion, a number of clear implications can be identi-
fied. Firstly, it appears that the problem of uncertainty in adoption of land
conservation practices is much greater and more far reaching than nor-
mally recognized. The fact that farmers have been slow to take up some
innovative land conservation practices is highly understandable when
viewed within the context of the issues raised here - even without con-
sidering the range of other negative influences on adoption of these
practices (Pannell and Schilizzi, 1999).
It does appear that uncertainty is an important cause of market failure
in this case. However, it is not clear whether government intervention can
reduce the extent of this failure. On one hand, government agencies may
be in possession of information from scientific research and other
sources that is in some sense better than that held by at least some
farmers. On the other hand, even if this is true, its accuracy at particular
sites may be unknown, and assessment of its management implications
for particular farmers will certainly be outside the capacity of agencies.
Given the heterogeneity discussed here, such an assessment depends
very much on local knowledge and individual circumstances. Farmers
understand this well, and so are most unlikely to be influenced by advice
from agencies that they should adopt particular practices. Even if the
advice is good, it will probably not be believed, and for sound and pru-
dent reasons. Information on biophysical aspects that does not attempt
to draw management implications for individual farmers is less suscepti-
ble to this problem.
One prominent government response to land degradation problems in
Australia has been the National Landcare Program, a central feature of
which is the formation of formal farmer groups. These playa role in col-
lection and sharing of information, and in this they appear to be partially
addressing the problems of uncertainty discussed here. In particular the
following advantages of the Landcare group approach might be ex-
pected.
• It can speed the flow of information between individuals in the
group,
• It may help to facilitate jOint trials. If farmers agree to share costs,
the problem of high trial costs can be partially avoided. (In practice,
this appears to be uncommon).
• Joint trials, because they are local and farmer-run, have greater
local relevance and credibility than agency information from other re-
gions.
• Perhaps the jOint effort involved may reduce the risk of poor im-
plementation.
Although these are important advantages, it appears that there has
been excessive optimism in some quarters about the extent to which the
Landcare approach can solve the problems of information and uncer-
tainty, especially for the most intractable problem of dryland salinity. In
particular, it seems unlikely that Landcare groups could do much to ad-
dress the following problems discussed earlier.
• The contribution of externalities to high uncertainty.
• The contribution of heterogeneity to high uncertainty.
• Long time scales.
• Cases where the minimum scale needed for trials is large.
• Low observability of some trial impacts.
• Low covariance of the behaviour of the innovation with traditional
practices.
• The high cost of ceasing a trial.
It may be worthwhile for government programs intended to promote
adoption to devote resources to attempting to devise innovative methods
for addressing these aspects of uncertainty.
Another strategy that would avoid several of these remaining problems
would be to attempt to develop technologies which are profitable in their
80 David 1. Pannell
own right, but which have resource-conservation benefits as a side ef-

fect. The Department of Conservation and Land Management (CALM) in
Western Australia is actively pursuing this strategy, in its program to de-
velop tree species that can be commercially viable on what have tradi-
tionally been crop and pasture-based farms (Bartle et al., 1996). Al-
though primarily motivated by a wish to tap into the profit motive of farm-
ers (e.g. Sinden and King, 1990; Cary and Wilkinson, 1997), an addi-
tional benefit of success by CALM would be that problems such as low
observability of below-ground hydrological impacts would become much
less important as an impediment to adoption.
Endnotes
1 No attempt is made here to rigorously define "sustainability". Its usage should be

interpreted broadly, in line with Pannell and Schilizzi (1999), to signal a concern for
conservation and the long term.
2 The framework can readily be extended to include risk aversion.
3 Calculation of "gross margins" should include any relevant spatial linkages between
the systems, such as reduced off-site impacts from erosion, or from spread of weeds
or pests.
References
Abadi Ghadim, A.K. and D.J. Pannell (1998), "The importance of risk in adoption of a
crop innovation: Empirical evidence from Western Australia". Paper presented at the
42nd Annual Conference of the Australian Agricultural and Resource Economics
Society, University of New England, Armidale, NSW Jan 19-21 1998.
Abadi Ghadim, A.K., M. Burton, and D.J. Pannell (1999), "More empirical evidence on
the adoption of chick peas in Western Australia". Paper presented at the 43rd An-
nual Conference of the Australian Agricultural and Resource Economics Society,
Christchurch, New Zealand, Jan 20-22 1999.
Abadi Ghadim, A.K. and D.J. Pannell (1999), "A conceptual framework of adoption of
an agricultural innovation", Agricultural Economics, 21: 145-154.
Antle, J.M. (1987), Econometric estimation of producers' risk attitudes, American
Journal of Agricultural Economics, 69: 509-522.
Bardsley, P., and M. Harris (1987), "An approach to the econometric estimation of
attitudes to risk in agriculture", Australian Journal of Agricultural Economics, 31 :112-
126.
Bartle, J.R., C. Campbell, and G. White (1996), "Can trees reverse land degradation"?
Australian Forest Growers Conference, Mt Gambier, South Australia.
Cary, J.W. and R.L. Wilkinson (1997), "Perceived profitability and farmers' conserva-
tion behaviour", Journal of Agricultural Economics, 48: 13-21.
Uncertainty andAdaptian afSustainable Farming Systems 81
Feder, G. and D. Umali (1993), "The adoption of agricultural innovations: a review",

Technological Forecasting and Social Change, 43: 215-239.
Leathers, H. D. and M. Smale (1992), "A Bayesian approach to explaining sequential
adoption of components of a technological package", American Journal of Agricul-
tural Economics, 68: 519-527.
Lindner, RK (1987), "Adoption and diffusion of technology: an overview", In: Techno-
logical Change in Postharvest Handling and Transportation of Grains in the Humid
Tropics, B.R. Champ, E. Highley and J.v. Remenyi (eds.), ACIAR Proceedings No.
19, ACIAR, Canberra, pp. 144-151.
Lindner, R.K., P.G. Pardey, and F.G. Jarrett (1982), "Distance to innovation source
and time lag to early adoption of trace element fertilizers", Australian Journal of Ag-
ricultural Economics, 26: 98-113.
Lockie, S. and F. Vanclay (eds.) (1997), Critical Landcare, Key Papers Series 5, Cen-
tre for Rural Social Research, Charles Sturt University, Wagga Wagga.
Marsh, S.P., D.J. Pannell, and RK Lindner (2000). The impact of agricultural exten-
sion on adoption and diffusion of lupins as a new crop in Western Australia. Austra-
lian Journal of Experimental Agriculture, 40(4): 571-583.
Myers, R.J. (1989), "Econometric testing for risk averse behavior in agriculture", Ap-
plied Economics, 21: 541-552.
Pannell, D.J. (1999), "Social and economic challenges in the development of complex
farming systems", Agroforestry Systems, 45: 393-409.
Pannell, D.J. and S. Schilizzi (1999), "Sustainable agriculture: A question of ecology,
equity, economic efficiency or expedience"? Journal of Sustainable Agriculture,
13(4): 57-66.
Pluske, J. and R. W. Fraser (1996), "Can producers place valid and reliable valuations
of wool price-risk information?" Review of Marketing and Agricultural Economics, 63:
284-291.
Rae, J. and N. Gruen (1997), "A Full Repairing Lease, Inquiry into Ecologically Sus-
tainable Land Management", Draft Report, September 1997, Industry Commission,
Canberra.
Rogers, E.M. (1995), Diffusion of Innovations, Free Press, New York.
Shapiro, B. I., B. W. Brorsen, and D.H. Doster (1992), "Adoption of double-cropping
soyabean and wheat", Southern Journal of Agricultural Economics, 24: 33-40.
Sinden, J. A. and DA King (1990), "Adoption of soil conservation measures in Manilla
Shire, New South Wales", Review of Marketing and Agricultural Economics, 58:
179-192.
Smale, M. and P.w. Heisey (1993), "Simultaneous estimation of seed-fertiliser adop-
tion decisions", Technological Forecasting and Social Change, 43: 353-368.
Tsur, Y., M. Sternberg, and E. Hochman (1990), "Dynamic modelling of innovation
process adoption with risk aversion and learning", Oxford Economic Papers, 42:
336-355.
Vanclay, F. (1997), "The social basis of environmental management in agriculture: A
background for understanding Landcare", In: S. Lockie and F. Vanclay (eds.) Critical
Landcare, Key Papers Series 5, Centre for Rural Social Research, Charles Sturt
University, Wagga Wagga, 9-27.
5
Incentive Design for Introducing
Genetically Modified Crops
Ross Kingwe/l
Agricultural applications of the commercialization of gene technology

have increased rapidly in the 1990s (Riley and Hoffman, 1999). Adoption
of genetically modified (GM) crops has been rapid in the United States
and large areas are sown to GM crops in Brazil, China and Argentina.
For example, in the United States by 1998 approximately 38 percent of
the soybean acreage and more than 40 percent of the cotton area was
planted to GM varieties (Carpenter and Gianessi, 1999; USDA/ERS,
1999). In Canada, by 1998, GM varieties of canola accounted for 44 per
cent of the area planted to canol a (Fulton and Keyowski, 1999).
Most GM seed currently used by farmers offers benefits of pest and
weed control. Examples include Roundup Ready® soybean and corn, Bt
cotton and corn, Buctril herbicide resistant cotton and Liberty herbicide
resistant corn. The on-farm benefits of these crops include decreased
pest management costs, increased yields and greater crop production
flexibility, although these benefits vary across regions (Klotz-Ingram et
al., 1999).
To generate a commercial return on their R&D investment in develop-
ing and protecting gene technology most biotechnology companies are
licensing or contracting the use of their GM products. For example, Mon-
santo imposes contractual obligations on growers opting to use their GM
products. Growers are not allowed to retain seed. Growers must allow
Monsanto or its nominee access to the farm's management records and
access to the fields in which the GM crops are growing, in order to in-
spect and test those crops. This right of inspection lasts for up to three
years after the last planting of the GM crop. Further, in the case of
Roundup Ready® crops no other glyphosate chemical can be used other
than Roundup®. Growers are required to pay technology fees to the seed
company that in turn passes these to Monsanto in return for receipt of a
Incentive Design for Introducing GM Crops 83
handling fee. These technology fees or seed premiums are typically

subject to discounting based on early purchase, volume discounts and
package deals for other seed or chemical products sold by the same
company (Hayenga, 199~.
However, this revenue received by gene technology developers is at
risk of "piracy" by potential users. This piracy can stem from other gene
technology developers illegally obtaining information and genetic prod-
ucts that are then incorporated in competitive R&D activity (Barton,
1998). The piracy can also come from growers using GM seed illegally.
Lindner (1999), for example, indicates Monsanto, as at February 1999,
had full-time Pinkerton investigators dealing with 525 cases of suspected
infringement and their workload was increasing. His understanding was
that the costs of enforcement would far outweigh payments for settlement
of proven infringement. A related comment by Wright (1996) is that: "In a
decentralized competitive farming sector, policing of replanting by farm-
ers seems to be a challenge. Private wheat seed markets are reported to
thrive only in parts of the United States where farmers have no on-farm
storage." (p. 573).
Policing piracy is necessary for commercial as well as legal reasons. It
needs to be cost-effective with the risk of detection and prosecution of
piracy being sufficiently large to protect the profits of the companies
rightfully selling the GM seed and associated crop inputs. Policing piracy
may also be a necessary part of supply-chain management, to ensure
identity preservation of GM and non-GM crops. Consumer and producer
concerns surrounding the food and environmental safety of GM crops is
requiring increased investment in the integrity of supply-chains to ensure
identity preservation (Kalaitzandonakes and Maltsbarger, 1998). Also
community concerns about GM crops and GM foods is causing many
governments to review their GM policies and to increase the regulation of
gene technology. For example, the Gene Technology Act in Australia
empowers a Gene Technology Regulator (ADHAC, 2001) to license, in-
spect premises, search and seize, monitor, enforce and prosecute
breaches of the Act.
This chapter develops a Simple model of producer behavior regarding
the availability of a GM crop and examines the role of incentive design in
influencing farmer adoption of the GM crop. The model is used to illus-
trate the importance of illegal and improper use of GM technology that
represents a leakage of technology fees to gene technology developers.
The approach in this chapter is drawn from studies of compliance to envi-
ronment schemes. In particular, this study initially follows the approach
84 Ross Kingwell
of Latacz-Lohmann and Webster (1999) who examined non-compliance

in agri-environmental schemes in Europe. Their approach is extended to
consider yield risk and risk aversion and is applied to GM crops.
5.1. A Model of Producer Response to GM Crops

To grow a GM crop typically requires a farmer to agree to license or con-
tract obligations that oblige the farmer to undertake a series of actions
and purchases as part of their production of the GM crop. Often farmers
are required to participate in a closed marketing loop whereby they sell
all grain harvested from the GM crop to a single firm. Often they are re-
quired to use particular chemicals at particular times and to adopt par-
ticular management practices such as planting buffer or refuge crops.
This set of contractual activities can be represented as n activities form-
ing the set A where A= {at, a2, , an}. The annual cost of this set of ac-
tivities, in some cases offset through receipt of cost-savings (e.g. less
herbicide or pesticide used), can be stated as Ca , and the income asso-
ciated with sale of the GM crop is Va.
Occasionally the management records and practices of a farmer may
be investigated to ensure the farmer complies with required practices.
Violation of the contractual agreement can be represented as a set of
activities C where C = {c t, c2" cm}. This differs from set A. The prob-
ability of violation detection can be represented by p(V) and the penalty
for violation, as specified in the license agreement, is V. If use of GM
seed is governed by contract law then legal judgments regarding contract
violations and liabilities will specify V. In practice, V could be a fixed fine
or some function of the revenue or profit from growing the GM crop (e.g.
V=f(Ya-CaJ)·
Employing the terminology of Latacz-Lohmann and Webster (1999),
farmer behavior can be modeled as amoral calculation. Assuming a
farmer is a risk-neutral amoral calculator indicates that the farmer's chief
interest is profit. The farmer will abide by or break agreements whenever
it is profitable. This assumption allows this behavioral extreme to be a
benchmark case.
Thus, the farmer's decision problem can be stated as maximizing profit
by selecting among the following choices:
Option 1: Legitimately adopting the GM crop and generating profit,

lrA.This requires utilising activity set A with lrA = Va-Ca·
Option 2: Not adopting the GM crop and generating profit, lra. To
generate this profit involves utilizing the activity set a where lra = Yb-Cb
Incentive Design for Introducing OM Crops 85
and B = {bt, b2, ... , brrJ. In this case set B includes activities required to
grow a traditional non-GM crop.
Option 3: Using the GM crop illegally. There are two main cases in
option 3. Firstly, a farmer may sign the contract to grow the GM crop yet
may knowingly or unwittingly not abide by all its terms and conditions.
This farmer's actions are represented by the activity set C where C = (Ct,
c2, , crrJ and the farm generates profit "C where "C = Yc-C c . The ex-
pected profit can be expressed as:
E("e) = lie - P(Vc)Ve (5.1 )
where, "C is the optimal profit generated by utilizing activity set C, the
penalty for scheme violation is Vc and the probability of detection is
P(VC)·
Secondly, a farmer may opt to not become a licensee yet the GM seed
is obtained and used illegally. In this case no contract would be signed
and the farmer's actions are activity set 0, the penalty for scheme viola-
tion is Vo and the probability of detection is p(Vo). Expected profit can
be expressed as:
(5.2)
Typically p(VC) > p(Vo) because a licensee, through contractual obli-

gations involving external monitoring and scrutiny, would be more likely
to have their contract breaches noticed than a farmer about whom a li-
censor would have no initial suspicion of illegal use of GM technology.
Practical evidence of P(VC) > p(VO) is the fact that measures such as
toll-free tip lines have accompanied large scale introduction of some GM
crops; in effect encouraging illegal users of GM crops to be identified by
members of their communities.
A licensor and, if applicable, a national GM regulator may also seek
greater legal and social redress from farmers who are not licensees and
who illegally grow the GM crops. For example, for GM canola in Canada,
Monsanto pays for radio advertisements that name farmers who have
been caught saving seed (Lindner, 1999). Also the Gene Technology Act
in Australia includes a feature of publishing the names of offenders (eg
those illegally growing GM crops).
In the United States some illegal users are prosecuted vigorously in
order that publicity about their cases acts as a deterrent to others. So, in
practice, it is likely that Vo > VC. However, for the purpose of illustration,
86 Ross Kingwell
the decision problem in Figure 5.1 portrays V as a linear function of Ya-

Ca and does not discriminate between licensees who act improperly and
those illegally acquiring GM seed and growing the GM crop.
As shown in Figure 5.1 if the net returns from legitimate adoption are
greater than Y"a-C"a then legal fully compliant adoption is the preferred
option for the risk neutral farmer. However, for returns in the range Y'a-
C'a to Y"a-C"a the farmer would prefer to either improperly or illegally
use the GM technology. In practice this might mean illegally obtaining or
retaining and using GM seed, falsifying records or failing to adhere to
various practices. Because p(VC) > P(VD) and given V is an assumed
linear function of Ya-C a , then in the range Y'a-C'a to Y*a-C*a the legal
yet improper use of the GM technology is preferred. From Y*a-C*a to
Y"a-C"a the illegal use of the technology is preferred.
No Improper Illegal Legal

adoption adoption adoption adoption
Farm
Profit
($)
P
Figure 5. 1: An Illustration of the Farmer's Decision Problem
For the farmer to accept the gamble that their violations will be de-
tected, as shown in Figure 5.1, then p(V) and V must be suffiCiently small
to provide the required incentive. If returns are less than Y'a-C'a then the
farmer would rather not adopt the GM crop.
The above decision problem highlights a few areas in which adoption
of GM crops by risk neutral farmers can be influenced and the illegal and
improper use of proprietary technology can be reduced. The options are
to increase p(V) or V or both. As shown in Figure 5.1 if V is a function of
Ya-Ca then increasing this difference will increase V. Increasing Ya-Ca
can be achieved in various ways. For example, a rigorous scrutiny of the
activity elements of set A may reveal better, fewer or cheaper ways to
grow successfully the GM crop and therefore reduce Ca. The size and
nature of the technology fees charged by the owners or licensors of the
GM technology is obviously an important component of Ca.
The proprietary technology may enable farmers to increase Ya , by
higher yields through better pest and weed control, better supply-chain
management and improved marketing. Assuming the increase in Ya is
also associated with increases in Ya-Ca and that V is positively related
to Ya-Ca as in Figure 5.1, then the greater size of V is a further disincen-
tive for illegal and improper use of the proprietary technology. Also, in
the future if price premia for GM crops arise, due to their quality im-
provements, then Ya may increase.
Increases in p(V) are possible through a range of measures such as
the licensor allocating more resources toward surveillance, rewarding
those who inform against illegal use of GM products and widely broad-
casting news about prosecutions. The purpose of such litigation would be
twofold; firstly to ensure the cost to a farmer of being detected (V) was
very high and secondly to publicize this cost and to create the impression
that the owners of the GM technology property rights were keen to detect
breaches of their proper use (Le. p(V) was not negligible). Further, in-
creases in p(V) may be possible due not to the actions of the owners of
the GM technology but rather due to the actions of either regulators or
purchasers of non-GM crops and crop products. To maintain consumer
confidence in the integrity of the non-GM status of their products some
purchasers may insist on testing the grain or product delivered to them,
thereby increasing the likelihood of detection of growers who use non-
GM marketing channels to sell their illegally grown GM crops. Failure of
growers to supply non-GM grain or product could result in fines or dock-
ages. Some government regulators may also engage in monitoring, in-
spection, policing and prosecution to safeguard community concerns
about the food and environmental safety of GM crops. Hence, with such
88 Ross Kingwell
activities p(V) and V could be sufficiently high to deter the illegal growing
of GM crops.
Increases in p(V) or V cause the lines rcc and rcD in Figure 5.1 to pivot
downwards from the points where they intersect the vertical axis (farm
profit). Eventually the farmer is restricted to choosing across a range of
Ya-Ca values to either legitimately adopt or not adopt the GM crop. Thus
it is possible to remove the problem of illegal and improper use of GM
technology by setting p(V) and/or V high enough. The effect of raising
p(V) and/or V would cause lines rcC and rcD in Figure 5.1 to eventually
pass through point P. Only at this point would the farmer be indifferent
between not adopting or legitimately adopting the GM crop.
In the preceding model the only risky decision involved the probabilistic
gambles of option 3. A risk-neutral farmer would avoid illegal and im-
proper use of GM technology if rcA > rcc and rcA > rcD. However, for a
risk-averse farmer contemplating adoption of GM crops a range of un-
certainties exist. The impact of such uncertainties will influence their
adoption decisions. For such a farmer to avoid illegal and improper use
of GM technology:
(5.3)
As shown in the appendix equation (5.3) may hold for a risk-averse

farmer but not reA > rec and reA > reo due to the possibility of there being a
greater variance of income associated with illegal use of GM crops. That
is, although expected profit from the legal use of a GM crop may be less
than expected profit from illegal use of the crop, a risk-averse farmer
may still prefer the legal use of the crop due to the dominating effect of
profit variance.
Hence, to reduce illegal and improper use of GM technology, there is a
range of factors to consider. Included are the relative profitability of
growing the GM crop, the probability of detection of improper or illegal
use of the GM crop, the probability of successful prosecution surrounding
such uses of GM seed, the severity of fines for fraud or contract
breaches and the risk attitude of the farmer. In the next section the rela-
tive importance of these factors is illustrated using an Australian example
of INGARD® cotton.
5.2. A Numerical Illustration: INGARD® cotton
The problem of illegal and improper use of GM technology and the re-
sponses to it can be illustrated using the case of INGARD® cotton grown
in Australia. The parameter values used in the numerical analysis are

outlined in Table 5.1.
Table 5. 1: Parameter Values for the Numerical Analysis
Conventional INGARD Improper Illegal

® use use
Cost of $/ha 541 384 354" 354"
production
Technology fee $/ha 155 155
Yield t/ha 1.516 1.526 1.526 1.526
variance 0.0231 0.0246 0.0246 0.024
Cotton price c/kg 230 230 230 230
Probability 0.3 0.06

of detection
Severity of fine $/ha 1000 2500
Note: a Excludes any fine associated with detection of improper or illegal use. This es-
timate assumes some cost-savings by farmers through to use of cheaper inputs and
avoidance of some management costs (eg provision of refuges).
Sources: Yield data came from Table 58, NSW lint yield 1984 to 1997, Australian Com-
modity Statistics 1998 (ABARE 1998). Price data came from Table 62, Australian raw
cotton prices 1984 to 1997, Australian Commodity Statistics 1998 (ABARE 1998). Pro-
duction costs are based on published farm surveys of Australian cotton growers (Pyke,
1998; Clark and Long, 1998). All prices and costs are in Australian dollars.
Complementing these parameter values in Table 5.1 is the risk attitude

of the farmer. A farmer's attitude to risk can be represented by the mean-
variance formulation of expected utility. See Hanson and Ladd (1991) for
arguments supporting this approach:
E(U( 1f)) = U(E( 1f )) + !...U"(E( 1f ))Yar( 1f) (5.4)

2
where, U(tr} is the utility function of profit and U'(tr}>O and U"(tr}<O.
Following Fraser (1991), the farmer's utility function of profit can be
represented by the constant relative risk aversion form:
U( 1f ) = 1fl-r /(1- R) (5.5)
where, R = -U"( 1f)1f /U'( 1f), and R is the farmer's coefficient of relative
risk aversion.
90 Ross Kingwell
Table 5.2 shows the expected profit, variance of profit, expected utility
and farm management decisions for a range of risk attitudes, given the
parameter values in Table 5.1. The results in Table 5.2 are the base
case findings. The results show the importance of risk attitude in influ-
encing farmer behavior. Risk neutral or slightly risk averse farmers would
use illegally the GM crops. Moderate and highly risk averse farmers
would accept and abide by INGARD® technology agreements. In adopt-
ing this technology this latter group of farmers would experience lower
expected profits but less profit variance. The switch in farmer behavior
from preferring illegal use to lawful adoption of the INGARD® technology,
as risk aversion increases, illustrates the potential dominating effect of
profit variance as outlined in the Appendix.
Table 5.2: Base Case Findings
Management Choices
Conventional INGARD® Improper Illegal Optimal

use use decision
E(Jf) 2946 2989 2701 3006
Var (Jf) 122199 130134 340134 482634
E(U( Jf)) 1472 1491 1359 1496 Illegal use

(R=O.1)
use is
(R=O.5) 108.6 109.1 103.3 108.9 INGARD®
slightly preferred
INGARD® use is
(R=O.9) 22.23 22.25 21.99 22.22
slightly preferred
In the model of producer response outlined in the previous section,

there are various factors that influence farmer behavior regarding the
adoption of GM crops. Sensitivity analysis reveals how these factors af-
fect the illegal and improper use of the GM technology by farmers. For
example, for the base case parameters, the results in Table 5.2 show the
problems of illegal and improper use decrease with increasing risk aver-
sion. Results in Table 5.3 also show the role of the probabilities of detec-
tion, the severity of fines, the cost of the INGARD® technology agree-
ments and the cotton price in influencing producer reactions to the
INGARD® technology.
Table 5.3: Sensitivity Analysis of Problem of

Illegal Production of GM Crops
Switching Unit Risk attitude Difference

values of a from base case
(R=0.1 ) (R=0.5) (R=0.5) (R=0.1 ) (R=0.5) (R=0.9)
Fine for illegal use $/ha 2671 2354 2149 171 -146 -351
Probability of no. 0.064 0.056 0.050 0.004 -0.004 -0.010
detection of
illegal use
Technology fee $/ha 144 167 190 -11 12 35
Cotton price c/kg 231 229 228 -1 -2
a These are the parameter values that cause £(U(1rA» = £(U(1rD» in equation (5.3).
In words, the farmer is indifferent between accepting and abiding by the terms of the
INGARD® technology agreement or using the technology illegally.
For producers with R=O.1, only relatively small changes in the prob-
abilities of detection, the severity of fines, the cotton price and the cost of
the INGARD® technology agreements are needed to overcome the
problem of illegal use. However, it also is worth noting that for producers
with R=O.5 or R=O.9, similarly small changes in the probabilities of de-
tection, the severity of fines, the cotton price and the cost of the
INGARD® technology agreements can stimulate illegal use of the tech-
nology.
The results in Table 5.3 point to only modest changes in one or a com-
bination of the following being necessary to form an incentive-compatible
policy to address the problems of illegal and improper use: the probability
of detection, the fine severity and the cost of the INGARD® technology.
Although the cotton price will also influence producer behavior, in prac-
tice it is unlikely to be accessible to control.
To-date Monsanto has shown its preparedness to adjust its INGARD®
technology fee. Initially in Australia the fee was set at $245 per hectare.
Subsequently the fee was lowered by $35 per hectare and in 1998 the
fee was further adjusted to be effectively $155 per hectare. These
changes in the technology fee were not a response to any perceived
problems of improper or illegal use of the INGARD® technology. Rather
the fee reductions were in response to growers' mixed experience con-
cerning the profitability of the technology (Clark and Long, 1998; Pyke,
1998). However, it could be inferred that such reductions in the technol-
92 Ross Kingwell
ogy fee would have lessened the likelihood of the problems of illegal and
improper use alluded to in this chapter.
Most studies of the risk attitude of Australian farmers (Bardsley and
Harris, 1987, 1991; Abadi, 1999) reveal a range of risk attitudes, with the
majority being identified as slight to moderately risk averse. If a similar
range applies to cotton farmers then the problems of illegal and improper
use could be addressed by relatively small changes in factors such as
the technology fee and the severity of fines.
5.3. Conclusions and Limitations

