Willingness to Pay for Area Yield Index Insurance of Rice Farmers in the Mekong Delta, Vietnam

Transcription

1 WAGENINGEN UNIVERSITY AND RESEARCH CENTER MASTER THESIS Willingness to Pay for Area Yield Index Insurance of Rice Farmers in the Mekong Delta, Vietnam Student : Nguyen Mai Trang Supervisor : Dr. Lei Pan Chair Group: Development Economics 1 WAGENINGEN, APRIL 2013

2 WAGENINGEN UNIVERSITY AND RESEARCH CENTER MASTER THESIS Willingness to Pay for Area Yield Index Insurance of Rice Farmers in the Mekong Delta, Vietnam Student : Nguyen Mai Trang Supervisor : Dr. Lei Pan Chair Group: Development Economics WAGENINGEN, APRIL 2013

3 Abstract Area-yield index insurance is among the latest initiatives by the Vietnamese government to support the country s rural rice-growing farmers. While expectations on its ability in dealing with covariant shocks and complementing farmers self insurance mechanisms are high, doubts have been cast on its feasibility and effectiveness. This thesis strives to find out the willingness to pay of Vietnamese farmers in the Mekong Delta for this area-yield index insurance, as well as the factors that influence it. I conducted a contingent valuation study on a sample of Vietnamese farmers in Dong Thap province, using a double bound dichotomous choice procedure. Farmers risk attitude was measured by a gambling game, while other interested variables were collected from a household questionnaire. Empirical analysis using a probit model for the willingness to join and an interval data model for double bound dichotomous choice found evidences that that while most farmers might be interested in joining the programme, they will only be willing to buy the insurance at a subsidized price. The perceived high probability of agricultural risk occurrence, farmers risk aversion, wealth, education and complementary risk management strategies such as the sale of assets statistically influenced the decision to take up insurance. Risk-averse farmers, those who do not grow rice in the risky season, and those intending to sign an interlinked contract, statistically had a lower willingness to pay. Households with non-farm employment and savings, better knowledge of insurance, and, surprisingly, the small farmers of the community were willing to pay more for the insurance. The findings highlight the importance of client education, development of better indices, and potential of tying insurance to an interlinked input supply contract. Future work on revealed willingness to pay and the cost-benefit analysis of insurance as a rural welfare improvement policy option is recommended. Key words: area yield index, agricultural risks, insurance, willingness to pay, contingent valuation ii

4 List of tables Table 1: Agricultural risk management strategies... 4 Table 2: Comparison area yield index and weather index insurance... 8 Table 3: Versions of double bounded dichotomous choice bids Table 4: Socio-economic conditions of sampled districts Table 5: Summary of respondent characteristics Table 6: Locally classified distribution of households in the sample districts Table 7: Household rice-growing land Table 8: Rice production of sample districts Table 9: Significant damage due to floods (in Vietnam Dongs) Table 10: Risk management strategies of respondents Table 11: Reasons for (not) joining the insurance scheme Table 12: Are you willing to join the insurance scheme? Table 13: Probit regression results for the Willingness to Join Table 14: How much are you willing to pay for the insurance? Table 15: Double-bound regression results for the Willingness to Join Table 16: Variables used for the estimation of the Willingness to Pay Table 17: Summary of estimated Willingness to Pay iii

5 List of figures Figure 1: Double bounded dichotomous choice procedure Figure 2: Map of Dong Thap province with 2 sample districts Figure 3: How likely or unlikely is it that you would personally experience catastrophes or epidemics causing damage to rice production? Figure 4: How small or large do you expect the negative consequences would be of catastrophes or epidemics causing damage to rice production? Figure 5: How important is it for you to prevent or reduce the negative consequences of catastrophes or epidemics causing damage to rice production? Figure 6: How do you judge your own ability to protect yourself from catastrophes or epidemics causing damage to rice production? Figure 7: Expected occurrence of future catastrophes and epidemics Figure 8: Self-judged relative risk exposure Figure 9: Who should be responsible for payment of damages caused by catastrophes and epidemics? Figure 10: "Risky gain" game results Figure 11: "Risky loss" game results iv

6 Acknowledgements I conducted this study with invaluable academic guidance from my supervisor, Dr. Lei Pan. Dr. Lei Pan has been following my study from the proposal building to the final thesis report with valuable and timely advices. Her research experiences, particularly with the issue of risks and field researches in the context of developing countries, helped me in designing and conducting my contingent valuation survey which is vital to the rest of the thesis. I would have overlooked many practical details that would have ruined my progress without her kind advices. Dr. Lei Pan also supported me in selecting a suitable analytical model for my data within the time and academic constraints that I faced. Moreover, she has always been a critical reader who goes through my work thoroughly and provides spot-on feedbacks. I am grateful for the support of Dr. Lei Pan in the completion of this thesis. I also owe great thanks for the help of the researchers from the Department of Development Economics, University of Economics of Ho Chi Minh (UEH), who have put me in contact with the local governments and cooperated with me in conducting the household survey. I am in particular grateful to lecturer Phung Thanh Binh at UEH, who is also a PhD candidate of ENR chair group, WUR, for helping me with logistics arrangement for my field trips, providing me with a team of qualified enumerators and facilitating my meetings with key informants. I would also like to thank all of my key informants, local guides in Thanh Binh and Tan Hong districts, my team of enumerators and all the farmers who participated in the survey and spent their time working and sharing valuable information for my study. I would not have enjoyed the process of writing my thesis as much without the emotional support of my friends, both in Vietnam and in Wageningen. In particular, I would like to dedicate my special thanks to Ms. Mai Le Quyen who has not only been my constant source of encouragement, but also provided me with useful advice as a senior student in International Development Studies. Finally, I would give the greatest thanks to my beloved parents who unconditionally support me in everything I do, including the writing of this thesis. v

7 Table of Contents Abstract... ii List of tables... iii List of figures... iv Acknowledgements... v Summary... viii Chapter I: Introduction Problem Statement Research objective, questions and scope Structure of the thesis... 3 Chapter II: Literature Review Risks and crop insurance in developing countries Agricultural risks and risk management strategies Agricultural insurance in developing countries Index based agricultural insurance Risk attitude and demand for insurance Risk perceptions and demand for insurance Area yield index insurance and other risk management and coping strategies Other determinants of the willingness to pay Chapter III: Research methodologies Contingent valuation method in studies on demand for insurance Double-bounded dichotomous question to elicit WTP Willingness to pay model Theoretical model of the willingness to pay vi

8 The analytical model of the dichotomous CV data The analytical model for double bounded CV data Measuring risk attitude Measuring other variables Chapter IV: Data Selection of study sites and sampling procedures Contingent valuation survey Descriptive statistics of the sample Household socio-economic characteristics Rice production Flood risk exposure and experiences Crop risk perception and management options Experience with insurance Risk attitude Chapter V: Results and discussion Farmers willingness to join the area-yield index rice insurance Farmers willingness to pay for area yield index insurance Limitations Chapter VI: Conclusions and implications References Appendices vii

