KOUAME Euphrasie Ben Houassa

MINISTERE DE L'ENSEIGNEMENT SUPERIEUR ET DE LA RECHERCHE SCIENTIFIQUE

UNIVERSITEFELIX HOUPHOUËT-BOIGNY COCODY-ABIDJAN

UNITE DE FORMATION ET DE RECHERCHE DES SCIENCES ECONOMIQUES ET DE GESTION (UFR-SEG) Année académique 2013-2014 L- THESE PHD présentée par : KOUAME Euphrasie Ben Houassa

Soutenue publiquement le 24 Janvier 2014

Pour obtenir Je grade de : DOCTEUR DES UNIVERSITES PUBLIQUES DE COTE D'IVOIRE

~!inetS. pjcialitl': Sciences Economiques/Economie du Développement

THESE dirigée par : M. AKE N'GBO Gilbert Marie Professeur Titulaire, Université Félix Houphouët-Boigny, Côte d'Ivoire

JURY: Président: M. DIA W Adama Professeur Titulaire, Université Louis Gaston Berger, Saint Louis, Sénégal Membres : M. AKE N'GBO Gilbert Marie Professeur Titulaire, Université Félix Houphouët-Boigny, Côte d'Ivoire M. KOUASSI Eugene Maître de Conférences, Université Félix Houphouët-Boigny, Côte d'Ivoire M. OUATTARA Wautabouna Maître de Conférences Agrégé, Université Félix Houphouët-Boigny, Côte d'Ivoire M. ADAIR Philippe Maître de Conférences Agrégé HDR, Université Paris Est Créteil, France M. ACCLASSATO Denis Maître de Conférences Agrégé, Université Abomcy-Calavi, Bénin M.BALLOZié Maître de Conférences Agrégé, Université Félix Houphouët-Boigny, Côte d'Ivoire RISK A VERSION AND AGRICULTURAL DECISION-MAKING UND ER UNCERTAINTY: EVIDENCE FROM COCOA FARMERS IN WESTERN

CÔTE D'IVOIRE

A Thesis presented to the Faculty of Economies and Management, Felix Houphouet-Boigny University ofCocody, Abidjan, Côte d'Ivoire

In Partial Fulfilment of the Requiremcnts for the Degree of Doctor of Philosophy

by Kouamé Euphrasie Ben-Houassa June,2013 DEDICATION

To my Nissy and my late Father

ii DISCLAIMER

The Felix Houphouet Boigny University will not give any approbation to the opinions found in this dissertation. They have to be considered as persona[ to the author.

iii ACKNOWLEDGEMENTS

First I am very much grateful to the African Economie Research Consortium (AERC) for giving me the opportunity to pursue my studies under his Collaborative PhD Programme.

I am highly indebted to my superviser, Professer N'Gbo Aké Gilbert Marie for bis guidance, support, and faith throughout the entire thesis process. Without him this thesis would not have developed the way it did. Neither words can fully articulate his role in the materialization of this effort.

Sincere thanks are extended to my internai defense committee members, Professer Marna Ouattara, Professer Wautabouna Ouattara and Professer Zié Ballo for their constructive criticism and recommendations for this study. I would like to extend my appreciation to Dr Mahmud Y esuf and Dr Abbi Kedir for their expert advice and opinions. I am also thankful to Mr. Carlin Ama from the National Institute of Statistics (INS) who helped me in organizing the fieldwork. Without the kind cooperation of the respondents (farmers), this study wouldn't have been possible. In this regard, I would like to thank the farmers for their hospitality and willingness to take substantial time off work to take part of the experimental sessions and complete the survey questionnaire. I also thank the staff of the Agriculture and Rural Development Bureau (ANADER-Soubré) for their invaluable advice during the data collection period. I would like to thank my friends and colleagues, Dr Narcisse Komenan, Dr Marcellin Brou Bosson, Dr Obert Pimihdzai, Dr Alastaire Alinsato, Edith Togba, Dorothé Yong and Christian Kamala for their continuous support, encouragement, and enthusiasm throughout the completion ofthis dissertation.

Finally, I am especially thankful to my family for their love and support. To my parents, for their prayers and encouraging words, which helped me to overcome many obstacles. In particular my best of gratitude goes to my mother for raising and taking care of my daughter Nissy. Thanks 'MAMA'. I am especially grateful to my lovely daughter Nissy. Nissy, you are truly the oxygen that sustains me. I wish also to express my special thanks to my sisters and brothers for the love and friendship that we have maintained through the years. Thank you ail for having been there for me. Last but not least I wish to express my special thanks to Michel Binaté for his continuous understanding, encouragement and faith in me.

iv TABLE OF CONTENTS

DEDICATION ii DISCLAIMER iii ACK.NOWLEDGEMENTS iv

TABLE OF CONTENTS V

LIST OF TABLES X LIST OF FIGURES xi ABSTRACT xii RESUME xiii LIST OF ACRONYMS AND ABBREVIAT I ONS xiv

GENERAL INTRODUCTION 1 1. Background of the study 1 2. Problem statement 3 3. Objectives of the study 4 4. Significance of the study 5 5. Outline of the dissertation 6

CHAPTER ONE: RISK, RISK AVERSION AND RISK MANAGEMENT IN AGRICULTURE. 8 1.1. Introduction 8 1.2. Sorne basic concepts and definitions 9 1.2.1. Defining risk and uncertainty 9 1.2.2. Measuring risk 10 1.2.2. l. Risk as chance of a bad outcome 10 1.2.2.2. Risk as variability of the outcomes 11 1.2.2.3. Risk as uncertainty of the outcomes 11 1.2.2.4. Limitations of the risk measures 11 1.3. Theories of decision making under risk and uncertainty 12 1.3 .1. The expected utility theory 13 1.3.1.1. Axioms of the EU theory 14 1.3.1.2. Violation of the EU theory 16 1.3.2. The non-expected utility model: The Prospect Theory 21 1.3.3. Risk aversion and stochastic dominance 24 1.3 .3 .1. Risk aversion and concavity of the utility function 24 1.3.3.2. Certainty equivalent and risk premium 25 1.3.3.3. Measures ofrisk aversion 27

V 1.3.3.4. Stochastic dominance 29 1.4. Types of risk in agriculture 30 1.4.1. Sources of risk 30 1.4.2. Spread and magnitude of risk 32 1.4.3. Impact ofrisk in farming 32 1.5. Risk management strategies in farming 34 1.5.1. fa-ante strategies 35 1.5.2. Ex-post strategies 37 1.6. Conclusion 39

CHAPTER TWO: OVERVIEW OF THE COCOA SECTOR IN COTE D'IVOIRE 40 2.1. Introduction 40 2.2. Country overview and the study area 41 2.2.1. Geographical settings 41 2.2.2. Population-Poverty situation and profile 43 2.2.2.1. Demographic situation 43 2.2.2.2. Profile of poverty in Côte d'Ivoire 44 2.2.3. The study area: Biophysical environment 44 2.2.3.1. The study region 44 2.2.3.2. The department of study 45 2.2.4. Relevance of the department for the study 46 2.3. Agronomie and historical background ofcocoa production 47 2.3.1. Origin and technological features of the cocoa tree 47 2.3.2. Production pattern 48 2.3.3. Harvest and Marketing Period 50 2.3.3.1. Themaincrop 50 2.3.3.2. The mid-crop 51 2.3.4. Cocoa production zones in Côte d'Ivoire 51 2.4. Cocoa marketing system 52 2.4. l. Pre-liberalization marketing system 52 2.4.1.1. Role of the CAISTA B until the beginning of the 1990s 52 2.4.1.2. The reforms of the l 990s 53 2.4.2. Total liberalization of the cocoa sector 53 2.4.2.1. Contents of the liberalisation 54 2.4.2.2. Consequences of the liberalisation 54 2.4.3. New reforms in the cocoa sector 55

vi 2.4.3.1. The refonns of the 2000s 55 2.4.3.2. Restructuring of the sector 56 2.4.3.3. Recent developments 57 2.4.4. Cocoa market structure 58 2.4.4.1. Major actors in the cocoa marketing chain 58 2.4.4.2. Cocoa value chain 60 2.5. Constraints faced by farmers in Côte d'Ivoire cocoa sector. 60 2.5.1. Domestic constraints 61 2.5.1.1. Pest/diseases of the plants 61 2.5.1.2. Taxes in the cocoa sector 62 2.5.1.3. Other domestic constraints 64 2.5.2. Global constraints 64 2.5.2.1. Price volatility 64 2.5.2.1. Determinants of cocoa price volattility 67 2.6. Conclusion 69

CHAPTER THREE: MODELLING FARMERS' BEHA VIOURAL RESPONSES TO RISKS 70 3.1. Introduction 70 3.2. Sample design and survey questionnaire 70 3.2.1. Sampling procedure 70 3.2.1.1. Sampling plan 71 3.2.1.2. Sample size determination 71 3.2.1.3. Distribution of the sample 72 3.2.2. Research instrument 72 3.3. Data collection 73 3.3.1. Training of data collectors 73 3.3.2. Pre-testing 74 3.3.3. Household survey 74 3.4. Assessing farmers' risk aversion and its detcrminants: The empirical methodology 76 3.4.1. Eliciting farmers' risk behaviour 77 3.4.1.1. Design of the experiment.. 81 3.4.1.2. Deriving risk aversion coefficient 83 3.4.2. Modelling risk aversion with persona! characteristics: The Ordered Logit mode!... 85 3.4.2.1. Explanatory variables and expected sign 86 3.4.2.2. Endogeneity of"farm income" 90 3.5. Risk perceptions and econometric analysis offarmers' risk management decision-making 91

vii 3.5.1. Behavioural Mode! (Discrete choice models) 93 3.5.2. Modelling determinants offanners' risk management decisions: The Multivariate Probit mode) 99 3.5.3. Explanatory variables and expected sign 103 3.6. Conclusion 106

CHAPTER FOUR: DETERMINANTS OF FARMERS' RISK ATTITUDES AND RISK MANAGEMENT STRATEGIES 107 4.1. Introduction 107 4.2. Household socio-demographic and economic characteristics 107 4.2.1. Socio-demographic attributes 107 4.2.2. Wealth composition and distribution 109 4.3. Farmers' risk preferences 112 4.3.1. Distribution ofpreferences to risk over games 112 4.3.2. Homogeneity in risk distribution 115 4.3.3. Nature of partial risk aversion 118 4.3.4. Risk aversion and socio demographic characteristics 120 4.4. Risk sources and farmers' risk management strategies 123 4.4.1. Farmers' perceptionofrisksources 123 4.4.2. Risk management strategies used by farmers 125 4.5. Econometric results and discussion 126 4.5.1. Determinants ofrisk aversion 127 4.5.1.1. Ordered Logit estimation results 127 4.5.1.2. Marginal effects on risk aversion categories 131 4.5.2. Determinants ofrisk management strategies adoption 134 4.5.2.1. Multivariate Probit estimation results: Correlation coefficients among risk management adoption decisions 134 4.5.2.2. Multivariate Probit estimation results: Parameters estimates 135 4.6. Conclusion 139

GENERAL CONCLUSION AND POLICY IMPLICATIONS 141 Summary of the major findings 141 Contribution of the thesis 143 Policy recommendations 143 Limitations of the study 145 Future Research Outlook 146

REFERENCES 147

viii APPENDICES 155 Appendix 1: The prospect theory (Kahneman and Tversky) 155 Appendix 2: The study area 156 Appendix 3: Cocoa production in Côte d'Ivoire 157 Appendix 4: Measuring farmers' level ofrisk aversion 160 Appendix 5: Testing for endogeneity of "fann income" 167 Appendix 6: Spearman's ranJc correlation coefficients 174 Appendix 7: Predicted probabilities 175 Appendix 8: The survey questionnaire (in French) 176

ix LIST OF TABLES

Table 1.1: The Allais paradox - the common consequence effect... 17 Table l.2: Kahneman and Tversky's choice - the common ratio effect... 18 Table 1.3: Preferences reversai bets 20 Table 1.4: Framing effect 21 Table 1.5: Relationships between the Shape of the Utility Function, Risk Attitude, Risk Premium and Marginal Utility 26

Table 2.1: Cocoa marketing periods in producing countries 50 Table 2.2: Cocoa exports from the ports of San Pedro and Abidjan, Côte d'Ivoire 59

Table 3.1: Sample distribution 72 Table 3.2: The basic structure of the experiment 81 Table 3.3: Regressors used in Ordered Logit mode} and expected sign (0=362) 89 Table 3.4: Definition of Variables in the MVP model and expected sign (0=362) 104

Table 4.1: Household socio-demographic characteristics (n=362) 108 Table 4.2: Distribution and mean offarm income per zone 109 Table 4.3: Distribution and mean of livestock value per zone 110 Table 4.4: Percentage share of cocoa revenue in total income (per zone) 111 Table 4.5: Percentage of household with other sources of income (per zone) 111 Table 4.6: Percentage distribution of risk aversion in different types of games and different game levels* 112 Table 4.7: Percentage distribution ofrevealed risk preferences in five experimental studies 114 Table 4.8: Tests of the risk distributions in the three game levels, the two types of game and the three zones 117 Table 4.9: Effect of payoff Scale on Partial Risk Aversion- 119 Table 4.10: Risk Aversion and Gender 120 Table 4.11: Risk Aversion and education 121 Table 4.12: Risk Aversion and Age 122 Table 4.13: Risk A version and Income 123 Table 4.14: Identification of risk sources and rank (n = 362) 124 Table 4.15: Mean scores and rank of sources ofrisks (n = 362) 125 Table 4.16: Proportion of Producers Adopting Different Combinations ofRisk Management Tools 126 Table 4.17: Ordered Logit models of risk aversion per game level 128 Table 4.18: Changes in Predicted Probabilities (marginal effects) by risk categorics 132 Table 4.19: Multivariate Probit Model Results: Correlation Coefficients of Risk Management Adoption Occisions 135 Table 4.20: Parameter estimates from the Multivariate Probit and Individual Probit Approach for estimating the factors affecting adoption of agricultural risk management strategies 136

X LIST OF FIGURES

Figure 1.1: Definition of risk 9 Figure 1.2: Risk aversion and sharp of the VNM utility function 25 Figure 1.3: CE and risk premium for a risk averse decision maker 26 Figure 1.4: Risks idiosyncratic and systemic 32 Figure 1.5: Supply in presence ofrisk 33

Figure 2.1: Cocoa production in Côte d'Ivoire and Ghana 48 Figure 2.2: World-wide distribution of cocoa production (in Tonnes) 50 Figure 2.3: Cocoa value chain for Côte d'Ivoire 60 Figure 2.4: Yield of cocoa farm in Côte d'Ivoire 62 Figure 2.5: Cocoa taxation in three major producing countries 63 Figure 2.6: World cocoa prices evolution 65 Figure 2.7: World cocoa price volatility 66 Figure 2.8: Nominal cocoa producer prices per kg, Côte d'Ivoire, 1966-2007 69

Figure 4.1: Comparison of risk distribution between the 100 FCF A gains-only game and the l 00 FCFA gains and tosses game in the three zones 118

xi ABSTRACT

In an attempt to investigate Ivorian cocoa farmers' behavioural responses to risks, this study first measures farmers' risk aversion by means of an experimental gambling approach and analyses the link between risk aversion and the household persona! characteristics using an Ordered Logit mode!. Second, the thesis identifies the main sources of risk faced by cocoa farmers and their risk perceptions, and then examines the determinants of producers' risk management adoption decisions in the absence of formai insurance, while taking into account the possibility of simultaneous utilization of multiple risk reducing strategies. ln particular, based on a Multivariate Probit specification, we use data frorn a survey on 362 households in Soubré to analyse factors influencing farmers' use of three major traditional risk management strategies: crop diversification, precautionary savings and social network.

We find more than 45 percent of the farm households to exhibit severe to extreme risk aversion, with a constant partial risk aversion coefficient of more than 2. With careful construction of the experiment, the natures of absolute and partial tisk aversion are examined and the data support the existence of Decreasing Absolute Risk Aversion (DARA) and lncreasing Partial Risk Aversion (IPRA) behaviour. Variables like wealth, education, age, gender and otigin of the farmer are found to affect his/her level ofrisk aversion. The results also show that the decision to use one tisk management strategy positively influences the decision to use the other strategies. Moreover, the Multivariate Probit approach points out the importance of risk aversion, farm size, household's size, household's head literacy and the engagement in off-farm activities as factors that increase the likelihood of using some tisk management strategies.

Key Words: Experimental approach, Risk Aversion, Ordered Logit, Risk Management, Agriculture, Multivariate Probit.

xii RESUME

L'existence de risque est un problème capital qui affecte différents aspects de la vie des populations dans le monde rural. Cela affecte la manière dont les individus maintiennent leurs ressources, et comment ces ressources sont transformées en revenu pour atteindre les objectifs de développement. En milieu rural en Côte d'Ivoire, le risque est présent dans toutes les décisions de gestion dans la production du cacao, comme le résultat de l'incertitude sur les prix des produits, sur les rendements et sur les ressources. Dans un environnement marqué par l'absence de marchés de crédit et d'assurance, le challenge principal auquel font face les paysans est comment gérer cette incertitude dont la conséquence directe est la baisse et la fluctuation de leur revenu. La survie du ménage agricole dépend alors de sa capacité à anticiper et à se couvrir du risque de revenu. A travers le temps, les ménages ont développé des stratégies traditionnelles, qui, quoi que sous optimales, les aident à réduire les effets des différents risques inhérents à la production du cacao sur leur bien-être.

Dans une tentative d'analyser le comportement des producteurs face au risque, cette étude se propose de mesurer dans un premier temps le degré d'aversion au risque des paysans à l'aide d'une méthode expérimentale et d'analyser la relation entre aversion au risque et caractéristiques personnelles du ménage par le biais d'un modèle Logit Ordonné. Ensuite, la thèse évalue la perception qu'ont les producteurs des risques auxquels ils font face et examine les facteurs qui influencent les décisions d'utilisation de stratégies de gestion du risque, tout en tenant compte de la possibilité d'une utilisation simultanée de plusieurs stratégies. En particulier, à partir de données primaires portant sur un échantillon de 362 ménages recueillies dans le Département de Soubré, un modèle Probit Multivarié est estimé afin d'analyser les facteurs qui influencent l'utilisation de trois stratégies traditionnelles majeures de gestion du risque: diversification des cultures, épargne de précaution, réseau social.

Les résultats révèlent que plus de 45% des chefs de ménage sont dans les classes d'aversion au risque sévère et extrême avec un coefficient d'aversion partielle au risque de plus de 2. Avec une construction minutieuse de l'expérience, les natures de l'aversion absolue et l'aversion partielle au risque sont examinées et les résultats supportent l'existence d'aversion partielle au risque croissante (ARPC) et d'aversion absolue au risque décroissante (ARAD). Les variables telles que la richesse, l'âge, l'éducation, le sexe et l'origine du producteur affectent son degré d'aversion au risque. Les résultats aussi montrent que la décision d'utilisation d'une des stratégies de gestion du risque accroît la probabilité que les autres stratégies soient utilisées. Par ailleurs, les résultats du modèle Probit Multivarié mettent en évidence l'importance de variables telles que l'aversion au risque, la superficie du champ, la taille du ménage, l'éducation et l'engagement dans des activités non agricoles qui affectent positivement la probabilité d'utilisation de stratégies de gestion du risque par le paysan.

Mots Clés: Approche Expérimentale, Aversion au Risque, Logit Ordonné, Gestion de Risque, Agriculture, Probit Multivarié

xiii LIST OF ACRONYMS AND ABBREVIATIONS

ADM Archer Daniels Midland ANADER Agence National d' Appui au Développement ARCC Autorité de Régulation du Café et du Cacao BCC Bourse du Café et du Cacao CAF Coût Assurance Fret CAISTAB Caisse de Stabilisation CARA Constant Absolute Risk Aversion CE Certainty Equivalent CIF Cost, Insurance, Freight CPRA Constant Partial Risk A version CRRA Constant Relative Risk Aversion CSSPPA Caisse de Stabilisation et de Soutien des Prix des Productions Agricoles DARA Decreasing Absolute Risk Aversion OPRA Decreasing Partial Risk Aversion DRRA Decreasing Relative Risk Aversion DUS Droit Unique de Sortie EU Expected Utility FAO Food and Agricultural Organization FCFA Franc de la Communauté Financière Africaine FDPCC Fonds de Développement et de la Promotion des Activités des Producteurs de Café et de Cacao FGCCC Fonds de Garantie des Coopératives de Café et de Cacao FOB Free On Board FOSD First-Order Stochastic Dominance FRC Fonds de Régulation et de Contrôle du Café et du Cacao GDP Gross Domestic Product GHI( Geweke-Hajivassiliou-Keane GVC Groupements à Vocation Coopérative IARA Increasing Absolute Risk A version ICCO International Cocoa Organization ICRISAT International Crops Research Institute for the Semi-Arid-Tropics IIA lndependence of Irrelevant Alternatives IID Independent and Identically Distributed IIN Independence from Irrelevant Nests IMF International Monetary Fund IPRA Increasing Partial Risk Aversion IRRA Increasing Relative Risk A version ITF International Task Force ML Maximum Likelihood MNL Multinomial Logit MNP Multinomial Probit MSM Method of Simulated Moments MVL Multivariate Logit MVP Multivariate Probit NMNL Nested Multinomial Logit

xiv PT Prospect Theory RUMs Random Utility Models SML Simulated Maximum Likelihood SOSD Second-Order Stochastic Dominance SSA Sub Saharan Africa SUR Seemingly Unrelated Regression UNCTAD United Nations Conference on Trade and Development USA United States of America VNM Von Neumann and Morgenstern

XV GENERAL INTRODUCTION

1. Background of the study

Risk is a central issue that affects many different aspects of people's livelihoods in the developing world. It affects whether people can maintain assets and endowments, how these assets are transformed into incomes via activities and how these incomes and earnings are translated into broader development outcomes. In rural area, risk is present in ail management decisions of agricultural systems, as a result of price, yield and resource 1 uncertainty • The existence of such risks has been found to alter household behaviour in ways that at first glance seem suboptimal. Indeed, farmers take their decisions in a risky environment so that the consequences of these decisions are often not known with certainty until long after those decisions occur. As a result, outcomes may be better or worse than expected. In the empirical literature, many researchers have found that risks cause farmers to be Jess willing to undertake activities and investments that have higher expected outcomes, but carry with them risks of failure (Adebusuyi, 2004; Alderman, 2008). For example, it bas been found that farm households use Jess fertilizer, improved seeds and other production inputs than they would have used if they simply maximized expected profits. It is also not uncommon to observe farm households in developing countries being reluctant to adopt new technologies even when those technologies provide higher returns to land and labour than traditional technologies (Yesuf et al, 2009). One aspect of this reluctance is reaction to risk. Hence, knowledge on how farmers make economic decisions under risk as well as their attitudes towards risks is important in determining strategies and formulating policies for agricultural development.

Agricultural risks are especially important if they result in income and consumption fluctuations. High income risk usually implies fluctuations in consumption and may be a cause of persistent poverty. This is likely when insurance and credit markets are incomplete or do not exist as it is the case for developing countries. The failure to cope with income risk is not only reflected in household consumption fluctuations but affects nutrition, health and education and contributes to inefficient and unequal intra-household

1 An extensive definition of the terms "risk" and ''uncenainty" is given at the beginning ofChapter One.

1 allocations (Dercon, 2002). Therefore, for agricultural economists, dealing with risk is an important research topic which has received particular attention throughout the past four decades.

There is strong evidence that poor farm households are risk-averse (Moscardi and de Janvry, 1977; Dillion and Scandizzo, 1978; Binswanger, 1980, 1981; Sillers, 1980; Antle, 1983, 1987; Wik and Holden, 1998; Yesuf, 2007). Risk aversion is a manifestation of people's general preference for certainty over uncertainty, and for minimizing the magnitude of the worst possible outcomes to which they are exposed", These general conclusions and observations have stimulated considerable research into the effects of risk on farmers' economic decisions. Sorne studies have focused on production decisions and choice of technology (Wolgin, 1975), other studies have analysed risk coping and risk management strategies (Udry, 1994; Townsend, 1995). In industrial countries formai risk• sharing institutions are widely available to help farmers overcome problems associated with uncertainty and risk. Farmers can usually borrow for production or consumption purposes to ease the transition from good years to bad. In many cases, they have access to a variety of privately provided insurance against specific types of risks, and they can trade in commodity futures and options markets. In developing countries, however, these kinds of institutions are usually much more rudimentary, and may not be available at ail for srnall-scale farmers or other impoverished households. Poor rural residents generally rely on a wide range ofinformal arrangements to reduce the multitude ofrisks they face. These traditional risk reducing strategies, however incomplete, are considered as alternative means to deal with risky incomes in rural area in developing countries. There is a vast literature which documents strategies used by rural households to offset the adverse effects of income shortfalls and entitlement failures. Alderman and Paxson (1994) presented a whole range of strategies and distinguished between risk management strategies and risk coping strategies.

Risk management strategies are decisions and actions taken ex-ante (before the realization of a risky event) to lower the probability of a risky event while risk coping strategies are decisions and actions taken ex-post after the risky event has occurred. Households' ex

2 The obvious definition of rrisk aversion" is "réluctance or antipathy to risk",

2 ante and ex post responses to risk have been extensively studied '. Ex-ante strategies aim at smoothing the flow of income of the household. Ex-post, farm households smooth consumption given the variability of income and other shocks. Risk-coping strategies also allow households to smooth consumption across households, through risk-sharing mechanisms such as gifts, transfers, and remittances. In agricultural sectors where risks are common, like in the cocoa sector in Côte d'Ivoire, it will be interesting to investigate how farm households respond to risks.

2. Problem statement

In Western Côte d'Ivoire, rural households are exposed to a variety of income uncertainties affecting their well-being. For cocoa farmers, examples of risks they are most likely to face include weather events, variation in prices of inputs and outputs, sudden and unfavourable changes in govemment policies and regulations, plants diseases and pests, and illness of the farm operator or his/her family", The issue of cocoa price variation or price risk is a recent phenomenon in Côte d'Ivoire. Before the liberalisation of the cocoa sector, farmers were isolated from price fluctuations in the world markets. However, the liberalisation of the sector in 1999 under the influence of the World Bank has given rise to new problems, the most important being that price risk is thrown back onto the local producers and intermediaries, who are ill equipped to deal with it.

In fact, the problem induced by the existence of the risks above cited is further compounded when one takes into account the fact that households' access to credit, road infrastructure, information and training are limited and that modem or formai risk management instruments do not exist in rural areas in Côte d'Ivoire. Therefore, cocoa farmers may struggle with a series of risks leading in fluctuations and fall in their incarne 5 and increasing their vulnerability • In this respect, it is intuitive to expect that farmers would adopt various strategies to deal with the risks they are confronted with, even if these strategies may not be optimal compared to modem risk management instruments.

3 See section 1.5 for a review ofhouseholds' risk management/coping strategies. 4 Detailed descriptions of these risks will be provided in later Chapters in the context of risk literature in general and rural Côte d'Ivoire in particular. 5 Vulnerable households are those which have difficulties to resist against a shock in order to prevent a decline in their well-being, Vulnerability is therefore primarily a function of the households' assets and their capacity to deal with risk.

3 However, the behavioural response of the farmers might depend on their level of risk aversion as well as on the type of risk they face.

Dealing with risk in Côte d'Ivoire cocoa sector remains of crucial importance not only for farrners but also for the government given the importance of cocoa in the economy of the country. Indeed, Côte d'Ivoire is the leading cocoa producing country in the word, supplying over 40% of the global cocoa production. This commodity generates 35% of the total export revenues of the country, accounts for 15% of the GDP and 20% of the tax revenue. The cocoa sector employs about 700,000 farmers and constitutes the source of livelihood for more than one third of the Ivorian population. 95% of Côte d'Ivoire's cocoa is produced by small-scale farmers. Hence, volatility of cocoa revenue contributes to the fragility of the Ivoirian economy and poses a stringent impediment to farmers' welfare and the country's poverty reduction objectives. How to manage best the risks faced by farmers in the cocoa sector should therefore be a key issue for govemment and policy• makers ifthey want to improve farmers' well being and considerably reduce the incidence of poverty in rural area.

The present dissertation then focuses on understanding the issue of risk in agriculture by answering the following questions: what are the risk preference schemes among cocoa farmers in Côte d'Ivoire? How is the farrner's attitude toward risk affected by its persona! characteristics? What are the main sources of risk faced by farmers in the I vorian coco a sector? What are farmers' general perceptions about different sources of risks? What are the determinants offarrners' risk management decisions?

3. Objectives of the study

The overall objective of the study is to analyse Ivorian cocoa farmers' decision-making behaviour under risk and uncertainty. The information will be useful for assisting policy makers in Côte d'Ivoire to evaluate the role of risk and implement measures to mitigate the risks faced by farrners. Specifically, this research intends to:

1. Give an overview of the cocoa sector in Côte d'Ivoire; 2. Measure the attitudes toward risk among cocoa farmers; 3. Determine the effects of persona! characteristics on farmers' risk aversion; 4. Identify the main sources ofrisk faced by cocoa farmers and their risk perceptions;

4 5. Examine the risk management strategies used by producers and investigate the factors that drive their risk management decisions, while taking into account the possibility of simultaneous utilization of multiple risk reducing strategies.

4. Significance of the study

It is undeniable that there is a clear link between the existence of farm risks and the incidence on national and individual farm income. This study contributes toward the goal of establishing a more fundamental empirical basis for risk analysis in rural area. The results of this study may provide useful information to formulate and implement appropriate policies to improve the welfare of farm households on one band and to revamp the lvorian cocoa sector which plays a preponderant role in the economy of the country on the other hand. More importantly, the relevance of this study can be stated as follows:

1. First, knowledge of attitude toward risk for particular categories of peasants defined by socioeconomic variables makes it possible, in tum, to determine packages of technological and institutional practices optimally tailored to peasants' economic behaviour. Such packages should greatly enhance the chances of success of rural development programs. Furthermore, an empirical basis on the determinant of risk aversion is necessary to qualitatively understand (and predict) how fanners (will) react to a change in their socioeconomic conditions. For example, is it the simple possibility of loss that drives aversion to risk or do the levels of potential gains and losses affect behaviour in response to risk in low-income rural settings? Does the build-up ofwealth at very low income levels affect risk behaviour? Do past successes within risky environments have an impact on future choices? Ail these questions are still largely open, but critically important to policy formation.

2. Second, little is also known about the basic household factors affecting risk behaviour. For example, in an environment of missing markets and low incomes, there might be important linkages between attitudes toward risk and seemingly disparate elements such as household fertility behaviour, educational levels and even gender. Working on these elements could improve outcomes for technology adoption. lt is therefore important to

5 better understand farm households' risk attitude through the analysis of the determinants of risk aversion.

3. Third, knowledge of perceived risks and risk management strategies across the smallholder cocoa fanners helps to adjust risk management policy intervention in the cocoa sector in particular and in the agricultural sector in general. It also enhances the satisfaction that fanners may get from externat support in the sphere of risk management in cocoa fanning in particular and in rural development in general. Furthermore, information of this sort provides enhanced risk communication among ail concerned parties ( e.g. producers, policy makers, donors, governmental and non-govemmental organizations, international agencies, researchers and local extension workers). Although, structural transformations are important in the long term, more immediate gains in poor households' welfare can be achieved through proper and better understanding of the farm household econornic decisions and knowing their risk sources and risk management strategies. Knowing the risk behaviour of farmers would enable policy makers to devise policies that can overcome some of the critical constraints they now face in meeting their basic needs.

4. Fourth, despite the contribution of cocoa to the livelihood of millions of households in Côte d'Ivoire and the inherent risks associated with agricultural activities, specific micro• economic empirical studies on the importance of risk to cocoa farming have not been found. As a consequence, knowledge about major sources of risk faced by Ivorian cocoa producers is largely missing. More importantly, little is known about farmers' risk attitude and risk perception as well as their strategies to deal with risks. This study is an attempt to fill that gap using rnicroeconornic data collected at the fann level alongside with an experimental gambling approach.

5. Outline of the dissertation

The objectives of this research stated in section 3, are elaborated and analysed in more details throughout this dissertation, which is organized into four chapters.

Chapter I reviews the basic literature on risk and uncertainty. It starts by defining some basic concepts used in this thesis. Theories which are commonly used to explain decision• making under risk and uncertainty are reviewed and evaluated. Specifically, a review of

6 the literature on expected utility and non-expected utility theories is provided. Sorne criticism of the descriptive inaccuracy of the expected utility model is presented and alternative descriptive models like prospect theory are discussed. Much attention will be devoted to the expected utility model and the measurement of attitudes toward risk. Then, the concepts of risk aversion and stochastic dominance are exposed. The typology of risks in agriculture is presented as well as the impact of risk in farming and finally the issue of risk management in farming is exposed.

Chapter 2 gives an overview of the cocoa sector in Côte d'Ivoire with an emphasis to the production pattern and the marketing system. The chapter presents the descriptive insights into the study site and also highlights the risks that may affect the livelihood of rural households producing cocoa in Côte d'Ivoire.

In chapter 3 the data set is introduced. The sampling and the survey procedures are exposed. The experimental gambling procedure to elicit farmers' risk aversion is exposed and the specific econometric methods used to fulfil the objectives of the study are developed.

Based on the theoretical and empirical analysis, chapter 4 finally presents the main results of the study. Sorne descriptive statistics of socio-economic and demographic characteristics of the sample are first presented followed by the results of the experimental games. Next, the econometric results on the determinants of risk aversion and the adoption of risk management strategies are exposed.

Eventually, the study ends with a general conclusion which gives a summary of major findings as well as the contribution of the thesis. Sorne policy recommendations, limitations of the study and suggestions for future research are also brought to the fore.

7 CHAPTER ONE: RISK, RISK AVERSION AND RISK MANAGEMENT IN AGRICULTURE

1.1. Introduction

Agriculture is often noted as an economic activity fraught with risk. Risk and uncertainty are universal characteristics of life in rural areas of Sub-Saharan Africa. On a daily basis, farmers are confronted with an ever-changing landscape of possible price, yield, and other outcomes that affect their financial returns and overall welfare. The consequences of decisions or events are often not known with certainty until long after those decisions or events occur, so outcomes may be better or worse than expected. When aggregate crop output or export demand changes sharply, for example, farm prices can fluctuate substantially and farmers may realize returns that differ greatly from their expectations. Agricultural producers regularly demonstrate concern for the economic uncertainty of the industry and major risk management tools such as futures markets have their origins in the agricultural sector. Similarly many farm support prograrns are justified primarily as risk safety net for agricultural producers.

Understanding risk in farming is important for two reasons. First, most producers are averse to risk when faced with risky outcomes. Someone who is risk averse is willing to accept a lower average retum for lower uncertainty, with the trade-off depending on the person 's level of risk aversion. Second, identifying sources of uncertainty helps farmers and others address the most important strategies for mitigating risk, and aids in circumventing extreme outcomes, such as bankruptcy. In spite of its manifest importance, risk management in agriculture is an under-researched topic relative to traditional concerns such as land tenure, technology adoption and food policy.

The purpose of this chapter is to introduce the concept of risk and its management in agriculture. The chapter begins with definitions of key concepts related to risk. Following this, the two cornrnon theories of decisions making under uncertainty are exposed. Then, the typology of risk in agriculture is discussed in section 4. The focus in section 5 is on risk management/coping strategies used by rural households. The paper concludes in section 6 with a surnrnary of major points discussed.

8 1.2. Some basic concepts and deffnltlons

1.2.1. Defining risk and uncertainty

Risk and Uncertainty are two basic terms to any decision making framework. Then, a considerable literature on defining risk and uncertainty spans the past several decades. Knight ( 1921) was the first to divide decision-making situations into risk and uncertainty. He defined the risk situation as one in which the decision maker knows both the alternative outcomes and the probability associated with each outcome. Under uncertainty, the decision maker does not know the probability of alternative outcomes. Furthermore, be may or may not know the different outcomes that can occur. A more common usage of these terms would state uncertain consequences. Risk is then viewed as the combination of the probability and the consequences of occurrence of an uncertain (bad) event (see Figure 1.1).

Figure 1.1: Definition of risk

High

Low L:erRisk

Low Magnitude of consequences High Source: Author

In this thesis, we adopt the definition given by Harwood et al. ( 1999) who described agricultural risk as uncertainty that affects an individual's welfare, and is often associated with adversity and loss. It may involve the probability of losing money, possible harm to human health, repercussions that affect resources, and other types of events that affect a person's welfare. Uncertainty (a situation in which a person does not know for sure what

9 will happen) is necessary for risk to occur, but uncertainty need not necessary lead to a risky situation. However, on the literature, Jess emphasis is usually placed on the differences between uncertainty and risk because both have similar effect on business enterprises. Throughout this dissertation therefore, we have tended to use both terms interchangeably, and indeed this is common practice by most practitioners (Varian, 1993;

Dixit and Pindyck, 1994). Nevertheless, it is important to understand the difference, as in some situations it may be necessary to apply the absolutely correct term to avoid ambiguity.

1.2.2. Measuring risk

Measuring risk has always been a complex exercise for researchers and experts working on the field of risks analysis. The risk measure adopted is always in line with the definition of risk considered. Hence, if risk can be defined in several ways, it implies that several interpretations can result and therefore, various approaches of measuring risk. Through the literature, three interpretations of the term risk are commonly used when it cornes of defining a measure of risk. These are risk as: the chance of a bad outcome, the variability or instability of the outcomes and the uncertainty about the future outcomes. Although seemingly similar, these three definitions imply quite different ways of measuring risk. Moreover, when formally defined they can be seen to be mutually inconsistent.

1.2.2.1. Risk as chance of a bad outcome

The chance of a bad outcome implies the probability of some defined unsatisfactory outcome happening. Assume for simplicity that there is a single measure of outcome, X, more ofwhich is always preferred to less. This definition ofrisk might be represented by the probability P* = P(X < X*), where P is probability, X is the uncertain outcome, and X* is some eut-off or minimally acceptable outcome level below which outcomes are regarded as 'bad'. In some cases, the value of X* might reflect some disaster level such as 'starvation' or 'bankruptcy', but in most cases it may be a Jess clear-cut notion, so that application ofthis measure ofrisk requires specification of the two parameters P* and X*.

10 1.2.2.2. Risk as variability of the outcomes

Risk as variability may be measured by some statistics of dispersion of the distribution of outcomes X, such as the variance or standard deviation of X, V= V[X] or SD, equal to the positive square root of V. Obviously, neither statistic alone tells anything about the location of the distribution of outcomes on the X axis. So it is common for those who think of risk as dispersion of outcomes to link V or SD with the mean or expected value E = E[X].

1.2.2.3. Risk as uncertainty of the outcomes

Finally, adopting the definition of risk as uncertainty of outcomes requires the whole distribution of X to be specified. Complete specification requires the probability density function, f(X), or equivalently and often more conveniently, the cumulative distribution function F(X). However, summary statistics including moments are commonly used to describe the probability distribution, implying some similarity with the measurements based on the definition of risk as dispersion. For a few special cases, such as the normal, the distribution of outcomes is fully defined by only the mean and variance. Other distributions might be approximated in terms of these first two moments, though higher order moments may be needed to tell more about the shape of the distribution. For some arbitrary distribution, however, description by moments will be an approximation; the adequacy of which is not easily judged.

1.2.2.4. Limitations of the risk measures

The weakness of the first two definitions of risk with their associated measures is that they neither 'tells the whole story' when a choice bas to be faced among risky alternatives. In regard to the first definition, it is clear from observing behaviour that not all risks with bad outcomes are rejected. For example, most people will travel by car for tourism - an activity that certainly increases the probability of death or serious injury in a road accident. Evidently, choices with chances of very bad outcomes (e.g. death or serious injury) are sometimes accepted, presumably because the benefits of the up-side consequences (e.g. seeing interesting sights) are sufficiently appealing to offset the relatively low chances of the bad outcome. It follows that to evaluate or assess a risky

11 situation we need to be able to consider the whole range of outcomes, good and bad, and their associated probabilities. Descriptions of risk expressed in terms of only the probability in the lower tail of the distribution of outcomes do not provide full information for proper risk assessment, and so may be seriously misleading.

A similar argument shows the limitation of variance alone as a measure for risk evaluation. Consider two normal distributions of outcomes of, say, net income, with identical variances but different means. On one hand, everyone will prefer the one with the higher mean. Many would describe the distribution located further to the right as the Jess risky of the two since the chance of getting Jess than any specified level of X is lower for this distribution than for the one with the lower mean. On the other hand, using variance as the measure of risk suggests that the two distributions are equally risky. Clearly, we could avoid such confusion by interpreting measures of dispersion simply as what they are, and not regarding them as 'stand alone' measures ofrisk.

Adopting the third option of defining risk as the full distribution of outcomes means that there is no one statistic that can be used to measure risk, so that it becomes impossible to compare distributions in terms oftheir 'riskiness'. What this third view ofrisk implies is that notions of 'more' or 'Jess' risk ('more risky' versus 'Jess risky' prospects) are unsatisfactory, and careful analysts will confine themselves to describing risky prospects as 'preferred' versus 'not preferred', or as 'risk efficient' versus 'not risk efficient'.

To end with this section, one should note that despite these limitations, measures of risk commonly used in theory of uncertainty have always been the variance and the standard deviation of outcomes, in liaison with the mean or expected value. Variance may then be described as the risk around the specified mean; the higher the variance, the higher the level of risk.

1.3. Theories of decision making un der risk and uncertainty

The neoclassical theory of the firm starts by a situation where ail the markets are perfectly competitive and information is perfect for all entities. This means that, when taking his production decision, the firm knows exactly the market price of inputs and outputs, as well as the quantity of output he will produce. In the short run, the firm maximises his

12 profit at the point where the marginal cost of a unit of output equal to the market price of that output. However, in practice, finns and especially agricultural enterprises rarely know with certainty the level of production they will have and at what price they will sell this production in the market. This can result in severe income losses and to fluctuations in consumption. Given their limited ability to offset these shocks, many rural households suffer from extreme farm income fluctuations.

Modelling decision making under uncertainty centres on two main theories: the classical expected utility model of individual choice under uncertainty and the non-expected utility model, which assumes that individuals do not maximize expected utility. This section gives an overview of the expected utility model - and its caveats - and the prospect theory as one of the possible alternatives to the expected utility mode! of individual decision making under uncertainty.

1.3.1. The expected utility theory

In general the expected utility (EU) mode! has been the dominant model for the last decades in modelling behaviour under risk and uncertainty. Von Neumann and Morgenstern - VNM (1947) are the major contributors to a large body of work that provides the justification for the use of the EU mode) by a rational decision maker. This model views decision making under risk as a choice between alternatives. Decision makers are assumed to have a preference ordering defined over the probability distributions for which the axioms of the EU mode! hold (Mas-Colell et al., 1995). Risky alternatives can be evaluated under these assumptions using the expected utility function U(.).

In maximizing the decision maker's utility, consider a risky prospect in which the decision maker does not know ex-ante which state of the world will occur. However he/she can list the various alternatives and can attach probabilities to them. For simplicity, assume two possible states of the world, state 1 and state 2, with respective probabilities p I and p2 and denote x, the individual's monetary gain if state 1 occurs and x, if state 2 occurs. The individual must choose ex-ante between the risky bundles (x,,x,). Ex-post, the individual gets x, or x, depending upon which state of the world has occurred. If the decision

13 maker's preference ordering over risky alternatives satisfies al! the axioms of EU, including the independence and continuity axioms (see next section), then there exists a VNM expected utility function. This VNM expected utility function reflects the decision maker's choice as if he/she maximizes utility of the different states weighted by the probabilities for each state to occur.

