THREE ESSAYS IN DEVELOPMENT ECONOMICS By Joshua Gill A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Agricultural, Food, and Resource Economics Doctor of Philosophy 2019 ABSTRACT THREE ESSAYS IN DEVELOPMENT ECONOMICS By Joshua Gill The three essays in this dissertation study how the rural poor in Pakistan make choices and how better program design can alleviate the constraints they face. The first essay investigates t he participation decision of smallholder paddy farmers in a Warehouse Receipts Financing (WRF) program which can mitigate the ir credit and storage constraints, allowing them to increase their incomes. We a discrete choice experiment approach to study the d ecision - making process and find risk aversion and transaction cost erode the benefits for smallholder farmers making it an unattractive prospect . We find that likelihood of participation can be increased through better contract design which lower cost of p articipation and reduc e exposure to price uncertainty. These findings have important implications for the optimal design of warehouse receipt financing contracts, as well as their general feasibility for marketing to small farmers. It also highlights that programs aimed to upli ft smallholder farmers should not only address constraints of circumstance (e.g. access) but also internal constraints (e.g. risk aversion). The second essay aims to alleviate information constraints regrading fertilizer usage as its i ndiscriminate and fa ulty use can affect soil health. Evidence shows that soil quality in Pakistan has been deteriorating which can be partially explain by poor nutrient management . In this study we conducted soil tests and provided recommendations on use of organic and inorga nic fertilizer . T his study uses an experimental design with two treatment arms and a control group which received no information and its soil was not tested. The base treatment provided farmers with information on their soil health an d recommended fertiliz er use condition on the crops they cultivate. The second treatment arm complemented this information with a peer comparison which was used an encouragement mechanism to improve the efficacy of information provided . The study highligh ts some important cons traints to information dissemination and provides some evidence on the use of peer comparison as a potential tool to improve efficacy of information campaigns. We see a statistically significant increase in manure usage and a heteroge nous impact on Urea us e but no impact on the overall fertilizer use . We find that farmers they were already using close to the recommended amount (within 1 bag deviation) increased their urea application rate . These findings suggest two underlying mechanis ms at play. First, it alludes to liquidity constraints as farmers increased manure use which is cheap and those who could already afford higher quantities of Urea were able to respond to the recommendation of increasing application rates. The fact that we do not see impact on D AP further gives credence to this assumption as DAP is close to 3 times the cost of Urea. A lternat ively, it could be that farmers who were away from the suggested fertilizer amounts did not trust the recommendations. The third essay studies the dynamics o f warehouse receipts financing (WRF) demand by small scale risk averse farmers in Pakistan. A dynamic model is used to investigate how risk and time preferences, transaction costs, and uncertainty reduce demand for WRF, and even lead to non - participation i n the program. The model is calibrated and solved for a representative small - scale farmer that grows paddy. Results show high transaction costs to be a major barrier to participation. Similarly, expectations about future prices also a ffect participation wh ich drops to zero if the subjective probability of prices falling goes beyond 10 percent. Copyright by JOSHUA GILL 2019 v ACKNOWLEDGEMENTS I am very grateful to Dr. Andrew Dillon for his invaluable support and guidance th roughout my PhD program. His advice was instrumental in helping me frame my research questions, secure grants, and run my projects as a graduate student. I am especially than kful to Dr. Maria Porter for agreeing to be my major professor in my last year of PhD and providing the guidance and support that was needed to finish my dissertation. I also want to thank the other members of my committee, Dr. Robert Myers, Dr. Vincenzina Caputo, and Dr. Chris Ahlin . Their advice was very helpful in adding rigor to the analysis. I am also appreciative of other faculty members in AFRE and the Economics department, particularly those who provided feedback at development seminars. This dissertation uses data collected with the support of International Food Policy Research Institute supervised by Dr. David Spielman . I am particularly grateful for his continuous support and guidance; without him i t would have been very difficult to carry out my research . I would also like to thank Global Center for Food Systems Inn ovation at Michigan S tate , Center for Economic Research in Pakistan , Dr. Asim Ijaz Khawaja and Dr. Farooq Naseer for their s upport . I was lucky to be surrounded by great friends during my time at Michigan S tate University. I would to thank Marie Steele , Hamza Haider , Mukesh Ray and Awa Sanou for their support. Sophia , Jo ey , and Asa were especially helpful in my final year. I am particularly grateful to Schanzah Khalid f or her continuous support during my PhD program , helping me with my research ideas, grant proposals, job applications, and insp iring me to excel. Finally, m y parents have been a source of constant support and I am grateful to them for their encouragement . Without their prayers and support, I would not have gotten so far. vi TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ ....................... viii LIST OF FIGURES ................................ ................................ ................................ ........................ x Essay 1 : Selling Cheap: Arbitrage in the Basmati Value Chain ................................ .................... 1 Abstract ................................ ................................ ................................ ................................ ........... 1 1.1. Intro duction ................................ ................................ ................................ .......................... 2 1.2. Background ................................ ................................ ................................ .......................... 4 1.3. Experimental Methods ................................ ................................ ................................ ......... 6 1.3.1. Experiment s ................................ ................................ ................................ .................... 6 1.3.2. Treatment Assignments ................................ ................................ ................................ .. 8 1.4. Estimation Procedures ................................ ................................ ................................ ........ 10 1.4.1. R isk Aversion ................................ ................................ ................................ ............... 10 1.4.2. Time preferences ................................ ................................ ................................ .......... 11 1.4.3. Discrete Choice Models ................................ ................................ ............................... 12 1.5. Data ................................ ................................ ................................ ................................ .... 16 1.6. Results ................................ ................................ ................................ ................................ 18 1.6.1. Risk and Time Preferences ................................ ................................ ........................... 18 1.6.2 . Program Participation ................................ ................................ ................................ ... 19 1.7. Conclusion ................................ ................................ ................................ .......................... 23 APPENDICES ................................ ................................ ................................ .............................. 25 APPENDIX A DISCRETE CHOICE EXPERIMENT ................................ ...................... 26 APPENDIX B RISK AVERSION ................................ ................................ ..................... 29 APPENDIX C TIME PREFERENCES ................................ ................................ ............. 34 BIBLIOGRAPHY ................................ ................................ ................................ ......................... 37 Essay 2 : Does peer comparison encourage adoption of best practices among farmers in Pakistan? ................................ ................................ ................................ ................................ 42 Abst ract ................................ ................................ ................................ ................................ ......... 42 2.1 Introduction ................................ ................................ ................................ ........................ 43 2.2 Conceptual Framework ................................ ................................ ................................ ...... 46 2.3 Expe rimen tal Design ................................ ................................ ................................ .......... 48 2.4 Intervention Description ................................ ................................ ................................ ..... 49 2.4. 1 Soil Health Cards ................................ ................................ ................................ .......... 49 2.4.2 Mental Accounting and Call Reminders ................................ ................................ ...... 51 2.5 Data and Empirical Strategy ................................ ................................ ............................... 51 2.6 Estimation Methods ................................ ................................ ................................ ............ 54 2.7 Results ................................ ................................ ................................ ................................ 56 2.7.1 Take - Up ................................ ................................ ................................ ........................ 56 2.7.2 Manure Use ................................ ................................ ................................ .................. 58 2.7.3 Fertilizer Use ................................ ................................ ................................ ................ 59 2.8 Conclusion ................................ ................................ ................................ .......................... 60 vii APPENDIX ................................ ................................ ................................ ................................ ... 63 BIBLIO GRAPHY ................................ ................................ ................................ ......................... 68 Essay 3 : A Dynamic Model for Warehouse Receipts Financing Demand in a Developing Country ................................ ................................ ................................ ................................ .. 72 Abstract ................................ ................................ ................................ ................................ ......... 72 3.1. Introduction ................................ ................................ ................................ ........................ 73 3.2. Dynamic Storage Model ................................ ................................ ................................ ..... 76 3.3. Model Parameterization ................................ ................................ ................................ ..... 78 3.4. Results ................................ ................................ ................................ ................................ 83 3.5. Conclusion ................................ ................................ ................................ .......................... 85 APPENDIX ................................ ................................ ................................ ................................ ... 88 BIBLIOGRAPHY ................................ ................................ ................................ ......................... 95 viii LIST OF TABLES Table 1.1: Cost Benefit of WRF ................................ ................................ ................................ ..... 6 Table 1.2: Choi ce Experiment Attribute and Levels ................................ ................................ ...... 8 Table 1.3: Treatment Groups ................................ ................................ ................................ .......... 9 Table 1.4: Balance Table ................................ ................................ ................................ .............. 17 Table 1.5: Risk Aversion and Time Discounting Parameters ................................ ....................... 18 Table 1.6: MXL - EC Model (Preference Space) ................................ ................................ ........... 20 Table 1.7: Actual and Predicted Choices from MXL - EC model (Preference Space) .................. 21 Table 1.8: MXL - EC Model Estimates by Treatments (WTP Space) ................................ ........... 22 Table 1.9: Risk Aversion Choice List 1 ................................ ................................ ........................ 31 Table 1.10: Risk Aversion Choice List 2 ................................ ................................ ...................... 33 Table 1.11: Sheet 1 ................................ ................................ ................................ ....................... 34 Table 1.12: Time Discounting Choices ................................ ................................ ........................ 36 Table 2.1: Balance Table ................................ ................................ ................................ .............. 53 Table 2.2: Predicted Probabi lities of Recalling Receipt of SHC ................................ .................. 57 Table 2.3: Predicted Probabilities of Trust in SHC ................................ ................................ ...... 57 Table 2.4: Tobit Estimates of Manure Appli ed (maunds/kanal) ................................ ................... 58 Table 2.5: Heterogenous Treatment Effect on Urea ................................ ................................ ..... 60 Table 2.6: Balance Table (Restricted Sample for He terogeneity ) ................................ ................ 64 Table 3.1: Lagrange Results ................................ ................................ ................................ ......... 78 Table 3.2: Time and Risk Parameters ................................ ................................ ........................... 79 Table 3.3: Expected Prices (PKR/maund) ................................ ................................ .................... 81 Table 3.4: Warehousing Cost (40kg bag) ................................ ................................ ..................... 82 ix Table 3.5: Base Case Parameter Values ................................ ................................ ....................... 83 Table 3.6: Simulation Results ................................ ................................ ................................ ....... 84 x LIST OF FIGURES Figure 1.1: Sampling strategy ................................ ................................ ................................ ...... 10 Figure 1.2: Switching Distribution ................................ ................................ ............................... 11 Figure 1.3: Sample Question (DCE) ................................ ................................ ............................. 28 Figure 1.4: Example Game ................................ ................................ ................................ ........... 29 Figure 1.5: Practice Game ................................ ................................ ................................ ............. 30 Figure 1.6: Sample Question (Risk Aversion) ................................ ................................ .............. 32 Figure 1.7: Calendar Sheet ................................ ................................ ................................ ............ 35 Figure 2.1: Study Design ................................ ................................ ................................ .............. 49 Figure 2.2: Soil Health Card ................................ ................................ ................................ ......... 66 Figure 3.1: Historical Paddy Prices (PKR/maund) ................................ ................................ ....... 80 Figure 3.2: Average Expected Paddy Prices (PKR/maund) ................................ ......................... 81 Figu re 3.3: Storage Sensitivity Analysis ................................ ................................ ....................... 85 1 Essay 1 : Selling Cheap: Arbitrage in the Basmati Value Chain Abstract Smallholder farmers in developing countries face numerous constraints in both input and output markets, that reduce their p rofit - generating potential. Warehouse Receipt Financing (WRF) has been promoted as an innovative solutio n that provides access to more remunerative markets and formal financial institutions. However, participation by smallholder farmers has been low despit e its potential for profit generation. This paper considers the case of a WRF intervention in Pakistan w hich saw a similar outcome and aims to identify reasons for low take - up and how better contract design can improve participation. This study finds that risk aversion as an important factor for lack of participation. It highlights that smallholder farmers a re unwilling to take on the entire risk of price uncertainty due to intertemporal arbitrage. Results from a choice experiment show that when smallholder farmers were offered the same WRF product with price certainty that resulted in a no loss scenario pred icted participation increased by 32 percent. This result highlights that participation of smallholder farmers can be significantly increased by designin g WRF products and contracts that meet the needs of the farmers. These findings have important implicati ons regarding the demand for WRF and its general feasibility of marketing to smallholder farmers. 2 1.1. Introduction In most developing countries agricultur al markets are marred with inefficiencies that lead to a suboptimal equilibrium. Smallholder farmers are affected the most under these conditions as they suffer both in the input and output markets. They are underserved by financial institutions and have l imited access to downstream buyers, as a result they rely on intermediaries who expropriate r ents. The outcome is a vicious cycle of low investment and low earnings which inhibits their upward economic mobility. Moreover, a poorly functioning agriculture s ector also hinders in achieving the development goals of poverty reduction and inclusive econ omic growth as majority of the poor live in rural areas and derive their livelihood from agriculture. The desire to break this cycle has motivated numerous interve ntions (e.g. subsidies, information, infrastructure) that have provided little benefit to sma llholder farmers (Barrett et al., 2012; Fafchamps and Minten, 2012; Fischer and Qaim, 2014; Jayne et al., 2018; Ricker - Gilbert et al., 2013; Shiferaw et al., 2011) . Warehouse receipts fi nancing (WRF) has lately been promoted as an innovative solution that can improve incomes of smallholder farmers by alleviating liquidity and market access constraints (Aggarwal et al., 2018; Basu and Wong, 2015; Burke, 2014; Omotilewa et al., 2018) . However, smallholder farmer participation in these programs has not been very encouraging (William and Kaserwa, 2015) . There is limited work which explores the reasons for low participation despite it being a profitable prospect (Miranda et al., 2017) . The purpose of this paper is to understand the marketing choices of smallhold er rice farmers in Paki stan under a WRF program. 