The increased commercialization of gene technology is giving rise to a
greater array of GM crops. As pOinted out in this chapter, there is a po-
tential problem of illegal and improper use associated with the release
and adoption of GM crops in farming communities. To generate a com-
mercial return on their R&D investment most biotechnology companies
are licensing or contracting the use of their GM technology. The revenue
received by these companies however, is at risk of "piracy" by potential
users.
In this chapter two forms of grower piracy are considered; growers who
use GM seed illegally and growers who fail to comply with technology
agreements and thereby lessen the revenue flow to the licensor. A model
of producer response to GM crops is developed that outlines these
problems and the factors influencing them.
This model is illustrated using the example of INGARD® cotton grown
in Australia. Ingredients of the mechanism design to reduce the problems
of illegal and improper use of GM technology are illustrated. Results
show that these problems are likely to increase with decreasing risk
aversion. As risk aversion decreases there is a switch in farmer behav-
iour away from the lawful adoption of the INGARD® technology, due to
the declining influence of profit variance. Results point to only modest
changes in the probability of detection of illegality, the fine severity
and/or the cost of the INGARD® technology being necessary to form an
incentive-compatible policy to address the problem of illegal and im-
proper use. Although such changes can address this problem, this paper
has not extended the analysis to consider the costs to the biotechnology
companies of implementing these changes, in spite of the changes po-
tentially being small. The producer model that underpins the analysis
also has not been extended to consider other sources of uncertainty and
differences between farms (apart from risk attitude). Further the model
does not account for the ease with which the GM technology can be ob-
tained illegally or used improperly. In the case of cotton farming, farmer-

saved seed is not the norm so policing cotton seed sales is fairly simple.
However for some other crops, use of farmer-saved seed is common so
the opportunity to illegally obtain or retain GM seed might be greater and
therefore the costs of policing could be higher. Although the findings in
this paper specifically relate to one case study, nonetheless the model
could be applied to other crops in other regions.
Appendix: Income Variance Associated with

the Illegal Use of a GM Crop
Consider the impact of yield risk upon the issue of moral hazard for a
risk-averse farmer. Such a farmer's profit per hectare from illegal use of
a GM crop is:
;rND = py - c, where p is the fixed price of grain, y is the uncertain crop
yield and c is the fixed per hectare cost of production and no detection of
illegal use occurs and,
;r D = py - c - f , where detection occurs and a fine, f, is imposed.
Given the probability of detection is r then the farmer's expected profit

from illegal use is:
E(;r) = py-c-rf (1-1 )
and the variance of profit is:
Var(;r) = f Y (l-r)«py-c)-E[;r])2 f(y)dy+fY r«py-c- f) -E[;r])2 f(y)dy (1-2)
After substituting for El;r} using equation (1-1), equation (1-2) can be
re-expressed and simplified to:
(1-3)
For a farmer not engaged in any illegal use of a GM crop their ex-
pected profit is:
94 Ross Kingwell
where, p is the fixed price of grain, y is the uncertain crop yield and d is
the fixed per hectare cost of production.
Their profit variance is:
Var(nr) = p2Var(y) (1-4)
Note that comparing equations (1-3) and (1-4) reveals that:
p2Var(y)+rj2(l-r) > p2Var(y) duetoO<r< 1 andf>O.
Hence the profit variance associated with the illegal use of the GM crop
exceeds that for the legal use. Thus, even if the expected profit from the
legal use of the GM crop does not exceed that from its illegal use, a risk-
averse farmer may still prefer the legal use of the crop because of the
dominance of income variance in the farmer's selection decision.
References
Abadi Ghadim, A.K. (1999), Risk, uncertainty and learning in farmer adoption of a
crop innovation. Unpublished PhD thesis, Faculty of Agriculture, University of
Western Australia.
ADHAC (2001), Gene Technology Act 2000, Australian Department of Health and
Aged Care, at http://scaletext.law.gov.au/html/pasteact/3/3428/top.htm, Oct 12,
2001.
Bardsley, P. and M. Harris (1987), "An approach to the economic estimation of atti-
tudes to risk in Australia", Australian Journal of Agricultural Economics, 31 (2): 112-
126.
Bardsley, P. and M. Harris (1991), "Rejoinder: An approach to the economic estima-
tion of attitudes to risk in Australia", Australian Journal of Agricultural Economics,
35(3): 319.
Barton, J.H. (1998), "The impact of contemporary patent law on plant biotechnology
research" in SA Eberhart et al. (Eds) Intellectual property rights III, global genetic
resources: access and property rights, pp. 85-97, Madison, WI:CSSA.
Carpenter, J. and L. Gianessi (1999), "Herbicide tolerant soybeans: Why growers are
adopting Roundup Ready varieties", AgBioForum, 2(2):65-72. Retrieved July 1999
from the World Wide Web: http://www.agbioforum.missouri.edu.
Clark, D. and T. Long (1998), "The performance of Ingard® cotton in Australia in the
1997/98 season", Cotton R&D Corporation Occasional Paper, Narribri, New South
Wales, pp. 51.
Fraser, R.w. (1991), "Price-support effects on EC producers", Journal of Agricultural
Economics, 42(1):1-10.
Fulton, M. and L. Keyowski (1999), "The producer benefits of herbicide-resistant
canola", AgBioForum, 2(2):85-93. Retrieved July 1999 from the World Wide Web:
http://www.agbioforum. missouri. edu
Hanson, S.D. and G.w. Ladd (1991), "Robustness of the mean-variance model with
truncated probability distributions", American Journal of Agricultural Economics,
73(2): 436-45.
Hayenga, M. (1998), "Structural change in the biotech seed and chemical industrial
complex", AgBioForum, 1(2), 43-55. Retrieved January 1, 1999 from the World Wide
Web: http://www.agbioforum.missouri.edu.
Klotz-Ingram, C., S. Jans, J. Fernandez-Cornejo, and W. McBride (1999), "Farm-level
production effects related to the adoption of genetically modified cotton for pest
management", AgBioForum 2(2):73-84. Retrieved July 1999 from the World Wide
Web: http://www.agbioforum.missouri.edu
Kalaitzandonakes, N. and R. Maltsbarger (1998), "Biotechnology and identity-
preserved supply chains", Choices, Fourth Quarter 1998:15-18.
Latacz-Lohmann, U. and P. Webster (1999), Moral hazard in agri-environmental
schemes, Mimeo, Agricultural and Resource Economics Group, University of
Western Australia.
Lindner, R.K. (1999), Prospects for public plant breeding in a small country. Paper
presented at the ICABR conference on The shape of the coming agricultural bio-
technology transformation: strategic investment and policy approaches from an
economic perspective at the University of Rome "Tor Vergata" Rome and Ravello,
June 17-19,1999.
Pyke, B. (1998), "Ingard survey results for the second year", The Australian Cotton-
grower, 19(6): 36-39.
Riley, PA and L. Hoffman, (1999) "Value-enhanced crops: biotechnology's next
stage"; Agricultural Outlook, March 1999, pp. 18-23.
United States Department of Agriculture (USDA), Economic Research Service (ERS)
(1999), Genetically engineered crops for pest management. Retrieved June 1999
from the World Wide Web: http://www.usda.gov/whatsnew/issues/biotech
Wright, B.D. (1996), Agricultural genetic research and development policy, Conference
proceedings of the Global Agricultural Science Policy for the 21 st Century, Mel-
bourne, pp. 559-580.
6
An Economic Risk Assessment of the
Impact on Producers of Removing
Quarantine Restrictions
Robert W. Fraser
One of the key outcomes of the Nairn Review of Australia's quarantine

policies was the proposal to establish "expert Working Parties to com-
plete specific components of a detailed risk analysis" (Nunn, 1997, p.
567). In particular, there would be "Scientific Working Parties conducting
detailed risk assessments and considering risk management options, and
Economics Working parties examining the potential economic loss due to
the introduction or establishment of any pests or diseases" (Nunn, 1997,
p. 567). On this basis, there is a clear distinction maintained in the con-
text of quarantine policy between the tasks of "risk assessment" and
"economic assessment".1 Moreover, this distinction has been perpetu-
ated to a large degree by the existing economic literature relating to the
assessment of quarantine policies. For example, the economic assess-
ment of Newcastle Disease in chickens by Hafi, Reynolds and Oliver
(1994) is conducted almost exclusively in terms of the impact of imported
chickens and the disease on producer and consumer surplus. Only in the
final chapter is it acknowledged that as estimated "the total cost of dis-
ease is only valid under strong assumptions of transition from a zero
probability of a disease state with quarantine to one of certainty (with-
out)" (p. 45). Hafi et al. (1994), then, proceed to consider briefly the role
of the probability of a disease outbreak in determining the relationship
between (expected) costs and benefits as measured by the changes in
producer and consumer surplus. But at no pOint is an assessment of the
impact of imports and the disease on a domestic producer's perception of
income risk even contemplated. Similarly, James and Anderson (1998)
undertake an economic assessment of Australia's quarantine policy of an
import ban on bananas using estimated changes in producer and con-
Economic Risk Assessment of Quarantine Restrictions 97
sumer surplus, thereby excluding any assessment of the impact of re-

moving the policy on the income risk of producers. 2 3
This focus of the existing economics literature on the producer surplus
(expected income) effects of quarantine policies is probably justified as a
first-order approximation. But, at the same time, it is not appropriate to
disregard economic consideration of the income risk effects on producers
- especially when risk assessment is the central focus of scientific con-
sideration of quarantine policy.
The aim of this chapter is to modify the distinction between "economic
assessment" and "risk assessment" by developing the understanding of
the impact of removing quarantine restrictions on the income risk of do-
mestic producers. 4 In so doing, it will show that removing an import ban
has four separate effects on the income risk of producers: two of which
are favorable, and two of which are unfavorable. Moreover, because of
this conflict of effects it will also show that a producer's perception of the
overall effects on income risk of removing an import ban may be favor-
able or unfavorable depending on a range of factors largely beyond the
control of the producer. Nevertheless, a characterization of the role of
these factors will help identify situations in which producers can be ex-
pected to perceive the impact of removing quarantine protection on their
income risk as to their advantage or not.
The structure of the chapter is as follows. Section 6.1 outlines the
model used to represent the impact of removing an import ban on the
income stream of a producer. It also distinguishes the four effects of in-
troducing imports on income risk and identifies the range of factors,
which govern a producer's perception of the overall effect of imports on
income risk. Section 6.2 undertakes a numerical analysis of this model,
including an evaluation of the importance of each of the factors govern-
ing a producer's overall perception of the impact on income risk. The
chapter concludes with a discussion of policy implications.
6.1. The Model

The model of a producer's uncertain income per period (/) assumes that
this income is a product of uncertain price (p) and uncertain yield (y).
In the absence of imports, expected income (Eo(l)) is given by:
Eo(/) = E(poyo)
= poro + cov(Po,Yo) (6.1 )
where, Po = expected price in the absence of imports
Yo = expected yield in the absence of imports
98 Robert W. Fraser
= covariance between price and yield.
Note that in the absence of imports the relationship between price and
yield is assumed in what follows to be represented by a negative covari-
ance. This assumption is consistent with a closed economy framework.
In addition, the variance of income in the absence of imports (Varo(l))
can be approximated by, 5
Varo(l) .. y;Var(po) + p;Var(yo) + 2poyo cov(Po,Yo) (6.2)
where, Var(po) = variance of Po

Var(yo) = variance of Yo
It is further assumed, that yield uncertainty can be represented by the

mutiplicative form, 6
y = By (6.3)
where, E(B) = 1
Var(y) = y2Var(e)
Var(B) = variance of s.
In considering the impact of the removal of the import ban, it is antici-

pated that both the expected price (;51) and the variance of price
(Var( P1)) with imports will be lower:?
P1<po
Var(p 1 ) < Var(po) (6.4)
In addition, it is anticipated that the negative covariance between price

and yield will be eliminated: 8
COV(P1,Y1) = 0 (6.5)
Finally, expected yield in the presence of imports (Yl) may be lowered

by the impact of disease:
(6.6)
Note that the presence of disease is assumed not to affect the sea-
sonal factors governing the variability of yield:
Variance(B1)=Variance(Bo), (6.7)
and so combining equations (6.3) and (6.6) shows that:

Var(Y1) s Var(yo) (6.8)
On this basis, expected income in the presence of imports E1(1) is

given by:
E 1 (I) = PIYI (6.9)

and the variance of this income (Var1 (I)) can be approximated by:
Var1(1) '" prVar(Yl) + yrVar(pl) (6.10)
A comparison of equations (6.1) and (6.9) gives:
(6.11 )
Based on equations (6.4) and (6.6) the first term on the right-hand-side
of (6.11) is clearly negative while the second term is positive as speci-
fied. Consequently, equation (6.11) is analytically ambiguous, although it
is anticipated that in virtually all cases the magnitude of the first term will
dominate that of the second and that overall the producer's expected in-
come will be lower in the presence of imports than in its absence:
E 1(I) < Eo(l) (6.12)
A comparison of equations (6.2) and (6.10) gives
Var1(1)-Varo{l) =~iVar(YI~- P~V~(Yo)) + 0r Var (PI)-

Yo Var(po))- 2 PoYo cov(Po,Yo) (6.13)
On the basis of equations (6.4), (6.6) and (6.8), the first two terms on
the right-hand-side of (6.13) are negative, while the third term is positive
as specified. The first term represents the decline in income risk with
imports due primarily to the lower expected price, while the second term
represents the decline associated primarily with less price variability.
Note also that the magnitude of both these terms is larger if disease af-
fects expected yield (Yl < Yo)' In contrast, the third term represents the
increase in income risk, which follows the removal of the negative co-
variance between domestic price and yield associated with the introduc-
tion of imports. It follows that the overall impact of the introduction of im-
ports on the variance of income is analytically ambiguous.
Moreover, these three terms do not represent all of the components of
the overall impact of introducing imports on the producer's perception of
income risk. A fourth component can be seen by approximating the pro-
ducer's expected utility of income (E(U(I))) as follows: 9
100 Robert W. Fraser
E(U(/)) z U(E(I)) + !... U"(E(I)) . Var(/) (6.14)

2
The second term on the right-hand-side of equation (6.14) shows that a

producer's perception of whether the impact of the introduction of imports
has been to make income risk (Var(l)) "better" or "worse", depends also
on that producer's attitude to risk as represented by the second deriva-
tive of the utility function:
U"(E(/))
Note that this term is negative for a risk averse producer. Based on
equation (12), it is anticipated that the introduction of imports will de-
crease the producer's expected income. Consequently, the associated
change in U"(E(I)) depends on whether:
<
U'I/(E(I)) o (6.15)
>
The standard assumption in this context is that the third derivative of

the utility function is positive, which is consistent with empirically-
supported functional forms such as the constant relative risk aversion
form,10
t· R
U(I) =- (6.16)
1- R
where, R= coefficient of relative risk aversion.
On this basis, because of the anticipated decline in E(I) associated
with the introduction of imports, it follows that the (negative) term U"(E(I))
will be larger in magnitude, and that as a consequence the producer will
perceive (from a lower expected level of income) an unchanged level of
income risk as "worse".
In summary, it can be seen that there are four components to the over-
all impact of introducing imports on a producer's perception of income
risk. Three of these are contained in equation (6.13) and relate to direct
effects of introducing imports on the variance of income, while the fourth
is contained in equation (6.14) and relates to the indirect effect of the
associated decline in expected income on the perception of a given level
of income risk. Moreover, two of the components in equation (6.13) are
negative, suggesting an improvement in the producer's perception of in-
come risk, while the third component in this equation and the fourth re-
lating to the indirect effect suggest a worsening in the perception of in-
come risk. It follows that, as with the impact on the variance of income,
a producer's overall perception of the impact of introducing imports on

income risk is also analytically ambiguous.
Nevertheless, based on equations (6.13) and (6.14) it is possible to
identify a range of factors which are likely to playa role in determining
whether the producer's overall perception of income risk is worsened or
improved with the introduction of imports:
(i) the strength of the negative covariance between price and yield in
the absence of imports (cov(Po,Yo)). For a producer this strength will
depend on the correlation between individual and aggregate production
(ii) the domestic elasticity of demand for the product. This elasticity
will determine the extent to which yield fluctuations in the absence of
imports generate domestic price fluctuations, with a more elastic demand
associated with less price variability (Varo(p))
(iii) the level of world price variability (Var(P1)). This level will deter-
mine the extent to which domestic price variability is reduced by the in-
troduction of imports
(iv) the relative level of the average domestic price in the absence of
imports to the average world price. This level will determine the extent to
which a producer's expected price decreases with the introduction of im-
ports (Po/PI)
(v) the level of yield variability associated with uncertain seasonal
conditions (Var(8)). This factor, together with the expected yield deter-
mine the variance of yield (Var(y))
(vi) the impact of disease on expected yield (yl!yo ). This factor will
determine the extent to which both the producer's expected level and
variance of yield declines with the introduction of imports
(Var(Y1)Nar(yo))
(vii) the producer's attitude to risk (UI/(E(I))). This factor, as explained
previously, will govern the extent to which the anticipated decline in ex-
pected income associated with the introduction of imports affects the
producer's perception of income risk.
The next section reports the findings of a numerical analysis of the
model developed in this section, including an evaluation of the role of
these seven factors in determining whether a producer's perception of
income risk is improved or worsened with the introduction of imports.
6. 2. Numerical Analysis
In order to undertake a numerical analysis of the model developed in
Section 1 it is necessary to specify a particular form for the producer's
utility function as well as values for the parameters of the model. In what
follows use is made of the constant relative risk aversion form of utility
function referred to in the previous section:
IloR
U(I)=- (6.17)
1- R
In addition, the following parameter values are assumed to represent

a Base Case before the introduction of imports:
Po = 10
a Po = 3 (Var(po) = 9)
a() = 0.3 (Var(8) = 0.09)
Yo = 100
Following the introduction of imports the Base Case values are as-
sumed to be:
PI = 6.1
a PI = 0
YI = 100
Note that PI has been specified at a level such that 90 percent of the
probability distribution of Po would exceed it. This assumption is subject
to a sensitivity analysis in what follows, as are the assumptions of zero
world price variability (a PI = 0) and zero impact of disease on expected
yield (YI = 100). Other sensitivity analysis relates to the assumed val-
ues for a Po and as·
Table 6.1 contains details of the Base Case results of producer per-
ception of the impact of introducing imports on income risk for a range of
attitudes to risk (R) and of correlation coefficients between price and
yield in the absence of imports. 11 Note that the value of this coefficient
is based on the specification: 12
(6.18)
In addition a perception index of improvement or worsening of income
risk is specified as follows:
U(Eo{l»)- U(EI{l») >

1 (6.19)
Eo{U{IJ)- EI(U(I1)) <
This specification is derived by combining and rearranging equation

6.14 for the pre and post import situations. 13
Table 6. 1: Base Case Results of the Perception of the

Impact of Introducing Imports on Income Risk
Perception Index b
(1 ) (2) (3) (4)
a c d
E\(1X vaf\(-}{ R = 0.8 R = 1.6
Eo (1) Varo(1)
Correlation
Coefficient (p)
-0.25 0.62 0.25 1.06 1.03
-0.5 0.64 0.37 1.02 0.97
-0.75 0.65 0.74 0.97 0.90
Notes: a: Var1 (I)Naro(i) = 1 for p = -0.81

b: Perception of Improvement: > 1 Perception of Worsening: <1
c: No change in perception: p = -0.57
d: No change in perception: p = -0.37
The results in Columns (3) and (4) of Table 6.1 show that, in general
terms, a producer is more likely to view the impact of introducing imports
on income risk as an improvement the smaller the magnitude both of the
(negative) correlation coefficient (p) and of the producer's risk aversion
coefficient (R). The first of these findings can be explained by reference
to the diminished role of the negative covariance as a pre-import stabi-
lizer of income for lower values of p. Consequently, the reduction in in-
come risk associated with imports is more powerful for smaller p as
shown by the results in column (2) of Table 6.1. Moreover, the second
finding is explained by the indirect impact of reduced expected income
(see column (1)) on the perception of income risk (the U"(E(/)) term in
equation (6.14)). In particular, the increase in magnitude of U"(E(/))
caused by the decline in E(I) is smaller for a less risk averse producer.
This means that a given decline in E(I) and Var(l) is more likely to be
perceived as representing an improvement in income risk by a less risk
averse producer. Finally in relation to Table 6.1, it should be noted that
for all values of p between -0.57 and -0.81 for R :;: 0.8, and between -0.37
and -0.81 for R :;: 1.6, the producer's perception of income risk is that it
has worsened with the introduction of imports, despite the fact that:
Var1(1) < Varo{l) (6.20)

for all these values of p.14
Consider now a sensitivity analysis of the Base Case assumptions with
a view to further evaluating the role of the factors outlined at the end of
Section 6.1. Table 6.2 contains details of the results for a more elastic
domestic demand as represented by a smaller level of pre-import price
variability ((JPo = 2 rather than 3).
Table 6.2: Perception of the Impact of Introducing

Imports on Income Risk: More Elastic Demanda
Perception Index
(1 ) (2) (3)
b c d
var\(% R = 0.8 R = 1.6

Varo(J)
Correlation Coefficient (p)
-0.25 0.49 1.03 0.99
-0.5 0.70 0.98 0.92
-0.75 1.23 0.94 0.85
Notes: a: (J Po = 2 (Base Case: (J Po = 3)

b: Var1 (1)/Varo(l) = 1 for p = -0.67
A comparison of column (2) in Table 6.1 with column (1) in Table 6.2
shows that this change results in a weaker tendency for the introduction
of imports to reduce the variance of income. In particular, for a producer
with a value of R = 0.8 and a value of p = -0.5, the introduction of imports
in this situation is perceived as worsening income risk whereas it was
perceived as improving income risk for the situation of less elastic de-
mand in Table 6.1. Moreover, Table 6.2 shows that, compared with Table
6.1, for producers with either level of risk aversion, there is a smaller
range of values of p for which a decrease in the variance of income is
perceived as an improvement in income risk.
Next consider a situation of greater seasonal variability (00)' Table 6.3
contains details of results for a value of 08 = 0.5 (Base Case: 08 = 0.3).
Table 6.3: Perception of the Impact of Introducing Imports

on Income Risk: Greater Seasonal Variabilitya
Perception Index
(1 ) (2) (3)
b c d
varl (%
Varo(I) R =0.8 R =1.6
Correlation
Coefficient (p)
-0.25 0.35 1.06 0.94
-0.5 0.49 0.99 0.86
-0.75 0.81 0.91 0.76
Notes: a: 06 = 0.5 (Base Case: 06 = 0.3)

b: Var1 (1)/Varo(l) = 1 for p = -0.82
A comparison of the results in Tables 6.2 and 6.3 shows that the im-
pact of greater seasonal variability is similar to that of a more elastic de-
mand. In both cases the change results is a weaker tendency for the in-
troduction of imports to reduce the variance of income. As a conse-
quence, in Table 6.3 there is for both levels of risk aversion a smaller
range of values of p for which a decrease in the variance of income is
perceived as an improvement in income risk. The essential difference
between the two cases is that, whereas for a more elastic demand there
is a smaller absolute gain in terms of reduced price variability, for greater
seasonal variability there is a smaller gain in reduced price variability
relative to total income variability.
Table 6.4 contains details of the results of a sensitivity analysis of the
impact of import-induced disease on expected yield. The Base Case as-
sumed that the introduction of imports had a zero impact on expected
yield (jil = 100).
The results in Table 6.4 are based on an assumption that disease in-
troduced by imports causes a 15% decrease in expected yield (YI =
85) .15 Compared with columns (1) and (2) in Table 6.1, columns (1) and
(2) of Table 6.4 show a more powerful impact of imports both on ex-
pected income and the variance of income. The first of these impacts
strengthens the negative perception of the impact of imports on income
risk (via U"{E(J))) while the second strengthens the positive perception by
directly reducing the variance of income. As a consequence, there is to a
large extent a neutralizing of these conflicting effects, and the general

pattern of results in columns (3) and (4) of Table 6.4 is similar to that for
Table 6.1.
Table 6.4: Perception of the Impact of Introducing Imports on

Income Risk: Disease Reduces Expected Yielda
Perception Index
(1 ) (2) (3) (4)
b c d
Var, (1)/ R =0.8 R = 1.6
/Varo (1)
Correlation
Coefficient (p)
-0.25 0.53 0.18 1.04 1.01
-0.5 0.54 0.27 1.01 0.97
-0.75 0.56 0.54 0.98 0.92
Notes: a: y, = 85 (Base Case: Yl = Yo = 100)

b: Var1 (I)/Varo(l) = 1 for p = -0.87
More specifically, there is no change in the range of values of p for

which a producer with R = 0.8 perceives the impact of imports on income
risk to be favorable (ps - 0.57), while there is a small reduction in this
range for R = 1.6 suggesting a dominance of the indirect effect on
U"{E{I)) in this case. But perhaps more significantly, the results of this
sensitivity analysis suggest that an absence of precise information about
the likely impact on expected yield of disease associated with introducing
imports is unlikely to affect significantly the extent to which producers
perceive imports as improving or worsening income risk. This finding
contrasts powerfully with the significance of this information in determin-
ing the impact on expected income.
Table 6.5 contains details of the results of a sensitivity analysis of the
assumed level of world price variability {Var (Pt)). The Base Case as-
sumes zero world price variability whereas the results in Table 6.5 are
based on a coefficient of world price variation of 10% (CT PI = 1). The re-
sults in this table are very similar to those in Table 6.2 and reflect a
similar situation of a smaller absolute reduction in income variability at-
tributable to a reduction in price variability. It follows that, the more vari-

able is the world price of the product in question, the less likely a pro-
ducer is to perceive the introduction of imports as improving income risk.

on Income Risk: Greater World Price Variabi/itya
Perception Index
(1 ) (2) (3)
(%
b c d
vali Varo(l)
R = 0.8 R = 1.6
Correlation Coefficient (p)

-0.25 0.32 1.03 0.98
-0.5 0.48 0.99 0.92
-0.75 0.97 0.95 0.86
Notes: a: Var{P1) = 1 (Base Case: Var{P1) = 0)

b: Var1 (1)/Varo(l) = 1 for p = -0.76
Finally, Table 6.6 contains details of the results of a sensitivity analysis

of the assumed expected level of world price.

on Income Risk: Higher Average World Pricea
Perception Index
(2) (3) (4)
b c d
Vali(l)/
/Varo(l)
R = 0.8 R = 1.6
Correlation
Coefficient (p)
-0.25 0.77 0.45 1.06 1.04
-0.5 0.78 0.67 0.99 0.93
-0.75 0.80 1.34 0.91 0.80
Notes: a: PI = 7.48 (Base Case: PI = 6.1)

b: Vaq (I)/Varo{l) = 1 for p = -0.67
c: No change in perception: p =-0.48
d: No change in perception: p =-0.34
Whereas the Base Case featured an expected world price which 90%
of domestic prices (in the absence of imports) would exceed, the results
in Table 6.6 are based on an expected world price (PI = 7.48), which
only 80% of domestic prices would exceed (in the absence of imports). A
comparison of columns (1) and (2) in Tables 6.4 and 6.6 shows that this
change has the opposite impact on the expected level and variability of
income with imports to that of a reduction in expected yield associated
with disease. As a consequence, conflicting impacts are once again cre-
ated on the producer's perception of income risk. But, unlike the results
in Table 6.4, the results in Table 6.6 show a clear distinction from those
in Table 6.1 at both levels of producer risk aversion. In particular, the
range of values of p for which an improvement in income risk is per-
ceived decreases for both types of producers. Therefore, this finding
suggests that the lower is the expected world price relative to the ex-
pected domestic price in the absence of imports, the more likely it is that
producers will perceive an improvement in income risk with the introduc-
tion of imports.
6.3. Conclusion
The aim of this chapter has been to develop understanding of the impact
of removing quarantine restrictions on the income risk of domestic pro-
ducers. In so dOing it is intended that the existing distinction between
"economic assessment" and "risk assessment" in the analysis of quaran-
tine policy be diminished.
Section 6.1 developed a model of the impact of removing import re-
strictions on a producer's perception of income risk. This model devel-
opment identified four separate effects of the introduction of imports on
the perceived income risk of domestic producers, as well as a range of
factors, which influence the magnitude of the various effects. Section 6.2
undertook a numerical analysis of the model developed in Section 6.1
and evaluated the role of each of the factors identified as influencing the
producer's perception of the impact of imports on income risk.
On the basis of this numerical analysis the following conclusions have
been reached. In particular, a producer is more likely to perceive the in-
troduction of imports as improving income risk:
(i) the weaker is the negative correlation between the producer's price
and yield in the absence of imports
(ii) the less risk averse is the producer
(iii) the less elastic is the domestic demand for the product in question
Economic Risk Assessment o/Quarantine Restrictions 109
(iv) the lower is the producer's level of seasonal variability

(v) the lower is the level of world price variability of the product in question
(vi) the lower is the level of expected world price relative to the ex-
pected domestic price in the absence of imports of the product in
question.
Finally, the numerical analysis showed that, in contrast to its signifi-
cance in determining the impact on expected income, information about
the impact on expected yield of disease associated with introducing im-
ports is unlikely to affect significantly the extent to which producers per-
ceive the introduction of imports as improving or worsening income risk.
Endnotes
1 See also Tanner (1997) and Tanner and Nunn (1998).

2 Although, like Hafi et a/ (1994). James and Anderson (1998) acknowledge the
role of the probability of a disease outbreak (pp. 431-2) as a form of risk.
3 Such economic analyses are reminiscent of the "hidden gains and losses" lit-
erature in Australia relating to the use of price stabilization policies in the Austra-
lian wool industry (e.g. Campbell, Gardiner, and Haszler, 1980), where the focus
on producer and consumer analysis overlooked the essential point that a price
stabilization policy is ostensibly a device for reducing the income risk of produc-
ers (Newbery and Stiglitz, 1981).
4 The (favorable) risk implications for consumers are fairly straightforward. In
particular, the introduction of imports is likely to stabilize both the domestic price
and quantity available of the product in question, with imports making up for any
shortfall of domestic production due to disease or seasonal conditions.
5 See Mood, Graybill and Boes (1974, p. 180).
6 See Newbery and Stiglitz (1981, p. 65).
7 See James and Anderson (1998).