VNM began by stating that utility maximization is a rational goal when a decision maker is faced with risky choices. In this framework, an individual will evaluate the expected value and objectively, given probability of occurrence of each alternative. This evaluation is carried out by first entering the probabilities and expected outcomes into an individual' s utility function. lt is then a matter of selecting the combination of available alternatives that maximizes the function. The manner in which individuals choose among available alternatives is then dependent upon their utility function. For this setting the VNM expected utility function can be specified as:

U(Pi,···,Pw··,P.) = Î,p;u(x;) (1.1) i=I

where U is the VNM expected utility function, u(x;) is the utility of the ith element of a

vector of possible outcomes, and p, is the probability of outcome X;, LP; = 1. The

VNM expected utility function U(pp···,A,···,PJ, defined up to a positive linear transformation, characterizes both the utility of the outcome and the individual 's attitude toward risk. The curvature of this utility function contains information about the degree of individual's risk aversion (Mas-Colell et al., 1995: 173). Further discussion on this point will be presented in section 1.6.

1.3.1.1. Axioms of the EU theory

There are three main axioms in the EU framework. They are defined over a binary relation where: i'.: denotes weak preference, >- denotes strict preference, and - denotes indifference.

14 The preferences over probability distributions p, q, r E P are defined over a common

( discrete or continuous) outcome vector X . The three axioms that are necessary and sufficient for the EU representation U(.) over preferences are:

• Axiom O (Order)

The binary relation >- on P is asymmetric and transitive. The asymmetric part of axiom 0 says that the decision maker will not both prefer p to q and prefer q to p . According to the EU theory, it is irrational to hold a defmite preference for p over q and a definite preference for q over p at a time. However, there is a possibility that neither p nor q is preferred (i.e.; p - q . the decision maker is indifferent between p and q ). The transitivity part of axiom O holds if and only if both >- and - are transitive, i.e., for ail p.q.r e P:

p (' :, q, and q (' :, r ==> p (' :, r p - q and q - r ==> p - r

Transitivity implies that it is impossible to face the decision maker with a sequence of pair wise choices in which preferences appear to cycle.

• Axiom C (Continuity)

For all p.q.re P, with p (' :, q, and q (' :, r, there exists there a,/Je[0,1] such that a p + (1- a )r (' :, q and q (' :, /J p + (1- /J)r . This axiom gives continuity to the preferences. Continuity means that small changes in probabilities do not change the nature of the ordering between two lotteries (see Mas-Colell et al., 1995: 171 ). Continuity rules out lexicographie preferences.

• Axiom I (lndependence)

For p.q.r e P, and for a E [0,1], if p (' :, q, then ap + (1-a)r (' :, aq + (I- a)r This axiom states that preferences over probability distributions should only depend on the

15 portions of the distributions that differ ( p and q ), not on their common elements ( r)

and of the level of a that defines the linear combination. In other words, ifwe mix each of two lotteries with a third one, then the preference ordering of the two resulting mixtures does not depend on the particular third lottery used.

Axioms 0, C, and I can be shown to be necessary and sufficient for the existence of a function U(.) on the outcomes x e X that represents preferences through ;:::, . The role

of the order, completeness and continuity axioms are essential to establish the existence of a continuous preference function over probability distributions. It is the independence axiom which gives the theory its empirical content and power in determining rational behaviour. That is, the preference function is constrained to be a linear function over the set of probability distribution functions, i.e. linear in probabilities (Machina, 1982: 278).

To sum up this point, one should note that the EU mode! is based on axioms explaining individual behaviour. These axioms are assumptions in choosing risk alternatives and describe how a rational individual should behave. Hence, since VNM (1947), the EU model has been the dominant mode! in predicting choice behaviour under risk and uncertainty. However, starting with the well-known paradox of Allais (I 953), a large body of experimental evidence has been documented which indicates that individuals, in some cases, tend to violate the axioms underlying the EU mode!.

1.3.1.2. Violation of the EU theory

There have been many advances in the economic analysis of decisions under uncertainty using the EU mode! (e.g., Sandmo, 1971; Antle, 1987). These contributions take the validity of the expected utility mode! as given. However, a serious challenge to the use of EU was made as early as 1950s to show that expected utility lacked complete predictive and hence descriptive validity. Experimental investigations revealed a variety of inconsistencies between observed choice behaviour and EU. The following subsections report some violations of the EU theory.

16 • Violation of the independence axiom

The independence axiom bas been the most extensively investigated axiom from an empirical perspective. Allais (1953) opened the way and reported experimental evidence showing systematic violations of independence. These violations are the so-called common consequence effect and common ratio effect. They say that outcomes are not independent and agents show a higher degree of risk aversion for losses than for gains. Most examples of the common consequence effect and common ratio effect have involved choices between pairs of prospects.

The common consequence effect: the well-known risky choice provided by Allais is given in a paper by Kahneman and Tversky (1979). They synthesized the work by Allais and by others who have shown experimental violations of expected utility. The Allais paradox depicted in Table 1.1 is the leading example of this class of anomalies. There are two different choice sets, for each choice set there are two lotteries from which you can choose. For example, in lottery AI there is a guaranteed payoff of $1 M and there is zero probability of winning nothing. In lottery A2 there is a 0.10 probability of winning $SM, a 0.89 probability ofwinning $1M, and a 0.01 probability ofwinning nothing. Then one bas to choose between Al and A2, and between A3 and A4. Where Al, A2, A3, A4 are lotteries.

Table 1.1: The Allais paradox - the common consequence effect Lotteries OM lM 5M Al 0 1 Choice set 1 0 A2 0.01 0.89 0.1 A3 0.9 0 - Choice set 2 0.1 A4 0.89 0.11 0 Source: Kahneman and Tversky ( I 979) Note: outcomes are in Dollars and IM = $1,000,000.

Many agents prefer lottery Al to A2 and prefer lottery A3 to A4. This empirical tendency directly contradicts EU theory. According to EU theory Al >- A2 if and only if lu($1M) > 0.10u($5M) +0.89u($1M) + O.Olu($0). Subtracting 0.89u($1M) from each side, it follows that O. llu($1M) > 0.1 Ou($5M) + 0.0 lu($0) . Adding 0.89u($0) to both si des, we have 0.1 lu($1M) +0.89u($0) > 0.10u($5M)+ 0.90u($0) which holds if and only if

A4 >- A3. Thus, from EU theory, one can conclude that Al>- A2 <=> A4 >- A3 . However,

17 many people choose Al over A2 and prefer A3 over A4. This pattern of choice violates the independence axiom and hence the EU theory. The Allais Paradox is then comrnonly known as a special case of a general empirical pattern called the comrnon consequence effect. The name cornes from the "common consequence" 1 M in gamble 1 and O in gamble 2. The independence axiom requires that preferences be unaffected by changes in a common consequence. According to this axiom, an individual's preferences in one event should not depend on the outcome in another event. lt can be shown that violation of the independence axiom explains the observed inconsistencies in the measurement of the VNM utility mode!. If an agent is an expected utility maximizer then he must prefer AI to A2 and A4 to A3. Agents may prefer Al to A2 because they like to be a millionaire with certainty, implying risk aversion. But in choice set 2 the gambles are quite different with a high probability in each lottery of not winning any money. So, the agent may simply choose A3 because the chance of winning $SM is very similar to the chance of winning $1M and $SM is much more. The typical agent responds in a more risk-averse manner in choice set 1 and more risk neutral in choice set 2.

The common ratio ejfect: another closely related violation of the independence axiom is the common ratio effect or certainty effect (Kahneman and Tversky, 1979). Individuals are asked to make a choice between the two gambles described in Table 1.2. In the EU model it requires that the choice between Al and A2 bas to be compatible with the choice between A3 and A4, i.e. if the more risky alternative A2 is selected in the first choice, the more risky alternative A4 must be selected in the second choice and vice versa.

Table 1.2: Kahneman and Tverskû cboice - the common ratio effect Lotteries $0 $3000 $4000 Al 0 Cholce set 1 0 A2 0.2 0 0.8 A3 0.75 0.25 Cholce set 2 0 A4 0.8 0 0.2 Source: Kahneman and Tversky ( 1979)

In their experiment using hypothetical payoff outcomes, 80% of the subjects choose Al over A2 in the first pair of choice and 65% choose A4 over A3. That is, in the comrnon• ratio effect, subjects choose $3000 for sure to a 0.80 chance at $4000 but also a 0.20 chance at $4000 to a 0.25 chance at $3000. This pattern also contradicts expected utility,

18 since the first pair of choice implies lu($3000) > 0.80u($4000), but the second pair implies

0.25u($3000) < 0.20u($4000). The independence axiom is violated in this example, since the second pair is constructed by taking 25% chance of the first pair and 75% chance of receiving $0. The effect gets its name because the ratio of the probability of winning $4000 to the probability ofwinning $3000 is the same for both choices (1/0.8 = 0.25/0.2).

• Violation of the order axiom

In addition to the violation of the independence axiom, there is experimental evidence suggesting that descriptive failures of expected utility may run deeper than violations of the independence axiom (Starmer 2000: 338). The two hidden assumptions in any conventional theory of choice are procedure invariance and descriptive invariance, which constitute another source of weak descriptive power for EU. Procedure invariance suggests that preferences over prospects and acts are independent of the method used to elicit them, whereas description invariance stipulates that preferences over prospects are purely a function of the probability distributions and do not depend on how these objects are described.

Preference reversai: the most serious blow for the procedure invariance assumption may have been the discovery of preference reversai. Preference reversai, first reported by Lichtenstein and Slovic (1971), describes experimental results that appear to indicate systematic violations of transitivity of preferences. In their experiment subjects were asked to choose between two bets and then to give their true certainty equivalents for the bets in the form of a selling and a buying price. In many cases the subjects set the lowest price for the preferred lottery. In other words, individuals were presented with two gambles, one featuring a high probability of winning a modest sum of money (the P bet), and the other featuring a low probability of winning a large amount of money (the $ bet). The typical finding is that people often choose the P-bet, but assign a larger monetary value to the $-bet. In their I 971 article Lichtenstein and Slovic presented the following pair of gambles (see Table 1.3).

19 Table 1.3: Preferences reversai bets P-bet {$4, 0.99; -$1, 0.01} Expected Outcome of the P-bet = $3.95 S-bet {$16, 0.33; -$2, 0.67) Expected Outcome of the $-bet = $3.94 Source: Lichtenstein and Slovic (1971)

The P-bet shows 99% chance of winning $4 and 1 % chance of losing $1 white the $-bet shows 33% chance of winning $16 and 67% chance of losing $2. Expected outcomes of the two lotteries are almost the same. The subjects were asked to choose which game they would like to play. Later they were told that they had the ticket to play the bet and were asked to name a minimum selling price for the ticket. Lichtenstein and Slovic found that 73% of the participants consistently have a higher price to the $-bet even though they had chosen the P-bet. The EU theory implies that the bet which is actually chosen also will be the one which will be assigned the largest selling or buying price. In an earlier study, Slovic and Lichtenstein (1968) had observed that choices among pairs of gambles appeared to be influenced primarily by probabilities of winning and losing, whereas buying and selling prices were more highly correlated with payoffs than with probability of winning. Following this observation they argue that, if the method used to elicit preferences affected the weighting of the gamble's components, it should be possible to construct pairs of gambles such that the same individual would choose one member of the pair but set a higher price for the other. This gamble when viewed from the standard theory perspective presents a puzzle. Both choices constitute ways of asking essentially the same question. In these experiments, however, the ordcring revealed appears to depend upon the elicitation procedures. Moreover, choice and valuation tasks may invoke a different mental process, which in tum generates different ordering of a given pair of prospects. Consequently, the ranking observcd in choice tasks cannot be explained with reference to a single preference ordering (Starmer, 2000).

Framing effect: there is also a long list of experimental observations showing that choice behaviour can be dramatically affected by the context in which it takes place. Such evidence contradicts the assumption of description invariance. One of the most persuasive examples reported is due to Schoemaker and Kunreuther (1979) in which the subject has to choose among the same pair of prospects successively framed as prospect and as an insurance decision problem (see Table 1.4).

20 Table 1.4: Framing_ effect Choice 1: A: a sure loss of$ IO Gambling_ B: 1% chance of$1,000 loss Choice2: lnsurance C: pay an insurance premiurn of $10 D: remain exposed to 1 % loss of$ 1,000 Source: Schoemaker and Kunreuther ( 1979)

The two pairs of options are stochastically equivalent. The only difference is that choice 1 description presents the information in terms of a sure loss while the information presented in choice 2 is in terms of an insurance payment. Shoemaker and Kunreuther found a very striking difference in response to these two presentations: 51 % of the subjects preferred A to B while 81 % preferred D to C. This is to say that minor changes in the presentation or framing of prospect can have dramatic impacts upon the choices of decision makers. Thus, it is obvious that expected utility cannot naturally account for such difference.

These empirical evidences have motivated researchers to develop alternative theories of choice under uncertainty able to accommodate the observed patterns of behaviour. A wave of theories designed to explain the violation of expected utility theory began to emerge at the end of the 1970. Examples are the prospect theory (Kahneman and Tversky, 1979), the dual theory (Yaari, 1987), the cumulative prospect theory (Tversky and Kahneman, 1992), and the rank-dependent utility (Quiggin, 1993). For a thorough review see Starmer (2000). In this dissertation, we will only discuss the dominant alternative theory of the empirical literature, which is the prospect theory.

1.3.2. The non-expected utility model: The Prospect Theory

The Prospect Theory (PT) was developed first by Kahneman and Tversky ( 1979). They develop their theory as an alternative to the expected utility theory for explaining the outcomes of individual decision making under uncertainty. They argued that choices that individuals make in risky situations exhibit several characteristics that are inconsistent with the basic axioms of EU theory. They argued that individuals underweight probable outcomes in comparison with outcomes that are certain. They called this phenomenon the certainty effect. They also pointed out that the certainty effect brings about risk aversion in choices involving certain gains and risk-seeking in choices involving certain losses (Kahneman and Tversky, 1979).

21 Kahneman and Tversky (1979) distinguished two sequential phases in a decision process: the editing phase and the evaluation phase. In the editing phase, decision makers contemplate the cboice situation and if possible simplify the problem. This includes the operation of coding by which outcomes are coded as gains or losses, prospects are simplified by combining probabilities associated with identical outcomes, risky components of a prospect are separated from the risk less component of the prospect, and finally components of choices that are common to all prospects are discarded. The edited prospects are then evaluated and the most highly valued risky outcome is chosen. PT employs two functions: a probability weighting function f((p) , and a value function v(x). These functions are combined to form the basic equation of the theory which determines the overall value of a prospect. The following is the equation used by Kahneman and Tversky (1979) for simple prospects with the form (x,p;y,q); a gamble between two outcomes (x,y) with associated probabilities (p,q) which bas at most two nonzero outcomes:

V(x,p ;y,q) = f((p)v(x) + f((q)v(y) (1.2)

When the prospects are strictly positive or negative, the evaluation follows a different rule. In the editing phase the prospects are separated into a risk Jess (the minimum gain or loss which is certain to be gained or paid) and a risky component (the additional gain or loss which is actually at stake ). Thus, if p + q = 1 and either x > y > 0 or x < y < 0, so

1(( q) = [ 1 - f((p)], then,

V(x,p;y,q) = v(y)+f((p)[v(x)-v(y)] (1.3)

One of the essential features of the PT is that the overall value of a prospect is based on changes in a decision maker's wealth reference point rather than on final wealth states, as in the case of the EU theory. Kahneman and Tversky propose the value function, one of the most widely used components of PT, a function that is commonly S-shaped. It is generally concave for gains (implying risk aversion) and commonly convex for losses (implying risk-seeking), and steeper for losses than for gains (see Figure A. l in Appendix 1 ). Another major departure of prospect theory from the EU theory is the treatment of the probabilities. In EU models the uncertain outcome is weighted by its probability whilc the

22 uncertain outcome in prospect theory is multiplied by the decision weight tr(p). The

weighting function, ;r , which relates decision weights to stated probabilities, is a monotonie function of p , with tr(O) = 0 and tr(l) = 1 but is not a probability and should not be interpreted as a measure of degree ofbelief.

According to the PT, very low probabilities are over-weighted, that is, the decision weight attached to the rare event is larger than the probability tr(p) > p . Furthermore, PT

suggests that for all O < p < 1, tr(p) + tr(l - p) < 1, this is sub-certainty. lt implies that as

low probabilities are over-weighted, moderate and high probabilities are underweighted, that is the decision weight is smaller than the probability tr(p)

(x,y) with associated probabilities (p,q): in EU maximization theory, the value of the utility function is U(X) = pu(x) + qu(y) and in PT the value function is V(X) = tr(p)v(x) + tr(q)v(y). In both cases the summed function is maximized and the highest value is chosen. PT and EU theory coincide when tr(p) = p for all p and u(x) = v(x). ln this case the EU of a lottery defined on U(X) equals the value V(X) of the gamble in PT.

In summary, the empirical findings that drive PT offer a profound challenge to EU models of decision-making. First, the finding that individuals demonstrate a non-linear response to probabilities clearly differs from EU theory expectations. lndividual's tendency to underweight moderate and high probabilities means that in these situations their utility calculations will grant more weight to the utility of a possible outcome than to its probability of occurring (the reverse happens when probabilities are overweighted). In contrast, in EU theory individuals possess linear probability functions; utility and probability estimates are given equal weight in the generation of individual's EU values. Second, individuals' attitudes toward risk, and thus their decisions, are likely to change depending on whether the same situation is seen as a gain or a Joss (a phenomenon known as preference reversai). This violates EU theory's assumption that people will possess

23 consistent and transitive preferences (Starrner, 2000). Finally, the core elements of PT's value calculations contradict key tenets of EU theory by showing that individuals tend to be driven by gains and losses relative to a reference point rather than by final wealth levels. Despite these limitations, however, the EU theory remains the dominant approach for modelling individual decision making under uncertainty (Schneider, 2004).

1.3.3. Risk aversion and stocbastic dominance

The EU mode! allows us to capture in a natural way the notion of risk aversion, which is a fundamental feature of the problem of choice under uncertainty. This section allows us to define and describe risk aversion, as well as the adjacent notions of certainty equivalent and risk premium. Sorne common measures of risk aversion are then presented. Furtherrnore, the concept of stochastic dominance is exposed.

1.3.3.1. Risk aversion and concavity of the utility function

The utility function developed by Von Neumann-Morgenstern assumes that under uncertainty, individuals maximise expected utility. Let u(.) be an individual's expected utility function for gambles over nonnegative levels ofwealth. A simple gamble takes the

forrn (P1W1,···,Pn wn), where n is some positive integer, the W; 's are nonnegative wealth

levels, and the nonnegative probabilities, Pi,···,Pn, sum to 1. We assume that the individual's VNM utility function, u(.), is differentiable with u'(w)>O for ail wealth

level w. The expected value of the simple gamble g offering W; with probability P; is given by E(g) = :r;=l P; W; . The agent's attitude toward risk is defined by the choice between accepting the gamble g on one band or receiving with certainty the expected value of g on the other. If u(.) is the agent's VNM utility function, these two alternatives are evaluated as follows:

u(g) = Ï, p,u( w;), (1. 4) l=I

(1. 5)

24 The first of these equations is the VNM utility of the gamble, and the second is the VNM utility of the gamble's expected value. If the preferences satisfy the axioms of the EU theory, the agent should prefer the alternative with the higher expected utility. Thus, the individual is said to be:

- Risk averse at g if u(E(g)) > u(g),

- Risk neutral at g if u(E(g)) = u(g), - Risk loving at g if u(E(g)) < u(g).

Each of these attitudes toward risk is equivalent to a particular property of the VNM utility function. Over the domain of wealth, the agent is risk averse, risk neutral, or risk loving if and only if his VNM utility function is strictly concave (V"(.)< 0), linear (V"(.)= 0), or strictly convex (V"(.)> 0) respectively (see Figure 1.2).

Figure 1.2: Risk aversion and sharp of the VNM utility fonction

Utility Utility Utility

U(w) U(w) U(w)

Wealth Wealth Risk averse Risk neutral Wealth Risk loving

Source: Mas-Colell et al., 1995

1.3.3.2. Certainty equivalent and risk premium

The certainty equivalent of any simple gamble g over wealth levels is an amount of

weaJth, CE, offered with certainty, such that u(g) = u(CE). It is the amount of wcaJth that makes the individuaJ indifferent between accepting that wealth with certainty and facing the gambJe g (JehJe and Reny, 2001). When a persan is risk averse and strictly prefers more money than Jess, the certainty equivaJent is Jess than the expected value of

25 the gamble. Indeed, a risk averse persan will "pay" some positive amount of wealth to avoid the gamble's inherent risk. This willingness to pay to avoid risk is measured by the risk premium. The risk premium is then an amount of wealth, P , such that u(g)=u(E(g)-P). Clearly, P=E(g)-CE. By definition, the preferences are risk averse if and only if P > 0 (see Figure 1.3).

Figure 1.3: CE and risk premium for a risk averse decision maker.

u(.)

u(E(g) u(g:

p ~ '

CE E(g) Source: Jehle and Reny, 2001

In this graph, the risk premium is the horizontal distance (bold line). Based on the EU theory, the relationships between the shape of a decision maker's utility function, risk attitude, risk premium and marginal utility can be summarized like in Table 1.5.

Table 1.5: Relatlonshlps between the Shape of the Utlllty Fonction, Rlsk Attitude, Risk Premlum and Marginal Utllity.

Shape of the utlllty Attributes fonction U(.) Rlsk attitude Rlsk premium Marginal utllity lE!;)-CE) !U'!·l>Ol Aversion Concave Positive Diminishing (E(g_)>CE) (U"(.)Oj Source: Author

26 1.3.3.3. Measures of risk aversion

Many times, researchers not only want to know whether someone is risk averse, but also how risk averse he/she is. Ideally, they would like a summary measure that allows them both to compare the degrec of risk aversion across individuals and to gaugc how the degree of risk aversion for a single individual might vary with the level of wealth. Because risk aversion and concavity of the VNM utility function in wealth are equivalent, the seemingly most natural candidate for such a measure is the second derivative, U"(w).

Pratt ( 1964) and Arrow (1965) proposed two indicators to compare differences in risk attitudes. The first is the absolute risk aversion coefficient A( w). Mathematically, this coefficient is calculated as follows:

A(w) U"(w) = - (1. 6) U'(w) where U'(w) is the first derivative of the VNM utility function.

This coefficient can be interpreted as the percentage change in marginal utility caused by each monetary unit of gain or loss (Raskin and Cochram, 1986). Thus, the coefficient A( w) takes either positive or negative values for risk-averse or risk-Ioving economic agents respectively. When the coefficient decreases as the monetary value increases we have decreasing absolute risk aversion (DARA). Altematively, if the coefficient increases under the same set of circumstances we have increasing absolute risk aversion (IARA). Finally, if the coefficient does not change across the monetary level, the decision maker exhibits constant absolute risk aversion (CARA), which implies that the level of the argument of the utility function does not affect bis or ber decisions under uncertainty.

Because A(w) is not a non-dimensional measure of risk aversion, its value is dependent on the currency in which the monetary units are expressed. To overcome the impossibility of comparing risk aversion among different economic agents, the relative risk aversion coefficient R(w) bas been developed as follows:

U"(w) R( w) = -w,_____:.---=- = wA( w) (1. 7) U'(w)

27 This second coefficient measures the percentage change of marginal utility in terms of the percentage change in the monetary variable. As with the absolute risk aversion coefficient, we can find decreasing, constant or increasing relative risk aversion behaviour (DRRA, CRRA and !RRA, respectively).

There is another measure of risk aversion introduced simultaneously by Menezes and Hanson (1970), and Zeckhauser and Keeler (1970): the partial risk aversion coefficient P(Wo,m) . These authors reported that individuals are generally confronted with multiplicative lotteries, which rarely affect their total wealth. Consequently, they suggested breaking up the total wealth of an individual in the following way: w = Wo + m where w0 represents the initial wealth, m is the certainty equivalent of a risky prospect, and w is the final wealth (Pratt, 1964; Arrow, 1965; Menezes et Hanson, 1970 et Zeckhauser et Keeler, 1970). The partial risk aversion coefficient is defined as follows:

U"(Wo + m) = mA(w + m) P(Wo,m)=-m U'(wo+m) 0 (1. 8)

(Note that this is equivalent to the relative risk aversion when initial wealth is zero)

The advantage of this risk aversion measure compared to the absolute and relative risk aversion coefficients is that for its measurement, it requires only that the risk associated with an activity varies, while the level ofrichness remains constant.

In sum, absolute risk aversion traces the behaviour of an individual toward risk when his/her wealth rises and the prospect remains the same. Partial risk aversion examines behaviour when the prospect changes, but wealth remains the same. Relative risk aversion looks at behaviour when both the initial wealth and the level of the prospect rise proportionally. A DARA implies that a person will be more willing to accept a risky prospect as wealth increases. An !RRA postulates that a person's willingness to accepta risky prospect declines when both the outcome and wealth increase proportionally. An Increasing Partial Risk Aversion (!PRA) indicates a decrease in the willingness to take a gamble as the scale of the prospect increases. Empirical studies have generally rejected CARA in favour of DARA, decreasing partial risk aversion (DPRA) in favour of IPRA, white there is no consensus on the nature of relative risk aversion.

28 1.3.3.4. Stochastic dominance

We have talked about ways to determine whether one person is more or less risk averse than another person. Now we shift our emphasis to asking whether a particular lottery is more or less risky than another one. Probably the most general way to compare the riskiness of lotteries is in terms of what is called stochastic dominance. First and second orders stochastic dominance are concepts used to describe the similarities in evaluation of risky prospects by a large class of agents.

First Order Stochastic Dominance: Let F.(.) and Fa(.), respectively, represent the cumulative distribution functions of two random variables (cash payoffs). We say that distribution F.(.) first-order stochastically dominates (FOSD) distribution Fa(.) if for every nondecreasing concave function U : R ~ R , we have:

Î u(x)dF,(x) ';?. Î u(x)dFa(x) 0 0

The above definition is equivalent to the following: distribution F,(.) FOSD distribution

Fa(.) if and only for every x :

First order stochastic dominance does not imply that every possible retum of the superior distribution is larger than every possible return of the inferior distribution. If F.,(.) first order stochastically dominates Fa(.), the expected value of F.(.) is higher than the expected value of Fa(.). The reverse is not necessarily true. ln this thesis, the first order stochastic dominance is considered to compare risky lotteries.

Second Order Stochastic Dominance: For any two distributions FA(.) and Fa(.) with the same mean, F.(.) second-order stochastically dominates (SOSD) Fa(.) if for every nondecreasing concave function U : R ~ R, it holds:

f u(x)dFA(x)';?. Î u(x)dF8(x) 0 0

29 Second order stochastic dominance is a weaker concept than FOSD in the sense that FOSD implies SOSD but SOSD does not imply FOSD. Second order stochastic dominance is an attractive property to work with because it corresponds to a frequently used abstraction known as a mean preserving spread. If F,(.) SOSD Fa(.) , then

distribution F8 (.) is a mean preserving spread of distribution F. (.) .

1.4. Types of risk in agriculture

Ail agricultural enterprises, most especially in developing countries, operate under a situation of risk and uncertainty. These risks can be classified following two basic criteria: the source of the risk and the magnitude or diffusion of the risk.

1.4.1. Sources ofrisk

Risk has always been a part of agriculture. The uncertainties of weather, yields, prices, government policies, global markets and other factors can cause wide swings in farm income. A classification of the type of risks, as described in man y studies ( see for example Harwood et al, 1999; Mahul and Stutley, 2010) is presented here. Five basic types of risks are generally considered in the farming business, according to their sources:

Production or technical risks: concem variations in crop yields and derive from the unpredictable weather events/climate change, technological change, pests and diseases, and other natural hazards on production. This uncertainty affects both the quantity and the quality of the commodities produced. The effects of these uncontrollable factors are heightened by the fact that time itself plays a particularly important role in agricultural production, because long production lags are dictated by the biological processes that underlie the production of crops. Although there are parallels in other production activities, it is fair to say that production uncertainty is a quintessential feature of agricultural production.

Price or market risks: include the variability in agricultural input and output prices. Such volatility may be due to demand fluctuations, which are particularly important when a sizable portion of output is destined for the export market. Production uncertainty as discussed earlier, however, also contributes to price uncertainty because price needs to

30 adjust to clear the market. In this process some typical features of agricultural markets ( a large number of competitive producers, relatively homogeneous output, and inelastic demand) are responsible for generating considerable price volatility, even for moderate production shocks.

Financial risks: result from different methods of financing the farm business, subject to credit availability, interest and ex change rate, etc.

Policy-based or institutional risks: include the impact of policies and institutions, notably markets and related systems of exchange, property-rights arrangements, and uncertainty surrounding the legal framework. Policy-based and institutional risk can exacerbate the impact of natural hazards and market fluctuations. Structural adjustment policies, which often change the institutional "rules-of-the-game" (through changes in trade policies, credit policies, macro-prices, land and environment policies, etc.) can have pro-poor impacts, but often increase the degree ofuncertainty, especially in the short term.

Human resource or persona/ risks: refer to factors such as problems with human health or persona! relationships that can affect the farm business. Accidents, illness and death are examples of human crises that can threaten a farm business. Health risks have a direct impact on productivity, lead to defensive and other expenditures, and have a non money• metric impact on welfare. Human resource risks can also be associated with unavailability of personnel.

Although all the types ofrisk have important implications on the farming activity, it is not possible to generalize about the relative magnitudes as well as farmers' perceptions on these risks. In fact, the impact ofa given risk (or combination ofrisks) is a function of the frequency, intensity, duration and the spread of the risk. Moreover, there is significant heterogeneity among household risk perceptions, and their complex and constrained 6 decision making environment •

6 ln a recent participatory risk mapping by Smith. et. al., 1999, households listed and ranked their perceptions of the sources of risk, their frequency of occurrence and their intensity. Considerable beterogeneity of risk perceptions was found among a seemingly homogeneous group ofrespondents.

31 1.4.2. Spread and magnitude of risk

Risks that affect agricultural activities can also be grouped considering their level of diffusion among individuals. Under this classification, one can distinguish between systemic risks and idiosyncratic risks. A risk is systemic if it affects simultaneously a large number of farmers in the same geographical region. On the contrary, a risk is idiosyncratic if it affects only a single farmer. The spectrum in-between these two extremes of individual idiosyncratic uncorrelated risk and systemic perfectly correlated risk are wide (see Figure 1.4). In general, production risk tends to be more idiosyncratic while price risk tends to be systemic.

Figure 1.4: Risks idiosyncratic and systemic

Death, illness, Drought, Low yield, quality production flood, price of the output, crop failure, asset risk, conflicts, revenue, etc. loss, death of contagious livestock, etc. diseases, etc.

Systemic risks Idiosyncratic risks

No Correlation Perfect Correlation

Source: Cordier (2008)

1.4.3. Impact of risk in farming

The obvious impact of farm-system risk is to make management of the farm more difficult. The outcome of a management decision under risk cannot be sure. Ex ante, outcomes may be good or bad. Ex post, with hindsight, there are likely to be regrets about what might have been. Ali that the farmer can do, given the imperfect information available, is to make that choice among the available risky alternatives which most appeals to him/her given bis/ber preference for outcomes and degrees of belief in their occurrence.

32 A second impact ofrisk, especially for small-scale farmers, is that it generally makes them cautious in their decision making. Farm household survival demands that greater cognizance be taken of possible adverse outcomes (i.e., downside risks) than of possible good outcomes. While good outcomes (i.e., upside risks) are inherently attractive, for resource-poor small farmers this attraction must be expected to be more than outweighed by concem for the possibly disastrous impact of adverse outcomes. Downside risks challenge farm-system survival, particularly if a series of adverse outcomes should occur.

Finally, in terms of impact, there is no doubt that uncertainty leads to occurrence of a risk premium in the reasoning of the producer (Anderson et Danthine, 1980). The risk premium induces a shift in the total cost function and reduces the attractiveness of the production activity. The risky sectors then become economically Jess efficient compared to the results at the optimum without risk. The more important the perceived risk is, the more the economic inefficiency increases. The inefficiency is represented by a shift to the left of curve of total cost by a vector representing the risk premium. Symmetrically, inefficiency can be explained by a fall in the curve of the expected revenue when the prudent producer considers the certainty equivalent of the revenue rather than bis expected value (see Figure 1.5).

Figure 1.5: Supply ln presence of risk

Price

emand Supply

Marginal Cost

p•

Q. Q* Output Source: Bonjean and Boussard, 1 999.

33 In presence of risk, a risk averse producer does not equate his/her marginal cost with the

mean price (P) but chooses a level of expected supply Q' by equating his/her marginal

7 cost to the certainty equivalent of the mean price • The certainty equivalent is normally far below the mean price, the more as price volatility is larger and the farmer poorer. The difference between the mean price and its certainty equivalent is the risk premium which must be added to the marginal cost to obtain the supply curve. The consequence then is a supply curve which lies above the marginal cost curve. The equilibrium point, instead of being at (Q*, P*) is rather at ( Q', P') with a smaller level of output and a higher price.

1.5. Risk management strategies in farming

The purpose of risk management is to control the possible adverse consequences of uncertainty that may arise from production decisions. Because of this inherently normative goal, stating the obvious might yet be useful: risk management activities in general do not seek to increase profits as such but rather involve shifting profits from more favourable states of nature to Jess favourable ones, thus increasing the expected well-being of a risk-averse individual (Moschini and Hennessy, 2001). This requires an evaluation of trade-off between changes in risk, expected returns and entrepreneuriat freedom among others. The focus must be on the "risk that matters." This may involve the prospect of losing money, possible harm to human health, repercussions that affect resources or other events that affect a person 's welfare.

For an individual farmer, risk management involves finding the preferred combination of activities with uncertain outcomes and varying levels of expected returns (Harwood et. al., 1999). From this point of view, risk management can be defined as choosing among alternatives for reducing the effects of risk on the farm, which in tum affects the farm 's welfare position. In developed countries, farmers usually have access to a number oftools which have a direct risk management role. These include contractual arrangements ( e.g., forward sales, insurance contracts) as well as the possibility of diversifying their portfolio by purchasing assets with payoffs correlated with the retums on production activities. However, in rural areas of developing countries, the provision of financial services faces

7 The certainty equivalent price is the least certain price that a risk averse producer would be willing to receive instead of the random price.

34 various challenges. As a consequence, insurance and credit markets are not well developed or do not exist at ail. In order to mitigate risk effects, households replace formai markets, as formai insurance and financial markets, with several kinds of informai arrangements and institutions even though these traditional strategies mitigate only a small part of overall risk (Alderman, 2008; Dercon, 2002).

Dealing with risk can take place at two different points in time. First, there are strategies that take place before the shock occurs. These strategies are usually called ex ante or risk• managing strategies. They aim at smoothing the flow of income of the household. Households take steps to protect themselves from adverse income shocks before they occur (Morduch, 1995). The second set of strategies takes place after the shock bas occurred. This is why they are called ex post or risk-coping strategies. These strategies aim to smooth consumption given the variability of income and other shocks. These strategies allow the household to smooth consumption across time (Alderman and Paxson, 1992).

Households, in their efforts to deal with risk, can adopt one or more strategies at the same time. Tuen households have to weigh the marginal benefits and costs of taking different strategies. While the distinction between risk management and risk coping strategies is very useful from a theoretical perspective, its importance is less crucial from a practical point of view. In their daily life, farrners experience "the fear and the fate" at the same time (Dercon, 2007). The following sub-sections review the main risk reducing strategies used by agricultural households in the developing world where access to financial and insurance markets is limited.

1.5.1. Ex-ante strategies

When full financial and insurance markets do not exist or are not functioning properly, households may choose among different income generating opportunities to reduce the variability of their income. Income smoothing strategies often include the diversification of agricultural activities ( crop and field diversification), conservative production plans, the choice of a diverse portfolio of occupations, migration, etc. Risk-averse households may choose lower average income in exchange for lower risk or lower income variability. The most comrnon way of reducing the variability of income ex-ante is through the

35 diversification of income sources. As long as the sources of income are not perfectly covariate, using different income sources reduces total income risk.

Income diversification is widespread in rural areas of developing countries. Households in rural areas can reduce risk by choosing crops whose yields or prices display low correlations, planting crops on scattered plots that are subject to different weather shocks, using a variety of production techniques, or choosing a blend of farm and non-farm occupations (Alderman and Paxson, 1992). In some places, it is common for the household to have different plots of land, sometimes separated by long distances, and for non-farm incarne to contribute to a big share of total incarne.

Smoothing income cornes at a cost; the households face a trade-off. They want lower income variability but, in most cases, they have to forgo average incarne. Morduch ( 1995) reviews the topic of incarne smoothing in developing countries and finds different mechanisms for income smoothing such as: favouring variability-reducing inputs and production techniques (more use of labour while some other inputs may be used Jess intensively) and diversification through off-farm activities (households with higher fann profit volatility are more likely to have a household member with steady employment).

Rosenzweig and Binswanger (1993) investigate the effects of risk on the allocation of production resources among farmers, differentiated by wealth. Using the panel data set from the ICRISAT villages in India and its information on investment, wealth and rainfall, they examine how the composition of productive and non-productive asset holdings varies across farmers with different levels of total wealth and across farmers facing different degrees of weather risk. The results show that farmers in riskier environments select portfolios of assets that are less sensitive to rainfall variation and Jess profitable.

Alderman and Paxson (1992) find that the trade-offbetween profit variability and average retums to wealth is significant, and the Joss in efficiency associated with risk mitigation is considerably higher among the poorer farmers. Risky portfolios yield higher retums per unit of wealth. They reveal that in India, poorer households tend to choose less risky production strategies and the tendency to shift to a Jess risky portfolio is greater among households with Jess inherited wealth. Farmers with borrowing constraints choose Jess

36 risky portfolios of crops and plots. More vulnerable households are more likely to diversify plots.

1.5.2. Ex-post strategies

Ex post strategies help households to smooth their consumption given their income risk and consumption shocks. In the absence of formai insurance and credit markets, households can dilute risk over lime (using intertemporal strategies), across other households (using risk-sharing strategies), or they can combine these two strategies. Limitations of ex post consumption smoothing mechanisms decrease the profits of the poorest households. ln the developing world, studies have shown that households in rural areas regularly use at least one of the following strategies to smooth consumption, but most of them use a mix of strategies. Households can engage in borrowing and lending activities, accumulating and selling assets and consumption goods (grain and livestock), giving and receiving transfers, and changing their labour supply, including their demand for child labour. If the rural households are risk averse, they should group together and share risk within social network. If risks are fully pooled, the growth in household consumption should track the growth in group average consumption and should not depend on the current income of the household (Townsend, 1995). Although, different studies have rejected the hypothesis that poor rural households in developing countries can fully insure themselves by using informai institutions, these households do use informai transactions to reduce the impact of shocks (Udry, 1994; Townsend, 1995). Households usually rely on networks of neighbours, family members and relatives to engage in informai credit transactions and transfers (Rosenzgweig, 1988; Udry, 1994; Fafchamps and Lund, 2003).

Informai credit transactions take on a special role when insurance markets are incomplete, by allowing households to smooth income and consumption over time. Frequently, informai credit transactions present some of the following characteristics: (i) transactions are made without any witness or written proof but, in most of the cases, they occur between people who know each other well and between people of the same village or town, thus reducing the information asymmetries between borrower and lender; (ii) the transactions are very informai; borrower and lender agree on the size of the loan but, in

37 consumption. In different parts of the world, households use their precautionary reserves as a strategy to smooth consumption. Sorne of these reserves are kept in the form of livestock, crop inventories, consumer durables, national or foreign currency, and, in some cases, even land".

1.6. Conclusion lt is abundantly clear that considerations of uncertainty and risk cannot be escaped when addressing most agricultural economics problems. The demands imposed on economic analysis are complex and wide-ranging, with issues that extend fi:om the pure theory of rational behaviour to the practicality of developing risk management advice. The economics profession at large, and its agricultural economics subset, bas responded to this challenge with a wealth of contributions. In this chapter we have emphasized theoretical analyses as they pertain to production decisions at the farm level. The EU mode), in spite of its limits, provides the most common approach to characterizing rational decisions under uncertainty, and it has been the framework of choice for most applied work in agricultural economics.

Models of decision making under risk and uncertainty bring to the forefront the fact that decisions will be affected in a crucial way by the agent's risk preferences, i.e., his/her attitude toward risk. Consequently, it is quite important to quantify the degree of agricultural producers' risk aversion. The limited access to formai financial and insurance markets makes households develop different schemes to deal with risks in developing countries. However, neither existing strategies nor government policies have solved the farmers' risk exposure problems. Risk continues to have the potential of adversely affecting farmers' welfare, as well as carrying implications for the long-run organization of agricultural production. The cocoa sector in Côte d'Ivoire is an example of agricultural sector exposed to many sources of risks which affect farmers' production decisions and welfare. The next chapter gives an overview of the Ivorian cocoa sector and highlights the main constraints faced by producers.

8 Townsend, 199S; Udry, 1995; Deaton and Paxson, 2000; Jalan and Ravallion, 2001; Rodriguez-Meza, Southgate and Gonzalez-Vega, 2004.

39 CHAPTER TWO: OVERVIEW OF THE COCOA SECTOR IN COTE D'IVOIRE

2.1. Introduction

Agricultural commodities play a key role in the economies of man y low-income countries and households. The United Nations Conference on Trade and Development (UNCTAD, 2005) estimates that one billion people depend on agricultural commodities for a substantial portion of their income. Cocoa in developing countries provides a perfect example of agricultural commodity dependence. Indeed, cocoa is grown in more !han 50 countries world-wide, but the production is limited to three main growing areas: West Africa, Latin America and South East Asia. West Africa is the major producing region, with a market share of around 69%. The increasing dominance of Côte d'Ivoire as the world's largest producer has perhaps been the most spectacular development in the cocoa industry in recent times (LMC International, 2001).

Côte d'Ivoire has occupied a special place in the world of cocoa since it captured the position of leading world producer of cocoa beans from Ghana in 1977. This place has been characterized above ail by exceptional growth in production. The country currently accounts for around 41 % ofworld output compared to just 15% in 1975. This commodity generates 35% of the total export revenues of the country, accounts for 15% of the GDP and 20% of the tax revenue. The cocoa sector employs about 700,000 fanners and constitutes the source of livelihood for more than one third of the Ivorian population. Because of its importance for the overall GDP, export eamings and employment, and its forward and backward linkages to the non-farm sector, the cocoa sector plays a key role in economic development and poverty reduction in Côte d'Ivoire. Consequently, the Ivorian economy is highly sensitive to ail uncertainties related to the production pattern as well as the marketing channel of this commodity. For exarnple, when international prices are highly volatile, the govemment has difficulties meeting debt service obligations and making appropriate development plans such as much needed investments in basic health, education and infrastructure, that is, reducing poverty. Moreover, because the majority of cocoa producers often operate at a small, specialized scale, they are particularly vulnerable to a number of constraints including price volatility, low productivity, pests

40 and diseases, high taxes, poor infrastructure and weak organisations. These difficulties tend to increase poverty levels among farmers and their families.