1 This study explores whether potential external and internal constraints made non - participation a subjectively rational decision. Three main factors are examined in this study: transaction costs, risk aversi on, and impatience. The objective of the paper 1 This program was implemented in 2017 and farmers cultivating less than 10 ac res were targeted. However, despite high interest shown during village meeting, take - up was less than 2 percent among the target population at rollout. 3 is to test how better product and contract design can address these constraints and in doing so, improve participation and earning possibilities of smallholder farmers. This paper uses a Discrete - Choice Experi ment (DCE), which has been used extensively to study consumer preferences and has been widely applied in many fields of applied and development economics (Caputo et al., 2013; Clark et al., 2014; Gibson et al., 2016; Lusk and Briggeman, 2009; Ortega et al., 2011; Scarpa and Willis, 201 0; Tanaka et al., 201 4; Ward and Singh, 2015) . We collect information on risk and time preferences through incentivized games using a multiple price list approach. (Andersen et al., 2008; Ward and Singh, 2015) . In addition, respondents were randomly assigned to one of three differe nt experimental groups offering different hypothetical levels of price guarantees: Group 1 was guaranteed a high price 3 months after storage; Group 2 was guaranteed a low price; Group 3 was not given any price guarantees. This study makes several contri butions to the literatu re on WRF. First, we use a DCE framework to evaluate whether a market for WRF exists among smallholder farmers, and their willingness to pay (WTP) for this service. There is very limited and mixed evidence on the benefits of WRF to s mallholder farmers (Aggarwal et al., 2018; Burke, 2014; Miranda et al., 2017; William and Kaserwa, 2015) , and prior studies do not explicitly measure WTP. Information on WTP can be ve ry useful in developing appropriate products for smallholder farmers, and in ascertaining whether such products would be financially feasible for providers. In addition, internal constraints can act as independent sources of disadvantage and can affect s mallholder tec hnology adoption decisions (Duflo et al., 2008; Liu, 2013; Ward and Singh, 2015) . This study contributes to this body of literature by examining the role of risk aversion, ti me discounting, and unc ertainty on decision to participate in WRF. 4 To our knowledge, only one other study has explicitly accounted for preference - related factors when evaluating the feasibility of WRF programs (Miranda et al., 2017) . The authors highlight the role of preferences in eroding the profitab ility of WRF and suggest that WRF is not feasible for smallholder farmers. We differ from this study by exploring ways to redesign WRF products to address these issues. Our results highlight that participation in WRF can be signif icantly increased if ex posure to price risk can be reduced, and that this factor is more important to farmers than either the cost of transport or borrowing. Farmers in the price guarantee groups select the WRF option more often (approximately 30 percent), which suggests that pr ice uncertainty is a major deterrent to participation. While marginal WTP decreases as interest rates increase, and increases if transport is provided, farmers in the price guarantee group are willing to pay more for credit in compar ison to those in the ot her two treatment groups. Farmers who receive a price guarantee also have a lower willingness to pay for transport compared to those in the other treatment groups. In addition, who are relatively less risk averse or do no discount fu ture payments are more likely to select WRF. The remainder of the paper proceeds as follows: Section 1.2 provides background information on the context of this study of Basmati rice farmers in Pakistan; Section 1.3 outlines the experimental design; Secti on 1.4 outlines the est imation strategy; Section 1.5 discusses the data; Section 1.6 summarizes the results; and Section 1.7 concludes. 1.2. Background Rice value chain in Pakistan is underdeveloped and smallholder farmers suffer the most in this environment. There is significant in tertemporal and spatial price variation as the output markets are inefficient and fail to move the product from surplus to shortage periods. Smallholder farmers suffer more as they lack access to storage and sell their output immedia tely at harvest 5 when pr ices are depressed low. In addition, the playing field is also tilted against them in the input market as they are underserved by formal financial institutions due to their small size and high poverty incidence. This not only leads t o liquidity constraints during the production cycle but also limits the ability of smallholder farmers to invest in better technology. Taken together, these input and output constraints severely inhibit the ability of smallholder farmers to take risks, mak e necessary investments , and improve their productivity. WRF has lately been promoted as a viable mechanism to develop the value chain and make it more inclusive. WRF can improve income of smallholder farmers through three channels. First, it provides acce ss to storage allowing smallholder farmers to transfer produce from periods of surplus to shortage. Second, it provides access to formal financial institutions by using stored grain as collateral. Finally, the warehouse acts as a clearing house where produ ct is graded and agglom erated thereby improving the bargaining power of smallholder farmers, reducing the transaction and search costs of doing business. In 2017 a WRF program was piloted in the district of Hafizabad, Pakistan. The project was implemented in 50 randomly selecte d villages with the objective of improving returns for smallholder farmers through storage, credit, and better market linkages. Approximately, 1500 farmers cultivating under 10 acres registered for the program and gave soft commitmen ts regarding the quanti ty they would store in the upcoming harvest. However, once the harvest season concluded less than 2 percent of the registered farmers had stored paddy at the warehouse. This outcome was very puzzling as WRF was a profitable prospect and the increase in inc ome was not trivial as shown in Table 1.1. 6 Table 1 .1: Cost Benefit of WRF Year Harvest Price Post - Harvest Price Net benefit Profit 2016 Rs 1300/40 kg Rs 1900/40 kg Rs 240/40 kg Rs 360/40 kg Rs 10,800/acre 2017 Rs 1550/40 kg Rs 2200/40 kg Rs 240/40 kg Rs 410/40 kg Rs 12,300/acre Note: This table shows the average prices at harvest and 3 month post - harvest in the project areas. The cost includes drying charges, storage charges, rent for jute bags, labor, mark - up on loan, and w eight - loss due to drying. The profit figure is calculated using an average yield of 1200 kg per acre. These figures were provided by the warehousing management company and reflect the prices at which paddy was bought and later sold by them. The cost of sto rage was also shared by them based on the actual expenses incurred during the season. ` This paper aims to understand the participation choice of smallholder farmers and test if alternate designs can improve take - up. The first question we address is whethe r there is demand for WRF among smallholder farmers. Risk aversion, time discounting, and transaction costs are explored as potential factors depressing demand. The choice of these factors is based on literature which shows smallholder farmers to be price sensitive, reluctant to engage in risk prospects and value immediate rewards more than future rewards. The second question this study addresses is can participation be increased through better contract designs. We specifically test how a reduction in uncer tainty on returns can improve participation by farmers. Finally, the paper quantifies the willingness to pa y under different contract designs. 1.3. Experimental Methods This section illustrates the experimental procedures. Respondents were first exposed to in centivized games to elicit risk and time preferences and then to DCE questions. The first subsection explains the games that were played to measure risk and time pref erences and the DCE that was implemented to elicit farmers preferences for WRF. The second subsection describes the hypothetical price regimes farmers were randomly assigned to before participating in the DCE. 1.3.1. Experiments This study utilizes risk and time discounting to explain farmer participation choice in WRF. Risk and time preference have been extensively used to explain a range of choice behavior such 7 as technology adoption, investment, migration, education attainment, and smoking (Ashraf et al., 2006; Chavas and Holt, 2006; Jensen, 2010; Lawless et al., 2013; McKenzie et al., 2013; Warnick et al., 2011) . Multiple price list approach wa s used to elicit risk preferences. Two price lists with 14 questions in each list was used and the respondent had to choose between two lottery options labelled Option A and Option B. Option A offered a sure small return and Option B offered a higher expec ted return but with higher risk. Graphical images were used to help the respondent underst and the probabilities involved under both lotteries. One price list with 10 questions was used to elicit the discount rate. Respondents were given a token at the st art of the game and informed that they could exchange it for real money based on the quest ions in the price list. Each question had two options labelled Option A and Option B offering different amount of money at different times in the future. Option A was for sooner smaller payment and Option B was for larger later payment. Picture of a calend ar with dates corresponding to Option A and Option B was used to help the respondents understand the choices better. These games were incentive compatible, and respon dents were informed that at the end of the game one of their responses will be randomly ch osen for actual payment. Experimental protocols for these games are shown in the Appendix. In the DCE, respondents were asked to choose between three alternatives; t wo warehousing products and an opt - out option. The warehousing products were described by three attributes as reported in Table 1.2. These three attributes were chosen based on the conversations with farmers. The cost levels reflect subsidized cost, true c ost, and true cost plus a value - added service fee. Credit up to 70 percent of the value of the stored product was offered to relax liquidity constraints due to storage. The interest rate levels are reflective of the rate charged by government sponsored 8 agr icultural loans, conventional banks, and microfinance banks. Transport is another critical attribute considered as smallholder farmers do not own a truck and renting can be very costly given their low volumes. Two levels of this attribute are selected base d either paddy is picked from farmgate or the farmer must arrange for the transportation. Table 1. 2 : Choice Experiment Attribute and Levels Warehouse Receipts Financing Attributes Description Levels Cost The cost of storage for a 40 kg bag of paddy. Rs 10 per month Rs 20 per month Rs 40 per month Interest The depositor has the option to take loan of up to 70% of the value of the product stored at harvest. No Interest 15 % per annum 30 % per annum Transport Paddy would be picked from the field for storage. Present Absent An orthogonal optimal de sign was used for the DCE as a full factorial design would have resulted in 324 possible choice questions. The questions were generated using Ngene software which showed that a minimum of 18 choice sets were required to achieve 100 percent D - efficiency. Th ese questions were further divided into two blocks so that each respondent answered a set of 9 questions. This was done to avoid mental fatigue and loss of interest by the respondent, which can lead to poor responses. Each choice se t had three options; two options of a WRF service and an opt out option. See appendix A for an example question. 1.3.2. Treatment Assignments Respondents were randomly assigned to a group with a hypothetical price regime to estimate its impact on participation choice. Table 1.3 shows the three experimental groups. 9 Table 1. 3 : Treatment Groups Groups Treatments Description Price Guarantee Group 1 Control No Price Guarantee -- Group 2 High Price High Price Guarantee Rs 1800 /40 kg Group 3 Low Price Low Pr ice Guarantee Rs 1600/40 kg Respondents in each group were read a script before the start of the DCE informing them regarding the hypothetical price scenario. Group 1 is the Control and respondents were informed that the expected price of paddy at harves t is Rs1300/bag and historical ly paddy prices tend to rise in the 3 months post - harvest. Hence, if they chose to store and sell later, they could earn a significantly higher profit. However, they were also informed that profits were not guaranteed as in an y other business and prices ca n fluctuate unexpectedly. Group 2 is the High Price treatment and respondents were informed that the expected price of paddy at harvest is Rs1300/bag and historically paddy prices tend to rise in the 3 months post - harvest. Hen ce, if they chose to store and sell later, they could earn a significantly higher profit. However, they were given the guarantee that the warehouse would purchase at Rs1800/bag 3 - month post - harvest irrespective of the market price. Group 3 is the Low Price treatment and respondents wer e informed that the expected price of paddy at harvest is Rs1300/bag and historically paddy prices tend to rise in the 3 months post - harvest. Hence, if they chose to store and sell later, they could earn a significantly higher profit. However, they were gi ven the guarantee that the warehouse would purchase at Rs1600/bag 3 - month post - harvest irrespective of the market price. The scripts are given in the appendix. The sampling frame for this study consisted of 1500 farmers in 50 villages who had earlier show n interest in the WRF. A subsample of 800 individual was randomly drawn for this study. On average 16 farmers were picked from each of the 50 villages. Farmers were then randomly assigned to one of the three experimental group s and then to one of the two D CE blocks. Figure 1.1 outlines the sampling strategy. 10 Figure 1. 1 : Sampling strategy 1.4. Estimation Procedures This section explains the estimation procedure for the preference data collected through th e incentivized games and choice data collected in the DCE. 1.4.1. Risk Aversion We estimate a Constant Relative Risk Aversion (CRRA) measure in this study and account for probability weighting when estimating risk aversion (Kahneman and Tversky, 1979) . Consider a game with two potential outcomes X and Y which can occur with the probabilities p and q, respecti vely. The value of the pr ospect can be given by where and and are jointly determined. Let us look at the following example for illustration. If a participant switches from the risk - free option (A) to the lottery (B) in question number 5 in both game 1 and game 2 then the following problem can be solv ed t o derive and Village 16 Respondents High Price 5 Respondents Block1 3 Respondents Block 2 2 Respondents Low Price 5 Respondents Block 1 2 Respondents Block 2 3 Respondents Control 6 Respondents Block 1 3 Respondents Block 2 3 Respondents 11 Figure 1.2 illustrates the switching points in the two games. As expect ed, we see a large r proportion of switches happening later in the games suggesting the that the farmers are risk averse on average. Figure 1. 2 : Switching Distribution 1.4.2. Time preferences We estimate discount rates while controlling for concavity of utility and included a front - end delay of two weeks in the games. Literature highlights correcting for concavity of utility as ignoring it results in underestimating the discount factor and over estimating the impatience of the respondent (Cheung, 2016) . Similarly, using a front - end delay is recommended to ensures that the 12 transaction cost of receiving money is consistent across sooner and later payments (Andersen et al., 2008) . Utility in each period can be expressed as where utility , the disco unt function , and the interest rate is .The respondent will select the sooner payment if the utility from it is higher compared to that in the future. is estimated from the risk game and is estimated from the choices made by respondents in the time preference game. 1.4.3. Discrete Choice Models DCE methodology is based on Lancaster theory of consumer demand and random utility theory (He nsher et al., 2015) . The former assumes utility is derived from the characteristics and properties of the good consumed rather than the good itself (Lancaster, 1966) . The latter assumes individuals to be rational beings who compare alternative s and select the option that maximize their utility (McFadden, 1974) . Formally, let represent the utility that an individual derives from alternativ e j at time situation t. The decision maker evaluates all the alternatives and choses i if and only if . However, the utility assigned by the decision maker is not known to the researcher and is partitioned into an observed and an unknown stochastic component . The functional form of this utility can be written as: The most common assumption is that the observed part of the utility is linear in some observed factors and can be expressed as where is a vector of product attributes, is the vector of decision characteristics, and is a ssumed to be independent and identically distributed. 13 We used a mixed logit model with error components ( MXL - EC) to analyze the choice data. Estimation of the model was carried out using a panel data structure which includes individual level risk aversion and discount rates as behavioral characteristics. MXL - EC was used as it is less restrictive and allows us to capture both systematic and random taste variations and accounts for correlations across utilities (Greene and Hensher, 2010) . In addition, it al so controls for the effects associated with the opt out option relative to the experimentally designed alternatives ( Caputo et al., 2013; Scarpa et al., 2005, 2007) . This utility structure can be expressed as follows: where is an indicator function that takes the value of 1 for experimentally desig ned alternatives; and is normally distributed with zero mean respondent - specific idiosyncratic error component associated with the experimentally designed alternative but not with the opt - out option. The rest of the elements in equation (2) are s ame as in equation (1). The unconditional probability of individual n choosing alternative j under MXL - EC is given by: where and are the probability densities over which the coefficients of and vary in the population. Three MXL - EC models are estimated, one for each treatment group (Control, Hi gh Price, and Low Price). All models account for random taste variation by allowing the coefficients of attributes (cost, interest, and transport) to vary in the population following a one - sided triangula r distribution 2 . Whereas, the OptOut is assumed to b e normally distributed as the utility from the 2 One - sided triangular dis tribution is used as symmetric distributions around zero, example normal can lead to implausible results, such as a positive coefficient for cost. Under the one - sided triangular distribution, the parameters in our specification are distributed as where is distributed between with mean and variance . 