9 Summary Farming households in developing countries face a multitude of risks, among which covariant shocks, e.g. catastrophic events that affect the whole communities, are not easily absorbed through individual self-insurance mechanisms. Area-yield index insurance is among the latest initiatives by the Vietnamese government to support the country s rural livelihoods, specifically rice-growing farmers who are prone to natural disasters and pest diseases. While expectations on the insurance scheme are high, doubts have been cast on the acceptance of the product by the farmers and appropriateness of insurance clauses in simulating demand. Addressing these doubts, this thesis strived to find out the willingness to pay (WTP) of Vietnamese farmers in the Mekong Delta for this area-yield index insurance, as well as the factors that influence it, especially farmers risk attitude, risk perception, understanding of the index, and risk management strategies. Literature review indicates the role of index insurance as an innovative alternative to traditional insurance products, while maintaining the potential in improving social welfare. The underwriting of contracts against a trigger independent of individual farm s performance is said to minimize asymmetric information consequences and considerably reduce transaction costs. The discrepancy between insurance payout and actual losses, however, can be a hindrance to the scheme s successes. Another frequently mentioned factor hampering insurance-uptake is rural households risk aversion, the effects of which on other beneficial farming activities exhibit similar patterns. Farmers perception of the surrounding risky environment also plays a role in the decision to buy insurance. Meanwhile, the application of other risk-management strategies by the farmers, e.g. savings, diversification, non-crop income, other types of insurance can have a complementary or substitution effects on the demand for index insurance. Farmers states of knowledge, wealth and trust have also been empirically found to influence the insurance purchase decision. To explore the WTP, I used a contingent evaluation survey on a stratified sample of farmers in Dong Thap province, Vietnam. Data on farmers socio-economic characteristics, risk perception, experience and understanding of insurance and risk-management options were collected via a questionnaire, while farmers risk attitude was measured through means of a hypothetical gambling game. A double-bound dichotomous choice (DBDC) procedure was carried out to find the maximal amount a farmer is willing to pay for the insurance premium before each cropping season to have their rice production insured. I ran a probit model to examine the factors determining a farming household s willingness to join (WTJ) the insurance scheme or not; and an interval data DBDC model using maximum likelihood to estimate the average WTP and its determinants. The results of the empirical analysis show that while most of the interviewed farmers were interested in the programme and willing to join once offered, they were only willing to buy the insurance at a subsidized price. Empirical analysis of the WTJ using a probit model shows that the decision to take up insurance was motivated by the perceived high probability of agricultural risk occurrence and hampered by farmers risk aversion. Wealthier farmers were unlikely to join the insurance scheme, while better-educated farmers were more likely potential insurance buyers. The statistics figures also predicts that insurance is more of a complementary measure against risks, viii

10 next to other risk management strategies such as the sale of assets. The farmers WTP for the insurance was statistically lower among risk-averse farmers, those who do not grow rice in the risky flood season, and those with an intention to sign an interlinked contract with local procurement companies. Meanwhile, the WTP for insurance was found to be higher among those with complementary non-farm employment and savings, better knowledge of insurance, and, surprisingly, the small farmers of the community. Basis risk and understanding of basis risk did not significantly affect the WTJ and WTP. The findings highlight the importance of client education, development of better indices, and potential of tying insurance to other credit-loosening mechanisms such as interlinked input supply contracts. Future work on revealed WTP farmers using the insurance uptake data so far, put in the context of a cost-benefit analysis of insurance, in comparison with emergency relief, rural infrastructure improvement and other safety net options, is recommended. ix

11 Chapter I: Introduction 1.1. Problem Statement Poor and vulnerable households in developing countries face high income risks as part of their daily lives. In response to climatic risks, economic fluctuations and individual-specific shocks, households resort to income-smoothing risk management strategies (e.g. diversification and low-risk activity taking) and consumption-smoothing risk coping mechanisms (e.g. self-insurance, precautionary savings), informal group-based risk-sharing and loans)(dercon, 2002). While these arrangements are more effective in dealing with idiosyncratic risks, formal insurance can be a necessary tool to deal with common/covariant (catastrophic in many cases) shocks that affect the whole community (Dercon, 2002; P. Hazell, 1992; Mechler, Linnerooth-Bayer, & Peppiat, 2006). Developing welldesigned insurance products that help farmers manage risks in agricultural production is therefore an important goal in developing countries(clarke, Das, Nicola, & Hill, 2012). Vietnam is an agricultural economy, with agriculture as an important source of economic growth. Approximately 70% of Vietnamese rely on agricultural production, whose major cash crop is rice paddy, for their livelihoods (H. N. Nguyen, 2007). Despite the structural economic change towards industry and services, the agricultural sector still accounts for 20% of the annual GDP(GSO (General Statistics Office), 2012). The country is exposed to various risks to agricultures, namely typhoons, flooding, landslides, seasonal drought, storm surge and coastal flooding. Its paddy cultivation is also prone to epidemics and pest diseases (RAP (Regional Office for Asia and the Pacific), 2011). The Vietnamese Mekong Delta (VMD), which accounts for around 50% of the country s rice production, is projected to be among the areas most impacted by sea level rises due to climate change, thus being more vulnerable to extreme events like severe flooding, salinity intrusion and storm surges. These climatic adversities pose threats to the region s rice production performance, such as water levels exceeding the physiological limits of the rice plant, excessive moisture and serious salinity intrusion (Wassmann, Hien, Hoanh, & Tuong, 2004). Among others, insurance instruments have been put on the world s agenda as an essential response to climate change risks(linnerooth-bayer & Mechler, 2006). Despite this foundation for the potential market segment, agricultural insurance in Vietnam remains a minor player in the insurance market, with agricultural insurance premiums only accounting for 0.08% of the whole insurance market in 2011 (MOF (Ministry of Finance), 2012a). The first multiple-peril-crop insurance scheme for rice farmer was launched in 1982 by the then state company Bao Viet and was not a success. Agricultural insurance has since then never favored by private insurers. In 2010 only Bao Viet (the former national insurance company) and Groupama (The French mutual agricultural insurance company) were underwriting very small agricultural insurance portfolios (RAP (Regional Office for Asia and the Pacific), 2011). A government-funded agricultural insurance pilot programme was initiated from July 2011 to support agriculture producers in coping with financial damage caused by catastrophes and epidemics, contributing to rural social security and agricultural production promotion as stipulated in Decision 315 by the Minister of Vietnam on Agricultural Insurance Pilot (Prime Minister of Vietnam, 2011, p. 1). This 1

12 insurance scheme, which covers paddy farming, livestock and poultry, and aquaculture production, provides farmers with premium subsidies according to their poverty levels, ranging from 20% to full premium subsidy. Regarding the paddy farming insurance product, the government adopted an area-yield index approach to insurance, in which payouts are made based on the productivity loss of the defined area, not on individual households rice production. Insurance specialists consider the index-insurance, with its advantages in transaction cost and moral hazard reduction, suitable for the farming situations in Vietnam (See Appendix A for a summary of the key features of the rice insurance products). The paddy crop insurance pilot has been implemented in selected communes of 7 provinces (Nam Dinh, Thai Binh, Nghe An, Ha Tinh, Binh Thuan, An Giang, Dong Thap) and is expected to end in The pilot has been reported by the Ministry of Finance to yield encouraging results and was supported and embraced by farmers (MOF (Ministry of Finance), 2012b). Dong Thap province, one of the most productive province in terms of rice production, for example, reported the increasing participation rate of households, compensation was paid according to the regulations, no sign of insurance fraud (Vietnam Government Portal News, 2013). Besides these optimistic reports, media attention, however, has also been paid to the underperformance of the insurance product. Until the end of August 2012, according to the Ministry of finance s figures, merely over 98,000 households participated in insurance scheme, accounting for only 3% of the number of households entitled to participate in the scheme (K. N. Nguyen, 2012). Several reasons for this low uptake were addressed at the National Forum for Farmers with the topic Farmers with Agricultural Insurance back in August 2008 (C. Nguyen, 2012). The risks covered by the scheme (e.g. for paddy insurance only a limited number of diseases and pests) were considered too limited. The insurance premium has also been considered to be too high, which explains the participation of mostly poor and marginally poor whose premium payment is 100% or 80% subsidized. Some interviewed farmers said that with the insurance premium added to the costs of ploughing, seedling, fertilizers and pesticides, the earned profit would be almost minimal. Others also considered the area-yield based approach to compensation as unfair and a barrier to farmers willingness to join the scheme. A farmer admitted When my crop was lost, I was hoping my neighbors crops were lost too so I would be paid compensation. There have also been concerns raised on the administrative delays in indemnity payment, as compensation can only be made upon official announcements on epidemics as listed covered risks. Others deemed the low awareness and knowledge of farmers on insurance products as a major drawback to the programme implementation (C. Nguyen, 2012). In response to these pitfalls, recently the Ministry of Finance and Ministry of Agriculture and Rural Development have made adjustments to the scheme, including the addition of new covered risks, increase in the trigger point in average yield and lower the premium fees for paddy crop insurance (MOF (Ministry of Finance), 2012c). Despite this attention to the challenges faced by crop insurance product in Vietnam, yet of now there is no study available to the public on the willingness to pay (WTP) of rice-growing farmers for this insurance product, nor empirical researches on the factors that influence their decision to purchase this insurance. Addressing these questions will be important to understanding the households need for insurance, re-considering the feasibility of the insurance product and enhancing the effectiveness of crop insurance as a solution to risk coping for farmers. As the end of 2