The objective of this chapter is to investigate the cocoa sector in Côte d'Ivoire by addressing the structure and economics of cocoa production as well as the pricing and marketing systems. The chapter also sheds light on the trend and main constraints inhibiting the contribution of the cocoa sector to socioeconomic development in the country. Specifically, the rerninder of the chapter is organised as follows: the next section gives an overview of the socio-demographic and geographic settings of the study country as well as the study area. Section 2.3 provides some insights on the agronomie and historical background of cocoa production in Côte d'Ivoire. Section 2.4 and 2.5 respectively present the marketing system and the main actors in the cocoa value chain. A picture of impediments to the Ivorian cocoa industry in general and to the smallholder cocoa farms in particular is presented in section 2.6, while section 2.7 gives the summary of the chapter.

2.2. Country overview and the study area

2.2.1. Geographical settings

Côte d'Ivoire is a Sub-Saharan nation in West Africa. The country demonstrates notable geographical diversity; its terrain transitions from equatorial rainforests in the south to grassy and shrub-covered savannah in the north. Its elevation pattern also follows a latitudinal trend, as the predominantly fiat landscape rises from the southern coast to the northern plateau and northwestern highlands. Historically home to diverse plant and animal species, the nation's ecosystems have been subject to substantial pressures over the last several decades. Foremost, the region's rich soi! and tropical climate have made it ripe for agricultural production. As a result, a large percentage of the country' s once dense and extensive rainforests has been cleared for cultivation or timber.

The country borders the Gulf of Guinea in Atlantic Ocean to the south (515 km of coastline) and five other African nations on the other three sides: Liberia to the southwest (716 km), Guinea to the northwest (610 km), Mali to the north-northwest (532 km), Burkina Faso to the north-northeast (584 km) and Ghana to the east (668 km). In total,

41 2 Côte d'Ivoire comprises 322 462 km , of which 318 000 km2 is land and 4 460 km2 is water.

Côte d'Ivoire has three natural regions: a narrow coastal strip dominated by large lagoons, a forest belt extending from the coastal fringe to the central interior, and a northern savanna.

./ The Coastal Lagoon Region

Running along the Gulf of Guinea, the country's coastal strip stretches from the Ghanaian border in the east to the mouth of the Sassandra River in the west. In addition to beaches and marshes, the terrain consists of numerous islands and sandbars interspersed by an extensive series of lagoons, many of which are interconnected. Human settlements and farmlands have replaced much of the natural vegetation along the coast; the existing plant life ranges from palm trees and shrubs to small patches of mangrove forest.

./ The Forest Belt

Côte d'Ivoire's dense forest belt extends from the southern coast to the central interior, where it transitions into the grass and woodlands of the northern savanna. In the southwest, this region encompasses the coastal area between the Liberian border, where the Cavally River empties into the gulf, and the mouth of the Sassandra River. In the south and southeast, it begins north of the lagoon region. As a whole, the forest belt has been severely affected by high rates of deforestation, as its valuable hardwood trees and fertile soil have been exploited by loggers and farmers. Most notably, cocoa and coffee plantations have replaced expansive tracts of dense forest.

./ The Savanna

A vast expanse of savanna - broadly defined as grassy plains scattered with scrub and occasional woodlands - covers the northern half of Côte d'Ivoire. In general, this region has a lower population density than the southern half of the country. The region, as a whole, serves as a transitional zone between the tropical rainforest to the south of the country and the Sahara Desert to the north of Africa .

42 2.2.2. Population-Poverty situation and profile

2.2.2.1. Demographic situation

According to the World Development Indicators (World Bank, 2008), the total population in Côte d'Ivoire is 20.59 millions, growing at an annual average rate of 2.3 per cent. The fertility rate is 4.6% and the life expectancy at birth is 57 years. The international migrants as a percentage of the total population is 12.3 compared to 22.3 in 1960. This decline in the percentage of migrants is due to the political crisis which started in 2002. The population is young with 43% being under age 15. The economically active population, as a percentage of the total population is 41 % with 33% of women and 67% of males. The agricultural sector represents 45% of the total economically active population. The adult literacy rate(% aged 15 and above) is 48.7%.

Côte d'Ivoire is home to over 60 ethnie groups. Many ofthese groups share cultural and ethnie affinities, especially after years of internai migration and intermarriage. Broadly, the Ivoirian population can be divided into four general groups (Mandé, Voltaïque, Akan and Krou) which in turn have various subgroups. In the northwest savanna region, the Mandé peoples form the major ethnie group. Historically, this group includes the Malinké, Dioula and Bambara. They eam their livelihoods from agriculture, animal husbandry, and trade. The Mandé peoples are predominantly Muslim. Predominant in the northeast, the Voltaïque, or Gur, group includes the subgroups of the Senufo, Lobi, and Bobo tribes. For the most part, these tribes practice either Islam, animism, or other indigenous faiths, although some may be Christian converts. The Senufo tribes, who comprise the largest of the Voltaïque groups, work primarily in agriculture. The Akan are the dominant ethnie group of the southeast. Many of them are descendents of migrants from the historie Ashanti Kingdom, and they therefore maintain strong cultural ties to Akan groups in Ghana and Togo. Major subgroups of the Akan include the Baoulé and Agni peoples. The Baoulé are agriculturists whose primary crops include coffee, cocoa, and yams. In the southwest, the Krou peoples comprise the major ethnie group, with the Bété tribe forming the largest subgroup. They are known as fishermen, stevedores, and skillful navigators. Many of them also work in agriculture in the forest region. The majority of them are Christi ans.

43 2.2.2.2. Profile of poverty in Côte d'Ivoire

With a per capita GDP of 1700 USD in 2008 growing at a rate of 2.3%, Côte d'Ivoire ranlcs among the poorest countries in the world according to the United Nations Human Development lndicators. Since the beginning of the civil war in 2002, political turmoil has continued to damage the economy, resulting in the loss of foreign investment and slow economic growth. GDP grew by nearly 2% in 2007 and 3% in 2008. Per capita income has declined by 15% since 1999. The unemployment may have climbed to 40-50% as a result of the civil war. Poverty Reduction Strategy Paper (PRSP, 2009) identifies any person with consumption expenditure below 661 FCFA9 per day, or 241,145 FCFA per annum to be poor. Today, one out of every two people is poor compared to one out of ten in 1985 and the number of poor people has been multiplied by 10 in the space of a generation. Poverty has, therefore, increased on a steady trend, going from 10.0% in 1985 to 36.8% in 1995 and to 33.6% in 1998 before increasing further to 38.4% in 2002 and eventually to 48.9% in 2008, as a result of the successive socio-political and military crises. Poverty is more acute in rural areas than in urban areas. The poverty rate increased from 49% in 2002 to 62.45% in 2008 in rural areas as against 24.5% and 29.45% over the same period in urban areas.

2.2.3. The study area: Biophysical environment

2.2.3.1. The study region

This study was carried out in the Bas-Sassandra region which is located in the rain forest region of the south-west of Côte d'Ivoire (see Figure A.2 in Appendix 2) and contains four (4) departments: San-Pédro, Soubré, Sassandra and Tabou. The region covers a surface of 26.960 krn2, that is, 1/8 of the national territory with a population bordering 2.500.000 inhabitants distributed in the four departrnents. The population is, in majority, young and rural. The native people are Kroumen (Tabou, San-Pédro, and Grand-Béréby), Bakwé (Méagui), Néyo and Godié (Sassandra and Guéyo) and Bété (Buyo, Guéyo and Soubre). But the economic assets of the region have attracted many populations from various localities of Côte d'Ivoire and also from foreign countries as Burkina Faso and Mali. Second economic axis of the country after the region of lagunes, the region of Bas-

9 1 USD= 470 FCFA

44 Sassandra is equipped with consequent basic infrastructures: port (first port of cocoa export in the world), airport, roads, modem and powerful equipment of telecommunication, a very diversified ban.king network and departments of health etc., which makes it an open region. However, the area is characterised by widespread poverty, severe deforestation through slash and bum agriculture, lack of property rights on land, poor access to credit and existence of several types of agricultural risks.

The climate in the region of Bas-Sassandra is wet tropical type with two (2) dry seasons and two (2) rainy seasons. The relief is made up of many inland valleys (bas-fonds) favourable to the development of subsistence crops. There are also some plates which in general dominate all the ground with the presence of some hills. The vegetation consists of tropical forest (characterised by large trees and green plants all the year). The precipitation is averaging 2,370 mm per year. The total yearly precipitation is divided into two rainy seasons, a longer period which lasts from March to June with smaller quantities of rain and a shorter, but more clear and intensive, rainy season from the end of August to the middle of November. A distinct dry season is missing in the south-west. However, during the months of December and January, precipitation, which only occurs on a few rainy days; rarely exceed the amount of 50 mm (Anhuf 1997 cited by Mund 1999). The monthly average temperature in this tropical lowland region varies between 24.5 and 26. 7°C, though a maximum of 28°C can be reached in March (Szarzynsky 1994 cited by Mund 1999).

The morphology of the soil in the region of Bas-Sassandra is ferralitic predominance. Deeply weathered and strongly leached ferralsols with a high plinthic content are the characteristic soit types on crests and upper slopes in south-west Côte d'Ivoire, as well as in many other regions of tropical West Africa. This type of soi 1, very rich, is particularly favourable to agriculture.

2.2.3.2. The department of study

Situated in the region of Bas Sassandra, the department of Soubré is the principal area covered by this study (see Figure A.3 in Appendix 2). It's about 452 km from Abidjan (economic capital of Côte d'Ivoire) and is located on 5°78' latitude, -6°60' longitude and at an altitude of 143 meters. According to the GeoNames geographical database (2008

45 estimates) the population of Soubré is 58492 inhabitants with two major ethnie groups (Bété and Bakwe). The study was carried out in the whole department subdivided in seven (7) Sous-Prefectures'": Buyo, Grand-Zattry, Meagui, Soubre, Okrouyo, Liliyo, Oupoyo.

The department of Soubré is subject to a climate type equatorial characterized by the sequence of a dry season (from November to February) and a long rainy season (from March to October with a flagging period between July and September). The annual precipitation is averaging 1,700 mm and the daily temperature is between 26°C and 27°C. As part of the Bas-Sassandra region, the department of Soubré is in the forest zone, favourable to agricultural development. This area is therefore very rural with a higher number of migrants from other region of the country and from the neighbouring countries. Major export commodities as cocoa, coffee, oil palm and hevea are produced in Soubré as well as many subsistence crops (yam, rice, plantain, tomatoes, groundnut, etc.).

2.2.4. Relevance of the department for the study

For the development process in rural area, tools and data are needed to deal with the risky environment inherent to agricultural activities. In this vein, this thesis gathered data in order to assess cocoa farmers' behavioural response to risks. The department of Soubré, which is at the heart of cocoa belt in Côte d'Ivoire, definitely appeared to be representative in terms of cocoa production capacity, population heterogeneity, and activity diversity. In fact, according to the Statistics Division of the Ministry of Agriculture 2007, this department is the first cocoa producing area in Côte d'Ivoire supplying alone 23% of the national production (see Table A.l in Appendix 3). As such, Soubré comprises several villages and encampments where native, national migrants, Burkinabe and Malian populations, as well as other nationalities do exist. Generally speaking, immigrants live primarily as much in the villages as in encampments. The majority ofthese immigrants live there permanently.

10 The Sous-Prefecture is an administrative subdivision (or a circonscription administrative déconcentrée) which carries out limited functions, overseen by an appointed Sub-Prefet who reports to a Prefet appointed by the National Govemment at the Departmental levcl.

46 2.3. Agronomie and historical background of cocoa production

2.3.1. Origin and technological features of the cocoa tree

The cocoa tree, with his scientific name Theobroma cacao belongs to the Sterculiacee family. 1t is native to Central and South America and is cultivated extensively for its seed, which is the source of cocoa, chocolate, and cocoa butter. The cocoa tree is a wide• branched evergreen that grows up to 7.5 m tall and has a lifespan from 25 to 40 years and reaches its full development towards the age of 10. The age of the cocoa tree is thus a significant determinant of output. The fruit, called cabossa, looks like the form of the papaya or the apple quince according to the variety. The dimension oscillates between 15- 30 cm of length and 7-10 of width. The externat wrap is marked by 5-10 furrows that it actually hardens to the maturation. The young is green fruit and it becomes yellow and red to the maturity. Every cabossa contains about forty seeds, called faves or beans. After extraction from the fruit, the beans are placed in piles, covered with banana leaves, and allowed to ferment; afterward they are dried to prevent moulding. They are then sacked and shipped to chocolate or cocoa manufacturers. There are three broad types of cocoa: the Forastero, the Criol/o and the Trinitario which is a hybrid of Forastero and Criollo. Within these types are several varieties.

• Forastero, which now forms the greater part of all cocoa grown, is hardy and vigorous producing beans with the strongest flavour. This is the type most widely grown in West Africa and Brazil. lt has a smooth yellow pod with 30 or more pale to deep purple beans. • Criollo with its mild or weak chocolate flavour is grown in Indonesia, Central and South America. Criollo trees are not as hardy and they produce softer pods which are red in colour, containing 20-30 white, ivory or very pale purple beans. • Trinitario plants are not found in the wild as they are cultivated hybrids of the other two types. Trinitario cocoa trees are grown mainly in the Caribbean area but also in Cameroon and Papua New Guinea. The mostly hard pods are variable in colour and they contain 30 or more beans of variable colour but white beans are rare.

47 2.3,2, Production pattern

The introduction of cocoa into Côte d'Ivoire dates from the second half of the nineteenth century, but the sector did not really begin to develop until the 1930s, under French colonial rule. Cocoa production began in the southeast of Côte d'Ivoire and today it is grown throughout the southern part of the country (forest areas) and mainly in the western zone (south-west, centre-west and west). Cocoa farms in Côte d'Ivoire are estimated to cover in average 2.5 million hectares with a structure of production characterized by the predominance of small scale producers (over 500 000) who own between 1.5 and 5 hectares and only a few medium - sized farms (only 12% of growers have more than 12 hectares).

There has been a remarkable expansion in production since 1960 (the year of the independence). As a result, Côte d'Ivoire has occupied a special place in the world of cocoa ever since it captured the position of leading world producer of cocoa beans from Ghana in 1977 (see Figure 2.1 ). This place then has been characterized by exceptionally high level of production, which since the mid-1990s reached 35-42 percent of world supply.

Figure 2.1: Cocoa production ln Côte d'Ivoire and Ghana

1 p 1600000 r 1400000

~ 1200000 u 1000000 1 ' C 800000 1 i 600000 0 400000 n 200000 'I T 0 0 n ~ ~ ! ~ n • ...,._Côte d'Ivoire -.-Ghana

Source: FAO Statistical Database (2009)

There are a number of reasons behind the cocoa boom in Côte d'Ivoire including govemment campaigns, the opening up of virgin land in the west through Jogging activities, the construction of roads and bridges leading to the developing port of San

48 Pedro and the relocation of farrners to the west either because of lowering yields elsewhere or through lands flooding under dam construction. The country also enjoyed extensively available land as well as a good supply of migrant seasonal labour from neighbouring countries. Cocoa provided a comparatively good income and enthusiasm for cocoa continued as the government struggled to keep the producer price of cocoa supported.

However, in spite of this exceptional growth in production, the quality of the Ivorian cocoa beans has always been considered mediocre and the beans tend to trade at a slight discount to the terminal market. In the early 1980s a consortium of British chocolate manufacturers (under the Biscuit, Cake, Chocolate and Confectionery Alliance) set up a project to investigate the reasons behind the difference in quality between Ivorian and Ghanaian cocoa beans. They found that the Ivoirian farmer was capable of producing Ghanaian type beans but that there was no commercial incentive to do so. Moreover, the annual yield per hectares is very low, between 260 and 560 kg/hectare. This is due to insufficient fertilizer application and old cocoa farms. More than 60 % of the plantations are between 11 and 30 years old.

The production of cocoa is extremely concentrated. The main producing regions are: • West Africa: Côte d'Ivoire which is the leading world producer, Ghana which grows some of the best quality cocoa in the world, Nigeria and Cameroon; • South America: Brazil and Ecuador; • Asia: Malaysia and Indonesia, where cocoa is a relatively new crop, are becoming increasingly important growing areas.

The following graph (Figure 2.2) shows the distribution of cocoa beans production in the world in 2009. As it can be seen, cocoa is largely produced in West Africa with Côte d'Ivoire and Ghana representing more than 60% of the world production. But it is important to note that cocoa production was dominated by Caribbean and South American countries till the early 20th century, then moved to Africa and is now spreading in Asia. lndeed, when the forest rent is over, cocoa production moves to another country or continent.

49 Figure 2.2: World-wfde distribution of cocoa production (ln Tonnes)

Cl Côte d'Ivoire Cl Ghana •Other Africa cAmericas CAsia

Source: FAO Statistical Database (2009)

2.3.3. Harvest and Marketing Period

Although fruits mature throughout the year, usually two harvests are made: the main crop and the intermediary crop (also called mid-crop), The mid crop is usually much smaller than the main crop; however the relative size varies according to the country. Harvest and marketing period also varies from a country to another (see Table 2.1 ).

Table 2.1: C ketin2 perfods i duci - tri Country Main crop Mid-crop Brazil October-March June-September Cameroon September-March May-August Côte d'Ivoire October-March May-August Ecuador March-June October-February Ghana September-March May-August Indonesia September-December March-July Nigeria September-March May-August Source: FAO (2008)

2.3.3.1. The main crop

In Côte d'Ivoire the main cocoa harvest runs from October to March, which is roughly five to six months after the wet season. The main crop accounts for roughly 75 to 80% of the total cocoa produced in Africa. Farmers begin the harvest in October when the cocoa pods start to ripen. After harvesting the pods, farmers open them to remove the beans, and then the beans undergo fermentation and drying process before being bagged for delivery. Cocoa prices tend towards weakness in anticipation of harvest and rally when harvest is at

50 risk. In November prices tend to bounce back as the market begins factoring in harvest problems or political strife, as cocoa is grown in generally unstable political regions.

2.3.3.2. The mid-crop

The mid-crop harvest runs from May through August. The mid-crop accounts for roughly 15 to 20% of the total cocoa produced in Africa. The cocoa beans from the mid-crop are generally ofsmall size and lower quality. For this reason, the govemment of Côte d'Ivoire by Decree No 2000-192/ March 17, 2000 has decided to withdraw that production from the export market. The mid crop which amounts generally to 250,000 tons is entirely sold to local processors.

2.3.4. Cocoa production zones in Côte d'Ivoire

Cocoa is widely grown in Côte d'Ivoire (see Table A.l in Appendix 3) but the production is quite concentrated. In 2007, the major producing regions were the following: • The Bas Sassandra region which is the first cocoa production area in the country with 42% of the national production; • The Sud Bandaman region, with 12% of the national production; • The Haut Sassandra region, with 9% of the national production ; • The Fromager region, with 8% of the national production; and • The Moyen Cavally region, with 7% of the national production.

Cocoa farms are estimated to cover in average 2 million ha with a structure of production characterized by the predominance of small scale producers who own between 1.5 and 5 ha. The yield per ha is between 400 and 500 kg/ha. The tree stocks consists essentially non-hybrid plant material and are ageing; more than 23% of the trees are more than 25 years old and a further 40% are between 12-25 years old (Biodiversity and Agricultural Commodities Program, 2010).

51 2.4. Cocoa marketing system

2.4.1. Pre-liberalization marketing system

Before the liberalization of the cocoa sector which started in 1990, the marketing of cocoa was organized under a centralized stabilization fund called "Caisse de Stabilisation et de Soutien des Prix des Productions Agricoles (CSSPPA)" introduced by the French in 1965 during the colonial area in the country. Commonly known as CAIST AB, the "Caisse de Stabilisation" was a system of fixed price from the farm gate to the export point.

2.4.1.J. Ro/e of the CAISTAB unti/ the beginning of the 1990s

At the beginning of each campaign, the government through the marketing board CAIST AB, determined a schedule of prices called the "Barême or Différentiel," a guaranteed price for ail the operators along the marketing chain from the exporters to the farmers. In other words there was a guaranteed price for the exporters, the buyers and the farmers during each season. When the exporters sold their cocoa at a higher price than the guaranteed export price, the difference was paid into the CAIST AB to fund the stabilization fund. When the export prices were below the minimum export price the CAIST AB drew from the stabilization fund to compensate the exporters.

Through the years, CAIST AB developed severe inefficiencies, including bloated operating costs and some degree of political capture. In developing the barême, CAIST AB often overestimated the freight and insurance costs, leading to a lower starting FOB11 price from which it calculated farm gate prices (Mclntire and Varangis, 1999). Compounding this consistent error, marketing costs and export taxes were extremely high, equalling nearly 25 percent of the producer price during the 1998/1999 season (Gilbert and Varangis, 2003). These growing inefficiencies made the practice unsustainable when cocoa prices remained low for an extended period in the 1980s, although CAISTAB was able to succeed for many years in providing nominal inter-annual price stability to cocoa producers.

11 FOB is "freight on board" (export price)

52 2.4.1.2. The reforms of the 1990s

The first refonns started at the beginning of the nineties after Côte d'Ivoire loosed the 12 "cocoa war" • In 1990, the CAISTAB was forced to reduce by half the official guaranteed producer price and also to abandon the objective of year-over-year price stabilization in favour of intra-year price stabilization (within seasons). This did little for producers since their major concem was price variation between annual harvest periods (Mclntire and Varangis, 1999). Indeed, the goal of inter-annual price stabilization was abandoned after the extended decline of world prices had made it impossible to maintain the producer price across seasons. This is consistent with the argument of Deaton ( 1992) that inter-annual stabilization is difficult because slumps tend to last longer than booms; in such an asymmetrical market, even large stabilization funds go bust.

Finally, even if nominal price stabilization limited price volatility to producers, it became fiscally unsustainable. The objective of producer price stabilization, after changing from inter-annual stabilization to intra-annual stabilization, was totally abandoned with the total liberalization of the Ivorian cocoa sector and the dissolution ofCAISTAB in 1999.

2.4.2. Total liberalization of the cocoa sector

The inefficiencies of the parastatal marketing board, combined with the numerous subsidies and frequent corruption within govemment-controlled marketing channel, became too costly for the govemment, which faced massive pressure from international donors in the 1980s and 1990s to trim expenditures and to eliminate price controls (Gibson, 2007). The total liberalization of the cocoa sector in Côte d'Ivoire occurs in 1999. The liberalization program aimed to improve productive efficiency through an alignment of domestic prices with world prices and to give cocoa farmers improved prices, which were considered to be low in relation to the FOB price (ul Hague, 2004). However, with the abolition of CAISTAB and the removal of the guaranteed price, the operators and especially small-scale farmers found themselves faced with price risks, which they were not prepared to confront (ITF, 2002).

12 Between July 1987 and October 1989, Côte d'Ivoire suspended his cocoa exportation in order to put pressure on the world prices. This tentative manipulation of the international market failed and this lead to the introduction ofa series ofreforms.

53 2.4.2.1. Contents of the liberalisation

The full liberalization of the Ivorian cocoa sector in Augustl 999, led to some radical changes in the organization of the sector. The market liberalization had two complementary objectives. The first was to ensure that farmers would receive a higher proportion of world prices than had been the case in the pre-liberalization period. This often involved a reduction in (implicit or explicit) export tax rates. The second objective was to align incentives with world prices, both for farmers and more generally in the marketing chain, in the expectation that production and marketing would be more efficient. It was often hoped that these incentives would increase both production and revenues in the liberalizing economies. This price realignment process then involved an ending of previous inter-annual and intra-annual price stabilization arrangements, and paralleled the simultaneous abandonment of attempts to stabilize international prices though commodity agreements. At the same time, previously monopsonistic marketing systems were opened up to competition. See Gilbert and ter Wengel (2001) and Akiyama et al. (2001) for a summary of these developments.

2.4.2.2. Consequences of the liberalisation

The liberalisation of cocoa sector was tailored to improve farmers' welfare. Although it has opened domestic and export markets to competition, the replacement of parastatal by imperfectly competitive agents with market power has led to widening the previously large gap between the farm gate price and the world price; as agents in a long supply chain reap rents that previously accrued to the govemment as taxes. This has arguably left farmers at the mercy of both the unstable world market and concentrated multinational corporations, increasing the level of uncertainties and risks within the sector. Farmers also became dependent on the intermediaries who buy cocoa at the farm gate. AU these consequences have resulted in increased the vulnerability of farmers. For national and foreign processing firms, uncertainty increased about volumes and qualities delivered, and how farmers and initial buying agents would be financed following stricter conditions of bank credit. Therefore, to hamper the negative effects and take advantage of the liberalisation process, some new reforms had been introduced.

54 2.4.3. New reforms in the cocoa sector

2.4.3.1. The reforms of the 2000s

In 2000, new institutions related to cocoa marketing were, with the World Bank support, organised to provide some form ofprice predictability and stability in the market. Instead of the Caisse, new managing bodies in the cocoa sector were built.

• The Coffee and Cocoa Regu/atory Authority (ARCC)

The ARCC (Autorité de Régulation du Café et du Cacao) was created by Decree No 2000-751/0ctober 10, 2000. It was an independent public agency with administrative control and regulation of the cocoa sector and its operators. Its responsibilities included approving exporters, buyers and collateral managers, arbitrating conflicts, implementing international agreements, monitoring and implementing agreement with processing firms, making proposais to the Govemment on improving and coordinating the management of the cocoa sector and liaise with the Bourse du Café et du Cacao (BCC) for establishing a minimum producer price, undertaking harvest forecast, and maintaining statistics. The activities of ARCC were financed with a levy of 2, 96 FCFA on each kg of cocoa sold.

• The Coffee and Cocoa Exchange (BCC)

The BCC (Bourse du Café et du Cacao) was the marketing wing in the organizational structure for the cocoa sector. It was established by ordinance No. 2000-583 of August 17, 2000 and modified by Ordinance No 2001-46 of January 31, 2001. lt was a private organization whose objective was to regulate the marketing and export of cocoa.

The responsibilities of BCC included the following: to define a mechanism of guarantee of a minimum revenue and remunerative price to producers; to improve producer's revenue; to set up mechanism of export sale; to determine the procedure, condition and manner of allocation of exports rights; register exports sales contracts; centralize and coordinate coffee and cocoa export operations; to liaise with ARCC to forecast the harvest and maintain statistics; to monitor the export market; stocks and declaration of purchases of commodity upcountry; to monitor also the agreement with processing firms; to promote

55 small and medium exporters and cooperatives; and to promote the quality and the Ivorian label. The activities of the BCC were financed by a levy of 7.25 FCFA on each kilo of cocoa exported.

• The Coffee and Cocoa Regulatory and Control Fund (FRC)

The FRC (Fonds de Régulation et de Contrôle) was set up by decree No 2001-668 of October 2001. lt was a private body whose mission was to determine, in liaise with ARCC and BCC, a minimum farm gate price for farmers, and guarantee that minimum price using the provident fund. The provident fund was constituted by a levy of 60 FCFA on each kilogram of cocoa exported. Cooperatives were exempted from paying that levy when exporting up to 7,000 tonnes of cocoa, and local exporters (small and medium exporters) were also exempted when exporting up to 5,000 tonnes of cocoa. The FRC was also in charge of collecting on behalf of ARCC and BCC, and for its own account the levies paid by the exporters, monitoring the financial situation of the exporters and their commitments vis-à-vis ARCC and BCC and promoting the small and medium exporters and exporting cooperatives.

• The Coffee and Cocoa Producers Promotion Fund (FDPCC)

The FDPCC (Fonds de Développement des activités des Producteurs de Café et Cacao) was set up by Decree No 2001-512 of August 28, 2001. lts mission was to promote the activities of cocoa producers by financing their training, the marketing of their products and any activities undertaken to improve the functioning of the sector and the living conditions of the farmers. The activities of the FDPCC were funded by a levy of 35 FCF A/kg on cocoa.

2.4.3.2. Restructuring of the sector

In 2008, the govemment launched an investigation into possible corruption in the cocoa sector that led to the arrest of 23 high-ranking cocoa officiais. After that, the operations of the cocoa institutions were halted and a committee for the management of the sector ( Comité de gestion de la filière café-cacao) that provisionally look over the tasks of the cocoa institutions - ARCC, BCC, FRC and FDPCC - was created. In 2009, a committee charged with the reform

56 r of the sector was formed (Comité chargé de la réforme de la filière café-cacao) and entered into consultations with ail the stakeholders of the cocoa sector. The purpose was to arrive at a consensus for a restructuring process, which simplifies the institutional set-up of the sector and tackles its current problems at the Ievel of production, commercialization and processing. However, this reform couldn't achieve its goals due to political tunnoil in the country.

2.4.3.3. Recent deve/opments

A major restructuring of the cocoa sector is under way in Côte d'Ivoire under the new government. Following consultations with farmers, buyers and international cocoa traders, the government has drawn up a package of reforms aimed at improving the management of the sector, giving a fair price to farmers and boosting productivity. As had been widely anticipated, the sector is to be managed by a single body modelled on the defunct CAIST AB, which will have overall responsibility for regulating and stabilising the sector.

The body will be state-owned, although the government has indicated that private companies could have stakes in it and sit on its advisory board. Ali previous cocoa marketing structures, including the temporary management committee, Comité de gestion de la filière café-cacao, will be abolished and their staff incorporated into the new body. Key targets for the new body include the following:

ensuring that a minimum price of 50-60% of the international price is paid to farmers, to be fixed at the start of the season; in previous seasons the cocoa price has been "indicative"; ensuring that 35-50% of national cocoa production is processed in Côte d'Ivoire by 2015; in 2009/10 only 20% was processed in country; entering into pre-season purchase agreements with farmers for 70-80% of their crop to ease any fluctuations in the international market throughout the season; and boosting training for farmers and improving the quality and productivity of the cocoa crop.

There are also plans to impose quarterly quotas on the larges! cocoa purchasers to prevent them from manipulating the market, as well as to scrap tax breaks for exporters with local grinding operations. In 2010/11 (October-September) just four companies - Cargill,

57 Archer Daniels Midland (ADM), Barry Callebaut and Olam - bought 50.4% of national production, giving them enormous market power. For its part, the IMF has insisted that total taxes and duties charged on cocoa production must not exceed 22% of its value in order not to stifle development of the sector. The reforms are to be introduced gradually throughout the 2011/12 season, with the new regime and pricing system to be fully operational by the start of 2012/13 next October.

2.4.4. Cocoa market structure

2.4.4.1. Major actors in the cocoa marketing chain

There are about 3 to 4 million people working in the cocoa sector in Côte d'Ivoire. These include ail the agents intervening from the harvest of cocoa beans to the final chocolate products: the farmers, the buyers (pisteurs/traitants), the cooperatives, the exporters and the local processing companies.

Producers are the first link in the coffee and cocoa marketing chain. They undertake harvest and post harvest operations. After the post harvest operations the farmers put their products in bags and sell them either to their cooperative, or to the buyers (pisteurs/traitants), exporters and processors.

The buyers, called traitants, undertook the collection and transport of cocoa to the main shipping ports where the exporters are based. The traitants based in the main cities of the producing areas act through their agents called pisteurs. The pisteurs go to the farm gate level to buy cocoa from the farmers on behalf of the traitants. The traitants were licensed bytheARCC.

With regards to cooperatives, one should note that the cooperative movement started at the end of the 1960's in Côte d'Ivoire, when the government encouraged the producers to get together into associations for the marketing of their products. These Associations called "Groupements à Vocation Cooperative or GVC" were not real cooperatives, because they didn't have the necessary legal basis to act as cooperatives. ln December 1997 a new cooperative law was passed by the parliament. That new cooperative law gives the possibility to cocoa producers to form real cooperatives and compete on equal foot with the other private operators, and thus it constitutes a great opportunity for cocoa

58 producers to gain the most from the new liberalized environment. But the cooperatives in the cocoa sectors are still confronted with a lot of constraints that if not resolved will surely prevent them from gaining the most from the liberalized marketing system.

Other important actors in the cocoa chain value are the exporters. The exporters buy cocoa from the traitants, process and export them on the international markets. The exporters were licensed by the ARCC. They are based in Abidjan and San-Pedro, the two shipping ports of the countries. Since the liberalization of the cocoa sector the export market is heavily dominated by big multinationals. According to statistics published in March 2008 and relayed by Oxfam (2009), 60.78 percent of the annual cocoa export via the ports of Abidjan and San Pedro was purchased by eight companies (see Table 2.2).

le2.2: C fi Ped ------7 Percentage purchase Purchaslng company (Country of Head Office) (San Pedro & Abidian. March 2008) Carzill (US) 15.96% SAF Cacao" (Côte d'Ivoire) 12.54% ADM Cocoa (US) 9.03% Barrv Callebaut (Switzerland) 7.72% Outsoan Ivoire-Clam (Singapore) 6.75% CIPEXI-Continaf (The Netherlands) 4.68% Tropival - ED&F Man (UK) 4.10% TOTAL 60.78% Source: Oxfam (2009)

Concerning local processing companies, five companies are active in Côte d'Ivoire: MICAO owned by Cargill has a total processing capacity of 125,000 tons of cocoa beans per year. SACO, member of the international Barry-Callebaut Group has three factories with a total processing capacity of 100,000 tons of cocoa beans per year. SACQ produces mainly semi- finished products (cocoa butter) and chocolate. UNICAO, owned mainly by ADM has a processing factory which can handle 65,000 tonnes of cocoa beans per year. CEMOI-Côte d'Ivoire, an affiliate of the French Group CEMOI has a factory with a nominal capacity of 60,000 tonnes of cocoa beans per year. CEMOI-Côte d'Ivoire buys cocoa beans directly from buyers and cooperatives.

13 SAF Cacao (Société Amer Frère Cacao) is the larges! domestically owned cocoa exporter in Côte d'Ivoire. The firm, based in the major cocoa exporting port of San Pedro, has capacity to grind around 30,000 tonnes of cocoa beans per year.

59 2,4,4,2, Cocoa valu, chain

The presentation of the main actors of the cocoa sector in Côte d'Ivoire as well as their specific role leads to the illustration of the value chain for this comrnodity as shown in Figure 2.3.

Figure 2.3: Cocoa value chain for Côte d'Ivoire Marketing Chain Active Stakeholders Cocoa Status Price Received

VIiiage <:::::::::::::::· ..... C= -~= .. ··:~:::::::::> Cocoa beans Farm gate price

Cocoa beans Buying center ----- (:_,T ::::> __ price ------...,._ - - - - - ...... Buying centre Cocoa beans 'Entrée usine' < ~.~~~~·~<) ( ~:'.~.~ ) price ------~-~-~-!------...... ::r············--···························· Port ( Local Processors { Exporters \ Cocoa beans ••••. \ (Local/Multinational) / F.O.B pricc

············:::~::::::::· ···z._- ~--::· - Europe-USA Cocoa Butter, Processed product '· Foreign :, Liquor, Powder price ···················-··------.. ············· Source: Adapted from Wilcox and Abbott (2004)

2.5. Constraints faced by farmers in Côte d'Ivoire cocoa sector

In their daily life, farm households producing cocoa face a lot of constraints both market• marketing related as well as production constraints. Before turning to details in the subsequent chapters on microeconomic (household) level analysis, it is essential to highlight the main constraints related to cocoa production in Côte d'Ivoire. The discussions in the followings subsections, thus, focus on domestic and global impediments currently surrounding the Ivorian cocoa sector.

60 2.5.1. Domestic constraints

The difficult economic conditions that have been experienced by the smallholder Ivorian cocoa farmers can be explained by some major constraints at the national level.

2.5.J.J. Pestldiseases of the plants

Many hundreds of insects and pathogens have been recorded on cocoa. The most significant disease attacking cocoa is the fungal disease 'black pod', which is responsible for an estimated yearly Joss of about 44% of total global production. Ake et al 2009 estimated that cocoa black pod disease in Côte d'Ivoire can cause crop losses up to 30%. The cocoa 'swollen shoot' virus is another damaging disease across the West African region. Farmers in Ghana, Côte d'Ivoire, Togo and Nigeria have not been spared the devastation. But the disease is found mainly in Côte d'Ivoire and Ghana and accounts for about 15% of the global Joss of the crop. Taking this in terms of the 2007/08 crop output of about 3. 7 million tons, the disease would account for the Joss of approximately 555,000 tons throughout West Africa. Insect pests are also serious constraints to cocoa production in West Africa. Cocoa mirids are widely perceived as the most damaging. In outbreak years, especially in areas where trees have been neglected, losses could be up to 75%.

The existence of pests and diseases, combined with high cost or unavailability of pesticides contribute to low yields in Côte d'Ivoire cocoa farms. In a study on cocoa and coffee investrnent decisions, Ruf and Agpko (2008) found yields to be generally low, with the majority offarms not producing over 500 Kg/hectare (see Figure 2.4).

61 Figure 2.4: Yfeld of cocoa farm in Côte d'Ivoire

'0 t ~ 70 u, 60 o 50 o.. 40 -O 30 ; 20 ~ 10 ::::, 0 D D = = z D u u u u:o D D = = ê ~ ~ 8 ~·ij ê ~ ~ § ê ~ ~ ~ ~ ~ ê ~ 0 ~ ~ § ~ ~ i!~ § ~ ~ i ~ ~ ~ ~ j ~ i ~ ~ ~ Yield classification (kg/ha)

Source: Ruf and Agpko (2008)

2.5.1.2. Taxes in the cocoa sector

Taxation of the cocoa sector has been a complex combination of fiscal and quasi-fiscal levies. ln Côte d'Ivoire export taxes are considered as an important source of fiscal revenue and also as an indirect way to tax the land used to generate profit. Fiscal levies form part of budget revenue; quasi-fiscal levies are collected to finance sector institutional structures and are not included in fiscal revenue.

Fiscal levies have been a combination of specific and ad valorem taxes:

• Export tax (droit unique de sortie, or DUS) is a specific tax set in FCFA per kg of cocoa beans and is periodically revised. It was 220 FCF A per kg for the 2008-2009 crop season. The government has reduced it to 210 FCFA for the 2009-2010 crop season. Exporters pay the DUS when the cocoa cargo is loaded onto the vessel. • Registration tax (taxe d'enregistrement) is an ad valorem tax set as a percentage of the price CIF14 and is periodically revised. Because international cocoa prices rose, the government in 2008 temporarily increased the tax from 5 to 1 0% but intends to reduce it to 5% starting with the 2009-2010 crop season. Exporters pay the registration tax first when they register with the government as an official cocoa exporter and annually thereafter.

14 Cost, lnsurance and Freight price or the world market price

62 • Other state levies: There is a 2 FCF A per kg tax on itinerant buyers, and the European Union bas docurnented over 20 other implicit state charges (EU, 2005).

The total amount of quasi-fiscal levies fluctuates from year to year. For the 2008-2009 season, they totaled 31 FCFA per kg, a substantial reduction from 47 FCFA a year earlier. The amount is a function of a complex power balance between the central government and several public entities (ARCC, BCC, FRC, and FDPCC) involved in management and investrnent in the cocoa sector.

It is bard to establish with precision how much tax the cocoa sector pays because of the unstable tax structure; the DUS and quasi fiscal levies changing regularly, and active use of specific taxes, the ad valorem equivalents depend on international cocoa prices, so they may change from year to year. Rough estimates based on the October 2008 world price and the tax structure for the 2007-08 campaign suggest that the total of DUS, the registration tax, and the quasi-fiscal levies represented about 40% of the export price.

This high level of taxes inevitably has a negative effect on the grower price of cocoa. Compared with other major producers, the World Bank estimated that Ivorian farrners get 35-40% of the international market price, while their counterparts in Ghana and Cameroon receive up to 70% (see Figure 2.5).

Figure 2.5: Cocoa taxation in three major producing countries

$3 000 •• ~ $2 500 u E $2 000 ~•• SI 500 ~.. ,u: $1 000 ••• = $500 r0. . $0 Côte d'Ivoire Cameroon Ghana

DFarmer Price CTrading margms and incountry transport DExportcr costs and local processmg DOccan Freight IITotal taxes"

Source: World Bank, 2008 • Note: Total taxes include both export tax and parapublic tax

63 2.5.1.3. Other domestic constraints

Besides the constraints from pest and diseases and high taxation in the cocoa sector, other domestic factors are also causes of a plethora of uncertainties surrounding cocoa income. For instance, cocoa farmers in Côte d'Ivoire have limited access to information, physical infrastructure (transport facilities, roads, warehouses), and incur higher costs for farm inputs such as fertilizers, chemicals and machineries. Ivorian cocoa farmers are also subject to persona[ and family risk as illness and death. Like in most developing countries, the cost of falling sick is substantial in Côte d'Ivoire. In cocoa growing areas, climate is wet and humid thereby creating a fertile ground for many parasitic human diseases such as malaria which impacts on production (Audibert et al, 2006). The weakness of farmers organisations is also another constraint to farmers' well being. In fact, the development of producers' associations is a key to increasing the bargaining power of farmers. Associations can provide information to their members about international prices and production techniques and gaining access to credit and price insurance. They can also enable farmers to do some initial processing which significantly raises the value of cocoa beans. Local processing significantly raises the proportion of value-added retained at a local level by producers. By marketing bigger volumes of production, cooperatives could also sell their products directly to exporters and international buyers. Ali these problems tend to increase vulnerability and poverty levels among farmers and their families.

2.5.2. Global constraints

2.5.2.J. Price volatility

Besides the domestic constraints mentioned in the previous section, at the international level, cocoa price volatility also severely impedes and endangers the livelihoods of millions of farmers. As illustrated in Figure 2.6, after dipping dramatically at the end of the 1970s and then again at the turn of the century, world cocoa prices start recovering in 2001. Indeed, cocoa prices are highly volatile in the international market. Even compared to other agricultural commodity groups, cocoa bas exhibited a high degree of price volatility over the past four decades (ul Hague, 2004).

64 Ff1ure 2.6: World cocoa prlces evolutton

World Cocoa Prices (US $ffonne) 4 000,00 $

3 500,00 $ _j

3 000,00$ Fi 2500,00$ - 2 000,00 $ 1-- 1 500,00 $ ,---

] 000,00 $ 1- --- 500,00 $ -t~

0,00$ 1-:-:-- :-,-,~r,----,---.----,,---:-~---:-··, r 1-1-,-~. ,.-~------:---.-~ 1·i r·-~-î--r...... ------1·-:,·-~-, -, ,-----s;··: ~~i~~R~~~~~~~~~~~~~~88~~~ -~~-~-~--~-~~--~~-~--~-~-~~-~-~-~-~-~-~-0NN0N00N0N Source: International Monetary Fund IFS Online database (2010)

Volatility is defined as the variability of the price series around its central value; i.e. the tendency for individual price observations to vary far from its mean value (Huchet• Bourdon, 2011 ). lt measures the extent of observed fluctuations in international agricultural commodity prices.

Two kinds of volatility are found in the literature: an historical (realised) volatility and an implicit (future) volatility. The historical volatility reveals how volatile a price series has been in the past and is based on observed past prices. Besides the implicit volatility corresponds to the markets' expectation on how volatile a price will be in the future. In this thesis we are interested in measuring only the historical volatility based on observed world cocoa prices .

Severa} methods exist for calculating historical price volatility. Through the literature, price volatility measures have often focused on daily price range or on the standard deviation of prices or on the coefficient of variation.

./ Volatility as daily price range

The range represents the spread in prices during a specific period. In a market with an open bidding process, the range is often defined as the average spread between the bid price and the ask price during a specific time period. When bid and ask prices are not available, the range is measured as the difference between the daily high price and the daily low price.

65 ./ Volatfllty as standard devlatfon

Standard deviation in average price represents an absolute measure of the actual price fluctuation over a specific period. This is probably the most common method used for measuring volatility. The standard deviation represents the expected deviation from the average market price during the given period. A higher standard deviation represents greater price volatility .