14 status quo option can both be positive or negative 3 . The risk variable measures the constant relative risk aversion (CRRA) and has a positive support. A valu e of 1 means the individual is risk neutral while va lues smaller than 1 imply risk aversion and higher values imply risk loving behavior. The time variable has a support between zero and 1, it measures the discount factor where zero implies future income h as no value and 1 implies that future income is not discounted at all. These variables were interacted with the Opt - Out option as we assume that there might be heterogeneity in preferences with respect to selecting the WRF products. The utility function is expressed as follows for all treatment g roups: ( 1.6 ) where OptOut is an alternative - specific constant representing the opt - out alternative, is a continuous variable indicating the monthly cost of storage for a bag of paddy; and are effects coded variables indicat ing the per annum interest rate. takes the value of 1 if the interest rate is 15 percent and takes the value 1 if the interest rate is 30 percent. If the interest rate is zero, then then variables are coded as - 1. Transport is al so effects coded and takes the value of 1 if the farmgate pick - up is provided and - 1 otherwise. Predicted probability of selecting an alternative in each choice set was computed using results from the MXL - EC model. Based on these probabilities a new varia ble was created which takes the value 1 if one of the WRF alternatives was assigned th e highest probability and zero otherwise. A probit model was then estimated with as the dependent variable and treatment assignmen t as independent variables. The specification is given as 3 The parameter is distributed as , where is the conditional mean and is the standard deviation. 15 where is a binary variable, is the average participation rate in th e control group and and are the treatment effects on participation relative to the control. In addition to estimating the impact on predicted choices a probit was also estimated on the actual choices made by the respondents in the DCE. Willi ngness to pay across the three treat ment groups was calculated in the WTP space. This outcome is important in establishing whether WRF has a market among smallholder farmers. WTP can be obtained by taking a ratio of the attribute and price coefficients. Ho wever, these estimators generally do not have finite moments as the ratios of the coefficients can have infinite variances under most distributions (Daly et al., 2012) . One solution is to fix the coefficient estimate on price but that inherently assumes that everyone values money similarly. An alternate solut ion is to estimate the model in the willingness to pay space as it relaxes the assumption to have a fixed price coefficient (Scarpa et al., 2008) . The coefficients can be directly interpreted as marginal WTP measures (Scarpa and Willis, 2010) and it is also a more feasible approach whe n making comparisons across treatmen ts (Caputo et al., 2017) . An extended utility framework was used by pooling the data and including dummy variables for treatment assignment which were interacted with the attributes (Bazzani et al., 2017; De - Magistris et al., 2013; Lin et al., 2019) . T he data was pooled as high price guarantee vs control, low price guarantee vs control, and high price guarantee vs low price guarantee. For each of the experimental groups the utility in willingness to pay space can be s pecified as where is a random positive scalar r epresenting the price parameter, is the willingness to pay for each of the attributes which are defined above, give the respe ctive treatment 16 effects of the experimentally designed attributes. The sign and significance of the determine how the willingness to pay for the different attributes varies under different treatments. 1.5. Data Data for this study comes from two source s: a household level survey conducted in October 2017 and lab in the field experiments implemented in Octo ber 2018. Balance among respondents across socio demographics, farm characteristics, liquidity constraints, and price expectations is shown in Table 1.4. First three columns show the mean and standard deviation (brackets) of variables in the three experim ental groups. The last three columns show the p - value for differences in means across the three groups. As shown in the table the variables are balanc ed across the treatment groups. Data for variables in panel A, B, and C comes from the household survey wh ile data for panel D comes from the lab in the field experiments. The average age of a respondent is 40 years with 7 years of schooling and a househo ld of 5 persons. A farmer cultivates rice on around 7 acres of land and sell 150 maunds of paddy on averag e. Around 64 percent of the farmers report that they are liquidity constrained and around 55 percent of them reported that they were able to acquire a loan. Among those who had taken loans, a large proportion of them were from middleman. The prices reporte d suggest that on average farmers expect the price to increase over the 3 - month period post - harvest. 17 Table 1. 4 : Balance T able Means p - values Variables C T1 T2 [T1=C] [T2=C] [T1=T2] A. Socio Demographic Age (years) 39.27 (0.87) 40.06 (0.95) 39.02 (0.86) 0.718 0.894 0.714 Education (years) 7.05 (0.31) 6.87 (0.36) 7.48 (0.34) 0.898 0.360 0.285 Household Size 4. 76 (0.14) 4.70 (0.14) 4.79 (0.15) 0.180 0.868 0.389 Asset Index 0.062 (0.19) - .026 (0.15) 0.131 (0.16) 0.352 0.976 0.376 B. Farm Characteristics Area (acres) 6.86 (0.33) 6.63 (0.34) 6.60 (0.33) 0.532 0.442 0.966 Commercial Farmer (0/1) 0.87 (0. 03) 0.87 (0.03) 0.90 (0.03) 0.900 0.705 0.628 Quantity Sold (maunds) 146.39 (8.74) 141.25 (9.72) 159.31 (10.08) 0.875 0.204 0.245 Farm Gate Sales (0/1) 0.73 (0.02) 0.73 (0.03) 0.74 (0.03) 0.743 0.678 0.738 C. Credit Constraints Cash Constrained (0/1) 0.64 (0.05) 0.64 (0.04) 0.62 (0.05) 0.877 0.839 0.998 Borrowing (0/1) 0.57 (0.04) 0.53 (0.04) 0.55 (0.05) 0.446 0.993 0.328 Borrowing from Middleman (0/1) 0.47 (0.02) 0.45 (0.02) 0.47 (0.02) 0.576 0.999 0.763 Borrowing for Ag Inputs (0/1) 0.54 (0.03) 0.52 (0.03) 0.54 (0.03) 0.547 0.765 0.809 18 Table 1.4: (cont ) Means p - values C T1 T2 [T1=C] [T2=C] [T1= T2] D. Subjective Price Expectations (Rs/40kg) Min Price at Harvest (Rs/40 kg) 1481.48 (25.82) 1479.19 (28.28) 1496.86 (26.03) 0. 909 0.764 0.541 Max Price at Harvest (Rs/40 kg) 1701.14 (30.44) 1681.40 (38.15) 1721.02 (34.82) 0.720 0.756 0.393 Min Price 3 - months Post Harvest (Rs/40 kg) 1953.96 (39.20) 1930.78 (38.52) 1933.57 (35.85) 0.671 0.679 0.945 Max Price 3 - months Post Har vest (Rs/40 kg) 2221.14 (43.27) 2185.32 (43.90) 2183.18 (49.07) 0.558 0.543 0.968 N 214 203 224 Note: The asset index score is was calculated using PCA and is composed of 23 items, which includes household items, savings in commodities, and modes of transport. 1.6. Results The first subsection reports risk aversion and discount factor estimates. The second subsection reports results from the MXL - EC model, predicted probability of participation calculation, and WTP outcomes. 1.6.1. Risk and Time Preferences Ta ble 1.5 shows the est imated constant relative risk aversion (CRRA) and the discount factor. The average CRRA measure is 0.52 and average discount factor is 0.84. These results imply that majority of smallholder farmers in our target area are risk averse a nd compared to a risk neutral person would require higher returns to participate in WRF. Similarly, the discount factor highlights that value of income earned in the future is eroded due to discounting. Hence, the return from WRF would need to compensate f or it as well to beco me an attractive prospect. Table 1. 5 : Risk Aversion and Time Discounting Parameters Mean SD Min Max Median CRRA 0.51 0.42 0.05 1.5 0.35 Discount Rate 0.84 0.13 0.46 0.99 0.88 19 The table also highlights that there is significant heterogeneity in the risk and time discounting traits as seen by the large range of the measures. This suggests that the utility derived from participating would also vary across farmers. 1.6.2. Program Participation The results show that farmers are price sensitive as higher cost of storage and interest rate lowers utility and probability of participation. Tables 1.6 reports the MXL - EC estimates which show coefficient estimates for cost and interest rat e to be negative and statistically si gnificant at the 1 percent level across all treatment groups. This implies that higher cost and interest rates reduce utility derived from WRF and are associated with a lower probability of participation. The coefficien t for transport is positive and stati stically significant at the 1 percent level across all models. This suggests that farmgate pick up service increases the marginal utility of WRF and increases the probability of participation. The result also shows dis persion around the mean of the sample population highlighting heterogeneity in preferences. The standard deviation (spread) estimates for all elements of the parameter vector are statistically significant at the 1 percent levels which implies that the resp ondents have different individual - spe cific parameter estimates which might be different from the mean parameter estimate. Therefore, the utility derived from WRF attributes could also differ across individuals. 20 Table 1. 6 : MXL - EC Model (Preference Space) Variable Poo led Hi Price Low Price Control Means Cost - 0.13*** (0.01) - 0.14*** (0.01) - 0.11*** (0.01) - 0.14*** (0.01) Low_Int - 0.32*** (0.05) - 0.30*** (0.09) - 0.28*** (0.08) - 0.47*** (0.10) Hi_Int - 1.85*** (0.09) - 1.58*** (0.14) - 1.85*** (0.15) - 2.11*** (0.17) Transport 1.02*** (0.05) 1.04*** (0.09) 0.91*** (0.08) 1.12*** (0.10) OptOut 4.78 (4.17) 10.65 (7.94) - 0.10 (5.85) - 1.30 (4.86) Standard Deviation Cost - 0.13*** (0.01) 0.14*** (0.01) 0.11*** (0.01) 0.14*** (0.01) Low_Int - 0.32*** (0.05 ) 0.30*** (0.09) 0.28*** (0.08) 0.47*** (0.10) Hi_Int - 1.85*** (0.09) 1.58*** (0.14) 1.85*** (0.15) 2.11*** (0.17) Transport 1.02*** (0.05) 1.04*** (0.09) 0.91*** (0.08) 1.12*** (0.10) Opt Out 4.02*** (0.40) 5.34*** (0.66) 5.90*** (0.61) 2.63*** (0.5 7) Error Component 4.02*** (0.40) 0.96 (0.99) 0.30 (0.45) 4.31*** (0.48) Heterogeneity in Mean OptOut x Risk - 4.68** (1.24) - 6.41** (2.67) - 3.56** (1.52) - 2.24** (1.14) OptOut x Time - 6.21 (4.25) - 14.00* (7.91) - 1.40 (6.09) - 1.32 (5.12) AIC 1.067 1.121 1.041 1.028 Sample 641 203 224 214 Observations 5769 1827 2016 1926 Note: The table reports parameter estimates in the preference space. The model was estimated in Nlogit 6 utilizing 500 Halton Draws. Risk aversion and discounting decrease util ity from participation in WRF. However, the effect is statistically significant and consistent across the treatment groups only for risk aversion. The coefficient on the interaction terms of opt - out and risk is negative and statistically significant at the 5 percent level showing that as risk aversion decreases the disutility from no t participating in the WRF program increases, ceteris paribus. Similarly, the coefficient on the interaction term 21 between opt - out and discount rate is negative suggesting that a s the discount factor increases (discounting declines) the disutility from not participating in the WRF program increases, ceteris paribus. However, this is only statistically significant for the high price guarantee group at the 10 percent level. A produ ct design which reduces exposure to risk, if financially feasible, could potent ially increase smallholder participation. Table 1.7 reports results from a probit model which estimated the probability of participation across the three treatment groups. As ex pected, respondents in the price guarantees groups have a higher likelihood of selecting the WRF option and the increase is statistically significant at the 1 percent level as shown by the p - values. Model 2 reports the predicted probability of selecting th e WRF option based on the MXL - EC results which controls for observable and unob servable preferences. The results are consistent with the actual choice but of higher magnitude. The probability of selecting the WRF is higher in the price guarantee groups and the difference between High Price and Low Price group is also statistically si gnificant. Table 1. 7 : Actual and Predicted Choices from MXL - EC model (Preference Space) (1) (2) Variable Actual Choices Predicted Choices Control 0.5 1*** (.03) 0.46*** (.01) High Price Guarantee 0.67*** (.03) 0.71*** (.01) Low Price Guarantee 0.68*** (.02) 0.78*** (.01) Hypothesis Test p - value H 1 : Control = High Price 0.001 0.000 H 2 : Control = Low Price 0.000 0.000 H 3 : High Price=Low Price 0.7 43 0.000 Note: Coefficients cannot be compared across different groups due to the scaling factor. Hence, we used the MXL - EC results to generate individual specific predicted probability of selecting an alternative in each choice set for the three groups. The probab ilities are then pooled across the treatment groups and a Probit model is estimated with village fixed effects and standard errors were clustered at the village level. The probabilities were calculated using the margins command in stata. Column 1 reports t he estimates on actual choices in the DCE for comparison. 22 WTP space estimates show that WTP is higher for credit in the price guarantee groups and lower for farmgate pickup. Table 1.8 reports the WTP estimates 4 with the corresponding p - values re lated to t he difference in marginal WTP for interest and transport attributes. Respondents in the High Price group show a higher WTP for credit compared to the control group. However, the difference is statistically significant only for the High Int level. WTP for t ransport on the other hand is lower in the high price guarantee group and the difference is statistically significant. Comparison between Low Price and Control groups shows that WTP for credit is higher with a price guarantee. However, in this ca se the dif ference is statistically significant only for the Low Int level. WTP for transport on the other hand is lower and the difference is statistically significant. Comparison between the High and Low Price guarantee groups shows WTP for credit is high er for cre dit and lower for transport in the high price guarantee. However, the differences are statistically significant only for Hi Int and transport. Table 1 . 8 : MXL - EC Model Estimates by Treatments (WTP Space) Variable Coefficie nt Standard Error p - value Low_Int 1.84 1.25 0.140 High Int 3.27*** 1.06 0.002 Transport - 2.53*** 0.83 0.002 Low Int 2.52** 1.27 0.0 48 High Int 1.01 1.15 0.380 Transport - 3.83*** 0.68 0.000 Low Int 0.55 1.59 0.723 High Int 2.98** 1.41 0.035 Transport - 2.05* 1.22 0.094 Note: The model was estimated in Nlogit 6 utilizin g 500 Halton Draws . 4 The cost variable was divided by 10 and the estimates were multiplied by 10 to calculate WTP. The model was not estimated with risk and time interactions for two reasons: 1) We are interested in the overall WTP acr oss the three treatment groups and 2) the models did not converge when these interactions were included. 23 These results show that if there is less uncertainty in returns farmers would be willing to pay a higher price for credit but would value farmgate pick - up less. These outcomes provide s ome evidence that contract designs which reduce ex posure to price uncertainty can increase participation and allow the provider to charge slightly higher prices for WRF. 1.7. Conclusion Smallholder farmers continue to face numerous constraints which discourage productivity enhancing investments. WRF has rece ntly be promoted as a viable potential solution to address financial inclusion and market access constraints of farmers. The introduc tion of WRF for smallholder farmers in Hafizabad, Pakistan is a case in point. However, participation by smallholder farmer s was very low despite it being a profitable prospect. We find that external and internal constraints combined erode the profitabil ity of WRF, making it a less attractive prospect. Results show that high transaction costs reduce the attractiveness of WRF as seen by the negative marginal utility on cost and interest rate parameter estimates. We also see that farmers value transport fro m the farm gate to the warehouse, a service which was not offered in the pilot. Risk aversion is another important factor t hat reduces the attractiveness of WRF as the program required farmers to take price risks. The experimental treatment design further validates this conclusion as we see an increase in predicted participation for WRF under the price guarantee groups. An imp ortant finding is that the increase in participation is also high under the low - price guarantee suggesting that smallholder farmers a nd they would be willing to participate as long they are assured that they will not incur a loss. Since, this study uses s tated preference data one can argue that the results suffer from hypothetical bias. This criticism is fair, but it does not diminish the findings of the study which highlights that the existing contract only focused on the constraints of circumstance and f ailed to 24 address the internal constraints of smallholder farmers which made participation an unattractive prospect. These findings ar e particularly relevant for Pakistan as the agriculture value chains are very underdeveloped and WRF has the potential to i mprove and develop these value chain. Importantly, the federal government also recognizes this potential of WRF and the state bank of Pakistan and developed a regulatory framework to operationalize WRF. However, establishing a warehouse network is an expen sive infrastructure endeavor and it is important to understand the markets appetite while designing WRF products, especially those th at would target smallholder farmers. These results also highlight the need to incorporate the preferences of the target pop ulation while designing programs aimed to improve their welfare. 25 APPENDICES 26 APPENDIX A DISCRETE CHOICE EXPERIMENT Now I will present you with some choices regarding a warehousing service. In each of these questions I will present you w ith two versions of a warehousing product and a no - purchase option. The warehousing options vary across t he questions in terms of cost of storage, pickup service from your farm, and future price after 3 months. For each of these questions, we would like to know if you are interested in either of the warehouse product. You can always choose the not interested option. If you are interested in the warehousing option, please also share what proportion of your harvest you would be interested in storing. For exam ple, you will only store 50 out of your 150 bags of paddy. There are a few things you should keep in min d when making these decisions: Assume there are no other options except for the ones we are showing you. All the choices are separate so do not try to remember your choices in the previous questions. In simple words treat each warehouse product option sep arately. Once you have made the choice you cannot go back and change it. Group 1 Script High Price Guarantee Suppose your paddy is ready for harves t and the selling price of paddy is Rs1300 per maund if you take it to the mandi. Looking at past paddy prices there is a good chance that the price of paddy will rise in the future and you can earn more if you sell la ter. A company is offering farmers lik e you the option to store your paddy and earn better income by selling later. The company is guaranteeing to buy the paddy at a premium of Rs1800 after 4 months even if the prices are lower, so you have a guaranteed hi gher income. Credit against the store d paddy is also available if required, for example if you store paddy worth Rs 100,000 you can get credit of upto Rs70,000 at a service charge. You can sell you paddy to anyone and anytime once you repay the credit. Th e sale option is also available at the warehouse. 