8 This specification is based on the assumption that domestic production is un-
related to world production. This assumption seems reasonable where world
production is very large relative to domestic production.
9 See Newbery and Stiglitz (1981, p. 90).
10 See Pope and Just (1991).
11 See Newbery and Stiglitz (1981, p. 105) for details of empirical evidence to
support these values of R.
12 See Mood, Graybill and Boes (1974, p.154).
13 More generally, an improvement or a worsening of income risk is character-

ized by:
U"(Edl)) . Vardl)
>
U "(EoCl)) . VaroCl)
<
where it should be recalled that the second term on the right-hand-side of
equation (6.14) is negative.
14 Note that for values of p nearer to -1 the role of the negative covariance as an
income stabilizer becomes dominant, and so its removal with imports results in
increased income variability (see equation (6.13)).
15 See Hafi et al. (1994).
References
Campbell, R., B. Gardiner, and H. Haszler (1980), "On the Hidden Revenue Effects of Wool
Price Stabilisation in Australia" Australian Journal of Agricultural Economics, 24(1 ):1-15.
Hafi, A., R. Reynolds, and M. Oliver (1994), Economic Impact of Newcastle Disease on the
Australian Poultry Industry ABARE Research Report 94.7, ABARE, Canberra.
James, S. and K. Anderson (1998), "On the need for more economic assessment of quar-
antine/SPS policies" Australian Journal of Agricultural and Resource Economics,
42(4):445-58.
Mood, A.M., FA Graybill, and D.C. Boes (1974), Introduction to the Theory of Statistics 3rd
Edition, McGraw-Hili, Kogakusha.
Newbery, D.M.G. and J.E. Stiglitz (1981), The Theory of Commodity Price Stabilization
Clarendon Press, Oxford.
Nunn, M. (1997), "Quarantine risk analysis" Australian Journal of Agricultural and Resource
Economics, 41 (4): 559-78.
Pope, R.D. and R.E. Just (1991), "On testing the structure of risk preferences in agricultural
supply analysis" American Journal of Agricultural Economics 73(3):743-8.
Tanner, C. (1997), "Principles of Australian quarantine" Australian Journal of Agricultural
and Resource Economics, 41 (4): 541-58.
Tanner, C. and M. Nunn (1998), "Australian quarantine post the Nairn Review" Australian
Journal of Agricultural and Resource Economics, 42(4): 445-58.
Part II
Case Studies
7
Risk Attitudes and Risk Perceptions
of Crop Producers in Western Australia
Amir K. Abadi Ghadim and David J. Pannell
It has often been suggested that risk and uncertainty have major influ-
ences on the rate of adoption of rural innovations (e.g. Lindner et al.,
1982; Lindner, 1987; Weisensel and Schoney, 1989; Tsur et al., 1990;
Leathers and Smale, 1992; Shapiro et al., 1992; Smale and Heisey,
1993; Feder and Umali, 1993). However, empirical evidence for or
against this view has been relatively scarce and weak. This is primarily
because very few of the many empirical studies of adoption of rural inno-
vations have attempted to include risk and uncertainty as explanatory
variables. Even the few existing attempts have tended to use crude proxy
variables, probably because of the practical difficulty of obtaining high
quality measures of the relevant variables, these being: farmers' percep-
tions of the impacts of the innovation on the levels of risk they face;
farmers' uncertainty about the innovation; and, farmers' attitudes to risk
and uncertainty.
A current extension program is encouraging crop farmers in Western
Australia to grow several grain legumes crops, or "pulses", that are new
to the region: chickpeas (Cicer arietinum) , faba beans (Vicia faba) and
lentils (Lens culinaris) (Hamblin, 1987; Siddique and Sykes, 1997). A
minority of farmers are trialling these crops and very few have reached
the stage of full adoption. Anecdotal evidence suggests that Australian
farmers view grain legumes as being more risky than other available en-
terprises, and that this is inhibiting their rate of adoption. Given this
situation and the fact that they are in the early stages of adoption, these
crops provided an ideal opportunity to conduct a study of the roles of risk
and uncertainty in the adoption process.
In that study the focus was on chickpeas to explore the risk attitudes
and risk perceptions of Western Australian farmers in the context of
114 Amir K. Abadi Ghadim and David J. Pannell
adoption of a new crop. It examined both the perceived riskiness of

chickpeas and the risk attitudes of farmers who are considering the
adoption of chickpeas. This chapter presents a subset of results from
that larger study, the theoretical basis of which was reported in Abadi
Ghadim and Pannell (1999). In this chapter we have focused on the dis-
cussion of some of the risk-related findings from the non-statistical
analysis of the data generated from a longitudinal survey of farmers who
participated in this project. The results of regression analysis of the sur-
vey data exploring causal factors in adoption decisions were reported in
Abadi Ghadim and Pannell (1998) and Abadi Ghadim et al. (1999). In
the remainder of the chapter we outline the relevant theory, describe
methods used to collect the data, present the results obtained, discuss
the results, and draw a number of conclusions.
7.1. Theoretical Framework
The choice to adopt an innovation is a decision that has to be made un-

der considerable risk and uncertainty. In the context of this study we re-
fer to risk as the outcome of a variable that is inherently and unavoidably
random. It may be possible to know for certain what the probability dis-
tribution of the variable is, but one cannot know what the value of the
next observation will be. Uncertainty refers to an individual's lack of
knowledge of a variable. An individual may have uncertainty about a
variable, even if that variable is actually non-random (i.e., non-risky)
(Hey, 1979; Anderson et al., 1977; Wright and Ayton, 1994; Hardaker et
al., 1997).
The distinction between the perceived riskiness of an innovation and
the decision-maker's attitudes or preferences regarding risk is also im-
portant. The perceived riskiness of an option is a subjective judgment
about how risky things are. It is distinct from the decision maker's atti-
tude to risk, which refers to the individual's personal preferences for or
against options that are more or less risky (Anderson et al., 1977; Lind-
ner, 1987; Smidts, 1990; Hardaker et al., 1997; Pannell et al., 2000).
7. 1. 1. Attitudes to risk
Studies of risk attitudes of Australian farmers (e.g. Bond and Wonder,
1980; Bardsley and Harris, 1987) have reported a range of risk attitudes,
with a majority of farmers being classed as risk averse. Risk aversion
implies that they are prepared to sacrifice some income in order to avoid
some risk or uncertainty. The most common method for representing and
Risk Perceptions in Western Australia 115
analyzing the impacts of risk aversion in decision-making is "expected

utility theory".
7. 1.2. Expected utility theory
A thorough review of the pros and cons of expected utility (EU) theory is
beyond the scope of this review. Anderson et al. (1977) and Hardaker et
al., (1997) provide detailed accounts of the theory and its application. In
this section, a brief review of the basic principles of this theory is pre-
sented in order to clarify how our methodology draws on this theory.
Under this theory, a decision such as adoption of an innovation in-
volves a choice from a set of alternative actions with consequences that
depend on the eventual outcome of risky or uncertain events. The deci-
sion-maker's choice depends on the subjective belief they hold about the
probabilities of these uncertain events as well as their preferences for the
possible consequences (Anderson et al., 1977).
EU theory is not the only available theory of behavior under risk. The
relevance of EU theory in the context of farm planning and adoption of
innovations has been debated widely in the literature (e.g. Machina,
1981; Schoemaker, 1982; Hertzler, 1997). However, models based on
EU continue to be used widely by agricultural economists, mainly due to
their tractability and the relative non-ambiguity of results produced by
them (Bar-Shira, 1992).
Abadi Ghadim and Pannell (1999) built a conceptual framework of
adoption under risk and uncertainty. They argued that while expected
utility theory is by no means a perfectly accurate description of real be-
havior, it is a useful tool in analytical studies of investment under risk. It
can be argued that, on pragmatic grounds, EU theory provides an inter-
nally consistent, and well developed framework which captures the key
aspect of risk averse decision making: it gives greater weight to down-
ward deviations in income than to upward deviations.
The concept of the certainty equivalents (CE) can be used to reflect in-
dividuals' risk preferences. The CE of a risky prospect is the smallest
sure amount that a decision maker would be willing to accept in ex-
change for the risky prospect (Hardaker et al., 1997). For a risk-averse
decision-maker, the CE is less than the expected value of the distribution
of returns of the risky prospect. The difference between the CE and the
expected monetary value of the risky prospect is termed the "risk pre-
mium" (Anderson et al., 1977).
Pratt (1964) and Arrow (1970) defined a coefficient that can be used to
classify the risk attitude of a decision-maker, reflecting the shape of his
116 Amir K. Abadi Ghadim and David 1. Pannell
or her utility function. According to the expected utility theory, the shape
of the utility functions of risk-averse, risk neutral and risk preferring indi-
viduals are respectively concave down, linear and convex down. The
Pratt-Arrow coefficient of absolute risk aversion is positive for risk-averse
persons and negative for risk-preferring individuals.
7.1.3. Influence of risk on adoption behavior
Some research evidence suggests that the allocation of land to innova-
tions declines with higher degrees of risk aversion (e.g. Feder, 1980;
Shapiro et al., 1992). Feder et at. (1985) reflected the view, which is
common amongst many in a large body of literature, that a risk-averse
decision-maker will take longer than a risk neutral person to decide to
adopt an innovation.
Lindner and Fischer (1981) refuted the generality of that claim. Their
analysis showed that, in some cases, risk aversion could speed up the
process of adoption. This may occur where the profitability of the innova-
tion is not highly correlated with that of the traditional or alternative en-
terprise that it potentially replaces. In this case partial adoption of the
innovation will reduce the variability of total profits simply because it be-
comes a form of portfolio diversification.
On the other hand, diversification can sometimes result in increased
risk. Specialization can result in greater efficiency and improved quality
in production and marketing of produce. This specialization can provide
some insurance against price and weather risks. Diversification may ac-
tually result in increased financial and business risk if it extends the
farmer beyond their competencies and knowledge base (Pannell et al.,
2000). In this case, the prospect of a drop in productivity from diversifi-
cation (e.g. adoption of an innovation) may discourage the farmer from
adopting as quickly as they might. Indeed, a farmer may be so highly
skilled at some technology that he or she faces substantial incentives to
persist with its use. Someone who is less skilled may find it optimal to
switch technologies regularly, and therefore enjoy long-run growth in
output (Jovanovic and Nyarko, 1996).
This argument can be extended to the situation where a divisible inno-
vation, such as a new species of a crop, is being considered by two
similar farmers who differ mainly in years of experience in farming. The
young farmer with relatively less experience in growing and managing
traditional enterprises (e.g. wheat) may find it more worthwhile to switch
than does an older farmer. This is because the older farmer, who usually
has more experience, would be very skillful and confident in growing the
conventional enterprises. For the older farmer the opportunity cost of the
conventional enterprise is higher than for the younger farmer.
There are also other studies that have found that farmers' perceptions
of riskiness of innovations influenced the probability and intensity of their
adoption (e.g. Smale et al., 1994). Several reviews of adoption studies
have highlighted the general scarcity of empirical studies that have ade-
quately addressed the role of risk and uncertainty in adoption. This scar-
city has been attributed to difficulties in observing and measuring risk
and uncertainty. It has been argued that this omission of risk has been a
serious limitation of many of the previous adoption studies leading to
poor model specification.
7. 1.4. Measuring risk
Norris and Kramer (1990) carried out a comprehensive review of tech-
niques for measuring risk and uncertainty in agriculture. They found that
the most reliable, accurate and acceptable method is to elicit subjective
judgments of individuals about probabilities of events that are important
to them.
Elicitation of subjective judgments requires the use of specific survey
techniques to gather information from individuals who make the adoption
decisions. Surveys that enable the researchers to collect panel data offer
the most suitable tool for this purpose. The use of panel data in adoption
research offers several major advantages over conventional cross-
sectional and time-series data sets. Longitudinal surveys can be de-
signed to detail individual farmer characteristics, preferences, percep-
tions and adoption choices made by them at each point in time (Hsiao,
1986).
Given the advantages of panel data sets we chose to collect the data
for this study using a longitudinal survey. Farmers were personally inter-
viewed at their property by trained interviewers using structured ques-
tionnaires. The purpose of the questionnaire was to obtain detailed in-
formation about the characteristics of the property and the farmer. In this
chapter we discuss only a subset of results drawn from the analysis of
the data from our survey.
The sample of farmers in this study was selected from the population of
growers in the central and eastern wheat belt of Western Australia. Most
of the farms in this region have a combination of crop and livestock en-
terprises on a mix of soils ranging from sands to clay soils (Kingwell et
al., 1995). Chickpeas, the crop innovation, can be grown on neutral to
alkaline loams and clays. These soils form a large proportion of the re-
gion (Stoneman, 1992; Lantzke, 1992).
7.1.5. Elicitation of subjective assessment of risk and uncertainty

The review of subjective probability techniques by Norris and Kramer
(1990) found that the direct approach to elicitation of probability density
function is superior to other techniques. They investigated the merits of
various methods used for this purpose and concluded that visual re-
sponse methods were the most reliable, accurate and practical. Visual
response methods use visual tools to aid the assessor in specifying a
probability density function. This conclusion has wide spread support in
the literature (e.g. Anderson et al., 1977; Hardaker et al., 1997).
With this technique, the researcher divides the range of possible out-
comes (e.g. yields of a crop over a 20-year period) into several incre-
ments or intervals. "Then, the assessor is asked to distribute a specific
number of counters over the different intervals in accordance with his or
her belief of the occurrence of the each interval. The probability assigned
to each interval is calculated as the ratio of the number of counters as-
Signed to the interval to the total available" (Norris and Kramer, 1990, p.
133).
Despite its superiority, it has been shown that there are several possi-
ble cognitive biases that can potentially reduce the accuracy of this
elicitation method. Norris and Kramer (1990) proposed that these biases
might be caused by three "mental operations". First is the issue of how
representative (or similar to a larger suite of events) the event under as-
sessment is perceived to be. Second is the issue of availability that re-
fers to the ease with which instances or occurrences can be brought to
mind. Third is anchoring and adjustment which might occur when asses-
sors make estimates by starting from an initial value which is adjusted to
yield the final value. A complete discussion of these biases is beyond the
scope of this chapter. It suffices to pOint out that the techniques used in
this project and the training procedures put in place for both farmers and
interviewers participating in this study, were intended to minimize the
likelihood and impact of these biases. It is worth noting that these biases
are not specific to elicitation of subjective probabilities. They occur also
in other survey research situations and their causes and remedies have
been widely documented in survey research literature (e.g. Alreck and
Settle, 1995; de Vaus, 1991).
Several distributions were elicited during each interview. These in-

cluded the distributions of yield and price of chickpeas, of wheat and of
the enterprise that would be replaced if chickpeas were to be adopted.
At the interview we started the elicitation procedure by asking the
farmer for subjective estimates of the lowest yield, the long-term average
yield and the highest yield for chickpeas on a specified paddock or field.
This was a paddock nominated by the farmer as the one most likely to be
suited to chickpeas.
The interviewer recorded the name and size of this paddock . The
farmer was asked to identify and describe the soil type, history and area
of a paddock that would be suitable for chickpeas . We asked that they
nominate a paddock with no previous history of chickpeas. In the subse-
quent interviews we recalled the name of the nominated paddock and
requested that the respondent answer our questions in the context of that
paddock.
Next, we gave the farmer 20 counters (coins), which were each to rep-
resent a season. The farmer was then asked to distribute the counters on
a scale of seven season types ranging from worst to best as illustrated in
Figure 7.1 . The number of counters placed on each yield level repre-
sented the farmer's subjective probability of that yield for chickpeas on
that paddock.
Yield of a chickpea crop on a specified paddock (t/ha)
Lowest Highest
0.2 0.43 0.66 0.89 1.12 1.35 1.58
Figure 7.1: A Sample Elicited Subjective Yield Distribution Table

7.2. Elicitation of Covariance Between the Old and the New
Covariance is a rather complex variable to elicit and compute. It required

elicitation of subjective jOint probability distributions. Hardaker et al. (1997)
showed that in the rare cases that researchers have attempted to elicit joint
distributions the methods have, by necessity, required simplifying assump-
tion to reduce the difficulty of the task. They illustrated a method where, for
two variables that are jOintly distributed, it is possible to assess the distri-
bution by allocating probability counters across a two-way table. They then
showed how to use the two-way table to compute an estimate of the co-
variance of these two variables. However, they also pointed out that most
people find it difficult to consider simultaneously the entire two-way table.
Consequently, the elicitation procedure developed for this project further
adapted the procedure recommended by Hardaker et al. (1997).
The method used in this study, involved elicitation of five yield distribu-
tions, conSisting of two unconditional distributions and three conditional
distributions. Initially, the interviewers were apprehensive about elicitation
of a variable that required so many distributions. However, they soon re-
ported that farmers had little trouble with the task and were consistent in
their responses. Formulating a set of questions that could be used to elicit
the covariance variable was not an easy task. However, as the results
show, the method proved to be an effective way of eliciting covariance.
The procedure for elicitation of the five yield distributions was as follows.
First the unconditional wheat and chickpeas yield distributions were elic-
ited. Then the farmer was asked to provide three new conditional distribu-
tions for chickpea yield. These were the distributions of chickpea yield that
were conditional on the season being the one that would result in the sec-
ond lowest, the mean and the second highest wheat yield from the wheat
distribution. The estimates from these five distributions were then used to
estimate the jOint probability distributions via a fitting procedure (to esti-
mate probabilities for parts of the jOint distribution that were not elicited).
7.3. Elicitation of Individual Risk Preferences
In this study, risk aversion was estimated with the use of subjective
probabilities in conjunction with a set of questions that prompted the
farmer to make a choice from a set of hypothetical wheat grain marketing
scenarios that exposed the farmer to different degrees of risk. The ques-
tions were specifically designed to create a realistic context for decisions
regarding the timing and the amount of the farm's wheat harvest offered
for sale under different marketing schemes. Using different approaches,
three attempts were made to elicit individuals' risk preferences in the

1995, 1996 and 1997 surveys. The technique used in the final interview
(1997) proved to be the only one which showed apparently consistent
and reliable estimates of absolute risk aversion coefficients that were
comparable to the earlier estimates of risk aversion among Australian
farmers (Bond and Wonder, 1980; Bardsley and Harris, 1987).
This technique required the farmer to first describe their perception of the
distribution of the price of wheat. Second, the farmer was asked to indicate
the premium they would be willing to pay for a specified price distribution
with lower variability. This elicited "risk premium" and other production pa-
rameters of the farm were then used to estimate a Pratt-Arrow risk aversion
coefficient for each of the respondents, as explained below.
The elicitation process began by asking the farmer for a set of pa-
rameters that were used in calculating the gross margin and the whole
farm profitability of the entire wheat crop of the property. They included
the total area of the wheat crop in the season preceding the interview
and the expected yield of a specified variety of wheat usually grown (af-
ter a pasture phase) on the soil most suited to chickpeas. It was decided
to make the simplifying assumption that this was the wheat yield across
the whole-farm since obtaining the yield of wheat crops on each soil type
and management strategy for each property would be impractical.
The next step involved the elicitation of a subjective price distribution for
the same variety of the wheat crop on the same soil type with the same
expected quality (protein content). The elicitation technique was similar to
that described previously for elicitation of chickpea yield. To do this the
farmer was asked to specify their estimate of the highest "cash-out pool"
price they were likely to get for the wheat grown in 1998. The interviewers
were instructed to stress that we were interested in the highest price that is
realistically possible in the current market outlook and not the highest the
farmer would hope for. The farmer was then asked for the lowest possible
price they could realistically get for their 1998 wheat crop.
This range was then divided into seven equal intervals and noted in-
side the top row of a grid as shown in Figure 7.1. The farmer was then
asked to distribute 20 counters (coins) within the grid to represent their
estimate of the chances of obtaining each of the possible prices for their
1998 wheat harvest. This provided a seven-increment discrete distribu-
tion for wheat price.
Next the interviewer created, from the elicited price distribution, a new
distribution conSisting of only five price intervals. The grower was told that
the difference between the new distribution and the elicited one was that
the new one had a smaller variability associated with it. The new distribu-
tion was created by "winsorising" the original elicited distribution at a point
equal to the average of the second and third highest prices specified by the
farmer. Winsorising is similar to truncation except that the probability
weight that would be removed by truncation is reallocated to the pOint of
truncation (Fraser, 1988; Fraser and Kingwell, 1997). The low end of the
distribution was winsorised in a similar way. Table 7.1 is a numerical ex-
ample of an elicited and modified price distribution for one of the respon-
dents.
Table 7.1: A Sample of an Elicited and a Contract Price Scheme

Distribution Used in Estimating the Risk Aversion Coefficient
of the Individuals in the Sample.
Farmer specified distribution for Modified distribution for

cash at harvest the contract scheme
Number of Number of
Increment Price ($/t) counters Increment Price ($/t) counters
150
2 158 1 162
3 166 2 166
4 174 3 174
5 182 4 182
6 190 5 186
7 198
The farmer was then asked to consider a hypothetical scenario in

which, for a limited time in that season, the Australian Wheat Board
(AWB) was offering an alternative payment system that guaranteed a
spread of prices with less variability, as indicated by the modified distri-
bution. If the grower chose to participate in this scheme, the AWB would
guarantee that at the time of harvest the prices would be bound by the
lower and upper limits specified by the new price distribution. This
scheme was in effect a tool for managing wheat price risk that the farmer
could choose to use. It was then explained that such a scheme would
have administrative costs. Consequently, the farmer would be required to
pay for the use of such a scheme. This cost would be in the form of a fee
to be deducted from the harvest payment made to the farmer. They were
asked to specify the highest fee they would be willing to pay and still be
willing to participate in the scheme. This fee constitutes the difference in
risk premium between the two distributions.
The farmer was then asked if he or she would consider using this
scheme for the entire production of their wheat in 1998, on the condition
that they were not committed to a pre-defined amount of grain to be de-
livered. The rationale for not asking the farmer to commit to a set ton-
nage was to remove the element of risk and uncertainty associated with
the yield of wheat so that we would be dealing exclusively with the price
risk aspect of the production of wheat. To determine whether risk prefer-
ences varied in response to the amount of money involved, the question
regarding the use of this scheme was repeated for 10, 25, and 50 per-
cent of the farmer's wheat harvest.
When this question was pre-tested, it was found that some farmers
preferred the more variable distribution to such an extent that their re-
sponse was to want to be paid to participate in this scheme. The reason
expressed by them was that, although the scheme reduced the downside
risk, it also excluded the possibility of the higher prices. Consequently,
the interviewers were asked to note the amount by which such farmers
would want to be compensated to participate in this scheme, reflecting a
degree of risk seeking.
Individuals' risk aversion coefficients could be inferred from the two
distributions and the risk premia we elicited from them. Solving for the
value of absolute risk aversion (P) satisfies equation (7.1).
CE(ll1 ; p) = CE(ll2 - R ; p) (7.1 )
where CE(Ilj ; p) is the certainty equivalent value of profit distribution i
given risk aversion coefficient p, and R is the difference in risk premia
between the two distributions, as specified by the grower.
By assuming a constant absolute risk aversion utility function of the
form,
U = a + b . exp (-pI1) (7.2)
p becomes the only unknown in equation (2). Non-linear programming

was used to derive the value of p for each farmer because the complex
structure of the equations did not permit us to solve directly for p.
7.4. Results and Discussion
7.4. 1. Expected yields

Table 7.2 shows the large variation that exists between the farmers in their
estimate of the mean yield of crops like wheat and chickpeas.
124 Amir K. Abadi Ghadim and David J. Pannell
Table 7.2: Perceived Mean Yields for Chickpeas and Wheat
Range of expected yields Proportion of farmers in each of the mean yield

(t/ha) from subjective dis- categories (%)
tributions
Chickpeas Wheat
1995 1996 1997 1995 1996 1997
0.2-0.3 1
0.31-0.4 7 2 2
0.41-0.5 7 5 2
0.51-0.6 17 9 8
0.61-0.7 20 12 17
0.71-0.8 16 27 21
0.81-0.9 17 17 16
0.91-1.0 9 13 12 2
1.01-1.1 6 9 2 2
1.11-1.2 3 2 5 7 2 3
1.21-1.3 4 3 4 9 3
1.31-1.4 1 3 9 2 8
1.41-1.5 8 12 5
1.51-1.6 11 5 7
1.61-1.7 2 10 9 8
1.71-1.8 8 10 10
1.81-1.9 9 8 8
1.91-2.0 8 9 8
2.01-2.1 6 8 5
2.11-2.2 2 4 10
2.21-2.3 4 7 6
2.31-2.4 2 5 6
2.41-2.5 4 4 4
2.51-2.6 2 1 2
2.61-2.7 1 2
2.71-2.8
2.81-2.9 2
2.91-3.0 2
3.01-3.1
Sample mean yield 0.65 0.76 0.79 1.73 1.77 1.79
Note: calculated from subjective yield distributions elicited in the 1995, 1996, and 1997
interviews.
The smaller variation in the estimates of mean yields for chickpeas may
be because their perceptions were formed on the basis of information from
similar origins (e.g. the extension agents of Agriculture Western Australia).
It also probably reflects the lower mean yields of chickpeas, so that the
absolute scope for variation is less.
The sample mean yield for wheat is about one ton per hectare larger
than for chickpeas (Table 7.1). Excluding four outlier farmers, the expected
yield of chickpeas is clustered between 0.4 and 1.4 tons per hectare. The
lowest mean yield in the sample for chickpeas is 0.2 tons per hectare. One
person in the sample thought that the expected yield could be as high as
2.2 tons. Lack of adequate knowledge of the agronomy of chickpeas in this
region may be the main reason for the extremely high or low estimates of
expected yield by some individuals
Table 7.1 also shows the cluster of opinions about the expected yield of
wheat. Except for the extremes, expected wheat yields in this sample are
clustered between one ton and 2.7 tons per hectare.
The variation in the mean of the population between the three interview
years was 0.06 tons per hectare for wheat but 0.14 tons per hectare for
chickpeas. This suggests that farmers in this sample are still learning about
the performance of chickpeas. Consequently, their perceptions are likely to
be influenced Significantly by the information they collect each year. How-
ever, for a crop like wheat with a long history on these farms, an additional
year of information makes little difference to perceptions. It is worth noting
that mean yields of chickpeas for the sample population over the three
years shows a shift towards higher yields, possibly reflecting improved
knowledge about its performance, but possibly also a result of increased
skill and knowledge about suitable agronomic practices.
7.4.2. Riskiness of the old versus the new
The coefficient of variation (CV) of crop yields provides an indication of
riskiness of the crop The CV was estimated from the subjective probabil-
ity distribution of crop yields elicited from each respondent. A comparison
of the CV of chickpea yield with the CV of wheat yield can be used as an
estimate of the relative riskiness of chickpeas. All farmers in the sample
had far more experience with wheat than with any other crop, so wheat
was used in this situation as a benchmark crop.
Every year, one or two farmers gave responses that chickpeas were
less risky than wheat on the basis of yield. However, the majority consid-
ered chickpeas to be more risky than wheat. This is further summarized
by the weighted-averages shown in the last row of Table 7.3.
Table 7.3: The Coefficient of Variation of Chickpea and Wheat Yields

Calculated from Individual Subjective Probability Distributions.
Range of CV of Proportion of farmers in each of the CV categories (%)

crop yield
(%) Chick peas Wheat
1995 1996 1997 1995 1996 1997
0-5
6 - 10 2
11 - 15 1 5 5 3
16 - 20 4 3 12 7 11
21 - 25 1 10 12 15 22 24
26 - 30 6 9 13 16 18 25
31 - 35 14 14 14 16 18 15
36 - 40 11 17 15 14 13 12
41 - 45 15 8 20 10 8 5
46 - 50 8 15 13 6 5 4
51 - 55 10 8 5 3 4
56 - 60 8 8 2 1
61 - 65 8 4
66 - 70 4 1 2
71 - 75 6
76 - 80 2
81 - 85 4
86 - 90 1
91 - 95
96 - 100
Pooled sample 49 41 37 31 30 29
mean CV (%)
For instance, in 1995, the average CV of yield for the entire sample
was 31 percent for wheat but 49 percent for chickpeas, while in 1997 the
average CV for the sample was 29 percent for wheat but 37 percent for
chickpeas.
There is a clear indication of tight clustering of the CV values for wheat
and a wider clustering for chickpeas. This indicates the extent of uncer-
tainty and lack of knowledge about chickpeas as compared to a more
traditional crop like wheat. For instance, the majority of the farmers in the
sample (83 percent) believed that the CV of wheat yield is between 16
and 45 percent (a range of 29 percent). For chickpeas, the majority of

farmers in the sample (84 percent) believed that the CV of chickpea yield
is between 26 and 70 percent (Table 7.3), (a range of 44 percent).
There is a clear indication of a shift over time in perceptions of the
riskiness of chickpeas. The average CV fell substantially from 1995 to
1996 and fell a little further from 1996 to 1997 (Table 7.3). This indicates
that from a whole-sample perspective, farmers had become more certain
about the yield of chickpeas over the three seasons. Trial evaluation of
chickpeas by the 37 percent who chose to grow this crop in 1995 and
non-trial evaluation from other sources may have contributed to this
change in perceptions. It is worth pointing out that, from one year to the
next, there is a small proportion of farmers who left the sample while
others were recruited into the study. Therefore, it is possible that a small
proportion of the change of perception of CV is attributable to this
change in the sample population from one year to next. However, our
view is that most of the observed shift is likely to be changes in percep-
tions.
7.4.3. Covariance
In a previous section we showed evidence from past research supporting
the idea that covariance affects a farmer's ability to learn from past expe-
rience. The covariance between wheat and chickpeas performance is
one way of quantifying how a farmer thinks chickpeas compare to wheat
in the way they responds to different seasonal conditions. This in turn
gives important clues about the extent to which it is possible to take ac-
cumulated knowledge of wheat yield responses to weather conditions
and apply it to chickpeas. Higher correlation implies increased ability to
learn from trialling chickpeas since the yield of wheat is well understood
and provides a clear benchmark for judging where the observed chickpea
yield lies within the long-run distribution of yields (Lindner and Gibbs,
1990).
The pooled sample mean covariance between the yield of wheat and
chickpeas is 0.08 (Table 7.4). There is a clustering of the perception of
covariance between 0.025 and 0.5.
Only two percent of the farmers thought that there was a negative cor-
relation between the yield of wheat and chickpeas. At the other extreme
only four percent of the farmers in the sample thought that there was
high (greater than 0.5) correlation between the yield of two crop. These
findings suggest that most farmers in this sample would interpret infor-
mation about chickpea crop yields in different seasons in the light of the
response of their own wheat crops in similar climatic conditions. The re-
sults are consistent with field observations regarding the correlation be-
tween the yield of wheat and legume crops.
Table 7.4: The Range of Covariance Values Observed among Farmers

in the 1997 Sample
Covariance range Proportion of farmers in each of the co-

(ton 2 ha- 2) variance categories (%)
<0 2
0-0.025 4
0.025 - 0.05 15
0.05 - 0.075 24
0.075 - 0_1 13
0.1 - 0.25 16
0.25 - 0.5 23
0.5 - 1 4
0.08 Pooled sample mean
7.4.4. Risk preferences of individuals