../ Volatility as coefficient of variation

The coefficient of variation is a relative measure of price movement and is calculated as the standard deviation divided by the mean value. It is a useful comparative measure of price volatility. The main advantage ofthis measure ofvolatility is that it does not depend on the unit of measurement. A higher coefficient of variation represents greater price volatility.

In order to measure price volatility in this thesis, the coefficient of variation ( the ratio between the standard deviation and the average value of the series) bas been calculated on the series ofprices over the period 1960-1960. More precisely, we use a moving average to conduct the statistical analysis of the indicator of volatility. This method is chosen as it is more relevant in such analysis conducted over a long history of price changes. Indeed using a moving average is the best way to compute volatility with annual data. Volatility is calculated here over three years based on the International Monetary Fund Online database ( see Figure 2. 7).

Figure 2. 7: World cocoa price voladllty

Percentage Change in World Cocoa Prices 60,00% ~------....::.... ! JI 50,00% 40,00% 30,00% 20,00% 10,00%

Source: Calculated by the author based on IFS Online database (201 l)

66 The figure shows that world cocoa prices experienced high volatility, especially during the period 1974-1977 where the price climbed to over 3.000 USD per tonne in the international market.

2.5.2.1. Determinants of cocoa price volattility

Severa! factors contribute to the formation of cocoa price at the international level. Cocoa prices mainly respond to cocoa supply and demand factors.

On the suppl y side, the major drivers of world prices volatility are: • Weather and natural disasters: since cocoa production is concentrated among a small number of countries, weather phenomena like drought or unusually high rain levels in one or more of the major producers can cause supply concerns. Early season drought in Côte d'Ivoire in 2003 caused world prices to jump simply on the prospect that the main harvest in the country might have been lower than normal (Elliott, 2003). • Conflict in producing countries: when conflict arises, it can damage cocoa producing lands and make it difficult for buyers to transport the dried beans to a market or port. Continuing conflict in Côte d'lvoire's cocoa growing region has driven prices up, and also endangers Ivorian producers' livelihoods (Guerriere, 2005). • Structural oversupply: there is a lag between movements in world cocoa prices and production changes. lt takes two to five years after planting for a cocoa tree to become productive, but once in production, the cocoa pods are generally harvested even if cocoa prices drop. Once the initial capital investment is made and the trees are planted, smallholders face relatively low marginal costs in harvesting the beans (Gilbert, 1997). As a result, high prices will lead to new plantings that, once mature, will drive down world prices for a long period oftime.

On the demand side, the major drivers ofvolatility in world cocoa prices are: • Business cycles in key markets: since chocolate is primarily consumed in wealthy, industrialized countries, any slowdown in the developed economies will negatively affect world cocoa prices (Gibson, 2007).

67 • Increasing market speculation: commodity derivatives have become more and more popular among persona! and institutional investors as alternatives to traditional equities markets. A sort of bandwagon effect can result, in which investors with no stake in the prices of the actual cornmodity can amplify price movements on the world commodity markets.

Within the overall trend ofvolatility in cocoa world market prices, producer prices in Côte d'Ivoire had been very stable in the pas! under the government price stabilisation system run by the CAIST AB. With the liberalisation of the cocoa sector, empirical results rcgarding absolute price levcls have been mixed; but as one would expect, producer price volatility has increased (ul Haque, 2004) as a result of the abolition of the price guarantee function. The operators and especially small-scale farmers found themselves faced with price risks, which they were not prepared to confront (ITF, 2002). When the international markets prices fluctuate, the producer's prices fluctuate accordingly. Prices are no longer stable, farm gate prices now fluctuate on a daily or weekly basis compared with the fixed seasonal price set in the past, causing difficulties and more uncertainty for farmers in making short-term investment decisions (ICCO, 2006).

Figure 2.8 highlights the movements of cocoa nominal farm gate prices with reference to three main periods: before liberalisation, the beginning of the liberalisation and the full liberalisation. One should note that the phenomenon of price fluctuation can be observed from the beginning of the Iiberalisation in 1990 till the full liberalisation of the sector which lead to an increase in the fluctuations. Price volatility at the national level is largely the result of price transmission between the international and the domestic cocoa markets. Y et, more rigorous analysis of transmission mechanisrns is needed to establish fundamental correlations.

68 Figure 2.8: Nominal cocoa producer prlces per kg, Cilte d'Ivoire, 1966-2007

1 Inter-year Stabilisation s:=:n Full Liberalisation i_ , ;! 800 700 600 500 400 300 200 100 0

Source: Ivorian authoritics, Bourse du Café et du Cacao.

2.6. Conclusion

ln this chapter, we described the cocoa sector in Côte d'Ivoire. We discussed the marketing system and the main constraints faced by farmers in tbeir production activities. These constraints are related to instable cocoa prices, instable weather conditions, diseases of the plants and attacks from insects. The high taxation found in the lvorian cocoa sector compared to other cocoa producing countries tends to lower producer price in Côte d'Ivoire. The low farm gate prices combined with the volatility of prices since the liberalisation of the sector has motivated fanners who may lack resources to cope witb price risks to shift to alternatives perennial crops. Price risk, combined witb ail the other risks inherent to agricultural production might affect lvorian cocoa farmers' welfare as they become more vulnerable, particularly when lacking formai financial and insurance markets. The following chapters empirically analyse cocoa farmers' risk attitudes and provide insights on the risks they are really exposed to as well as the strategies they use to deal with such risks when financial and insurance markets are missing or imperfect.

69 CHAPTER THREE: MODELLING FARMERS' BEHAVIOURAL RESPONSES TO RISKS

3.1. Introduction

This chapter outlines the data and the research methodo]ogy used in order to achieve the objectives of the study. The purpose of this research is to contribute to the empirical literature on the nature of behavioural responses to risks in rural areas of low income countries. There are two aspects to the research. The first part is an experiment that seeks to help us understand how lvorian rural houscholds rcspond to the presence of risky alternatives. This is to determine the risk preferences schemes among farmers as well as identify the potential link between farmers' risk aversion behaviour and their own characteristics. The risk behaviour findings are then used as explanatory variable in a multivariate Pro bit model that tries to exp Iain the factors affecting farmers' adoption of risk management strategies in the absence of insurance market.

The quality of data collected for a study has always been a good indicator of the robustness and validity of the results obtained through the empirical analysis. In choosing the research method that would best fulfil the purpose of this study and answer the research questions most effectively, it is felt that a survey research method would be most appropriate. The chapter briefly presents the survey design which includes the sampling procedure, the survey questionnaire, the training of data collectors and the data collection process. Then the methodology for data analysis is discussed both for risk aversion and risk management decision-making.

3.2. Sample design and survey questionnaire

3.2.1. Sampling procedure

As stated earlier, this study was carried out in rural area in Western Côte d'Ivoire and concerned cocoa producers in the department of Soubré. Thus, the data were exclusively collected from cocoa producers. A former was considered as cocoa producer if and only if he possessed a least one cocoa field among others. Therefore, non-cocoa producing farmers were not included in the sample.

70 3.2.1.1. Sampling plan

The survey uses a Iwo-stage stratified sampling design with gcographical proximity as the stratification variable, the villages as the primary sampling unit and the households as the secondary sampling unit. A total of fifteen ( 15) villages from the seven (7) Sub• Prefectures of the department of Soubré are allocated to three strata (by geographical proximity, see figure 3.2). The selection of the villages is based on unequal probability sampling method. This method ensures that big villages (in terms of total number of cocoa producers) will be selected. Alongside the 15 villages, 91 encampments connected to the different villages were also surveyed (see Table A.2 in Appendix 3). The number of producers per strata is then determined proportionally to the size of the strata. In a sample village, a household is a qualified respondent if they reside in the village for over 6 months and engage in cocoa production activities (not necessarily within the village).

3.2.1.2. Sample size determination

Let P denotes the proportion of farmers exposed to risks in their management decisions. The interval estimation of P with a confidence level at 95%, in the case of multistage samples is as follow:

[ P- 2J P(l -=)dejf p + zJ P(l -=)dejf ] (3. 1) where de.If is the design effect 15 and n is the total size of the sample. If we impose

P(l - P)dejf . . 4 P(l - P) d = 2 ~---"---=- as the random error m response, we then obtam n = * de.If J n d 2 being the sample size.

For P=0.7, de.If =1.5 and d=0.06 we get n = 350. Considering a non-response rate of 5%, we finally have n = 367.S.

15 The design effect - the ratio of the variance of a statistic with a complex sample design to the variance of that statistic with a simple random sample or an unrestricted sample of the same size - is a valuable tool for sample design. However, a design effect fouod in one survey should not be automatically adopted for use in the design of another survey. A design effect represents the combined effect of a number of components such as stratification, clustering, uncqual selection probabilities, and weighting adjustments for non-response and non• coverage. Usually il lies between I and 2. ln this study, a design effect of 1.5 has bccn used for consistency.

71 In conclusion, according to statistic estimations, for a precision random crror set at 6% and a non-response rate of 5%, the expected size of the sample is 368, if we consider the variable of interest which is the proportion of farmers facing risks in their management decisions.

3.2.1.3. Distribution of the sample

The distribution of the sample within strata, including the number of villages per strata as well as the number (mean) of producers to be interviewed per village, is presented in the Table 3.1: s Sample Numberof Number (mean) of producers Sub-prefectures Strata Size villages to be interviewed per village

Zone] Buyo-Grand Zattry 138 5 27

Zone2 Okrouyo-Soubré-Liliyo 122 5 24

Zone3 Méagui-Oupoyo 108 5 22

Total 7 368 15 -

Source: Own computation from samphng

3.2.2. Research instrument

In line with the objectives of this thesis; that is, find the risk preferences of cocoa farmers and analyse the factors influencing adoption of possible risk-reducing strategies, the questionnaire include a special feature. lndeed, the questionnaire comprises two distinct parts: the first part is made of a number of experimental sessions designed to capture the farmers' attitude towards risk, while the second part is a standard household survey.

The questionnaire for the survey has been designed so that the questions go from those general in nature to those more specifically meant to gather the survey's target information (see Appcndix 8). It is divided into 6 sections and includes 68 questions in total. The formats of the questions are both closed and open-ended, depending on the purpose of the questions. The first section of the survey questionnaire concems general information such as the farmer gcnder, origin, age, education, size of the household, the size of cultivated land by the farmer, the total number of cocoa farms or fields managed by a farmer, etc.

72 The second section instead gathers information on the farmer's economic situation; that is the average amount of cocoa produced by the household and the received price, the household shares of cash income derived from cocoa, livestock ownership, etc. The third section seeks to find farmers' marketing decisions white the fourth includes questions about the production facilities used by the farmers. Explicitly, it concems information on how farmers' fund their production activities, their access to credit market, their labour decisions, the quality and quantity of inputs and the cost of chemicals used on the cocoa farms during the three previous cocoa seasons, etc. The fifth section relates to farmers' perception of risk sources and their risk management strategies used in absence of insurance market. This section also asks questions on farmers' interest in potential minimum price insurance. Lastly, the sixth section covers questions relative to farmers' access to information.

3.3. Data collection

3.3.1. Training of data collectors

Field data collectors conducted interviews with farmers in two weeks (from April 20 to May 03, 2009). Teams of two data collectors included a designated team leader, travelled to the different Sub-Prefecture to collect data. For the seven (7) Sub-Prefectures in the departrnent of Soubré, there were in total fourteen (14) data collectors (two (2) for each Sub-Prefecture), Ali members of the staffwere experienced field interviewers since ail of 1 them are employees at ANADER 6, the national agency for rural development in Côte d'Ivoire. They were also close with farmers in the department as they used to work with them. This special closeness existing between interviewers and farmers was a significant advantage that favours farmers' openness to the questionnaire.

The data collectors underwent extensive training over a ten-day period at a centralized location in Soubré, to be precise at a school. The training was led and overseen by statisticians who also help for the sampling and questionnaire design, with my coordination. The training included supervised on-site data collection in scveral farm

16 ANADER stands for Agence Nationale d'Appui au Développement Rural. lt aims at assisting farrners in their farm activities. There are many ANADER bureaus ail over the country and especially in rural areas. For our survey, we worked with the regional bureau of Soubré.

73 households, as well as in-depth coverage of the methodology and administration of each survey instrument. Training procedures included large group instruction and demonstration, and closely supervised small group practice sessions. A training manual was provided and included detailed procedures for each component of the data collection and question-by-question specifications for each instrument. Approximately 20 hours (the first three days of this training) were spent training data collectors to the experimental gambling approach. All data collectors were tested for proficiency in administering the experimental game as well as the survey questionnaire before being cleared to work with actual respondents. Additional topics covered in the training included gaining respondents' cooperation, respondents' rights, answer reporting, and quality control and monitoring. To make sure that the data collectors had a good understanding of the training sessions, a pilot survey or pre-test was very useful.

3.3.2. Pre-testing

This is an important stage in the preparation and training of data collectors. The questionnaire was pre-tested to obtain valuable feedback. This was to ensure that any possible problems with regards to wordings, the timing, the format of the questionnaire, the design of the experiments, and the clarity of the questions and instructions will be brought to light and corrected. The questionnaire was pre-tested on two (2) of the seven (7) Sub-Prefectures with a few respondents. The two Sub-Prefectures chosen were Soubré and Okrouyo. After the pre-test, some amendments were made before the final questionnaire was administrated to farmers. The pre-test and the questionnaire improvement took two weeks.

3.3.3. Household survey

The finalised questionnaire was administrated within a period of two wccks by facc-to• face interviews with primary decision makers (head of household) in April-May 2009. Face-to-face interviewing is the oldest mode of interview since it does not rely on modem communication technologies. Because it provides for the maximum degree of communication and interaction between the interviewer and the respondent, face to face interviewing is often associated with good data quality and is viewed by many survey researchers as the preferred mode of data collection for most survey topics.

74 The actual survey was conducted among 368 households in fifteen (15) villages and forty ( 40) encampments across the seven (7) Sub-Prefectures of the department of Soubré. In order to explain the objectives of the study and to encourage the farmers to be patient and open to the questionnaire, a meeting was held in each of the selected villages with them before the survey. Prior to the meeting, official letters with the heading of my university was sent to each Sub-Prefet; this one should be aware of the study so that he could inform his population in advance of the upcoming survey. This procedure also facilitated farmers' participation in the survey.

As already stated, two types of data were collected. The first set of data was collected

17 from "artefactual field experiments " used to elicit households' risk aversion behaviour through a number of experimental sessions or games. These data were obtained by asking respondents to choose between six lotteries with varying expected return and variance, where higher return can only be obtained by accepting higher variance. The subject's choice among these alternative prospects is taken as an indication of the degree ofher risk aversion. The games were played with giving the actual payoffs to farrners. The amounts of money were relatively large compared to average income in the area (the highest possible gain was approximately 20% of average total annual income in rural area). It was clear that the respondents enjoyed this part of the survey. Details on the experimental game will be given later in this chapter, especially in the methodology section.

The second data set came from a household survey executed aftcr the experimental sessions. Respondents were asked questions in order to fulfil the survey questionnaire described in section 3.4.1. The survey took approximately one (!) hour for each respondent. Each interviewer was assigned to survey a maximum of ten households per day in order to avoid weariness and therefore preserve data quality.

17 Our experiments are "artefactual field experiments" in the terminology of Harrison and List [2004]. Thal is, they involve taking procedures from the laboratory and applying them in the field. The use of field experiments in developing countries bas grown dramatically in recent years, and is reviewed by Cardenas and Carpenter [2007].

75 Other important aspects of the survey design, such as the use of language are handled directly by the data collectors, which therefore represent a filter between the people interviewed and the data processing unit. The questionnaire was made available in French. However, interviewers were free to resort to the local language where it was necessary in order to facilitate the communication with the interviewee. In the case the interviewer didn 't know any language spoken by the former, he also had the possibility to request the assistance of a third-party translator. Most of the farmers had completed the questionnaire satisfactorily. However, telephone interviews had to be conducted for few farmers if it is deemed necessary to obtain answers for questions which were not answered or were not clearly answered.

After the data have been collected, CSPro 3.3 software was used for the data entry and data processing. Due to missing data on one or several of the explanatory variables and in some cases farmers' difficulties in understanding the experimental games, six (6) individuals were drop out of the sample. The actual sample size used in the following analysis is then 362. The theoretical framework and the methodology for data analysis are presented in the next sections.

3.4. Assessing farmers' risk aversion and its determinants: The empirical

methodology

The existence of agricultural risk and its effects in low income countries where many key markets (such as credit and insurance markets) are missing is well-known. lt is largely acknowledged that the Jack of insurance causes inefficiency in production choices (Fafchamps, 1999; Dercon and Christiaensen, 2007; Alderman, 2008). The response of farmers to risk will depend critically on their risk preferences. Broadly speaking farmers can be risk takers, risk neutral and risk averse. The choice of practices will be unaffected by risk where farmers are risk neutral. They will seek to maximise profit within a set of financial, environmental and social constraints. Risk takers, on the other band, will tend to choose a set of actions that have higher profit but are more variable in terms of the potential outcomes. Risk averse farmers, in contrast, will tend to sacrifice some profit and adopt several strategies to reduce variability in profit. Then, an analysis of factors affecting the adoption of risk management strategies by farmers cannot be conducted

76 without looking first at farmers' risk aversion behaviour as well as its determinants. To carry out such analysis in this thesis, we are fortunate to have some direct information on farmers' risk attitudes, obtained from an experimental gamble with real payoffs.

3.4.1. Eliciting farmers' risk behaviour

Considerable research bas attempted to provide empirical evidence of farmers' risk attitudes though most of the existing empirical studies in developing countries are very recent. According to Young ( 1979), Lins et al. (1981) and Robison et al. (1984) there are three basic methods of measuring the attitudes to risk of agricultural producers: Direct estimation of the utility function, Experimental methods and Observed economic behaviour.

•:• Direct estimation of the utility function

This method involves direct interaction with the decision maker, who expresses his or her preferences among various alternatives. Regression techniques then enable us to obtain their utility function. Examples can be found in Francisco and Anderson ( 1972), Lin et al. (1974), Dillon and Scandizzo (1978), Bond and Wonder (1980), Hamal and Anderson (1982), and Feinerman and Finkelshtain (1996). The study by Scandizzo & Dillon (1978) was among the first and the famous to investigate attitudes towards risk. Scandizzo & Dillon found a significant larger tendency to take risks when subsistence was at risk than when it was assured, but the farmers were in general not significantly deviating from risk neutrality. Bond and Wonder ( 1980) assessed risk attitude in an expected utility framework without eliciting utility functions. Their major finding is that a significant proportion of individuals, more than many observers might have expected, had negative risk premiums in one or more of the risks. In Hamal and Anderson (1982) study, attitudes toward risk are explored for a sample of rice growers on small farms in Nepal, in the context of the subjective expected utility maximization model. Farmers are found to be generally averse to risk, with diverse levels of absolute risk aversion that tend to diminish as wealth increases, both for individuals and in a cross-sectional sense. Relative risk aversion is argued to be the most comparable measure for contrasts of attitudes toward risk.

77 •!• Experimental methods

This can be regarded as a variant of the previous method, in which real bets are used instead of hypothetical gains and losses. The experimental approach is based on hypothetical questionnaires regarding risky alternatives or risky games with real payments. Early empirical work in this line was conducted in rural India. Binswanger ( 1980) used an experimental gambling approach with real payoffs to estimate the structure of risk preferences of 240 Indian farmers. He found that, at high payoff levcls, virtually ail individuals are moderately risk-averse with Iittle variation according to persona! characteristics. Wealth tends to reduce risk aversion slightly, but its effect is not statistically significant. Belaid and Stanley (1987) used Binswangcr (1980) experiment to elicit farmer utility functions in the eastem high plateau of Algeria. Based on the results, the hypothesis is tested that farrners' risk attitudes are modified by the agroecological zone in which they live, by the crops they grow, and the type of sector (private or socialist) in which they produce. It is concluded that, while the studied farmers are risk averse, no intrinsic difference exists in farrners' attitudes between sectors or sites. Later in 1998, Wik and Holden used the same method to find a wide spread in risk aversion for 143 farrners is Northem Zambia. The experimental measures indicate that on average, more than 80% of the farmers are moderately to extremely risk averse; that they exhibit decreasing absolute risk aversion and increasing partial risk aversion; and, that they are more risk averse in games with gains and losses than in games with gains only. Holt and Laury (2002) found that income has a mildly negative effect on risk aversion. They use their observations to argue that increased incentives appear to change risk attitudes, leading to greater risk aversion. With real laboratory payoffs of several dollars, most subjects are risk averse and few are risk loving. In particular, increase in risk aversion is observed when payoffs are scaled up. More recently in 2007 Harrison, Lau and Rutstrôm carried out an experiment in Denmark using a representative sample of 253 people between 19 and 75 years of age, to find that the average Dane is risk averse, and that risk neutrality is an inappropriate assumption to apply. They also find that risk attitudes vary significantly with respect to several important socio-demographic variables such as age and education. However, they do not find any effect of sex on risk attitudes. Mahmud Y esuf (2007) used a random sample of 262 farm households from seven villages to participate in an experiment in Ethiopia. He observed that a majority of the farm households exhibit intermediate, severe,

78 and extreme risk aversion. The results also indicate increasing partial risk aversion in which individual farm households are more risk averse as the size of the game increases.

•:• Observed economic behaviour

This method is based on the difference between the observed behaviour and that predicted by the empirical models. Furthermore, these models rely on either production theory under uncertainty (econometric models) or cropping pattern selection (mathematical programming). Moscardi and Janvry (1977), Antle (1987 and 1989), Myers (1989), Chavas and Holt (1990 and 1996), Pope and Just (1991), Saha et al. (1994), and Bar-Shira et al. (1997) present good examples of the first category, while for the latter we have Wiens (1976) and Brink and McCarl (1978). The pioneering work by Moscardi and de Janvry ( 1977) used a safety-first rule approach to find that the measurement of behaviour toward risk is explained by a set of socioeconomic and structural variables that characterize peasant households. He found for example that age, years ofschooling of the household head and family size affect the risk attitude of the household. He found that older farmers tend to be less prone to take risks than younger ones. Later, individual risk attitudes have been elicited assuming expected utility maximization. One of the best• known methods is that of Antle (1987), who used a moment-based mode! to estimate risk attitudes with Indian data. The results show that the population is characterized by Arrow• Pratt and downside risk aversion and there is a considerable heterogeneity of risk attitudes in the population. Pope and Just ( 1991) proposed and implemented an econometric test for distinguishing the class of preferences for potato suppl y response in Idaho. The data reject constant absolute and partial relative risk aversion and are congruent with constant relative risk aversion. Chavas and Holt (1996) presented a method for estimating jointly technology and risk preference parameters. The method was applied to U.S. aggregate corn and soybean acreage response decisions, incorporating both price risk and production risk. The results indicated that com-soybean farmers are risk averse and that they exhibit decreasing absolute risk aversion and downside risk aversion. The analysis also provides useful information on the influence of risk on acreage decisions and on the farmers' implicit cost ofprivate risk bearing. Saha et al. (1994) developed an econometric method to permit joint estimation of risk preference structure, degree of risk aversion, and production technology. Evidence rejected the null hypothesis of risk neutrality and

79 suggested that Kansas farmers exhibited decreasing absolute risk aversion and increasing relative risk aversion.

Ali the above approaches have their drawbacks (see Young, 1979; Binswanger, 1980 and Lins et al., 1981 ), which arc most important in the direct estimation mcthod due to interviewer bias, the selection of probabilities and/or insufficient experience on the part of the decision maker in the evaluation of hypothetical situations.

With respect to observed economic behaviour there are also some difficulties, such as the influence of other non-monetary objectives in the decision-making process ( e.g. leisure, management complexity, etc.) and constraints (financial limitations, lack of technical information, etc.) that "contaminate" attitudes to risk. Ifthis method is adopted, therefore, it would not be correct to explain any behaviour that differs from profit maximization purely in terms of risk aversion. In fact, when considering possible resource constraints faced by economic actors, it may appear as if individuals are more risk averse than they truly are (Binswanger, I 982). This is particularly important in developing countries where market imperfections are prominent and consumption and production decisions are non• separable (Wik and Holden, 1998).

Even though these limitations can be reduced, to certain extent, by adopting the experimental method, this has oftcn proved difficult to implement in practice, since the financial cost involved in a real situation with many producers is too high. However, when working with developing countries' poor farmers where the daily wage is low, this disadvantage become obviously less constraining than in developed countries.

To our knowledge, no earlier published studies on risk attitudes of peasants in Côte d'Ivoire exist. Côte d'Ivoire peasants face market imperfections such as rationed credit markets, missing land markets, missing insurance markets, limited access to financial and cornrnodity markets. Therefore, for the purpose ofthis study in Côte d'Ivoire, we found an experirnental approach to be the most appropria te to measure cocoa farrners' risk aversion. As we use an experimental approach, we attempt to avoid the hypothetical bias problem by using real payoffs.

80 3.4.1.1. Design of the experiment

ln our experiment, subjects were confronted with a series of choices among sets of alternative prospects (gambles) involving real money payment. The subject's choice among these alternative prospects is taken as an indication of the degrec of his/her risk aversion.

We follow an experimental design developed by Binswanger (1980) and recently used by Yesuf (2007) to reveal risk preferences of farm households in Western Côte d'Ivoire. A series of schedules of prospects (called games) similar to those shown in Table 3.2 were presented to each subject. Each game Iists six prospects, each with 50% probability of winning. Each subject was asked to select one of the six prospects: 0, A, B, C, D, or E.

Table 3.2: The basic structure of the exeeriment Choices Bad Good Expected Standard CPRA Risk Classification* Outcome Outcome Gain deviation or Coefficient "Heads" "Tails" Spread (S) 0 100 100 100 0 eo to 7.47 Extreme

A 90 180 135 45 7.47 to 2.00 Severe

B 80 240 160 80 2.00 to 0.85 Intermediate

C 60 300 180 120 0.85 to 0.32 Moderate

D 20 380 200 180 0.32 to O Slight to neutral

E 0 400 200 200 0 to -oo Neutra) to preferring

Source: Adapted frorn Binswanger ( 1980) • According to Binswanger (1980) classification.

Once chosen, a coin was tossed and the subject received the left hand amount if the coin showed heads and the right hand amount if the coin showed tails. Every game contains a safe alternative, which is alternative O. The selection of alternative O is equivalent to not playing the game. If the subject selected alternative 0, she received 100 FCFA whether she got a head or a tail. If she chose alternative A instead of 0, her expected gain increased by 35 FCFA, but a bad luck alternative (heads) would now give her 10 FCFA less in return than she would have received with the safe alternative O. In other words, in choosing A instead of 0, the standard deviation in gain is increased from O to 45 FCFA. For the successive alternatives, A to B, B to C, and C to D, the samc is true: the expected

81 gain increases, but so does the spread between the two outcomes. Alternative D and E have the same expected gain, but alternative E has larger spread.

When risk is viewed in terms of uncertainty in gains, income or wealth, as in utility based choice thcories, the alternatives involve more risk the further down you gct in Table 3.2. This means that a decision-maker possessing a utility function concave in wealth, incarne or gain would demand a higher risk premium to accept prospect B rather than its expected outcome than he would demand to accept prospect A rather than its expected outcomc. Whether he prefers prospect B over prospect A depends on the degree of concavity of his utility function. The different prospects are classified from extreme risk aversion (alternative 0) to neutral to preferring (alternative E). The classification is the samc as the one used by Binswanger (1980).

In order to observe a farm household's behaviour following different outcomes, and hence the nature of partial risk aversion, the experiments were made to be conducted at diffe rent levels (see the full format of the games in Table A.3 in Appendix 4). Though the amounts may seem low, it must be recalled that incarnes in the study area are very low, so the amounts listed indeed provide significant incentive for respondents to carefully consider the options and reveal their true preferences. On average, each household won a sum of 900 FCFA, which is about three rimes the rural area daily wage in the country. Each individual played games 1 to 4. Ail gains-only games (game 1 to 3) were derived from the 100 FCF A game by multiplying all amounts by 10 and 50. Furthermore, to test for significant differences in behaviour when faced with the possibility of lasses as opposed to gains-only, a choice set sirnilar to the gains-only games, but involving actual losses to farm households was incorporated into the experiment. The game involving loss (game 4) was derived from game 1 (100 FCFA game) by subtracting all amounts by the certain amount of alternative O. So instead of recciving 80 FCF A for choosing alternative B and throwing a «heads», the subject would have to pay 20 FCFA to the interviewer. By using this gains-and-tosses format, preferences among prospects involving tosses as well as gains can be observed directly. In this way it is possible to test the hypothesis of asset integration. If asset-integration holds, preferences revealed in the gains-only format (opportunity gains and lasses) should not be significantly different from preferences revealed when playing games with real tosses.

82 The three gains-only games were played, at 100, 1000 and 5000 FCF A level. The first two games were real, i.e., the individual actually received the payment. But because ofbudget• restrictions, we also included some hypothetical games. Thus, the third game was set hypothetical. lt is also difficult and hard to defend morally to ask poor peasants to participate in games with real losses and put their own money at risk. To avoid the possibility of major financial losses for households, the gains-and-Iosses experiment (game 4) was done hypothetically and no actual gains were won or Iosses incurred. The individuals were told before playing the games that some of the games would be hypothetical, but they were not told which games. They were paid after playing the four games, told that game 3 and 4 were the hypothetical ones. In this way we hoped the individuals would play ail games as ifthey were real.

The newness of the experiment to the farmers necessitated extra care in explaining the process. lt was vital to the experiment that participants clearly understood that they will receive the amount of money they had won at the end of the game sequences. The six alternatives were carefully explained. Because the majority of the farmers were illiterate, posters with the money stuck on thern for each alternative were presented (See Table A.4 in Appendix 4). Farmers were gambling separately to avoid that the "luck-factor" of a successful player would bias results of successive players. Two hypothetical rounds at the 100 FCFA scale were played to familiarize the subjects with the games and to determine and correct any potential problems before starting the experiment. In general, farmers had no problems in understanding the game.

3.4.1.2. Deriving risk aversion coefficient lt is typically useful to compute a risk aversion coefficient which can serve as a measure of household level of risk aversion from farmers' responses through the expcriment. For this purpose, we employa Constant Partial Risk Aversion (CPRA)18 utility function of the forrn U = (l-S)c<1-s1 following the work of Binswanger (1980), where S the approximate partial is risk aversion coefficient, and c is the certainty equivalent of a risky prospect. If a respondent is indifferent between two consecutive prospects (say I and 2) given that both

18 Note that in our experiment, prospects' size was varied by the factors 10 and 50 whilc wcalth was lcft virtually constant. Thus, partial risk aversion measurc is most suitable. See footnote 11 for more details.

83 prospects have equal probabilities of a good or bad outcome, then we have E(U,) = E(U,)

, and hence (l-S)c,°-51 = (l-S)c,0-s,. Since there is no algebraic solution to this equation, we salve for S using a standard numerical method in excel. So, the partial risk aversion coefficient Sis computcd by solving the equation for indiffcrcnce ( cqual cxpcctcd utility) between two consecutive alternatives, using the CPRA utility function. The upper and lower limits of the CPRA coefficients for each prospect of our experiment are given in

19 Table 3.2 in the previous section • As one can see, the choice of an alternative docs not yield a unique value of the parameter S but rather an interval value. To obtain a unique value for S for each alternative, the geometric mean of endpoints was used". ln fact, the literature hypothesises that individuals get more risk averse when the size of the gamble increases. If this hypothesis holds, individuals have non-linear, risk averse utility functions, which exhibit IPRA. To measure risk aversion coefficients, therefore, one should use an IPRA utility function. Such a function is discussed in Binswanger (1980). However, the function has the disadvantage that its parameters must be estimated from the observed choices that an individual has made at two games levels. Thus, the partial risk-aversion coefficient for any indifference point will then not be unique but will depend on the rate at which partial risk aversion increases, i.e., on the choice paths across the game scale. For each individual one would have to approximate the risk aversion associated with a game at a given level separately, depending on the individual's choices. To simplify matters and obtain a unique risk aversion coefficient for each game level, Binswanger (1980) proposed the use a CPRA function as an approximation. Fortunately, Sillers ( 1980) showed that the values using local risk aversion measures derived from fitting a CPRA function on observed choices offer very close approximations to the corresponding local risk measure values which would be computed directly from an expo• power IPRA function. He argued this is cssentially because the partial risk aversion

19 Details on the computation of the upper and Iower limits of the CPRA coefficients for each prospect are f.resented in Appendix A3 .1. 0 The unique values were as follows: prospect 2, S=3.86; prospect 3, S=l.30; prospect 4, S=0.52; prospect 5, S=O .16. For prospect 5, at one of the endpoints, S=O and the geometric mean of both endpoints would be zero. Therefore, the arithmetic mean was used. ln the case of alternative 6, S=O, but for logarithmic transformations 0.05 was chosen arbitrary. For prospect 1, the upper bound for Sis equal to infinity, while its lower bound is 7.47. Following Binswanger (1981), this value was increased by 10% to give ofS=S.21, taking into account the fact that generally in experimental game, not too many people choose alternative 1. So their risk aversion should no! exceed 7.47 by very much (this result is confirm in our experiment).

84 implicit in individual's choices increases rather slowly with the scale of the experimental prospects.

3.4.2. Modelling risk aversion with persona! characteristics: The Ordered Logit mode!

Bcforc tuming to the analysis of the consequcnces of risk aversion on fanncrs' choicc between risk management strategies, we examine how persona! and farm characteristics are correlated with risk aversion. This is an issue which has not been subject to much empirical analysis in the literature. To analyse the determinants of risk aversion we first need to decide how to treat the dependent variable. Binswanger (1980) based on the lottery selections made by the respondents, computed point estimates of partial risk aversion coefficients S and regressed these (in logarithms) on various persona! characteristics such as gender, occupation, age, income, financial and nonfinancial assets, and schooling. However, at the simples! level, one can choose to define the dependent variable on another way; that is, to give numbers one(!) to six (6) to choice O to E, and use these numbers as regressors. Binswanger (1982) reported little impact ofusing either lnS or the choices 1 to 6 as dependent variable on regression results. Furthennore, our experimental data fits into an ordinal econometric mode!; i.e. an ordered Logit (Probit) model (Maddala, 1983; Wooldridge, 2002). An ordered Logit (Probit) mode! exploit the fact that the dependent variable outcomes, categories of risk aversion, have a natural (ordinal) ranking ranging from 1 (extreme risk aversion) to 6 (risk loving behaviour). This mode! has an advantage in that we need not assume a particular functional fonn of the utility function to analyse the risk aversion behaviour of farm households. Therefore, we simply use the underlying latent variable mode! to analyse the observe choices. Using this mode!, the different hypotheses on risk aversion are tested and factors affecting risk aversion of farm households are analysed.

The ordered Logit (Probit) mode! for the observed variable y (conditional on explanatory variables x) can be derived from a latent variable mode!. Assume there is a latent variable i measuring the degree of risk aversion of the ith decision maker that can be described as:

i =x,P+u, (3.2)

85 for a k x 1 parameter vector p, stochastic disturbance term u; and a vector of regressors x to be specified in the next section. The six outcomes for the observed variable y, are assumed to be related to the latent variable through the following obscrvability critcrion:

Y; = m if am-i ~ i < am for m = 1, ... , 6 (3.3)

for a set of unknown eut points ( or threshold parameters) a0 to a6 to be estimated jointly with the parameter vector p, where: a0 < a1 < a2 < a3 < a4 < as < a6, a0 = -oo and a6 = co .

• 1 => extreme 1 if a0 = -oo ~ y1 < a1 2 => severe if a, ~y; int ermediate ifa2 ~y; mo erate if a3 ~Y,·< a4 5 => slight - to - neutral if a, ~y; neutral - to - prefering • 1i f as ~ Y; < a6 = oo We assume the disturbance term has a logistic distribution yielding the ordered Logit model".

The maximum likelihood (ML) estimation is used to estimate the parameters of the ordered Logit model. Once we have estimated the response probabilities by ordered Logit, we can easily estimate E(y I x) for any value of x, for example, .x. Estimates of the expected values can be compared at different values of the explanatory variables to obtain partial effe cts for discrete x1 •

3.4.2.1. Explanatory variables and expected sign

In this section we discuss why and how some persona} and farm characteristics might be correlated with some of the variation in farmers' risk aversion.

Wealth indicators: From theory and the common assumption of DARA, we would expect wealthier individuals to be less risk averse than poorer households. To capture the effect

21 A standard normal distribution could also be assumed, which would lead to an ordered Probit model. Our results using both approaches are similar and so only the ordered Logit results are presented.

86 of wealth on risk aversion, we include in the mode! two wealth indicators which are value of livestock and farm income. One of the major forms of wealth owning by cocoa producing farm households in Western Côte d'Ivoire is the stock of wealth kept in the form oflivestock. We therefore include the market value oflivestock as a wealth indicator in the mode!. To capture the effect of cash income on risk aversion, we used the farm annual income measured as the mean of annual gross sale throughout the past three years. It has been suggested that asset market imperfections severely constrain substitution between these different forms of wealth (Reardon and Vosti, 1995; Holden et al., 1998). Under such conditions each asset category may have an independent correlation with risk aversion. Nevertheless, on the whole, we expected ail the wealth variables to be negatively correlated with risk aversion.

Househo/d (head) characteristics: We also include a number ofhousehold and household head characteristics. These variables include household head age, education, gender, origin, matrimonial status, percentage share of cocoa revenue in total income and household size. Although Moscardi and deJanvry ( 1977) write « it is generally assumed that older farmers tend to be less prone to take risks than younger ones ... », we find this to be an assumption without any theoretical grounding, and include the variable capturing the age of the decision-maker, without any a priori expectation of the sign. According to previous empirical studies (Binswanger, 1980; Holden et al., 1998), education may influence risk aversion negatively. This makes sense, as education can be seen as human 22 capital, and thus as an indicator of wealth • We therefore expect a negative sign. The gender variable is included taking into account the different roles of men and women in rural area. lt is a durnrny-variable taking the value of 1 for men and O for women. We believe that different attitudes to risk might reflect gender differences in the society. In this study, we expect women to be more risk averse than men. Women have more responsibilities for providing and preparing the food and for feeding and caring for the children. Traditionally, in most societies, men were warriors, and supposed to engage in dangerous and risky activities. In view of the origin variable, we consider three durnmies taking the value of I for native (for migrants from Côte d'Ivoire or for migrants from neighbouring countries) and O otherwise. We expect native to be less risk averse than

22 ln this thcsis, we consider the theory underlying the Education For Ail Development Index which assumes that ail persons with a complete primary school education are literate.

87 migrants, as land ownership could be perceived to be uncertain by migrants and relatively certain by native households. Another variable likely to affect the farmers' attitude towards risk is the matrimonial status. lt is a dummy taking the value I if the head of household is married " and O for unmarried head. We expect married head to be more risk averse compared to an unmarried head who doesn't have any family responsibility.

Household size can have two opposing effects on risk aversion. On the one hand, houschold size can be vicwcd as a wealth variable. A larger family could reprcscnt an increased labour force for the household and thus have a negative effect on risk aversion. On the other hand, a larger family means more people to feed, which may increase risk aversion. Earlier studies are inconclusive. Moscardi and deJanvry (1977) found that increased family size is leading to more cautious and conservative behaviour, while Dillon and Scandizzo ( 1978) found that farmers with larger households were Jess risk averse. This variable is thcrefore included without any a priori cxpectations of the signs. The variable percentage share of cocoa revenue in total incarne is considered here to measure the household level of dependence on cocoa revenue. It is a categorical variable taking the value 1 if cocoa revenue represents Jess than 70% of the total household incarne, 2 if cocoa revenue represents 70 to 90% of the total household incarne and 3 if cocoa revenue represents 90% to 100% of the total household incarne. Taking into account the high volatility in cocoa price which lead to fluctuations in incarne, we expect farm household with high dependence on cocoa revenue to be more risk averse. Therefore, we include this variable with a positive expected sign.

Game variable: Our mode) takes into account one aspect of the game likely to influence the choice made by farmers; that is, the effect of luck factor through successive games. Indeed, to believe that past experience with a random process (such as tossing a coin), would influence a person's next choice, is not common in economic theory. Psychologists, on the other hand, would find it surprising to think that such past experience would not influence future choices. To check whether previous luck had a significant effect on subject's choices, we included a dummy variable defined as LX,,

23 ln the survey, Iwo people are considered as married if they usually reside together as a couple in a relationship in the nature of a civil marriage (with a written contract), a customary/traditional marriage or a religious union.

88 where i is the game number of previous games (for game 4, i would be 1, 2, 3), and X takes the value of I when the person wins (tails), -1 when he loses (heads) and O when neither wins nor !oses (alternative 0). Binswangcr (1980) found the cffcct of previous luck to be negative and highly significant. We expect to find that previous luck does affect people's choices. Subjects who have experienced previous luck will be more willing to take risk than subjects who have experienced previous tosses.

Dummy for location: Finally, it should also be possible to test the correlation between area ofresidence and the measures of attitudes towards risk. It is difficult from the data to say anything definitive about the sign in this relation, we therefore include three dummy for the three zones of our study without a priori sign.

The following Table 3.3 summarises the independent variables with their expected sign .

------c.1.------"c...... , •.• . ,.. -••- •..••. ..,..,".,.,...., .:,ac.u u -.1uM Expected Variables Description Sien •f-.., Wealth c.:, lndicators 0 Farm income Total aonual farm income ofhousehold head (in FCFA) ,..;i - Value of livestock Economie value oflivestock (in FCFA) Q - Household ~ Characteristics ~ Ave Aae of the household head ? § Education Dummv=l ifhcad ofhousehold is literate 0 Gender Dummy=J ifhead ofhousehold is a male - Origin ~ Native Dummy=I ifhead ofhousehold is native ofSoubré ~d Migrantl Dummy=l ifhead ofhousehold is migrant from Côte d'Ivoire + rn Q Mionmt2 Dummv=l ifhead ofhousehold is migrant form another countrv + ~o Matrimonial status ,..;i~ Dummv =! ifhead ofhousehold is married + Househo/d size Total number of people in the household ? ~ Level of Categorical variable taking values I: < 70% dependence, 2: from ~ dependence in 70 to 90% dependence and 3: 100% dependence on cocoa + :> cocoa revenue revenue Game variable ~ Previous /uck ~ Dummy variable defined as LX, , and X takes the value of 1 - Qz when the person wins (tails), -1 when he loses (heads) and O when ~ neither wins nor loses ( alternative 0) ~Il. Location Zone 1 Dummy=l ifhousehold lives in Zone 1 ? ~ Zone2 Dummy=l ifhousehold lives in Zone 2 ? Zone3 Dummy=l ifhousehold lives in Zone 3 ? Source: Author

89 3.4.2.2. Endogeneity of "farm income"

In traditional setting of regression equation right-hand side variables (regressors) determine the left-hand side (dependent) variable. However, it could be true that depcndcnt variable tends to explain one or more explanatory variables. Those explanatory variables which are also explained by the dependent variable are called endogenous variables. In fact some regressors are endogenous because they correlate with the residuals and they corne at the cost of creating bias in estimation.

An important econometric issue we need to consider in this study is potential problem of endogeneity for the variable farm income. This variable is likely to be endogenous because of the possible presence of measurement error in farmers' income. To address this issue, we test for endogeneity of farm income, using the two-step instrumental variable approach suggested by Rivers and Vuong (1988) and Nakamura and Nakamura (1998). As a version of the Hausman test for endogeneity, this method simply consists in the first• stage of estimating the auxiliary equation or reduced form equation (income on all exogenous variables plus instrumental variables) and adding the predicted error of this regression in the original equation or structural equation (augmented regression). If the student statistic shows that the residual variable is insignificant, we reject the hypothesis of endogeneity. Conversely, when this predicted error is significant, it means that we have to deal with endogeneity (see also Wooldridge 2002 for a simple discussion of the procedure and Alvarez and Glasgow (2000) for properties of the Rivers and Vuong (1988) 24 estimator) • In case of endogeneity, the endogenous variable is replaced in the second• stage by its predicted value obtained from the first-stage. The complete description of the procedure used to test for endogeneity offarm income is presented in Appendix 5.