27 Once the product arrives at the warehouse it will be cleaned, dried, graded, weighed, and bagged by the staff at the warehouse. After the bagging is complete your bags will be tagged using your id number a nd stacked in the warehouse. The wareh ouse staff will regularly check the paddy to ensure that the moisture level is controlled and will be responsible for the safety of the product. The warehouse charges are meant to cover these expenses. The service char ge on loans is meant to cover the cost of employees and travel. Group 2 Script No loss Guarantee Suppose your paddy is ready for harvest and the selling price of paddy is Rs1300 per maund if you take it to the mandi. Looking at past paddy prices there is a good chance that the price of pad dy will rise in the future and you can earn more if you sell later. A company is offering farmers like you the option to store your paddy and earn better income by selling later. The company is guaranteeing to buy the paddy at a premium of Rs1800 after 4 m onths even if the prices are lower, so you have a guaranteed higher income. Credit against the stored paddy is also available if required, for example if you store paddy worth Rs 100,000 you can get credit of upto Rs7 0,000 at a service charge. You can sel l you paddy to anyone and anytime once you repay the credit. The sale option is also available at the warehouse. Once the product arrives at the warehouse it will be cleaned, dried, graded, weighed, and bagged by the staff at the warehouse. After the bagg ing is complete your bags will be tagged using your id number and stacked in the warehouse. The warehouse staff will regularly check the paddy to ensure that the moisture level is controlled and will be responsible for the safety of the product. The wareho use charges are meant to cover these expenses. The service charge on loans is meant to cover the cost of employees and travel. Group 3 Script No Price Guarantee 28 Suppose your paddy is ready for harvest and the selling price of paddy is Rs1300 per maund if you take it to the mandi. Looking at past paddy prices there is a good chance that the price of paddy will rise in the future and you can earn more if you sell la ter. However, the company gives no guarantee about future price. Credit against the stored paddy is also available if required, for example if you store paddy worth Rs 100,000 you can get credit of upto Rs70,000 at a service charge. You can sell you paddy to anyone and anytime once you repay the credit. The sale option is also available at the warehouse. Once the product arrives at the warehouse it will be cleaned, dried, graded, weighed, and bagged by the staff at the warehouse. After the bagging is compl ete your bags will be tagged using your id number and stacked in the warehouse. The wareho use staff will regularly check the paddy to ensure that the moisture level is controlled and will be responsible for the safety of the product. The warehouse charges are meant to cover these expenses. The service charge on loans is meant to cover the cost of employees and travel. Figure 1. 3 : Sample Question (DCE) Option 1 Option 2 Option 3 29 A PPENDIX B RISK AVERSION Instructions This game gives you two options to choose from and the two options are labelled Option A an d Option B like before. Please look at the handout which shows an example of the game. Figure 1. 4 : Example Game If you select Option A you draw a ball from the cylinder on the left and you will get Rs100 if you pick a blue ball w hich will be true every time as all balls are blue. If you select Option B then you draw a ball from the cylinder on the right. You will ge t Rs 2600 if you pick the green ball and Rs 500 if pick the yellow ball. The chance if picking a green ball is one ou t of ten. Here again, there are no correct or incorrect answers. We just want to know what you prefer and would choose. Now play a pr actice game so that you understand the game better. Practice Game As you see each question had two Options: A and B to earn some candies. These two options differ in the number of candies you can win depending on your choices. For example, in question num ber 1, under Option A, you always draw a blue ball and earn three candies for sure. For the 30 same question, if you cho ose Option B, you can earn either five candies if you draw green ball or you earn only 1 candy if you draw the yellow. The number of candie s you can earn in Option A under all the questions remains the same but in Option B, the number of candies you can wi n with the green ball increases, but if you get yellow colored ball it remains the same (1 candy), in all the questions. Figure 1. 5 : Practice Game suppose that question 1 gets selected for actual payment. If you had chosen o ption A you will get 3 candies. Suppose you selected option B for question 1, ask the respondent to pull out one ball from the bag (which h as 9 yellow and 1 green ball), if it is green pay five candies or 31 if it is yellow, then give one candy. Do you under stand the game? If yes, play the real game now. Game Instructions I will ask you to choose between option A and Option B for each of the pairs. The amount of money you can win in Option A is the same Rs 200. On the other hand, in Option B, you can win up to Rs 5600 if you draw a green ball (1 out of 10 chances) and you win Rs 500 if you draw the yellow ball (9 out 10 chances). So, lo ok carefully at the questions as they come, and select your preferred option. You have the complete freedom to selec t Option B for any question or not select Option B at all. Table 1.9 shows the first price list that was used in the risk aversion game. T able 1. 9 : Risk Aversion Choice List 1 32 The respondent was shown one choice task at a time i n a graphical manner (shown below) and asked to make a choice. The respondent answered all 14 questions even if they switched earlier. Figure 1. 6 : Sample Question (Risk Aversion) The following set of questions are the same as be fore except for the fact that now that there are 7 green balls and 3 yellow balls. So the chance of winning the bigger reward are higher as compared to before. The respondent was shown one choice task at a time in a graphical manner (as above) and asked t o make a choice. The respondent answered all 14 questions even if they switched earlier. 33 Table 1. 10 : Risk Aversion Choice List 2 34 A PPENDIX C T IME PREFERENCES Instructions I am giving you a token (hand over a token) and you can exchange this token with me for real money. I will give you two options labelled Option A and Option B offering different amount of money at different times in the future. How much money you receive w ill depend on which option you cho ose. Let me give you an example, [hand over the sheet 1 with the example] you will get Rs1000 in 2 weeks if you select Option A and Rs 1100 in 8 weeks if you select Option B [Show the calendar to the respondent to make the choice easier to understand]. So you get additional money for waiting. As you go down the questions, you can see that Option A does not change but how much you earn in Option B increases. As you can see Option A gives you money in two weeks and Option B g ives you money 2 months later than Option A. We would like to know which option you prefer for each question. I would like to reiterate that there is no wrong or correct answer. We just want to know what options you prefer and hope that you would find the games interesting. Table 1. 11 : Sheet 1 Task Horizon in Months Option A Option B 1 2 1000 1100 2 2 1000 1200 3 2 1000 1300 35 Figure 1. 7 : Calendar Sheet We will now randomly select a question, suppose question 2 in th is game is selected for actual payment. The reward will now be based on your choice 1000 in two weeks if you selected Option A or 1200 in 8 weeks if you had selected Option B. Now let play the actual game. The structure of this game is very similar to the practice game. Option A is for sooner smaller payment and Option B is for larger later payment. The questions differ in the waiting period and the amount in the later period. Let m e remind you again that these questions are eligible to be selected fo r a real payment at the end of the survey. Do you have any questions? If no, then let start the game. 36 Table 1. 12 : Time Discounting Choices Task Horizon in Months Earlier Payment Delayed Payment Annual Interest Rate 1 4 500 517 10 2 4 500 525 15 3 4 500 533 20 4 4 500 542 25 5 4 500 550 30 6 4 500 558 35 7 4 500 567 40 8 4 500 575 45 9 4 500 583 50 10 4 500 592 55 37 BIBLIOGRAPHY 38 BIBLIOGRAPHY Aggarwal , S., Francis, E., and Robinson, J. 2018. Grain today, gain tomorrow: Evidence from a storage experiment with savings clubs in Kenya, Journal of Development Economics , Advance Access published 2018: doi:10.1016/j.jdeveco.2018.04.001 Andersen, S., H arrison, G. W., Lau, M. I., and Rutström, E. E. 2008. Eliciting risk and time preferences, Econometrica , Advance Access published 2008: doi:10.1111/j.1468 - 0262.2008.00848.x Ashraf, N., Karlan, D., and Yin, W. 2006. Tying odysseus to the mast: Evidence from a commi tment savings product in the Philippines, Quarterly Journal of Economics , Advance Access published 2006: doi:10.1162/qjec.2006.121.2.635 Barrett, C. B., Bachke, M. E., Bellemare, M. F., Michelson, H. C., Narayanan, S., and Walker, T. F. 2012. Small holder p articipation in contract farming: Comparative evidence from five countries, World Development , Advance Access published 2012: doi:10.1016/j.worlddev.2011.09.006 Basu, K. and Wong, M. 2015. Evaluating seasonal food storage and credit programs in eas t Indone sia, Journal of Development Economics , Advance Access published 2015: doi:10.1016/j.jdeveco.2015.02.001 Bazzani, C., Caputo, V., Nayga, R. M., and Canavari, M. 2017. TESTING COMMITMENT COST THEORY IN CHOICE EXPERIMENTS, Economic Inquiry , Advance Access pub lished 2017: doi:10.1111/ecin.12377 Working Paper , Advance Access published 2014 Caputo, V., Lusk, J. L., and Nayga, R. M. 2018. Choice experiments are not conducted in a vacuum: The effects of external pri ce information on choice behavior, Journal of Economic Behavior and Organization , Advance Access published 2018: doi:10.1016/j.jeb o.2017.11.018 Caputo, V., Nayga, R. M., and Scarpa, R. 2013. Food miles or carbon emissions? Exploring labelling preference fo r food transport footprint with a stated choice study, Australian Journal of Agricultural and Resource Economics , Advance Access p ublished 2013: doi:10.1111/1467 - 8489.12014 Caputo, V., Scarpa, R., and Nayga, R. M. 2017. Cue versus independent food attribut es: The effect of adding attributes in choice experiments, European Review of Agricultural Economics , Advance Access published 201 7: doi:10.1093/erae/jbw022 Chavas, J. - P. and Holt, M. T. 2006. Economic Behavior Under Uncertainty: A Joint Analysis of Risk P references and Technology, The Review of Economics and Statistics , Advance Access 39 published 2006: doi:10.2307/2109935 Cheung, S. L. 2016. Recent developments in the experimental elicitation of time preference, Journal of Behavioral and Experimental Finance Clark, M. D., Determann, D., Petrou, S., Moro, D., and de Bekker - Grob, E. W. 2014. Discrete Choic e Experiments in Health Economics: A Review of the Literature, PharmacoEconomics Daly, A., Hess, S., and Train, K. 2012. Assuring finite moments for willingne ss to pay in random coefficient models, Transportation , Advance Access published 2012: doi:10.1007 /s11116 - 011 - 9331 - 3 De - Magistris, T., Gracia, A., and Nayga, R. M. 2013. On the use of honesty priming tasks to mitigate hypothetical bias in choice experiment s, American Journal of Agricultural Economics , Advance Access published 2013: doi:10.1093/ajae/aat 052 Duflo, E., Kremer, M., and Robinson, J. 2008. How high are rates of return to fertilizer? Evidence from field experiments in Kenya, in American Economic R eview Fafchamps, M. and Minten, B. 2012. Impact of SMS - based agricultural information on Indian fa rmers, World Bank Economic Review , Advance Access published 2012: doi:10.1093/wber/lhr056 Fischer, E. and Qaim, M. 2014. Smallholder farmers and collective ac tion: What determines the intensity of participation?, Journal of Agricultural Economics , Advance Access published 2014: doi:10.1111/1477 - 9552.12060 Gibson, J. M., Rigby, D., Polya, D. A., and Russell, N. 2016. Discrete Choice Experiments in Developing Cou ntries: Willingness to Pay Versus Willingness to Work, Environmental and Resource Economics , Advan ce Access published 2016: doi:10.1007/s10640 - 015 - 9919 - 8 Greene, W. H. and Hensher, D. A. 2010. Modeling ordered choices: A primer Hensher, D. A., Rose, J. M., and Greene, W. H. 2015. Applied choice analysis Jayne, T. S., Mason, N. M., Burke, W. J., and Ari second - generation agricultural input subsidy programs, Food Policy Jensen, R. 2010. The (Perceived) returns to e ducation and the demand for schooling, Quarterly Journal of Economics , Advance Access published 20 10: doi:10.1162/qjec.2010.125.2.515 Kahneman, D. and Tversky, A. 1979. Kahneman & Tversky (1979) - Prospect Theory - An Analysis Of Decision Under Risk.pdf, E conometrica Lancaster, K. J. 1966. A new approach to consumer theory., The Journal of Political Ec onomy , vol. 74, no. 2, 132 57 Lawless, L., Drichoutis, A. C., and Nayga, R. M. 2013. Time preferences and health behaviour: a 40 review, Agricultural and Food Ec onomics Lin, W., Ortega, D. L., and Caputo, V. 2019. Are Ex - Ante Hypothetical Bias Calibration Met hods Context Dependent? Evidence from Online Food Shoppers in China, Journal of Consumer Affairs , Advance Access published 2019: doi:10.1111/joca.12202 Liu, E . M. 2013. Time to change what to sow: Risk preferences and technology adoption decisions of cotto n farmers in China, Review of Economics and Statistics , Advance Access published 2013: doi:10.1162/REST_a_00295 Lusk, J. L. and Briggeman, B. C. 2009. Food va lues, American Journal of Agricultural Economics , Advance Access published 2009: doi:10.1111/j.146 7 - 8276.2008.01175.x McFadden, D. 1974. Conditional logit analysis of qualitative choice behaviour, in Frontiers in econometrics McKenzie, D., Gibson, J., and Stillman, S. 2013. A land of milk and honey with streets paved with gold: Do emigrants have over - optimistic expectations about incomes abroad?, Journal of Development Economics , Advance Access published 2013: doi:10.1016/j.jde veco.2012.01.001 Miranda, M. J ., Mulangu, F. M., and Kemeze, F. H. 2017. Warehouse Receipt Financing for Smallholders in Developing Countries: Short on Logic, Long on Imagination.: Omotilewa, O. J., Ricker - Gilbert, J., Ainembabazi, J. H., and Shively, G. E. 2018. Does improved storage technology promote modern input use and food security? Evidence from a randomized trial in Uganda, Journal of Development Economics , Advance Access pub lished 2018: doi:10.1016/j.jdeveco.2018.07.006 Ortega, D. L., Wang, H. H., Wu, L., and Olynk, N. J. 2011. Modeling heterogeneity in consumer preferences for select food safety attributes in China, Food Policy , Advance Access published 2011: doi:10.1016/j.f oodpol.2010.11.030 Ricker - f input subsidy programs in africa, Applied Economic Perspectives and Policy , Advance Access published 2013: doi:10.1093/aepp/ppt001 Scarpa, R., Campbe ll, D., and Hutchinson, W. G. 2007. Benefit estimates for landscape improvements: Sequential Bayesian des experiment, Land Economics , Advance Access published 2007: doi:10.3368/le.83.4.617 Scarpa, R., Ferrini, S. , and Willis, K. 2005. Performance of Error Component Models for Status - Quo Effects in Choice Experiments , in Applications of Simulation Methods in Environmental and Resource Economics Scarpa, R., Thiene, M., and Train, K. 2008. Utility in willingness to p ay space: A tool to address confounding random scale effects in destination choice to the Alps, American Journal of Agricultural Economics , Advance Access published 2008: doi:10.1111/j.1467 - 41 8276.2008.01155.x Scarpa, R. and Willis, K. 2010. Willingness - to - p ay for renewable energy: Primary and - generation tec hnologies, Energy Economics , Advance Access published 2010: doi:10.1016/j.eneco.2009.06.004 Shiferaw, B., Hellin, J., and Muricho, G. 2011. Improving m arket access and agricultural productivity growth in Africa: What role for producer organizations and col lective action institutions?, Food Security , Advance Access published 2011: doi:10.1007/s12571 - 011 - 0153 - 0 Tanaka, M., Ida, T., Murakami, K., and Friedm alternative fuel vehicles: A comparative discrete choice a nalysis between the US and Japan, Transportation Research Part A: Policy and Practice , Advance Access published 2014: doi:10.1016/j.tra.2014.10.019 War d, P. S. and Singh, V. 2015. Using Field Experiments to Elicit Risk and Ambiguity Preferences: Behavioura l Factors and the Adoption of New Agricultural Technologies in Rural India, Journal of Development Studies , Advance Access published 2015: doi:10.1080/ 00220388.2014.989996 Warnick, J. C. E., Escobal, J., and Laszlo, S. C. 2011. Ambiguity aversion and portf olio choice in small - scale peruvian farming, B.E. Journal of Economic Analysis and Policy , Advance Access published 2011: doi:10.2202/1935 - 1682.2331 Wi lliam, J. G. and Kaserwa, N. 2015. Improving smallholder farmers access to finance through warehouse rece ipt system in Tanzania, International Journal of Economics and Financial Research , Advance Access published 2015 42 Essay 2 : Does peer comparison encourage adoption of best practices among farmers in Pakistan? Abstract Poor nutrient management has contributed towar ds deteriorating soil quality in Pakistan. Despite years of experience, non - convergence to optimal practices suggests constraints to lear ning. Informing farmers about optimal organic and inorganic fertilizer use, given soil and crop type, can address this challenge. This study takes an experimental approach with treated farmers receiving this information either with or without a peer compa rison, while the control farmers receive no information. Results show that majority of the soils are in poor health, ma nure use is very low, and 93 percent of farmers are underusing fertilizers. The intervention increased manure and urea use in the peer co mparison group. However, urea increase was concentrated among farmers that were applying fertilizers closer to the reco mmended quantities at the baseline. These findings provide some evidence on the effectiveness of peer comparisons and highlights the pres ence of other constraints in addition to information. 