13 the insurance pilot is coming closer, these considerations would have important implications for both Vietnamese policy makers, who are seeking the best way to support the farmers livelihood resilience next to the costly relief programmes; and insurance providers, who in the case of Vietnam have to balance monetary profits and social objectives in their business operations Research objective, questions and scope This thesis strives to investigate the willingness to pay for area-yield based paddy crop insurance of Vietnamese farmers in the Mekong Delta region and the determinants of this WTP, providing empirical evidence necessary for addressing the challenges currently faced by the insurance product in Vietnam. To reach the aforementioned objective, the research questions to be answered in this thesis are: Are the farmers willing to join the rice area yield index insurance scheme? How much are they willing to pay for the insurance? What are the determinants of farmers willingness to join (WTJ) and WTP for the area-yield index rice insurance product? For the last question, the sub-questions are: How do risk preference, risk perception and other risk management strategies of the farmers affect their WTJ and WTP? Do the basis risk, an inherent characteristic of index insurance, and understanding of basis risk, influence these WTJ and WTP? How do other factors play a role? The area-yield based index paddy crop insurance is being implemented by Bao Viet and Bao Minh in 7 districts of different regions of Vietnam. Despite sharing several common features, these regions have distinct geographical features that influence the rice cultivation practices and thus the agricultural adoption decision. This research focuses on the Mekong Delta region where the largest rice yield is produced in the country. Next to the official paddy crop insurance product as stipulated in the Decision 315, there exists another insurance product provided by Bao Minh Insurance Company, with the technical support of the Netherlands Development Organization SNV in Vietnam. This product is provided in a limited number of villages in Nghe An province, with some features slightly different from the official scheme. This product is not the objective of my thesis. However, I will provide some comparisons and relevant implications in the final chapter of this thesis Structure of the thesis The thesis is organized into 6 chapters as follows: The Introduction chapter is followed by a Literature review, in which I covered the most relevant and current literature on agricultural insurance, with a special focus on crop insurance and the determinants of demand for insurance in developing countries. The third chapter describes the methods I used to explore the WTP of Vietnamese farmers in Mekong Delta for the paddy farming insurance, including the WTJ and WTP models. The forth chapter details the data I collected, including descriptive summaries of the interested variables. The fifth chapter advances to the findings generated based on the methods detailed in chapter three on the collected data. Moreover, this chapter provides comparisons and discussions with previous works. The last chapter concludes my study and lists some recommendations I have for policy makers regarding agricultural insurance. 3

14 Chapter II: Literature Review 2.1. Risks and crop insurance in developing countries Agricultural risks and risk management strategies Rural communities in developing countries are exposed to a variety of risks in their daily lives. Most of them are involved in agricultural livelihoods, which are characterized by a higher degree of risk and uncertainty, due to the dependence on natural factors and the long production cycle. Authors such as Mishra (1996) decomposed agricultural risks into output risk, which is caused by natural hazards, and price risk, which results from economic uncertainty mainly coming from price fluctuation of agricultural products. These risk forms play a more detrimental role in developing countries, where volatile weather conditions, market imperfections and social insecurity prevail (Mishra, 1996). The focus of this research is on the first type of risks which are basically dealt with by crop insurance. Dercon (2004) classifies risks based on the covariant or idiosyncratic nature of risks. While covariate (common, aggregate) risks have an impact on all members of a community or region, idiosyncratic (individual) risks affect only a particular individual (household) (Dercon, 2004). In this line, the agricultural shocks in Mishra s classification would fall well into the category of covariate risks. Faced with risks, households in developing countries have adopted a range of measures to deal with consequences of shocks. Although variations in terms exist, most literature on risks distinguishes risk-reducing (management, ex-ante) strategies from risk-coping (ex-post) strategies. The strategies can also be divided into informal and formal arrangements; household-level and community-level measurers; market-based or publicly-provided interventions (Jaffee, Siegel, & Andrews, 2010). The most common actions taken to deal with agricultural risks are summed up in Table 1. Table 1: Agricultural risk management strategies Informal risk management strategies Farm household (mitigating risk) Ex-ante Savings Buffer stocks Enterprise diversification Low risk, low return cropping patterns Production techniques Ex-post Sale of assets Reallocation of labour Reduced consumption Borrowing from relatives Community-level (sharing risk) Food crop sharing Common property resource management Social reciprocity Rotating savings/credit Sale of assets Transfers from mutual support networks 4

15 Formal risk management strategies Market-based (share/transfer risk) Ex-ante Contract marketing Financial hedging tools (options) Traditional agricultural insurance Weather index insurance (WII) Contingent funds for disaster relief Ex-post Savings Credit Publicly-provided (transfer-absorb risk) Pest/disease management Physical crop/food stocks Price guarantees or stabilization funds Input subsidies Public agricultural insurance Disaster assistance Social funds Cash transfers Waiver (cancellation) of crop loans Source: (Jaffee et al., 2010, p. 25) The nature of risks determines the appropriateness of coping measures in case of consequences. While dealing with idiosyncratic shocks can take place at a household level or be shared within a close-knit community, in case of covariate risks, local risk-coping strategies need to be complemented by risk pooling arrangements that cut across small rural communities. Apparently, in case of weather events causing consequences for a local geographical area, risk-sharing mechanisms such as inter-household transfers, local credit and assets are not sufficient to mitigate the impact of shocks for households. This necessitates the role of formal banking and insurance institutions thanks to their geographically spanning portfolio (P. B. R. Hazell et al., 2010; P. Hazell, 1992) Agricultural insurance in developing countries Proponents of agricultural insurance as a risk-coping strategy have promoted its behavior-altering functions: farmers will follow more profitable cropping patterns, increase technology adoption and invest more, improving their welfares and securing national food supplies (Binswanger-Mkhize, 2012). Attention has been especially paid to the beneficial role of agricultural insurance in leveraging small farmers access to rural finance. Examples include the anticipated win-win situation created by bundling agricultural insurance with rural credit provision and input supplies, where the farmers can increase their credit-worthiness and banks can protect their solvency level. However, it has been recognized that agricultural insurance alone is not the full solution to improving farmers income, but it should be implemented in tandem with other essential agricultural services, efficient marketing channels, and other mitigation measures in case of increasingly frequent and intense events (RAP (Regional Office for Asia and the Pacific), 2011). Regarding the relation between insurance and credit, empirical evidence from studies such as that by Gine&Yang (2009) shows that the studied farmers with the bundled loan were surprisingly reported to have lower take-up, which was explained by the implicit insurance from the limited liability clause in the loan contract (Giné & Yang, 2009). It has been also argued that crop insurance will not solve the problem of agricultural credit provision as they are both exposed to the same underlying cost, moral hazard and covariance problems (Binswanger-Mkhize, 2012; P. Hazell & Pomareda, 1986). 5