The main assumption of this approach is that we have a good instrument for farm income. A good instrument must satisfy the following conditions: first, it needs to be sufficiently correlated with the endogeneous variable (i.e. it must not be weak); and, second, it can neither have a ( direct) influence on the dependent variable - degree of risk aversion - nor be correlated with the error term in the original equation. A natural instrumental variable to include in this test is the total labour force of the farm household. This instrument is

24 Severa! papers use Rivers and Vuong ( 1988) estimation procedure to test for the presence of endogeneity (See for e.g. McGranahan (2000)) and to control for endogeneity (See Costa (1995), Glewwe and Jacoby ( 1995)).

90 considered to be correlated with income and should fix the problem of correlation, if any, between the original variable and residual of the regression.

3.5. Risk perceptions and economctric analysis of farmers' risk management

decision-making

The literature on the deterrninants of risk management bchaviour has produced relevant, but sometimes puzzling results. For instance, the role of risk aversion in management behaviour appears ambiguous; with some researchers, finding a strong relationship between risk aversion and the use of risk management instruments whilc othcrs do not ( Pennings & Garcia, 2001; Rabin & Thaler, 2001). According to Tomek and Hikaru (2001 ), farmers are assumed to select a combination of strategies that, for example, maximize net expected retums (profits) subject to the degree of risk they are willing to accept. Clearly, optimal risk management strategies in agriculture vary with farm characteristics and the risk environment (Hope and Lingard, 1992). Farmers' risk perceptions, risk attitudes, objectives as well as the available resource base, influence their decisions and actions (Teague et al., 1995). The relationship between farm characteristics, farmers' risk perceptions and their use of risk management strategies is crucial to the design of risk management strategies.

Martin (1996) examined the importance that producers attached to different risk management strategies in New Zealand deregulated farming environment. She demonstrated that there is not necessarily an automatic linkage between specific risk sources and the use of particular risk strategies to alleviate that risk. She found that in New Zealand's deregulated environment, farmers are most concemed about market risk. However, obvions marketing strategies which can reduce this risk, such as forward contracting and the use of futures, were not even considered as viable risk responses by them. More preferred marketing strategies were the use of market information in conjunction with short-term flexibility to enhance prices. With US farmers, Patrick and Musser ( 1997) found that risk perceptions and risk management strategies used by farmers are influenced by farm and farmer characteristics in a study on large-scale farmers from corn-belt states. They concluded that the large-scale US combelt farmers saw liability

91 insurance, financial/credit reserves, debt/ leverage management, and (also) forward contracting as important managerial responses to risk.

Akcaoz and Ozkan (2005) examined risk sources and strategies among farmers in Turkey. Factor analysis has been conducted on information obtained from 112 farmers in 2000. From the findings of the research, risk sources were labelled as environmental, price, catastrophe, input costs, production and technological, political, finance, persona!, marketing, health and social sccurity. The dimensions of risk stratcgics were named as diversification, off-farm incarne, marketing, planning, financing and security. Using also factor analysis, Abebaw (2006) grouped Ethiopian farmers' risk perception into five dimensions. In descending order, these perceived risk dimensions have been identified to be: diseases and pest, labour supply, financial risk, price or market risk and natural events. He found that factors like information, human capital (education, experience), wealth (total land, cattle and small stock) and location (road, district) are statistically significant in explaining the perceived importance of various sources of risk to coffee income. As with risk perceptions, his results have also shown that farmers' have diverse preferences for risk managements strategies in coffee farming. He found six main dimensions of risk management strategies that are: crop diversification, technical aspect within the coffee farming, off-farm employment, forest products extraction, marketing and agro-forestry. Among others, risk perceptions significantly influenced farmers' preferences for many of the risk management strategies.

Ayinde et al (2008) analysed risk attitudes and management strategies of small-scale crop producer in Kwara state, Nigeria. The results showed that farming households in the study area placed different preferences on the risk attitude namely risk taking, risk neutral and risk averse. lt was revealed that the most used risk management strategy employed by the respondents was crop diversification and followed by income diversification and cooperatives. Again in 2008, Pennings et al used a choice bracketing framework to examine the factors that determine the combinations of risk management tools used by US producers. The determinants of producer risk management choices on each bracketing level are evaluated using multinomial logit models. The results show that different strategies are selected on different bracketing levels. Further, when comparing the determinants ofproducer's risk management decisions across bracketing levels it appears

92 that more general characteristics (farm size, age) are important drivers on ail bracketing levels, while more specific characteristics (innovativeness, risk aversion) are significant only on the narrow bracketing level.

More recently in 2009, Velandia et al examined the factors affecting the adoption of crop insurance, forward contracting, and spreading sales as farm risk management strategies using multivariate Probit approach that accounts for simultaneous adoption and/or correlation among the three risk management adoption decisions. They found that some factors significantly affecting the adoption of the risk management tools analyzed are proportion of owned acres, off-farm income, education, age, and level of business risks.

Although the literature suggests a clear link between farmers' risk attitudes, risk perceptions and choices of risk management strategies, similar studies in an Ivoirian context are not available. This Jack of knowledge presents significant challenges to policymakers who may wish to design risk management strategies to facilitate farm operations on the rural area and especially in the cocoa sector. Therefore, this dissertation is set up as an exploratory study.

3.5.1. Behavioural Mode! (Discrete choice models) ln many fields, the choices made by individuals will determine the effectiveness ofpolicy. Understanding what drives people's choices and how these choices may change is critical for developing successful policy. Discrete choice modelling provides an analytical framework with which to analyse and predict how people's choices are influenced by their persona! characteristics and by the different attributes of the alternatives available to them. Discrete choice analysis consists oftwo interrelated tasks: specification of the behavioural mode! and estimation of the parameters of that mode!. This section describes the features that are comrnon to ail discrete choice models. We star! by discussing the choice set, which is the set of options that are available to the decision maker. We then define choice probabilities and discuss different econometric models based on discrete choices.

• The choice set

Discrete choice models describe decision maker's choices among alternatives. The decision makers can be individual people, households, firrns, government agency or any

93 other decision-making unit, and the alternatives might rcpresent completing products, or any other options or items over which choices must be made. Discrete choice problems involve choices between two or more discrete alternatives, such as entering or not entering the labour market, or choosing between modes of transport. Such choices contrast with standard consumption models in which the quantity of each good consumed is assumed to be a continuous variable.

The choicc set is the set of alternatives that arc available to the persan. For a discrete choice model, the choice set must meet three requirements. First, the set of alternatives must be exhaustive, meaning that the set includes all possible alternatives. This requirement implies that the pcrson necessarily does choose an alternative from the set. Second, the alternatives must be mutually exclusive, meaning that choosing one alternative means not choosing any other alternatives. This requirement implies that the persan chooses only one alternative from the set. Third, the set must contain a finite number of alternatives, meaning that there are a countable number of alternatives in the set. This third requirement distinguishes discrete choice analysis from regression analysis in which the dependent variable can (theoretically) take an infini te number of values.

• Deflning Choice Probabilities

A discrete choice model specifies the probability that a person chooses a particular alternative, with the probability expressed as a function of observed variables that relate to the alternatives and the person. Discrete choice models are usually derived under an assumption of utility-maximizing behaviour by the decision maker. Models that can be derived in this way are called random utility models (RUMs). It is important to note, however, that models derived from utility maximization can also be used to represent decision making that does not entai} utility maximization. The models can also be seen as simply describing the relation of explanatory variables to the outcome of a choice, without reference to exactly how the choice is made.

Random utility models are derived as follows: A decision maker, labelled n , faces a choice among J alternatives. The decision maker would obtain a certain level of utility (or profit) from each alternative. The utility that decision maker n obtains from alternative j is Unj, j = 1, ... , J. This utility is known to the decision maker but not, as we

94 see in the following, by the researcher. The decision maker chooses the alternative that provides the greatest utility. The behavioural mode! is therefore: choose alternative i if and only if U,, > U,j, Vj *- i.

Consider now the researcher. The researcher does not observe the decision maker's utility. The researcher observes some attributes of the alternatives as faced by the decision maker, labelled x,j Vj , and some attributes of the decision maker, labelled s., and can specify a function that relates these observed factors to the decision maker's utility. The function is denoted V,1 = V(x,.,s,)Vj and is often called representative utility. Usually, V depends on parameters that are unknown to the researcher and therefore estimated statistically; however, this dependence is suppressed for the moment. Since there are aspects of utility that the researcher does not or cannot observe, V.i * U ,.. Utility is decomposed as

U,. V, + &, , where &,j captures the factors that affect utility but are not included in V,, . = 1 1 1 This decomposition is fully general, since "•i is defined as simply the difference between true utility U,1 and the part of utility that the researcher captures in V.i . Given its definition, the characteristics of "•i , such as its distribution, depend critically on the researcher's specification of V.i. ln particular, c,j is not defined for a choice situation in essence. Rather, it is defined relative to a researcher's representation of that choice situation. This distinction becomes relevant when evaluating the appropriateness of various specific discrete choice models.

&, The researcher does not know 1 vi and therefore treats these terms as random. The joint density of the random vector &, = (c.,, ... ,&,.,) is denoted f(c.) .With this density, the researcher can make probabilistic statements about the decision maker's choice. The probability that decision maker n chooses alternative i is:

P., = Pr ob(U,, > U,j Vj *- i) P., = Pr ob(V,,, + c., > V,. + &,j Vj * i) (3. 4) P., = Prob(c,j -&,, < V,, -V.i Vj *- i).

95 This probability is a cumulative distribution, namely, the probability that each random

-&,, • term c.; is below the observed quantity V., -V,,1 Using the density f(c.), this cumulative probability can be rewritten as:

P,,, = Prob(c.; - c,, < V., -V.1 Vj * i) (3. 5) P,,, = f, l(c,1 -c,, < V., -V,} Vj * i)f(c.)dc,, where I (.) is the indicator function, equalling l when the expression in parentheses is true and O otherwise. This is multidimensional integral over the density of the unobserved portion of utility, f ( c.). Different discrete choice models are obtained from specifications ofthis density, that is, from different assumptions about the distribution of the unobserved portion ofutility.

• Specific Models

As stated earlier, in the context of discrete choice modelling, the most common approach is based on random utility theory (McFadden, 1974). According to this theory, each individual n has a utility function U,1 associated to each of the alternatives}, choosing the one which maximises his (her) utility. This individual function can be divided into a systematic component ~; , which considers the effect of the explanatory variables

(measurable or observable by the modeller attributes), and a random component c,1 that takes into account ail the effects not included in the systematic component of the utility function; for example, the incapacity of the modeller to observe ail the variables that have an influence in the decision, measurement errors, differences between individuals, incorrect perceptions of attributes and the randomness inherent to human nature. Depending on the assumptions made for the distribution of the random error term, diffcrcnt choice models can be derived.

Discrete choice models can first be classified according to the number of available alternatives: binomial choice models which imply two available alternatives and multinomial (multivariate) choice models involving more than two available alternatives. When the dependent variable is binary or discrete taking the value of l or 0, some assumptions on linear regression estimation procedures, such as the OLS where the

96 dependent variable is continuous, do not hold. Alternatives methods have to be used; namely the binary Probit and Logit models. The probit mode! is obtained by assuming that the stochastic terms are distributed normal, and the logit mode! is obtained when the stochastic terms have logistic distribution. The logit and probit models are developed to deal with categorical (dichotomous and polytomous) dependent variables that are common in behavioural or rational choice studies (Aldrich, 1984).

Quanta! response models involving more than two possible outcomes are either multinomial or multivariate. Multinomial models are appropriate when individuals can choose only one outcome from among the set of mutually exclusive, collectively exhaustive alternatives. The analytical approach that is cornmonly used in discrete decision study involving multiple choices is the Multinomial Logit (MNL) which is the generalization of the binary logit mode!. The advantage of using a MNL mode! is its computational simplicity in calculating the choice probabilities that are expressible in analytical form (Tse, 1987). The main limitation of the mode! is the independence of irrelevant alternatives (IIA) property, which states that the ratio of the probabilities of choosing any two alternatives is independent of the attributes of any other alternative in the choice set (Hausman & McFadden, 1984; Tse, 1987). In fact, unobserved factors related to one alternative might be similar to those related to another alternative.

To remedy the IIA problem, McFadden ( 1978) generalized the MNL mode! to the Nested Multinomial Logit (NMNL) mode! which partly overcomes this limitation of the MNL mode!. Indeed, the NMNL avoids IIA by permitting correlations among the random utilities associated with similar alternatives. The NMNL mode!, in a manner analogous to the one-way analysis of variance, partitions a choice set into mutually exclusive subsets (or "nests"), The random utilities of similar alternatives within the same subgroup are correlated, whereas the utilities of alternatives in different subgroups are independent (independence from irrelevant nests - IIN). The mode! posits a hierarchical decision process: alternatives are clustered into nests; a random decision maker first selects a nest and having done so selects an alternative within the nest. The NMLN thus consider the existence of an additional error component, which represents correlation in a group of alternatives.

97 Another way to avoid the BA assumption encountered in the MNL model is the use of Multinomial Probit (MNP) models. MNP models assume that the random errors follow a multivariate normal distribution and are correlated across choices. However, despite its well-known advantages over the popular MNL model (i.e., its relaxation of the restrictive IIA assumption), the MNP model has rarely been used as a model of choice in applied work. The lack of use of MNP stems from the computational burden involved in its estimation. Even though there are from some time ago powerful tools that yield its estimation by simulation (McFadden 1989; Pakes and Pollard 1989) and raise renewed interest in MNP as a model of choice, the MNP model is still been timidly incorporated to practice.

Multivariate choice models are another kind of discrete choice models in which the assumption that choices are mutually exclusive is relaxed. These models allow simultancous choice of the available alternatives by the decision maker. The common forms of multivariate discrete models are the Multivariate Probit (MVP) model and the Multivariate Logit (MVL) model which are generalisation of the standard probit and logit models used to estimate several correlated binary outcomes jointly. For example, if it is believed that the decisions of sending at least one child to public school and that of voting in favour of a school budget are correlated (both decisions are binary), then the multivariate discrete model would be appropriate. Another possible application of multivariate discrete models is in study of farmers' strategies or technology adoption decisions. If fanners can choose to adopt simultaneously multiple strategies, thus taking into account this simultaneity as well as the potential correlation among these adoption decisions should be modelled in a multivariate discrete framework. Until recently, joint estimation of three or more equations with dichotomous dependent variables was computationally infeasible. However, in the last twenty years, and especially in the last decade, several methods for estimating MVP models, MVL models and other types of correlated binary response regression models have been developed.

The MVP model bas received the most attention because it is based on the multivariate normal distribution. In principle, the MVP model is a generalization of the bivariate probit case considered in Greene (2003) by extending to more than two the number of outcomes. The practical obstacle to such an extension is primarily the evaluation of higher-order

98 multivariate normal integrals (Greene, 2003). But as argued in the previous paragraph, recent developments have produced methods of producing quite accurate estimates of multivariate normal integrals. The most popular method applied in the estimation of MVP mode] is the SML that uses the Geweke-Hajivassiliou-Keane (GHK) smooth recursive conditioning simulator (Borsch-Supan et al., 1992; Borsch-Supan and Hajivassiliou, 1993; Keane, 1994). The GHK simulator exploits the fact that a multivariate normal distribution function can be expressed as the product of sequentially conditioned univariate normal distribution functions, which can be easily and accurately evaluated".

3.5.2. Modelling determinants of farmers' risk management decisions: The Multivariate Probit mode!

In modelling the factors affecting the decision to adopt risk reducing strategies, the study has benefitted from the work of Shapiro and Brorsen ( 1988); Knight, Lovell, Rister, and Coble (1989); Makus, Lin, Carlson, and Krebill-Prather (1990); Goodwin and Schroeder (1976). Common to ail these studies is the modelling of adoption of risk management strategies in the context of a subjective expected utility framework. The underlying premise of this framework assumes farm operators seek to maximize their determined subjective utility subject to profit considerations, risk, risk preference, and other factors, let say X; (Shapiro and Brorsen). This assumption leads to the expectation that if variables comprising the X; vector change in a way so as to increase expected returns, then tbese changes will cause an increase in the likelihood of adopting some risk management strategies. Similarly, any changes in the elements of X; which act to decrease the variance of returns will cause the likelibood of using any of these strategies to increase.

"An example of the SML method for the trivariate probit case is given in Cappellari and Jenkins (2003). More• over, the GHK simulator is more efficient in terms of the variance of the estimated probabilities than other simulators (Borsch-Supan and Hajivassiliou, 1993). ln addition, the SML estimator is consistent as the number of draws and the number of observations tends to infinity. Thercfore, simulation bias can be reduced by raising the number of draws with the sample sizc (Cappcllari and Jenkins, 2003).

99 • Application of a multivariate discrete choice mode/

This study attempts to examine the factors that influence cocoa producers' risk management adoption decisions while taking into account the possibility of simultaneous utilisation of multiple risk reducing strategies and the potential correlations among these adoption decisions. ln particular, we examine factors influencing farmers' use of the following risk management strategies in the absence of formai insurance: crop diversification, precautionary savings and being in a social network": Since the adoption of each of the three strategies corresponds to a binary indicator taking the value of one ( 1) if a given strategy is adopted and zero (0) otherwise, we will use a probit mode!.

Moreover, it is possible that these risk reducing strategies are related with each other. First, the same unobserved characteristics of the individuals may affect ail three adoption decisions or plans, making the error terms of the individual models correlated with each other. Secondly, the adoption decision may be more directly related through simultaneity.

27 We therefore mode! the variables as a multivariate probit (MVP) system , a technique that assesses the influence of a set of covariates on the incidence of multiple possible events, where the errors across the equations may be correlated and the strength of these correlations is estimated. The MVP mode! thus has a structure similar to that of a seemingly unrelated regression (SUR) mode!, except that the dependent variables are dichotomous indicators. This simultaneous equations system provides more efficient estimates when contemporaneous correlation is present (Maddala, 1983). Because this mode! is, in principle, simply an extension of the bivariate probit mode! by adding equations, the difficulty of evaluating higher-order multivariate normal integrals, has historically limited its use in more applications (Greene, 2003). However, recent developments have reduced the obstacles to evaluating multivariate normal integrals. The reader is referred to Greene (2003 ), as well as articles by Chib and Greenberg ( 1998) and Bock and Gibbons (1996) for further discussion ofthis subject.

26 These three risk management practices were chosen because they are the risk management strategies most frequently adopted by producers in our sample. Crop diversification concems diversification by other perennial crop as oil palm and hevea. 27 Note that multivariate probit estimation has already been used in a number of studies that evaluate factors that affect adoption of agricultural technologies (see Gillespie, Davis, and Rahelizatovo; Femandez-Comejo, Hendricks, and Misbra, Seo and Mendelsohn (2007); Velandia et al (2009). For example, Gillespie, Davis, and Rahelizatovo use this approach to estimate factors that affect adoption of four breeding technologies in hog production. They argue that modeling adoption decisions using a multivariate probit framework "allows for increased efficiency in estimation in the case ofsimultaneity of adoption".

100 lnstead of a MVP rnodel, it is possible to adopt a much simpler approach and estimate independent probit models for each of the three strategies as functions of farmer/farm characteristics. In such a case, the decision to use a particular risk reducing strategy is hypothesized to be independent across existing risk reducing strategies ( e.g., Schnitkey et al. 1992). In other words, producers' decision of using for example crop diversification as a risk reducing strategy is an independent process from the one that evaluates the use of precautionary saving as a strategy to reduce their exposure to income fluctuation. However, what is more likely is that producers combine various risk reducing strategies to make farm business decisions. Sorne combinations of strategies might be viewed by farmers as being complementary, while other combinations might be viewed as being competing. lt may therefore be important to take into consideration this correlated decision structure when analyzing the factors influencing the adoption of different risk reducing strategies by farmers. lgnoring this correlation in analyzing the simultaneous adoption of risk reducing strategies may lead to biased estimates of the choice probabilities and incorrect estimates of the standard errors of the parameters (Kiefer, 1982).

• The formulation of the Multivariate Probit mode/

This section details the specification of the Multivariate Probit mode! that is used to fit the choice of different risk management strategies by cocoa farmers in Western Côte d'Ivoire. To reiterate, it is not uncommon for a producer to utilize several risk management strategies, rather than just a single strategy, to manage risk. To take into account this simultaneity, which induces endogeneity risks that lead to biased coefficients (Wooldridge, 2002), we estimate a simultaneous equation system in which the three dependent variables are dichotomous; that is a multivariate (trivariate) probit mode! (Cappellari and Jenkins, 2003).

The three risk management strategies considered as dependent variables are y, (practice of some kind of crop diversification activities), Yi (adoption of precautionary saving behaviour) and y3 (participation in a social network) as described in Table 3.4 in the next section. Each of the three dependent variables is a binary indicator that takes a value of

101 one if the farmer has adopted a given strategy, and zero otherwise, Generally, the multivariate probit mode! can be written as:

(3. 6)

where y" (j = 1, ... ,m) represent the risk management alternatives faced by the i'' producer

(i = 1, ... , n), X" is a 1 x k vector of observed variables that affect the risk management adoption decision (see section 3.5.2.3 for presentation and definition of these variables), P; is a k x l vector of unknown parameters (to be estimated), and e" is the unobserved error term. In this specification, each Y; is a binary variable and, thus, Equation (3.6) is actually a system ofm Probit equations (m = 3 in our case).

Let the following hold for the given three latent variables y;, y;, y; (i.e. Y; = l if y~ > O; 0 otherwise), the system of equations to estimate is then given by:

y~ =a, +xp, +e,

Y2 = a2 + X P2 + &2 (3. 7) ly; =a3 +xp3 +&3 with X a vector of variables; the set of explanatory variables included in vector X is identical in the three equations, assuming that the same decision-making process underlies each choice (the set of explanatory variables is presented in the next section).

a1 and P; (with j = 1, ... ,3), the parameters to be estimated, &1, &2, &3, three error terms distributed according to law of multivariate normal distribution, with mean equal to O for each variable and a variance-covariance matrix V, such that variance are normalized to unity (for reasons of parameter identifiability) ". The variance-covariance matrix V is given by: Pn] V =[~21 A3 (3. 8) A, l

28 ln the standard MV Probit model, estimated with cross-sectional data, each of the error variances is normalised to unity - Ibis is required for identification. (Cappellari and Jenkins, 2003 ).

102 Of particular interest are the off-diagonal elements in the covariance matrix, pkJ which represents the unobserved correlation between the stochastic component of the kth and the jth risk reducing strategy. It thus measures how far the unobservcd factors influence use of enterprise diversification, precautionary savings and social networks as risk management strategies. Moreover, because of symmetry in covariances, we necessarily have p , p, . 1 = 1

Only in the case of independent error terms cij ( p is not significantly different from zero) it is possible to deal with the above equation system (3.7) separately as single (independent) probit models (Maddala, 1983). In this study, pairwise correlation of the error terms associated with each risk management adoption decision is computed and its significance is tested to justify the use of the MVP model,

With the assumption of multivariate normality, this system of three simultaneous equations is estimated according to the method of Simulation of Maximum Likelihood• SML (as the estimation implies the calculation of a triple integral with the likelihood function). The Geweke-Hajivassiliou-Keane (GHK) simulator is generally used (Hajivassiliou 2000). It corresponds to the mvprobit procedure of Stata developed by Cappellari and Jenkins (2003). The use of the GHK simulator implies that results depend on a number of random draws used to calculate the simulated likelihood function. Cappellari and Jenkins (2003) reeommend choosing a number of draws equal to at least 29 the square root the sample size . Consequently, the choice of 50 draws allows us to be relatively confident in the estimated parameters (50 > ./36ï'.).

3.5.3. Explanatory variables and expected sign

The producer choice of a particular risk management strategy is explained by the determinants of risk management behaviour. Because we do not have a priori knowledge about whether the determinants that influence risk management behaviour have different influence on different risk management strategies, we bypothesize that they play a similar

29 The mvprobit command implements this mode! using a simulated maximum likelihood (SML) estimator, which, "under standard conditions ... is consistent as the number of observations and the number of draws tend to infinity, and is asymptotically equivalent to the true maximum likelihood estimator as the ratio of the square root of the sample size to the number of draws tends to zero" (Cappellari and Jenkins 2003). The specific simulator that mvprobit uses is the Geweke-Hajivassiliou-Keane (GHK) simulator, which "exploits the fact that a multivariate normal distribution function can be expressed as the product of sequentially conditioned univariate normal distribution functions, which can be easily and accurately evaluated" (Cappellari and Jenkins 2003).

103 role on all risk reducing strategies. We hypothesize that the choice of risk management strategy is influenced by farm characteristics, producer characteristics, risk indicators, access to information, and location dummies. Table 3.4 summarizes the definition of the variables and provides the respective expected sign.

Table 3.4: Definition of Variables in the MVP model and expected sign (n=362) Expected Variables Description Sien

E- 4r.ri. Crop Diversification Dummy variable = 1 if the producer adopt any kind of crop NA z ~ diversification, 0 otherwise. ~ ...:l ~~ Precautionary Dummy variable = 1 if the producer resort to precautionary saving to NA Savings cope with risk, 0 otherwise. ~~ ~> Social Network Dummy variable = 1 if the producer is membcr of a social nctwork, 0 othcrwise. NA Farm Characterlstics Cocoa [arm size Size (in hectares) ofland used by the household for cocoa production. - Land ownership Dummy variable=I if the household owns the land -/+ Farmer/Household Characteristics Number of years the household head has been engaged in cocoa Experience production - Education Dummy=I ifhead ofhousehold is literate ? Access to credit Dummy = 1 if head of household has benefited from any kind of credit -/+ (formai/informai) in the past three years fi.) ~ Household size Total number of people in the household + ...:l Off-farm employment Dummy variable=! if the household is engaged in off-farm activitics - Risk lndicators Perception of risk source Output price risk Dummy= 1 if head of household perceived fluctuation in cocoa price as a + major risk iE- 4 z Pest/disease Dummy=I ifhead ofhousehold perceived pest/disease ofcocoa plants as + ~ a major risk Q Input access Dummy=I ifhead ofhousehold perceived difliculties for inputs access as + z a major risk ~ ~ 111/death Dummy=I if household experienced illness or death of a household + ~ member in the past ten years and consider this as major shock Q Risk aversion z High risk aversion Dummy=I ifhead ofhousehold fell into extreme to severe risk + - categories Medium risk aversion Dummy=I ifhead ofhousehold fell into intermediate to moderate risk + categories Low risk aversion Dummy=l ifhead ofhousehold fell into neutral to preferring risk - categories Information Access to information Dummy=I ifbousehold own a radio/TV -

Location Zone I Dummy=I ifhousehold lives in Zone 1 ? Zone2 Dummy=I ifhousehold lives in Zone 2 ? Zone3 Dummy=I ifbousebold lives in Zone 3 ? Note: NA denotes "not applicable", + dcnotc a positive efîect on the dependent variables, - denote a négative c!Tcet on the dcpcndcnt variables,? denotes no sign a priori.

104 As mentioned above, the three risk management adoption decisions that serve as the dependent variable for the multivariate probit estimation procedure are crop diversification, precautionary saving, and social networks. With respect to farm characteristics as independent variables, previous studies identified farm size and land ownership as farm characteristics relevant for risk management decisions. Farm size is hypothesized to have a negative effect on the use of risk management tools. Larger farm size is related to a larger asset base from which to draw resources. Consequently, a higher farm size signais a larger capacity for bearing risk and a lesser need for risk management instruments (Velandia et al, 2009). Whether the farmer owns the farm or is actually in a sharecropping system may have different effects on the farmer's adoption decision. The relationship between land ownership and adoption of different risk management strategies appears ambiguous (and depends on the particular strategy).

The producer characteristics considered here are experience, education, access to credit, household size and ojf-farm employment. Farmers with low experience generally lack some farming skills. The lack of skill then increases their fear of the risks inherent to agricultural production and consequently their needs to resort to risk management strategies. We therefore hypothesize a negative relationship between experience and the farmers' adoption decision. Furthermore, it is typically hypothesized that in general, producers with more education tend to adopt more sophisticated risk management tools. Because in this study the risk management tools considered are the traditional ones, we include this variable without a priori sign. Access to credit is a wealth indicator which tends to improve farmer ability to bear risk. But increase in wealth can also favour the use of risk reducing strategy like crop diversification. A big household size is hypothesized to favour adoption ofrisk management strategies. Off-farm employment represents a form of diversification that would have an impact on producers risk management behaviour. The existence of off-farm incomes may indicate a greater capacity to bear risks and may reduce incentives to adopt risk management tools.

Risk variables incorporated in this study comprise of farmers' perception of risk source and risk aversion. On one hand, the perception of risk source ( output price risk, pest/disease risk, input access and ill-health/death risk) is posited to influence the choice of risk management strategies. In general, there is a positive relationship between risk

105 perception and producers' use of risk management strategies (Pennings and Leuthold, 2000). On the other hand, findings reveal that risk aversion is an important driving force for adoption of risk management strategies by farmers (Pennings and Leuthold, 2000; Moschini and Hennessy, 2001; Mishra and El-Osta, 2002; etc). In this study, farmers' risk attitudes are measured through an independent experiment (as described in section 3.5.1 ). From the results of the experiment, it was possible to identify six distinct risk aversion categories among the respondents, according to their Iotteries choices. To avoid having too many parameters in this variable, and in the mode! above (MVP mode!), we group the six risk aversion categories into three risk aversion dummy variables: "high risk aversion", which will be equal to one if the respondent chose lottery I or 2; "medium risk aversion", equal to one if the respondent chose lottery 3 or 4; and "low risk aversion", equal to one if the respondent chose lottery 5 or 6. Producers with higher risk aversion may have more incentives to adopt risk management tools. We therefore hypothesize a positive relationship between high risk aversion and adoption of risk management strategies and a negative relationship between low risk aversion and the adoption decision.

Finally we includc farmers' access to information captured by radio and/or TV owncrship. Access to information may lower farmers' risk aversion and therefore negatively affect the adoption of risk management strategies. Three location dummies represented by the three zones considered in the study arca are also incorporate to take into account the geographical heterogeneity of the sample.

3.6. Conclusion

This chapter has outlined the economic methodology that is used to determine cocoa farmers' risk aversion as well as the persona! determinants of risk aversion in an experimental framework. The chapter has also modelled the adoption of risk management strategies by farmers using a simultaneous equation approach. The next chapter will provide the experimental and econometric results.

106 CHAPTER FOUR: DETERMINANTS OF FARMERS' RISK ATTITUDES AND RISK MANAGEMENT STRATEGIES

4.1. Introduction

This chapter presents some descriptive statistics and the estimation results of the analysis based on the cmpirical modcls developed in chaptcr three. The Ordered Logit mode! and the Multivariate Probit estimation procedures are employed to model the nature of farmers' behavioural responses to risks. Policy implications and simulations derived from these results are discussed. STATA (version 10.0) is the cconometric package uscd to generate the coefficients of the models.

4.2. Household socio-demographic and economic characteristics

4.2.1. Socio-demographic attributes

In total, 362 farmers were interviewed. Sample characteristics are summarised in Table 4.1. The larges! group of the interviewed people (44.75%) is between 30 and 45 years old, followed by the 46 to 59 years class (33.98%). Only 7.46% of the interviewees are younger than 30, while 13.81% were older than 60. The average age of a household head is 45 years. The Majority of houschold heads (86.74%) arc married. In general, the respondents (typically household "heads") are men, who have no formai education. lndeed, more than 57% of the sampled household heads are illiterate. Approximately 26% and 17% have primary and secondary school education level respectively. No one has higher school education in our sample. Religion is dominated by Christians, with almost half of the sample (45.86%), followed by Muslim (27.90%). More than 53% of the respondents are lvorian migrants from other areas in the country. The natives represent only 21.55% white migrants from neighbouring countries like Burkina Faso and Mali account for 25.14% of the total respondents.

107 Table 4.1: Houschold socio-demographic characteristics (n=362) Variable Percentage (%) Age composition (%) Age (less than 30 years) 7.46 Age (from 30-45 years) 44.75 Age (from 46-59 years) 33.98 Age ( + 60 years) 13.81 Marital Status (%) Married 86.74 Unmarried 13.26 Gender(%) Male 96.13 Female 3.87 Education(%) No education 57.46 Primary school education 25.69 Secondary school education 16.85 Religion(%) Christian 45.86 Moslem 27.90 Other (traditional, no religion, other) 25.41 Origin (%) Native 21.55 Migrants from Côte d'Ivoire 53.31 Migrants from neighbouring countries 25.14 Household size (mean)* 12.61 Number offarms (mean)* 1.44 Farm size (%) Fann size [0.5 to 5 Ha] 52.49 Farm size ]5 to 10 Ha] 29.28 Farm size ]10 to 15 Ha] 9.67 Fann size] +15 Ha [ 8.56 Farm experience (mean)* 22.07 Land ownership system(%) lnheritance 27.90 Purchase 53.04 Gift 16.85 Sharecropping 2.21 Location(%) Zone 1 36.46 Zone2 32.87 Zone 3 30.66 Information (yes=l, %)** 77.90 Source: Own computation from survey data Notes: "For these variables, the value reported here is the mean of the distribution. • "This variable captures the farmer 's access to any source of information (radio/Tl/).

Households tend to be big, with a mean of about 12 people. This is the general tendency in rural area and especially for cocoa producing households where family labour is very important. Average number of farms per household is 1.44 and highest number of farmers bas small size of farms. Farms up to 5 hectares belong to 52.49% of the fanners, 29.28%

108 of farmers have farm size comprised between 5 and 10 hectares. Only 9.67% and 8.56% of the farmers own farm size of 10 to 15 hectares and more than 15 hectares respectively. Experience in farming is high with a mean of 22 years. The majority of the farmers (53.04%) purchased their land from natives, 27.90% inherited the land, and 16.85% received the land as a gift while a minority (2.21 %) exploit the land as tenant and not landowner. The sample size is approximately the same in each of the three zones interviewed with the highest proportion of interviewees in zone 1 (36.46%), followed by zone 2 (32.87%) and zone 3 (30.66%). Information is accessible for 77.90% of the farmers.

4.2.2. Wealth composition and distribution

Heterogeneity of income source across households is important as a socio-economic variable to explain consumption behaviour (Chem et al., 2003). Household head income also serves as an indicator of household poverty status. This section makes a descriptive statistics of the research sample, considering the different sources of income (wealth) available to farmers and taking into account the geographical heterogeneity of the sample. The distribution offarm annual income within the three zones is presented in table 4.2.

Table 4.2: Distribution and mean of farm income per zone Annual farm income* Zones (in FCFA) Zone 1 (n=I32) Zone 2 (n=119) Zone 3 (n=l 11) <=500°000 40.91 29.41 24.32 1500°000; 1 °000°0001 23.48 23.53 28.83 l 1 °000°000;2°000°0001 25.00 30.25 28.83 >2°000°000 10.61 16.81 18.02 Mean of income** 857°971 962°868 1°899°017 Source: Own computation from survey data Notes: *Farm income per year. •• Mean of total farm income per year.

The table provides a mix of reports with respect to each class of farm income. Indeed, at lower farm income level, that is annual farm income Jess or equal to 500°000 FCF A, we find 40.91% of farmers from zone 1 while 29.41% and 24.32% of households in zone 2 and zone 3 respectively fall into this category of income. We also find zone 3 to contain the highest percentage offarmers (28.83%) falling into the second class (between 500°000

FCFA and I 0000°000 FCFA). The third class is dominated by zone 2 with 30.25% of the

109 farmers having an annual farm income between 1 °000°000 FCF A and 2°000°000 FCF A. At the highest level of income, zone 3 cornes in first with 18.02% of farmers while zone I encloses only I 0.6 I % of farmers. As a whole, zone 3 appears to be the area with highest annual farm income. The mean of annual income is about )0899°017 FCFA in zone 3, followed by zone 2 with a mean of 962°868 FCF A and finally we have zone 1 with 857°971 FCFA.

Another wealth indicator considered in this thesis is the market value of livestock owned by the farmer. In general, the value of animais owned by farmers in our sample is low with a mean of252°665 FCFA considering the entire sample. Table 4.3 gives the statistics per zone. In opposition to the rcsults in the previous section, zone 1 has the highcst mcan of the value of livestock owned by farmers, which is 309°015 FCFA. Zone 3 has the lowest mean value oflivestock (189°324 FCFA) with only 6.31% offarmers falling in the highest class. Farmers in zone 1 and zone 2 invest more in livestock as compared to farmers in zone 3.

Table 4.3: Distribution and mean oflivestock value Eer zone Market value of Zones Iivestock (inFCFA) Zone 1 (n= 132) Zone 2 (n=l 19) Zone 3 (n=ll l) <=100°000 75.00 76.47 76.58 ) l 00°000;500°000) 10.61 10.08 17.12 >500°000 14.39 13.45 6.31 Mean of livestock 309°015 249°243 1s9°324 market value Source: Own computation from survey data

The level of dependence of a household on cocoa revenue is weighed up by considering the share of cocoa revenue in the total income, see table 4.4. The first class refers to household with cocoa revenue representing Jess or equal to 70% of their total income. Farmers in this class corne in majority from zone 1 with 16.67% and zone 2 with 15.97%. Only 5 .41 % of farmers from zone 3 fall into this class.

110 Table 4.4: Percentage share of cocoa revenue in total income (per zone) Share of cocoa in Zones total incomc (%) Zone 1 (n=I32) Zone 2 (n=l 19) Zone 3 (n=l 11) Total (n=362) <= 70 16.67 15.97 5.41 12.98 ]80;90] 53.03 34.45 26.13 38.67 ]90;100] 30.30 49.58 68.47 48.34 Source: Own computation from survey data

The second class characterizes farmers with cocoa revenue representing 80% to 90% of their total income. This second class is also dominated by farmers from zone 1 and zone 2. However, when it cornes of the last class which represents farmers with total income constituted of 90% to 100% of cocoa revenue, we find that zone 3 has the highest percentage of farmers falling in this category. This result is in Iine with findings from table 4.1 where zone 3 has the highest annual farm income. Considering the entire sample, we find that almost half of the sample households rel y exclusively on revenue from cocoa production.

Others sources of income analysed are farmer's enrolment in off-farm activities and farmers access to informai credit (as formai credit markets are not functioning in rural area). The statistics are shown in table 4.5. Incomes are also not very diversified. For instance, in the sample, only 8.84% of households participate in off-farm activities. In such an environment, consumption and production decisions are made jointly by households. This implies that endowments, such as wealth and family size and household characteristics, will affect production outcomes (Jacoby, 1993).

Table 4.5: Percentage of household with other sources of income (per zone) Others sources of Zones income (%) Zone 1 (n=132) Zone 2 (n=l 19) Zone 3 (n=I 11) Total (n=362) Off-farrn activities 3.79 1.68 22.52 8.84 Informai credit 30.30 68.91 44.14 47.24 Source: Own computation from survey data

Within the zones, zone 3 is the area with the highest proportion of households participating in off-farm activities (22, 52%) compared to households in zone 2 where only 1.68% of the farmers are engaged in off-farm activities. Instead, farmers from zone 2

111 resort more on credit as other source of income. 68.91 % of the farmers in zone 2 have access to informai credit market while in zone 1 and zone 3 we have 30.30% and 44.14% respectively. In the whole sample, 47.24% have had access to informai credit. This result shows the importance of informai credit market in rural area.

4.3. Farmers' risk preferences

4.3.1. Distribution of preferences to risk over games

The risk-aversion distributions corresponding to different game levels are given in Table 4.6. The first panel shows the distribution for the games with gains only, while the second panel shows the distribution for the game with gains and tosses. Observe that at low level of the game, the distribution was rather evenly spread over all classes of risk aversion, but as the game level rose, the distribution shifted to the left and became more peaked, i.e., risk aversion increased. Even at the lowest level of the game, which is the 100 FCF A game, we have 17 .96% of farmers chose the alternatives representing severe to extreme degree of risk aversiorr'". When the game level increased to 1000 FCFA, 32.88% chose these two most risk averse alternatives and at the 5000 FCF A game level, we found more than 45% falling into these risk aversion classes.

Table 4.6: Percentage distribution of risk aversion in different types of games and different game levels* Extreme Severe lntermediate Moderate Slight to Neutral to (%) (%) (%) (%) Neutral Preferring Games {%} {%} (/) Games with on/y gains 100 FCF A. (game 1, real) 8.84 9.12 28.73 22.93 17.68 12.71

1000 FCF A. (game 2, real) 14.92 17.96 33.15 20.72 6.91 6.35

5000 FCF A. (game 3, hyp.) 16.85 29.28 30.94 14.92 4.42 3.59

(Il) Game with gains and fosses 100 FCFA. (game 4, hyp.) 19.34 34.53 27.07 14.36 2.76 1.93

Source: Own computation from experimental data Note: • Percentage shares are calculated for each game level, where 100 FCF A is the lowest and 5000 FCF A is the highest game level. A total of 362 households participated in ail the games.

30 At the lOOF game level, we have "Extreme (8.84%)" + "Sevère (9.12%)"= 17.96% offarmers falling into the scvcre to cxtremc risk aversion classes.

112 Considering the slight-to-neutral and neutral-to-preferring alternatives, the percentage choosing these alternatives was reduced from 30.39% at the 100 FCFA game level to only 13.26% and 8.01% in the 1000 FCFA level and 5000 FCFA level respectively. The share of responses falling into the interrnediate and moderate risk aversion categories remain stable between games 1 to 2 (51.06% and 53.87%), but decline to 45.86% in game 3 due to increases in the severe and extreme risk aversion categories. These results seem to indicate increasing partial risk aversion in which individual farrn households are more risk averse as the size of the game increases.

The second panel shows the distribution for the game with gains and losses. In this game, we see an inclination of people being more risk averse than in the othcr games with gains only. Comparing this game to the 100 FCFA game with gain only, we fund a sharp increase in the severe to extreme risk aversion categories from 17.96% in the gain only game to 53.87% in the game with gains and losses. The neutral to preferring class tends to disappear in the game with gainS and losses with only 4.69% of farrners falling into this class. This result could be explained by people being more risk averse when they have to put their money at risk.

Similar games have been played with peasant farrners in other areas in the developing world since the first field experiments by Binswanger (1980) with Indian farrners. Table 4.7 compares results from applying Binswanger's general experimental methods in farrning communities in the Phillipines (Sillers 1980), Zambia (Wik and Holden, 1998) and Ethiopia (Y esuf, 2007).

To facilitate comparisons among the experiments and to give a sense of the experimental pay-offs in terms of local incomes, ail pay-offs were expressed in experiment-specific «daily wage» (DW) units, equal to the daily wage received by an unskilled agricultural labourer in the study area in question. These results are shown together with our results from Côte d'Ivoire.".

31 Average daily wage rate in rural area in Côte d'Ivoire was 333 FCFA per day (Perspectives économiques en Afrique, BAD/OCDE, 2007).