43 2.1 Introduction Sustainable agricultural systems are essential fo r addressing poverty and ensuring food security. The current agricultural systems have significantly degraded the natural resource base, especially soils, reducing its productive capacity and efficiency of l and, making food security and poverty alleviation an increasingly difficult task to achieve. According to the Global Land Degradation Assessment (GLADA) land degradation is increasing and almost one - quarter of the total global land area has undergone it. E stimates show that in 2007 about 1.5 billion peop le in some way depended on degraded areas for their livelihoods while in excess of 42 percent of the (Nachtergaele et al., 2010) . Pakistan has also been severely affected by land degradation due to unsustainable land management practices and increasing demands on natural resources (Zia et al., 2004) . Around 68 million hectares of land across Pakistan has been affected and the issue has been worsening in the absence of effective policies (Khan et al., 2012) . Several studies have highlighted that land resource degradation due to nutrient mining is a dominant factor for poor productivity in Pakistan (Ali and Byerlee, 2002; Khan et al., 2012; Malik et al., 2017; Pingali and Shah, 2001) . Since, agriculture continues to be e economy, growth in this sector is important for achieving economic growth and broader developmental goals. Improper use of fertilizers is one of the core factors for declining soil health as it leads to nutrient mining and yield stagnation (Concepcion, 2007) . Each soil and crop ty pe has a specific mineral requirement and under o r overuse of fertilizers can cause damage to the productive capacity of soils (Ahmed and Gautam, 2013; Bumb and Baanante, 1996; Coulter, 2008) . Existing practices in Pakistan have caused t he soil organic matter concentration to fall to extremely low levels reducing its fertility and its ability to efficiently use fertilizers. This is also one of the factors 44 behind decreasing returns to ferti lizer application (Ali et al., 2017; Lal, 2014) . Field experiments in Pakistan suggest that use of compost manure along with inorganic fertilizers can significantly improve soil health and yields (Ibrahim et al., 2008; Sangeetha et al., 2013; Sarwar et al., 2007) . Non convergence to ideal practices de spite decades of experience using fertilizer tech nology suggests barriers to complete adoption. Two streams dominate this discourse, one argues is that there is heterogeneity in cost and/or returns and farmers might have possibly internalized the cost of t his resource degradation. Alternatively, there mi ght be informational and behavioral factors such as perceived risks, learning, and limited attention driving incomplete technology adoption (Janvry et al., 201 6) . This study alleviates the information con straint by providing fertilizer recommendations based on soil tests which provide an objective means to ascertain fertilizer requirements based on crop and soil type. This in itself is not a novel idea but the impact of providing customized information on nutrient management has not been encouraging (Corral et al., 2016; Fishman et al., 2016) . Therefore, this study complements information provision with peer comparison which has been shown to be a usef ul, low cost nudge to alter behavior in various contexts such as energy conservation, residential water use, and charitable giving among other t hings (Cooter et al., 2008; Goldstein et al., 2008; Martin and Rand al, 2008; Schultz et al., 2007) . The paper addresses two main questions regarding fertilizer application: (1) Does information on soil quality improve fertilizer use? and (2) Does including peer comparison to information improve fertilizer use? Moreover , since small farmers face numerous constraints and the intervention only alleviated the information constraint. The study focuses on heterogenous impacts. Incomplete adoption of fertilizers is relatively understudied. Most of the existing work on fertili zer adoption has focused on heterogenous adoption of fertilizer across farmers due to 45 availability, affordabilit y, risk and information (Jack, 2013) . While some authors have looked at differen tial marginal returns and barriers to learning as potential factors for incomplete adoption (Duflo et al., 2008; Janvry et al., 2016; Marenya and Barrett, 2009) . Recently, a few papers have evaluated the role of information and k nowledge on fertilizer application (Corral et al., 2016; F ishman et al., 2016) . However, the results are mixed where Corral et al. (2016) find an increase in fertilizer rate but it is not statistically significant. While Fishman et al. (2016) found no impact of providing the information. This study adds to this sparse literature by evaluating the impact of information on agricultural practices such as fertilizer. Results show that 87 percent of the far mers could recall receipt of the SHC and 75 percent of the reported that they trusted the information on it whic h is very encouraging as Fishman et al. (2016) attribute lack of trust as the reason for null effect. Secondly, the study also contributes to t he literature on social comparisons by testing this nudge in an agricultural setting. A random subset of farmers was also given peer comparisons in addition to information to induce behavioral change. Results show that manure use increased in the peer comp arison group while there was no change in the information group. These results validate earlier finding that inf ormation provision might not be sufficient to change practices. Another contribution this paper makes is that it measures the heterogenous impa ct of information provision which the earlier two studies did not. We find a statistically significant increase in urea use among farmers who were given the peer comparison and were applying urea within one bag of recommended quantity at baseline. These fi ndings provide some evidence that peer comparisons can be a useful level to alter behavior. The remainder of th is paper is as follows. Section 2 outlines the conceptual framework, section 3 explains the experimental design, followed by intervention detail s in section 4, section 5 46 describes the data, section 6 gives the empirical strategy and results, and section 7 gives the conclusion. 2.2 Conceptual Framework improve their performance and conform with their peers (Festinger, 1954). A generalized utility maximization framework with a social comparison function following (Levitt and List, 2007) to understand farmer response to information in the SHCs is given below where each farmer chooses an action , incurs a cost , captures heterogeneity that is inversely proportional to the wealth o f the farmer, is the benefit from the action given other inputs and farmer characteristics , and the Peer Comparison function depends on action , deviation from peers assumed to be quasi - The model shows that farmer behavior is influenced by the subjective cost - benefit, self - evaluation again st an objective standard, and relative to peers. The SHCs provide info rmation regarding (d) and (s) to the farmer following an experimental design (explained in the following section). One random subset of farmers only received (d) in the form of soil heal th which provides the farmer information of his own performance as a s teward of his land against an objective measure. The other random subset of farmers receives this information and information about the soil health of a peer for comparison. The remainin g farmers serve as the control group and receive no information. 47 T he design of our experiment allows us to test the following hypotheses: Hypothesis 1: SHC informs farmers on their optimal use of fertilizers, and farmers adjust their fertilizer use acc ordingly, provided they do not face other binding constraints The firs t and most important assumption that this hypothesis rests on is that information is the only binding constraint. Second, once the information is provided farmers would be able to unders tand it. Third, once farmers understand the information, they will tru st it and update their beliefs regarding optimal fertilizer quantities. Fourth, once the farmers update their beliefs they will act on these changed beliefs. This is strongly linked with the first assumption that information is the only binding constraint. If the farmer faces other constraints such as liquidity, water availability, and fertilizer availability then it will lead to a null result as those constraints are not removed through this intervention. As we expect farmers in developing countries to fa ce these constraints we will focus on the heterogeneous impact of this intervention. Hypothesis 2: Peer Comparison increases the efficacy of information and results in higher complianc e to the fertilizer recommendations. The second hypothesis rests on t he idea that augmenting information with social comparison ons will also raise the aspirati ons of the farmers. For example, without information on the soil health of peers, a farmer could easily assume that they are performing equally as well as everyone else. Social comparison will inform them that a better outco me is possible, encouraging them to adopt the recommendations provided on the card. 48 2.3 Experimental Design The study had three experimental groups (two treatment and a control) shown in figure 2.1. Randomization was done at the individual level and farmers were randomly assigned to one of the groups in each of the 90 villages in the study area. Treatment 1 (Information Only) group received information on soil health score 5 and recommendations on manure and fertilizer use based on the soil and crop type. Trea tment 1 provided farmers with an objective feedback on the fertility of their soil which was also reflective of their ability to manage their land. The aim was to improve learning and knowledge of the farmer and induce change in practices . Treatment 2 (Pee r Comparison) group received the same information and a nudge in the form a peer comparison. Farmers in this group were also given information on the best soil health score among their peers. Treatment 2 was designed learning from earlier studies which had failed to bring change from inf ormation provision. The peer comparison was aimed to stimulate the competitive nature of farmers by comparing them against a better performing peer. Secondly, it also aimed to encourage them and raise their aspiration for a better equilibrium. The interve ntion was implemented in 90 villages spread across four districts of south Punjab. A random sample of agricultural households 6 in each village was selected for the intervention and randomly assigned to the treatment and control groups as shown in Fig 2.1. 5 Soil health information was provided as a score out of ten for simplification. A higher number represented b etter soil health. This information was also presented graphically so that it was cognitively less demanding. 6 Households engaged in the production of wheat or cotton were selected as these are the main crops. This exercise let to the identification of a pproximately 4200 households. 49 Figure 2. 1 : Study Design 2.4 Intervention Description Learning in the agricultural setting is a complicated process for many reasons. First, farmers can choose from numerous inputs and there could potentially be severa l optimal bundles. Secondly, there is a significant unobservable component of the underlying production function which is not completely understood. Third, there are exogenous stochastic shocks to farm output in the form of weather. Fourth, the production process takes a long time and updating of beliefs can usually happen once a year. Finally, r eturns to inputs can be heterogenous conditional on the underlying resource base. This is particularly true in the case of fertilizer response and land quality. Th is intervention provided farmers with objective information on soil health and fertilizer re quirements based on soil testing. In addition, mental accounting and reminders were also used. Details are provided in the sub - sections below. 2.4.1 Soil Health Cards S oil samples were collected, tagged, and delivered to the district soil testing laboratories. Soil samples were not collected from all the farmers but limited to farmers assigned in the treatment groups only. This was done to avoid an effect on the control group through the soil collection Farming Households in the village Control Information Only (T1) Peer Comparison (T2) 50 activity as it could have led them to seek out information on soil testing and fertilizer recommendations leading to some changes in practices. Once the soil testing was complete the information was used to create a soil h ealth score. Organic matter content and availability of macro nutrients (Nitrogen, Phosphoro us, and Potassium) was used to assign a soil health score between 0 to 10 points, with 10 being the ideal score. This score was highlighted on the front of the card along with a labelling of the soil as very poor, poor, medium, and fertile. To make the int erpretation of the score results easier this information was also presented graphically showing the ideal score r. The peer comparison group also included the best soil health score of a peer farmer to ac t as a behavioral nudge. 7 The back side of the card had detailed information on the quantity of fertilizer to be used for kharif crops. The delivery of the cards st arted in end of March before the planting of cotton which starts in the first week of April. Enumerators went through training before the delivery of the card and were also given a script to follow. The soil script was translated in the local language for the enumerators and covered the following points: I. Explain the soil health status to the farmer and help them understand the score given to them. II. Inform them that one of the major reasons for poor soil health is the poor nutrient management by the farmer. III. Explain to the farmer that their soil health can be improved if they use the correct quantities and types of fertilizers and use manure to improv e the organic content of their soils. Farmers were also encouraged to call on the toll - free number given on t he soil health card if they had any follow up questions. 7 The best soil health card was chosen to avoid the boomerang effect as seen in oth er studies where sharing average scores resulted in better performing subjects gravitating towards the mean. The best performing peer was chose n from a cluster of villages based on their proximity. The cluster consisted of 6 villages on average. 51 2.4.2 Mental Accounting and Call Reminders Once the enumerator went through the information o n the soil health card, farmer was asked the amount of land that assigned to cultivation in the upcoming seaso n for cotton. The enumerator then calculated the fertilizer requirement for the farmers based on the land reported and the results of the soil hea lth card. The farmers were then asked to plan when they would go to get the fertilizer for the upcoming season . They were asked to plan the day, time of the day, and the location from where they would procure the fertilizers. This information was also reco rded by the enumerator and reminder calls were made to the farmers around their planned date of purchase. 2.5 Dat a and Empirical Strategy This paper utilizes three waves of data which were collected under the Punjab Economic Opportunities Program. Two rounds of survey were done preintervention and one post intervention. The first survey wave was conducted in August 2 016 and a random sample o f 12,700 households were surveyed in this round across the 90 villages. This dataset was used as the sampling frame to identify households engaged in farming. The second survey wave was conducted in February 2018 and is used as the baseline for the study. The intervention was rolled out in April 2018 (details in the next section) and the post treatment survey was conducted in April 2019. This study focuses on cotton production as it is the main cash crop in this region, has signifi cant forward market linka ges, and is also a very input intensive. The following regression approach was used for balance checks on pre - treatment socio - demographic, soil characteristics, and production variables. 52 The left - hand - side v ariable, , refers to the value for household i in village j. is a dummy village level. Table 2.1 shows the summary statistics and checks for balance ac ross the treatment and control groups. The first three columns show the mean of the variables and the last three columns give the difference in means. Standard errors are reported in parenthesis and th e asterisks show if the differences are statistically s ignificant. Panel A shows household levels variables and Panel B shows plot characteristics for the largest plot cultivated by the household. Panel C reports yield and inputs, averaged across all plots managed by the household. Panel A shows that househol ds were balanced across the experimental groups for non - business assets, agricultural assets, land ownership and education. The asset indices were generated using principal component analysis (PCA) and then normalized. The non - business assets comprised of 34 items such as household furniture, electrical equipment, vehicles, and livestock while the agricultural asset index comprised of 12 items which included different kinds of agricultural machinery and equipment. There is imbalance for age of the household head and size of the households. Household heads are older in the Peer Comparison group (T2) with an average age of approximately 51 years as compared to 50 years in the other two groups. Similarly, t he household size is larger in the Peer Comparison grou p (T2). The mean size is approximately 7.1 in T2 and 7 in the other groups. However, as can be seen in the table these differences even though statistically different are not very different in absolute terms. 53 Table 2. 1 : Balance Table Control T1 T2 T1 vs C T2 vs C T1 vs T2 Panel A. Household Demographic Non - Business Asset Index 0.472 0.447 0.514 - 0.057 0.010 - 0.066 (1.450) (1.309) (1.564) (0.066) (0.082) (0.07 8) Agricultural Asset Index 0.553 0.564 0.598 0.010 0.065 - 0.060 (1.843) (1.687) (1.935) (0.095) (0.106) (0.112) Land Owned (kanals) 24.398 24.581 22.645 0.656 - 1.440 1.625 (54.646) (36.904) (40.620) (2.239) (2.533) (2.219) Education (years) 3. 