16 Three basic institutional frameworks exist for agricultural insurance: public sector with usually heavily subsidized insurance programme; purely private commercial agricultural insurance markets; and public-private partnerships (PPP) (RAP (Regional Office for Asia and the Pacific), 2011, p. 26). Due to the fundamental role of agriculture in many developing economies, this formal risk management intervention has mostly been undertaken by governments. In fact, often heavily subsidized by governments, agricultural insurance is among the few classes of insurance subject to such high levels of government intervention. Policy interventions in insurance can be justified on many grounds: the protection of rural households from poverty and financial institutions from loan defaults, the production security of rural farming households, the financing on disaster relief and the development of social safety nets. Despite the expected role, the performance of public sector subsidized crop insurance (a more specific branch of agricultural insurance that covers damage to growing crops but not forestry, livestock, aquaculture, etc ) has historically been poor (RAP (Regional Office for Asia and the Pacific), 2011). Reasons for the failure were early on discussed by Hazell (1992): premium rates that were set too low; blurry boundaries between commercial and social goals; lack of management and control over risk selection and loss assessment due to information asymmetries; high transaction costs; and the liquidity and solvency of insurers in case of a significant catastrophe (P. Hazell, 1992). From being considered as an effective method of enabling farmers in developing countries to tackle risk, thereby removing one main obstacle to technical change and innovation in the 1960s and 1970s, until the 1990s, the trendy view was that crop insurance mostly multiple peril based doesn t work (Mishra, 1996). Responses to address the pitfalls have included the transfer responsibility to the private insurance sector (e.g. in Mexico, Brazil) and development of public-private partnership (e.g. in China, Japan). In Vietnam, the pilot insurance programme is PPPbased. Alternative designs for crop insurance programme, in place of individually assessed crop insurance products, have also been seen as necessary developments for betterment of crop insurances. These so-called index-based crop insurance products are discussed in the following section Index based agricultural insurance Traditional crop insurance, whose indemnities are based on individually-assessed losses, such as named peril crop insurance and multiple peril crop insurance, has been widely considered financially unsustainable and plagued with moral hazard and adverse selection problems. Traditional crop insurance also suffers from high transaction costs, notably loss adjustment ones, which hampers both the profitability of insurers and affordability for smallholders (Binswanger- Mkhize, 2012; Carter, Galarza, & Boucher, 2007; P. Hazell, 1992). In the past decade, index contracts emerged and have been promoted as a powerful financial solution to address the pitfalls of traditional insurance products. Smith & Watts (2009) and more recently Binswanger-Mhkize (2012) thoroughly reviewed the landscape for index insurance in developing countries. The basis of this approach is the underwriting of the contracts against specific perils or events trigger recorded at a local level. Examples of the triggers are area yield index or rainfall index. In the case of area yield index insurance, a payment is made when the 6

17 measured yield for the region falls below a certain predetermined limit critical threshold or strike-point (Smith & Watts, 2009). In the payment scheme implemented in Vietnam, compensation is made when the commune s total yield falls below 80% (recently adjusted to be 90%) of the pre-estimated average yield which is a moving average of the three previous respective crop seasons. Expected favorable features are listed for such a contract: Elimination of adverse selection and moral hazard as the expected indemnities paid out by the insurer are independent of the characteristics of insured farmers, and no individual farmer alone can influence the probability of an insurance pay-off (Carter et al., 2007; Skees, Hazell, & Miranda, 1999). Administratively, as there might not need any individual contract to write (rather a universal form with only varied insured units); there is also less individual loss assessment need. The data used is generally reliable and available, obtained from the crop cutting experiments conducted by the state governments for purpose of crop estimation (Skees et al., 1999) and much less open to dispute (Rao, 2010, p. 195). In the case of Vietnam the commune s yield is reported by the statistics offices in districts, or the relevant agricultural division unit of the district s People s Committee. For these advantages, pilot index insurance programmers have been hailed for its feasibility and affordability" as an effective, market-mediated solution to promote agricultural development and can benefit clients beyond agricultural producers (Binswanger-Mkhize, 2012, p. 187;189). However, so far, the international performance of index insurances doesn t seem to live up to the promises its promoters anticipated. Binswanger (2012) provided an intensive oversight of the underachievement of index insurance products. He cited the reasons for the low demand, reflected by the small proportion of purchasers, the small coverage purchased and the limited participation of the poor who would benefit the most as follows: wealthier farmers may already be well-insured with other risk management strategies and don t need formal insurance contracts, unless the cost is lower than that of their self-insurance mechanisms. Meanwhile, poor farmers are not adequately self-insured and would benefit from formal insurance arrangements, but are credit-constrained and cannot afford the premium before the harvest, while receiving the payouts only after the harvests. In the end, relief programmes are still needed even in the availability of the insurance scheme (Binswanger-Mkhize, 2012). Although the index contract can remove most of the systematic correlated risk that an individual faces and significantly reduce transaction costs, idiosyncratic yield variation exposes the farmers to basis risk known as the discrepancy between insurance payout and actual losses. Index contracts are essentially a trade-off between basis risk and transaction costs, and the high basis risk can hamper the attractiveness of the insurance (Skees et al., 1999). To minimize this basis risk, a crop insurance scheme based on the area approach should be a homogenous area so that the productivity of a majority of farmers will fluctuate in the same direction with the average area s yield. The basis risk from area yield index insurance is usually considered to be lower than that from weather index insurance (Carter et al., 2007). This is partly because area yield insurance can cover more perils than weather based insurance (Mahul & Verma, 2012). Table 2 below compares these two mostly-used index insurance products in India, where index insurance products are most fully developed and applied. 7

18 Table 2: Comparison area yield index and weather index insurance Technical and practical aspects Geographic Basis Risk Product Basis Risk Product Design Basis Risk Area yield index All peril cover (drought, excess rainfall, flood, pest infestation, etc.) Easy-to-design index (estimated aggregate yields in a given area) Low start-up costs Slow claims settlement Arises when the Insurance Unit size is too large and is not homogenous in terms of agricultural production level Yield index insurance covers risk from sowing till harvesting. As Yields are estimated at harvest stage, losses if any suffered after harvest are not reflected in the yield index. Trigger yield used in yield index insurance is a function of moving average of past 5 years yield and coverage level, which may range from 60 to 90 percent. In other words, the shortfall between normal yield and trigger yield is not protected Weather based index Single (sometimes multiple) peril cover (drought, excess rainfall, low temperature, etc.). Technical challenges in index design (peril, crop, farming practices, agrometeorological zone, etc.) High start-up costs Faster claims settlement Arises when a weather station is referenced for a larger geographical area, covering areas far off from weather station Weather index covers risk arising out of deviations in parametric weather exigencies only. Risks outside these parametric weather (like pests, diseases, hailstorm, flooding etc.) are not covered Arises because of imperfect correlation between weather index and the production process (yield) Source: (Mahul & Verma, 2012, p. 11; Rao, 2010, p. 200) Due to the presence of basis risks, indemnities paid in index insurance are imperfectly linked to a farm s actual loss experience and thus index insurance becomes less attractive (Smith & Watts, 2009). The homogeneity requirement of the defined insured area is in practice troublesome, as even small differences in cultivation land and conditions can lead to substantial differences in the yield damage, which leads to surprisingly low yield correlation between plots within the same village, or even within the same farm. In a study in Ethiopia, willingness to continue purchasing microinsurance in the subsequent years was found to be affected by basis risk as 30% of their respondents refused to continue with the product if there was no pay-out when rains failed their farm. Understanding how basis risk, especially downsize one (when participants have a bad crop and don t receive a payout) affects farmers interest in the insurance, is important to assessing the long-term sustainability of an index insurance market (Hill, Hoddinott, & Kumar, 2011). Ramasubramanian (2012) found a positive impact of understanding basis risks on the willingness to join index-insurance scheme (Ramasubramanian, 2012). As farmers willingness to join an insurance product depends on the extent to which a risk management strategy provides protection against losses when they occur, and idiosyncratic part of income risks is relatively large (Dercon, 8