As discussed earlier, risk preferences of the farmers in the sample were
calculated by eliciting a set of responses that allowed us to compute a
Pratt-Arrow absolute risk-aversion coefficient for each individual. As de-
scribed earlier, risk preferences were computed from responses to a set
of questions regarding the price distribution of wheat. The risk prefer-
ences for the sample of farmers that we interviewed are shown in Table
7.5.
The risk aversion coefficient implied by the farmer's response to the
contract offered to them varied somewhat depending on the size of the
contract. However, the overall pattern in responses was reasonably con-
sistent. The majority of farmers were found to be risk averse, consistent
with Bardsley and Harris (1987) and Bond and Wonder (1980).
In general, farmers showed higher risk aversion when the question in-
volved committing small proportions of their wheat harvest to the con-
tract. In discussing the reasons for this response, most farmers argued
that locking their entire crop into a contract would exclude them from the
possibility of getting the highest price in the more risky distribution. An
analysis of the reasons for such apparently risk-seeking behavior was
beyond the scope of this study. Depending on the proportion of harvest

sold to contract, between 13 and 23 percent of farmers provided re-
sponses that could not be used to compute the Pratt-Arrow coefficient.
Table 7.5: Proportion of Farmers with Different Levels

of Risk Aversion in the 1997 Sample.
Range of Pratt-Arrow Percent of total wheat crop

risk aversion coefficient harvest sold to contract
100 50 25 10
1.00E-02 to 1.00E-03 1 1
1.01 E-03 to 1.00E-04 13 25 23
1.01 E-04 to 1.00E-05 40 46 35 26
1.01 E-05 to 1.00E-06 21 18 13 14
1.01 E-06 to 1.00E-07 6 3 1
1.01 E-07 to O.OOE+OO 2 3
O.OOE+OO to -1.00E-07 3 2
-1.01 E-07 to -1.00E-06
-1.01 E-06 to -1.00E-05 10 2 4
-1.01 E-05 to -1.00E-04 0 0
-1.01 E-03 to -1.00E-02 1
Non-computable responses 13 13 18 23
This occurred, either because they were internally inconsistent, or in-

consistent with responses that could be considered "rational" within ex-
pected utility theory.
7.5. Conclusions
Three aspects of risk and uncertainty have been studied in the context of
a new crop currently being evaluated for adoption by a group of farmers:
farmer perceptions of the riskiness of the crop, farmer perceptions of the
covariance between the yield of the new crop (chick peas) and a tradi-
tional crop (wheat), and farmer attitudes toward risk (i.e. their risk aver-
sion or risk preference). All three elements are relevant in explaining
farmers' responses to the new crop (Abadi Ghadim and Pannell, 1999).
It was found that the riskiness of the new crop (as indicated by the co-
efficient of variation of the farmers' subjective yield distribution) was
higher for the new crop than for the old crop. This would be expected
given the farmers' lack of experience with and greater uncertainty about
the new crop. It was also shown that the perceived riskiness of the new
crop diminished over time, while the perceived riskiness of the traditional
crop was stable over time. The fall in perceived riskiness was presuma-
bly due to farmers gaining experience and knowledge of the new crop,
allowing them to better estimate its true performance and/or allowing
them to better identify most appropriate production methods and input
levels.
The farmers' perceptions of covariance in yield between the new crop
and a traditional crop were consistent with field evidence. The observed
moderate positive covariances are likely to be of value to farmers by al-
lowing them to interpret information about chickpea crop yields in differ-
ent seasons in the light of the responses of wheat crops in similar grow-
ing seasons.
Elicited risk preferences of farmers were consistent with previous re-
sults for Australian farmers, in that most were found to be risk averse
and that the observed range of risk attitudes was wide. The coefficients
of absolute risk aversion inferred from the elicitation process were high
relative to some results in the literature.
The study raises some interesting questions for possible future proj-
ects:
• To what extent does the diversity of perceptions that we noted in
this study reflect uncertainty versus real differences in circumstances? If
the dominant factor is uncertainty, it implies a greater potential for learn-
ing and refinement of decision-making.
• What determines how rapidly uncertainty falls as trialling pro-
ceeds?
• This chapter focused on on-farm trialling, based on a hypothesis
that it is the crucial source of information used by farmers in their adop-
tion decisions. Relatively little data was collected about off-farm sources
of information about innovations. However, an important question is,
how much more effective is on-farm trialling than off-farm information at
reducing uncertainty?
• Does the quality and relevance of off-farm information determine
the timing and intensity of on-farm trialling?
• Are there differences between various types of innovations with
respect to the importance of risk and uncertainty in their adoption? The
innovation. studied here is one with a rapid financial payoff from suc-
cessful adoption. But some innovations, such as land conservation
practices, have a much slower payoff, and often their trials reveal infor-
mation more slowly (e.g. see Pannell, Chapter 4 in this volume). To what
extent does this affect the rate and level of adoption?
Acknowledgments: This project was partly funded by the Rural Industries Research
and Development Corporation (RIRDC). We also wish to thank the Cooperative Re-
search Centre for Legumes in Mediterranean Agriculture and Agriculture Western
Australia for their support. Finally, we are grateful for the kind contribution of the
farmers who partiCipated in this study.
References
Abadi Ghadim, A. K., M. Burton, and D.J. Pannell (1999), "More empirical evidence on
the adoption of chick peas in Western Australia", paper presented at the 43rd An-
nual Conference of the Australian Agricultural and Resource Economics Society,
Christchurch, New Zealand, Jan 20-221999.
Abadi Ghadim, A. K., and D.J. Pannell (1999), "A conceptual framework of adop-
tion of an agricultural innovation", Agricultural Economics, 21: 145-154.
Abadi Ghadim, A. K., and D.J. Pannell (1998). "The importance of risk in adoption
of a crop innovation: empirical evidence from Western Australia", 42nd Annual
Conference of the Australian Agricultural and Resource Economics Society,
University of New England, Arimdale, New South Wales, pp. 15.
Alreck, P. L., and R.B. Settle (1995). The Survey Research Handbook. Sydney:
Irwin.
Anderson, J. R., J.L. Dillon, and B. Hardaker (1977). Agricultural Decision Analy-
sis. Armidale: University of New England.
Arrow, K. J. (1970), Essays in the Theory of Risk-Bearing. London: North-
Holland.
Bardsley, P., and M. Harris (1987), "An approach to the econometric estimation
of attitudes to risk in agriculture", Australian Journal of Agricultural Economics,
31: 112-126.
Bar-Shira, Z. (1992), "Nonparametric test of expected utility hypothesis", Ameri-
can Journal of Agricultural Economics, 74: 523-533.
Bond, G., and Wonder, B. (1980). "Risk attitudes amongst Australian farmers",
Australian Journal of Agricultural Economics, 24: 16-34.
de Vaus, D. A. (1991), Surveys in Social Research. North Sydney: Allen and Un-
win.
Feder, G. (1980), "Farm size, risk aversion and the adoption of new technology
under uncertainty", Oxford Economic Papers, 32: 263-2893.
Feder, G., R. E Just, and D. Zilberman (1985), "Adoption of agricultural innova-
tions in developing countries: a survey", Economic Development and Cultural
Change, 33: 255-297.
Feder, G., and D.L. Umali (1993), "The adoption of agricultural innovations: a
review", Technological Forecasting and Social Change, 43: 215-239.
Fraser, R. W. and R. Kingwell (1997), "Can expected tax revenue be increased
by an investment-preserving switch from ad valorem royalties to a resource
rent tax"? Resources Policy, 23: 103-108.
Fraser, R. W. (1988), "A method for evaluating supply response to price under-
writing", Australian Journal of Agricultural Economics, 32: 22-36.
Hamblin, J. (1987), "Grain legumes in Australia", 4th Australian Society of Agron-
omy Conference, , La Trobe University, Melbourne, Victoria, pp. 65-81.
Hardaker, J. B., R.B. Huirne, and J.R. Anderson (1997), Coping with Risk in Agri-
culture. Wallingford: CAB International.
Hertzler, G. (1997), "A new theory for explaining the paradoxes in decision making
under risk and for measuring time and risk preferences", in R. B. M. Huirne, J. B.
Hardaker, and A A. Dijkhuizen (eds.) Risk Management Strategies in Agriculture.
State of the art and future perspectives. Wageningen, Wageningen Agricultural Uni-
versity, pp. 65-86.
Hey, J. D. (1979), Uncertainty in Microeconomics. Oxford: Martin Robertson.
Hsiao, C. (1986), Analysis of Panel Data. Vol. 1. 1 vols. Cambridge: Cambridge
University Press.
Jovanovic, B., and Y. Nyarko (1996),"Learning by doing and the choice of tech-
nology", Econometrica, 64: 1299-1310.
Kingwell, R. S., A.K. Abadi Ghadim, S.D. Robinson, and J.M. Young (1995),
"Introducing Awassi sheep to Australia: an application of farming systems
models", Agricultural Systems, 47: 451-471.
Lantzke, N. (1992). "Soils of the Northam Advisory District - the Zone of Ancient
Drainage", Bulletin. Department of Agriculture Western Australian.
Leathers, H. D., and M. Smale (1992). "A Bayesian approach to explaining se-
quential adoption of components of a technological package", American Jour-
nal of Agricultural Economics, 68: 519-527.
Lindner, R., and M. Gibbs (1990), "A test of Bayesian learning from farmer trials
of new wheat varieties", Australian Journal of Agricultural Economics, 34: 21-
38.
Lindner, R. K. (1987), "Adoption and diffusion of technology: an overview", in B. R.
Champ, E. Highley and J.V. Remenyi (eds.), Technological Change in Postharvest
Handling and Transportation of Grains in the Humid Tropics, ACIAR Proceedings
No. 19, ACIAR, Canberra, pp. 144-151.
Lindner, R. K., and A.J. Fischer (1981), "Risk aversion and the innovation time
lag", University of Adelaide, Australia.
Lindner, R. K., AJ. Fischer, AJ. ArnOld, and M. Gibbs (1982), "Measuring beliefs
about future events and innovation profitability", Department of Economics,
University of Adelaide, Australia.
Machina, A. (1981), "Rational decision making versus "rational" decision model-
ing", Journal of Mathematical Psychology, 24: 163-75.
Norris, P. E., and R.A Kramer (1990), "The elicitation of subjective probabilities
with applications in agricultural economics" Review of Marketing and Agricul-
tural Economics, 58: 127-147.
Pannell, D.J., R.L. Malcolm, and R.S. Kingwell (2000), "Are we risking too
much? Perspectives on risk in farm modeling", Agricultural Economics, 23: 69-
78.
Pratt, J. W. (1964), "Risk aversion in the small and in the large", Econometrica,
32: 122-136.
Schoemaker, P.J. (1982), "The expected utility model", Journal of Economic Lit-
erature, 20: 529-63.
Shapiro, B.I., B.W. Brorsen, and D.H. Doster (1992), "Adoption of double-
cropping soyabean and wheat", Southern Journal of Agricultural Economics,
24(3): 33-40.
Siddique, K.H.M., and J. Sykes (1997)" "Pulse production in Australia past, pres-
ent and future", Australian Journal of Experimental Agriculture, 37: 103-111.
Smale, M., and P.W. Heisey (1993), "Simultaneous estimation of seed-fertilizer
adoption decisions", Technological Forecasting and Social Change, 43: 353-
368.
Smale, M., R.E. Just, and H.D. Leathers (1994), "Land allocation in HYV adoption
models: an investigation of alternative explanations", American Journal of Agri-
cultural Economics, 76: 535-546.
Smidts, A. (1990), Decision Making Under Risk. A Study of Models and Meas-
urements Procedures with Special Reference to the Farmers Marketing Be-
haviour. Wageningen Economic Studies 18. Wageningen: Pudoc.
Stoneman, T. C. (1992), "An introduction to the soils of the Merredin Advisory
District- description, illustration and notes on nine common soils", Bulletin.
Western Australian Department of Agriculture.
Tsur, Y., M. Sternberg, and E. Hochman (1990), "Dynamic modelling of innova-
tion process adoption with risk aversion and learning", Oxford Economic Pa~
pers, 42: 336-355.
Weisensel, W. P., and R.A. Schoney (1989), "An analysis of the yield-price risk
association with specialty crops", Western Journal of Agricultural EconomiCS,
14: 293-299.
Wright, G. and P. Ayton (1994), "Subjective probability: what should we be-
lieve?", in G. Wright, and P. Ayton (eds) , Subjective Probability, Chichester:
John Wiley and Sons, pp. 163-184.
8
Valuing Pest Control:
How Much is Due to Risk Aversion?
Terrance M. Hurley and Bruce A. Babcock
Understanding the economic and biological determinants of farmers' pest

control decisions is key to understanding the impacts of pesticide regu-
latory decisions that aim to reduce farmers' pesticide use. The first step
in gaining this understanding is to recognize that pesticides are damage
control inputs. This biological characteristic implies that they affect yield
only in the presence of damaging pests (Lichtenberg and Zilberman,
1986). Second, if the presence of pests is stochastic, and damage con-
trol pesticides must be applied before the pest populations can be ob-
served, then the producers use pesticides under uncertainty. This un-
certainty will influence pesticide decisions if pest populations affect prof-
its nonlinearly or if producers are not expected profit maximizers.
Thus, in general, there are two potential motivations for farmers to want
to reduce pest damage. The first is to an increase the expected value of
crop yields. The value of this increase equals the product of market price
(assumed to not change) and the change in yield. Both risk averse and
risk neutral producers will value this increase in mean. The second moti-
vation is to affect higher-order moments of the yield distribution. The
willingness to pay for a change in 'these moments depends on the pro-
ducer's degree of risk aversion. Only the utility of risk-averse producers
will be affected by this change.
The risk literature is full of studies that attempt to determine if agricul-
tural inputs are risk increasing or risk reducing. The classic definition of a
risk reducing input uses the expected utility framework as its guide and
states that an input is risk reducing if a risk-averse firm will use less of it
than a risk-neutral firm. This definition makes intuitive sense, but empiri-
cal determination whether an input is risk reducing or risk increasing de-
pends not only on the form of the utility function, but also on the as-
Valuing Pest Control 135
sumptions used to model the stochastic production process of the firm.

Just and Pope (1979) showed that common production function specifi-
cations used at the time imposed restrictions on how inputs affect risk.
Researchers have moved towards adopting flexible stochastic relation-
ship between inputs and the distribution of yields (Nelson and Preckel,
1989; Babcock and Hennessy, 1996) in an effort to let available data
determine how input use changes the distribution of yields.
The objective of this analysis is to present an analytical framework that
can be used to estimate the benefits of reductions in yield damage from
the application of pesticides to control a single pest. The framework is
presented in the context of estimating the value of using Bt corn to re-
duce the damage to corn yields from European corn borers infestation in
Iowa.
The motivation for the Bt corn application comes about is the common
belief that that one of the benefits of new transgenic crops, such as her-
bicide-tolerant crops and pesticidal crops, is their risk management
benefits due to a decrease in the variability of yields (Kalaitzandonakes,
1999). That is, because adoption of a transgenic crop lowers the risk of
crop damage, then the variability of yields must be reduced. A logical
conclusion of this perception is that studies that attempt to estimate the
value of these transgenic crops should account for changes in yield vari-
ability and risk-averse behavior by farmers. Ignoring variability and risk
aversion could lead to an underestimate of the value of adopting the
transgenic crop.
The analytical method will be to first estimate probability distributions
of corn yield with and without Bt corn on a representative field under dif-
ferent specifications of corn borer infestations and damage. Then, the
willingness to pay to move from the non-Bt corn yield distribution to the
Bt corn yield distribution will be estimated under different risk aversion
levels. The proportion of willingness to pay for the Bt distribution ac-
counted for by the change in mean yield is estimated to determine the
relative importance of risk aversion in determining the total benefits of
adopting Bt corn.
8.1. Modeling Crop Yields with and without Bt Corn
How yields are affected by a Single pest, European corn borer, can best
be modeled by assuming that harvested yield depends on two factors:
the level of yields that would occur without corn borer damage (the pest-
free yield); and, the level of corn borer damage. The pest-free yield is a
random variable, depending on many factors, including weather, applied
136 Terrance M. Hurley and Bruce A. Babcock
inputs, and the amount of damage from other pests (including other in-
sects, weeds, and disease). The level of corn-borer damage depends on
corn-borer population and control efforts. Corn borer populations in most
of the U.S. Corn Belt vary widely from year to year, thus damage is also
a random variable. In addition, most corn farmers do not treat for corn
borers, thus damage control efforts without Bt corn are zero.
Abstracting from the level of applied inputs (assumed to be set at ex-
pected-profit maximizing levels), and assuming that the profit-maximizing
level of damage control is zero, harvested yield depends on the level of a
Single pest infestation as follows:
Yh = Yf[l- D(p)Dmaxl (8.1 )
Where Yh is harvested yield, Yf is pest free yield which depends on all

other random inputs, O(P) is corn borer damage (expressed as the pro-
portion of maximum damage), p is pest population, and Omax is the
maximum proportion of the crop that can be lost from corn borer, with 0 s
Omax s 1. The properties of O(P) are generally consistent with that of a
cumulative distribution function: 0 s O(P) s 1 and 0 '(P) ~ O.
If corn borer damage is completely eliminated from use of Bt corn, then
the distribution of Bt corn yields equal the distribution of Yf. Given the
distribution of Yf, we could estimate the distribution of Yh. We could then
use the two distributions to calculate expected utility with and without Bt
corn and find the components of willingness to pay for Bt corn.
Using equation (8.1) and the transformation of random variables tech-
nique, if Yf - 9(yf)' then f(Yh)- 9 [Yh /S(p)] (1/S(p)), where
S(p) = [1- D(p)Dmaxl. The beta distribution can be used to show how the
transformation of variable technique can be used. Suppose that Yf is
beta-distributed with a minimum of zero (the crop gets hailed out, a
maximum of b, and shape parameters a and f3:
a-l(b )p-l
1 Yf -Yf
(8.2)
fey f) = B(a, [J) ba+p+1
Then f(Yh) conditional on p, also follows a beta distribution with mini-

mum of zero, a maximum of bS(p), and shape parameters a and [J.
Proof:
1 1
f(Yh I p) = B(a,[J) S(p)
1 1 S(p)2-a- p /t-l(S(p)b - Yf )p-l

B(a,p) S(p) ba + p - 1
1 yf-\S(p)b - Yf l-l (8.3)

B(a,p) (S(p)b)a+p-l
The distribution of Bt corn is identical to the distribution of non-Bt corn

except that the maximum yield of non-Bt corn depends on the infestation
level of corn borers. This is a reasonable outcome, because in the ab-
sence of corn borers, yields of Bt corn and non-Bt corn should be identi-
cal assuming no "yield-drag" on Bt corn.
8.2. Willingness to Pay for Bt Corn
Given acres A, a normalized output price of unity, and a per-acre cost of

production of c, expected utility for Bt corn is given by
b
EU( 7t f ) = fU((y - c)A)g(y f )dy (8.4)
o
Because we want to calculate the willingness to pay for this level of

expected utility, we need to calculate expected utility of non-Bt corn at
the same price and per-acre cost levels. Given a pest infestation level of
p, conditional expected utility of non-Bt corn is
S(p)b
EU( 7th I p) = EU( 7th) = f U((y - c»Af(Yh )dy . (8.5)
o
Assuming that the corn borer infestation level is independent of pest-

free yields, then unconditional expected utility for non-Bt corn is found by
integrating conditional expected utility with respect to the distribution
function of p, h(P).
P.
EU( 7th) = fEU (7th I p)h(p)dp
PI
Ph S(p)b
= f f U((y - c)A)f(Yh I p)dydp (8.6)
PI 0
The money metric for expected utility is certainty equivalent returns

(CER). CER values under the two distributions are defined by
EU(7rj) = U(CERj) and EU(7rh) = U(CER h )

The increase in CER is a money-metric measure of the benefit that ac-
crues from adoption of Bt corn. Whether a farmer adopts Bt corn de-
pends on whether the increase in cost from planting Bt corn (not consid-
ered in the above analysis) is greater than or less than CERf - CERh.
Now we can turn our attention to whether a risk-averse producer has a
greater willingness to pay for Bt corn than a risk-neutral producer. If not,
then we can say that the primary benefit of Bt corn is the increase in ex-
pected yield. If the willingness to pay under risk aversion is greater than
under risk neutrality, then we need to estimate the proportion of willing-
ness to pay that is accounted for by the increase in expected yield and
the proportion that is accounted for by a concave utility function.
8.3. Estimation of Willingness to Pay
To simulate the willingness to pay for Bt corn requires a utility function, a

level of risk aversion, an estimated probability distribution of corn borer
infestations, and parameter estimates of the beta distribution that de-
scribes corn borer-free yields, and an estimated damage function.
Choice of a utility function for this type of analysis is less important
than choice of a level of risk aversion. We will employ the CARA utility
function and limit the range of risk aversion coefficients to a range that
corresponds to a ratio of the risk premium to the standard deviation of
profits between 0.10 and 0.70. This range should include most risk-
averse individuals who are active in farming. (Babcock, Choi, and Fein-
erman, 1993).
Data to estimate the probability distribution of corn borer infestation
was collected from 1960 to 1970 across the Corn Belt (Calvin, 1996).
Population data collection efforts ended in most locations at that time
because researchers did not feel that infestation levels in a given year
could be predicted from past infestations. We calculate willingness to pay
for Bt corn in two locations. At the Boone County site (located in Central
Iowa) the sample mean and standard deviation of per-plant corn borer
larvae were 1.1 and 0.95. In Hall County, Nebraska (located in the irri-
gated corn area of Nebraska) the mean was 1.98 with a standard devia-
tion of 0.986. The higher mean in Nebraska agrees with recent observa-
tions that corn borer pressure on irrigated corn in Nebraska is greater
than in dryland corn areas of the Central Corn Belt.
Clearly, given these sample moments, the infestation distributions are

skewed to the right and limited from below by zero. A reasonable func-
tional form to model this distribution is the log-normal distribution:
h(p) = _1_ex p (_ (log (p) -

p& 20"2
pf)
Method of moments estimates of It and 0" are -0.183 and 0.557 for Boone
County and -0.572 and 0.222 for Hall County.
Expected pest-free yields are set at 128 bu/ac with a standard deviation
of 33 bu/ac in Boone County and 154 and 40 in Hall County. Maximum
pest-free yields are set to 200 bu/ac and 240 bu/ac in Boone and Hall
County respectively. Minimum yield is set to 0 bu/ac in both counties. Us-
ing method of moments implies that p and q of the beta distribution equal
4.78 and 2.69 for Boone County, and 4.67 and 2.61 for Hall County. The
reasonableness of these distributions is demonstrated by the implied crop
insurance rates of approximately 2.2 percent in Boone County and 1.8 per-
cent in Hall County at a coverage level of 65 percent of expected yield.
These parameterizations of the corn borer-free yield are meant to reflect
some of the benefits of Bt corn. The actual crop insurance rate charged in
Boone County is 3.03 percent and in Hall County it is 2.6 percent.
Data to estimate the damage function is extremely limited. What is
commonly used in analyses of alternative corn borer treatment options is a
constant proportion of damage per corn borer. We use four percent dam-
age per corn borer with Dmax = 1.0. Thus, if there are more than 25 corn
borers per plant, yield of non-Bt corn is zero.
Figures 8.1 and 8.2 present the yield density functions with and without
Bt corn in Boone County and Hall County respectively. The Bt corn yield
densities are given by the beta density in equation (8.2). The non-Bt yield
density is the expected yield density, where the expectation is taken with
respect to the density function of corn borer infestation. That is,
f !(Yh I p)h(p)dp
Ph
!(Yh) = (8.7)
PI
The integration needed to find the non-Bt corn density and all subse-
quent integrations used to find expected utility levels were carried out using
Monte Carlo integration techniques.
A comparison of the density functions in Figures 8.1 and 8.2, shows that
adoption of Bt corn shifts the density to the right. This shift to the right clearly
shows the benefits of Bt corn. There is a higher expected yield and a lower
probability of very low yields. But the rightward shift is not a simple mean
shift because maximum and minimum yields are not affected. The maxi-
mum yield is not affected because of the assumption that the yield capacity
of Bt corn and non-Bt corn are identical. And there is no reason to expect
that the minimum yield of Bt corn is any different than non-Bt corn.
1000
900
800
700
..,.
u 600
c::
:J 500
~ 400
300
200
100
0
0 50 100 150 200
Vie Id (bu/ac)
Figure B. 1: Effect of Bt Corn on Distribution of Corn Yields in Boone County,

Iowa
1000
900
800
700
Conventional corn
...
- - - - - - - - - - --------------- -----~~ --- .,.~--:--------------
>- 600
..
u
c
:::I
tI'
500
~ 400 ---~~-~ __---'b__~-__l
300
200
100
0
0 50 100 150 200
Yield (bu/ae)
Figure B.2: Effect of Bt Corn on Distribution of Corn Yields in Hall County,

Nebraska
Besides a shift in the density, there is also a spread in the density. This
spread is more apparent in Hall County than in Boone County. This in-
crease is indicated by lower peaks in the Bt corn densities. The cause of
this increase in spread is that Bt corn has a larger positive affect on yield
when corn borer-free yield is high than when it is low. That is, the assump-
tion of proportionate damage (a common assumption in nearly all pest
control studies) means that elimination of a damaging agent has a smaller
effect when pest-free yields are low than when they are high.
Table 8.1 presents the results of this analysis. The risk aversion coeffi-
cients are reported in the first column. The next two columns express this
level of risk aversion as a ratio of the corresponding risk premium to the
standard deviation of returns under Bt corn. This ratio shows the reason-
ableness of the range of risk aversion coefficients used in this study. The
last two columns report the change in CER after adopting (costlessly) Bt
corn. This change in CER is the per-acre maximum a producer would be
willing to pay for Bt corn (in terms of bushels of corn).
At a risk aversion level of 0.0 (risk neutrality) the maximum willingness to
pay is simply the change in expected profit. This amounts to 5.03 bu/ac in
Iowa and 11.2 bu/ac in Nebraska. This amount reflects the change in ex-
pected yield that would occur from adoption of Bt corn. The change in ex-
pected yield varies directly with expected corn borer pressure. If current
corn borer pressure is higher than that used in this analysis then the
change in expected yield would be greater than that reported in Table 8.1.
Table 8.1: Per-Acre Willingness to Pay for Bt Corn
Ratio of Risk Premium to Change in

Standard Deviation of Returns" Certainty Equivalent Returns
Risk
Aversion Iowa Nebraska Iowa Nebraska
Level b
0.0 0 0 5.03 11.18
0.001 0.02 0.02 4.76 10.40
0.005 0.09 0.10 4.77 10.87
0.009 0.16 0.19 4.62 10.55
0.01 0.18 0.21 4.58 10.45
0.02 0.36 0.44 4.14 9.34
0.03 0.55 0.66 3.68 8.19
"The standard deviation of returns is with Bt corn.