542 3.760 3.507 0.083 - 0.169 0.214 (4.141) (4.283) (4.181) (0.233) (0.215) (0.217) Household Head Age (years) 49.730 49.464 50.989 - 0.047 1.283* - 1.555* (13.463) (13.375) (13.320) (0.807) (0.725) (0.786) Household Size 6.833 6.945 7.084 0.077 0.2 55* - 0.164 (2.785) (2.886) (2.773) (0.159) (0.151) (0.166) Panel B. Land Characteristics Suffer Waterlogging 0.183 0.184 0.199 0.001 0.017 - 0.015 (0.387) (0.388) (0.400) (0.024) (0.020) (0.017) Suffer Salinity 0.165 0.169 0.172 0.00 4 0.007 - 0.003 (0.371) (0.375) (0.378) (0.023) (0.022) (0.023) Suffer Soil Erosion 0.169 0.134 0.143 - 0.035* - 0.026 - 0.009 (0.375) (0.341) (0.350) (0.020) (0.018) (0.018) Fertility Score (Likert Scale 1 - 5) 4.077 4.075 4.086 - 0.002 0.009 - 0.011 (0.638) (0.626) (0.672) (0.032) (0.035) (0.030) Soil Type: Sandy 0.161 0.139 0.154 - 0.022 - 0.007 - 0.015 (0.368) (0.346) (0.361) (0.018) (0.017) (0.015) Soil Type: Sandy Clay 0.537 0.543 0.559 0.006 0.023 - 0.016 (0.499) (0.499) (0.497) (0.027) (0 .027) (0.027) Soil Type: Clay 0.135 0.141 0.128 0.005 - 0.007 0.013 (0.342) (0.348) (0.334) (0.017) (0.018) (0.018) Soil Type: Clay Loam 0.059 0.050 0.062 - 0.009 0.002 - 0.011 (0.236) (0.219) (0.241) (0.013) (0.012) (0.017) Soil Type: Loam 0.102 0.122 0.096 0.020 - 0.006 0.027 (0.303) (0.328) (0.294) (0.017) (0.015) (0.019) Access to Tubewell 0.891 0.911 0.896 0.020 0.005 0.015 (0.311) (0.285) (0.305) (0.019) (0.023) (0.016) Access to Canal 0.740 0.750 0.736 0.010 - 0.004 0.015 (0.439) (0.433) (0.441) (0.032) (0.034) (0.020) 54 Table 2.1: ( c ont ) Control T1 T2 T1 vs C T2 vs C T1 vs T2 Panel C. Production (Cotton) Yield (maund/kanal) 1.775 1.784 1.818 0.008 0.043 - 0.035 (1.268) (1.130) (1.241) (0.079) (0.084) (0.064) Manure (maund/kanal) 4.007 4.040 3.047 0.034 - 0.960* 0.993 (11.124) (19.349) (9.158) (0.876) (0.482) (0.828) DAP (kgs/acre) 40.902 41.804 42.794 0.902 1.892 - 0.990 (32.531) (33.656) (39.455) (1.950) (1.899) (2.021) Urea (kgs/acre) 71.734 70.072 72.311 - 1.662 0.577 - 2.239 (58.049) (55.801) (61.854) (3.570) (3.173) (3.382) Use Potassium 0.008 0.010 0.011 0.002 0.003 - 0.001 (0.089) (0.100) (0.106) (0.005) (0.006) (0.004) N 1,012 597 617 1,609 1,629 1,214 Notes: This table shows the balance test for cotton farmers. Columns 1 - 3 provide the mean of each characteristic for the control group (C), information group (T1), and peer comparison group (T2), respectively. Columns 4 - 6 show the difference in mean ac ross the groups. Standard errors are give n in the parenthesis and the asterisks denote statistical significance. Panel A reports household variables which are continuous. Panel B reports soil characteristics of the largest plot managed by the household. Al l the variables are binary expect for fer tility which is a Likert scale. Panel C checks for balance in production variables where yield is given in maunds per kanal. Urea and DAP is given in kg per acre. Manure is maunds per kanal. All variables are contin uous nary variable. Statistical significance is denoted by: *** p<0.01, ** p<0.05, * p<0.1 Panel B shows that plots across the experimental groups are similar across a broad range of properties such water logging, salin ity, fertility, soil type, and access to irrigation. The only statistically significant difference is for soil erosion where farmers in the control group report a higher incidence. However, the difference again is not large in absolute magnitude Panel C sh ows balance on output and inputs. Cotton yield along with DAP, Urea and Potash application is balanced across the experimental groups. However, manure application is not balanced and statistically significant. Manure application is lower in the Peer Compar ison group (T2) where manure application is a maund lower per kanal. This difference is also large in terms of absolute magnitude. Therefore, regressions are also estimated after controlling for these variables. 2.6 Estimation Methods First, I estimate recall of SHC receipt and trust in the informat ion as they are the prerequisite for having any impact on farmer practices. Then the main outcome of interest; manure application 55 and heterogenous impact on fertilizer application is estimate d. The empirical specifications are explained in the subsections below. Since the assignment to treatment and control groups was random, we can measure the causal impact of the two treatment arms on outcomes of interest. The regressions are run at the hou sehold for cotton. First, a simple linear regression model withou t controlling for any village level fixed effects or pre - treatment covariates is estimated. Village fixed effects and controls are these subsequently included in the specifications. ANCOVA is used to improve the power of the study as it allows us to retain observations with missing values at the baseline. These observations were assigned a zero value and controlled for through an indicator variable in the regressions. The improvement in power is significant over the difference - in - differences (DID) estimator as a DID with equal power to the ANCOVA would require twice the sample size (McKenzie, 2012) . The main specification used to measure the impact is given below. where is the outcome variable of individual i in village j at the endline, and are indicator variables for treatment assignment, is a vector of controls, is an indicator variable if the information for the household is missing in the baseline, is the error term. and are the coefficients of interest and capture the intent - to - treat (ITT) effect of provin g the inform ation. Farmers in developing countries face numerous constraints, alleviating only one constraint (information in our case) might not be sufficient to induce behavioral change (Karlan et al., 2014) . Hence, heterogenous impact of the intervention are estimated. The sample is restricted to those who were underusing fertilizer compared to the r ecommended q uantities, which is 90 percent of the sample. The deviation from the recommended quantity at baseline is used to generate an indicator variable which takes the value 1 if the deviation is less than a bag and zero otherwise. 56 Balance was also tes ting for thi s restricted sample and the results were same as for the original sample. The table is shown in the appendix. This indicator is a proxy for liquidity as those already using higher quantities of fertilizer should not be facing this constraint. T he main spec ification used is given below where all the variables are the same except for the interaction term where Dev is the indicator for deviation from the recommenda tion at the baseline. I also estimate a Difference in Difference version of the above specification for robustness. 2.7 Results This section presents the results of the study. The first subsection reports the take - up of the SHC followed by the i mpact of receiving information and peer comparison on our outcomes of interest organic and inorganic fertilizer use. 2.7.1 Take - Up R ecall and trust variables provide important information on the receptiveness of the SHC s. We use these two variables as proxy for the take - up of the information as they are prerequisite to any change in beliefs or nutrient management practices of farmers in our treatment sample . Farmers were asked if they can recall r eceiving a SHC and did they trust the information provided 8 . R esults show that close to 90 percent of the respondents could recall receiving a SHC and close to 70 percent stated that they trusted the information on the SHC. l ogit regressions with village fixed effects and cluster the standard errors at the village le vel were performed. Table 2.2 8 Only fa rmers in the treatment groups were asked these question as we felt that asking this question from the control group would prime them and lead them to misreporting. Also, based on our knowledge there was no other intervention happening during this period wh ich delivered SHC to farmers. 57 and 2.3 reports the average probabilities of being able to recall receipt of the card and trusting the information. the two treatment arms, across the different sp ecifications. Table 2. 2 : Predicte d Probabilities of Recalling Receipt of SHC (1) (2) (3) VARIABLE Received SHC Received SHC Received SHC Information a .932*** .879*** .878*** (.012) (.011) (.011) Peer Comparison b .918*** .858*** .860*** (.014) (.011) (.010) Observation s 1888 1082 1082 p - value (a vs b) 0.267 0.353 0.393 Village Fixed Effects No Yes Yes Controls No No Yes Note: This table reports the average probabilities of recalling receipt of SHC based on a logistic regression (standard errors are reported in pare nthesis). The results show that the average probability of recalling receipt of SHC would be around 90 percent if everyone in the data was given the information or the peer comparison treatment. This question was only asked from the treatment group who wer e delivered the card under the treatment. Including village fixed effects reduces sample as there was no variation in responses within the village. The sample for this analysis includes everyone who was given the card irrespective of the crop they grow. St atistical significance is denoted by: *** p<0.01, ** p<0.05, * p<0.1 Table 2. 3 : Predicted Probabilities of Trust in SHC (1) (2) (3) VARIABLE Trust SHC Trust SHC Trust SHC Information a .742 *** .720*** .720*** (.024) (.009) (.009) Peer Comparison b .762 *** .742*** .742*** (.021) (.009) (.008) Observations 1888 1742 1742 p - value (a vs b) 0.249 0.236 0.236 Village Fixed Effects No Yes Yes Controls No No Yes Note: This table reports the average probabilities of trust in SHC based on a logistic regression (standard errors are reported in parenthesis). The results show that the average probability of trusting the information on the SHC would be around 72 percent if everyone in the data was given the information treatment or the peer comparison treatment. This question was only asked from the treatment group who were delivered the card under the treatme nt. Including village fixed effects reduces sample if there was no variation in responses across the village. St atistical significance is denoted by: *** p<0.01, ** p<0.05, * p<0.1 58 2.7.2 Manure Use In this subsection, we test whether farmers change their practice of using manure. This was one of the primary advices given to farmers as the organic matter in soils was be low the ideal threshold. The dependent variable is the quantity of manure applied in maunds per kanal for cotton. Tobit model was estimated with the lower bound at zero and no upper bound and the results are reported in Table 2.4. Table 2. 4 : Tobit Estimates of Manure Applied (maunds/kanal) (1) (2) (3) VARIABLE Manure (maund/ kanal) Manure (maund/kanal) Manure (maund/kanal) Information a 4.65** 4.66** 4.79*** (2.03) (2.00) (2.03) Peer Comparison b 6.731*** 5.57*** 6.58* ** (2.45) (2.34) (2.38) Manure baseline 0.23* 0.19 0.19 (0.12) (0.12) (0.12) Missing Dummy - 0.67 0.53 0.57 (2.87) (3.45) (3.44) Constant - 19.38*** - 17.57*** - 9.47 (4.82) (4.07) (8.28) Observations 2226 2226 2226 p - value (a vs b) 0.333 0.3 67 0.416 Village Fixed Effects No Yes Yes Controls No No Yes Notes: The dependent variable is maunds of manure used in a kanal of land. This table provides the ITT estimates of the treatments on manure use. Information and peer comparison are binary var iables indicating treatment assignment. Manure baseline is the q uantity of manure that was applied at baseline. Control variables include education, household size, non - business and agricultural asset indices, soil type, agriculture training, soil fertilit y, access to irrigation, waterlogging, salinity, erosion, and la nd ownership. All regressions are estimated using Tobit and the standard errors are clustered at the village level. Statistical significance is denoted by: *** p<0.01, ** p<0.05, * p<0.1 The results show that manure use increased in both treatment arms a nd the increase was statistically significant. The increase in manure application is higher in the peer comparison group as shown in the table, however, the difference is not statistically sig nificant but substantially large in absolute terms. This outcome provides some support to the hypothesis that peer comparisons can act as a good encouragement device and improve the efficacy of information. Since an 59 ANCOVA specification was used, the basel ine value of manure use was also included in the regression alon g with an indicator variable to control for missing values in the baseline. The results show that they are not strongly correlated to manure use at endline. 2.7.3 Fertilizer Use Earlier interventi ons in India and Mexico which only alleviated the information constraints on fertilizer use have not been very successful (Corral et al., 2016; Fishman et al., 2016) . Lack of trust and availability of fertilizers were cited as the factors responsible in India and Mexico respec tively. These results validate the finding by Karlan et al. (2014) that smallholder farmers face many constraints which need to be removed simultaneously to induce behavioral change. Therefore, this paper focuses on the heterogenous impac t as compared to t he overall change in fertilizers as that might be attenuated. First, I restrict my sample to farmers that were underusing fertilizer (90 percent of the sample). Then I generate a variable that captures deviation from the recommended quant ity at baseline 9 . This variable takes the value 1 if the deviation is less than a bag and zero otherwise. Results are given in table 2.5 which show a statistically significant increase in Urea application for farmers in the peer comparison group that were underusing by les s than 1 bag. Urea use increased in peer comparison group by 10 kg per acre among farmers that were within 1 bag deviation compared to those who were further away. The coefficient is statistically significant at the 10 percent level and consistent across the three specifications. The impact in the information only group is negligible suggesting the peer comparison was a successful encouragement design for this sub - sample of farmers. 9 Soils tests were only conducted for treatment groups and exact deviation from mean cannot be calculated for the control group. I impute values for these observations and use the median recommended quantities at village leve l. 60 Table 2. 5 : Heterogenous Treatment Effect on Urea (1) (2) (3) VARIABLE Urea (kg/acre) Urea (kg/acre) Urea (kg/acre) a 2.63 2.20 2.05 (4.51) (4.26) (4.35) b 10.03** 9.22* 9.11* (4.63) (4.83) (4.88) Dev - 5.48 4.84 5.97 (6.09) (4.21) (4.21) Inform ation - 2.89 - 2.89 - 2.91 (3.08) (2.62) (2.56) Peer Comparison - 1.35 - 0.64 - 0.47 (2.84) (2.64) (2.63) Missing Dummy 18.58*** 2.16 1.45 (6.98) (5.41) (5.43) Urea baseline 0.18* - 0.01 - 0.03 (0.09) (0.06) (0.06) Constant 79.02*** 73.36*** 74.38** * (6.66) (3.65) (10.00) Observations 2035 2035 2035 p - value (a vs b) 0.182 0.198 0.199 Village Fixed Effects No Yes Yes Controls No No Yes Notes: This table provides the coefficient estimates for heterogenous treatment effects based on the devi ation from recommended amounts at the baseline. Dev is an indicator variable which takes the value of 1 if deviation was less than 1 bag and 0 otherwise. The the impact for farme rs in the treatment groups that were within a deviation of 1 bag. Urea baseline is the quantity used at baseline and missing dummy controls for values that we re missing in the baseline. Control variables include education, household size, non - business and agricultural asset indices, soil type, attended agriculture training, soil fertility, access to canal and tube well, waterlogging, salinity, erosion, and land owned. Standard errors are clustered at the village level. Statistical significance is denoted by : *** p<0.01, ** p<0.05, * p<0.1 2.8 Conclusion I find that approximately 90 percent of farmers could recall receiving the SHC and around 72 percent stated th at they trusted the information provided on the cards. This suggests that the intervention was suc cessful in proving improved information to majority of the farmers. Delivery of SHC led to an increase in manure application in the both treatment groups. The magnitude of impact was substantially larger in the peer comparison group but not statistically s ignificant. Urea use also increased among across the two treatment groups, however, in this case the impact was 61 only statistically significant in the peer com parison group among those farmers that were close to the recommended quantities at the baseline. The results provide some evidence that peer comparisons can be used in an agricultural setting to promote more sustainable practices. The results show that th e nudge works and leads to impacts of higher magnitude, however it is not strong enough to have an impact which is statistically different than the traditional mechanism of information dissemination. The heterogenous impact on fertilizer use also provides suggestive evidence that other binding constraints might be limiting use of fertilizers according to recommendations. Farmers who were already using high quantities of fertilizer might not be facing as stringent liquidity constraint as those very far from it and when they were given information on correct fertilizer recommendations that could improve t heir soil health, they changed their practices. The impact on manure application and the lack of impact on DAP further supports the financial constraints hypo thesis as the cost of DAP is almost twice as that of Urea whereas manure is usually available at a very low cost. However, an alternative mechanism could be that those who were operating far from the recommended quantities discounted the recommendations. This points towards confirmation bias, because if the farmers felt they were using the correct or near correct quantities they would not accept recommendation that asks them to change application by more than a bag per acre. However, we cannot identify th e mechanism at play for the heterogenous impact and more research is needed to disentangle the rea sons for low adoption. The findings of this study have important policy relevance as soils in Pakistan have been severely affected by land degradation and de sertification due to farm practices and environmental factors. Degrading soils have made increasin g agricultural productivity an increasingly difficult task to achieve. This has important food security implications for Pakistan which is has a 62 burgeoning po pulation and an economy that is heavily dependent on agriculture. This experiment provides some ev idence on the role of information dissemination programs to encourage adoption of sustainable practices through soil health cards. It also highlights that the impact of such programs can be intensified through low cost peer comparisons. Finally, the study highlights that in the presence of other binding constraints (e.g. liquidity, apriori bias) programs that only alleviate the information constraint would not be enough. 63 APPENDIX 64 APPENDIX Table 2. 6 : Balance Ta ble (Restricted Sample for Heterogeneity ) (1) C (2) T1 (3) T2 (4) [CvsT1] (5) [CvsT2] (6) [T1vsT2] Panel A. Household Demographic p - values Non - Business Asset Index 0.502 (0.096) 0.495 (0.101) 0.588 (0.109) 0.007 - 0.085 - 0.093 Agricultural Asset Index 0.592 0.616 0.644 - 0.024 - 0.052 - 0.028 (0.094) (0.100) (0.108) Land Owned (kanals) 25.259 26.096 23.461 - 0.836 1.798 2.634 (2.806) (2.229) (2.297) Household Head Education 3.593 3.943 3.672 0.349 - 0.079 0.270 (0 .229) (0.229) (0.245) Household Head Age 49.871 49.588 50.932 0.283 - 1.061 - 1.344 (0.677) (0.718) (0.789) Household Size 6.892 7.010 7.115 - 0.117 - 0.222 - 0.105 (0.106) (0.142) (0.154) Panel B. Land Characteristics Suffer Waterl ogging 0.177 0.199 0.207 - 0.023 - 0.030 - 0.008 (0.031) (0.038) (0.034) Suffer Salinity 0.169 0.178 0.179 - 0.009 - 0.010 - 0.001 (0.027) (0.032) (0.030) Suffer Soil Erosion 0.154 0.134 0.124 0.020 0.030* 0.010 (0.016) (0.020) (0.019) Fertility (Likert Scale 1 - 5) 4.078 4.069 4.089 0.009 - 0.010 - 0.020 (0.027) (0.033) (0.035) Soil Type: Sandy 0.166 0.142 0.156 0.025 0.010 - 0.015 (0.022) (0.023) (0.027) Soil Type: Sandy Clay 0.527 0.540 0.559 - 0.014 - 0.033 - 0.019 (0. 027) (0.028) (0.031) Soil Type: Clay 0.139 0.140 0.137 - 0.001 0.002 0.002 (0.017) (0.018) (0.019) Soil Type: Clay Loam 0.054 0.052 0.055 0.002 - 0.001 - 0.003 (0.009) (0.012) (0.012) Soil Type: Loam 0.106 0.123 0.090 - 0.016 0.016 0.032 (0.013) (0.022) (0.017) Access to Tubewell 0.882 0.900 0.893 - 0.018 - 0.010 0.008 (0.028) (0.028) (0.028) Access to Canal 0.769 0.774 0.753 - 0.005 0.016 0.021 (0.045) (0.041) (0.043) N 883 522 531 65 Table 2. 6 : ( c ont ) (1) C (2) T1 (3) T2 (4) [CvsT1] (5) [CvsT2] (6) [T1vsT2] Panel C. Production (Cotton) Difference in Means Yield (maund/kanal) 1.763 1.818 1.846 - 0.55 - 0.083 - 0.028 (0.091) (0.094) (0.096) Manure (maund/kanal) 4.007 4.040 3.047 - 0.034 0.960* 0.993 (0.445) (0.898) (0.461) DAP (kgs/acre) 43.078 42.793 44.443 0.285 - 1.364 - 1.649 (1.396) (1.606) (1.814) Urea (kgs/acre) 76.065 72.490 75.294 3.574 0.771 - 2.803 (3.319) (3.228) (3.433) Use Potassium 0.007 0.011 0. 