19 2002), farmers will want to protect against basis risk and find it advantageous to complement the coverage offered by agricultural index insurance by investing in other risk-management options (Clarke, Das, Nicola, & Hill, 2012). Risk preference of households, which is discussed in the following sections, is also linked to the impact of basis risk on insurance take-up. Basis risk is likely to severely limit voluntary take-up from the most risk averse, even in the presence of large subsidies, since purchase worsens the worst that could happen: without indexed insurance the worst that could happen is that a farmer loses her entire crop, but with indexed insurance a farmer could lose her entire crop and have paid an insurance premium yet receive no claim payment due to basis risk (Mahul & Verma, 2012) Risk attitude and demand for insurance Risk attitude is economically an important determinant of individual choice. Agricultural activities in low income countries are inherently risky and farm households must make investment and production decisions within a multitude of risks. The implications for production decisions under risks are illustrated by the classic theoretical models of farmers attitude towards risks within the framework of a neo-classical model of farm production by Ellis (1988). According to these models, low-income farmers are risk-averse. Consequently, there is inefficiency of resource use at the farm level - cropping patterns designed to safeguard family security rather than to maximize profit; reduction of input levels, e.g., of fertilizers, if input costs are certain and output risky; discouraged or delayed beneficial innovations. Empirical studies have aimed at testing these and similar propositions. Many researchers have found the majority of farmers to be risk averse, and risks to hamper farmers willingness to take high-expected-return investment decisions, such as in fertilizers and machineries, sacrificing profit-maximizing portfolios for lower risks (H. P. Binswanger & Sillers, 1983; Binswanger-Mkhize, 2012; Foster & Rosenzweig, 2010; Yesuf & Bluffstone, 2009). Risk preference has important implications for formal insurance as a risk-management strategy and a beneficial activity. In the specific domain of index insurance, the most notable works on linking risk preference and index insurance through theoretical foundations are by Gine, Townsend & Vickery (2008) and Clarke (2011). The former developed a so called benchmark model, in which the willingness to pay for insurance contract will be increasing in risk aversion similarly to what is suggested by the neoclassical framework on expected utility and full insurance - and decreasing in basis risk (Gine, Townsend, & Vickery, 2008). Meanwhile, in Clarke s theory, optimal demand for index insurance is hump shaped in risk aversion, which is, first increasing and then decreasing in risk aversion. This is caused by the combination of actuarially unfair premiums, whereby premiums are greater than expected claim income, and basis risk, the risk that the income from index insurance will not accurately reflect the incurred loss. A farmer who is extremely risk averse will put a lot of weight on the event of non-payment in spite of loss, thus will not be willing to purchase the insurance, even if it is inexpensive. These factors cause demand to be low from both the risk neutral, for whom insurance purchase decreases mean income, and the very risk averse, for whom index insurance purchase decreases the minimum possible income. Only those with intermediate levels of risk aversion optimally purchase index insurance (Clarke, 2011). 9

20 Many recent papers investigated the empirical relationship between risk preference and farmers uptake of insurance products. Gine, Townsend, & Vickery (2008), in testing their own benchmark model, found that risk-averse households somewhat eschew purchasing rainfall insurance, the opposite of the predicted result which was explained by uncertainty about the insurance policy, making insurance a risk itself. Kouame and Komenan (2011) found a highly significant effect of risk aversion on cocoa farmers minimum price insurance take-up decision and that high risk aversion inhibits the demand for insurance (Kouame & Komenan, 2012). The choice experiment game executed by Galarza & Carter (2010) in Peru revealed a quadratic-shaped relationship between the decision to choose among different insurance options with varied risk preference levels highly risk averse subjects prefer uninsured loans while those with moderate risk aversion adhere to insured loans on area-yield index basis (Galarza & Carter, 2010). Using the evidence from rural Bangladesh, Clarke et al (2012), however, found the degree of risk aversion to insignificantly determine the weather insurance take-up overall (Clarke, Das, Nicola, & Hill, 2012). To elicit individuals risk attitudes, Kouame and Komenan (2012) reviewed two basic approaches: econometric and experimental approach. The econometric approach, though advantageously derived from actual behaviors, may suffer from confounding risk behavior with other factors. The experimental approach, based on hypothetic questionnaires regarding risky alternatives or risky games with or without real payments, has been increasingly considered to be a more desired method and widely used to measure risk preference (Kouame & Komenan, 2012). Numerous experiments with real payouts have been conducted to explore the risk preferences among rural households in developing countries and their heterogeneous nature. Examples in this line include the pioneering work by Binswanger (1980) on Indian farmers, who were virtually all found to be risk averse with little interpersonal variation (H. Binswanger, 1980). Data collected by Mosley and Verschoor (2005) in Uganda, Ethiopia and India confirmed this early finding that measures of vulnerability like income and wealth are unrelated to risk aversion (Mosley & Verschoor, 2005). More recently, Yesuf & Bluffstone (2009) concluded from their study on Ethiopian households that high levels of risk aversion are prevalent, and households circumstances, including their wealth and previous successes have important impacts on risk aversion (Yesuf & Bluffstone, 2009). Experiments conducted by Tanaka, Camerer & Nguyen (2006) in Vietnamese villages showed that people are less loss-averse in villages with higher income (Tanaka, Camerer, & Nguyen, 2006) Risk perceptions and demand for insurance Data on individual s perceptions of the risks involved in the environment he or she operates are particularly relevant for understanding behaviour in the developing world where risks are pervasive (Weerdt, 2005). Under the framework of expected utility theory, an insurance purchase can be conceptualized as a decision in which the consumer is faced with a risk that has some distributions of losses across probabilities. Expected benefits of insurance cover depend upon the perceived likelihood of occurrence and the size/severity of the conditional loss. The subjective perception of risks is a cultural phenomenon, which is influenced by personal experience, age, gender, information of consequences of alternate actions, as wells as the social, political and cultural environment (Zweifel & Eisen, 2012). If individuals underestimate the likelihood of a future disaster, they often do not protect themselves and insurance demand is decreasing in perceived 10