113 Table 4.7: Pcrcentagc distribution of revealed risk ereferences in five cxeerim cntal studies. Extrcm c to Intcrm cdiatc to Risk-neutral to Num bcr Severc risk M oderate risk Risk-preferring of Games aversion aversion reseonses

Games with on/y gains lndia (Binswanger, I 980) 50 Rupee (14.3 DW) 8.4 82.2 9.4 (107) 500 Rupee" (143 DW) 16.5 82.6 0.9 (115)

Philippines (Sil/ers, 1980) 50 Peso (7 .1 DW) 10.2 73.5 16.3 (49) 500 Peso (71 DW) 8.1 77.6 14.3 (49)

Zambia (Wik and Ho/den, 1998) 1000 Kw (1.8 DW) 29.l 46.4 24.5 (423) 10000 Kw"(l8 DW) 36.7 52.5 11.0 (137)

Ethiopia (Yesuf, 2007) 5 Birr (0.5 DW) 45.4 33.6 21 (262) 15 Birr" (1.5 DW) 55.7 27.5 16.8 (262)

Côte d'Ivoire 1000 FCFA (3.0 DW) 32.88 53.87 13.26 (362) 5000 FCFA" (15.0 DW) 46.13 45.86 8.01 (362)

Games with gains and tosses Philippines (Sil/ers, I 980) 50 Peso (7 .1 DW) 10.2 79.6 10.2 (49)

Zambia (Wik and Ho/den, 1998) 5000 Kw (9 DW) 45.3 42.4 12.4 (137)

Ethiopia (Yesuf, 2007) 5 Birr (0.5 DW) 21.1 44.8 34.1 (76)

Côte d'Ivoire 100 FCFA (0.30 DW) 53.87 41.43 4.69 (362)

Source: Author adaptation from Wik and Holden, 1998 H: Hypothetical game

Sillers and Binswanger (1983) suggest that most villagers in farming cornrnunities hold rather similar pure preferences toward risk. They ask for more research to confirrn this general pattern, but propose that a universal "moderate" risk aversion in such communities would appear to be a reasonable hypothesis on which to base our thinking about the agricultural investment and production behaviour of peasants. Our results from Côte d'Ivoire may indicate that the picture is probably different. Farmers in our sample

114 showed a much larger variation in risk aversion. A bigger incidence of farmers chose the alternatives representing extreme or severe risk aversion than in the Asian studies, and at the same time more farmers (with the possible exception of the Philippines experiment) were choosing the risk neutral to risk preferring alternatives,. These results are more in line with studies of African farmers' risk attitude like Wik and Holden's (1998) results from Zambia and Yesuf (2007) results from Ethiopia, than the general pattern of moderate risk aversion found in studies in Asia. Though we cannot dismiss the possibility of the big spread in our data being a result of subjects choosing alternatives randomly, these results could mean that risk preferences are indeed different from area to area, and that it is not possible to draw universal conclusions about peasant's risk preferences. ln the next section, we formally test for the homogeneity of various risk attitude distributions in our sample using some statistical tests.

Comparing our results to the Ethiopian and Zambian experiments, we found the proportion of farmers in the extreme to severe risk category to be higher in the Ethiopian experiment, but lower in the Zambian case. These results suggest that farm households in Côte d'Ivoire are Jess risk averse than in Ethiopia but are much more risk averse than in Zambia (and in Asia).

4.3.2. Homogeneity in risk distribution

This section addresses some important questions often raised in risk aversion studies. The first question is related to the "universality" of conclusions about farmers' risk attitude. Do the peasants, regardless of site, exhibit similar risk attitudes for a same payoff levels? To answer this question, we test for the homogeneity of various risk attitudes distributions using Pearson's chi-square test also known as chi-square test for independence. As can be seen from the chi-square tests (bottom of table 4.8), the null hypothesis that the distributions are independent of the site is rejected in almost ail the cases.

At the same level of payoff, the distributions of risk preferences are not equivalent from an area to another. The results are statistically significant and evident at the three levels of game. Only for the case of the gain and lasses game, comparison of risk distribution between zone I and 3 (l' vs. 3 ') revealed that farmers exhibit similar risk attitudes. However, this result is not confirmed when comparing the risk distribution by taking into

115 account ail the three zones ( 1' vs. 2' vs. 3 '). Hence, drawing universal conclusions about farmers' risk aversion could be misleading. Risk aversion may vary from a country to another, from a city to another, from a village to another, from an individual to another. It would be then important to analyse the determinants of peasants risk aversion; this is done later in this thesis.

The second question refers to farmers' behaviour at different levels of games. From table 4.6, we found a general tendency of farmers being more risk averse when the level of the game rose. This could mean that the risk distribution is different from a game level to another. The results of the Pearson's chi-square test in Table 4.8 comparing the risk distribution in the same area but at different payoff level ( 1 vs. 4 vs. 7; 2 vs. 5 vs. 8; 3 vs. 6 vs. 9) are highly significant and reveal non similarity in farmers' risk attitude from a game level to another.

116 Table 4.8: Tests of the risk distributions in the three game levels, the Iwo types of game and the three zones. Identification No. of Zone Payoff Scale 0 A B C D E No. Observations

1 Zone 1 IOOFCFA 12 li 54 24 17 14 132 2 Zone 2 100 FCFA 7 10 29 34 18 21 119 3 Zone 3 IOOFCFA 13 12 21 25 29 li Ill

4 Zone 1 IOOOFCFA 13 16 64 25 9 5 132 5 Zone2 IOOOFCFA 12 30 34 24 8 li 119 6 Zone 3 lOOOFCFA 29 19 22 26 8 7 111

7 Zone 1 5000FCFA 22 31 54 18 3 4 132 8 Zone2 5000FCFA 9 47 31 21 6 5 119 9 Zone 3 5000 FCFA 30 28 27 15 7 4 111

l' Zone 1 IOOFCFA 19 40 46 23 3 1 132 2' Zone2 IOOFCFA 30 47 20 18 3 1 119 3' Zone3 IOOFCFA 21 38 32 Il 4 5 111

Distributions tested**

Test Degrees Test Degrees Identification of Identification of No. x' Freedom P-Value No. x' Freedom P-Value

1 vs. 2 vs. 3 27.073 10 0.003 I' vs. 2' vs. 3' 21.383 10 0.019 1 vs. 2 11.403 5 0.044 I' vs. 2' 13.247 5 0.021 1 vs. 3 16.422 5 0.006 I' vs. 3' 7.953 5 0.159 2 vs. 3 10.068 5 0.073 2' vs. 3' 9.5429 5 0.089

4 vs. 5 vs. 6 39.151 lO 0.000 lvs.4vs.7 35.645 lO 0.000 4vs. 5 15.181 5 0.010 2vs.5vs.8 47.280 10 0.000 4vs. 6 25.652 5 0.000 3vs.6vs.9 42.784 10 0.000 5 vs. 6 12.795 5 0.025

7 vs. 8 vs. 9 29.524 10 0.001 1 vs. I' 39.798 5 0.000 7 vs. 8 15.667 5 0.008 2 vs. 2' 73.787 5 0.000 7 vs. 9 10.519 5 0.062 3 vs. 3' 44.319 5 0.000 8 vs. 9 17.327 5 0.004

Source: Author from experimental data Note: 'identification numbers 1 ', 2' and 3' represent the distribution for the 100 FCFA game with gain and Iosses in zone 1, zone 2 and zone 3 respectively. ++The chi-square statistics are calculated based on the distribution of risk preferences given in the first panel of the table. The degrecs of freedom are calculated as df = (r-1 )*(c-1 ), where r is the number of categories (6 in our case) and c is the number of columns to be compared (3 or 2 in our case).

Another question considered in this thesis is about the validity of the asset integration hypothesis in our sample. The results of the Pearson' s chi-square tests (1 vs. 1 '; 2 vs. 2 '; 3 vs. 3 ') reveal that the null hypothesis that the subjects' risk preferences are equivalent in both kinds of games is rejected for each zone of the study area. There is therefore a significant difference in risk preference between gains-only and gains and-losses games.

-- 117 People tend to be more risk averse when it cornes to losses than to gains (see Figure 4.1 for illustration). This strongly suggests the absence of asset integration and the presence of loss aversion by our farm households. The implication of this finding for policies such as agricultural extension is that farm households can be expected to be more responsive to the possibility of agricultural losses than stochastic output gains. Providing some type of insurance or other support to farmers who try new, but risky technologies is therefore suggested by our findings.

Figure 4.1: Comparlson of risk distribution between the 100 FCFA galns-only game and the 100 FCF A gains and losses game in the three zones.

100%

C 90% C, ~ 80% &=l Preferring 70% i Neutra! Q 60% Moderate •t>•I ) 50% !! lntermediate C 40% .•..• •Severe •..• 30% Il. •Extreme 20% 10% 0% 2 3 l' 2' 3'

Source: Author from experimental data 1 refers to the distribution for the 100 FCFA game with gain-on/y in zone 1. 1 ' refers to the distribution for the 100 FCFA game with gain and lasses in zone 1. 2 refers to the distribution for the 100 FCFA game with gain-on/y in zone 2. 2' refers to the distribution for the 100 FCFA game with gain and lasses in zone 2. 3 refers to the distribution for the 100 FCFA game wilh gain-on/y in zone 3. 3 ' refers to the distribution for the 100 FCFA game with gain and lasses in zone 3.

4.3.3. Nature of partial risk aversion

In studying the behaviour of an individual, the partial risk aversion parameter becomes relevant when initial wealth is maintained constant whereas the scale of the gamble is varied by a scalar k. An increasing value of S relates to the decrease in the willingness to accept a gamble as the scale of the gamble increases. In the course of our experiment, payoffscale was varied by a factor of 10 (rcal game) and 50 (hypothetical game), whereas mean wealth was left virtually constant. Past studies in developing countries (Binswanger,

118 1980; Sillers, 1980) indicate that, often, subjects' choices tend to shift to the more risk averse alternatives as payoff scale rises, thus implying an increasing partial risk aversion. This shift is fully supportcd by the results of this study (table 4.9).

Table 4.9: Effect of pa~off Scale on Partial Risk Aversion + Mean ofS Mean ofS Dcgrccs of Game Scale Small Scalc Large Scalc Frecdom

Zone 1 100 FCF A vs. 1000 FCF A 1.72 2.01 -1.056 262 100 FCF A vs. 5000 FCF A 1.72 2.88 -3.792 262 1000 FCF A vs. 5000 FCF A 2.01 2.88 -2.811 262

Zone2 100 FCFA vs. 1000 FCFA 1.30 2.29 -3.417 236 100 FCF A vs. 5000 FCF A 1.30 2.57 -4.691 236 l 000 FCF A vs. 5000 FCF A 2.29 2.57 -0.977 236

Zone3 100 FCFA vs. 1000 FCFA 1.79 3.20 -3.588 220 100 FCF A vs. 5000 FCF A 1.79 3.59 -4.693 220 1000 FCF A vs. 5000 FCF A 3.20 3.59 -0.920 220

Ali (n=362) 100 FCFA vs. 1000 FCFA 1.60 2.47 -4.645 722 100 FCF A vs. 5000 FCF A 1.60 3.00 -7.490 722 1000 FCF A vs. 5000 FCF A 2.47 3.00 -2.651 722

Source: Author from experimental data. Nole: + The intervals for the classification of the Partial Risk Aversion Coefficient (y) are: l =Extreme (1-«- :> to 7.47), 2= Severe (y=2.00 to 7.47), 3=intermediate (y=0.85 to 2.00), 4=moderate (y=0.32 to 0.85), S=slight to neutral (y=O to0.32), 6=neutral to loving(O to - oo).

ln each zone, we found an increase in the mean of the partial risk aversion coefficient (S) when the game level rises. This result confirms the presence of increasing partial risk aversion and is consistent with Binswanger ( 1980) and Y esuf (2007). The observed shift in S raises the question, "Would an increasing partial risk aversion (IPRA) utility function be more appropriate in representing individuals' behaviour?" To answer the question, risk aversion coefficients derived through both the IPRA and the CPRA functions ( at all payoff scales) could be compared". However, the CPRA function as used in this study is simply a local approximation of partial risk aversion for individuals who are indifferent between two consecutive prospects. No claim is made that the partial risk aversion is globally constant.

32 This is out of the scope ofthis thesis.

119 The results of the t-test shown in table 4.9 indicate significant difference (at 1 % level) between the mean of constant partial risk aversion coefficients at different game level and within area in almost ail the cases. Only in zone 2 and zone 3, we failed to reject the null hypothesis of equal mean of CPRA coefficients between the 1000 FCF A and the 5000 FCFA games. Also in zone 1, the null hypothesis of equal mean cannot be rejected between the 100 FCFA and the 1000 FCFA games. However, considering the whole sample, the results significantly show difference in the mean of S between games. The results confirm increasing partial risk aversion in peasants' behaviour with a mean'? of 2.35. This implics that the mean former in our samplc falls into the severe risk aversion category.

4.3.4. Risk aversion and socio demographic characteristics

This section examines variations in risk attitude across major socio demographic factors like gcnder, education, age and incarne with refercnce to the 1000 FCFA game.

• Risk Aversion and Gender

Researchers have only recently explored the issue of differences in risk aversion by gender. Anecdotal evidence suggests that women are more risk averse than men. A number of studies have confinned this finding34 even when controlling for the effects of other individual characteristics such as age, education, and wealth. Table 4.10 gives the distribution ofrisk aversion by gender in our sample.

Table 4.10: Risk Aversion and Gender Female Male Risk Aversion Class % %

Extreme 21.43 14.66 Severe 57.14 16.38 Intermediate 21.43 33.32 Moderate 0 21.55 Slight to Neutra/ 0 7.18 Neutra/ to preferring 0 6.61 Source: Author from experimental data.

33 This is the mean of the CPRA coefficient throughout the three game levels and considering the whole sample; that is (J.6o+2.47+3.00)/3. 34 Sec literature review in Chapter three.

120 The statistics seem to indicate an inclination of women to be more risk averse than men given that about 80% of the women fall into the extreme to severe risk aversion categories compared to only 31.04% of men. Moreover, there is no woman in the moderate to preferring risk aversion classes whereas more than 35% of the men fall into these categories.

• Risk Aversion and Education

The risk attitude distribution by education is presented in Table 4.11. The table shows a tendency of farrners with no education to be more risk averse than educated farrners. Indeed, more than 40% of non-educated chose the alternatives representing severe to extreme degree of risk aversion compared to only 22.58% for peasants with primary school education. Farrners with secondary school education seem to have the lowest degree of risk aversion with I I .48% falling in the risk preferring category while only 3.85% of non-educated farrners choose this alternative. These statistics then indicate that risk aversion tends to decrease with increase in education level. The econometric results on deterrninants of risk aversion will discuss this relationship more empirically.

Table 4.11: Risk Aversion and education No education Primary School Secendary School Education Education Risk Aversion Class % % %

Extreme 13.94 12.90 21.31 Severe 26.44 9.68 1.64 lntermediate 34.13 33.33 29.51 Modera/e 17.31 23.66 27.87 S/ight lo Neutra/ 4.33 11.83 8.20 Neutra/ to e_ref!rring_ 3.85 8.60 11.48 Source: Author from cxpcrimcntal data.

121 • Risk Aversion and Age

Age is a demographic characteristic that has long been hypothesized to affect an individual's degree of risk aversion. The results on the effects of age on risk tak.ing are mixed. From our sample, Table 4.12 gives the distribution of peasants risk attitude with respect to their age. We find oldest peasants with more than 60 years old to be more averse than the others. In the neutral to preferring class, there are not farrners with more than 60 years. In this age class, the dominant risk categories are extreme, severe and interrnediate risk aversion. However, we aiso find the youngest farrners with less than 30 years to fear risky alternatives. Only 3.70% of the youngest farrncrs chose the riskiest alternative. The highest percentage in choosing the riskiest alternative (9.88%) is found in the 30-45 years class.

Table 4.12: Risk Avenion and A.!i,e Age(· 30 years) Age (3045 years) Age (46-S9 years) Age(+ 60 ycars) Risk Avenion Class % % % %

Extreme 18.52 12.35 16.26 18.00 Severe 7.41 12.96 23.58 26.00 lntermediate 29.63 33.95 29.27 42.00 Moderate 29.63 20.37 21.95 14.00 S/ight to Neutra/ li.li 10.49 4.07 0.00 Neutra/ to preferring 3.70 9.88 4.88 0.00 Source: Author from experimental data

• Risk Aversion and Income

The association between income distribution and risk aversion categories is presented in Table 4.13. We see that the highest proportion of farrners falling into the extreme to severe risk aversion categories cornes from high income class. This is in contras! with the literature which predicts wealthier peasants to be Jess risk averse than the poor. But in general the table points up that peasant tend to fear risky alternatives, regardless of the income level.

122 Table 4.13: Risk Aversion and lncome In corne in FCF A Income in FCF A Jncomc in FCF A Income in FCF A <=500000 J 500000; 1000000 J J 1000000;2000000) >2000000 Risk A version Class % % % %

Extreme 15.52 16.48 7.92 24.07 Severe 11.21 18.68 24.75 18.52 Intermediate 38.79 35.16 27.72 27.78 Moderate 19.83 18.68 26.73 14.81 Slight to Neutra/ 9.48 5.49 5.94 5.56 Neutra/ to p_re~rrins_ 5.17 5.49 6.93 9.26 Source: Author from experimental data

4.4. Risk sources and farmers' risk management strategies

4.4.1. Farmers' perception of risk sources

A total of 8 sources of risk were presented to respondents in the survey. Farmers were asked to identify the sources of risk they have experienced and to score each source of risk (on a 5-point Likert scale ") from 1 (no impact) to 5 (high impact) to express how significant they considered each source to be of risk in terms of its potential impact on the performance of their farm. In doing this, farmers selected and ranked the different sources of risk from the Jess important to the most important (i.e. the source of risk they feared the most). The identified risk sources and their order of importance are presented in Tables 4.1436.

The majority of the farmers stated drought, pest/diseases, input access, output price fluctuation and ill health/death as important risk sources. Not as high as 7% of the respondents considered each of the risk sources of floods, unsecured land and labour important. The risk sources ranked between the top three (i.e. with 5 or 4 or 3 as score) by the majority of the farmers are output price risk, pest/diseases and input access risk; 88% of respondents ranked output price risk, 73% ranked pest/diseases, and 48% ranked input access in the top three sources of risk.

35 A Likert scale is psychometric scale commonly used in questionnaires, and is the most widely used scale in survey research. When responding to a Likert questionnaire item, respondents specify their Jevel of agreement to a statement. The scale is named after its inventor, Rensis Likert. 36 In table 4.14, a risk source is considered to have the rank 1 (2 or 3) if it has been given the score 5 ( 4 or 3) respectively by farmers.

123 Table 4.14: Identification of risk sources and rank (n = 362) RANK Risk Sources 1 2 3 A* B* Production Drought 13 73 12 27% 102 Pcst/diseasc 114 109 43 73% 273 Floods 0 Il 9 5.5% 68

Market Input access 43 69 64 48% 196 Outputprice 167 73 81 88% 329

Persona/ Ill-health/death 15 79 li 29% 115

Unavailability of Labour 0 0 3 0.8% li Land( unsecured) 0 1 4 1.4% 19

Source: Own computation from survey data Note: A• - Proportion of total respondents ranking each risk source among the most imponant three. B* - Number of respondents mentioning each risk source. For example 196 respondents (54% of total respondents) mentioned difficulties to access input as a risk source but only 176 (48% of total respondents) put il among the most important three.

Analysis of ranking of risk sources by cocoa fanners using mean of importance rating or score indicated output ( cocoa) price fluctuation as the first risk source with highest mean score of 4.01 (see Table 4.15). This could have been expected from Table 4.14 since 167 of the respondents ranked it as the most important risk source with a score of 5. Next in importance are pest/diseases with a score of 3.66. The importance of this risk source can be explain by the fact that cocoa farmers in Côte d'Ivoire has experienced these last year some important failure in yicld due to diseases like black pod and swollen shoot. The third risk source is related to input access with a mean score of 2.85 followed by ill• health/death of a member of the household. The highly affected input prices are those of fertilizers which are very unstable and high since a fcw years. ln conclusion, cocoa farmers in Côte d'Ivoire ranked market risks, to be precise output price risk (the risk that actual cocoa price will be different than originally expected price ), as the most important source of risk.

124 Table 4.15: Mean scores and rank of sources of risks (n = 362) Risk Sources Importance Score Rank by Mean Score Mean (n=362) Production Drought 2.14 5 Pest/disease 3.66 2 Floods 1.30 6

Market Input access 2.85 3 Output price 4.01 1

Persona/ Ill-health/death 2.64 4

Unavailability of Labour 1.03 8 Land (unsecurcd) 1.16 7

Source: Own computation from survey data Note: For labour risk, ail the farmers stated that it had no impact on the performance of their farm since il is not source of risks.

4.4.2. Risk management strategies used by farmers

There are several strategies that farm operators can use to reduce the farm exposure to risks. The strategies can be classified into modem and traditional risk management tools. The modem instruments include insurance, forward contract, options, futures etc. In the absence of modem risk management tools (this is the case in rural Africa), fanners can rely on some traditional strategies to deal with risk. This section summarizes the most important traditional risk management strategies used by the surveyed cocoa fanners in Côte d'Ivoire; these are: crop diversification, precautionary savings and participating in social network. The percentage of producers in the sample using some form of crop diversification is 32.87%. Producers who utilize precautionary savings represent 57.46% and those who are members of social network comprise 42.82% of the sample. The detailed proportions of producers using different combinations of these risk management strategies are presented in Table 4.16.

125 Table 4.16: Proportion of Producers Adopting Different Combinations of Risk Management Tools Possible Risk Management Strategies Number of Proportion(%) Combinations* farmers ( I) Use no risk management strategy 94 25.96 (2) Use crop diversification only 9 2.50 (3) Use prccautionary saving only 87 24.03 ( 4) Use social network only 26 7.18 (5) Use crop diversification and prccautionary saving 17 4.70 (6) Use crop diversification and social nctwork 25 6.90 (7) Use precautionary saving and social network 36 9.95 (8) Use ail three risk management strategies 68 18.78 Total 362 100 Source: Own computation from survey data Note: • The different combinations of risk management strategies above should serve as the basis for coding the dependent variable in a multinomial Probit/ Logit specification. The dependent variable then will be coded such that Y,= 1, ... ,8 and only one eombination (among the eight) will be chosen by the producer. ln spite of the risk sources mentioned by cocoa farmers, more than 25% of farmers are not using any of the risk management tools considered here. Less than 35% of producers in the sample used only one risk management strategies (2.50%, 24.03%, and 7.18% of cocoa farmers used only crop diversification, only precautionary savings or only social network respectively) to deal with risk. Note that relying on precautionary saving to deal with risk was the strategy most using by cocoa farmers in the sample. Also note that more than 40% of the producers used different combinations of risk management tools with 18.78% using simultaneously ail the three risk management strategies considered in this study. This information suggests that the decision to use one risk management strategy might be correlated with whether or not other management strategies will be used. This hypothesis will be evaluated by calculating pairwise correlation coefficients across the three risk management strategies in the estimation of the MVP mode! further of this thesis. These coefficients measure the correlation between the risk management strategies, after controlling for the influence of the observed factors (Greene 2003).

4.5. Econometric results and discussion

The empirical results of the study are presented in this section. ln assessing farmers' behavioural response to risks, we first analysed how persona) characteristics of individuals are correlated with his/her level of risk aversion. Next, empirical investigations of factors affecting the adoption of risk management strategies are presented.

126 4.5.1. Determinants of risk aversion

Many researchers have found that risk aversion matters for economic behaviour under uncertainty. Knowledge on farmers' attitudes towards risks is then important in determining the strategies for agricultural dcvclopment. In an attempt to explain variations in farmers' risk attitudes, an ordered Logit model has been estimated.

4.5.1.1. Ordered Logit estimation results

This section presents estimation results, using the methodology described in Section 3.5.2. First, the exogeneity of farm income was tested, using Rivers and Vuong's procedure. Tables A.5, A.6, A. 7 and A.8 in Appendix 5 present first stage estimates for the endogenous variable, farm income, at each game level. Total farm labour is used to instrumentfarm income. At the four levels of the game, the instrumental variable has the expected effects: total fann labour has strong explanatory power for fann income and an increase in the amount of agricultural labour leads to increase in farm income. This significance of the coefficient of total fann labour satisfies one of the requirements for a valid instrument, i.e., that the instrument be correlated with the endogenous variable. The endogeneity test results from the augmented regressions displayed in Table A.9 in Appendix 5 show that the coefficients, 0, of the residuals from the first stage regressions are not significant in the four specifications. As mentioned in Section 3.5.2.2, the test

: H0 B = 0 is also a test for the existence of endogeneity. As the coefficient is not significant, we cannot reject the null of no endogeneity of farm income at the four game levels. This gives us confidence in the rcsults discussed below.

The results of the ordered Logit model are given in Table 4.17 where the dependent variable is the respondent's risk aversion category (1 "extreme risk aversion" to 6 "risk preferring"). To correct for possible heteroscedasticity, the White estimator of variances (White, 1980; StataCorp, 2001) is used, instead of the conventional ML variance estimator. The reported standard errors are, therefore, robust standard errors. First of all, a note of caution is warranted about interpretation of the results in an ordered model. In our ordered Logit model, the dependent variable is an order (rank) of risk aversion where extreme risk aversion takes rank number one (1) and risk-loving is indicated by rank number six (6). Therefore, a positive coefficient sign indicates a reduction in the degree of

127 risk aversion. Howevcr, the coefficient docs not provide an indication of the magnitude of the effecr".

Table 4.17: Ordcrcd Lo!i!t models of risk aversion ecr ~amc lcvcl lOOFCFA 1000 FCFA 5000 FCFA 100 FCFA+ Variables Coef. Std. Err Coef. Std. Err Coef. Std. Err Coef. Std. Err

Wealth Indicators Farm incarne l.40c-7•• 5.78e-8 l.22e-8 9.46e-8 -1.1 lc-7 6.85c-8 -6.70e-8 7.42e-8 Value oflivestock 2.12e-7••• 7.09c-8 2.32e-7 •• 9.96e-8 l.65c-7 2.17c-7 7.51e-8 l.lOe-7

Household/Head Characteristics Age -0.047••• 0.010 -0.027* •• 0.009 -0.018' 0.010 -0.022 •• 0.010 Education 0.516 •• 0.238 0.586 •• 0.229 0.293 0.215 0.211 0.219 Sex (Male) 1.668 ••• 0.580 J.423 ••• 0.367 1.338··· 0.495 0.804 •• 0.370 Origin'" Native 0.910* •• 0.309 0.302 0.281 -0.360 0.309 -0.601 • 0.357 Migrant2 0.396 0.254 0.257 0.254 -0.417· 0.245 -0.558 •• 0.243 Matrimonial status -0.272 0.268 -0.182 0.300 0.067 0.322 -0.377 0.346 Household size 0.006 0.014 -0.021 0.017 -0.022 0.015 0.018 0.014 Dependence in cocoa 0.516 •• 0.237 -0.173 0.238 -0.213 0.237 -0.158 0.250

Game variable+++ Luckl - - 0.292°• 0.111 Luckl2 - - - - 0.289••• 0.079 Luckl23 - - - - 0.120• 0.070

Location++ Zone J -0.607 •• 0.264 -0.229 0.255 -0.444• 0.232 0.357 0.270 Zone3 -0.654 •• 0.280 -0.581 •• 0.296 -0.666** 0.275 0.350 0.269

Threshold parameters Cuti (ai) -2.805 0.700 -2.131 0.559 -1.98 0.658 -1.924 0.586

Cut2 (a2) -1.919 0.711 -1.001 0.539 -0.377 0.644 -0.224 0.575

Cut3 (a3) -0.335 0.707 0.543 0.530 1.118 0.643 1.165 0.586

Cut4 (a4) 0.786 0.703 1.852 0.543 2.400 0.661 2.772 0.645 Cut5 (a,) 2.017 0.710 2.709 0.567 3.268 0.709 3.699 0.721

Log likelihood fonction -573.871 -567.268 -540.557 -521.826 Wald Chi-Squared 84.42 67.99 56.18 43.30 P-value 0.000 0.000 0.000 0.000 Pseudo R2 0.20 0.14 0.13 0.11 Numberof 362 362 362 362 observations

Dependent variable: degrees ofrisk aversion (l=Extreme, ... ,6=Neutral to Risk Loving). + This refers to the 100 FCFA game with gains and tosses. ++ Migrant! is the reference for Origin and Zone 2 is the reference for Location. +++ Luckl is the luck from gamet (100 FCFA game with only gains); Luckl2 is combincd luck from gamet and game2 (1000 FCFA game); Luckl23 is combined luck from gamel, game2 and game3 (5000 FCFA game). •••, ••, • indicate significance levels at 1 %, 5%, and 10% levels, respectively.

37 The magnitude of the effect is provided by the marginal effects.

128 The Wald Chi-Square test that at least one of the predictors' regression coefficients is not equal to zero in the mode!, which is a measure of the overall goodness of fit of the mode!, provides evidence of a strong fit (p-value of 0.000) in the four regressions. The least value of the Pseudo R2 is O.! 1, an acceptable value in (cross-sectional) studies like tbese. In general, most variables bave a significant effect and the coefficients have the expected signs particularly at low game level.

• Effect of wealth on risk aversion

Ali the wealth indicators (farm incarne, value of livestock) are significant and have the expected negative effect on risk aversion (i.e. positive sign in the ordered Logit results) at low stake indicating that more wealth is indeed correlated with a lower degree of risk aversion. This result is consistent with the literature and the DARA hypothesis. The presence of DARA indicates the existing significant difference in risk behaviour between relatively poor and wealthy farm households. At higher game level (5000 FCFA), tbese wealth variables do not appear to affect risk aversion behaviour at ail. This means that at highcr level of investment (when the amounts of money at risk are important), the farmer' s risk aversion behaviour is independent of bis level of wealth. Other factors cou Id explain farmers' attitudes towards risk at higher investment levels, but not the level of wealth. This insignificant effect of wealth on risk aversion can also be observed when farmers face games with losses. The level ofwealth does not affect farmers' risk attitudes in situations involving real losses.

• Effect of household and household head characteristics on risk aversion

The effects of the second group of independent variables on farmers risk behaviour are mixed. Parameters estimates on matrimonial status and household size are not statistically significant at ail game levels. With respect to household size, this finding is contrary to Wik and Holden (1998) results from Zambia where they found household size to be significant and negatively correlated with risk aversion. But our result is in line with the Ethiopian study by Y esuf (2007) who found an insignificant effect of household size in the total sample. Howevcr, we found age to be highly corrclatcd with level of risk aversion. The coefficient is always negative (positive relationship) at ail game levels and

129 also in the gains and lasses game, indicating that older farmers are more risk averse than younger farmers. This result is consistent with many studies (Belaid and Stanley, 1987; Brüntrup, 2000; Gomez-Limon et al, 2002; Yesuf, 2007). Education bas been found to affect risk aversion negatively. This is in line with our expectation and the result ofYesuf (2007). In line with Wik and Holden (1998) and Yesuf (2007), we also found risk aversion to be significantly related to gender of the fanner where male heads are Jess risk averse than female heads as we expected.

Whether the fanner is a native or a migrant also bas an impact on his/her degree of risk aversion. At low game level (100 FCFA), we found a significant and negative relationship between being a native and the fanner's degree of risk aversion. This means that native people, relative to migrants from Côte d'Ivoire, migrant] (the reference modality), are Jess risk averse (because of the positive sign). This result can be interpreted in liaison with the education level as in our sample; only 40% of migrants from Côte d'Ivoire are literate compared to 80% for native farmers. But when considering the gains and losses game, this significant relationship became positive; this means that native people, relative to migrants from Côte d'Ivoire become more risk averse. Considering migrants from neighbouring countries migrant2, the results show an insignificant relationship with risk aversion at low game levels (100 FCFA and 1000 FCFA). However, at higher game level and for game involving lasses, we found a significant positive relationship. Relative to the reference modality which is Ivorian migrants, foreign farmers are more risk averse. This makes sense considering the existence ofland disputes always faced by foreign farmers.

A significant parameter estimate on fanner dependence on cocoa revenue is only observed at the lowest level of the game (100 FCFA). Consistently insignificant but positively correlated to the degree of risk aversion through the las! three games, we found level of cocoa dependence to be significant and negatively correlated with risk aversion in the first game. It means that the more the farmer's dependence in cocoa revenue, the Jess his/her degree of risk aversion. This is an unexpected result as we hypothesized fann household with high dependence on cocoa revenue to be more risk averse.

130 • Effect of previous luck on risk aversion

As expected, we found a highly significant negative relationship between prior success and degree of risk aversion, as indicated by a significant positive coefficient estimate for the previous luck variables. Our result is in line with findings by Binswanger ( 1980), Wik and Holden (1998) and Yesuf (2007) who found this variable to be highly significant and ncgativcly correlated with risk aversion. This implies that people rcvisc thcir cxpcctations as the game level progresses even if the actual probability of success remains constant (coin toss). Similar behaviours could also be observed in actual farrn investment decisions where farm households who had encountered a series of droughts may be more reluctant to undertake risky investment decisions, at least for a white, even when probabilities and wealth levels are unchanged throughout those periods. But one should note that when it cornes of the gains and tosses game, previous luck does not appear to strongly influence the farmer's degree of risk aversion.

• Effect of geographical location on risk aversion

Finally, significant location dummies indicate systematic, but unobservable differences in risk aversion across study zones. But this result is supported by the descriptive statistics on the distribution of risk aversion coefficient by zone where we found farrners from zone 2 ( the reference zone) to be less risk averse.

4.5.1.2. Marginal effects on risk aversion categories

The estimated coefficients of the ordered Logit model only describe the general direction of change, not specific to each category of risk aversion. Changes in the predicted probabilities (marginal effects) of the observed outcomes ofrisk aversion computed at the means of ail variables are provided in Table 4.18; only the results for the significant variables discussed above are presented. Sorne simulation results on predicted probabilities from varying levels of selected independent variables are also illustrated in Appendix 7. The marginal effects represent the change in the probability of the outcorne variable for a unit change in a specific explanatory variable when ail other explanatory variables in the rnodel are held constant at their means.

131 Table 4.18: Changes in Predicted Probabilities (marginal effects) bl risk categories. Extreme Scvcre Intermediate Moderatc Ncutral Preferring Marg. EIT P. Marg. Eff P. Marg. Eff P. Marg. Eff P. Marg. Eff P. Marg. EIT P. Variables value value value value value value

100 FCFA game Farm income -8.6c-9** 0.024 -8.8e-9** 0.024 -1.7e-8** 0.021 6.5e-9* 0,057 l.5e-8** 0.020 1.2e-8** 0.015 V. of livestock -1.3e-8••• 0.009 -1.3c-8*** 0.007 -2.6e-8••• 0.004 9.9e-9** 0.029 2.3e-8*** 0.004 l.9e-8*** 0.004 Education -0.031 •• 0.046 -0.032** 0.044 -0.063** 0.027 0.021• 0.056 0.056** 0.034 0.048** 0.038 Age 0.003*** 0.000 0.002••• 0.000 0.005*** 0.000 -0.002••• 0.002 -0.005••• 0.000 -0.004••• 0.000 Sex (Male) -0.198* 0.066 -0.123••• 0.002 -0.044 0.416 0.146*** 0.005 0.134••• 0.000 0.085••• 0.000 Native -0.045**• 0.001 -0.049*** 0.001 -0.118*** 0.006 0.013 0.314 0.097··· 0.002 0.102•• 0.022 Migrant2 -0.022* 0.093 -0.023* 0.100 -0.050 0.139 0.013* 0.053 0.043 0.120 0.039 0.163 Dep. in cocoa -0.031** 0.034 -0.032•• 0.034 -0.062** 0.036 0.023** 0.046 0.056** 0.032 0.047•• 0.039 Zone/ 0.040** 0.030 0.040•• 0.031 0.069** 0.020 -0.032* 0.058 -0.065** 0.020 -0.052•• 0.021 Zone3 0.045** 0.040 0.044** 0.036 0.012•• 0.012 -0.038* 0.059 -0.069** 0.016 -0.054** 0.039

1000 FCFA game Luckl -0.031 •• 0.018 -0.030••• 0.010 -0.001 0.684 0.034••• 0.010 0.015** 0.013 0.014•• 0.019 V. of livestock -2.5e-8•• 0.025 -2.4e-8•• 0.024 -1.5e-9 0.693 2.7e-8•• 0.025 1.2e-8** 0.036 l.le-8•• 0.022 Education -0.062••• 0.008 -0.059** 0.016 -0.007 0.467 0.067** 0.011 0.031** 0.023 0.030** 0.027 Age 0.003••• 0.004 0.002••• 0.010 0.000 0.702 -0.003••• 0.006 -0.001 •• 0.016 -0.001 •• 0.016 Sex (Male) -0.240*** 0.003 -0.099••• 0.000 0.111•• 0.020 0.139••• 0.000 0.047••• 0.000 0.041*** 0.000 Zone3 0.069* 0.083 0.058** 0.044 -0.006 0.640 -0.067** 0.044 -0.028** 0.042 -0.026** 0.046

5000 FCFA game Luck/2 -0.036••• 0.000 -0.035••• 0.001 0.024••• 0.005 0.029••• 0.001 0.009*** 0.005 0.008*** 0.002 Sex(Male) -0.243** 0.033 -0.063** 0.031 0.159*** 0.009 0.096••• 0.000 0.028••• 0.002 0.022••• 0.003 Age 0.001 0.117 0.001 0.124 -0.001 0.126 -0.001 0.115 -0.000 0.141 -0.000 0.172 Migrant] 0.056 0.117 0.047* 0.069 -0.039 0.138 -0.040* 0.078 -0.013* 0.082 -0.010• 0.086 Zone/ 0.058* 0.076 0.052•• 0.043 -0.040* 0.074 -0.043* 0.057 -0.014* 0.076 -0.012• 0.078 Zone3 0.091•• 0.036 0.013••• 0.005 -0.063** 0.042 -0.063** 0.011 -0.020•• 0.016 -0.017** 0.022

100 FCF A game with losses Luck/23 -0.017* 0.088 -0.012 0.102 omz- 0.090 0.012• 0.093 0.002 0.135 0.001 0.173 Sex (Male) -0.145* 0.062 -0.039••• 0.003 0.095** 0.035 0.067*** 0.009 0.013•• 0.027 0,009•• 0.047 Age 0,003•• 0.026 0.002•• 0.044 -0.002•• 0.026 -0.002•• 0.032 -0.000• 0.096 -0.000 0.107 Native 0.097 0.137 0.047•• 0.014 -0.067 0.128 -0.057* 0.057 -0.011 •• 0.049 -0.008* 0.081 Migrant] o.css« 0.034 0.046••• 0.009 -0.062** 0.040 -0.054** 0.014 -0.011 •• 0.027 -0.008•• 0.041

Dependent variable: degrees of risk aversion. •••, ••, • indicate significance levels at the 1 %, 5%, and 10% levels respectively.

From this table, we can give a very substantive interpretation of our regression results and confinn the basic conclusions of table 4.17. For example, holding all other variables at their means in the first game, one year increment in age increases the probabilities of falling into extreme, severe and intermediate risk categories by 0.003, 0.002 and 0.005 units respectively, but at the same time reduces the probabilities of falling into the moderate, neutral and risk preferring categories by 0.002, 0.005 and 0.004 units

132 respectively. Similarly, holding ail other variables at their means, a unit increment in wealth - whether in the fonn of farm income or value of livestock - reduces the probabilities of falling into the extreme to intennediate risk aversion categories but at the same time increases the probabilities of ending up in the moderate, neutral and risk preferring categories by the respective values provided in the table. This indicates the presence ofDARA as we hypothesized.

Furthermore, past successes (i.e. previous luck) tend to have empirically important effects in view of results from games 2 and 3. For example in game 2, if the respondent succeeded in one more game previously out of all those played there is a reduced probability of being in the extreme risk aversion category of 3.1 % and a 3.0% lower chance of exhibiting severe risk averting behaviour. At the same time, previous luck tends to increase the probabilities of falling into the moderate, neutral and risk preferring categories by 3.4 percent, 1.5 percent and 1.4% respectively. Similar marginal effects of previous luck on risk aversion are observed in game 3 and game 4. We can conclude from these results that past successes build on each other suggesting that first introducing agricultural extension measures with very high probabilities of success can quickly help villagers being more comfortable with taking on subsequent risks.

For dummies, the results can be interpreted in a similar way but as discrete changes from 0 to 1. For example, men are more likely to fall into the moderate to risk preferring categories, while women are likely to be found in the extreme to intcnnediate risk categories. In game 1, being a man incrcases the probabilities of falling into the moderate, neutral and risk preferring categories by 14.6%, 13.4% and 8.5% respectively but reduces the probability of being in the extreme, severe and intermediate risk aversion categories by 19.8%, 12.3% and 4.4% respectively. We also find an empirically important link between education (literate or not) and risk aversion. lndeed, being literate increases the probabilitics of falling into the moderate to risk preferring categories, while illiterate heads are more likely to fall into the extreme to intermediate categories. Similar interpretations can be made for other variables.

To better illustrate these marginal effects, we graph the relationship between some major explanatory variables like wealth (value of livestock) and age, and the predicted probability of falling into a specific risk aversion category using the results from the two

133 real games (game 1 and 2), as represented in Appendix 7. In computing the predicted probabilities, we evaluated ail other controls, except value of /ivestock and age, at their sample means. We can see from the figures that a similar conclusion supporting DARA and !PRA can also be drawn. Farm households are inclined to undertake relatively more risky decisions at higher level of wealth. We can also observe farmers being more risk averse in game 2 ( 1000 FCF A game) compared to their risk behaviour in game 1 ( 100 FCFA game). For real economic choices under risk with much higher values involved and longer lime horizons, for instance minerai fertiliser purchase, investment in animal traction equipment or even abandonment of subsistence production, it can be expected that risk aversion will be even higher.

4.5.2. Determinants of risk management strategies adoption

This section presents the estimation results on the factors affecting the adoption of risk management strategies by farmers. The explanatory variable age has been removed from the mode! due to high collinearity with experience. The correlation coefficients between the error terms in each equation are presented as well as the parameters estimates from the MVPmodel.

4.5.2.J. Multivariate Probit estimation results: Correlation coefficients among risk management adoption decisions

Pairwise correlation coefficients across the three risk management adoption equations are presented in Table 4.19. These coefficients measure the correlation between the risk management decisions considered, after the influence of the observed factors has been accounted for (Greene 2003). These coefficients are essentially the pairwise correlation between the error terms in the system of equations in the multivariate probit mode!. Ali of the correlation coefficients are positive and highly statistically significant at the 1 % level, except the correlation between precautionary savings and social network which is significant at 10% level. This supports our hypothesis that the error terms in the risk management adoption cquations arc correlated, and justifies the use of a multivariate probit approach instead of three independent probit models. The perceived interaction between risk management strategies (which is unobserved) is accounted for in the multivariate probit approach. Moreover, the positive signs of the correlation coefficients

134 suggest that the decision to adopt one particular risk management strategy may make it more likely that another strategy will be adopted. For example, a producer who uses crop diversification to deal with risk may also tend to use precautionary savings as risk management strategy (once the observable factors are controlled for), and vice versa (see Velandia et al 2009 for a similar result). As an alternative explanation, it could be argued that producers who adopt one kind ofrisk management strategy (say, crop diversification) tend to be highly risk-averse such that, behaviourally, they are also more likely to adopt other risk management strategies (say precautionary savings and/or social network).