bConstant absolute risk aversion is assumed.
The change in CER as risk aversion increases reflects the effects of Bt

corn on all the moments of the yield distribution. That is, risk-averse pro-
ducers are affected by a change in expected yield as well as yield variabil-

ity and yield skewness and kurtosis. Risk-averse producers place a posi-
tive value on increased expected yields. But they place a negative value on
yield variability and negative skewness. As can be seen in the density
functions in Figures 8.1 and 8.2, Bt corn increases the mean, seems to
increase the variability of yields, and definitely increases the leftward
skewness of yields.
The Table 8.1 results show that the net effect of these changes on the
yield distribution is a lower willingness to pay for Bt corn by risk-averse
producers than by risk-neutral producers. Furthermore, the gap in willing-
ness to pay increases as risk aversion increases. This demonstrates that
the benefits of Bt corn are due to the increase in mean yields, rather than
its effects on higher-order moments of the yield distribution.
These effects on the yield distribution are due to the way that corn borers
affect non-Bt corn as modeled by equations (8.1) and (8.3). These models
assume that: pest damage only occurs when pests are present; damage is
proportional to the number of pests present; and pest pressure and pest-
free yields are independent. How reasonable are these assumptions? The
first assumption is tautologically true. The second assumption says that
corn borer damage from a given level of infestation is twice as high if pest-
free yield were 200 bu/ac than if pest-free yield were 100 bu/ac. This as-
sumption of proportion damage is adopted in nearly all conceptual and em-
pirical analyses of pest control (see for example, Hennessy, 1997; Fox and
Weersink, 1995; Lichtenberg and Zilberman, 1986; Saha, Shumway, and
Havenner, 1997.) The rationale for this assumption is that if a pest affects
a given proportion of plants in a field, then the lost yield varies directly with
the yield that would have occurred had the pest not been present. The third
assumption is appropriate for European corn borers because relatively
unimportant weather factors (from the point of view of crop yields) can
have dramatic effects on corn borer pressure. For example, a single heavy
rain at the time of moth flight can devastate corn borer populations yet
have little affect on crop yields.
8.4. Implications
At first glance the results indicate that a risk-averse producer values the
pest control benefits of Bt corn less than the risk-neutral producer. This
implies that given a fixed treatment cost, a risk-averse producer is less
likely to adopt Bt corn than his risk neutral counterpart. That is, increases
in risk aversion decreases treatment. Does this mean Bt corn is a risk-
increasing technology?
Pope and Kramer (1979) define an input to be risk increasing if the risk-
averse firm uses less of the input than a risk neutral producer. A discrete
analog of this definition would define a risk-increasing technology as one
that would be more readily adopted by a risk-neutral firm than a risk-averse
firm. Or, equivalently, a technology could be said to be risk increasing if
increases in risk aversion decrease the willingness to pay to adopt the
technology. The results in Table 8.1 indicate that this definition leads one
to conclude that Bt corn is indeed a risk increasing technology.
But an examination of Figures 8.1 and 8.2 shows that a characterization
of Bt corn as a risk-increasing technology is extremely misleading. Bt corn
decreases the probability of achieving low yields and increases the prob-
ability of achieving high yields. How could such a technology be charac-
terized as risk-increasing?
A concave utility function simply means that a producer prefers yield sta-
bility over yield variability because the utility gain from positive yield shocks
is less than the utility loss from negative yield shocks. As shown in the fig-
ures, Bt corn shifts the distribution of corn yields to the right and increases
the variability of yield. Both risk averse and risk neutral producers value
equally the shift to the right in the distribution. But risk-averse producers
lose utility, relative to the status quo, because the standard deviation of
and the leftward skewness of yields increases with Bt corn. This disutility
occurs even if the increase in standard deviation is caused by an increase
in upside risk. It is in this sense that expected utility theory characterizes
Bt corn as risk-increasing.
8.5. Concluding Remarks
This empirical analYSis shows the limitations of the agricultural economics

literature in providing meaningful explanations and characterizations of the
role of risk in determining the production decisions of producers. Expected
utility theory has led us to adopt characterizations of technologies that pro-
vide little inSight into the motivations of firms to adopt new technologies.
We find that Bt corn is valued more by risk-neutral producers than by risk-
averse producers even when Bt corn yields can never be lower than con-
ventional corn yields in every state of nature. That is, Bt corn first-order
stochastically dominates conventional corn, yet the change in CER caused
by adoption of Bt corn decreases as risk aversion levels increase.
Economists and others who are interested in explaining the impacts of
new pest control technologies should concentrate on careful estimation of
the effects the technology will have on the distribution of crop yields, taking
into account any physical or biological constraints or characteristics of the
technologies. Such estimation will provide a complete characterization of

the impacts the technology will have on expected returns, which should
give good insight into the potential adoption rate of the technology.
References
Babcock, B. A, E. K. Choi, and E. Feinerman. {1993}, "Risk and Probability Premiums
for CARA Utility Functions." Journal of Agricultural and Resource Economics, 18:
17-24,
Babcock, B. A, and D. Hennessy (1996), "Input Demand Under Yield and Revenue
Calvin, D. D. (1996), "Economic Benefits of Transgenic Corn Hybrids for European
Corn Borer Management in the United States." Public Interest Document Supporting
the Registration and Exemption from the Requirement of a Tolerance for the Plant
Pesticide Bacillus thuringiensis subsp. kurstaki Insect Control Protein as Expressed
in Corn (Zea mays L.),.
Fox, G. and A Weersink (1995), "Damage Control and Increasing Returns." American
Journal of Agricultural Economics, 77(1}: 33-39.
Hennessy, D. A (1997), "Damage Control and Increasing Returns: Further Results."
American Journal of Agricultural Economics, 79(3}: 786-91.
Just, R. E., and R. D. Pope (1979), "Production Function Estimation and Related Risk
Considerations." American Journal of Agricultural Economics, 61: 276-84.
Kalaitzandonakes, N. (1999), "A Farm-Level Perspective on Agrobiotechnology: How
Much Value and for Who?" AgBioForum 2(2}: 61-64.
Lichtenberg, E. and D. Zilberman (1986), "The Econometrics of Damage Control: Why
Specification Matters." American Journal of Agricultural Economics, 68: 261-173.
Nelson, C. J. and P. V. Preckel (1989), 'The Conditional Beta Distribution as a Sto-
chastic Production Function." American Journal of Agricultural Economics 71 (2):
370-378.
Pope, R. D. and R. A Kramer {1979}, "Production Uncertainty and Factor Demands
for the Competitive Firm." Southern Economic Journal; 46(2): 489-501.
Saha, A, C. R. Shumway, and A Havenner (1997), "The Economics and Economet-
rics of Damage Control." American Journal of Agricultural Economics; 79(3}: 773-85.
9
The Influence of Price Risk on
Set-aside Choice in the EU
Hild Rygnestad and Robert W. Fraser
Set-aside was introduced in the European Union's (EU's) Common Agri-

cultural Policy (CAP) for the first time in 1988. Due to changes in world
prices and supply of crops supported with intervention purchasing and
regulated with set-aside (mainly cereals, oil seeds and protein crops), key
policy factors have been adjusted in later years. These factors include
set-aside rates, hectare and set-aside payments (i.e. compensatory
payments) and intervention prices 1. From 1993 to 1995 a distinction was
made between a Rotational set-aside scheme where a different part of
the farm was taken out of production each year, and a Non-rotational
option, which constituted longer-term diversion of a larger part of the
farm (i.e. a larger set-aside rate). Although this distinction was reduced in
subsequent policy amendments because the set-aside rates were uni-
fied, it was still possible for the farmer to choose either option depending
on land heterogeneity (Rygnestad and Fraser, 1996).
The Agenda 2000 reform for cereals feature a retention of the set-aside
policy, but an additional reduction in price support, thereby further ex-
posing EU cereal growers to price risk (European Commission, 1999).
The following analysis focuses on a specialist wheat grower's choice
between the two set-aside management options in the context of sub-
stantial price risk as proposed by Agenda 2000. The aim is to illustrate
the combined effect of land heterogeneity and price risk on the choice of
set-aside option - using the Danish situation as an example. Subsequent
effects on output reduction and nitrate leaching are also discussed.
Previous studies have shown that the level of output reduction is de-
pendent on set-aside management. On the one hand, there can be yield
benefits in the Rotational scheme when an area is brought back into pro-
duction after a year in fallow (set-aside). On the other hand, the Non-
146 Hild Rygnestad and Robert W. Fraser
rotational option makes it possible to maintain production on highly pro-

ductive land while poor land is set-aside (Babcock et a/., 1993; Brown,
1993; Fraser, 1994; Hoag et a/. 1993; and Rygnestad and Fraser,
1996)2. Evidently the latter point was recognized and emphasized among
EU policy makers, since at the introduction of the Non-rotational option in
1993/94 the set-aside rate was set higher than that of the Rotational
scheme (Agra Europe, 1993 a and b).
With regard to nitrate leaching, previous studies point out that set-aside
can provide environmental benefits to surrounding wildlife as well as re-
duced fertilizer use and subsequent leaching. However, the benefits are
highly dependent on appropriate management of set-aside land, for ex-
ample with the Rotational and Non-rotational options (Firbank et a/.,
1993; Magid et a/., 1994; Newbold and Rush, 1993; Simmelsgaard, 1991;
Sotherton, 1998; Waagepetersen, 1992; and Webster and Goulding,
1995). By considering both output and leaching, Rygnestad (1999) con-
cludes that the farmer's optimal choice of set-aside management for out-
put reduction varies with land heterogeneity, whereas the Non-rotational
option always is the optimal choice to reduce nitrate leaching. Thus, due
to the integrated relationship between output and leaching, there is typi-
cally a conflict between these policy goals.
Beyond land heterogeneity, a farmer's decision making with regard to
set-aside management is also affected by different risk factors. Both
Fraser (1993) and Roberts et a/. (1996) analyze the effect of risk on the
decision to participate in set-aside in the EU. The main factors affecting
farmer participation in set-aside were found to be policy variables such
as the set-aside and compensatory payments. Secondly, farm and re-
gional level differences in cost structures and yield levels were signifi-
cant. Lastly, factors such as price levels, price variations and support
prices had an effect on set-aside uptake. Similarly, Krause et a/. (1995)
found for the US situation that the expected price, the support price and
the price variance had an effect on the level of planting to program crops
(e.g. wheat and conservation set-aside). This is expected to be particu-
larly the case as EU farmers are exposed to more risk when price sup-
port for wheat is reduced further with the Agenda 2000 reforms. Because,
for a specialist wheat grower, variance in the wheat output price only af-
fects income earned on productive land, the farmer's revenue can be di-
vided between the risky income from wheat production and the guaran-
teed income from set-aside policy payments. Thus, by joining a set-aside
scheme the farmer is partially insured against output price variations. As
mentioned in Roberts et a/. (1996) risk and farm level factors are ex-
The Influence of Price Risk on Set-aside Choice in the EU 147
pected to be integral to the choice between different set-aside manage-

ment options. This is explored further in the next sections.
The simulation model from Fraser and Rygnestad (1999) and Rygnestad
and Fraser (1996) is developed in Section 9.2 to include output price risk in
a mean-variance framework. Section 9.3 shows the combined effect of
price risk and land heterogeneity on choices of set-aside management as
well as on output and nitrate leaching. Policy implications of the findings
are discussed and conclusions are drawn in Section 9.4.
9.1. Methods
In this section a theoretical model for price risk is developed based on a
mean-variance framework. This price risk is then built into the simulation
model from Fraser and Rygnestad (1999) and Rygnestad and Fraser
(1996) through a utility function based on the expected profit level, the
profit variance and the individual farmer's attitude to price risk. Note that
as only output price risk is considered in this analysis, all costs are
known with certainty.
Price risk can be described through a probability density function. And
as presented in Fraser (1994) the support system in the EU can lead to a
price distribution as in Figure 9.1.
Frequency
Price
p E(pw)
Figure 9.1: Winsorising a Normal Price Distribution

Figure 9.1 assumes that the underlying world price for wheat has a
normal distribution with an expected value, P, and a known variance,
var(p), and that the intervention price, j3, underwrites the expected world
price. Because the farmer is guaranteed a price of at least j3, the price
support scheme leads to a higher expected producer price, E(pw) , and a
smaller variance, var (Pw).
The expected wheat price and the price variance can be described as
in equations (9.1) and (9.2), respectively (Fraser, 1994):
. . . - ali Z ( P)
E ( Pw ) = F ( P ) P + ( 1 - F ( P ) ) [ p + (1 _ F (p)) J
(9.1 )
var (p ) = [ 1- F ( P)] u? 1 _[ Z ( P~ ]2 + ( P- p) Z ( P~
w P (1- F ( p)) Up (1- F ( p))
(9.2)
where, E(p~w) = expected price with intervention p~rchasing;

F( P) = cumulative probability of price =P ;
P = intervention price;
P = expected world price;
O'p = standard deviation of world price;
var(pw) = variance of price with intervention purchasing;
_1_ -0 5 [ ( p- p ) 1 2
Z(p) = .J2IT e ap
- apZ(p)
E
= p+ 1-F(p)
From the expected price and price variance it is possible to calculate

the expected profit and profit variance for a farm participating in the set-
aside scheme. Extending the model of Rygnestad and Fraser (1996) to
include price variation gives the expected yearly profit, E( 1rtJ, and the
variance of yearly profit, var (1rtJ as (Fraser,1993; Newberry and Sti-
glitz,1981, p.86):
E(Jrl )=2: ~~JL(t,)-a,))Y(N) )E(pw )-~L(t,)-a,))ky"
(9.3)
var( 1rt )= ~~j var( 1rj )+2 ~j-icov( 1rj,1r;)J (9.4)
where, J number of land qualities;

L area of land, ha;
Itj % of farm in year t of land quality j;
atj % set-aside in year t of land quality j;
y{Nj) = actual yield on land quality j, t/ha;
Pw = producer price of wheat, EUR/t;
k = hectare premium, EUR/t;
Ya reference yield, t/ha;
s set-aside premium, EUR/t;
Nj nitrogen application rate for land quality j, t/ha;
Pf = price of fertilizer, EUR/t;
Pmmj semi-fixed production costs, EUR/ha;
FC fixed costs, EUR/ha.
The yield function is maintained as:
Y(Nj) = mj (1- dj e- bjNj ) (9.5)
where, Y wheat yield, t/ha;

N'J = nitrogen application rate for land quality j, t/ha;
mj, dj, bj = parameters of the function;
j = land qualities of the model.
Thus, the total variance of profit is the sum of profit variance from each
land quality with adjustments for the covariance. The covariance between
profit from different land qualities can be described as (Heimberger and
Chavas, 1996):
= E [('1 - E(1rjJ) (nj - E{TrJ) J (9.6)

=E [{ L (Ij - ajJ Yj Pw - L (Ii - au Yi E{pw)}
{L (Ij - ajJ Yj Pw - L (Ii - au Yi E(pw) } J
= L2 {Ij - ajJ (Ii - au Yj Yi E [ (Pw - E(Pw))2 J
= L2 (Ij - ajJ (Ii - au Yj Yi var{pw)
Substituting equation (9.5) into equation (9.4) leads to the formula for
calculating the variance of profit:
(9.7)
Profit variance is a function of farm size, land heterogeneity, the set-

aside rate, wheat yields and the price variance. Together with the ex-
pected profit and a utility function the variance aids the estimation of the
expected utility of profit by a second order Taylor series approximation
shown in equation (9.8):
E( U( p ))= U (E( p ))+~U"( E( p ))var( p) (9.8)

2
For this study the yearly utility function is assumed to be a function of

the profit and the farmer's individual risk preference:
rr t-R
U(rr ) = - - (9.9)
1- R
where, 3 the constant coefficient of relative risk aversion,
R = _U"( I( ) I( .
U'( I( )
The yearly expected utility of profit is discounted over the six year pol-
icy period to arrive at a Net Present Value (NPV) of expected utility to
compare different set-aside schemes:
NPV=~l' rE(V(I(,))
L..,=l (1+r /,-1) (9.10)
The wheat output level is also affected by the introduction of output

price uncertainty through the fertilizer use and yield response functions.
The optimal rates of nitrogen use, Nj, are obtained by maximizing the
expected utility of profit (equation (9.8)) with respect to Nj and setting it
equal to zero (first order condition):
(9.11 )
+!.- U/I/(E( 1(/)) c5 E( 1(/) var( 1(/ )

2 c5 Nj
+!.- U"(E( 1(/ )) c5var( 1(/ ) = 0

2 c5 N;
By differentiating equations (9.3) and (9.7) with respect to Nj one obtains:

(9.12)
(9.13)
On a farm with homogeneous land quality j (Le. Itj = 1 and ati = lti = 0),
equations (9.12) and (9.13) simplify to:
(9.14)
ovar (1() 2 2 0Y
_--,---...::.t~= 2 L (1- at j) var ( Pw ) - (9.15)
ONj ON
By substituting equations (9.14) and (9.15) into (9.11) the optimal nitrogen
rate for land quality j can be found:
~=g(L.aif ,k,s,ya,E(PwJ,P"pp, FC,mj,dj,bj,var(PwJ,R) (9.16)
Note that, compared to the nitrogen rate without risk, the new optimal
rate is not only dependent on the input price and yield response function
parameters. It is also dependent on farm size, the set-aside rate, compen-
satory payments, regional yield, variable, semi-fixed and fixed costs as well
as the expected level and variance of the output price and the farmer's at-
titude to risk. If, on the other hand, the farm has heterogeneous land qual-
ity, the optimal fertilizer rates must be determined by solving a set of si-
multaneous equations due to the covariance (see equation (9.6)).
The leaching function is maintained from the original model:
N',
X t = :1:'7= d ( I tj - atj ) x j e°.7 ( n: -1) + atj Yj 1 (9.17)
where, X nitrate leaching, kg/ha;

Xj standard nitrate leaching (N/lnj = 1) from productive
152 BUd Rygnestad and Robert W. Fraser
land - land quality j, kg/ha;

= ratio between actual and standard nitrogen applica-
tion
on land quality j;
= standard nitrate leaching from set-aside land - land
quality j, kg/ha.
Table 9.1 shows parameter values for prices for season 1998/99. The
world and intervention prices are obtained from publications, and the above
formulas lead to an expected wheat price of 126.13 EU R/t.
Table 9.1: Prices and Risk, 1998/99
Parameter Expected Price Coefficient

Price, EUR/t variance of variation
World price p 112.00 ' 669.36 0.2312
Intervention price P 119.19' 0.00 0.000
Producer price Pw 126.31 159.47 0.100
Notes: 1. Commission of the European Communities (2000) T4.1.5.5, T3.3.1;

2. Fraser (1993).
The economic model used in this paper is based on Rygnestad and Fraser
(1996). The essential modifications are the use of yield response functions es-
timated from the results of Danish scientific field trials (Landsudvalget for Plan-
teavl, 1985-98), and the use of Danish cost and income structures for the
1998/99 growing season, including the appropriate subsidy income (see Table
9.3). Note that whereas the parameter values for the average land quality yield
response function (see Table 9.2) are estimated directly from the results of the
field trials, the parameter values for the poor and good land quality functions
are established by subjectively modifying these direct estimates 4 . In addition
note that the definition of utility in equation (9.9) is not applicable if the expected
profit level is negative. However, with the above base scenario, the estimated
profit from wheat and set-aside are negative on many land quality combinations
- as is also the case in Danish farm statistics on the average net profit of wheat
production (SJFI, 2000 b). A less partial analysis would include other crops as
well as livestock production, thereby leading to a positive farm profit. According
to farm statistics, the average profit on full time farms in Denmark was close to
Table 9.2: Parameter Values for Three Yield Response Functions,

Denmark
Land qualities
Unit Poor Average Good
mj 5.03 8.03 10.03
0 0.49 0.51 0.52
bj 10.24 10.26 10.27
Note: Own estimations based on field trial data from Landskontoret for Planteavl (1985-1998).
Table 9.3: Cost-structure and Other Parameter Values for

Three Land Qualities, no Price Risk, Denmark 1998/99 1
Land qualities
Fertilizer costs, Pr (If EUR/ha 101 134 150
Production costs, Pp m EUR/ha 167 267 333
Fixed costs EUR/ha 838 838 838
Total costs EUR/ha 1,106 1,239 1,321
Note 1. Other data used in the analysis:

k 54 EUR/t; s 69 EUR/t;
a(R) 10%; a(NR) 10%;
Ya 5.22 t/ha; T 6 years;
L 29.2 ha; r 4%
Pf 658 EUR/t.
The fertilizer price is the average price for commercial fertilizer in 1998 weighted by nitrogen content.
Agricultural conversion rate: 1 EUR =7.43 DKR.
Sources: SJFI (2000 b); Commission of the European Communities (2000) ; Statistics Denmark
(2000).
22.000 EUR in 1998/99 (SJFI, 2000a). Thus, to avoid problems of negative

profits, the simulated profit for an average farm (with rotational set-aside and
without price uncertainty) is scaled to match this average number. And this
same scaling is applied to all other farms. When price risk is added this does
lead to small increases in optimal nitrogen rates (and yields and leaching).
However, because the discussion is based on comparing the rotational and
non-rotational set-aside options, the scaling does not affect the conclusions.
Finally, note that Table 9.4 provides details of standard leaching rates for three
land qualities, while Table 9.5 provides details of prices and risk after imple-
mentation of the Agenda 2000 reform and assuming no changes to the world
price for wheat.
Table 9.4: Standard Leaching Rates for Three Land Qualities, Denmark 1
Land qualities
Standard leaching winter wheat, x kg/ha 39
Standard leaching Rot set-aside,r kg/ha 31
Standard leaching NR set-aside, r kg/ha 19
Note 1. Sources: Simmelsgaard, 1991; Schou and Vetter, 1994;

2. Rot =Rotational set-aside; NR = Non-rotational set-aside.
Table 9.5: Prices and Risk after Implementation of the Agenda 2000 Reform
Parameter Expected Price, Price Coefficient of

EUR/t variance variation
World price P 112.00 ' 669.36 0
p
A
Intervention price 101.31 2 0.00 0

Producer price Pw 117.85 345.29 0
Note: 1. Commission of the European Communities (2000) T3.3.1 ;

2. Council of the European Union (1999);
3. Fraser (1993).
9.3. Numerical results

It should be recognized at the beginning of this section that the full re-
sults of the numerical analysis are very complex, commencing with
changes in optimal nitrogen rates from the base case of zero price risk
applying to each of the land qualities. In addition, results for the full
spectrum of land quality combinations requires a three-dimensional
specification of quality variation (see Rygnestad and Fraser, 1996).
Nevertheless, our comprehensive examination of the results shows that
in general the recognition of price risk and risk aversion dilutes the role
of land quality in determining the optimal choice of set-aside option. Spe-
cifically, this effect is manifested as an overall tendency for the ratio of
expected utility for the two options to shift towards unity. This is illus-
trated by Figure 9.2 which is a two-dimensional representation of land
quality variation between poor and average, and which shows for R=0.6
the ratios of expected utility for all land quality combinations before and
after the introduction of price risk. But more significantly, for some land
quality combinations this shift results in the ratio of expected utility
switching from less than unity (i.e. a preference for the Non-rotational
option) to greater than unity (i.e. a preference for the Rotational option).
Moreover, the significance of this "switching" feature of the numerical re-
sults is only emphasized by adjusting the specification of wheat price risk
from that prevailing in 1998-99 (see Table 9.1) to that following from the
Agenda 2000 agreement (see Table 9.5). As indicated by a comparison of
these tables, the implementation of the Agenda 2000 reform would see a
substantial increase in the level of price risk faced by EU producers (i.e.
the CV of producer prices increases from 0.10 to 0.158). And as shown in
Figure 9.3, this increase in price risk results in a strengthening of the ten-
dency for producers with heterogeneous land qualities to switch from pre-
ferring the Non-rotational set-aside option to preferring the Rotational op-
tion. Consequently, this analysis suggests that implementation of the
Agenda 2000 reform would see farmers switching away from a preference
for Non-rotational set-aside towards a preference for Rotational set-aside.
1.015
1.010
it
z
t
e. 1.005
~
~ 1.000
~II
'I--R=O
- 6 - R=0.6
~ 0.995 !
'0
>
IL 0.990
'0
0
i:
II: 0.985
0.980 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _---'
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
% Poor land (reat average)
Figure 9.2: Ratio of Expected Utility for Different Levels of

Poor Land and Risk Aversion.
156 HUd Rygnestad and Robert W. Fraser
Furthermore, this "switching" feature associated with exposing EU ce-

real growers to significant price risk as is intended with the implementa-
tion of the Agenda 2000 reform might be considered inconsequential but
for two reasons:
(i) statistics regarding land quality in Denmark indicates an average
proportion of: 34 percent poor; 52 percent average; and 14 percent good
(Landbrugets Radgivningscenter, 1999). Moreover, as indicated by Fig-
ure 9.3, this "average" farm specification is very close to the range af-
fected by the "switching" feature identified previously. For example Table
9.6, which provides details of the expected utility ratio for a farm with a
land quality combination very close to that of the "average" farm, shows
that a switch of preference for the Non-rotational option to the Rotational
option would occur with the implementation of the Agenda 2000 reform.
(ii) as is also shown by Table 9.6, the implications for output control
and nitrate leaching of this "switching" feature are substantial. In par-
ticular, the output ratio results in Table 9.6 show that for the farm type in
question, the implementation of the Agenda 2000 reform and the associ-
ated switch to the Rotational set-aside option would see an improvement
in output control of 0.7 percent. Moreover, the nitrate leaching implica-
tions of this switch are even more Significant, with an increase in leach-
ing associated with the switch to the Rotational set-aside option in ex-
cess of 3.5 percent.
1. 004 1
« 1.002 '
z
l 1.000 +--,----~__=;::J::::::;::::::*~::::;-==-____:::T:::::,--i
~ . -+- NPV ratio, R=O
:g 0.998
__ Basis, R=0.6
:; ___ Agenda 2000, R=0.6
¥ 0.996
.=-
'0 0.994
0
;},
0.992
0.990
30% 33% 35% 38% 40% 43% 45% 48% 50%
% Poor land (rest average,
Figure 9.S: Ratio of Expected Utility for Different Levels of Poor Land
and Risk Aversion, with the Agenda 2000 Reform.
Table 9.6: Relative Expected Utility (Net Present Value) with Price Risk,
Heterogeneous Land Quality and Different Levels of Risk aversion
Land quality
40% poor, 45% average and Expected util- Output ratio Leaching
15% good soil ity ratio ratio
Risk aversion = 0 0.9967 0.9932 1.0360
Risk aversion = 0.6 0.9998 0.9931 1.0356
Agenda 2000 & 1.0001 0.9928 1.0355
Risk aversion = 0.6
It may be concluded that the implementation of the Agenda 2000 re-

form, in so far as the behavior of Danish cereal growers can be taken as
indicative of the cereal growing sector of the EU, will result in a substan-
tial switching of preference among cereal growers from the Non-
rotational set-aside option to the Rotational option and, associated with
this, improved output control but significantly worse consequences in
terms of nitrate leaching.
9.4. Conclusion
The primary aim of this chapter has been to extend the theoretical model
of set-aside choice outlined in Rygnestad and Fraser (1996) to include
the role of price risk and risk aversion on this choice in the context of the
EU set-aside policy.
It has been shown that the overall effect of the recognition of price risk
and risk aversion is to dilute the significance of the role of heterogeneous
land quality in determining a farmer's preference between the two set-
aside options. In particular, the inclusion of price risk and risk aversion
typically shifts the ratio of the expected utility of the two options towards
unity for all land quality combinations.
More significantly, this shift results in a preference switch from the
Non-rotational set-aside option to the Rotational option for a range of
farms with heterogeneous land quality. In addition, this "switching" fea-
ture is exacerbated by the inclusion of price risk at a level consistent with
the implementation of the Agenda 2000 reform. Specifically, it was shown
for the case of cereal farming in Denmark that the average land quality
combination across all farms is very close to those combinations with the
"switching" feature. Moreover, those farms, which do switch to set-aside
options will exhibit improved output control, but significantly worse nitrate
leaching.
Consequently, it was concluded that the implementation of the Agenda
2000 cereal reforms could see a substantial switch of farmer preference
towards the Rotational set-aside option and, associated with this switch,
improved output control but also increased nitrate leaching. These find-
ings have clear implications for nitrate abatement policies currently oper-
ating within the CAP.
Endnotes
1 Environmental requirements for set-aside areas are detailed in the individual

member countries, and large differences exist (Ansell and Vincent, 1994). It ap-
pears that management rules are based on environmental concerns, and that
these concerns vary greatly between countries. Requirements include establish-
ment of cover crops, allowed methods of set-aside maintenance, and when and if
various farm operations such as chemical use are allowed.
2 Output reduction may also vary because of the enhanced opportunity to reallo-
cate inputs under the Non-rotational option and to supplement grass production
under the Rotational one.
;rl-R
U(;r)=- ; U"( ;r ) = - R;r- R-I
J -R
Leading to:
R = _U"( ;r );rR -rl = - U"( ;r );r = _ U"( ;r );r
;r-R U'(;r)
4 The adjusted R2 for this estimation was 0.54. with all parameter values signifi-
cant at the 5% level. The basis of subjectively modifying these direct estimates
was by reference to the results from Danish field trials for low, average and high
yielding land (Landsudvalget for Planteavl, 1985-1998).
References
Agra Europe (1993a). Changes to EC set-aside will increase 'slippage'. No. 1550, July
9, P/1-P/2.
Agra Europe (1993b). Commission paper exposes weaknesses of set-aside. No. 1543,
May 21, P/1-P/2.
Ansell, D.J. and SA Vincent (1994). An evaluation of set-aside management in the
European Union with special reference to Denmark, France, Germany and the UK.
Centre for Agricultural Strategy Paper 30.
Babcock, B.A., W.E. Foster, and D.L. Hoag (1993). Land quality and diversion deci-
sions under U.S. commodity Programs. Review of Agricultural Economics, 15(3}:
463-471.
Brown, C. (1993). CAP reforms in historical and international perspective. Department
of Agriculture, University of Queensland, Agricultural Economics Discussion Paper
193.
Commission of the European Communities (2000). The agricultural situation in the
Community 1999 report. Office for Official Publications of the European Communi-
ties, Luxembourg.
Council of the European Communities (1999). Regulation amending regulation (EC)
No. 1766/92 on the common organisation of the market of cereals and repealing
regulation (EEC) No. 2731/75 fixing standard qualities for common wheat, rye, bar-
ley, maize and durum wheat. Official Journal of the European Communities, No.
1253/1999, L 160/18-20, 26 June.
Danmarks Statistik (2000). Priser pa vigtige handelsgf2ldninger (Prices of important
commercial fertilisers). Danish Statistics, Statistiske efteretninger 2000:3, Copenha-
gen.
European Commission (1999). Berlin European Council: Agenda 2000, conclusions of
the Presidency. Newsletter.
Firbank, L.G., H.R. Arnold, B.C. Eversham, O.J. Mountford, G.L. Radford, M.G. Telfer,
J.R. Treweek, N.R. Webb, and T.C.E. Wells (1993). Managing set-aside land for
wildlife. Institute of Terrestrial Ecology, Research publication no. 7.
Fraser, R. W. (1993). Set-aside premiums and the May 1992 CAP reforms. Journal of
Agricultural Economics, 44(3}: 41 0-417.
Fraser, R. W. (1994). The impact of price support on set-aside responses to an in-
crease in price uncertainty. European Review of Agricultural Economics, 21: 131-
136.
Fraser, R. W. and H. Rygnestad (1999). An assessment of the impact of implementing
the European Commission's Agenda 2000 cereal proposals for specialist wheat-
growers in Denmark. Journal of Agricultural Economics, 50(2}: 328-335.
Heimberger, P.G. and J.P. Chavas (eds.) (1996). The economics of agricultural prices.
New Jersey: Prentice Hall.
Hoag, D. L., B.A. Babcock, and W.E. Foster (1993). Field-level measurement of land
productivity and program slippage. American Journal of Agricultural Economics,
February, pp. 181-189.
Krause, M. A., J.H. Lee, and W.W. Koo (1995). Program and nonprogram wheat acre-
age responses to prices and risk. Journal of Agricultural and Resource Economics,
20(1}: 96-107.
Landbrugets Radgivningscenter (1999). Handbog i plantedyrknin~ 1999 (Handbook for
arable farming 1999). the Danish Agricultural Advisory Centre, Arhus.
Landsudvalget for Planteavl (1986-1998). Oversigt over landsforsf2lgene. Forsf2lg og
undersf2lgelser i de landf2lkonomiske foreninger (Summary of agricultural experi-
ments). the National Committee on Crop Production, Arhus.
Magid, J., N. Christensen, and E. Skop (1994). Vegetation effects on soil solution
composition and evapotranspiration - potential impacts of set-aside policies. Agri-
culture, Ecosystems and Environment, 44: 267-278.
Newbery, D.M.G. and J.E. Stiglitz (eds.) (1981). The theory of commodity price stabili-
zation. A study in the economics of risk. Oxford: Clarendon Press.
Newbold, C. and A. Rush (1993). Set-aside and extensification of agricultural produc-