21 loss probability, all else being equal (Petrolia, Landry, & Coble, 2011). Households are expected to spend more on mitigation investments if the potential loss accounts for a large proportion of their income. The frequency with which loss occurs (yield volatility) and expected severity of the loss frequent loss experiences tend to generate more mitigation efforts by farmers (Smith & Watts, 2009). Clearly the views of the farmers on the risks associated with their farming activity matter when it comes to production-related decisions like insurance purchases and arguably the best way to obtain this information is directly asking the targeted respondents questions on how he or she perceives the risks around them (Manski, 2004; Weerdt, 2005). Manski (2004) reviewed social scientists efforts in measuring probabilistic expectations using subjective probabilities elicited from survey respondents. Attitudinal researchers used questions, typically in Likert scale format to measure expectations of respondents on a specified event. This approach is said to suffer from the interpersonal and even intra-personally incomparability of responses, due to the ambiguity involved in modifiers and phrasing. Studies such as by Zimmer (1983, 1984), however, argued that expectations should be expressed in verbal rather than numerical forms as the human information processing prefers verbal thinking model. Increasingly, economists have built on Juster (1966) early suggestion to improve on the verbal approach of attitudinal research by combining verbal expressions of likelihood with numerical probabilities (Manski, 2004). Morgan and Herion (1990) discussed the problems involved in eliciting probability distributions from respondents. People will use heuristics to report probabilities: The heuristic of availability implies that a person s expectation on the occurring of an event depends on the availability of past events and the extent of the event s imaginability; the heuristic of representativeness implies that respondents believingly take into account the structure of the event (e.g. string of coin tosses headtail-head-tail-tail-head or head-head-head-tail-tail-tail) in its representativeness; the heuristic of anchoring implies the sensitivity of the respondents towards any suggested value (Weerdt, 2005). Weerdt (2005) designed a risk perception module to elicit probability distribution of future output in particular activities, in which farmers were encouraged to think about the risks involved in such activity, construct the upper and lower boundaries of the distribution and subsequently derive a number of intervals for which the respondent will be requested to provide a probability of occurring. This module involved thorough communication of the probability concept to farmers, including using allocation of objects such as magnets or stones onto different intervals to represent the probability. Petrolia, Landry & Coble (2011) recognized the possible difficulties in estimating probabilities in measuring flood risk perception through such an approach and used ordered categorical scale to measure concern over the flood risk; the respondents expected number of hurricanes that strike the area; and estimation of the damage as a proportion of the total housing structure if such a disaster happens. 11

22 2.4. Area yield index insurance and other risk management and coping strategies Agricultural insurance is one of the financial tools as part of farmers comprehensive agricultural risk management strategies. The demand for insurance product, therefore, depends on the availability, quality and cost/benefits of other risk management strategies: diversification, irrigation, non-crop income and other direct payment (government and NGO aid in the face of disasters); social network participation and/or access to savings and borrowings. If those alternative risk management strategies are relatively inexpensive and effective, then a farm may not be willing to pay very much to obtain agricultural insurance, whether it is individual or indexbased (Binswanger-Mkhize, 2012; Clarke & Kalani, 2011; Smith & Watts, 2009). If insurance is expensive, then retaining risk through individual or group savings or contingent credit may be more efficient than purchasing insurance for fairly severe, but not catastrophic shocks (Clarke, Das, Nicola, & Hill, 2012). Bingswagner-Mkhize (2012) claimed the ignorance of the interaction between self-insurance mechanisms and demand for insurance to be the main cause of overestimating the demand for insurance pilots (Binswanger-Mkhize, 2012). Additionally, for index insurance, due to the basis risk involved in index insurance products, improved insurance coverage against some of the specific idiosyncratic shocks, such as life insurance and group savings, can mitigate basis risk and increase the willingness to join index insurance programme. The substitution effect, however, suggests that as the price of index insurance increases, demand for non-index insurance also rises. Therefore, other types of insurances, namely self-insurance-individual savings, life and disability insurance can either increase the demand or decrease the demand for index insurance. Individual savings, on the other hand, can be a substitute for insurance, as a way to opt out of purchasing any type of insurance, as suggested above. The direction of impact of other types of insurance and risk hedges on demand for index-insurance is therefore an empirical question (Clarke, Das, Nicola, Hill, et al., 2012). Choiceexperiment games conducted by Clarke et al. (2012) on Bangladesh farmers found cross-price elasticities between area-yield insurance and non-index insurance products and individual savings, confirming the substitution effect, while the presence of group savings doesn t affect demand for other insurance products offered (Clarke, Das, Nicola, Hill, et al., 2012). Ramasubramanian (2012) found available risk coping avenues (such as savings, borrowing and diversification) to play important roles in the amount a household is willing to pay for the index based crop insurance and only the residual risk is passed on to insurance (Ramasubramanian, 2012). 12

23 2.5. Other determinants of the willingness to pay Although index insurance is considered preferable to traditional insurance products due to its potential elimination of adverse selection and moral hazard, it is still worthwhile to check the proneness of the area-yield based index insurance to information asymmetry. Adverse selection suggests that demand for insurance is higher among farmers with higher risk exposure. The risk level can be evaluated in terms of weather exposure and geographic situation. Meanwhile, moral hazard implies changes in the behavior of the farmers once they have adopted an insurance product, for example using less input or neglecting poor crops (Enjolras & Sentis, 2011). An indicator for the presence of moral hazard can also be in terms of output: Farmers who are insured will produce low yields more frequently than uninsured farmers with similar observed characteristics (Quiggin, 1993). Time series data requirement for this indicator, however, can be difficult to obtain for a new product within a limited pilot time like the insurance scheme in question. Expectations of farmers on different types of government support also play a role in the willingness to join insurance scheme. Smith and Watts (2009) observed that even in settings where government or aid agencies can afford to subsidize all administrative and operation costs, many of the poorest farmers will still be unwilling to participate in voluntary insurance programmes (Smith & Watts, 2009). In Vietnam, it has also been identified that the current extensive emergency policies of government can be a hindrance to the insurance market. If households expect unconditional support from government in the face of disasters, they are less inclined to buy insurance products (IPSARD, 2009). This tendency not to insure or take any other mitigation measures as a result of the reliance on expected financial assistance from relief programmes or donations by other individuals is known as charity hazard (Raschky & Weck-Hannemann, 2007). Understanding the relationship between this support and farmers willingness to purchase insurance is important to the feasibility of the product. For farmers in developing countries, an important barrier to the uptake of new methods in general and insurance products in particular has been the lack of understanding and trust. Clients knowledge of the insurance products not only has implications for the insurance take-up rate, but also the appreciation and satisfaction with the benefits of the insurance. More educated households have been found to be those more likely to take up insurance (Dercon, Kirchberger, Gunning, & Platteau, 2008). (Patt, Suarez, & Hess, 2010) observed a clear correlation between the desire to purchase insurance with the level of product understanding. The wealth of the households affects the purchase decision of index insurance in 2 ways: as an indicator of credit and liquidity constraints that play a key role in the decision-making; and as an important determinant of risk aversion which affects insurance take-up (Clarke & Kalani, 2011). Clarke and Kalani (2011) provide evidence for the hump-shaped relationship between weather index insurance demand and wealth based on a study in rural Ethiopia: Insurance demand is lower in the poorest and richest households and is highest in households with intermediate wealth. 13