Table 4.19: Multivariate Probit Model Results: Correlation Coefficients of Risk Management Ado.e,tion Decisions Risk Management Correlation Coefficient ' Standard Deviation Strategies

Crop diversification and precautionary saving 0.2892*** 0.1023 Crop diversification and social network 0.1002••• 0.0637 Precautionary saving and social network 0.1713* 0.0936

' Correlation coefficients between the residuals from the multivariate Probit equations. •••, • indicate statistical significance at the 1 % level and 10 % level respectively.

4.5.2.2. Multivariate Probit estimation results: Parameters estimates

The parameter estimates from the multivariate probit and (for comparison) the individual

probit models are presented in Table 4.20. The Likelihood Ratio test of P,1 (positive) allows us to justify again the estimation of this multivariate probit and not of three independent probits: the hypothesis Ho of conjoint nullity of p., can be rejected (p-value

= 0.0000). The Wald chi2 test allows us to reject also the H0 hypothesis of conjoint nullity of variable coefficients included in the estimation.

Based on the multivariate mode!, the observed factors that tend to significantly affect adoption of crop diversification are farm size, literacy, off-farm emp/oyment and cocoa price risk. The effect of cocoa farm size on adoption of crop diversification gives an unexpected positive sign. Although this variable is related to a larger asset base and thus to a higher capacity to bear risk, a large farm size also lead to an improvement in the financial capacity of the farmers. This increases financial capacity enables risk averse peasants to undertake investments in the production of other crops in a risky environment.

135 Table 4.20: Parameter estimates from the Multivariate Probit and Individual Probit Approach for estimating the factors affecting adoetion of agricultural risk management strategies Parameters Estimates from the Multivariate Parameters Estimates from the Individual Probit Model Probit Model

Adoption Equation Adoption Equation Crop Precautionary Social Crop Prccautionary Social Network Indeeendent Variables Diversification Savins Network Diversification Savins Farm Characteristics Cocoafarm size 0.0469* •• 0.0154* 0.0451*** 0.0466*** 0.0106 0.0427*** (0.0144) (0.0092) (0.0170) (0.0144) (0.0103) (0.0133) Land ownership -0.3186 -1.6987** -0.6215 -0.3622 -1.6267*** -0.5136 (0.6149) (0.7298) (0.4846) (0.5846) (0.6210) (0.4998)

Farmer/Household Characteristics Literacy 0.3728** -0.0074 0.2894* 0.4042** 0.0059 0.3117** (0.1755) (0.1539) (0.1567) (0.1849) (0.1585) (0.1590) Experience 0.0011 -0.0197* -0.0277** 0.0015 -0.0184* -0.0252** (0.0122) (0.0120) (0.0109) (0.0122) (0.0108) (0.0113) Access to credit -0.0552 0.2572 -0.0190 -0.1088 0.2596 -0.0559 (0.1867) (0.1648) (0.1718) (0.1952) (0.1682) (0.1709) Household size -0.0063 0.0698** 0.0631** -0.0036 0.0680** 0.0642** (0.0298) (0.0328) (0.0302) (0.0360) (0.0301) (0.0310) Househo/d size square -0.0001 -0.0013* -0.0011• -0.0001 -0.0013* -0.0011 (0.0005) (0.0007) (0.0006) (0.0008) (0.0007) (0.0007) Ojf-fann employment 0.8862*** 0.4511 * 0.5854** 0.8969*** 0.4523 0.5648** (0.2880) (0.2733) (0.2779) (0.2957) (0.3012) (0.2818)

Risk Indicators Risk Perception Output price risk 0.4509* 0.0293 -0.2707 0.4707* 0.0562 -0.2402 (0.2720) (0.2468) (0.2708) (0.2786) (0.2664) (0.2484) Pest/disease -0.1222 -0.6187*** 0.2572 -0.1029 -0.6245*** 0.2713 (0.2177) (0.1774) (0.1813) (0.2158) (0.1924) (0.1885) Input access 0.1628 0.3832** -0.3596** 0.1897 0.3975** -0.3329* (0.1975) (0.1738) (0.1721) (0.1879) (0.1763) (0.1715) lllness/death 0.1765 0.3091* 0.5450*** 0.1790 0.2998* 0.5369*** (0.1845) (0.1645) (0.1660) (0.1910) (0.1724) (0.1709) Risk aversion High risk aversion 0.2973 0.4227*** 0.3929** 0.2373 0.4357*** 0.3788** (0.1765) (0.1584) (0.1586) (0.1798) (0.1654) (0.1602) Low risk aversion -0.3789 -0.4266 -0.5957* -0.3791 -0.4401 -0.6654* (0.3529) (0.3036) (0.3361) (0.3854) (0.3360) (0.3621)

Information Access to information -0.1915 0.4713** 0.0403 -0.2598 0.4670** 0.0236 (0.2017) (O.l 963) (0.1941) (0.2081) (0.1939) (O.l 90 l)

Location Zone/ -0.3360 -0.5576** -0.0984 -0.4083* -0.5612** -0.1052 (0.2714) (0.2184) (0.2231) (0.2487) (0.2202) (0.2173) Zone] 1.6756*** 0.4359** 0.8015*** 1.6496*** 0.4398* (0.8171 )••• (0.2438) (0.2145) (0.2242) (0.2460) (0.2262) (0.2203)

Log-likelihood value -533.7521 -152.4718 -206.8515 -204.2310 Wald test chi2 (S l) 295.52*** 59.6046*** LR test of p11 LR test chi2 ( 17) - 153.55*** 80.05*** 85.88*** Pseudo R2 0.3349 0.1621 0.1737 Number of observations 362 362 362 362 Note: Figures in parenthesis are robust standard errors. •,••,and••• represent statistical significance at 10%, 5%, and 1% levels, respectively.

Likelihood Ratio Test H0 : p11 = p31 = P32 = 0, z(;) = 59.6046, p-value = 0.0000

136 More educated cocoa fanners tend to adopt crop diversification. They have more ability to assess the merit of crop diversification as a strategy to deal with the fluctuation on their income. This is in line with results from Tavernier and Onyango (2008) who fund a positive relationship between education and respondents who use enterprise diversification as a major approach to managing risk on their fann. Moreover, education of the fanner plays vital role in efficiency in production, e.g., a better educated fanner will do better in growing a crop than the less educated fanner growing the same crop. We also find the existence of off-farm fann employment to favour the adoption of crop diversification. In fact producers who are engaged in some off-farm activities are more likely to be highly risk averse and, behaviourally, they will adopt some risk management strategies. Volatility of cocoa prices has been fund to be the unique source of risk that significantly favours a fanner adoption of crop diversification as risk management strategy. In response to fluctuations in cocoa price since the liberalisation of the cocoa sector in 1999, cocoa fanners in Western Côte d'Ivoire find investments in other perennial crops like oil palm and hevea to be a mean to reduce growing uncertainty on their incarne. Contrary to our expectation risk aversion is not a motivation for adoption of crop diversification. The positive and significant parameter estirnate on zone 2 durnmy reflects a higher likelihood of adopting crop diversification in this zone relative to the omitted zone (zone 3).

For the precautionary savings adoption equation, the significant variables in the multivariate probit approach are farm size, land ownership, farming experience, household size, off-farm employment, pestldisease of cocoa plants, input access, illness/death of a member of the house, risk aversion and famer 's access to information. As with the crop diversification results above, producers with larger farm size tend to use precautionary savings as risk management strategy. On the other hand, producers who own their land do not tend to adopt risk management strategies. Considering land as a wealth variable, owning the land may lead to lower the fanner's level of risk aversion and consequently decrease the likelihood of adopting risk management strategies as precautionary savings. As expected, fanning experience has been fund to decrease the likelihood of using risk management strategies. This funding may be explain by the fact fanners with low experience generally Jack some farming skills. The lack of skill then increases their fear of the risks inherent to agricultural production and consequently their

137 needs to resort to risk management strategies. The opposite is true for farmers with high experience. We also find a large household size to be positively related to the use of precautionary savings to deal with risk. An increase in family size means more people to feed, to care and then increases the level of vulnerability of the household. Hence, taking into account the amount of uncertainty regarding the future, it is reasonable for large households to smooth their present consumption (by making saving) in order to secure their future welfare. As with crop diversification, the existence of off-farm employment is positively related to the precautionary saving behaviour. As expected, we find a positive relationship between risk perception variables and precautionary saving behaviour. Particularly, we find household who experienced illness/death of a member in the past to be more likely to save for precautionary motive. The difficulties to access input are also found to increase the likelihood of household to adopt a precautionary saving behaviour. However, we surprisingly find a negative relationship between the presence of pest/disease of cocoa plants and the adoption of risk management strategies. As expected, our results show a highly significant (1%) positive relationship between risk aversion and precautionary saving motive. This funding is in line with the theory of precautionary saving which posits that risk aversion enhances precautionary savings. At opposite, low risk aversion has a negative effect on the use of precautionary saving, but this effect is not significant. This may be due to the low proportion offarmers falling into this risk aversion category. Access to information, unexpectedly, increases the likelihood of the household to adopt a precautionary saving behaviour. Relative to zone 3, we find again a positive and significant parameter estimate for zone 2 dummy but a negative and significant parameter for zone 1 dummy.

Variables significant in the social network participation analysis are fann size, literacy, experience, household size, off-farm employment, input access, il/ness/death and risk aversion. Again, producers with large farrn size are more likely to participate in a social network. Literate producers are more likely to be members of social network and those with higher farming experience are less likely to use social networks as a mean to cope with risk. As with precautionary saving results, we also find a large household to be more likely to participate in social network. This is a kind of insurance for ail members of the household. In case of risk occurrence, a household member of a social network will share the risk with other members of bis network. Off-farm employment is again found to be

138 positively related to this risk management strategy. Producers who perceived difficulties to access input as a major risk source are less likely to participate in social network; only those who have experienced illness/death of a member of the household have a higher likelihood to participate in social networks. Being a part of a social network guarantees social and financial supports from a network of friends in case of illness or death of a household's member. Again high risk aversion increases the likelihood of participating in a social network while low risk aversion is negatively and significantly correlated to the use of social network to deal with illness/death risk. Producers from zone 2 again have a higher likelihood of participating in social networks relative to the omitted zone (zone 3).

The results discussed above are for the parameter estimates from the multivariate probit approach. Comparing these results with the parameter estimates from the individual probit approach, we find that in general, the signs of estimated parameters in both approaches are fairly similar. However, examination of the significance of parameters shows some differences in results in both approaches. Note first that some variables which were not significant in the individual probit models became significant when taking into account the endogeneity between risk management strategies in the MVP mode!. These variables include risk aversion in the crop diversification equation, Jam, size and off-farm employment in the precautionary saving equation, and household size square in the social network equation. The second point of divergence between results from both approaches is related to the level of significance of some variables. The individual probits for example reported zone 2 dummy in the precautionary saving equation and input access in the social network equation to be significant at 10% while the MVP reported significance at 5% level for these variables. Hence, by ignoring the correlation between the different risk management strategies, the individual probit approach gives biased estimates which could lead to wrong conclusions about the significance of the parameter estimates.

4.6. Conclusion

Throughout this chapter, we have looked at factors influencing farmers' decision-mak.ing under uncertainty. Based on the information gleaned from the ordered logit mode! and the multivariate probit procedure, policy makers and other risk management information providers could be able to tailor their programs better. Therefore, some policy

139 implications arc provided in the following section which dcals with the general conclusion of the thesis.

140 GENERAL CONCLUSION AND POLICY IMPLICATIONS

~ Summary of the major findings

This dissertation is intended to contribute to the ongoing debate on agricultural risk management in developing countries by providing empirical evidence on Ivorian cocoa farmers' risk aversion and the determinants of producers' adoption of traditional risk management strategies to mitigate their exposure to risks. Using time-series data, an experimental gambling approach as well as micro-econometrics data analysis, the findings from the study are the followings: lvorian cocoa sec/or: The high taxation found in the Ivorian cocoa sector compared to other cocoa producing countries tends to lower producer price in Côte d'Ivoire. The low farm gate prices combined with the volatility of prices since the liberalisation of the scctor have motivated farmers shift to alternatives perennial crops.

Farmers' risk aversion: Using an experimental gambling approach with real payoffs, we investigated farmers' attitudes toward risk and found a relatively high level of risk aversion among Ivorian cocoa farmers. Explicitly, the results showed that more than 45% of the households exhibited severe to extreme risk aversion, with a constant partial risk aversion coefficient of more than 2. With careful construction of the experiment, the nature of partial risk aversion was examined and the data supported the existence of IPRA; i.e. risk aversion increased with the level of the game. This finding is consistent with Binswanger ( 1980) and Y esuf (2007). The validity of some of the major predictions of the expected utility theory ( concavity and asset integration) was also tested and the predictions of risk neutrality for smaller stakes ( concavity) and of similar preferences for gains and losses (asset integration) were not supported by our experiment. Particularly, we found more than 15% of households to be severely to extremely risk averse even at lower level of the game, and also an inclination of farm households to be more risk averse in game involving actual losses.

Farmers' risk-aversion and households ' persona/ variables: Using an ordered Logit mode!, the effects of farm/farmer's characteristics on risk aversion were analysed. Household wealth (value of livestock, education, and farm income) was found to be

141 negatively related to the risk aversion behaviour of farmers. The empirical finding that wealthier farmers are more risk taking is confirmed by this study. This result then is consistent with the literature and supports the DARA hypothesis. Our results indicated that older farmers are more risk averse than younger. We also found risk aversion to be significantly related to the gender of the farmer, where male heads are less risk averse than female heads. Native farmers, relative to migrants were found to be less risk averse decision makers. The level of dependence on cocoa revenue was also found to be positively related to farmers' degree of risk aversion. The more the dependence of the household in cocoa revenue, the more risk averse is the head. Finally, the results revealed that farmers who have experienced some previous luck in the past, tend to be Jess risk averse than the others.

Risk sources and farmers' risk perceptions: Using 5-point Likert scale, we found perceptions on risk sources to be very persona) and specific across cocoa farmers. However, farmers were relatively in agreement when evaluating the impacts of some sources of risks: the majority of the farmers ranked market risk, to be precise output price risk (the risk that actual cocoa price will be different than originally expected price) as the most important source of risk. Technical risks (pest/diseases) were ranked next. The risks related to the difficulties to access input were ranked as the third source of risk in cocoa production, followed by the risk of illness/death of one or more household members.

Farmers' adoption of risk reducing strategies: Using a multivariatc Probit mode! which accounts for simultaneity between the dependent variables, we analysed the factors affecting farmers' utilization of three major traditional risk management strategies ( crop diversification, precautionary savings and social networks). The results showed that risk management adoption decisions are indeed correlated (even after controlling for observable factors). Furtherrnore, our analysis suggested that the decision to adopt one risk management strategy positively influences the decision to adopt the other strategies. The results clearly revealed that risk aversion is a strong motivation to farmers' adoption decisions. Furthermore, farm size, household's size, household's head literacy and the engagement in off-farm activities had also been found to incrcase the likelihood of adopting risk management strategies. However, we found the need for risk management

142 strategies to decrease with farming experience and ownership of the cultivated land. Low risk aversion also reduces the probability of adoption a risk management strategy.

> Contribution of the thesis

In addition to its overall contribution which is participating on the ongoing debate about farmers' behavioural responses to risks, the proposed dissertation makes four important contributions:

1. This is a pioneer work in Côte d'Ivoire using an experimental approach to assess individual's behaviour under uncertainty, and thereby testing the validity of some predictions of the expected utility theory.

2. The study provides good estimates of farmers' risk aversion and analyses the determinants of risk aversion. This is a very important variable to consider in the formulation and implementation of rural development policies.

3. The dissertation directly explores the different sources of risks faced by farmers in the Ivorian cocoa sector and points up the most important risks which need to be addressed by development policy, based on farmers' risk perceptions.

4. The study explores the risk management practices in the cocoa sector in Côte d'Ivoire and investigates factors affecting farmers' risk management strategies adoption decisions. The role of risk aversion in farmers' risk management decision making is highlighted.

> Policy recommendations

The effects of risks on rural households' well-being have important implications for development strategies. From the policymaker's perspective, it is essential to understand how to reduce farmers' risk aversion and the riskiness of rural incomes through different interventions. Sorne of these intervention strategies are of long-term concern, whereas some of them are of an immediate necessity to reduce poverty among lvorian cocoa farmers, which is induced by low cocoa prices and high variability of farmers' income. A wide range of intervention stratcgies might be considercd but here we limit our self to options, which follow from the research outcomes of this thesis:

143 1. Taxation in the sector should be reduced toward 20% of fob, in line with other countries and other domestic sectors. This is to increase the farm gate price and thereby increase farmers' incarne. Moreover, an explicit development of coordination mechanisms is necessary to reduce transaction costs and risks of opportunism in the liberalized cocoa chain value.

2. Whether in technology development or for policy formulation, the risk aversion of farmers must be taken into account. For example agricultural extension should start modestly by investments with high probability of success before asking households to take on larger investments involving higher degree of risk in terms of loss in agricultural incarne. Because farmers' level of risk aversion decreases with past successes, this way, we ensure that they will be less reluctant to undertake larger investments.

3. A major focus should be on community education, in order to reduce farmers' risk aversion and equip them with the skills to work in new environments, which subsequently leads to dissemination ofnew farm technologies.

4. A stronger consideration should be given to risk management in development programmes. In dealing with risk management in the Ivorian cocoa sector, the policy maker should start first by addressing the issue of price risk which is ranked as the most important source of risk by farmers. In the short run, this should include institute minimum guaranteed producer prices as before the liberalisation. In the long run, broad based economic development including the development of credit and insurance markets are needed to correct the existing price risk and reduce its negative impacts on different forms offarm investrnent decisions.

5. More attention should be given to making input available to farmers and reinforcing the technical assistance, particularly in the fight against pests/diseases of the cocoa plants.

6. Incarne diversification opportunities inside ( crop diversification) and outside ( off-farm activities) the agricultural sector should be identified and pursued vigorously in order to reduce farmers' dependency on cocoa revenue. Studies of the feasibility of replacing cocoa acreage on an orderly basis, and developing a broader diversification strategy, and an implementation plan, might contribute to a Jess specialized and more robust economy.

144 7. Collecting more and better information: Acting on the best information available is a clear desirability in farming. Rural development strategists should therefore ensure that arrangements are in place for farmers and others to have access to reliable information that will help them do a better job of managing their risks. These arrangements will usually involve a mix of public and private suppliers. Access to better marketing information will be influenced by various infrastructural elements, such as telecommunications and the better functioning of local markets themselves through reduced transaction costs that corne frorn better transport infrastructure.

8. In the long run, an optimal way to improve farmers' capacity to bear risks may be to offer them the opportunity to access futures contracts and options available on the world's cocoa market and commodity exchanges. Therefore, in the short run, it should be important for policy makers to carry out a research on the operational feasibility of this projcct.

);a, Limitations of the study

This section identifies some limitations of this dissertation that, at the same time, indicate interesting directions for future research and extensions. The focus of this dissertation has been the analysis of farmers' behavioural response to risks. This analysis evaluates households' perceptions on each source of risks and then investigates the factors affecting farmers' risk management strategies adoption. But the analysis does not provide information on farmers' perceptions on the different risk management strategies in regards to their efficacity for mitigating each risk factor. For the policymakers, it is important to know not only the factors affecting the choice of a specific risk management strategy by farmers', but also how important or valuable is each strategy for the farmers in terms of its (the strategy) ability to reduce their exposure to risk. This is important in order to irnplement policies that would help some of these farm households to move out of poverty.

Another limitation of this study is related to the debate on providing formai insurance to farmers. The study only investigates the traditional risk management strategies used by farmers to deal with risks. Y et, the interest should be on developing complementary products that will enable poor farmers to better manage agricultural risks, especially by

145 helping them accumulate assets and increase productivity even in the face of shocks. It is therefore imperative to explore the issue of insurance in farming. In particular, it is important to know what type of insurance (price, crop, health, etc ... ) is farmers' interest about and most importantly, what is the potential demand among farmers and which factors drive this demand.

~ Future Research Outlook

Based on the analysis made on this thesis and the limitations pointed out in the previous section, it is warranted to suggest areas requiring further research. A follow-up research is suggested to address the following issues.

Risk and vulnerability to poverty: To obtain a full understanding offarm households' risk attitudes and risk perceptions, it is necessary to complement the findings of this research with an analysis on the nature and degree of household vulnerability to poverty, particularly vulnerability to consumption. Research on risk and vulnerability to poverty in a wider area may enable to identify the group of households that are most vulnerable to risks.

Risk aversion and production efficiency: lt would also be interesting to explore the relationship between risk aversion and the technical efficiency of households in an environment characterized by a high incidence of exogenous shocks and natural risks. This is to investigate the ability of the household to get the maximum amount of output possible, given the level of available inputs and the technology, and taking into account the farmer's degree of risk aversion.

Wi//ingness to pay for agricu/tural insurance: Finally, another possible extension of this research is to assess the potential demand and willingness to pay to benefit from (price - weather) insurance among cocoa farmers in Côte d'Ivoire. This study is necessary to adjust contract sizes according to their needs if one wants to make market-based risk management instruments (options, futures) available to farmers.

146 REFERENCES

Abebaw, D. (2006), "The Economies of Smallholder Coffee Farming Risk and its Influence on Household Use of Forests in Southwest Ethiopia", PhD Dissertation, University of Bonn, Germany. Adebusuyi, B. (2004), "Stabilisation of Commodity Market of Intercst to Africa", Paper Presented at the Workshop on Constraints to Growth in Sub-Saharan Africa, organized by the lntergovemmental Group of Twenty-Four in Pretoria, South Africa. Akcaoz, H. and B. Ozkan (2005), "Determining risk sources and strategies among farmers of contrasting risk awareness: A case study for Cukurova region of Turkey", Journal of Arid Environments, vol. 62: 661-675. Akiyama, T., J. Baffes, D. Larson and P. Varangis (2001), "Market reforms: Iessons from country and commodity experiences", chapter I of Akiyama, T., J. Baffes, D. Larson and P. Varangis, Commodity Market Reforms: Lessonsfrom Two Decades. Washington DC, World Bank. Amemiya, T. (1974), "Bivariate Probit Analysis: Minimum Chi- Square Methods", Journal of the American Statistical Association, vol. 69: 940-44. Anderson, R. and J. Danthine (1980), "Hedging and Joint Production: Theory and Illustrations", The Journal of Finance, vol. 35: 487-498 Antle, J. (1983), "Testing the stochastic structure of production: a flexible moment based approach." Journal of Business and Economie Statistics, vol. 1 (3): 92-201. ------(1987), "Econometric estimation of producers' risk attitudes", American Journal of Agricultural Economies, vol. 69: 509-552. ------( 1989), "Nonstructural risk attitude estimation", American Journal of Agricultural Economies, vol. 71(3): 774-784. Anton, W. (2005) "The Choice of Management Practices: What Determines the Design of an Environmental Management System?" Selected Paper prepared for presentation at the American Agricultural Economies Association Annual Meeting, Providence, Rhode Island. Allais, M. (1953) "Le Comportement de l'Homme Rationnel devant le Risque: Critique des Postulats et Axiomes de l'Ecole Américaine", Econometrica, vol. 21: 503-546. Alderman, H. and C. Paxson, ( 1992), "Do the poor insure? A Synthesis of the Literature on Risk and Consumption in Developing Countries", International Economies Association Moscow, vol.4. Alderman, H., (2008), "Managing risks to increase efficiency and reduce poverty", Background Paper for the W orld Development Report, W orld Bank. Arrow, K. (1965), "Aspects of the Theory of Risk-Bearing", Yrjo Jahnsson Saatio, Helsinki. ------(1971), "Essays in the theory of risk bearing", Amsterdam: North Rolland. Ashford, J. and R. Sowden., (1970) "Multivariate probit analysis", Biometrics, vol. 26: 535- 546 Audibert, M., J-F. Brun, J. Mathonnat, and M. Henry (2006), "Malaria, Production and Income of the Producers of Coffee and Cocoa: an Analysis from Survey Data in Côte d'Ivoire", Etudes et Documents CERDI, E.31. Ayinde, O., O. Omotesho and M. Adewumi (2008), "Risk attitudes and management strategies of small - scale crop producer in Kwara State, Nigeria: A ranking approach" African Journal of Business Management, vol. 2(12): 217-221. Bar-Shira, Z., R. Just and D. Zilbennan (1997), "Estimation for farrners' risk attitude: an econometric approach", Agricultural Economies, vol. 17(4): 211-222.

147 Belaid, A. and F.M. Stanley (1987), "Measuring Farmers' Risk Attitudes: A Case Study of the Eastern High Plateau Region of Algeria", Western Journal of Agricultural Economies, vol. 12(2): 198-206. Binswanger, H. P. and D. A. Sillers (1983), "Risk Aversion and Credit Constraints n Farmers' Decision Making: Reinterpretation", Journal of Development Studies, vol. 20: 5-21. Binswangcr, H. P. (1980), "Attitudes toward Risk: Experimental Measurcmcnt in Rural lndia", American Journal ofA gricultural Economies, vol. 62: 395-407. ------(1981 ), "Attitudes toward Risk: Theoretical Implications of an Experiment in Rural India", The Economie Journal, vol. 91: 867-890. ------(1982), "Ernpirical Estimation and Use of Risk Preferences: Discussion", American Journal ofA gricultural Economies, vol. 64: 391-393. Biodiversity and Agricultural Commodities Program (2010), "Market Transformation Strategy for Cocoa". Boggess, W. G., K. A. Anaman and G.D. Hanson (1985), "Importance, Causes and Management Responses to Farm Risk: Evidence from Florida and Alabama", Southern Journal ofA gricultural Economies, vol. 17(2): 105-116. Bond, G. and B .Wonder (1980), "Risk attitudes amongst Australian farrners", Austra/ian Journal ofA gricultural Economies, vol. 24 (1): 16-34. Bonjean, C.A. and J-M. Boussard (1999), "La stabilisation des prix aux producteurs agricoles: approches micro-économiques" Revue Tiers-Monde, Vol 40 (160): 901 - 928 Bonjean, C.A. and G. Chambas (2001), "Impact du mode d'organisation des filières agro• alimentaires sur la pauvreté: La filière cacao en Côte d'Ivoire", CERDI-Université d'Auvergne, France. Brink, L. and B. McCarl (! 978), "The tradeoff between expected return and risk among combelt farmers", American Journal ofA gricultura/ Economies, vol. 60 (2): 259-263. Brüntrup, M. (2000), "The levels of risk aversion among African farmers - results of a gambling approach", Working paper, Deutscher Tropentag in Hohenheim. Burger, K. (2008), "Optimal export taxes - the case of cocoa in Côte d'Ivoire" Paper prepared for presentation at the 107th EAAE Seminar, Sevilla, Spain. Cappellari, L. and S. Jenkins (2003), "Multivariate Probit Regression using Simulated Maximum Likelihood", The Stata Journal, vol. 3: 278-294. Chavas, J. and M. Holt (1990), "Acreage decisions under risk: the case of corn and soybeans", American Journal ofA gricultural Economies, vol. 72 (3): 529-538. ------(1996), "Economie behaviour under uncertainty: A joint analysis of risk preferences and technology", Review of Economies and Statistics, vol. 78 (2): 329- 335. Collier, P. and J. Dehn (2001) "Aid, Shocks, and Growth", Washington, DC: World Bank Development Research Group, Policy Research Working Paper. No 2688. Cordier, J. (2008), "La gestion des risques en agriculture, de la théorie à la mise en œuvre : éléments de réflexion de l'action politique", Note et Etudes Economiques, vol. 30: 33- 71. Costa, L. (1995), "Pensions and Retirement: Evidence from Union Army Veterans," The QuarterlyJournal a/Economies, vol. 110 (2): 297-319. Chem, W.S., K. Ishibashi, K. Taniguchi and Y. Tokoyama (2003), "Analysis of the food consumption of Japanese households", FAO Economie and Social Development Paper, No. 152, pp. 72. Deaton, A. (1992), "Saving and incarne smoothing in Côte d'Ivoire", Journal of African Economies, vol. 1 (!): 1-24. Deaton, A. and C. Paxson (2000), "Growth and Saving Among Individuals and Households" Review of Economies and Statistics, 82(2): 212-225.

148 Dercon, S. (2002), "Income Risk, Coping Strategies, and Safety Nets", World Bank Research Observer, Vol. 17 (2): 141-166. ------. (2007), "Fate and Fear: Risk and its consequences in Africa", Working Paper, Oxford University. Dercon. S.; L. Christiaensen (2007), "Consumption risk, technology adoption and poverty traps: evidence from Ethiopia", World Bank Policy Research Working Paper No. 4257, Washington, D.C. Dercon, S., T. Bold and C. Calvo (2008), "Insurance for the poor?" University of Oxford Working Paper No. 125. Dillon, J. and P. Scandizzo (1978), "Risk attitudes of subsistence farmers in northeast Brazil: a sampling approach", American Journal of Agricultural Economies, vol. 60 (3): 425- 435. Dixit, A.K., R.S. Pindyck (1994), "Investment under uncertainty," Princeton University Press, Princeton, NJ. Elliot, D. (2003), "Commodities Report: Cocoa Prices could go Higher if Dry Conditions continue", Wall Street Journal. New York: Eastern Edition, Cl3. European Union (2005), "Etude des Prélèvements sur la Filière Café et Cacao en Côte d'Ivoire", Amsterdam: Nederlands. Ezeh, C. and J. Olukosi (1991), "Farmers' perception of Risks and Their Responses in Dry Season Farming: A Study of the Kano River Project". The Nigerian Journal of Agricultural Extension, vol. 6(1 & 2): 12-17. Fafchamps, M. (1999), "Rural Poverty, Risk, and Dcvelopment", Food and Agriculture Organization, Social Development Paper No 144. Fafchamps, M., and S. Lund (2003), "Risk-sharing Nctworks in Rural Philippines." Journal ofDevelopment Economies, vol. 71(1): 261-287. Feinennan, E. and 1. Finkelshtain (1996), "Introducing socio-economic characteristics into production analysis under risk", Agricultural Economies, vol. 13 (2): 149-161. Francisco, E. and J. Anderson (1972), "Chance and choice west of the darling", Australian Journal ofA gricultural Economies, vol. 16 (!): 82-93. Gibson, J. (2007), "Consistently In consistent: Addressing income volatility arnong coco a producers in Ghana and Côte d'Ivoire", International Institute for Sustainable Development (USD), Working paper. Gilbert, C. L. (1997), "Cocoa Market Liberalization: Its Effects on Quality, Futures Trading and Prices", The Cocoa Association of London. Gilbert, C. L. and J. ter Wengel (2001), "The production and marketing of primary commodities", in Common Fund for Commodities, Commodities and Development at the Turn of the Millennium, Amsterdam, Common Fund for Commodities, 21-62. Gilbert, C. L. and P. Varangis (2003), "Globalization and International Commodity Trade with Specific Reference to the West Afiican Cocoa Producers", NBER Working Paper No. W9668. Gillespie, J., C. Davis and N. Rahelizatovo (2004), "Factors Influencing the Adoption of Breeding Technologies in U.S. Hog Production", Journal of Agricultural and Applied Economies, vol. 36: 35-47. Glewwe, P. and H. Jacoby, (1995), "An Economie Analysis of Delayed Primary School Enrollment in a Low Income Country: The Role of Early Childhood Nutrition," The Review ofEconomics and Statistics, vol. 77 (!): 156-169. Goodwin, B., and T. Schroeder (1996), "Human Capital, Producer Education Programs, and the Adoption of Forward Pricing Methods", American Journal of Agricultural Economies, vol. 76: 936-47.

149 Guerriere, A. (2005), "Cocoa Prices Soar On News of Unrest from Ivory Coast", Wall Street Journal. New York: Eastern Edition. Greene, W. (2003), "Econometric Analysis", 5th cd. Upper Saddlc River, NJ: Prcnticc Hall. Hardaker, J., B. Huirne, and J. Anderson (1997), "Coping with Risk in Agriculture", New York: CAB International. Hama!, K. and J. Anderson (1982), "A note on decreasing absolute risk aversion among farmers in Népal", Australian Journal of Agricultural Economies, vol. 26 (3): 220-225. Harrison, G. and J. List (2004), "Field Experiments," Journal of Economie Literature, vol. 42(4): 1013-1059. Harwood, J., R. Heifner, K. Coble, J. Perry, and A. Somwaru (1999), "Managing Risk in Farming: Concepts, Rescarch and Analysis", Agricultural Economie Report No. 774. Market and Trade Economie Division and Resource Economies Division, Economie Research Service U.S. Department of Agriculture. Henrich, J. and M. Richard (2002), "Arc Pcasants Risk-Averse Occision Makers?" Cun·ent Anthropo/ogy, vol. 43(1). Holden, D., P. Hazell, and A. Pritchard (1991), "Risk in Agriculture", Proceedings of the Tenth Agriculture Sector Symposium, World Bank: Washington, D.C. Holden, S., B. Shiferaw and M. Wik (1998), "Poverty, market imperfections and time preferences: of relcvance for environmental policy?" Environment and Development Economies, vol. 3: 105-130. Holt, C. and K. Laury (2002), "Risk Aversion and Incentive Effects", American Economie Review, vol. 92(5): 1644-1655. Hope, J. and J. Lingard (1992), "The Influence of Risk Aversion on the Uptake of Set-Aside: A MOTAD and CRP Approach". Journal of Agricultural Economies, vol. 43(3): 401- 411. ICCO (2006) "Assessment of the movements of global supply and demand" Document of the market committee. MC/6/4. Jehle, G. and P. Reny (2001), "Advanced Microeconomic Theory", second edition, Boston: Addison Wesley. Kagel, J. and A. Roth (1995), "The Handbook of Experimental Economies", Princeton University Press. Kahneman, D. and A. Tversky, (1979), "Prospect theory: An analysis of decisions under risk", Econometrica, vol. 47: 313-327. Kiefer, N. ( 1982): "Testing for Dependence for Multivariate Probit Models." Biometrika, vol. 69: 161-166. Knight, T., A. Lovell, M. Rister, and K. Coble (1989), "An Analysis ofLenders' Influence on Agricultural Producers' Risk Management Decisions", Southern Journal of Agricu/tural Economies, vol. 21: 21-33. Lichtenstein, S., and P. Slovic (1971), "Reversais ofpreference between bids and choices in gambling decisions", Journal of Experimental Psychology, vol. 89: 46-55. Lins, D. A., S. C Gabriel and S. T. Sonka (1981), "An analysis of the risk aversion offarm operators: an asset portfolio approach", Western Journal of Agricultural Economies, vol. 6 (!): 15-29. LMC International (2001), "The World Cocoa Market Outlook", Ghana Conference Paper, 3- 29. Long S. J. and J. Freese (2006), "Regression Models for Categorical Dependent Variables Using Stata", 2nd edition, College Station, Stata Press. Losch, B. (2002), "Global Restructuring and Liberalization: Côte d'Ivoire and then End of the International Cocoa Market?", Journal ofA grarian Change, vol. 2(2): 206-227.

150 Machina, M. (1982), "Expected Utility Analysis without the Independence Axiom", Econometrica, vol. 50(2): 277-323. Maddala, G. S. (1983), "Limited-Dependent and Qualitative Variables in Econometrics", Cambridge et al: Cambridge University Press. Mahul O. and C. Stutley (2010), "Public support to agricultural insurance: challenges and options for developing countries", The World Bank. Washington DC. Makus, L., B. Lin, J. Carlson and R. Krebill-Prather (1990), "Factors Influencing Farm Level Use of Futures and Options in Commodity Marketing", Journal of Agribusiness, vol. 6: 621-631. Martin, S. ( 1996), "Risk management strategies in New Zealand agriculture and horticulture" Review of marketing and agricultural economics, vol. 64 ( 1). Mas Colell, A., M. Whinston, and J. Green, ( 1995) "Microeconomic Theory", Oxford University Press. McGranahan, L. (2000), "Charity and the Bequest Motive: Evidence from Seventeenth Century Wills," Journal of Political Economy, vol. 108 (6): 1270-1291. Mclntire, J. and P. Varangis (1999), "Reforming Côte d'Ivoire's cocoa marketing and pricing systems", World Bank Policy Research Working Paper 2081. Washington, DC. Menezes, C.F. and D.L. Hanson (1070), "On the theory of risk aversion", International Economie Review, vol. 11: 481-487. Meuwissen, M, R. Huime, J.B. Hardaker (2001), "Risk and risk management: an empirical analysis of Dutch livestock farmers", Livestock Production Science, vol. 69: 43-53. Mishra, A., and H. El-Osta (2002) "Managing Risk in Agriculture through Hedging and Crop lnsurance: What Does a National Survey Reveal?" Agricultural Finance Review, 62:135-48. Morduch, J. (1995), "Income smoothing and consumption smoothing", Journal of Economie Perspectives, vol. 9: 103-114. ------(2002), "Consumption smoothing across space: Testing theory of risk sharing in the ICRISAT study region of South lndia", Research paper, World Institute for Development Economie Research (UNU-WIDER). Morimune, K. (1979), "Comparisons of normal and logistic models in the bivariate dichotomous analysis", Econometrica, vol. 47: 957-976. Moscardi, E. and A. de Janvry (1977), "Attitudes toward risk among peasants: an econometric approach", American Journal ofA gricultural Economies, vol. 59 (4): 710-716. Moschini, G. and D. Hennessy (2001), "Uncertainty, risk aversion, and risk management for agricultural producers", Chapter in Handbook of Agricultural Economies, vol. lA ( B.Gardner and G.Rausser) Elsevier Science, 87-153. Mund, J. (1999), "Rice production on inland-valleys soils (Bas-Fonds) of the south-west Ivory Coast - agroecological conditions and specific pedological determinations", Sustainable Technology Development in Crop Production, Berlin Session, http://www.staff.unimainz.de/mund/ Abstractsff ropentag-99 .pdf Myers, R. ( 1989), Econometric testing for risk averse behaviour in agriculture, Applied Economies, vol. 21(4): 541-552. Nakamura A. and M. Nakamura (1998), "Madel specipication and endogeneity", Journal of. Econometrics, vol. 83: 213-237 Nkamleu, G., Y. Keho, J. Gockowski and S. David (2007), "Investing in agrochemicals in the cocoa sector of Côte d'Ivoire: hypotheses, evidence and policy implications", African Journal of Agriculture and Resource Economies, vol. 2(1): 145-166 Pennings, J. and R. Leuthold (2000), "The role of farmers' behavioural attitudes and heterogeneity in futures contracts usage", American Journal of Agricultural Economies, vol. 82(4): 908-919.

151 Pennings, J. and P. Garcia (2001), "Measuring producers' risk preferences: A global risk attitude construct", American Journal ofA gricultural Economies, vol. 83: 993-1009. Pennings, J., O. Isengildina-Massa, S. Irwin and D. Good (2008), "Produccrs' Complcx Risk Management Choices", Agribusiness, vol. 24 (!): 31-54. Pesaran, H. and Y. Shin (1995), "Autoregressive distributed lag modelling approach to cointegration analysis", DAE Working Paper Series, No. 9514, Department of Economies, University of Cambridge. Pesaran, H., Y. Shin and R. Smith (2001), "Bounds testing approaches to the analysis oflcvcl relationships", Journal ofA pplied Econometrics, vol. 16: 289-326. Pleitez-Chavez, R. (2004), "Rernittances as a Strategy to Cope with Systemic Risk: Panel Results from Rural Households in El Salvador", unpublishcd Ph.D. dissertation, Columbus, Ohio: The Ohio State University. Pope, R., R. Just (1991 ), "On testing the structure of risk preferences in agricultural suppl y analysis", American Journal ofA gricultural Economies, vol. 73 (5): 743-748. Pratt, J. (1964), "Risk Aversion in the Small and in the Large", Econometrica, vol. 32: 122- 136. Quiggin, J. (1993), "Testing between Alternative Models of Choice under Uncertainty: Comment", Journal of Risk and Uncertainty, vol. 6: 161-164. Rabin, M. (2000), "Risk aversion and expected-utility theory: A calibration theorem", Econometrica, vol. 68(5): 1281-1292. Rabin, M. and R. Thaler (2001), "Anomalies: Risk aversion", Journal of Economie Perspectives, vol. 15(1): 219-232. Raskin, R., M. Cochran, (1986), "Interpretations and transformations of scale for the Pratt• Arrow absolute risk aversion coefficient: implications for generalized stochastic dominance", Western Journal ofA gricultural Economies, vol. 11 (2): 204-21 O. Reardon, T. and S. Vosti (1995), "Links between rural poverty and the environment in developing countries: Asset categorics and investrnent poverty'', World Deve/opment, vol. 23(9): 1495-1506. Rivers, D and Q. Vuong (1988), "Limited Information Estimators and Exogeneity Tests for Simultaneous Probit Models," Journal ofE conometrics, vol. 39: 347-366. Rosenzweig, M. and H. Binswanger (1993), "Wealth, Weather Risk and the Consumption and Profitability of Agricultural Investments", The Economie Journal, vol. 103(416): 56-79. Ruf, F. and J-L. Agkpo (2008), "Etude sur les revenus et les investissements des producteurs de café et de cacao en Côte d'Ivoire", Rapport pour l'Union Européenne, Agrisystems Consortium, Abidjan, 90 p. Sadoulet, E. and A. de Janvry (1995), "Quantitative Development Policy Analysis", Johns Hopkins University Press, Baltimore. Saba, A., C. Shumway and H. Talpaz ( I 994 ), "Joint estimation of risk preference structure and technology using expo-power utility", American Journal of Agricultural Economies, vol. 76 (2): 173-184. Salimonu, K. and A. Falusi (2009), "Sources of risk and management strategies among food Crop farmers in osun state, Nigeria" African Journal of Food, Agriculture, Nutrition and Development, vol. 9: 7. Sandmo, A. (1971), "On the theory of the competitive firm under price uncertainty", American Economie Review, vol. 61: 65-73. Schneider, P. (2004), "Why should the poor insure? Theories of decision-making in the context ofhealth insurance" Health Policy and Planning, vol. I9(6): 349-355 Schoemaker, P. and H. Kunreuther (1979), "An experimental study of insurance decisions", Journal ofRisk Jnsurance, vol. 46 (4): 603-618.