tion: Implications and opportunities for nature conservation and the river engineer. In
Agriculture and the environment. Chichester, England: Ellis Horwood Limited.
Roberts, D., J. Froud, and R.W. Fraser (1996). Participation in set-aside: What deter-
mines the opting in price? Journal of Agricultural Economics, 47(1): 89-98.
Rygnestad, H. (1999). The conflict between farmer choice flexibility and policy goals
using set-aside in the Common Agricultural Policy. the University of Western Aus-
tralia, Agricultural and Resource Economics Group, PhD thesis, Perth, Australia.
Rygnestad, H. and R. W. Fraser (1996). Land heterogeneity and the effectiveness of
CAP set-aside. Journal of Agricultural Economics, 47(2): 255-260.
Schou, J. S. and H. Vetter (1994). Regulering af arealanvendelsen i vandindvinding-
somrader (Regulating land use in water catChments). Statens Jordbrugs0konomiske
Institut (Danish Institute of Agricultural Economics), Report No 79, Copenhagen.
Simmelsgaard, S. E. (1991). Estimering af funktioner for kVCElstofudvaskning (Estima-
tion of nitrate leaching functions). In Rude (ed.) KVCEIstofg0dning i landbruget - be-
hoy og udvaskning nu og i fremtiden (Nitrogen fertilizers in Danish Agriculture - pre-
sent and future application for leaching), Statens Jordbrugs0konomiske Institut
(Danish Insitute of Agricultural Economics), Rapport no. 62, Copenhagen.
SJFI (2000a). Landbrugsregnskapsstatistik 1998/99 (Agricultural accounts statistics
1998/99). Danish Institute of Agricultural and Fisheries Economics, Report 83 A.
SJFI (2000b). 0konomien i Landbrugets Driftsgrene 1998/99 (Economics of agricul-
tural enterprises 1998/99). Danish Institute of Agricultural and Fisheries Economics,
Report 83 B (forthcoming).
Sotherton, N. W. (1998). Land use changes and the decline of farm land wildlife: An
appraisal of the set-aside approach. Biological Conservation, 83(3): 259-268.
Waagepetersen, J. (1992). BraklCEgningens betydning for N-udvaskning fra landbrug-
sarealer (the effect of set-aside on N-Ieaching from agricultural areas). in Brak-
ICEgning Planteproduktion og Milj0 (Set-aside, Crop Production and the Environ-
ment) , Statens Planteavlsfors0g (Danish Institute of Agricultural Sciences), No. S
2224.
Webster, C.P. and W.T. Goulding (1995). Effect of one-year rotational set-aside on
immediate and ensuing nitrogen leaching loss. Plant and Soil, 177: 203-209.
10
Production Risks, Acreage Decisions, and
Implications for Revenue Insurance Programs
JunJie J. Wu and Richard M. Adams
In 1996, the Federal Agriculture Improvement and Reform Act ended

over sixty years of direct government subsidies for seven program crops
in the United States. Declining commodity prices since then and associ-
ated farm hardships in many rural areas refocused interest on farm in-
come protection. In 1998, the U.S. Congress approved one-half billion
dollars of funding to develop agricultural insurance. In addition, many
farm organizations now offer their own revenue insurance programs for
farm income protection, covering a variety of agricultural commodities
ranging from Iowa corn to Idaho potatoes. With the increasing use of
revenue insurance for farm income protection, questions have been
raised about their economic and environmental implications. For exam-
ple, will revenue insurance affect cropping patterns and input use? If so,
what are the environmental implications of these insurance programs?
Numerous studies examine the impact of government commodity pro-
grams on crop acreage (e.g., Houck and Ryan, 1972; Chavas and Holt,
1990). More recently, county-level acreage response models are used in
several studies. For example, Wu and Segerson (1995) estimate a
county-level model to examine the effect of agricultural policy on crop-
ping patterns and non-point source pollution in Wisconsin. Lichtenberg
(1989) uses such a model to examine the interaction between land qual-
ity, cropping patterns, and irrigation development. Hardie and Parks
(1997) use a similar model to analyze the impact of land quality on land
allocation between agriculture and forest. These county-level studies,
however, do not examine the impact of insurance programs on cropping
patterns.
The primary objective of this chapter is to explore the relationship be-
tween production risks, cropping patterns, and alternative revenue insur-
162 JunJie J. Wu and Richard M. Adams
ance programs in the Corn Belt (Iowa, Illinois, Indiana, Ohio, and Mis-
souri). Specific objectives include estimating the effects of production
risks on farmers planting decisions, and then measuring how cropping
patterns would be affected by two revenue insurance programs. The ef-
fects of production risks on planting decisions are first measured by es-
timating a system of acreage response equations that use risk measures
as independent variables. The effects of revenue insurance are then es-
timated by simulating the effect of changing production risks and ex-
pected returns on cropping patterns.
The economic and fiscal performance of insurance as a means of farm
income protection is investigated in several studies (Turvey 1992a
1992b; Gray, Richardson, and McClasky, 1994; Hennessy, Babcock, and
Hayes, 1997). Research also focuses on the effect of crop insurance on
chemical use at the intensive margin. For example, Horowitz and
Lichtenberg (1993), in an analysis of Midwestern corn farmers, find that
crop insurance increased fertilizer use by 19 percent and pesticide ex-
penditures by 21 percent. In contrast, Smith and Goodwin (1996) and
Babcock and Hennessy (1996) conclude that crop insurance decreases
fertilizer and chemical use. These studies, however, do not include the
effect of crop insurance on cropping patterns, which may be of greater
importance in affecting total chemical use and environmental quality than
per-acre input use. Wu (1999) examines the effect of federal crop insur-
ance on cropping patterns and chemical use and find that the "extensive
margin" effect of crop insurance can dominate the effect of crop insur-
ance on chemical application rates, leading to an increase in total chemi-
cal use and nonpoint source pollution. Wu (1999), however, focuses on
crop rather than revenue insurance programs on cropping patterns.
In the next section, we present a model to examine the effect of pro-
duction risk on farmers' acreage decisions in the presence of a revenue
insurance program, followed by a discussion of empirical specification,
data, and estimation procedures. The empirical results and policy impli-
cations are discussed in the last two sections.
10.1. Modeling the Effect of Revenue Insurance

on Acreage Decisions
Suppose land quality on a farm is represented by a scalar variable s,

which is normalized such that 0 s S s 1 and its density function
is f (s) . 1 Assume that the farmer allocates the land between two crops.
Crop 1 has higher value but is riskier to grow. Let Ri(s, w) be the per-
acre revenue of crop i on type-s land, where w is a random variable re-
Production Risks, Acreage Decisions, and Revenue Insurance Programs 163
flecting the random state of nature and stochastic market factors that
affect crop yields and prices. w is normalized such that 0 :$ W:$ • De-
note the density function of w as g(w). Let TC i (s, W) be the per-acre
profit from growing crop i on type-s land. The assumption that growing
crop 1 is more profitable and riskier on soil type s implies that
ETC 1(s, w) > ETC 2 (s, w) and var(TC I (s, > var(TC 2 (s, w» Let liCs) w».
be the proportion of type-s land allocated to crop 1. Then the total acre-
age of crop 1 is
1
Al =f r1 (s)f(s)ds (10.1 )
o
The average revenue per acre of crop 1 for a given level of w is
(10.2)
Suppose a revenue insurance program is offered to producers of crop

1, under which the farm will receive an indemnity payment if Rl is below
a guaranteed revenue floor ali (Hennessy, Babcock, and Hayes, 1997),
where R is the insurable revenue, which is defined as the historic (e.g.,
ten-year) average of revenue from growing the crop, and a is the cover-
age level (Le., the percentage of insurable revenue guaranteed). Thus,
the farmer's total net return under the insurance program is
h(s)[aR - c{ (R,a, s) - I(a)j+ (1-Ij(S»1Z"2(S, w)V(s)ds

~oh(s)[~(S,
if ~ < aR
IT(Ij(s),a, w).. 0 (10.3)
w) - c{ (R,a, s) - I(a)j+ (1-Ij(s»1Z"2(s, w)V(s)ds if Rl ~ aR
where J(a) is the premium rate, and c{ (R,a,s) is the per-acre production
cost on type-s land under revenue insurance. Note that revenue insur-
ance may change the profit distribution for crop 1 in two ways: a) by
truncating the revenues, and b) by altering the farmer's input use which
in turn affects production costs and profits. 2 Thus, the production cost
also depends on R and a. Even if the farmer does not receive any in-
demnity payment, profit may still be affected by the insurance payment
because of the effects on input use.
164 lunJie 1. Wu and Richard M. Adams
Assume the farmer's objective is to maximize the expected utility of

profits, then his or her crop mix and insurance decisions can be repre-
sented by
)
Max EU "'fU(II(r) (s),a, w»)g(w)dw, (10.4)

y,(s),a
o
s.t. Osr)(s)sl,
Osasa,
where, the utility function a U(-) is assumed to be increasing, concave,

and twice differentiable, and is the maximum coverage level the farm can
choose.
By using the Lagrangean function, the first-order condition for the opti-
mal land allocation under the revenue insurance program is derived (see
the derivation in the Appendix):
E1C{ (s) - E1C2 (s) - RACov(II, 1C{ -1C2) + fJ(s) - yes) = 0, (10.5)
where, 1C{ (s) is the per-acre profit from crop 1 under revenue insurance
(defined in the Appendix), RA '" -U"(EII)jU'(EII) is the Arrow-Pratt abso-
lute measure of risk aversion, and fJ(s) and yes) are the Lagrange multipliers
for the constraints of 0 s '1(s) s 1and satisfy the Kuhn-Tucker conditions:
P(sh (s) = 0, yes )(1 - f) (s» = O. The third term in (10.5) measures the
farmer's risk premium to grow crop 1. If the additional profit from growing
crop 1, E[1C{ (s) -1C2(S)] , is greater than the risk premium, yes) is positive.
From the Kuhn-Tucker condition, 1j(S)=1, i.e., the type-s land will be allo-
cated to crop 1. On the other hand, if the additional income from growing
crop 1 is less than the risk premium, the land will be allocated to crop 2.
Equation (10.5) indicates that revenue insurance affects crop mix in two
ways. First, it affects crop mix through its impact on farmers' expected
profit. This result is especially relevant because federal crop insurance has
historically been subsidized in the U.S. For example, in March 2000, the
U.S. Congress passed a bill to subsidize insurance premiums by up to 60
percent. The subsidy would increase farmers' expected profits for insured
crops, which would result in more land allocated to these crops. Second,
revenue insurance affects crop mix through its impact on the risk premium.
As shown in the appendix, the risk premium under revenue insurance is
smaller than without insurance. Thus, the farmer will be more likely to plant
crop 1 even if the insurance program is actuarially fair. The more risk
averse the farmer, the larger the effect of the revenue insurance on land
allocation.
This result has clear environmental implications. If the high-value risk-

ier crop uses more chemicals, as is usually the case, then offering reve-
nue insurance may lead to more chemical use at the extensive margin.
The significance of these effects can only be determined through empiri-
cal analysis. In the rest of this chapter, we examine empirically how pro-
duction risks affect the crop mix in the Corn Belt.
10.2. An Empirical Analysis of the Crop-Mix Effect

of Risks and Insurance
10.2. 1. Specification of acreage response models
Consider crop choice for a parcel of land. Let i be an index of crop, with i
= 1, 2, 3 representing corn, soybeans and other uses. Let Uj ;;; Uj(Xj ) + Bj
be the expected profit plus risk premium from growing crop i, where X j is
a vector of variables that affect the profit and risk premium from growing
crop i , and G j is an error term. The first-order condition for the optimal
land allocation problem (10.5) indicates that the farmer plants crop i if
and only if Uj = max(u\>u2,U3) . Under the assumption that the error terms
j are independently distributed with the extreme value distribution
(Maddala, 1983, p. 60), the probability that the farmer will choose to plant
crop i on the parcel, ~, is
(10.6)
(10.7)
where, U3(X 3) is normalized to zero (see Greene, 1993, pp. 697-699). In

empirical applications, ~ is often estimated as the share of land allo-
cated to crop i, that is, 11 = Ai/TA, where TA is total land area, and ~ is
the acreage of crop i. Thus, by taking a log of the ratio of (6) and (7), we
obtain
(10.8)
In this empirical application, Uj (Xj) is specified as

Uj(Xj) = '70j + .L~)'7lkiE(Rd+'72k;V(Rd]+'73iCOV(RI>R2)+'7'4jZj +Vj

where, E(Rj ) is the expected revenue from growing crop i, V(RJ is the
perceived variance of revenue for crop i, Zj is a vector of other variables
that affect the acreage decision, and vi is the error term. Specifically, the
following logistic regression equations are estimated:
(10.9)
with i=1 ,2 and where subscripts j and t are added to indicate county and
year because county-level, time-series data are used to estimate the
equations (see the discussion of data below). .Li= Aijt = TA j is the
total acreage of cropland and potential cropland In c6unty j, which is
assumed to be constant over time. TA j is defined as the maximum of
the 1982, 1987, and 1992 acreages of cropland, pastureland, and
rangeland in county j. The years of 1982, 1987 and 1992 were chosen
because agricultural censuses were conducted in these years and data
were available. One advantage of the logistic specification is that it en-
sures that predicted land use proportions remain between zero and one
and sum to one. In addition, the logit model is a flexible functional form,
and has been shown to outperform other flexible functional forms, in-
cluding the translog (e.g., Lutton and LeBlanc, 1984).
From the logistic regression model, the acreage elasticity of crop i with
respect to an independent variable x in county j in year t can be derived
to be (see Greene, 1993, p. 697):
(9.10)
where, '7kx is the coefficient of x in the equation for crop k, and Sijt is the
share of crop i in county j in year t. Since the acreage elasticity depends
on several estimated coefficients, it is important to provide measures of
statistical significance for the estimated elasticities.
10.2.2. Estimating the effects of revenue insurance on crop mix
Federal revenue insurance programs reduce farmers' production risk by
guaranteeing a revenue floor. The resulting censored distributions of
crop revenue affects both the expected value and variance of revenue.
Since the effects of censoring are best understood in the context of a
normal distribution (Chavas and Holt, 1990), we examine the effect of
censoring on the expected value and variance of revenue by assuming

that crop revenue is normally distributed. The normal distribution has
been widely used for modeling the censoring and truncation effects of
federal commodity programs on crop prices (e.g., Shonkwiler and Mad-
dala, 1985; Holt and Johnson, 1989; Chavas and Holt, 1990). Define the
censored variable
RI = {ali if R<aR
(10.11)
R if R"i! aR.
The expected value and variance of RI are (Chavas and Holt, 1990)
E(R i ) = E(R) + V(R)lj2 [¢(h) + h<l>(h)] , (10.12)
V(RI) = V(R){I- <I>(h) + h¢(h) + h 2<1>(h) - [¢(h) - h<l>(h)] 2 } , (10.13)
where, h = (aR - E(R»/V(R)i/2 , and ¢O and <1>0 are the density and dis-
tribution functions of the standard normal, respectively.
Given the effect of revenue insurance on the expected value and
variance of revenue, the change in the share of crop i in county j under
revenue insurance is estimated by 3
e Wijt
(10.14)
1+ k 2
e Wkjl
where,
2 I
Wijl = 7Joi + }: [1JikiE(~jl)+1J2kr(~jt)]+'l3iCol(,Rljt,R2jl)+17 4i Zi,
k=i
I
Wijl = 7Joi +
}:2 I nI
[1Jiki E(!?kjl) + 1J2kr(~jt)] + 'l3iCol(.l?ljl,R2jl) + 17 4i Zi'
I
k-i .
The correlation between corn and soybean revenues is determined by
weather conditions and by corn and soybean markets. Because corn and
soybeans require similar weather and soil conditions and corn and soy-
bean prices are highly correlated, revenues from growing corn and soy-
beans are also highly positively correlated. This positive correlation re-
duces the effect of revenue variances on the corn-soybean mix because
when revenue from growing corn is low, revenue from growing soybeans
also tends to be low. The correlation variable was included in the acre-
age response equations to reflect this relationship between corn and
soybean. Since revenue insurance neither affects weather nor directly
affects corn and soybean markets, we assume that the correlation vari-
able is not affected by revenue insurance in the simulations.
10.2.3. Data
Historical data on county crop acreage collected by the National Agri-
cultural Statistic Service (NASS) were used for this analysis (downloaded
from the NASS website). The pooled time-series and cross-sectional
data cover counties in the Corn Belt from 1975 to 1994.
Following Gardner (1976), the expected prices for corn and soybeans
were specified as the average futures prices in the planting season,
which were estimated as the average of the first and second Thursday
closing prices in March on the Chicago Board of Trade (CBT) for Decem-
ber corn and November soybeans. During the study period, the expected
price for corn was influenced by government price supports, which
should be reflected in the futures price for corn. In addition, the Acreage
Reduction Programs (ARP) directly affect acreage decisions. Thus, the
ARP rate for corn was included in the model as an independent variable.
The ARP rate was taken from Green (1990). A dummy variable for 1983
is included to account for the effect of the payment-in-kind (PIK) program
offered in that year. Time-series data on input prices were also included
as independent variables, which include farmer wage rates and the
prices (indices) paid by farmers for seed, fuel, and chemicals. All prices
are normalized by the index of prices paid by farmers for all inputs in-
cluding interest, taxes, and wages (U.S. Department of Agriculture, 1971-
1997).
The perceived variances of corn and soybean prices are estimated
following Chavas and Holt (1990). Specifically,
3
V(Pu) = ~ WAPi,l-J - Et-J-1(Pi,t-)f ' (10.15)
where, the weights wJ are .5, .33, and .17, and E t - J - 1 is the expectation,
at planting time in period t-j, of the price for crop i at harvesting in period
t-j. Sensitivity analysis indicates that the results of this analysis are in-
sensitive to the choice of the weights.
Given the expected prices and yields and their variance measures, the
expected revenue and the variance of revenue for corn and soybean are
estimated by the following expressions (Bain and Engelhardt, 1992, p.177):
E(R) = E(p)E(y) + Cov(p,y) , (10.16)
VCR) e E(y)2 V(p) + E(p)2 V(y) + 2E(p)E(y)Cov(p, y) (10.17)
The mean and variance of corn and soybean yields during the study pe-
riod were estimated following Chavas and Holt (1990). Specifically, corn
and soybean yields were first regressed on a trend variable for each
county. The resulting predictions were taken as expected yields, and the
estimated residuals were used to generate the variance of yields and the
correlation between price and yield. The non-truncated correlation be-
tween price and yield was estimated to be -0.293 for corn and -0.149 for
soybeans. Historical data on corn and soybean yields for each county in
our study region were taken from the National Agricultural Statistics
Service.
The land quality variables were derived by linking the National Re-
source Inventories (NRI) with the Natural Resource Conservation Serv-
ice's (NRCS) Soil5 database. The NRI is conducted every five years by
the NRCS to determine the status, condition, and trend of the nation's
soil, water, and other related resources at more than 800,000 sites
across the continental U.S. Each NRI site is assigned a weight (called
the expansion factor or xfactor) to reflect the acreage the site represents.
Each NRI sample site is linked to the NRCS's SOIL 5 database, provid-
ing detailed soil profile information from soil surveys. From the data, av-
erage measures of soil properties for the top soil layers were estimated.
These include average organic matter percentage, clay percentage, soil
pH, and permeability at each NRI site. The data also include information
about soil texture and land capability class. By using the expansion
factor at each NRI site, the average land quality for each county was es-
timated.
Weather data from 1975 to 1992 were obtained from the Midwestern
Climate Center. The mean and variance of temperature and precipitation
during corn and soybean growing seasons were estimated from these
weather data and included in the crop choice model.
The dependent variables, lagged by one year, were also included in
the model to reflect the impact of crop rotation on crop choice. If soy-
beans were planted in the previous year, the current year's crop is more
likely to be corn because a corn-soybean rotation is the most commonly
used cropping system in the Corn Belt. A trend variable is included to
capture the systematic effects of any omitted variables on acreage deci-

sions over time.
10.2.4. Econometric issues

Since pooled time series and cross-sectional data are used, potential
econometric problems associated with the use of panel data may be pre-
sent. Specifically, because the disturbances that affect one crop in one
year may also affect the same crop in other years, autocorrelation may
exist. The presence of autocorrelation was tested using the Durbin test
(Johnston, 1984, p. 318). The null hypothesis of no autocorrelation was
rejected for both corn and soybean equations at the 1% level. 4 In addi-
tion, because county size, cultivation history, and other disturbance fac-
tors differ across counties, heteroskedasticity may also exist. Hetero-
skedasticity was tested using the Lagrange multiplier test suggested by
Greene (1993, p. 467). The null hypothesis of no heteroskedasticity is
rejected at the 1% level for both the corn and soybean equations. 5 Fi-
nally, with land allocation imposing joint production decisions and distur-
bances for different crops reflecting common factors (e.g., climate and
the general state of the economy), contemporaneous correlation (i.e.,
correlation between error terms for different crops) is likely. Contempora-
neous correlation was tested using the Lagrange Multiplier test sug-
gested by Breusch and Pagan (Greene, 1993, p. 515). The null hypothe-
sis of no contemporaneous correlation is rejected at the 1% level of sig-
nificance. 6 Because heteroskedasticity, autocorrelation, and contempo-
raneous correlation all appear to be present, the equation system in
(10.9) was estimated using SUR-HEAR - a procedure that combines the
Seemingly Unrelated Regression technique with Kmenta's method of
handling heteroskedasticity and autocorrelation problems (Wu and
Brorsen, 1995).
10.2.5. Results and implications
The empirical analysis performed here evaluates two aspects of crop

acreage response developed earlier in this chapter. The first involves a
statistical evaluation of the historical relationship between crop acreage
for corn and soybeans and a set of economic and environmental vari-
ables, based on the data derived from 421 counties in the Corn Belt. 7
Acreage response elasticities with respect to these variables are also
measured.
The second aspect of the empirical analysis focuses on the effects of
two alternative crop revenue insurance programs on crop mix. Specifi-
cally, the effects on crop mix of programs guaranteeing 50, 75 and 100
percent of the average revenue during the study period for each crop in
isolation, and then in combination are evaluated. This is achieved by using
Table 10. 1: Estimated Coefficients for the Logistic Regression Model

of Corn and Soybean Acreage Responses
Corn Soybeans
Variable Coefficient t-Statistic Coeffi- t-Statistic
cient
Intercept 0.578** 3.35 -0.341 * -2.14
Price and Policy variables

Expected Corn Revenue 0.237** 15.46 0.009 0.60
Variance of Corn Revenue -2.662** -19.99 -1.439** -12.00
Expected Soybean Revenue -0.142** -6.80 0.094** 4.95
Variance of Soybean Revenue -2.915** -14.89 -1.157** -6.46
Corn & Soybean Revenue Correlation 6.319** 16.57 3.084** 8.90
ARP Rate for Corn -0.006** -9.11 -0.008** -13.98
Chemical Price -0.129** -6.60 -0.228** -12.58
Seed Price 0.111** 7.26 0.162** 11.60
Fuel Price -0.045** -5.51 -0.008 -1.06
Wage Rate -30.954** -18.46 -23.928** -15.57
Land Quality and Weather Variables

% Land of NRCS class I & II 0.014 0.72 -0.058** -3.21
Average Land Slope -0.007** -3.30 -0.027** -11.88
% Medium Textured Soil 0.096** 6.11 0.035* 2.46
% Fine Textured Soil 0.069* 2.48 -0.011 -0.46
Mean Max. Daily Temp. 0.003* 2.09 0.011** 8.00
Mean Precipitation -1.436** -6.34 -1.139** -5.35
Other Variables
Log(corn acres_1/other crop_1) 0.969** 263.46 0.047** 11.82
Log(bean acres_1/other crop_1) 0.012** 2.83 0.925** 184.99
Dummy Variable for 83 -0.662** -48.13 -0.420** -34.35
System Weighted R-Square 0.99
the results from the estimated relationships between crop acreage and
the explanatory variables to simulate changes in county-level crop
acreages under each insurance option. The resulting changes in crop
mix are also explored in terms of their implications for environmental

performance in counties most influenced by the crop revenue programs.
Table 10.1 presents the statistical results from the corn and soybean
acreage response functions, based on the general specification presented in
equation 10.14.
One and two asterisks indicate statistical significance at the 5% and
1% level, respectively.
As is evident from the table, the statistical results are quite robust;
most variables in both equations are significant at least at the 1 percent
level, and only two of the price and policy variables are not significant at
the 5 percent level. Among the weather and land quality variables, pre-
Cipitation and slope are significant at the 1 percent level in both equa-
tions. One and two asterisks indicate statistical significance at the 5%
and 1% level, respectively.
Table 10.2 translates the coefficients from table 10.1 into acreage re-
sponse elasticities, following equation 10.7. In general, signs on the
acreage elasticities are as expected. Variables of particular interest to
the analysis are the revenue and revenue variability measures; increased
revenue has a positive affect on acreages of each crop, while increased
variance has a negative effect. Also, the response in acreage of one crop
is inversely related to the revenue performance of the alternative (and
competing) crop. For example, the acreage response of corn is positively
affected (but inelastic) by an increase in corn revenue but negatively
(and again, inelastic) by an increase in soybean revenues. In fact, all
elasticities are inelastic. Thus, acreage response in those counties is
relatively unresponsive to changes in these variables. This is not sur-
prising, in view of agronomic (rotational) constraints and the relatively
few crops grown in each county.
Most other variables also perform as expected. An increase in the ARP
rate for corn reduces corn and soybean acreage, as does an increase in
chemical, fuel, and labor costs. The acreage elasticities with respect to
the land quality variables indicate that high-quality (NRCS soil classes I
and II) and medium to fine textured soils are more likely to be planted to
corn than to other crops. In contrast, land with steep slopes is more
likely to be allocated to other crops (e.g., hay) than to soybeans. Finally,
a negative acreage elasticity with respect to precipitation for both corn
and soybeans reflects the increasing levels of precipitation and de-
creasing shares of corn and soybean acreage as one moves from west to
east across the study region.
Table 10.2: Estimates of Acreage Elasticities
Corn Soybeans
With Respect To Elasticity t-Statistic Elasticity t-Statistic
Price and Policy variables

Expected Corn Revenue 0.319** 18.51 -0.093** 5.88
Variance of Corn Revenue -0.393** 20.03 -0.106** 6.17
Expected Soybean Revenue -0.194** 9.91 0.160** 9.22
Variance of Soybean Revenue -0.280** 15.75 -0.025 1.58
Corn & Soybean Revenue Correlation 0.598** 16.93 0.121** 3.85
ARP Rate for Corn -0.016** 6.21 -0.032** 13.60
Chemical Price -0.105** 3.62 -0.339** 12.89
Seed Price 0.270** 4.81 0.567** 11.38
Fuel Price -0.133** 6.29 0.021 1.10
Wage Rate -0.507** 17.00 -0.307** 11.49
Land Quality and Weather

% Land of NRCS class I & II 0.015* 1.99 -0.030** 4.37
Average Land Slope 0.002 0.52 -0.064** 13.33
% Medium Textured Soil 0.043** 6.57 0.002 0.36
% Fine Textured Soil 0.004** 3.11 -0.002 1.83
Mean Max. Daily Temp. -0.024 0.35 0.600** 9.12
Mean Precipitation -0.107** 5.84 -0.068** 4.02
The second, and most important, aspect of the empirical analysis con-
cerns how these variables act collectively to determine the aggregate
crop acreage and crop mix within the 421 counties examined here under
conditions of a simulated crop revenue insurance program. The basis for
this simulation of crop acreages, and hence changes in crop mix, is
contained in equation 10.11. The results, for each crop separately and
then for a revenue insurance program encompassing both crops, is re-
ported in Table 10.3.
The effects of a revenue insurance program for either crop, in isolation,
are as expected. Specifically, a revenue insurance program for corn re-
sults in an increase in corn acreage under the 50, 75 and 100 percent
programs. A "corn only" program results in soybean acreage reduction
for the 50, 75, and 100 percent options as soybeans acreage is diverted
to corn. When only soybeans are covered, soybean acreage increases.
However, corn acreage also increases. This is due to the rotational con-
sequence of soybean expansion, given that little continuous soybeans

cUltivation occurs in the Corn Belt. Thus, increases in soybeans result in
a slight increase in corn acreage. In both cases, the acreage response is
much larger under a 100 percent revenue insurance program, as ex-
pected. The increases in the crop acreage for both crops is from land
diverted from the "other" crop category (neither corn nor soybeans).
Table 10.3: Changes in Crop Mix under Revenue Insurance
Coverage Level Corn Soybeans Others
Revenue Insurance for Corn Only