24 Chapter III: Research methodologies 3.1. Contingent valuation method in studies on demand for insurance Empirical literature on agricultural insurance demand has determined the WTP of farmers for insurance by the main three methods: Contingent valuation (CV) method (asking producers how much they are willing to pay and other relevant questions in a survey), revealed preference (RP) technique (inferring the willingness to pay from analysis of production patterns and producers behavior) and the indirect approach (combining microeconomic household information and market information to estimate indirectly the appropriate premiums) (Kouame & Komenan, 2012). While there is a strong case for using RP when the market information is complete, stated preference (SP) techniques become necessary when the WTP information that is needed cannot be inferred from markets and/or individuals actual decisions (Bateman, 2002). In Vietnam, the agricultural insurance market is underdeveloped and almost exclusively run by state insurance companies. Besides, the area yield index insurance scheme is a new product in its pilot phase with very small geographical coverage, thus having limited records of clients data. The product is also subsidized by the government and the farmers who join the scheme so far have mostly been those with premium support. Additionally, there exist option users and non-users of the insurance: Some farmers want the insurance to be available although they do not purchase it at the moment or have no intention of actually using the products, but mostly because of aesthetic reasons: they support all initiatives by the state to stabilize rural livelihoods or they want more vulnerable individuals to be supported. These conditions make SP the most relevant method to elicit the WTP for the area yield index rice insurance of the farmers in Mekong Delta. Whittington (2010) draw essential lessons learned from stated preference research in lessdeveloped countries. These include new approaches to overcome the hypothetical bias (e.g. cheap talk scripts, time-to-think experiments, and drop-off protocols); presentation of hypothetical baselines/scenarios; and that the concern about overestimation of WTP was not as serious as skeptically thought of. Despite their pitfalls, stated preference techniques are still among the most widely used tools for economic evaluation. Nowadays, households in developing countries are more exposed to news, technologies and a better range of consumer goods and literacy rates have increased. Additionally, SP studies are now undertaken by researchers with a developing country background who are more culturally sensitive (Whittington, 2010). I took these lessons into consideration when designing the CV questionnaire and conduct my CV study. A SP study can be executed by either a contingent valuation questionnaire or a choice modeling (CM) questionnaire, of which CM is preferable when WTP for individual attributes of a good is required (Bateman, 2002). As my research scope is on the WTP for the insurance product with the same attributes as stipulated by the government s pilot documents, I employed a CV questionnaire. As reviewed by Kouame & Komenan (2011), there are few studies that used CV for specifically agricultural insurance (Patrick 1980, Vandeveer and Loehman (1994) and Sarris, Karfakis and Christiaensen (2006). This method has been quite frequently used recently to elicit the WTP for 14

25 microinsurance in developing countries, see for example (Gustafsson-Wright, Asfaw, & van der Gaag, 2009), Donfouet et al (2012) on micro-health insurance, Ramasubramanian (2012) for index crop insurance in India, Kouame&Komenan (2011) on price insurance in Cote Divoire Double-bounded dichotomous question to elicit WTP In the CV survey, after the presentation of the valuation scenario, the provision mechanism and the payment mechanism, respondents are asked to determine how much they would value the good if confronted with the opportunity to obtain it, under the specified terms and conditions. There are several ways to structure the elicitation process, including open-ended questions, the bidding game, payment card, single-bounded dichotomous choice, double-bounded dichotomous choice (Bateman, 2002). The open-ended direct elicitation format, in which the respondents are straightforwardly asked how much they are willing to pay for a certain good, is not subject to anchoring bias and respondent-specific, which is very informative and requires simple statistical techniques. However, this format has been criticized for its high non-response rate, protest answers, zero answers and outliers, because respondents are likely to find it difficult to give a WTP having either little familiarity with the product or the thought of valuing it. The close-ended formats (bidding game, payment card, and dichotomous choice) help to simplify the cognitive task faced by respondents and simulate the daily market transactions which involve the decision to buy or not to buy at fixed prices (Bateman, 2002). For the surveyed farmers who have little knowledge of insurance and the specific pilot programme, a close-ended format is preferred. I conducted double-bounded dichotomous choice (DBDC) procedure, as illustrated in Figure 1. This DBDC has the same limitations with single bound format in terms of starting bias and yea-saying, and the two responses may not correspond the to the same underlying distribution (Bateman, 2002; Yoo & Yang, 2001). Compared to most other elicitation methods, however, this procedure has significant statistical efficiency gains (M. Hanemann, 1991; Yoo & Yang, 2001). More information about each respondent s WTP is elicited than in a single-bounded dichotomous choice which only provides the upper or the lower bound of the WTP and requires larger samples. In a study that has limited pretest results such as mine, the double-bounded format helps to provide insurance against too low or too high a choice for the starting bid (W. M. Hanemann & Kanninen, 1998). 15

26 Figure 1: Double bounded dichotomous choice procedure WTJ Yes No WTP 1st bid Yes No WTP 2nd higher bid WTP 2nd lower bid Yes No Yes No Under this method, each respondent is asked if s/he is willing to pay the first bid. If s/he says yes to the first bid, a second higher bid will be given and her/his willingness to pay is asked. If s/he says no to the initial bid, a second lower bid will be provided. If s/he says no to both the first and the second bids then s/he will be asked to mention the maximum that s/he is willing to pay. Under this elicitation procedure, we have two discrete responses from every individual. To correct for starting bias, 4 different starting bids were randomly asked. Before the bidding procedure took place, as a way to lessen the hypothetical bias, enumerators were also asked to include a cheap-talk script in their presentation of the scenario. In designing different bid levels, I consulted various information sources. The informant from Bao Viet provided the prevailing premium rate (VND 37,000 1 per công after 60% of premium support from the government) and the premium rate that he deemed feasible to be enough to bring break-even point profit for the insurer (VND 20,000 per công ). The farmers in the in-depth group discussion thought VND 14,000 per công (the premium they have to pay after both government support and AGPPS - (An Giang Plant Protection Joint Stock Company) are cheap for them. I used these numbers in constructing the bidding scheme, following literature on double-bounded CV with overlapping intervals and follow-up bid amounts are placed at increasing distances from starting bid value (W. M. Hanemann & Kanninen, 1998). Table 3 shows the 4 versions of the DBDC bids. Equal numbers of questionnaire for each version of starting bid were distributed randomly among enumerators and respondents. 1 local area unit equivalent to 0.1 hectare 16

27 Table 3: Versions of double bounded dichotomous choice bids Questionnaire version Lower bid Starting bid Higher bid 1 10,000 15,000 20, ,000 20,000 30, ,000 30,000 45, ,000 45,000 65,000 Call the first bid amount t 1 and the second one t 2, each individual will fall into one of the following: Case First answer Second answer Implication for t 1 and t 2 Implication for WTP 1 Yes No t 2 > t 1 t 1 WTP < t 2 2 Yes Yes t 2 > t 1 t 2 WTP < 3 No Yes t 2 < t 1 t 2 WTP < t 1 4 No No t 2 < t 1 0< WTP <t Willingness to pay model Theoretical model of the willingness to pay The starting point of CV is to place monetary value on a change in a household s welfare. The analysis of CV data therefore follows standard welfare economic theory (Bateman et al., 2002). We assume that an indirect utility function can be derived for a household, given their income (Y), the prices of goods (P) and provision level (Q) of the insurance product, and other socio-economic characteristics of the household (X). A household s indirect utility function is denoted as: The provision of the insurance product is assumed to be beneficial, which means the utility enjoyed by the household will be greater at level - with insurance than level - without insurance, hence: When posed with the contingent valuation scenario, households are assumed to be comparing their well-being at the two levels of insurance provision with insurance or without insurance. Their maximum willingness to pay (WTP) is the monetary payment such that: We can also derive a function for WTP from the other parameters in the model, denoted as: (1) Besides, a household s maximum WTP for the insurance product is bounded by their ability to pay discretionary income: 17

28 The WTP is also assumed to be non-negative, as the good can simply be ignored if it does not provide utility to the respondent The analytical model of the dichotomous CV data In practice, there are two approaches to modeling the willingness to pay function: the random utility model (utility difference approach) and the random willingness to pay model (expenditure different/bid function approach). For the sake of more direct specifications of the analyzed function and more straightforward practical calculations, I used the later. While the random utility model bears proximity to the neoclassical economic theory of constrained utility maximization, the random WTP model allows straightforward marginal impacts of variables on household WTP. After all, the two approaches can be identical for dichotomous choice models as any bid function could be derived from some specific formulation of the underlying indirect utility function (Bateman, 2002; Haab & McConnell, 2002). The random utility model The random utility model is the basic model used for analyzing dichotomous CV responses (Haab & McConnell, 2002). The indirect utility for respondent j can be written: i = 1 in the condition with insurance and i=0 is the status quo without insurance purchase. The determinants of the utility are the discretionary income of the respondent j, as a vector of the household characteristics and other factors discussed in the literature review and elicited from the CV survey, and composing of preferences known to the individual respondent but not observed by the researcher. (2) From (1) and (2), respondent j will answer yes to a payment amount (bid) of associated with insurance purchase exceeds that of the status quo: if the utility (2) The presence of the unobservable random element be made on the dichotomous choice of the respondent. means that only probability statements can (3) (2) can be specified as additively separable in deterministic and stochastic preferences: (4) (3) then becomes: 18