011 - 0.005 - 0.005 0.000 (0.003) (0.004) (0.004) N 883 522 531 Notes: This table provides a check on the randomization for the full data. Columns 1 - 3 provide the mean (and standard errors) of each baseline characteristic for the control group ( C), information group (T1), and peer comparison group (T2), respectively . Columns 4 - 6 give difference in means. Statistical significance is denoted by: *** p<0.01, ** p<0.05, * p<0.1 66 Figure 2 .2: Soil Health Card 67 Figure 2 68 BIBLIOGRAPHY 69 BIBLIOGRAPHY Growth Structure, Polic ies, Performance, and Impacts, in Agriculture and the Rural Economy in Pakistan Punjab: A Decomposition Analysis, Economic Development and Cultural Change, Advance Acc ess published 2002: doi:10.1086/342759 Bumb, B. L. and Baanante, C. A. 1996. Role of fertilizers in food security and protecting the environment to 2020, IFPRI Discussion Paper Concepcion, R. 2007. Sustainable Fertilization Management of Croplands: The Phi lippines Scenario.: Cooter, R. D., Feldman, M ., and Feldman, Y. 2008. The misperception of norms: The psychology of bias and the economics of equilibrium, Review of Law and Economics, Advance Access published 2008: doi:10.2202/1555 - 5879.1222 Corral, C., Gi né, X., Mahajan, A., and Seira, E. 2016. A Mu ltiple Intervention Approach to Increasing Technology Adoption with a view towards scaling - up: evidence from Mexico: Coulter, J. K. 2008. Unwelcome Harvest: Agriculture and Pollution. By G. R. Conway and J. N. P retty. London: Earthscan Publications Ltd (19 91), pp. 672, £15.00. ISBN 1 - 85383 - 036 - 4., Experimental Agriculture, Advance Access published 2008: doi:10.1017/s0014479700020226 Duflo, E., Kremer, M., and Robinson, J. 2008. How high are rates of return to fertilizer? Evidence from field experiments in K enya, in American Economic Review Fishman, R., Kishore, A., Rothler, Y., Ward, P. S., Jha, S., and Singh, R. K. P. 2016. Can information help reduce imbalanced application of fertilizers in India?, IFPRI Disc ussion Paper, Advance Access published 2016 Gold stein, N. J., Cialdini, R. B., and Griskevicius, V. 2008. A Room with a Viewpoint: Using Social Norms to Motivate Environmental Conservation in Hotels, Journal of Consumer Research, Advance Access published 2 008: doi:10.1086/586910 Ibrahim, M., Hassan, A. U., Iqbal, M., and Valeem, E. E. 2008. Response of wheat growth and yield to various levels of compost and organic manure, Pakistan Journal of Botany, Advance Access published 2008 Jack, B. K. 2013. Constrain ts on the adoption of agricultural technologies in developing countries., Literature review, Agricultural Technology Adoption Initiative, J - PAL (MIT) and CEGA (UC Berkeley)., Advance Access published 2013 70 Janvry, A. de, Macours, K., and Sadoulet, E. 2016. Learning for adopting: Technology adoption in de veloping country agriculture, in Policy briefs from the workshop organised by FERDI and SPIA on June 1 - 2, 2016 in Clermont - Ferrand Karlan, D., Osei, R., Osei - Akoto, I., and Udry, C. 2014. Agricultural decis ions after relaxing credit and risk constraints, Quarterly Journal of Economics, Advance Access published 2014: doi:10.1093/qje/qju002 Khan, M. A., Ahmad, M., and Hashmi, H. S. 2012. Review of available knowledge on land degradation in Pakistan Lal, R. 201 4. Societal value of soil carbon, Journal of Soi l and Water Conservation, Advance Access published 2014: doi:10.2489/jswc.69.6.186a Levitt, S. D. and List, J. A. 2007. What Do Laboratory Experiments Measuring Social Preferences Reveal About the Real World? , Journal of Economic Perspectives, Advance Acce ss published 2007: doi:10.1257/jep.21.2.153 Malik, S., Shujat, A., Riaz, K., and Whitney, E. 2017. Chapter 2: Agriculture, Land, and Productivity in Pakistan, in Agriculture and the Rural Economy in Pakistan Marenya, P. P. and Barrett, C. B. 2009. State - co nditional fertilizer yield response on Western Kenyan Farms, American Journal of Agricultural Economics, Advance Access published 2009: doi:10.1111/j.1467 - 8276.2009.01313.x Martin, R. and Randal, J. 2008. How is donation behaviour affected by the donations of others?, Journal of Economic Behavior and Organization, Advance Access published 2008: doi:10.1016/j.jebo.2007.08.001 McKenzie, D. 2012. Beyond baseline and follow - up: The case for more T in experiments, Journal of Development Economics, Advance Access published 2012: doi:10.1016/j.jdeveco.2012.01.002 Nachtergaele, F. O., Petri, M., Biancalani, R., and Lynden, G. Van. 2010. An Information database for Land Degradation Assessment at Global Level:, date last accessed at Global Land Degradation Information System ( GLADIS ) Pingali, P. and Shah, M. 2001. Policy re - directions for sustainable resource use: The rice - wheat cropping system of the Indo - Gangetic Plains, Journal of Crop Production, Advance Access publ ished 2001 Sangeetha, S. P., Balakrishnan, A., a nd Devasenapathy, P. 2013. Influence of Organic Manures on Yield and Quality of Rice (<i>Oryza sativa</i> L.) and Blackgram (<i>Vigna mungo</i> L.) in Rice - Blackgram Cropping Sequence, American Journal of Plant Sciences, Advance Acc ess published 2013: doi:10.4236/ajps.2013.45141 Sarwar, G., Hussain, N., Schmeisky, H., and Muhammad, S. 2007. Use of compost an environment friendly technology for enhancing rice - wheat production in Pakistan , Pakistan Journal of Botany, Advance Access pub lished 2007 Schultz, P. W., Nolan, J. M., Cialdini, R. B., Goldstein, N. J., and Griskevicius, V. 2007. The constructive, destructive, and reconstructive power of social norms: Research article, 71 Psychological Science, Advance Access published 2007: doi:10. 1111/j.1467 - 9280.2007.01917.x Zia, M. S., Muhmood, T., Baig, M. B., and Aslam, M. 2004. Land and environmental degradation and its amelioration for sustainable agriculture in Pakistan., Sci Vision, Advance Access published 2004 72 Essay 3 : A Dynamic Model for Ware house Rece ipts Financing Demand in a Developing Country Abstract This paper studies the dynamics of warehouse receipts financing (WRF) demand by small scale risk averse farmers in Pakistan. A dynamic model is used to investigate how risk and time preferenc es, transa ction costs, and uncertainty reduce demand for WRF, and even lead to non - participation in the program. The model is calibrated and solved for a representative small - scale farmer that grows paddy. Results show high transaction costs to be a major barrier to participation. Similarly, expectations about future prices also affect participation which drops to zero if the subjective probability of prices falling goes beyond 10 percent. 73 3.1. Introduction Warehouse receipts fina ncing (WRF) has been promoted as a potential solution to some of the fundamental issues that affect agricultural markets in developing countries (Aggarwal , Francis, & Robinson, 2018; Basu & Wong, 2015; Burke, 2014) . WRF essentially provides small farmers access to storage facilities where their product is weighed, graded, and agglomerated. The storage organization issues a receipt st ating the value of the product stored. This warehouse receipt can be used as collateral to acquire loans from financial institutions such as banks. WRF gives farmers bargaining power as they can choose time of sales, provides access to downstream buyers du e to quality control an d higher volumes, and most importantly access to formal financial institutions for loans. The establishment of a mechanism under which warehouse receipts can be traded has the potential of modernizing the agricultural value chains, i mproving market practic es on quality control, standardization of product, and promoting exports. However, the demand for WRF programs has been weak and these services mostly end up being used by large farmers, traders, and processors (Willi am & Kaserwa, 2015) . Several authors have postulated that poor farming hou seholds face multiple constraints which need to be relaxed simultaneously (Duflo, Kremer, & Robinson, 2011 ; Karlan et al., 2014) . Lack of WRF take up also suggests that access to a storage or financial institutions are not be the only binding constraints. Other aspects such as risk a version, present bias, and uncertainty also play an important role in the d ecision making of farmers. Since, WRF has the potential to significantly improve the agricultural value chains in developing countries it is important to understand under what condi tions small farmers would be willing to participate. WRF requires farmers to take on the role of speculators who are expecting to make money off future prices increases. However, in the absence of financial products that can be used to hedge, 74 farmers have to take on the entire risk of price volatility. We have ample evidence tha t small farmers are generally risk averse might not invest in options which have risky cash flows (Feder, Just, & Zilberman, 1985; Liu, 2013; Ward & Singh , 2015) . Given the possibility of a downward price shock it is possible that WRF is not a profitable venture for small farmers under the expected profit margins. Apart from risk considerations, storage is also a choice between foregoing current income f or future income. This is important for farmers as they earn bulk of their income in lump sum at harvest and engaging in WRF would mean less cash available in the current period to payoff outstanding debt, meet current consumption needs, invest in next cro p cycle, and plan for anticipated and unanticipated expenses during the post - harvest period. These factors individually or collectively can make WRF non - participation a subjectively rational choice, making it important to understand the conditions under wh ich participation is feasible for a small farmer. Risk and time preferences introduce dynamic considerations into storage and marketing choices of small farmers. The main goal of t his paper is to explicitly model the dynamics of WRF in the context of Paki stan. This paper adds to the existing literature on WRF programs, most of which have taken a reduced form approach to evaluate the benefits of WRF (Aggarwal et al., 2018; Basu & Wong, 2015; Burke, 2 014) . Miranda et al. (2017) develop a dynamic model to evaluate WRF for an average smallholder Ghanaian farmer while and find WRF not to be a viable product for small farmers. The motivation for this paper is to estimate demand for WRF in Pakistan which has two agricultural cycles instead of one due to irrigation, most farmers are commercial growers, and there is a lot interest by th e government to establish a WRF system in the country. This paper employs dynamic programming to solve a two - period model for an agricultural year. Even though the model is for two period its findings are clear into how risk, time, and uncertainty undermi ne the benefits of WRF. The model is simulated for a representative farmer 75 who grows rice as a cash crop on 5 acres of land and has an average harvest of 150 maunds. The future expectations, and credit constraints. Results derived show that farm ers will not store using modern approach of storing in silos due to high cost. However, under the traditional system which is 25 percent cheaper farmers would will store 24 percent of the harvested paddy. Sensitivity analysis further supports this claim as participation falls to zero if interest rate on loans offered against stored grain goes beyond 15 percent. This also highlights that liquidity is an important constraint for farmer s who need the cash advance through WRF to participate. Results also show t hat farmers will not participate if the subjective probability of prices falling is more than 10 percent. Similarly, risk aversion also depresses the demand for storage as expected and the quantity stored increases as the farmer becomes less risk averse. T hese results highlight that reducing exposure to risk would also improve demand for WRF. Hence, a product design which reduces the transaction cost of participation and reduces futu re price uncertainty can significantly improve participation. Transaction c osts can be decreased by registering farmers and forming small community groups so that costs can be reduced by economies of scale. Similarly, providing information about past price s and future expected prices can help the farmers develop price expectation which are closer to the true distribution. Alternatively, developing financial products through which the warehouse management companies can hedge their risk to price shocks can re duce uncertainty in future prices making WRF an investment with less volati le cash flows for warehousing companies, banks, and the farmers. The rest of the paper is organized as follows. Section 2 outlines the dynamic model for a standard WRF product and decision rules are then presented under these scenarios. Derivations are in itially shown for a scenario without liquidity constraint and perfect risk pooling, these 76 constraints are then included in the model and their effect on demand is investigated. Sect ion 3 lays out the specific functional form and explains the data collected to calibrate its parameters for a representative rice grower in central Punjab. Section 4 discusses the results from the analysis and shows how delayed payments and a reduction in price uncertainty through price floor can increase demand. Section 5 gives concluding comments. 3.2. Dynamic Storage Model Consider a small farmer who has the option of participating in a WRF program, which allows intertemporal arbitrage and access to more re munerative markets through quality control and agglomeration of grain. Howe ver, intertemporal arbitrage inherently carries the risk of loss if future prices do not rise sufficiently. Also, it requires a compromise between current income over future income. Finally, beliefs about future also effect the choice about either selling now or storing and selling in the future after considering several factors. I use a stylized model to represent the decision making of the farmer under risk aversion, time preferen ces and uncertainty. Risk and time preference play an important role in the decision making of individuals especially when it involves makes choices which have a time dimension and uncertainty such as technology adoption, investment, migration, education, and health (Ashraf, Karlan, & Yin, 2006; Engle Warnick, Escobal, & Laszlo, 20 11; Jensen, 2010; Lawless, Drichoutis, & Nayga, 2013; McKenzie, Gibson, & Sti llman, 2013) . In any given year there are two time periods the paddy harvest period and the wheat production period. In each period utility is a function of income and the obje ctive of the farmer is to maximize the present value of utility derived ove r these two periods. Period one starts with the stock of freshly harvested paddy and there are no carryovers from the previous period. The farmer decides how much of this paddy to s tore and how to sell at the prevailing price. The income generated from sal es is then allocated 77 between saving and investment in the wheat crop which yields a return at the end of the second period. The farmer has a subjective probability regarding price o f paddy at the end of period two (when he liquidates the stock) and believe s that the prices will increase with a probability to earn him a profit from storage. sed as follows. I closely follow the approach used by Karlan et al. (2014): Before grain is stored it needs to be transported, dried, cleaned, and graded. Once stored the farmer receives a warehouse receipt which can be used to acquire a loan which is equal to a fraction of the value of the gr ain stored. The loan is repaid at the end of period two when the farmer liquidates the stored grain. Another aspect that the farmer needs to consider is the return from other investment opportunities such as growing wheat. The resource allocation decision is made subject to where is an increasing and concave utility function, is income in period 0, there are two states good and bad , is income in period 1 for each state, is probability of each state, is the discount factor, is the amount of storable commodity in period 1, is quantity stored, is investment in wheat crop whose price is normalized to one, is the net s aving in financial market, is the advance rate set by the warehousing firm which allows the farmers to receive credit against the value of the stored grain, is the return on th e credit market or the cost 78 of borrowing from the credit market or the WRF , is the per unit cost of storage, and is the probability of being in bad state. Since, the main interest is storage choice under risk and credit constraint. The model is s olved using the Lagrange method under four different market scenarios i n which access to credit and risk pooling is varied. Table 3. 1 : Lagrange Results Scenarios Storage Wheat Perfect Credit Market and Risk Pooling Incomplete Credit Market and Perfect Ris k Pooling Perfect Credit Market and Incomplete Risk Pooling Incomplete Credit Market and Risk Pooling 3.3. Model Parameterization In order to provide a sense of the magnitude of the value of the insurance under various market cond itions the model is fully parameterized so that we can solve it numerically. The utility function is assumed to take the form where > dictates the curvature of the function and can be thought of as a m easure of risk. The model parameters are calibrated to represent an average smallholder farmer. The average farmer sell approximately 150 maunds immediately at harvest as the quality of the paddy starts to deteriorate if it is not dried within a few days. We assume that a farmer would store paddy for 90 days as the prices stabilize at a higher equilibrium by that time. However, small farmers are also 79 known to have high dis count rates which can erode the profitability of investments. I collected data on tim e preferences of farmers using price lists which involve choosing between a smaller amounts sooner versus a larger amount later. The interest rate is gradually increased t o elicit the subjective discount rate of the agent. However, if the utility is concav e, then this approach leads to overestimation of the discount rates. This can be corrected by estimating discount rates conditional on the curvature of the utility functio n which is also a measure of risk aversion. The risk task was based on the design of Holt and Laury (2002) where the agent was given a choice list of two lotteries. The agent chooses one lottery over the other and the expected value of the two lotteries i s adjusted to elicit the subjective risk preferences of the agent. These games are ca refully designed so that the pair of switching rounds can be used to identify the lower and upper consistent. The midpoint of this range is then used this as an approximation of the provides summary statics of both the risk aversion and time preference parameters and shows that there is significant heterogeneity in the preferences of respondents. The actual games are detailed in the appendix. Ta ble 3. 2 : Time and Risk Parameters Count Mean SD Discount Rate 622 .21 .19 Risk Aversion 622 .55 .42 The decision to participate in WRF is not only shaped by preferences but expectations about future outcomes as well. Subjectiv e price expectation is an important aspect in the decision - making process as the support of the expected price of paddy could make WRF an unprofitable prospect making non - participation a subjectively rational choice. Figure 3.1 below shows the historical p rices at harvest and 3 - month post - harvest as reported by farmers in the study area. 80 Figure 3. 1 : Historical Paddy Prices (PKR/maund) Since these prices are reported by the farmers, we can consider them as subjective prices based on which the farmers would form expectations about future prices. The figure also shows t hat on average post - harvest prices have been higher in the past three years. However, we also see that there is an overlap in the harvest and the post - harvest pr ice distributions suggesting that there is a positive probability that the farmers could receiv e a lower price in the post - harvest compared to - harvest were colle cted before the rollout of the warehouse receipts financing program. Farmers were asked to repo rt the maximum and minimum price for the upcoming 2018 harvest and 3 - months post - harvest. Table 3. 