50% 357,029 -29,224 -327,806
(1.65%) (-0.14%) (-0.71%)
75% 2,117,281 -459,921 -1,657,359
(9.78%) (-2.23%) (-3.59%)
100% 4,222,922 -1,361,414 -2,861,507
(19.50%) (-6.59%) (-6.20%)
Revenue Insurance for Soybean Only
50% 23,834 6,855 -30,689
(0.11%) (0.03%) (-0.07%)
75% 294,674 13,282 -307,956
(1.36%) (0.06%) (-0.67%)
100% 321,319 336,894 -658,213
(1.48%) (1.63%) (-1.43%)
Revenue Insurance for both Corn and Soybeans
50% 387,739 -42,337 -345,402
(1.79%) (-0.20%) (-0.75%)
75% 2,556,264 -700,501 -1,855,763
(11.81%) (-3.39%) (-4.02%)
100% 4,912,135 -1,638,812 -3,273,322
(22.69%) (-7.93%) (-7.09%)
Note: The simulations are based on the 1993 price and acreage levels. The insur-
able reve nue levels are assumed to be the average of corn and soybean
revenue from 1985 to 1994. Percentage changes are in parentheses.
In the case of a revenue insurance program that covers both crops, the
results differ across the program options. For example, under the 75 per-
cent revenue option, corn acreage increases by 12 percent, while soy-
bean and "other" acreage declines. Under a 100 percent revenue cover-
age option, corn acreage increases by nearly 23 percent, and by neces-
sity soybean and other crop acreage decreases to accommodate the ex-
pansion in corn acreage. It should be noted that these changes in acre-
age are understated due to the failure to consider moral hazard possibili-
ties. For example, revenue insurance may reduce farmers' use of risk-
reducing inputs. Qualitatively, the "intensive margin" effects would in-
crease the crop mix effect of insurance. That is, adjusting input uses in
the presence of the revenue insurance will increase the profit from the
insured crop, making it more attractive relative to the uninsured alterna-
tive. 8 To adequately address the role of input adjustments requires farm-
level data and specifications, which is not possible using the county-level
specifications and data employed here. However, other research indi-
cates a substantial moral hazard issue in designing federal crop insur-
ance programs (e.g., Chambers, 1989; Goodwin, 1993). Wu (1999) esti-
mated the crop-mix effect of federal crop insurance programs using farm-
level data of crop insurance participation in the central Nebraska Basin
and found a much larger adjustment in crop mix.
One implication of the theoretical model developed in this chapter (and
discussed elsewhere in the literature) is that revenue insurance pro-
grams may affect the environmental performance of agriculture by
changing crop mix. For example, corn and soybeans account for over 70
percent of total cropland acreage in the Corn Belt and use more chemi-
cals than other crops in the study region. Thus, switching from other
crops to corn or soybeans implies negative environmental conse-
quences, such as greater soil erosion or increased use (and runoff) of
agricultural chemicals. More importantly, counties with the largest per-
centage increase in corn acreage also have the greatest potential envi-
ronmental problems, as reflected in the land characteristics data. Spe-
cifically, the average slope for farmland in counties where the percentage
increase in corn acreage is above the median is much steeper than the
average slope for farmland in other counties (4.2 percent slope vs. 2.7
percent slope for all counties). In addition, the percentage of high quality
land (land of NRCS soil class I and II) is lower in counties where the per-
centage increase in corn acreage is above the median; 55 percent of the
land is classified as high quality soils in these counties vs. 70 percent in
counties where the percentage increase in corn acreage is below the
median. The presence of a corn revenue insurance program thus is likely
to lead to an increase in production from environmentally sensitive land,
with associated increases in soil erosion and other types of externalities.
10.3. Conclusions
Revenue insurance programs are an increasingly popular alternative to

direct price supports or Federal farm income support programs. In an era
when the federal government is attempting to extract itself from direct
involvement in agricultural income maintenance, revenue insurance is
viewed as a means of sustaining agricultural producers during periods of
low prices. Such programs are likely to have effects on cropping pat-
terns, particularly when corn or both corn and soybeans are covered.
These effects on cropping patterns may, in turn, have unintended envi-
ronmental consequences.
In this chapter, we explored the relationship between production risk,
cropping patterns and revenue insurance programs. These relationships
are first examined using mathematical and statistical models of farmer
acreage response. An empirical analysis of these relationships is then
performed using economic and environmental data from 421 counties in
the Corn Belt. Specifically, a series of county level crop acreage re-
sponse relationships for corn and soybeans are estimated and then used
in a simulation analysis to predict changes in crop acreage under three
revenue insurance programs. The results from this empirical analysis
confirm that revenue insurance will alter cropping patterns. When an in-
surance program covers both corn and soybeans, increases in acreage
are achieved by reductions in other crops. Whether the insurance pro-
gram covers 50, 75 or 100 percent of revenue alters the magnitude of
acreage changes. Finally, the effects of these acreage changes are likely
to involve environmental consequences, as the counties most prone to
acreage shifts are also those with higher potential for environmental
damage, based on measures such as slope and soil productivity.
This study examines how revenue insurance reduces production risk
and how the reduction in production risk in turn affects crop mix. It does
not model input adjustments and moral hazard possibilities associated
with an insurance program. Nevertheless, the empirical findings corrobo-
rate the theoretical findings reported here and elsewhere and should pro-
vide evidence to policy makers and others concerning the potential ef-
fects of such insurance.
Acknowledgement: This chapter was first published as a paper in the Canadian

Journal of Agricultural Economics. We thank the Journal for giving permission to in-
clude it in this volume.
Endnotes
1. The model presented in this section builds on previous studies of cropland

allocation decisions. Lichtenberg (1989) and Caswell and Zilberman (1985) in-
cluded land characteristics in their models of land allocation, but did not consider
production risk. Chavas and Holt (1990) examined cropland allocation under un-
certainty but did not include land characteristics. Our model of crop mix and in-
surance decisions incorporates both land characteristics and production risks.
2. We thank an anonymous referee for pointing this out.
3. No assumption about insurance premiums was made in the policy simulation
because insurance premiums primarily affect participation decisions.
4. A procedure equivalent to the Durbin test is used here because nvar(bd ~ 1
(see Johnston, 1984, p. 318). The hypothesis of no autocorrelation is rejected
for both corn and soybean equations.
5. The Lagrange Multiplier statistic for group-wise heteroskedasticity is 79000
for the corn equation and 103610 for the soybean equation.
6. The Lagrange Multiplier statistic for contemporaneous correlation is 6197.
7. Sixty-four counties were dropped from the analysis because of lack of land
quality or weather data.
8. We thank an anonymous referee for pointing this out.
Appendix: Derivation of Equation (10.5)
The choice of the coverage level a and land allocation {rl (s)} influences
the range of w in which the farmer receives an indemnity payment. Sup-
pose an increase in w increases the revenue potential. Then the critical
value of w below which the farmer receive indemnity payment, w C , is
defined by
I
fRI (s, wC)rl (s)f(s)ds = AaRI . (A1)
o
Thus, the Lagrangean function for the maximization problem in (10.4)
can be written as
we I
L = fU(rr(rl (s),a, w»)g(w)dw + fu(rr(r1 (s),a, w»)g(w)dw + fla + v(a - a)

o we
I I
+ f J31 (s)'l(s)f(s)ds+ fr1(s)(1-'l(s»f(s)ds, (A2)
o 0
where II(rl (s),a, w) is defined by (10.3), Jl and v are the Lagrange multi-
pliers for the constraint of Os a sa, and fil(s) andrl(s) are the Lagrange
multipliers for the constraints of Os rl (s) s 1.
The first-order conditions for the optimal land allocation can be derived
by differentiating (A2) with respect to 'i(s) for a given s:
w"
-- =
OL U (IT(rj (s),a, we )-av"- + fU'(IT)--g(w)dw
m
orj (S) itj(s) orj (s)
o
c I
- U(II(Ij(s),a, 11"c ~ + JU'(II)~g(W)dW + fil (s) - rl(s) = 0. (A3)
il)(s) il)(s)
w'
The first term and the third term in (A3) offset each other. By substi-
tuting (10.3) into (A3) , we get
~
orj (s)
=
W
f 'U'(I1)[aR; - Cj(R,a,s) -lea) -JZ"2(S, w)k(w)dw
o
°
I
+ JU'(I1)[R 1(s, w) - C1(R',a,s) -1(B) - Jr2(s, w)k(w)dw+ fil (s) - rl (s) =

w'
I
= JU'(II)[Jr{ (s, w) - Jr2 (s, w) k(w)dw + fil (s) - rl (s)

o
= EU(II)[En{ (s,w)-EJr2(S,w)l+co~U(II),n{ - Jr2) + A(S)-rl(S) =0, (A4)
where,
1 _!Rl(S,W)-Cl(S)-I(B) ifw~wC
Jrl (s, w) = _ (A5)
aRI-C1(S)-I(B) ifw<wc
Dividing both sides of (A5) by EU'(II) gives
1 Cov(U'(II) (Jr{ - Jr2»

EJrl(S,W)-EJr2(S,W)+ E~' +fi(s)-r(s) =0, (A6)
where, fi(s) = fil (s)/ EU' and res) = rl (s)/ EU'. By substituting the first-order
Taylor expansion of the marginal utility function U'(II) "" U'(ElI) +
U"(ElI)(II - ElI) into the third term in (A6), it can be simplified to
RP1 -= Cov(U'(II),Jr{ - Jr2) R C (II 1 )

(A7)
- "" A ov Jrl - Jr2 ,
EU'(II) ,
where RA == -U"(ElI)/U'(ElI) is the Arrow-Pratt absolute measure of risk

aversion. Substitute (A7) into (A6), we get equation (10.5).
Now, we show that the risk premium (RP) under the revenue insurance
is smaller than without the insurance program. By definition,
RPJ (s) - RPo (s) = RA [Cov(ll,Jr{ - Jr2) - Cov(llo ,Jrl - Jr2)]
= RA [COV(ll,Jr{ - Jrd + Cov(ll - llo ,Jrl - Jr2)]
= RA [Cov(ll,Jr{ - Jr\) + Cov(ll - llo ,Jr1)] sO (A8)
where the equality holds only when the farmer is risk neutral (i.e.,
RA = 0). The difference is always non-positive because both co vari-
ances in the last brackets are negative. Intuitively, as weather improves,
the indemnity payment Jr{ - Jrl goes down, whereas the total profit goes
up. So the first covariance is negative. Also, as weather improves, 1rl
goes up, and the difference between profits with and without revenue
insurance goes down. So the second covariance is also negative.
References
Babcock, A.B., and D. Hennessy (1996), "Input Demand Under Yield and Revenue
Caswell, M., and D. Zilberman (1985), "The Choice of Irrigation Technologies in
California." American Journal of Agricultural Economies, 67: 224-234.
Chambers, R.G. (1989), "Insurability and Moral Hazard in Agricultural Insurance Mar-
kets." American Journal of Agricultural Economics, 71: 604-16.
Chavas, J.-P., and M.T. Holt (1990), "Acreage Decisions Under Risk: The Case of
Corn and Soybeans." American Journal of Agricultural Economics, 72: 529-38.
Gardner, B. L. (1976), "Futures Prices in Supply Analysis." American Journal of Agri-
cultural Economics, 58:81-84.
Gray, A. W., J. W. Richardson, and J. McClasky (1994), "Farm Level Impacts of
Revenue Assurance." Unpublished manuscript, Department of Agricultural Eco-
nomics, Texas A&M University.
Green, R.C. (1990), "Program Provisions for Program Crops: A Database for
1961-90." Agriculture and Trade Analysis Division, Economic Research Service,
U.S. Department of Agriculture. Staff Report No. AGES 9010. March.
Greene, W. H. (1993), Econometric Analysis. 2nd edition. New York: Macmillan.
Goodwin, B.K. (1993), "An Empirical Analysis of the Demand for Multiple Peril Crop
Insurance." American Journal of Agricultural Economics, 75:425-34.
Hardie, I.W. and P.J. Parks (1997), "Land Use with Heterogeneous Land Quality: An
Application of an Area Base ModeL" American Journal of Agricultural Economics,
79: 299-310.
Hennessy, DA, BA Babcock, and D.J. Hayes (1997), "Budgetary and Producer
Welfare Effects of Revenue Insurance." American Journal of Agricultural Econom-
ics, 79: 1024-34.
Holt, M.T. and S.R. Johnson (1989), "Bounded Price Variation and Rational Expecta-
tion in an Endogenous Switching Model of the U.S. Corn Market." Review of Eco-
nomics and Statistics, 71: 605-13.
Horowitz, J.K. and E. Lichtenberg (1993), "Insurance, Moral Hazard, and Chemical
Use in Agriculture." American Journal of Agricultural Economics, 75:926-35.
Houck, J. P., and M. E. Ryan (1972). "Supply Analysis for Corn in the United States:
the Impact of Changing Government Programs." American Journal of Agricultural
Economics, 54: 184-191.
Johnston, J. (1984), Econometric Methods. Third Edition. New York: McGraw-HilI.
Lichtenberg, E. (1989), "Land Quality, Irrigation Development, and Cropping Patterns
in the Northern High Plains." American Journal of Agricultural Economics,
71 :187-194.
Lutton, T.J., and M.R. LeBlanc (1984), "A Comparison of Multivariate Logit and
Translog Models for Energy and Nonenergy Input Cost Share Analysis." Energy
Journal, 5:35-44.
Maddala, G.S. (1983), Limited-Dependent and Qualitative Variables in Econometrics.
Cambridge: Cambridge University Press.
Shonkwiler, J.S., and G.S. Maddala (1985), "Modeling Expectations of Bounded
Prices: An Application to the Market for Corn." Review of Economics and Statistics
67: 697-701.
Smith, V.H., and B.K. Goodwin (1996), "Crop Insurance, Moral Hazard, and Agricul-
Turvey, C.G. (1992a), "Contingent Claim Pricing Models Implied by Agricultural Stabi-
lization and Insurance Policies." Canadian Journal of Agricultural Economics,
40:183-98.
Turvey, C.G. (1992b) "An Economic Analysis of Alternative Farm Revenue Insurance
Policies." Canadian Journal of Agricultural Economics, 40: 403-26.
U.S. Department of Agriculture (1971-97), Agricultural Statistics, Washington, D.C.
Wu, J. (1999), "Crop Insurance, Acreage Decisions, and Non-Point Pollution." Ameri-
can Journal of Agricultural Economics, 81: 305-20.
Wu, J. and B. W. Brorsen (1995), "The Impact of Government Programs and Land
Characteristics on Cropping Patterns." Canadian Journal of Agricultural Economics,
43: 87-104.
Wu, J. and K. Segerson (1995). "The Impact of Policies and Land Characteristics on
Potential Groundwater Pollution in Wisconsin." American Journal of Agricultural
Economics, 77:1033-47.
11
The Effects of Crop Insurance and
Disaster Relief Programs on Soil Erosion:
The Case of Soybeans and Corn
Barry K. Goodwin and Vincent H. Smith
The consequences of government agricultural subsidy policies for indi-

cators of environmental quality, including soil loss, food safety, and
groundwater and river contamination, have received considerable atten-
tion and have been lightening rods for high profile political debates since
at least the 1960s. Federal crop insurance and other agricultural disaster
assistance programs have not been immune from controversy in this
policy context, perhaps especially in relation to the use of chemical in-
puts. Thus, previous research has examined the effects of federal crop
insurance programs at the intensive margin on the use of potentially en-
vironmentally damaging inputs fertilizer and other chemical input use
(Smith and Goodwin, 1996; Horowitz and Lichtenberg, 1993; Babcock
and Hennessy, 1996). More recently, the interaction of government pol-
icy, especially risk management programs, and crop choice and other
land use issues has also become the focus of considerable debate.
Thus, some recent research has evaluated the extent to which risk man-
agement assistance, through the provision of ad-hoc disaster assistance
and federally subsidized crop insurance, has affected land use patterns.
The debate about the impact of crop insurance and other disaster relief
programs on land use has been complicated by the fact that government
programs are quite heterogeneous in terms of their effects on individual
crops and specific regions of the country. To the extent that the support
provided to one crop may be different than the support provided to oth-
ers, government programs may indeed affect crop planting decisions,
both in terms of crop mixes and, at the extensive margin, the allocation
of land between, crops, pasture, and grazing. In addition, farm policy
may affect production practices. Smith and Goodwin (1996), for example,
182 Barry K Goodwin and Vincent H. Smith
demonstrated that Kansas wheat farmers, who purchased subsidized

crop insurance, tended to use significantly less fertilizer and chemical
inputs.
Empirical research that has directly addressed the extent to which land
use decisions are affected by policy is relatively sparse. A study by Grif-
fin (1996) that focused simply on land use concluded that disaster as-
sistance programs could have offset about one third of the reductions in
soil erosion achieved by the Conservation Reserve Program (CRP). Grif-
fin's research was later extended by Keeton, Skees, and Long (2000),
who estimated that expansions in risk management programs in the
1980s led to the introduction of about 50 million new acres of U.S. crop
land into use, where use was defined to include planted acres, idled
acres, and land in conservation reserve uses. A large proportion of this
increase, approximately 35 million acres, was accounted for by land put
into the Conservation Reserve Program.
An alternative analysis by Young and Vandeveer (2000) suggested
very modest aggregate U.S. acreage responses to the provision of risk
management programs. Most recently, Goodwin and Vandeveer (2000)
found that expansions in participation in the crop insurance program
likely only had a modest effect on acreage decisions for corn and soy-
beans in the corn belt. Their empirical results implied that a fifty percent
decrease in premium rates would only lead to about a two percent in-
crease in corn acreage and a half percent increase in soybean acreage.
In fact, the effect of policy on land use and production practices has
important implications for how policy may influence environmental fac-
tors, including soil loss and chemical and fertilizer runoff. If policy
changes result in growers shifting from less erosive crops toward more
erosive crops (e.g., from soybeans to corn), the agricultural sector may
realize an increase in soil erosion. Likewise, if policy results in de-
creased use of fertilizer and agricultural chemicals, concerns of runoff
and groundwater contamination may be diminished. Concern over such
environmental impacts has been relevant to recent debates over the role
of risk management policies in the U.S. agricultural policy mix.
The modeling exercises presented in previous research on land use
are complicated by the fact that the effects of policy on production and
land use patterns almost certainly vary significantly across regions and
across crops. Griffin's (1996) findings were based on two simple Single
equation empirical models of land use in the Great Plains States. The
first estimated the determinants of the proportion of all acres allocated
either to wheat or used for pasture that were planted to wheat. The
Crop Insurance and Disaster Relief Programs 183
model was estimated using a time-series/cross-sectional data set for the

period 1978-92 in which the spatial units of observation were counties. A
serious problem with Griffin's data set is that information on pasture land
acreage was available only for four agricultural census years (1978,
1982, 1987 and 1992) and thus data for the missing years were obtained
through simple linear interpolation. 1 Griffin's second empirical model ex-
amined changes in land use in the same counties using cross-sectional
data across two distinct three-year time periods (1974-77 and 1989-91).
In this model, the dependent variable was the proportion of total land al-
located to six major field crops and pasture utilized in the production of
the six major field crops.
In contrast, Goodwin and Vandeveer (2000) considered only soybean
and corn production in the corn belt. Their analysiS used pooled county
level data for the period covering 1983-1993, accounting for the fact that
risk management policies were constantly changing over that period. The
importance of recognizing the importance of accounting for the fluidity of
these policies is highlighted by the recent expansions in crop insurance
and the introduction of revenue insurance that have resulted in signifi-
cant expansions in risk management program participation levels.
Moreover, the policy innovations have almost certainly had important
effects on land use choices. The 1996 FAIR Act, for example, signifi-
cantly changed the policy environment underlying crop and acreage de-
cisions by eliminating base acreage restrictions.
In this chapter, we focus on one particular aspect of the debate over the
interaction of risk management policy and the environment soil erosion. We
present econometric results based on time-series/cross-sectional data that
examine the effects of crop insurance programs on soil erosion. The ap-
proach is to estimate a simultaneous equations model of the effects of fed-
eral crop insurance programs, disaster assistance programs and other
government programs on: participation in crop insurance programs for
corn and soybeans at the county level; fertilizer and chemical use; and,
most importantly in the context of this study, observed soil erosion in each
county. The results we report indicate that the effects of federal crop insur-
ance programs on soil erosion may depend on the mix of field crops that
are planted at least as much as on the degree to which land is attracted out
of pasture. These results also provide additional support for the hypothesis
that, farmers with federal crop insurance coverage use smaller amounts of
fertilizer and other chemical inputs.
184 Barry K. Goodwin and Vincent H. Smith
11.1. The Data
Estimating the effects of government programs such as disaster assis-

tance and federal crop insurance on land use and soil erosion is data
intensive. If the unit of analysis is the county, information is needed on
participation in crop insurance programs, crop insurance premiums and
premium rates, and insurance indemnities. In addition, county level data
are needed on farm income subsidies provided through other government
programs such as the commodity programs (deficiency payments, etc.),
federal agricultural loan programs, and other disaster assistance pro-
grams. Also, information on crop planting by relevant crop is required,
as are data on chemical and fertilizer input use. Many studies (for exam-
ple, Goodwin, 1993; Goodwin and Kastens, 1993; Smith and Baquet,
1996; Just and Calvin, 1990; and Griffin, 1996) have emphasized the role
of risk in influencing crop insurance participation decisions. The most
relevant measures of risk in relation to multiple peril crop insurance
(MPCI) are measures of the dispersion of yields (such as the relative
yield variance). Thus, county level historical data on yields for the crops
of interest are also needed.
Data on soil erosion are central to the analysis. Two measures of soil
erosion are adopted in this study. The first is the universal K-factor, a
measure of the inherent erodibility of the soil in each county. This is
measured using data collected by the U.S. Natural Resources Conserva-
tion Service (NRCS) for each county through the NRCS National Re-
sources Inventory (NRI) program. The NRI, a survey of 320,000 sample
units drawn from every county in the contiguous 48 states,2 was admin-
istered in 1982, 1987, 1992, and 1997, although the data for 1997 were
not available for this study. The universal K-factor describes the inherent
erodibility of the soil mapping unit at the point of sample.
Actual soil erosion is determined not only upon the inherent erodibility of
land (which is mainly a function of soil type and location) but also by the
manner in which the land is used. The second measure of soil erosion
used in this study, the soil erosion index constructed by the NRCS, is en-
dogenous to cropping and conservation practices. 3 The NRCS takes ac-
count of the following factors in constructing the soil erosion index: the ki-
netic energy produced by the frequency and intensity of rainfall; the univer-
sal K-factor; the L-factor, which identifies the degree to which run-off is
concentrated; the S-factor which describes the gradient in percent of the
slope of the land; the C-factor which represents cropping patterns and
management (including crop rotation, crop tillage practices, crop history
and crop residues); and the P-factor which describes erosion control prac-
tices in place (such as terracing, contour farming, etc.). The soil-erosion
index is influenced by cropping patterns and production practices, and,
thus, we utilize simultaneous-equations estimators to evaluate determi-
nants of soil erosion as well as the effects of erosion on production prac-
tices, including participation in the federal crop insurance program.
Out of necessity, therefore, the data used in the study are drawn from
several different sources. Unpublished countywide data on premium
rates and participation in federal crop insurance programs for corn and
soybeans were obtained from the Federal Crop Insurance Corporation
Office of Risk Management. Data on county level crop plantings, the K-
factor, and the soil erosion index were obtained from the NRI. Informa-
tion on disaster payments at the county level were obtained from the En-
vironmental Working Group Agricultural Disaster Data Base. Data on
other government payments, fertilizer and other chemical expenditures,
and crop revenues were obtained from the U.S. Department of Com-
merce Regional Economic Information System county tables. Finally,
data on county yields and acreage were obtained from the National Agri-
cultural Statistical Service (NASS).
A complete data set that included land use variables and the endoge-
nous soil erosion index was constructed for three Agricultural Census
years - 1982, 1987 and 1992. A continuous data set that included the
universal K-factor, insurance program participation data, and relevant
economic variables was constructed for the period 1982 through 1992.
We utilize the USDA Economic Research Service's farm resource region
that represents the main corn and soybean producing region designated
as the Heartland. This includes all of Illinois, Indiana, and Iowa as well as
large portions of Missouri and Ohio, and smaller parts of Ohio, Kentucky,
Minnesota, Nebraska, and South Dakota. Farm resource regions are de-
fined to include relatively homogeneous counties. Detailed descriptions
of the variables used in the analysis are presented in Table 11.1 together
with sample means and standard deviations.
11.2. Empirical Results

Several important questions arise in relation to the potential effects of
federal crop insurance programs on soil erosion. One is whether farmers
with more erodible land are more likely to participate in the federal crop
insurance program. Another is whether the federal crop insurance pro-
gram has resulted in more soil erosion. These two questions, while closely
Table 11. 1. Variable Definitions and Summary Statistics

Standard
Variable Definition Mean Deviation
Participation Ratio of net insured acres to planted acres 0.2488 0.1627

(Soybeans)
Participation Ration of net insured acres to planted 0.2907 0.1819

(Corn) acres
Normalized Ratio of real liability purchased to maxi- 0.1562 0.1049

Liability mum possible real liability
Normalized Ratio of real liability purchased to maxi- 0.1721 0.1155

Liability (Corn) mum possible real liability
Premium Rate Premium rate (percent) 0.0453 0.0235

(Soybeans)
Premium Rate Premium rate (percent) 0.0484 0.0215

(Corn)
Rate Per Acre Real premiums paid per insured acre 0.0519 0.0177
(Soybeans)
Rate Per Acre Real premiums paid per insured acre 0.0715 0.0230
(Corn)
Risk Coefficient of variation for preceding ten 16.6642 4.9016
(Soybeans) years of soybean yields
Risk (Corn) Coefficient of variation for preceding ten 22.2885 6.8209

years of corn yields
Acres Planted Thousand acres of soybeans planted (t-1) 72.3897 48.3138

(t-1) (Soybeans)
Acres Planted Thousand acres of corn planted (t-1) 87.0014 59.2954

(H) (Corn)
Crop Revenue Ratio of crop revenues to total marketings 0.5480 0.2003

Proportion revenues
ctd.
Table 11.1 continued
Government Ratio of government payment receipts to 0.0645 0.0309

Payments (t-1) total farm revenues (t -1 )
Fertilizer Ratio of expenditures on fertilizers and 0.0300 0.0133

Expenditures agricultural chemicals to total farm revenues
Disaster County average real disaster payment 5.2840 4.9953

Payments receipts between 1986 and 1994
Universal Universal soil erosion k-factor. represent- 0.3247 0.0460

K-Factor ing inherent erodibility of farm land
Erosion Index NRI index of wind and water soil erosion 4.5418 2.9088
Corn Proportion of agricultural land devoted to 0.4159 0.1400

Proportion corn
Soybean Proportion of agricultural land devoted to 0.3460 0.1082

Proportion soybeans
Proportion Proportion of agricultural land planted to 0.1192 0.1015

Planted to pasture
Pasture
Proportion Proportion of agricultural land planted to 0.0501 0.0400
Planted to grass
Grass
CRP Proportion Percentage of agricultural land enrolled in 0.0421 0.0470
CRP program
related, are clearly distinct. A third important related issue is whether

acreage in particular crops has expanded in areas, which are inherently
more erodible. Multivariate regression analysis needs to be utilized in
order to fully evaluate the relationships among these variables and other
characteristics of individual counties that may be relevant to erosion and
program partiCipation. This evaluation is centered around the first two
fundamental issues.
11.2.1. Do farmers with land that is inherently more erodible have higher
participation rates in federal crop insurance programs?
To obtain some preliminary inSights about this question we estimated
standard single equation ordinary least squares and state level fixed ef-
fects models of participation in federal crop insurance programs for corn

and soybeans in which the universal K-factor is included as an explana-
tory variable. Results for representative models are reported in Table
11.2 using two measures of countywide participation for each crop. The
first, Participation, is the ratio of net insured acres to planted acres for
each crop in each county. The second, Normalized Liability, is the ratio
of the real liability purchased to the maximum possible real liability that
could have been purchased in each county.4 In addition to the universal
K-factor, explanatory variables also included the premium rate, a risk-
rate interaction term (where risk is measured as the coefficient of varia-
tion for county wide crop yield over the previous ten years), fertilizer ex-
penditures, the proportion of total marketing revenues provided by crops
(versus livestock), per-acre county average disaster payments received
over the period 1986-94, government program payments per-acre in the
previous year, and acres planted to the crop in the previous period.
Interestingly, for corn, the regression results reported in Table 11.2 in-
dicate that there is a statistically significant direct relationship between
the universal K-factor, the index of inherent soil erodibility of land, and
participation in crop insurance. In these states, for corn, farms with more
erodible land do appear to be more likely to participate in federal crop
insurance programs. However, statistical significance does not imply a
quantitatively substantial relationship between two variables. Given the
estimated coefficients reported in table 11.2 for corn, the results imply
that an increase in the soil erosion by two standard deviations from its
mean value increases crop insurance program participation (using either
measure) by less than one tenth of a percentage point. For soybeans,
the results reported in table 11.2 indicate that there is no evidence of any
relationship between the inherent erodibility of a farm's soil and its par-
ticipation in federal crop insurance programs. The important finding,
therefore, is that there is no evidence that soybean growers who pur-
chase crop insurance in these states have inherently more erodible land
than those who do not while the relationship between erodible land and
crop insurance program participation appears to be quantitatively unim-
portant for corn.
A second interesting finding is that the results reported by Smith and
Goodwin on chemical use for wheat in Kansas are confirmed for corn and
soybeans in Illinois, Iowa, Minnesota and North Carolina. There is a sta-
tistically significant inverse relationship between such outlays and par-
ticipation in crop insurance. Other explanatory variables also generally
have the expected signs. Finally, the results indicate that producers with
Table 11.2: Parameter Estimates and Summary Statistics for Models of Insurance Purchases
Participation Liability
Soybeans Corn Soybeans Corn
Fixed Fixed Fixed Fixed \:)

""'I
Variable OLS Effects OLS Effects OLS Effects OLS Effects .g
Intercept 0.4444- 0.2339- 0.2600- 0.1781- ;;;-
...,
(0.0219) (0.0351) (0.0173) (0.0230) l::
~
;::
!")
Rate -0.0307- -0.0277- -0.0403- -0.0370- -2.1448- -2.1582- -4.1109- -3.670- (1)
(0.0051) (0.0046) (0.0043) (0.0039) (0.3159) (0.2924) (0.3634) (0.3358) $::)
;::
I:l..
Risk*Rate -0.0001 0.0011- 0.0011- 0.0013- 0.0168 0.0719- 0.1037- 0.1100- t::I
(0.0002) (0.0002) (0.0001) (0.0001) 1:;'
(0.0133) (0.0145) (0.0106) (0.0101)
...,$::)
Acres Planted (t-1) 0.0055- 0.0042- 0.0078- 0.0023 0.0038- 0.0280- 0.0033** 0.0022** ~
""'I
(0.0015) (0.0014) (0.0017) (0.0017) (0.0010) (0.0090) (0.0010) (0.0110) ~
(1)
Crop Revenue -0.0119 0.0741- 0.0543 0.1550- -0.0036 0.0583- 0.0365* 0.0846-
-
~
"'tl
Proportion (0.0337) (0.0311) (0.0373) (0.0348) (0.0210) (0.0200) (0.0227) (0.0271)
Fertilizer Expenditures -0.0455- -0.0223- -0.0169- -0.0042 -0.0311- -0.0175- -0.0074'

~
-0.0157** ~
(0.0054) (0.0055) (0.0065) (0.0068) (0.0036) (0.0036) (0.0043) (0.0044) ...,S!
Disaster Payments 0.0143- 0.0092- 0.0125- 0.0098- 0.0090- 0.0063- 0.0087** 0.0079-
(0.0013) (0.0014) (0.0016) (0.0018) (0.0008) (0.0009) (0.0010) (0.0011)
Universal K-Factor 0.0002 -0.0013 0.0104- 0.0042- -0.000004 -0.0010 0.0062** 0.0026'
(0.0018) (0.0018) (0.0022) (0.0021) (0.0016) (0.0011) (0.0014) (0.0013)
~
Price Elasticity at -0.6483 -0.5758 -0.9909 -0.9196 -0.6274 -0.6255 -1.1736 -1.0544 00
\0
Means
higher relative yield variation are less responsive to premium changes.