152 Shapiro, B., and B. Brorsen (1988), "Factors AfTecting Farmers' Hedging Decisions." North Central Journal ofA gricultural Economies, vol. 10: 145-53. Sillers, D. A. (1980), "Measuring Risk Preferences of Rice Farmers in Nueva Ecija, Philipines: An Experimental Approach", Ph.D. Dissertation, Yale University. Slovic, P. and S. Lichtenstein (1968), "The Relative Importance of Probabilities and Payoffs in Risk Taking", Journal of Experimental Psychology Monographs, 78, p. 1-18. Starmer C. (2000), "Developments in Non-Expected Utility Theory: The Hunt for a Descriptive Thcory of Choicc undcr Risk", Journal of Economie Litera/ure, vol.38: 332-382. Tavernier E. and B. Onyango (2008), "Utilization of farm management risk strategies al the rural/urban fringc", African Journal ofA gricultural Research, Vol. 3(8): 554-565. Tomek, W., and H. Hikaru (2001), "Risk Management in Agricultural Markets: A Review". The Journal of Futures Markets, vol. 21(10): 953-985. Townsend, R. (1995), "Consumption Insurance: An Evaluation of Risk-Bearing Systems in Low Income Economies", The Journal ofE conomie Perspectives, vol. 9(3): 83-102. Tversky A. and D. Kahneman (1992), "Advances in Prospect Theory: Cumulative Representation of Uncertainty", Journal ofRisk and Uncertainty, vol. 5: 297-323. Udry, C. (1994), "Risk and Insurance in a Rural Credit Market: An Empirical Investigation in Northern Nigeria", The Review of Economie Studies, vol. 61(3): 495-526. ul Hague, I. (2004), "Commodities under Neoliberalism: The Case of Cocoa", UNCTAD, Discussion Paper Series, No 25, 35 p. UNCTAD (2005), "Commodity Policies for Development: A New Framework for the Fight Against Poverty.", Background note by the UNCTAD secretariat for a meeting of the Commission on Trade in Goods and Service and Commodities, Geneva, 6-10 February 2006. Varian, Hal R. (1993), "Intermediate Microeconomics", W.W.Norton and Co. Velandia M., R. Rejesus, T. Knight and B. Sherrick (2009), "Factors Affecting Farmers' Utilization of Agricultural Risk Management Tools: The Case of Crop Insurance, Forward Contracting, and Spreading Sales" Journal of Agricultural and Applied Economies, vol. 41(1): 107-123. Von Neumann, J. and O. Morgenstern (1947), "The Theory of Game and Economie Behaviour", 2nd Ed, Princeton University Press. Wiens, T.B. (1976) "Peasant risk aversion and allocative behaviour: A quadratic programming experiment", American Journal of Agricultural Economies, vol. 58(4): 629-635. Wik, M., and S. Holden (1998), "Experimental studies of peasants' attitudes toward risk in Northern Zambia", Discussion Paper D-14. Department of Economies and Social Sciences. As, Norway: Agricultural University ofNorway. Wilcox, M. D. and P.C. Abbott (2004), "Market Power and Structural Adjustment: The Case of West African Cocoa Market Liberalization", Selected Paper prepared for presentation at the American Agricultural Economies Association Annual Meeting, Denver, Colorado. Wooldridge, J. (2002), "Econométrie Analysis of Cross-Section and Panel Data", MIT Press, Cambridge, MA. Yaari, M. (1987), "The Dual Theory ofChoice under Risk", Econometrica, vol. 55: 95-105. Yang, D. and H.J. Choi (2007), "Are Remittances Insurance? Evidence from Rainfall Shocks in the Philippines", The World Bank Economie Review, forthcoming. Yesuf, M. (2003), "Time preferences of farm households in Ethiopia", PhD Dissertation, Department of Economies. Gothenburg, Sweden: Gothenburg University.

153 ------(2007), "Risk Aversion in Low Income Countries: Experirnental Evidence from Ethiopia", International Food Policy Research Institute, Discussion Paper 715, Environment and Production Technology Division. Yesuf, M., M. Kassie, and G. Kôhlin (2009), "Risk Implications of Farm Technology Adoption in the Ethiopian Highlands" Environment for Development Discussion Paper 09-13. Young, D. (1979), "Risk preferences of agricultural producers: their use in extension and research", American Journal of Agricultural Economies, vol. 61 (5): 1063-1070. Zeckhauser, R. and E. Keeler (1970), "Another Type of Risk Aversion," Econometrica, vol. 38(5): 661-065.

154 APPENDICES

Appendix 1: The prospect theory (Kahneman and Tversky)

Figure A.1: Hypothetical value fonction and Weighting fonction

Value v( :r)

-x X

Losses

0 p* 1

155 Appendlx 2: The study area

n.11•1(M /I

:.11onn.11 /

•

156 Appendix 3: Cocoa production in Côte d'Ivoire

~ ------•.•.. ------• ----~------•.•.•...• --n• •.•. -• •.... •• ••• • ---~ Regions Production ner 1 ear of camnaisn Departments 2001/2002 2002/2003 2003/2004 2004/2005 2005/2006 2006/2007 Abenzourou 81 122 86 634 35 049 33 048 36 098 41 571 MOYENCOMOE Aznibilékrou 0 0 4 572 2 125 3 804 5 805 TOTAL MOYEN COMOE 81 122 86634 39 621 35173 39902 47 376 Azboville 12 026 12 852 18 565 16 905 19 589 19 547 AGNEBY Adzoné 24 811 26 511 37 894 28 629 41 274 36 571 TOTAL AGNEBY 36837 39 363 56 459 45534 60 863 56118 Bondoukou 2 881 3 078 252 57 13 0 ZANZAN Tanda 5 856 6 258 678 489 1 108 1 253 TOTAL ZANZAN 8 737 9336 930 546 1 121 1253 Abidian 19 233 20 553 25 552 40986 13 388 5 347 Aléné 0 0 lO 361 7 768 10464 7465 LAGUNES Dabou 0 0 18 138 5 777 17 344 12 989 Grand Labou 6474 6 918 19669 Il 662 7 985 16 151 Tiassalé 14612 15 615 19460 27 325 20 571 20 794 TOTAL LAGUNES 40 319 43 086 93180 93 518 69 752 62 746 Aboisso 30067 32 131 15 602 9 371 20 036 18 144 SUDCOMOE Adiaké 0 0 238 128 582 154 Grand Bassam 0 0 2 179 2496 3 832 2 984 TOTAL SUD COMOE 30067 32 131 18019 11 995 24450 21282 Béoumi 71 76 0 23 28 0 V ALLEE DU BANDAMA Sakassou 68 73 0 0 0 TOTAL VALLEE DU BANDAMA 139 149 0 23 28 0 Bonzouanou 12614 13480 4946 6633 4 533 4 583 Daoukro 7 346 7 850 2937 2 341 926 1923 N'ZICOMOE Dimbokro 282 301 0 27 52 21 M'Bahiakro 295 315 30 59 82 66 TOTAL N'ZI COMOE 20537 21946 7 913 9060 5 593 6593 Toumodi 3 691 3 944 3 161 2 131 2474 3 043 LACS Tiébissou 0 0 21 0 Il 58 Yamoussoukro 6 498 6945 5 914 3 099 5 253 4 569 TOTAL LACS 10189 10 889 9 096 5 230 7738 7670 Bouaflé 25 229 26962 21978 25 946 24 990 20 187 MARAHOUE Sinfra 44 388 47436 35 946 30 858 37 000 31 826 Zuénoula 1 775 1 897 229 95 397 542 TOTAL MARAHOUE 71392 76295 58153 56899 62387 52 555 Table A2. l continued over page

157 Table Al.! continued Regions Production per year of campaign Departments 2001/2002 2002/2003 2003/2004 2004/2005 2005/2006 2006/2007 Daloa 121 269 129 596 80 915 78 570 77 400 75 787 HAUT SASSANDRA Issia 82 139 87 779 96 353 69 974 79 970 37 079 Vavoua 92 607 98 966 891 525 36 232 TOTAL HAUT SASSANDRA 296 015 316 341 178159 149 069 157 406 113 098 WORODOUGOU Sézuéla 418 447 34 0 0 0 TOTAL WORODOUGOU 418 447 34 0 0 0 Divo 85 885 91 782 126512 113 369 125 705 118216 SUDBANDAMA Lakota 30 875 32 995 16467 17 764 18 114 22 194 TOT AL SUD BAN DAMA 116 760 124 777 142 979 131133 143 819 140 410 Gagnoa 75 967 81 183 102 587 76 212 89 929 70 521 FROMAGER Oumé 45 422 48 540 36 574 39 526 42 Oil 34 172 TOTAL FROMAGER 121389 129 723 139161 115 738 131940 104 693 Man 10 383 11 096 52 162 228 34 Bangolo 8 285 8 854 4 012 6 549 7 486 10 348 MONTAGNES Biankouma 305 326 0 0 0 Danané 17 236 18 419 1 303 1 174 388 42 TOTAL MONTAGNES 36 209 38 695 5 367 7 885 8102 10424 Duékoué 55 272 59 067 74 751 87 560 56 924 64648 MOYEN CA V ALLY Guiglo 71 309 76 205 18 669 25 080 24276 26 643 Toulépleu 506 541 71 252 0 TOTAL MOYEN CA V ALLY 127 087 135 813 93 491 112 892 81200 91291 San Pédro 95 309 101 853 236 594 265 619 239 402 187 852 Sassandra 19 840 21 203 21 723 19 580 35 725 33 992 BAS SASSANDRA :ci\'( ::.:< ,_:\:)'... ~--- 1.., -c:,:c>• ::c::, _,,·::tffrgi~Y?ti l""'V-!U'.I\""' l;iiJFJIJ..(.41\E · 1S2°3iS ;; lff3bl"349} :x2:rs,~1ô;*11I ,,. . ,~' L/.<ë# •Ii1;s)l~ lt11 .<218.915 Tabou 9 872 10 550 4 955 7 826 10 059 13 640 TOTAL BAS SASSANDRA 267 491 285 921 564 621 511635 614 545 514 399 TOT AL COTE D'IVOIRE 1264 708 1351546 1407 213 1286330 1408 854 1229 908 Source: Ministère de l'Agriculture - Direction des statistiques, de la documentation et de l'informatique (2008)

158 Table A.2: Details of the survev areas """~.;-:' .. /.•;,ia:Eni:amnmen& Buvo 1 Gblialo /V3) 1 Alexandrekro; Aboukro· Léonkro· Kouamekro; Gblialo V3 Bonikro; Bakaridougou; N Drikro; Zouzoukro; N Goran N Drikro; Buyo 1 Kodaya 2 Rémikro Dafibougou; Yaokankro; Petit Katiola;; Yaokro; Kambourkro; Buyo 1 Tchétaly 3 Aldomiabla; Yacoubakro· Tchétalv Salif; Gouabouo 1; Kcffikro; Kouamekro; Akakouamékro; Kouakoukro: 2 Gouabouo 4 Kouadiokro; Kaborédougou 2 Zakoeoua s Zakoeoua Il ,_.,..,,~ Ettiennkro; Konédougou; Remykro; Lessiri; Alberkro; Jackro; Koda Liliyo 3 Lessiri 6 Centre Kouadiokro; Dékayo; Anzoua Koffîkro; Dramanekro; Kouakou Blékro; Okrouyo 4 Bogreko 1 7 Kouamékro; Raphaelkro 1 Germainkro; Jbkro; Kouassikankro; Bertinkro; Juleskro; Ngatta Okrouyo 4 Mabéhiri 8 Kouassikro; Assantikro Mafoukro; Djakro; Konankro; Galléa; Amonkro; Dioulabougou; Soubré s Galléa 9 Miangabougou Soubré s Savo 10 Savo;Konankro;Kouakoukro;Djèkro ' 'li/il!), ,•.; '' - i)\]l[;,l;hll Dozo ,.. __ r,_ ____ v -----.t.1---· "' Méagui 6 Gnititouadji 2 11 Kouad Brahimakro; Ladjikro; Kouassikro; Yaokro; Sakiare; Marcel Koro; Koffi Méagui 6 Sakiaré 12 Amonkro Kiemde Dougou; Kouadiokro; Yaokro; Attakro; Touadji 2; Méagui 6 Touadji 2 13 Douanierbrokro Gnankoradji; Yao Carrefour; Vieux Père Gendarme; Cpt De Jean; Gblétia; Oupoyo 7 Gblétia 14 Kouassikro Oupoyo 7 Walebo 15 Loukoukro; Kouadiokouamékro; Yao Blékro; Bobokro; Jeromekro

159 Appendix 4: Measuring farmers' level of risk aversion

Table A.3: Risk Preferences games

Bad Good Expected Standard Risk classification Choices Outcome Outcome Gain deviation "Heads" "Tails" =E =SE 0 100 100 100 0 Extreme A 90 180 135 45 Severe B 80 240 160 80 lntennediate C 60 300 180 120 Moderato D 20 380 200 180 Slizht to Neutra] E 0 400 200 200 N eutral to Prcfcrring

.. ~ · ~ · , Bad Good Expected Standard Risk classification Outcome Outcome Gain deviation Choices "Heads" "'Talls" =E =SE 0 1000 1000 1000 0 Extreme A 900 1800 1350 450 Severe B 800 2400 1600 800 lntennediate C 600 3000 1800 1200 Moderato D 200 3800 2000 1800 Slight to Neutra! E 0 4000 2000 2000 Neutra! to Preferrina uame .3: :,uuu rl.-rl\ zarne wnn zarn-omv invnomenca1J Bad Good Expected Stand~rd Risk classification Cbolces Outcome Outcome Gain devlation "Heads" "Tails" =E =SE 0 5000 5000 5000 0 Extreme A 4500 9500 7000 2500 Severe B 4000 12000 8000 4000 Intennediate C 3000 15000 9000 6000 Moderate D 1000 19000 10000 9000 Slight to Neutra( E O 20000 10000 10000 Neutra( to Preferrinz

Bad Good Expected Standard Risk classification Choices Outcome Outcome Gain devlation "Heads" "Tails" -E -SE 0 0 0 0 0 Extrcmc A -10 80 35 45 Severe B -20 140 60 80 lntennediate C -40 200 80 120 Moderate D -80 280 100 180 Slizht to Neutra! E -100 300 100 200 Neutra! to Preferrinz

160 Table A.4: Games Posten

Jeu de 100 FCFA Choix Mauvais Résultat Bon Résultat

Loterie 1 0

Loterie 2 A

Loterie] B

Loterie4 C

Loterie S D

Loterie 6 E 0

161 Jeu de 1000 FCFA Choix Mauvais Résultat Bon Ruultat "Pile" "Face"

Loterie 1 0

Loterie 2 A

Loterie 3 B

1000 Loterie 4 C

Loterie S D

Loterie 6 E 0

162 Jeu de 5000 FCFA Choix Mauval1 Résultat Bon R61ultat "Plie" "Face" _,,""' Loterie 1 0

., ,,,.•

Loterie 2 A

Loterie 3 B

1000 Loterie 4 C

1000 Loterie S D

10000 Loterie 6 0 E

163 Jeu avec Perte

Choix Mauvais Résultat Bon Résultat "Pile~ PERTE "Face" c:: :::> GAIN

Loterie 1 0 0 0 PERTE GAIN

PERTE Loterie 2 A GAIN

PERTE Loterie3 B

GAIN PERTE

Loterie4

GAIN

PERTE

Loteries D

GAIN PERTE

Loterie 6 E

GAIN

164 Appendix 7: Predicted probabilities Predlcted Probabllilles by Age (Gam!_1l Predlcted Probabilities by Value of liver:11:ock (Game 1) 0.4 •.. ~ • ,!! 03 i 03 :0 ~m .l! ê (? O.. 02 O.. 02 i i l"_ 0.1 l"_ 0.1 ~- · ~lc:jl ::alc:jl:~l::a:-S=t--- - ~- .. on •••• on •• Ill Ill "' ---- •••.•.. . 1:== --- • =I Predicted ProbablllHes by Age (Game 2) Predlcted Probabllllies by Value of llvelilock (Game 2) 0.4 OA

i 0.3 i 0.3 "'m i ê ê ~ 0.2 o.. 02 jl i " l 0.1 l"_ 0.1

0.0

Ill 4000000 8000IX)O 8000IX)O •• e!ii• "' "' ---- •.... Yaœdlwaltd(ln FCFA) I:==.. . -- •..•... . =I

175 Computation the upper and Iower limits for the Constant Partial Risk Aversion Coefficient

The Iimits are determined by the value of S (the partial risk aversion coefficient) for which the expected utility of a prospect G is equal to that of prospect G-1 (highest possible S) and G+ 1 (lowest possible S). In this case where there are only 2 possible outcornes (since coin toss); the expected utility of prospect G is sirnply:

E (G) = O.S*[utility (good outcome) + utility (bad outcome))

Thus, using the CPRA utility function of the form: U = (1- S)cci-s, , we can derive the range of the CPRA coefficient for each prospect in the 100 FCFA game with gain only as follows:

1- Equality between prospect O and A is only achieved when S=7.47, since only for this value, will it be true that:

Expected utility of prospect O = Expected utility of prospect A

This implies solving:

0.5[ (1-S)IOO(I-S) + (I-S)I00°-SJ J = 0.5[ (1-S)180(l-S) + (I-S)90(l-S) J

(I-S)[10011-S> + 100 + 90°-S)J

1 ---- 1 : 1 oo(l-S) + 1 oo : '------· We solve the last equation for S using a standard numerical method in Excel and obtain S = 7.47. Hence the lowest possible S for anyone choosing alternative Ois 7.47. We know this person is at least as risk averse as that.

2- For equality between prospect A and prospect B, S = 2.00 is obtained by solving:

Expected utility of prospect A = Expected utility of prospect B

This implies solving:

: J 80(1-S) + 90(!-S) = 240(1-S) + 80(1-S) : L------1

165 3- For equality between prospect Band prospect C, S = 0.85 is obtained by solving:

Expected utility of prospect B = Expected utility of prospect C

This implies solving:

1 1 : 240(1-S) + so

4- For equality between prospect C and prospect D, S = 0.32 is obtained by solving: Expected utility of prospect C = Expected utility of prospect D

This implies solving:

1 1 : 3QQ(I-S) + 60(1-S) = 38Q(I-S) + 20(!-S) : '------·

5- For equality between prospect D and prospect E, S = 0 is obtained by solving: Expected utility of prospect D = Expected utility of prospect E

This implies solving:

1 1 : 38Q(I-S) + 20(1-S) = 400(!-S) + Ü(I-S) : '------·

Note: Because other game levels (game 2 and game 3) are obtained by multiplying the outcomes in game 1 (100 FCFA game) by a constant factor, the upper and lower limits of the CPRA coefficients computed for game 1 are the same for ail game levels.

166 Appendix 5: Testing for endogeneity of "farm income"

A - The procedure

Rivers and Vuong (1988) and Nakamura and Nakamura (1998) developed a two-step instrumental approach to test for the presence of endogeneity in discrete choice models.

Consider the following discrete choice model:

(A. 1) y,=I [J.°>0] Y, =0 [y,' <0]

Where y,' is the unobserved latent variable capturing the fanner's degree of risk aversion (

Avers), X, is a set of exogenous variables (presented in Table 3.3), y2 is the continuons variable suspected to be endogenous, that is farm income (Inc) and µ, is the error term.

Considering the ordered nature of the dependent variable, degrees of risk aversion ( Avers ), ranging from m = I (extreme risk aversion) tom= 6 (risk Ioving behaviour), Equation (A.I) which is the structural equation can then be rewritten as follow:

(A. 2) Avers= m [ À.m_, ~Avers'< À.]

First step: the potential continuons endogenous variable is regresscd (OLS) on ail exogenous variables X, plus an instrumental variable. The instrument we use for fann income is the

total number of farm workers; that is the total labour force (X,). The potential endogenous variable is then modelled as:

(A. 3)

and the residuals û, are saved. The linear projection in equation (A.3) is called a reduced form equation for the endogenous explanatory variable Inc.

167 Second step: the first stage residuals û, are included as regressors the structural cquation (A.2) and we obtain the augrnented equation as follows:

(A. 4)

To derive the procedure, Rivers and Vuong (1988) first show that µ, =01u2 +e1• We recall that the essence of the endogeneity problem is the correlation between the explanatory

• variable Inc and the error term µ1 An advantage of the Rivers and Vuong (! 988) two-step

approach is that the t statistic on Û2 is a valid test of the null hypothesis that Inc is

: exogenous, i.e., H0 0, = O.

: "# If H0 81 0, there is evidence of endogeneity of Inc and this problem can be corrected by replacing Inc in the structural equation by its predicted value obtained from the first-stage,

168 B - The results of the endogcncity test at ail game levels

Game level 1: 100 FCF A game

X1 = (Sex,Age,Education,HH _Size,Val _Livestock,Native,Migrant2,Mat _Status, Dep _ Cocoa, Zone 1, Zone2)

X2 = Farm_Labour

Table A.5: Ordinary Least Square Estimation results of Income Dependent variable - Farm lncome Explanatory Variable coefficients Standard error P>t

Instrument Total farm labour 345939.6*** 39464.76 0.000

Exogenous variables Sex (Male) 189940.4 322621.2 0.556 Age 491.667 6164.065 0.936 Education -17435.22 141911 0.902 Household size 35924.69*** 9943.104 0.000 Value of livestock -.1164342 .0814315 0.154 Native -223672.8 169696.8 0.188 Migrant2 151721.4 150852.1 0.315 Matrimonial status 13292.03 186392.2 0.943 Dependence Cocoa -145201.8 134076.4 0.280 Zone) -470278.5*** 149725.6 0.002 Zonc3 78622.66 155649.2 0.614 constant 464386.9 413906.l 0.263

Number of obs =362 F-stat = 12.81 R-squared = 0.3057 Prob > F =0.000

Note:***,**,* indicate significance lcvels at l %, 5%, and 10% levels, respcctively.

169 Game level 2: 1000 FCFA game

X,= (Luckl,Sex, Age.Education.Hll _Size, Val _Livestock,Native,Migrant2, Mat_ Status, Dep _ Cocoa, Zone!, Zone2)

X2 = Farm _ Labour

Table A.6: Ordinary Least Square Estimation results of Income Dependent variable - Farm Income Explanatory Variable coefficients Standard error P>t

Instrument Total fann labour 347504••• 39513.2 0.000

Exogenous variables Luckl -59089.72 65435.3 0.367 Sex (Male) 198230.8 322837.1 0.540 Age 281.0911 6170.103 0.964 Education -23762.38 142121.4 0.867 Household size 36525.61 *** 9967.971 0.000 Value oflivestock -.1181693 .0814757 0.148 Native -240719.8 170788.1 0.160 Migrant2 138503.4 151600.3 0.362 Matrimonial status 10133.34 186474.3 0.957 Dependence Cocoa -149088.2 134180.9 0.267 Zonel -450561.8*** 151348.4 0.003 Zone3 92587.18 156456.4 0.554 constant 468421.3 414039.7 0.259

Number of obs =362 F-stat = 11.88 R-squared = 0.3073 Prob > F = 0.000 Note: •••, • •, • indicate significance levels at 1 %, 5%, and 10% levels, rcspectively.

170 Game level 3: 5000 FCFA game

Reduced form cquation: Inc= 821X, + 8,,X, + v2

X, = (Luck12, Sex, Age, Education, HH _ Size, Val_ lives Iock, Native.Migrant'L, Mat _ Status, Dep _ Cocoa, Zone!, Zone2) X, = Farm _ Labour

Table A. 7: Ordinary Least Square Estimation results of Income Dependent variable - Farm lncome Explanatory Variable coefficients Standard error P>t

Instrument Total farm labour 352220.7 ••• 39602.18 0.000

Exogenous variables Luckl2 -69699.14 45532.52 0.127 Sex (Male) 227582 322939.7 0.481 Age 47.88636 6159.064 0.994 Education -26157.23 141753.3 0.854 Household size 36354.21 ••• 9927.991 0.000 Value oflivestock -.1246653 .081453 0.127 Native -261800.2 171192.8 0.127 Migrant2 117120.7 152249.9 0.442 Matrimonial status 26591.72 186237.3 0.887 Dependence Cocoa -158250.7 134090.4 0.239 Zone) -447247_4••• 150193.8 0.003 Zone3 110675.9 156755.4 0.481 constant 451572.9 413196.7 0.275

Number of obs = 362 F-stat = 12.05 R-squared = 0.3104 Prob > F =0.000

Note: •••, ••, • indicate significance levels at 1 %, 5%, and 10% levels, respectively.

171 Game level 4: game with losscs

Reduced form equation: Inc= o,1X1 + o22X, + v,

X1 = (Luck123,Sex, Age, Education,HH _ Size,Va/ _ livestock, Native,Migrant2, Mat _Status,Dep _ Cocoa,Zonel,Zone2) X, = Farm _ labour

Table A.8: Ordinary Least Square Estimation results of lncome Dependent variable - Farm Income Explanatory Variable coefficients Standard error P>t

Instrument Total farm labour 346919.6 ••• 39425.08 0.000

Exogenous variables Luck123 -51949.51 38491.53 0.178 Sex (Male) 223655.9 323209 0.489 Age -18.37939 6168.41 0.998 Education -32134.26 142162.1 0.821 Household size 36508.42 ••• 9940.835 0.000 Value of livestock -.1276491 .0817592 0.119 Native -255586.6 171138.9 0.136 Migrant2 127087.4 151776.4 0.403 Matrimonial status 28793.04 186527.1 0.877 Dependence Cocoa -163077.6 134572.3 0.226 Zonel -451386 ••• 150203.4 0.003 Zone3 103212.1 156530.2 0.510 constant 463414.1 413420.5 0.263

Number of obs =362 F-stat = 11.99 R-squared = 0.3093 Prob > F =0.000

Note:•••.••,• indicate significance levels at 1%, 5%, and 10% levels, respectively.

172 Table A.9: Residuals from the augmented regressions

Dependent variable - Degree of risk aversion

Avers' =ô,X, +a.lnc+ûu, +e,

Aversl Avers2 Avers3 Avers4 Coef. P-value Coef. P-value Coef. P-value Coef. P-value

Residuall 4.89e-08 0.739 Residual2 -2.28e-08 0.892 Residual3 -1.51e-07 0.426 Residual4 - 7 .06e-08 O. 702

Note: the residuals from the first-stage regressions are no! significant in any of the augmented regressions - No endogeneity problem.

173 Appendix 6: Spearman 's rank correlation coefficients

A: variables used in the Ordered Lo~it model (Observation= 362) 1 Sex HH Size Age Education Incorne Livestock M Status Sex \ 1. 0000 HH Size 1 0.0645 1.0000 Age I 0.0409 0.3560 1. 0000 Education \ 0.0857 -0.1333 -0.2869 1.0000 Incorne 1 0.0149 0.3236 0.1166 -0.0480 1. 0000 Livestock 1 0.0479 0.2353 0.0205 -0.0365 0.1232 1.0000 M Status 1 0.1750 0. 1398 0.2216 -0.0919 0.0970 0.0568 1. 0000 Native 1 0.0006 -0.1281 -0.0684 0.3916 -0 .1165 -0 .1156 -0.1517 Dep_Cocoa \ 0.0507 -0.1074 0.0180 0.0397 -0.0854 -0.2185 0.0522 Migrant2 1 0 .1162 0.0827 0.2117 -0.2925 0.0935 0.0403 0.0764 Zonel 1 0.0627 0.0974 -0.0737 -0.0947 -0 .1090 0.0535 -0.0084 Zone3 1 -0.0220 0. 0213 -0.0380 0.1670 0.1144 -0.0528 0.0304 Native Dep_Cocoa Migrant2 Zonel Zone3 Native 1 1.0000 Dep_Cocoa 1 -0.1305 1. 0000

Migrant2 1 -0.3037 -0.0126 1. 0000 Zonel 1 -0.1039 -0.2735 0.0240 1. 0000 Zone3 \ 0.0449 0.2678 -0.0263 -0.5038 1.0000

B: variables used in the Trivariate Probit model (Observation= 362} Education low avers high avers ill/death output p disease intrant ris Education 1.0000 low avers -0.2044 1. 0000 high avers 0.1588 -0.1735 1. 0000 ill/death 0.0609 -0.0510 -0. 0611 1. 0000 output_price 0.0978 -0.1173 0.0346 0.0512 1.0000 disease/pest -0.0667 0.0168 -0.0023 -0.0789 -0.0025 1.0000 intrant risk -0.0043 -0.0070 -0.1172 0.2469 0 .1130 -0.1778 1.0000 ownership -0.0227 -0. 0112 0.0364 -0.0992 0.0177 -0.0858 -0.0629 HH size -0.1333 0.0902 -0.1417 -0.1577 0.0892 0.0444 0.0678 HH size2 -0.0999 0.0759 -0.0952 -0.1229 0. 0677 0.0510 0.0639 infori;;ation -0.1141 0.0601 0.0335 0.0257 0.0314 0.0364 -0 .1386 credit 0.0476 -0.1297 0.0618 0.2220 -0.0656 -0.1922 0.0601 off_frnact 0.0667 -0.0385 0.0099 0.0174 -0.0704 -0.1612 0.0132 experience -0.1365 0.1787 -0.1287 -0.1125 -0.0881 0.0933 -0.0160 zonel 1 -0.0947 -0.0588 -0.0827 0.0994 0.1203 -0.0606 0.4438 zone2 1 -0.0669 0.0319 0.0367 0.0277 0.0173 0.1947 -0.1585 ownership HH size HH size 2 inforrn credit off_frnact experience ownership 1 1. 0000 HH_size 1 0.0533 1.0000 HH size2 1 0.0496 0.9326 1. 0000 radio 1 0.0124 -0.2112 -0.1700 1.0000 information 1 0.0293 -0.0614 -0.0921 0.1574 1. 0000 off_fmact \ 0.0468 0.0560 0.0524 -0.0902 0.1537 1.0000 experience 1 0.0216 0.3234 0.2426 -0.0962 0.0343 -0.0695 1.0000 zonel 1 -0.1594 0.0974 0.1131 -0.2654 -0.2570 -0.1348 -0.0798 zone2 1 0.1052 -0.1206 -o .1153 0.4264 0.3038 -0.1765 -0.0613 zonel zone2 zonel 1 1.0000 zone2 1 -0.5301 1. 0000

174 Appendix 8: The survey questionnaire (in French)

Présentation du projet: Ce questionnaire a été conçu dans un but essentiellement académique. Vos réponses seront analysées sans référence explicite à votre identité et serviront à la rédaction d'une thèse PhD. Ce questionnaire vise essentiellement quatre objectifs : (i) Identifier l'attitude du paysan face à des alternatives risquées, (ii) Identifier les différentes sources de risques auxquelles font face les paysans ainsi la perception qu'ils ont de l'impact de ces risques sur leur revenu, (iii) Identifier les différentes stratégies utilisées par les paysans pour gérer ces risques et enfin (iv) Evaluer l'intérêt que les paysans portent à une potentielle assurance qui leur permettrait de se couvrir du risque-prix du cacao. Votre participation au déroulement de cette enquête est une marque d'encouragement à la recherche académique et scientifique. Nous vous en remercions d'avance.

SECTION O: IDENTIFICATION

001 -Nom et code de l'enquêteur /_/_/

002 - Date de l'interview /_/_/ /_/_ / 2009

003 - Zone : LJ Nom et code de la Sous-préfecture /_ / _ /

004 - Nom et code du village /_/_/

005- Nom du Campement. /_/_/

006-- Nom et code de l'enquêté ./ _ / _ /

007 -Contact de l'enquêté Tel: BP : .

Confidentialité: les informations collectées au cours de cette enquête sont strictement confidentielles. Elles ne peuvent en aucun cas être utilisées à des fins de contrôle ou de répressions économiques.

176 SECTION 1 : CARACTERISTIQUES GENERALES DU PRODUCTEUR No Questions Réoonscs/Codcs 101. Sexe 1-Masculin 2- Féminin 102. Quel est votre statut matrimonial ? 1- Célibataire 2- Marié(e) 3- Divorcé(e)/séparé(e) 4- Union libre/ Concubinage 5- Veuf(ve) 103. Quel âge avez-vous ? (Laisser /'enquêté répondre librement et inscrire son âge LLI Ans exact dans l'espace; lui proposer les intervalles s'il hé.

- Nombre d'enfants du chef du ménage : LLI - Autres personnes membres du ménaze : 1 1 1 109. Quelles sont vos activités courantes génératrices de A- Agriculture revenu? B-Commerce (Plusieurs 111odalùés possibles : Ne pas suggérer les C-Elevage modalités) D- Salarié public/privé (Pension) E-Rente F- Autre (à préciser) ...... 110. Quelle est votre principale activité génératrice de 1- Agriculture revenu? 2-Commercc (Ne pas suggérer les modalitis) 3-Elevage 4- Salarié public/privé (Pension) 5- Rente 6-Autre (à préciser) ...... 111. Combien de champs de cacao avez-vous ? LLI champs de cacao

177 112. Quelles sont les superficies et les années de I Champs Superficie Année création création des différents champs de cacao ? en hectare (Si le producteur II plus de 4 champs, retenir les 4 premiers champs les plus grands) 2

4 113. Quelles sont les autres principales cultures pratiquées par le ménage? {_Citer les modalités du tableau ci-dessous) Cultures Pratiquées Superficie Année de création Variation de la actuelle (ha) superficie sur les 5 Ecrire «oui» ou « non ,. selon la réponse de /'enquêté dans la dernières années colonne en face de la culture l-Augmenté 2-Pas de changement 3-Diminué Café 1 1 Hévéa 1 1 Palmier à huile 1 1 Maraichers (tomate, aubergine, piment, oignon) LJ Féculents (Patate, Banane, Manioc, Igname) 1 1 Céréales (riz, maïs, mil, sorgho) LJ 114. 1 Les cultures vivrières (maraichers, féculents, céréales) 1 A=Consommation pratiquées par le ménage sont-elles destinées à la consommation du ménag_e ou à la vente? B=Vente

SECTION 2: SITUATION ECONOMIQUE ET FINANCIERE DU PRODUCTEUR No Questions I RéJ>onses/Codes 201. Combien de kilogrammes de cacao avez-vous I Années I Quantités (Kg) 1 Prix d'achat (FCFA) 1 produit au cours des cinq (5) dernières années? 2008 (Insister pour avoir la réponse) 2007 :: !e:m':,:n::-:a:::~ ':::;::~ ~n,. ~=~::: ,«;~;;~;: 1 ~:: 1 ...... 1 ...... 1 valeur en kilogramme dans /.a colonne « quantttis » 2004 202. A combien estimez-vous en moyenne le revenu annuel provenant de la vente des autres cultures industrielles autre que le cacao ( café, hévéa, palmier à huile ... )? 1 .•...... •...... •..••• FCFA (En cas hésitation, donner le revenu issu des ventes de l'année dernière. Insister pour aw,ir la riJJOnse) 203. A combien estimez-vous en moyenne le revenu annuel provenant de la vente des cultures vivrières (tomate, banane, igname, riz, mais, etc.) ? 1 FCFA (En cas hisitation, donner le rnenu issu des ventes de l'annie derniire., Insister pour avoir la réponse)

178 204. A combien estimez-vous en moyenne le revenu annuel provenant des activités non agricoles (commerce, élevage, rente, etc.)? (En cas hésitation, donner le revenu de la l'année ...... FCFA dernière. Insister pour avoir la réponse) 205. Quelle est la part de revenu issu du cacao dans 1- 1/10 votre revenu total? 2-2/10 3- 3/10 4-4/10 5- 5/10 6- 6/10 7- 7/10 8- 8/10 9- 9/10 10- 10/10 206. Possédez-vous des animaux ? 1-0ui 2- Non - question 209 207. Si oui, quels sont les animaux que vous Animaux Nombre de tête possédez? Volaille 1 1 1 8œufs 1 1 1 Moutons 1 1 1 Cabris 1 1 1 Porcs 1 1 1 Autresl : ...... 1 1 1 Autres2: ...... 1 1 1 208. Quel usage faites-vous de votre revenu total ? A- Entretien familial NB : Poser la question suivante (209) même si la 8- Equipement domestique réponse à cette question est dijfé rente de «Epargne » C- Réinvestissement dans l'agriculture D- Epargne E-Autres (à préciser) ...... 209. Quel usage faites-vous de votre épargne? A- Prestations sociales (éducation, alimentation, aide, etc.) 8- Investissement dans l'agriculture (outils, produits, etc.) (Citer les modalités) C- Investissement non agricole (construction de maison, achat de véhicule, etc.) D- Support pour les chocs (décès, maladie, accident, etc.) E- Pas d'épargne F- Autres (à préciser) ...... - ········································································· 210. Comment a évolué cette épargne par rapport à son }-Augmenté niveau il y a 2 ans? 2-Diminué 3- Stable 211. Avez-vous été en contact avec une structure 1-0ui financière (banque, rnicrofinance) au cours de ces 2- Non - Question 301 deux dernières années ? 212. Quelles opérations avez-vous entrepris auprès de A- Prêt ses structures ? 8- Epargne C- Subventions D-Autres (à préciser) ......

179 SECTION 3 : VIE COMMUNAUTAIRE ET COMMERCIALISATION No Questions Réponses/Codes 301. Etes-vous membre d'une organisation paysanne agricole (OPA)? 1-0ui NB : Poser la question (301) même si le producteur 2-Non répond ,, non » à cette question 302. Pourquoi?

303. 1 Quels sont les services offerts aux membres? A- Ventes groupées (Cit,r les modalités) B- Accès au crédit C- Informations sur les prix D- Fourniture des intrants E- Transport de produits F- Assistance technique G- Formation H- Autre_(à préciser) . 304. 1 Quelle appréciation faites-vous de ces services ? 1- Bonne 2-Moyenne 3- Mauvaise 4- Très mauvaise 305. Avez-vous déjà passé un contrat (a"angement de 1- A chaque campagne vente avant la récolte) commercial avec des 2- Assez souvent acheteurs durant les 5 dernières années? 3-Rarement 4-Jamais 306. Pourquoi avez-vous choisi de passer un contrat ? A- Limiter les fluctuations de prix (Citer les modaütés) B- Obtenir le meilleurs prix possible C- Garantir un écoulement facile D- Obtenir des avances sur la récolte E-Autre (à préciser) . 307. Si vous avez déjà passé un contrat, avec qui ? A-VotreOPA (Ne pas sugglr,r de modalités) B-Pisteurs C- Acheteurs D- Exportateurs E-Autre (à préciser) . 308. 1 Si vous ne passez pas de contrat, à qui vendez-vous I A- Votre OPA généralement votre produit ? B- Pisteurs C- Acheteurs D- Exportateurs E- Autre (à préciser) . 309. 1 Pourquoi?

180 SECTION 4 : MOYENS DE PRODUCTION No I Questions Réponses/Codes 40 l. 1 Comment financez-vous vos activités de I A- Epargne production? B- Crédit (formel/informel) C- Aide familiale D- Groupe d'entraide E- Avance sur récolte F- Autre (à préciser}. . 402. 1 S'il s'agit d'un crédit, à quelle(s) institution(s) 1 A- Banque faites-vous recours? B- Tontine C-OPA D-Ami E- Pas de crédit F- Autre (à préciser) . 403. A combien d'hectares pouvez-vous estimer la totalité des terres (que vous cultivez) occupées par ...... Hectares la culture de cacao? 404. 1 Comment avez-vous obtenu les terres que vous I A- Héritage (familiale) cultivez ? B- Location C-Donation D- Métayage (Abusan) E-Achat F- Prêt temporaire gratuit G-Autre (à préciser) . 405. Comment jugez-vous la fertilité du sol par rapport à l-Bonne ilya5ans? 2-Moyenne 3-Mauvaise 406. Avez-vous utilisé des intrants (engrais, produits !-Oui phytos) au cours des 3 dernières années ? 2-Non 407. Combien de personnes de votre famille vous assistent dans la pratique de la cacao-culture ? I I I 408. De combien de manœuvres disposez vous actuellement pour l'ensemble de votre exploitation /_/_/ de cacao? 409. Quelle est l'origine de la main d'œuvre que vous A-Allogène B- Allochtone employez habituellement ? C- Autochtone D- Pas de main d' œuvre

41 O. 1 A combien pouvez-vous estimer le coût de la main d'œuvre dans la culture du cacao au cours d'une 1 F CFA année? 411. 1 A combien estimez-vous votre coût total de production (Intrants + main d'œuvre) de votre 1 ••••••••••••••••••••••• F CFA cacao par année? 412. 1 Est-ce que le prix actuel du cacao suffit à couvrir 11-0ui votre coût de production ? 2- Non

181 SECTION 5 : PERCEPTION ET GESTION DU RISQUE No _Questions Réponses/Codes 501. Dans votre activité de culture du cacao, quels A- Sécheresse sont les problèmes/catastrophes qui ont B- Pluies prématurées négativement affectés votre condition de vie ces C- Maladie ou décès d'un membre clé du ménage 10 dernières années ? 0- Baisse inattendue et volatilité du prix du cacao (Citer les modalités) E- Difficultés d'accès aux intrants F- Les risques phytosanitaires (attaques d'insectes et de maladies des plants) G- Pénurie de la main d' œuvre H- Expulsion, perte de terre 1- Autres (à préciser) . 502. Quelle appréciation faites-vous de l'impact de LJ- Sécheresse chacun de ces problèmes sur la de votre activité LJ- Pluies prématurées agricole, en particulier l'impact sur votre LJ- Maladie ou décès d'un membre clé du ménage revenu? LJ- Baisse inattendue et volatilité du prix du cacao LJ- Difficultés d'accès aux intrants 1- Pas d'impact LJ- Les risques phytosanitaires (attaques d'insectes et de 2- Impact faible maladies des plants) 3- Impact modéré LJ- Pénurie de la main d'œuvre 4- Impact fort LJ- Expulsion, perte de terre 5- Impact très fort LI- Autres (à préciser) .

(Inscrire le score dans la case approprile) 503. Que faites-vous pour vous protéger de ces A- Epargne propre problèmes (risques) ou pour y faire face quand B- Travail non agricole ils se produisent? C- Aide de la famille (Citer les motlalitis) D-Aide non familiale (groupe d'entraide, coopérative) E- Diversification des cultures F- Migrer vers une autre région G-Autres (à préciser) .

504. Comment appréciez-vous les mouvements de 1- Généralement stable prix bord-champ du cacao sur les 10 dernières 2- lnstable années? 3- Très instable 505. Quels sont les prix rrummum et maximum auquel vous aviez vendu votre produit ces 10 !-Prix minimum: FCFA dernières années ? 2- Prix maximum: FCFA

506. Pensez-vous que le prix bord champ du cacao 1- Augmentation augmentera ou baissera dans un futur proche ? 2- Stable 3- Diminution 507. Comment réagissez- vous en cas de baisse du A- Diminution du nombre de manœuvres prix du cacao ? B- Diminution des dépenses en intrants (engrais, pesticides, etc.) (Citer les modalités) C- Diminution des dépenses familiales 0- Abandon de certaines parcelles de cacao E- Diversification des cultures (hévéa, palmier à huile, etc.) F- Ne réagit pas G-Autres (à préciser) .

182 508. Comment réagissez- vous en cas de hausse du A- Augmentation du nombre de manœuvres prix du cacao ? B-Augmenlation des dépenses en intrants (engrais, pesticides, etc.) (Citer les modalités) C- Augmentation des dépenses familiales D- Création de nouveaux champs de cacao E- Ne réagit pas F- Autres (à préciser) ...... - ...... 509. A combien voulez-vous qu'on achète le kilozramme de votre cacao à la prochaine traite? ...... FCFA/Kg 510. Si une assurance-prix vous était proposée, 1- Oui - Question 512 seriez-vous prêt à payer un certain montant 2-Non AUJOURD'HUI pour garantir ce prix? 511. Si non, pourquoi ? ...... -+ Section 6 ...... 512. Combien êtes-vous prêt à payer par kilogramme pour garantir ce prix ? ...... FCFA/Kg

SECTION 6 : ENVIRONNEMENT INFORMATIONNEL No Questions Réponses/Codes 601 Avez-vous un poste de radio ? 1- Oui 2-Non 602 L'écoutez-vous souvent? 1-0ui 2-Non 603 Quelles informations vous intéressent ...... particulièrement quand vous écoutez la radio ? ··········································································· ...... 604 Avez-vous un poste de télévision? 1-0ui 2-Non 605 Le regardez-vous souvent ? 1-0ui 2-Non 606 Quelles informations VOUS intéressent ··········································································· particulièrement quand vous regardez la ··········································································· télévision ? ...... 607 Disposez-vous du réseau téléphonique ? 1-0ui 2-Non 608 Quel type de connexion téléphonique avez-vous ? A- Téléphone portable Be Téléphone fixe 609 Avez-vous l'information sur le prix indicatif du 1-0ui 2-Non cacao avant de procéder à la vente ? 610 Si oui, d'où tenez-vous cette information? A- Télévision B-Radio C-Jownaux D-OPA E-Autre (à préciser) ......

FIN DU QUESTIONNAIRE MERCI POUR VOTRE COLLABORATION!!!

183