3 shows the summary statistics of prices reported by farmers. 81 Table 3. 3 : Expected Prices (PKR/maund) N Mean SD Min Max Min Harvest 2018 661 1482 272 1100 2100 Max Harvest 2018 661 1702 352 1200 2500 Min Post Harvest 2018 661 1938 378 1200 2800 Max Post Harvest 2018 661 2207 453 1350 300 0 On average the farmers expected prices to be between rupees 1500 to 1700 at harvest and between rupees 1900 to 2200 post - harvest. I use the minimum and maximum expected prices to calculate averages and the distribution is shown in figure 3. 2 which shows an overlap in the price distributions. This suggests that farmers do assign a positive probability to prices falling in the future. Figure 3. 2 : Average Expected Paddy Prices (PKR/maund) Another important consideration for parti cip ating in the WRF would be the transaction cost. A small holder will only use WRF if the premium earned from the stored grain adjusted for cost of warehousing is positive. Data on warehouse charges was obtained from a local warehouse management company w hic h offered these services to small farmers in the study area. Table 3. 4 reports the cost of warehousing which is substantial even when non - pecuniary costs such as 82 arranging for labor and vehicles for transportation are not incorporated. Grading and weigh tin g charges are assumed to be zero. Loading and dispatch is the labor cost for taking the paddy off the trucks and then loading them back at sale time. Preclearing is the cost of labor used to feed the paddy to the dryers. The moisture content needs to be re duced before its stored, mechanical drying is done by passing the paddy through dryers which dry it and clear it of impurities such as dust and empty kernels. Under mechanical drying paddy is stored in silos where moisture level is automatically maintai ned . In the case of traditional storage paddy is dried in the sun and overturned several times. Paddy is stored in jute bags and manually monitored to ensure moisture levels are within acceptable levels. The cost of drying and storage are higher for mecha nic al drying due to high energy costs and using off grid sources of electricity due to load shedding. Jute bags are available on rent from the warehouse or can be bought by the farmer themselves, a jute bag normally costs around PKR 250 and often short in sup ply at harvest time. The transport cost reported in the table is the mean, but cost can double depending on the distance from warehouse and availability at the time of harvest. Table 3. 4 : Warehousing Cost (40kg bag) Cost per bag Mechanical Traditional Unloading and Dispatch 10 10 Preclearing 15 0 Drying cost 50 24 Storage 90 days 120 60 Jute bag rental 90 days 0 36 Transportation cost to warehouse 15 15 Total Warehousing Cost 210 145 If the farmer stores the paddy, i t leads to 6 percent loss of weight under mechanical storage and a 2.5 percent weight loss under the traditional sun drying method on average. The weight loss is higher in the mechanical drying as it is more efficient in dr ying and removing impurities such as empty kernels, dust, and other debris. The total cost of mechanical drying is approximately 83 PKR60 per maund higher than the traditional drying, and the premium for mechanically dried paddy can vary between 0 to PKR 200 per maund. The farmer is issued a warehouse receipts which records the amount of paddy stored and its value at the time of storage based on its quality. This warehouse receipt can be used to acquire a loan which needs to be repaid at the time of liquidati ng the stored grain. The advance rate is fixed at 70 percent of the value of the crop at the time of storage at an annualized interest rate of 28 percent. Base - case model parameters are set to the values given in Table 3. 5. Table 3. 5 : Base Case Parameter Values Sy mbol Value Description 1550 Price/maund at harvest 2100 Price/maund 3 - month post - harvest in the good state 1300 Price/maund 3 - month post - harvest in the bad state 0.10 Subjective probability of being in the bad sta te. 0.79 Subjective annual discount rate q 150 Harvest quantity in maunds ra 0.55 Risk Aversion a 145 Per maund storage cost for 90 days a 0.025 Weight loss due to drying and cleaning R 1.14 Annualized interest rate 0.70 Advance rat e for loan against warehouse receipts b 1.25 Return on investment in wheat a Warehousing cost and weight loss are given for traditional drying and cleaning. b Return on wheat is based off the calculations provided by the agriculture department. 3.4. Results The model is calibrated for a smallholder farmer and solved under the four market conditions derived earlier. Under complete credit market the farmer can borrow and solve from the market under the prevailing interest rate. Future price of paddy is assume d to be known and equal to the average of the good and bad state. However, to ensure that results are reflective of actual market outcome and unrealistic amounts are not borrowed, investment in wheat capped at PRK 200,000 84 which is the cost of production fo r 5 acres of land 10 . Results show that under these conditions the farmer would store the entire 150 maunds, borrow PRK 164,000 from the market, and invest PKR 200,000 in wheat production. The model is then solved with incomplete credit market, and to make t hi s binding the farmer cannot borrow but can save at the prevalent rate. Other conditions are same where expected future price is known with certainty and the maximum amount that can be invested in wheat is capped. Results show that the farmer will store t he entire 150 maunds again, invest PKR 55,000 in wheat and not save anything. Next the credit market is assumed to be functional but there is uncertainty in future paddy prices. Results show that the farmer will store the entire harvest, borrow lower quant it ies from the market and invest PKR 200,000 in wheat. Finally, the model is solved under the final market scenario where there is no credit market or risk pooling. The results show that the farmer will store 36 maunds of the produce, invest PKR 108,000 in w heat, and not save any money. A summary of these outcomes is given in table 3. 6 . Table 3. 6 : Simulation Results Scenarios Paddy Stored (maunds) Borrowing (PKR) Investment in Wheat (PKR) Perfect Credit Market and Risk Pooling 150 164,000 200,000 Incomplete Credit Market and Perfect Risk Pooling 150 0 55,000 Perfect Credit Market and Incomplete Risk Pooling 150 158,000 200,000 Incomplete Credit Market and Risk Pooling 36 0 108,000 I then conduct a sensitivity analysis on how t he demand for storage will change under different interest rates, probability of a bad state, and risk aversion. All of these parameters are changed individually to identify parameter values at which they becom e binding. Interest rates are varied from 7 pe rcent to 20 percent with steps of 1 percent, probability of being in the bad state is varied from 5 percent to 20 percent, with steps of 1 percent. The subjective risk aversion parameters is 10 I use the production costs for 2018 - 19 under average conditions in Punjab developed by the local agriculture department 85 varied one standard deviation above and below the mean. Figure 3 . 3 reports these sensitivity analysis results under the traditional storage method. The results show that risk aversion does not impact storage when risk pooling is possible. When risk pooling is not possible risk aversion decreases quantity stored and as relative risk aversion decreases the farmer higher proportion of paddy. Similarly, at low interest rates the farmer would store the entire stock of paddy but as the interest rate increases the quantity of padd y stored decrease and goes to zero. The subjec tive probability of bad state also reduces participation, and farmers under incomplete credit markets are most sensitive this variable. Figure 3 .3 : Storage Sensitivity Analysis 3.5. Conclusion This paper develops a dynamic model of demand for a standard WRF product in the context of Pakistan in the presence of liquidity constraints, risk aversion, and price uncertainty. This paper 86 adds to the existing literature by analyzing the feas ibility of WRF in a context where there are two com plete agricultural cycles. Most of the WRF evaluations and technical reports have been in contexts where there is primary growing season followed by a dry season in which limited agricultural activities ca n be carried out. I solve a two - period model for an agricultural year, and even though the model is for two period its findings are clear into how preferences and uncertainty reference and solved under four different market co ndition. Results show that farmers will store if either the credit market or insurance market is functioning. However, in a scenario where access to credit is constrained and risk cannot be pooled demand f or storage would fall. I also find that farmers ar e sensitive to transaction cost of storage and would not store under the mechanized method as it is more costly, similarly the demand for storage also falls as interest rates increase. Risk also plays an important role in the decision to participate. Risk averse farmers have low demand when there is no risk pooling. Similarly, as the subjective probabilities of the prices falling in the future increases non - participation becomes the dominant strategy. The se results suggest that transaction costs of wareho using are significant and erode the profitability of WRF. The farmer is better off selling the paddy at harvest especially under mechanical storage as the weight loss due to drying and the cost of storage are very high and the premium over traditional meth ods is not certain. Participation in the WRF can be increased by offering lower interest rates on credit and reducing the exposure to risk. One mechanism could be the price risk in paddy against other commodities or buying futures. However, absence of these market in developing countries make the success of WRF difficult as it requires the small farmer to bear the entire risk. 87 A limitation of this model is that it assumes disutility from losses to be proportional to the utility from gains. Prospect theory highlights that this might not be the case as people derive higher disutility from losses. This might be particularly true in the case of small farmers who might be highly loss averse as one negative shock could push them into poverty. Hence, the results from this numerical solution must be viewed an upper bound to the storage demand. Future work can include loss aversion as an additional parameter to test the feas ibility of this program. Government and international organizations promoting warehouse receipts financing need to take the transaction and risk factor into account while planning this program. The direct impact of improving profitability of small farmers through intertemporal arbitrage would be limited unless access to this program is improved through measure which protect small farmers against large negative shocks. If we look past the distributional impact of WRF, it has many other benefits such as it c ould promote the development of agricultural value chains by making it more efficient and imposing quality controls and standards. It would also reduce the price volatility of the commodities, therefore reducing uncertainty and increasing investment. 88 APPENDIX 89 APPENIDX Preferences over income in the first period and in the various states of the second period with probability of state and a discount factor are where is an increasing and concave utility function, is income in period 0, there are two states good and bad , is income in period 1 for each state, is probabilit y of each state, and is the discount factor. The farmer maximizes (1) u nder the following constraints is the amount of storable commodity in period 0, is quantity s tored, is investment in wheat crop whose price is normalized to one, is the net saving in financial market (negative if borrowing and positive if saving), is the advance rate set by the warehousing firm against the value of the stored grain 11 , is the c oncave wheat production function, is the return on the credit market or the cost of borrowing from the WRF, is the weight loss due to storage, and is the per unit cost of storage. Additional side conditions are imposed so that the farme r cannot st ore negative amounts or store more than quantity produced and investment in wheat is nonnegative. Since the main interest is in storage choice under risk and credit constraint, storage rules under these environments are deriv ed. The pro blem is initially solved under perfect credit and risk 11 F or simplicity I assume that the farmer would always take the maximum amount of cash advance through WRF . 90 pooling to arrive at a well - known intertemporal arbitrage optimal decision rule. Credit constraint and imperfect risk pooling are then introduced into the model, first individually and then together. Case 1: Perfect credit market and risk pooling. Under this market condition the farmer does not face liquidity constraints and consumes an average of the good and bad states due to perfect risk pooling { . This extreme case of perfect insurance later helps us to focus on the implications of a binding credit constraint when risk plays no role in the decision to allocate resources. The setup of the problem under L agrange method is given below. The first order conditions are Using the above results, we can derive the decision rules for the farmer. Assuming . ndent of re sources (q) and preferences. The investment is only determined by the cost of interest and the return from investment which depends on the production function and prices. Similarly, the choice of storage depends on expected price, current price, and the co st of storage. Neither reduction in risk or access to credit would influences the choices under this scenario. 91 Case 2: Imperfect credit market and perfect risk pooling. In this case credit market is non functions and binding which essentially adds an ad ditional constraint . Risk pooling is still available, so the farmer continues to earn the expected income. The setup of the problem under Lagrange method is given belo w. The first order conditions are Using the above results, we can derive the decision rules for the farmer. Assuming . Due to the liquidity constraint, investment in wheat is constrained and lower than under functioning credit market. The choice of storage now depends on the relative return from wheat. The choice of storage depends on the expected future prices, return in the credit market, advance rate in WRF, current prices, and return from investment in wheat. Preferences still do not enter the decision maki ng for either of the investment options. 92 Case 3: Imperfect risk pooling but a perfect capital market. In this scenario there is no risk pooling, but the capital market is complete, and the farmer can engage in income smoothing across the two time perio ds. The setup of the problem under Lagrange method is given below. The first order conditions are Using the above results, we can derive the decision rules for the farmer. Assuming . The choice of inves tment in z is independent of resources (q) or preferences and only depends on the cost of credit and the re turn from wheat which depends on its production function and prices. The decision rule for storage is given by Assuming for simplicity 93 Storage choice is influence d by return in the credit market, advance rate, current prices, cost of storage, future price in different states and their probabilities, and the preferences of farmers. Case 4: Imperfect Credit and R isk Pooling In this scenario there is no access to a credit market or risk pooling. The setup of the problem under Lagrange method is given below. The first order conditions are Using the above results, we can derive the decision rules for the farmer. Assuming . Due to the liquidity constraint, investment in wheat is constrained and lower than under functioning credit market. The choice of storage now depends on the relative return from wheat. The decision rule is given by 94 Assuming for simplicity 95 BIBLIOGRAPHY 96 BIBLIOGRAPHY Aggarwal, S., Francis, E., & Robinson, J. (2018). Grain today, gain tomorrow: Evidence from a storage experiment with savings clubs in Kenya. Journal of Development E conomics . https://doi.org/10.1016/j.jdeveco.2 018.04.001 Ashraf, N., Karlan, D., & Yin, W. (2006). Tying odysseus to the mast: Evidence from a commitment savings product in the Philippines. Quarterly Journal of Economics . https://doi.org/10.1162/qjec.2006.1 21.2.635 Basu, K., & Wong, M. (2015). Evaluat ing seasonal food storage and credit programs in east Indonesia. Journal of Development Economics . https://doi.org/10.1016/j.jdeveco.2015.02.001 e in Kenyan Villages. Working Paper . Chavas, J. P., Despins, P. M., & Fortenbery, T. R. (2000). Inventory dynamics under transaction costs. American Journal of Agricultural Economics . https://doi.org/10.1111/0002 - 9092.00023 Duflo, E., Kremer, M., & Robinso n, J. (2011). Nudging farmers to use fertiliz er: Theory and experimental evidence from Kenya. American Economic Review . https://doi.org/10.1257/aer.101.6.2350 Engle Warnick, J. C., Escobal, J., & Laszlo, S. C. (2011). Ambiguity aversion and portfolio choic e in small - scale peruvian farming. B.E. Journ al of Economic Analysis and Policy . https://doi.org/10.2202/1935 - 1682.2331 Feder, G., Just, R. E., & Zilberman, D. (1985). Adoption of Agricultural Innovations in Developing Countries: A Survey. Economic Develop ment and Cultural Change . https://doi.org/10. 1086/451461 Jensen, R. (2010). The (Perceived) returns to education and the demand for schooling. Quarterly Journal of Economics . https://doi.org/10.1162/qjec.2010.125.2.515 Karlan, D., Osei, R., Osei - Akoto, I., & Udry, C. (2014). Agricultural decisions af ter relaxing credit and risk constraints. Quarterly Journal of Economics . https://doi.org/10.1093/qje/qju002 Lai, J. Y., Myers, R. J., & Hanson, S. D. (2003). Optimal On - Farm Grain Storage by Risk - Averse Farmers . Journal of Agricultural and Resource Econom ics . Lawless, L., Drichoutis, A. C., & Nayga, R. M. (2013). Time preferences and health behaviour: a review. Agricultural and Food Economics . https://doi.org/10.1186/2193 - 7532 - 1 - 17 Liu, E. M. (2013). Time to cha nge what to sow: Risk preferences and technol ogy adoption 97 decisions of cotton farmers in China. Review of Economics and Statistics . https://doi.org/10.1162/REST_a_00295 McKenzie, D., Gibson, J., & Stillman, S. (2013). A land of milk and honey with streets paved with gold: Do emigrants have over - optim istic expectations about incomes abroad? Journal of Development Economics . https://doi.org/10.1016/j.jdeveco.2012.01.001 Miranda, M. J., Mulangu, F. M., & Kemeze, F. H. (2017). Warehouse Receipt Financing for Sm allholders in Developing Countries: Short on Logic, Long on Imagination. Saha, A., & Stroud, J. (2006). A Household Model of On - Farm Storage under Price Risk. American Journal of Agricultural Economics . https://doi.org/10.2307/1243663 Ward, P. S., & Singh, V. (2015). Using Field Experiments to Elicit Risk and Ambiguity Preferences: Behavioural Factors and the Adoption of New Agricultural Technologies in Rural India. Journal of Development Studies . https://doi.org/10.1080/00220388.2014.989996 William, J. G., & Kaserwa, N. (2015). Improving sma llholder farmers access to finance through warehouse receipt system in Tanzania. International Journal of Economics and Financial Research .