This result confirms the earlier findings of Goodwin (1993) and Smith and
Baquet (1996). The results reported in Table 11.2 correspond to price
elasticities ranging from a minimum of -0.65 for participation to a maxi-
mum of -1.17 for liability. These estimates are in the higher end of the
ranges reported in previous research. 5
11.2.2. Does the federal crop insurance program result in higher levels of
soil erosion?
The findings reported in Table 11.2 only provided initial insights about
the issue of whether farmers with more highly erodible land participate in
the federal crop insurance program because they are estimated using
single equations techniques and examine only the inherent erodibility of
land and not actual soil erosion. The issue of whether soil erosion in-
creases as a result of the federal crop insurance program is more appro-
priately evaluated in a simultaneous equations model, in which farmers
are assumed make decisions about program participation for crops and
input use (both with respect to land and agricultural chemicals) at the
same time. As previously noted, actual erosion is likely to be endoge-
nous with respect to production and conservation practices as well as
such policy parameters as crop insurance, disaster relief, and the con-
servation reserve program.
Thus, a four-equation model was estimated in which the endogenous
variables are the county-level soybean and corn MPCI participation
rates, the level of expenditures on fertilizers and other chemical inputs,
and the soil erosion index. The latter takes indirect account of effects
resulting from land use in soybean, corn, and other crop plantings, pas-
ture, and other agricultural uses, because changes in land use drive
changes in the soil erosion index at the county level. The four-equation
simultaneous systems model was estimated using three-stage least
squares for different sets of explanatory variables, including state spe-
cific effects. Results, which were invariant to the inclusion of state-
specific fixed effects, are reported in Table 11.3.
In the simultaneous equations model, the participation equations for
corn and soybeans include the same set of variables as do the single
equation models in Table 11.2, except that the universal K-factor is re-
placed by the (endogenous) soil erosion index in the simultaneous equa-
tions model for which results are reported in Table 11.3. In this model,
there is no evidence of any causal relationship from estimated soil ero-
sion to participation in the soybean MPCI program but there is a positive

relationship from estimated soil erosion to participation for corn. Other
explanatory variables generally have the expected signs in the participa-
tion equations.
The relationship between fertilizer and chemical use and the soybean
MPCI programs in the Heartland region are qualitatively identical to those
reported by Smith and Goodwin for wheat in Kansas. Farmers who use
more of these inputs have lower partiCipation rates while farmers that
participate in MPCI reduce their use of these inputs. This is not, how-
ever, the case with respect to corn. There appears to be a positive rela-
tionship between participation in the corn program and fertilizer and
chemical use. However, these results may simply reflect the fact that
different levels of fertilizer and chemical use are associated with corn
and soybean operations. Our measure of fertilizer and chemical usage
applies to the entire farm and thus changes in the mix of crops, which
have different input requirements may result in changes in fertilizer and
chemical usage.
The most interesting results, from the perspective of the effects of
MPCI programs on soil erosion, are those for the soil erosion index
equation. This equation includes variables such as the proportions of
agricultural land planted to soybeans and corn, the proportions utilized
for grass and pasture, and the proportion enrolled in the CRP. These
proportion variables account for differences in crop planting and land use
across counties. The universal K-factor was also included to account for
variations in the inherent erodibility of the land. The key finding is that
while the soil erosion equation coefficient for corn MPCI participation is
positive and significant at the one percent level, the coefficient for soy-
bean MPCI partiCipation is negative and significant at the 1 percent level.
Moreover, the magnitudes of the two coefficients are relatively similar.
The results suggest that partiCipation by corn producers in the federal
MPCI program is correlated with increased soil erosion. This may reflect
the fact that participation and the intensity of production of these crops
tends to be higher in those parts of the country that experience signifi-
cant soil erosion (for example, Iowa and Illinois). Higher premium rates,
such as those mandated for many crops under the 1994 Crop Insurance
Reform Act, would tend to lower participation rates among producers for
both crops. To the extent that acreage in the two crops is similar, the ef-
fects of these changes on soil erosion, through incentives for changes in
land use, appear to be largely offsetting.
Table 11.3. Three-Stage Least Squares Parameter Estimates and

Summary Statistics for Structural Model of Insurance
Purchases, Fertilizer Use, and Soil Erosion
Soybean Corn Soil Erosion Fertilizer &
Variable Participation Participa- Index Chemical
tion Use
Intercept 0.1353" 0.1104" 0.6130" 4.7963"
(0.0179) (0.0216) (0.0568) (0.3820)
Rate -1.0079" -2.6068"
(0.2110) (0.2839)
Rate*Risk 0.0402" 0.0581"

(0.0091) (0.0070)
Acres Planted (t-1) 0.0013 0.0034"

(0.0063) (0.0007)
Disaster Payments 0.0101 ** 0.0079"

(0.0007) (0.0008)
Crop Revenue Propor- -0.1160** 0.0241

tion (0.0235) (0.0269)
Gov't Payments (t-1) -2.4308"

(0.6825)
Fertilizer Expenditures 0.0134** -0.0013
(0.0056) (0.0073)
Soil Erosion Index -0.0010 0.0090"
(0.0011) (0.0013)
CRP Proportion 1.9956"
(0.4540)
Pasture Proportion -5.8180"
(0.4281)
Grass Proportion -8.7654"
(1.0000)
Universal K Factor -3.4614"
(0.9928)
Soybean Participation -2.2776" -6.7468"

(0.5900) (1.4312)
Corn Participation 3.8861" 8.8073"

(0.5474) (1.2128)
Price Elasticities at -0.2922 -0.7330
Means
System-Weighted R2 0.4622
aNumbers in parentheses are standard errors. Single and double asterisks indicate statistcal
significance at the a = .10 and .05 levels, respectively. Number of observations is 625.
The empirical results presented in the soil equation in table 11.3

should, however, be viewed with some caution. A map-based spatial
analysis (not reported in detail here) indicated that corn MPCI participa-
tion rates are higher than soybean MPCI participation rates on more
highly erodible land. Thus, it is possible that association rather than cau-
sation is being identified in the estimation process.
11.3. Summary and Concluding Remarks
This analysis evaluates the relationship between soil erosion and partici-
pation in the federal crop insurance program for corn and soybeans in six
important states. The results can be summarized as follows. First, par-
ticipation in crop insurance programs tends to be lower in counties where
land is inherently more erodible. Second, over time participation in crop
insurance programs has increased more rapidly in areas that are inher-
ently more erodible, suggesting that participation could actually be higher
in such areas at some pOint in the future. Finally, when endogenous soil
erosion is considered simultaneously with crop insurance participation for
corn and soybeans as well as with fertilizer usage patterns, the results
indicate that insurance purchases by corn producers are associated with
increased soil erosion while insurance purchases by soybean producers
are associated with less erosion.
These results are pertinent to the issue of "extensive margins;" that is,
acreage that is marginally productive and typically considered to be more
highly erodible. However, it is important to note that these margins exist
at two different levels. The first involves production in areas of the coun-
try not usually thought of as important in production of a particular crop.
An example might include dryland wheat production in West Texas. A
second extensive margin involves the marginal land located in major
producing regions. Our analysis is directed toward the latter margin as
we believe the effects of commodity-specific programs such as the MPCI
insurance program are likely to be more significant in such areas. 6
Although our research provides important implications regarding soil
erosion and agricultural pOliCies, much work remains to be done. The
NRI database contains precise measures of the productive capability of
land.? Future research is needed to examine the cultivation capability of
land and its relation to insurance program participation and crop produc-
tion. Further, other crops, including wheat and grain sorghum, may also
have important soil erosion effects that should be investigated in this
context. Finally, the margin in areas with less significant levels of pro-
duction also merits further analysis.
Endnotes
lSuch linear interpolation is commonly applied in analyses that combine Cen-

sus data with data from sources which are reported on a more frequent basis. In
situations where the variable being observed is relatively slow to change from
year-to-year, the measurement error induced by such interpolation may be minor.
However, when variables change substantially from one year to the next, linear
interpolation often introduces significant measurement error.
2Sample units whose primary uses were nonagricultural business or commerce,
recreation, residential, wildlife reservation, transportation, or nonagricultural
waste were not included in the analysis. Likewise, sample units that were cov-
ered with water were dropped from the analysis. Individual sample units were
aggregated to obtain acreage-weighted average measures of the NRI variables.
3The NRCS soil-erosion index plays an important role in determining participa-
tion in the CRP program.
4The average yield over the preceding ten years was used to represent ex-
pected yield for the purposes of calculating a maximum possible liability.
5See Knight and Coble (1997) for a recent review of premium rate elasticity es-
timates.
6That is, if interest is mainly in the effects of the MPCI program for soybeans
on soil erosion, one is best served by examining the important soybean produc-
ing regions rather than regions with minor levels of production.
7The NRI classifies each unit into one of eight capability classes, where the top
four categories are considered to be capable of being cultivated. Future research
could examine production at various marginal levels of cultivation capability.
References
Babcock, B. and D. Hennessy (1996), "Input Demand Under Yield and Revenue In-
surance." American Journal of Agricultural Economics, 78: 416-427.
Goodwin, B. K. (1993), "An Empirical Analysis of the Demand for Multiple Peril Crop
Goodwin, B. K. and T. Kastens (1993), "Adverse Selection, Disaster Relief, and the
Demand for Multiple Peril Crop Insurance." Unpublished research report, Kansas
State University.
Goodwin, B. K. and M. Vandeveer (2000), An Empirical Analysis of Acreage Distor-

tions and Participation in the Federal Crop Insurance Program. Unpublished paper
presented at the Economic Research Symposium on Crop Insurance and Soil Ero-
sion, September 15.
Griffin, P. W. (1996), Investigating the Conflict in Agricultural Policy Between the Fed-
eral Crop Insurance and Disaster Assistance Programs and the Conservation Re-
serve Program. Unpublished Ph.D. dissertation, University of Kentucky, Lexington.
Horowitz, J. and E. Lichtenberg (1993) "Insurance, Moral Hazard, and Chemical Use
in Agriculture." American Journal of Agricultural Economics, 75: 926-35.
Just, R. E. and L. Calvin (1990), "An Empirical Analysis of U.S. Participation in Crop
Insurance." Paper prepared for University of Maryland, Saskatchewan Conference
for the Improvement of Agricultural Crop Insurance, April 1-3, Regina, Saskatche-
wan.
Keeton, K., J. Skees, and J. Long (2000), AThe Potential Influence of Risk Manage-
ment Programs on Cropping Decisions at the Extensive Margin, unpublished manu-
script.
Knight, T. 0., and K. H. Koble (1997), A Survey of U.S. Multiple Peril Crop Insurance
Literature Since 1980. Review of Agricultural Economics, 19: 128-157.
Smith, V. H. and B. K. Goodwin (1996), "Crop Insurance, Moral Hazard, and Agricul-
Smith, V. H. and A. Baquet (1996), "The Demand for Multiple Peril Crop Insurance:
Evidence from the Great Plains." American Journal of Agricultural Economics, 78:
189-201.
Young, E. and M. Vandeveer (2000), Presentation given in organized symposium
AMeasuring Market Distortions of Crop Insurance Subsidies, MEA Meetings,
Tampa, Florida.
Conclusion
Bruce A. Babcock, Robert W Fraser, Joseph N. Lekakis
As stated in the Introduction, the motivation for this book has been
prompted by two main developments in relation to agriculture over the
last decade: (i) the increasing integration of agricultural and environ-
mental policy as society's perception has grown of the relationship be-
tween agricultural activity and the quality of the environment; (ii) the
growing role for risk management skills and instruments in agriculture as
farmers have been increasingly exposed over this period to market as
well as other sources of risk. And, although at first glance it may be
thought that this book comprises a diverse range of contributions to a
very broad topic, a closer scrutiny shows that two key themes run
strongly through both the theoretical and case studies parts of the book.
The first of these, addressed by the theoretical contributions of Paul
Mitchell and David Hennessy (Chapter 3) and David Pannell (Chapter
4), and in the case studies of Amir Abadi and David Pannell (Chapter 7)
and Bruce Babcock (Chapter 8), relates to the importance of a clear un-
derstanding of the factors governing the adoption of environmentally
sensitive farming practices in uncertain conditions. The chapter by David
Pannell is particularly inSightful in this context in its highlighting of three
key "states of farmer awareness" that must be achieved for adoption to
take place:
• awareness of the innovation.

• perception that it is feasible and worthwhile to trial the innova-
tion.
• perception that the innovation promotes the farmer's objec-
tives.
Moreover, David argues that although farmers typically act out of "self
interest", this self-interest is potentially considerably broader than simply
the profit motive. In addition he argues that uncertainty has a number of
negative influences on these "states of farmer awareness". Conse-
quently, he suggests that the opportunity to trial an innovation takes on
particular value when these features of and impediments to the adoption
process are recognized. Specifically, the opportunity to trial allows the
Conlusion 197
farmer to develop a closer working relationship with the new practice,

thereby both reducing uncertainty associated with its value, and promot-
ing its role in the farmer's overall self-interest. Moreover, since the case
studies of this theme emphasized the role of farmer aversion to risk in
the adoption process, it follows that new farming practices which can be
trialled easily and cheaply are more likely to be successfully adopted into
on-going farming systems. This finding is of considerable significance of
the future of sustainable agriculture, so the developers and proponents of
sustainable farming practices must come to terms with the key roles of
farmer self-interest and aversion to risk in the adoption process.
Turning to the second key theme of the book, the theoretical contribu-
tions of Ross Kingwell (Chapter 5) and Rob Fraser (Chapter 6), and the
case studies of Hild Rygnestad and Rob Fraser (Chapter 9), JunJie Wu
and Richard Adams (Chapter 10) and Barry Goodwin and Vince Smith
(Chapter 11), addressed the theme of the importance for policy-makers
both of an understanding of the impact of their policies on producer be-
havior, and of the central role of risk and risk aversion in determining this
impact. These chapters showed that as farmers became increasingly
exposed to market risk, they will increasingly assess policies and policy
changes from the perspective of risk management. It has perhaps been a
recognition of this importance of risk management skills and instruments
in the contemporary agricultural sector that has led to the recent spate of
government studies in this area (European Commission, 2001; USDA,
1999; MAFF, 2001). Moreover, such studies have revealed, particularly
in the EU, an absence of research evidence concerning farmers' atti-
tudes to risk and risk management. It follows that, as useful as the stud-
ies contained in this book have been in highlighting the value of research
in the area of risk management in agriculture, considerably more re-
search waits to be done. In particular, this research needs to be funded
by policy-makers in order to help them formulate risk management in-
struments that work, and this research needs to be directed at assessing
the factors which determine a farmer's willingness-to-pay for risk man-
agement instruments. Only in this way will policy-makers be able to tune
the supply-side of the risk management instrument market to the demand
of the farming sector for such instruments.
But in highlighting the important role of risk and risk aversion in deter-
mining the impact of policy and policy changes on farmer behavior, these
chapters have also identified a key consideration for policy formulation
which aims to capitalize on the fundamental insight of principal-agent
theory: the centrality of the notion of incentive-compatibility to successful
policy formulation. In particular, as shown in the chapter by Ross King-

well, this incentive-compatibility needs to deal with the issue of moral
hazard as it arises in an uncertain production environment. Specifically,
policymakers can capitalize on the growing exposure of farmers to mar-
ket risk and on the pervasive finding that farmers are risk averse to
"manage" them into policy-compliance. This involves the use of penalties
if detected "cheating", and of the role of monitoring systems in determin-
ing the probability of detection to promote farmer policy-compliance,
which is in society's best interests. This finding is of particular signifi-
cance in the context of agri-environmental policy, where it is the case
that farmer self-interests, and what is in the best interests of society, are
much more likely to be at odds, and therefore where an element of policy
enforcement is required. Appropriate recognition by policy-makers of the
potential for manipulating farmers' use of risk management strategies to
include being policy-compliant may well ease the task of policy-
formulation in the context of the contemporary agricultural sector.
Therefore, we can conclude this book by observing that although its
proposal may well have seemed a venture into deep and expansive wa-
ters, its realization has been the identification of the defining features of
the future research agenda for risk management and the environment in
agriculture.
References
European Commission (2001) "Risk Management Tools for EU Agriculture" Working
Document, Agriculture Directorate-General, January.
MAFF (2001) "Risk Management in Agriculture" Discussion Document, Economics
and Statistics Group, Janauary.
USDA (1999) "Managing Risk in Farming: Concepts, Research and Analysis"
Agricultural Economics Report No. 774, Market and Trade Economics Division and
Resource Economics Division.
Author Index
Abadi Ghadim, A.K. 68, 92,114, Caswell, M. 177

129 Chambers, R.G. 11, 12, 15, 18,
Abbey, A.43 19, 20, 21, 22, 23, 24, 25, 26,
Aldy, J.E. 53 54, 175
Alreck, P. L. 118 Chavas, J.P. 149, 161, 166,
Anderson, J.R.114, 115, 118 169,177,
Anderson, K. 96, 109 Choi, E. K. 138
Ansell, D.J. 158 Christensen, N.
Antle, J.M. 69 Clark, D. 89,92
Arnold, A.J. 132, 159 Conrad, J.M. 40
Arnold, H.R. Cooper, J. 40, 53
Arrow, K.J. 11,15,115 Crocker, T. 26
Ayton, P.114 Cummings, F.H. 54
Babcock, B.A. 3, 53, 54, 59, Davenport, S. 43

135,138,146,162,163,181 de Vaus, D.A. 118
Baquet,A. 184, 190 Debreu, G. 11, 15
Bardsley, P. 40, 43, 69, 92, 114, Dillon, J.L. 131
121, 128 Dixit, A. K. 31
Bar-Shira, Z. 115 Doster, D.H. 81, 132
Bartle, J.R. 80 Drees, B.M. 66
Barton, J.H. 83
Black, F. 41 English, B.C. 40
Boes, D.C. 109 Ernst, U.F.W. 54
Bond, G. 114,121,128 Eversham, B.C. 159
Brorsen, B. W. 81, 132
Brown, C. 146 Faires, J.D. 39
Burden, R. L. 39 Feather, P. 53
Burton, M. 80, 131 Feder, G. 67,113,116
Feinerman, E. 138
Calvin, D.D. 138 Fernandez-Cornejo, J. 95
Calvin, L. 40, 184 Firbank, L.G., 146
Campbell ,C. 80 Fischer, A.J. 116
Campbell, R. 109 Foster, W.E. 159
Carpenter, J. 82 Fox, G. 142
Cary, J.W. 70,80 Francis, D.D. 57
Cashin, P. 40 Fraser, R.W. 69, 89, 122, 145,
200 Author Index
146,147,148,154 Hennessy, D.A 3, 54, 59, 135,

Froud, J. 160 142,162,163,181
Fulton, M. 82 Hertzler, G. 31, 115
Hey, J.D. 114
Gardner, B.L. 109,168 Haag D.L. 146, 159
Gianessi, L. 82 Hochman, E. 81
Gibbs, M. 127 Hoffman, L. 82, 84
Goble, K. 1 Holt, M.T. 161, 166, 167, 169,
Goodwin, B.K. 3, 54, 162, 175, 177
181,182,183,184,190 Horowitz, J.K. 3,162,181
Goulding, W.T. 146 Houck, J.P. 161
Gray, A W. 162 Hrubovcak,J.53
Graybill, F.A. 109 Hsiao, C. 117
Green, R.C. 168 Huirne, R.B. 132
Greene, W.H. 53, 165,170 Hull, J.C. 36,38,39
Griffin, P. W. 182, 184
Grimmett, G.R., 31,32 James, S. 96, 109
Grossman, S. 22 Janes, S.95
Gruen, N. 67 Jarrett, F. G. 81
Jarrow, R.A. 36
Hafi, A, 96, 109, 110 Johnson, S.R. 167
Hamblin, J. 113 Johnston, J. 170, 177
Hancock, W.M. Jovanovic, B. 116
Hanson, S.D. 89 Just, R.E. 40, 109, 131, 135,
Hardaker, J.B. 114, 115, 118, 184
120
Hardie, I.W. 161 Kalaitzandonakes, N. 83, 135
Harman, J. 31 Karagiannis, G. 8
Harper, J.K., 63 Kastens, T. 184
Harris, M. 69, 92, 114, 121, 128 Keeton, K. 182
Harrison, J.L. Keim, R. 66
Hart, 0.22 Keyowski, L. 82
Harwood, J. 1 King, D.A. 70, 80
Haszler, H. 109 Kingwell, R.S.117, 122
Havenner, A 142 Klotz-Ingram, C. 82
Hayenga, M. 83 Knight, T.O. 194
Hayes, D.J. 162, 163 Koble, K.H. 194
Heisey, P.W. 67, 113 Koo, W.W. 159
Hellerstein, D. 66 Kramer, R.A. 66,117,118,143
Heimberger, P.G. 149 Krause, M.A. 146
Author Index 201
Newbold, C. 146
Ladd, G.W. 89 Norris, P. E.117, 118
Lantzke, N. 118 Norton, G.W. 66
Latacz-Lohmann, U. Nowak, P. 53
Leathers, H.D. 67, 113 Nunn, M. 96, 109
LeBlanc, M.R. 166 Nyarko, Y. 116
Lee, J.H. 159
Lichtenberg, E. 3, 134, 142, Oliver, M. 96
161,162,181 Olson, K. 53
Lindner, R.K. 31, 67, 68, 69, Osborn, C. 63
81,83,113,116,127
Lockie, S. 67 Pannell D.J. 67, 68, 78, 80, 81,
Long, J. 182 114,129
Long, T. 89, 92 Pardey, P.G. 81
Luenberger, D. G. 14 Parks, P.J. 161
Lutton, T.J. 166 Perry, J. 1
Petzel, T.E. 40
Machina, A. 115 Pindyck R. S. 31
Maddala, G.S. 165, 167 Pluske, J. 69
Magid, J. 146 Pope, R.D. 109, 135, 143
Mahul, 0.40 Pratt, J.W. 115
Malcolm, R.L. Preckel, P. V. 135
Malmquist, S.14 Pyke, B. 89
Maltsbarger, R. 83
Marsh, S.P. 70 Quiggin, J. 3,11,12,15,18,19,
McBride, W. 95 20, 21, 22, 24, 25, 26, 27, 40,
McClasky, J. 162 43, 54
McFadden, D. 16
McPherson, R.M. 66 Radford, G.L.159
Merton, R.C. 36, 38 Rae,J.67
Miranowski, J.A. 54 Rajotte, E.G. 66
Mirrlees, J. 24, 25 Rendleman, C. 66
Mjelde, J.W. 66 Reynolds, R. 96
Mood, A.M. 109 Ribaudo, M. 66
Mountford, O.J. 159 Richardson, J. W. 162
Myers, R.J. 69 Riley, P.A. 82
Nayda, W.1. 40 Rister, M.E. 66
Roberts, D. 146
Nelson, C.J. 135 Robinson, S.D. 132
Newbery, D.M.G. 109, 148 Rogers, E.M. 74
202 Author Index
Rush, A. 146 Telfer, M.G. 159

Ryan, M. E. 161 Tollefson J.J. 57
Rygnestad, H. 145, 146, 147, Treweek, J.R. 159
148, 154 Tsur, Y. 67,113
Turvey, C.G. 162
Saha, A. 142
Savage, L.J. 12 Umali, D.L. 67,113
Schepers, J. S. 57
Schilizzi, S. 78 Vanclay, F. 67
Schoemaker, P.J. 115 Vandeveer, M. 182, 183
Scholes, M. 41 Varvel, G. E. 57
Schoney, RA 113 Vasavada, U. 53
Schou, J.S. 154 Vetter, H. 154
Segerson, K. 161 Vincent, S.A. 158
Settle, R.B. 118
Shapiro, B.1. 67,113,116 Waagepetersen, J. 146
Shephard, 14 Way M.O. 66
Shogren, J. 26 Webb, N.R. 159
Shonkwiler, J.S. 167 Webster, C.P. 84, 146
Shumway, C. R. 142 Weersink A. 142
Siddique, K.H.M. 113 Weisensel, W. P. 113
Simmelsgaard, S. E. 146 Wells, T.C.E. 159
Sinden, J.A. 70, 80 Westra, J. 53
Skees, J. 182 White, G. 80
Skop, E. 159 Wilkinson, R.L. 70, 80
Smale, M. 67,113,117 Wonder, B. 114, 121, 128
Smidts, A. 114 Wright, B.D.
Smith, V. H. 3, 54, 162, 181, Wright, G. 83, 114
184, 190 Wu, J. 53,161,162
Sotherton, N.W. 146
Spike, B.P. 57 Young, J.M. 132
Stanton, J. 8 Young, E. 182
Sternberg, M. 81
Stiglitz, J.E. 109, 148 Zilberman, D. 131, 134, 142,
Stirzaker, D. R. 31, 32 177
Stokes, J. R. 40
Stoneman, T. C.118
Sykes, J. 113
Tanner, C. 109
Subject Index
Acreage awareness of, 68

decisions, 162 perception of, 69
response models, 165
Adoption Insurance
of agricultural innovations, Crop, 181
67, 68 Green, 52, 56, 61
behavior 116 IPM,53
of best management initia- Revenue, 53, 161, 174
tives, 52, 59
impact of uncertainty on, 70 Investment theory, 30
process, 71 Irreversibility, 78
of sustainable farming, 67
uncertainty of, 70 Land
Agenda 2000,2, 145 heterogeneity, 76, 146
management practices, 52
Contracts
forward, 1, 36 Moral hazard, 3, 22
futures, 36
production, 36 Optimal decisions, 35
Options
Cost functions, 15, 16 on formal contracts, 36, 39
on the environment, 33, 36
Crop on futures, 36, 37
genetically modified, 82 on resource stocks, 41
insurance, 182
yields, 135 Pest control, 134
Pesticides, 17
Distri bution, parametrized, 25 Pollution runoff, 22
Production
Externalities, 22, 75 functions, 22, 23
technology, 26
Fertilizers, 17 risks, 161
under uncertainty, 11
Hedging, 34 Profit
function, 72
Innovation maximization, 84
adoption of, 67, 68
204
Programming, 31
Quarantine restrictions, 96
Regression model, 171

Revenue
insurance, 53, 161
non-stochastic, 16
Risk
attitudes, 113, 114
aversion, 19, 33, 134
complements and substi-
tutes, 17
environmental, 12
and insurance, 165
measurement of, 117
neutral farmers, 87
preferences, 120, 128
premiums, 14, 33
price, 145
subjective assessment of,
118
and uncertainty, 118
Soil erosion, 184

Sustainable farming, 71, 72
Uncertainty
about conservation innova-
tion, 74
and risk, 118
Utility
expected, 29, 115
marginal, 33
maximization
Wealth,46

Risk Management and The Environment: Agriculture in Perspective

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Risk Management and The Environment: Agriculture in Perspective

Uploaded by

Copyright:

Available Formats

RISK MANAGEMENT

AND THE ENVIRONMENT:

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

ISBN 978-90-481-6158-4 ISBN 978-94-017-2915-4 (eBook)

Printed an acid-free paper

AII Rights Reserved

The State-contingent Approach to

2 The Precautionary Principle in Practice:

3 Factors Determining Best Management

4 Uncertainty and Adoption

5 Incentive Design for Introducing

6 An Economic Risk Assessment of the

Part II Case Studies

7 Risk Attitudes and Risk Perceptions

8 Valuing Pest Control:

9 The Influence of Price Risk

10 Production Risks, Acreage Decisions, and

11 The Effects of Crop Insurance and

Author Index 199

Table 1. 1: Problem Taxonomy 20

3. 1: Sequence of Events for Each Production System 58

5. 1: Parameter Values for the Numerical Analysis 89

5.2: Base Case Findings 90

5.3: Sensitivity Analysis of Problem of Illegal Production

6. 1: Base Case Results of the Perception of the

6.2: Perception of the Impact of Introducing Imports on

6.3: Perception of the Impact of Introducing Imports

6.4: Perception of the Impact of Introducing Imports on

6.5: Perception of the Impact of Introducing Imports

6.6: Perception of the Impact of Introducing Imports

7. 1: A Sample of an Elicited and a Contract Price Scheme

7.2: Perceived Mean Yields for Chickpeas and Wheat 124

7.3: The Coefficient of Variation of Chickpea and Wheat

7.4: The Range of Covariance Values Observed among

7.5: Proportion of Farmers with Different Levels

8. 1: Per-Acre Willingness to Pay for 8t Corn 141

9. 1: Prices and Risk, 1998/99 152

9.2: Parameter Values for Three Yield Response

9.3: Cost-structure and Other Parameter Values for

9.4: Standard Leaching Rates for Three Land Qualities,

9.5: Prices and Risk after Implementation of the Agenda

9.6: Relative Expected Utility (Net Present Value) with

10. 1: Estimated Coefficients for the Logistic Regression

10.2: Estimates of Acreage Elasticities 173

10.3: Changes in Crop Mix under Revenue Insurance 174

11. 1. Variable Definitions and Summary Statistics 186

11. 2: Parameter Estimates and Summary Statistics for

11. 3. Three-Stage Least Squares Parameter Estimates and

Figure 2. 1: Prices of a European Call Option on Futures 38

2.2: Values of an Option on Forward Contracts 40

2.3: Prices of a Put Option on Futures 41

2.4: Prices of a Call Option on Resource Stocks 42

2.5: Prices of a Call Option on Financial Stocks 43

2.6: Prices of a Call Option on the Environment 44

2.7: Prices of a Put Option on Futures 45

5. 1: An Illustration of the Farmer's Decision Problem 86

7. 1: A Sample Elicited Yield Distribution Table 119

8. 1: Effects of Bt Corn on Distribution of Corn Yields in

8.2: Effects of Bt Corn on Distribution of Corn Yields in

9.1: Winsorising a Normal Price Distribution 147

9.2: Ratio of Expected Utility for Different Levels of

9.3: Ratio of Expected Utility for Different Levels of