29 (5) where is an m-dimensional vector of parameters so that From this point on it is necessary to specify a parametric version of the preference function. The utility function can be given in the form, for example of a linear, log linear or Box-cox function (Haab & McConnell, 2002). The simplest form - a linear utility function will give: The change in deterministic utility is: Assuming a constant marginal utility of income between the two CV states ( and, (7) becomes: (6) (7) (8) Assuming are independently and identically distributed (IID), the probability of yes for respondent j can be estimated as: Suppose, let then and we have the probit model: (9) (10) In practice, parameter estimation comes from the maximization of the likelihood function. With T as the sample size and for a yes answer, the likelihood function to be maximized is: (11) Follow from equation (8), when is distributed logistic with mean zero and variance. Normalizing by gives. For the logistic distribution, the cumulative distribution has a closed form solution and the probability that respondent j answer yes (8) becomes: In the logit model, the right hand side of the equation above is substituted for likelihood function (11). in the Typically the differences between probit and logit models are slight (Haab & McConnell, 2002). I will from now on use the probit model. 19

30 The WTP, as the amount of money that makes the respondent indifferent between the status quo and the scenario, in line with the linear random utility functions (7) and (8), can be defined as: Solving this equation for WTP yields (12) (13) Expectation of the WTP: (14) The sample mean of the expected WTP can be calculated based on the mean vector of exogenous variables: (15) The random willingness to pay model The willingness to pay function for dichotomous choice CV questions can also be directly modeled. The dichotomous CV question can be interpreted as the question of whether willingness to pay exceeds the price that is posed to the respondent. A respondent answers yes when willingness to pay exceeds the required payment: which is true if We can then make the equivalent probability statements: (16) If we let the WTP function be linear in attributes with an additive stochastic preference term: (17) Where is symmetric, IID with mean zero, and a vector of parameters to be estimated for explanatory variables associated with respondent j. (18) When is, diving by will convert the problem to a standard probit, then: (19) 20

31 where is and is the standard cumulative normal as before. As shown above, the linear WTP function and the linear utility difference function yield identical model and identical estimates of WTP. The expression for the expected willingness to pay and median willingness to pay (with to be symmetric around zero): assumed (20) represents the vector of coefficients associated to each one of the explanatory variables and coefficient capturing the amount of the bid in the probit model. the The analytical model for double bounded CV data Under the assumption that there is a single valuation function behind both answers, we can define and as the dichotomous variable capturing the first and second answers. Denote (s stands for yes, n stands for no ) Following from the random WTP/bid function (17), we have the probability expressions for each of the 4 cases given as below: Case 1: Case 2: Because by definition t 2 > t 1 : 21

32 Case 3: Case 4: These equations for 4 cases do not readily correspond to a pre-existent model. The (log) likelihood function that needs to be maximized in order to find the parameters of the model: takes the value of 1 or 0 depending on the relevant case for each individual. From this we can obtain the estimated values and for calculation of WTP using the simple WTP = Measuring risk attitude As reviewed earlier, risk attitudes of farmers can be experimentally measured by a survey or lottery with real-payouts. The use of a survey to measure risk attitude, either by the reservation price for a lottery ticket and self-assessment on a given scale has also been found to apparently generate valid indicators of the more expensive experimentally measured risk attitude (Ding, Hartog, & Sun, 2010; Dohmen et al., 2011). Due to the time and financial constraint, I used the survey approach and incorporated these questions into the CV questionnaire. I formulated the survey questions based on the experimental games with multiple price list design as conducted in Voors et al. (2012) and Brick, Visser & Burns (2012) among others. In these games, participants are presented with 2 choices: either receiving a specific amount of money with certainty (choice A), or play a simple gamble in which they either have a probability of winning some money or nothing (choice B). This choice-making is made in a number of consecutive turns, in which the payoff associates with choice A declines systematically, while the expected value of choice B remains the same. The point at which a subject switches from the safe to the risky alternative allows determining the respondent s degree of risk aversion: A risk neutral respondent will choose to switch from A to B when the expected values of both are approximately the same. 22

33 Meanwhile, a risk-loving respondent with choose B in the first turn and a risk-averse respondent will still choose A even in the fifth turn (Brick, Visser, & Burns, 2012). I used the 50/50 probability of winning in choice B, as this can be more easily demonstrated to and understood by farmers. The enumerators demonstrated the 50/50 probability by tossing a coin and showing the participant either sides (head or tails) of the coin. Turn Choice A Choice B (Expected 500,000 VNDs) Risk preference class if B is chosen 1 900,000 VNDs 50% 1,000,000 VNDs, 50% 0 Risk loving 2 700,000 VNDs 50% 1,000,000 VNDs, 50% ,000 VNDs 50% 1,000,000 VNDs, 50% 0 Risk neutral 4 300,000 VNDs 50% 1,000,000 VNDs, 50% 0 Risk averse 5 100,000 VND 50% 1,000,000 VNDs, 50% 0 As people react differently towards gains and losses, questions are formulated in 2 ways: Farmers are either faced with the chance to earn money (gambling) or lose money. Turn Choice A Choice B (Expected 500,000 VNDs) Risk preference class if A is chosen 1-900,000 VNDs 50% - 1,000,000 VNDs, 50% 0 Risk averse 2-700,000 VNDs 50% - 1,000,000 VNDs, 50% ,000 VNDs 50% - 1,000,000 VNDs, 50% 0 Risk neutral 4-300,000 VNDs 50% - 1,000,000 VNDs, 50% 0 Risk loving 5-100,000 VNDs 50% - 1,000,000 VNDs, 50% 0 As literature review suggested, the adoption of new technologies/crops/inputs depends largely on the risk aversion level of farmers and farmers who are more risk-loving will be more inclined to adopt the new production methods. The following question was formulated for this inclination: If a new crop is being introduced to the village, will you be among the first people who accept and plant the crop? (1=yes, 0=no) 3.5. Measuring other variables The respondent/household socio-economic characteristics, their rice production, flood risk exposure and experiences, perceptions of the insured agricultural risks, as well as their experience with insurance, are elicited from the CV survey. The CV questionnaire is included in Appendix B of this thesis. I described the construction of the CV survey, which was part of the data collection process, in the following Chapter IV. 23

34 Chapter IV: Data 4.1. Selection of study sites and sampling procedures I exercised a multi-stage sampling strategy, keeping in mind rice farmers in Mekong Delta as the targeted population of the study. The selected cluster representative for this region is Dong Thap province. The whole province is in the tropical climate zone, with two clearly defined seasons: rainy season from May to November, and dried season from December to April of the following year. The hydrology of the province is under the influence of 3 factors: floodwater from upper Mekong River, in-field rain and ocean tides. The hydrologic regime is divided into 2 seasons: exhausted season from December to June of the following year and flood season from July to November. Benefiting from a large river, channel and spring system, frequently silt-aggraded soil, and permanently fresh and non-saline water source, the province is the 3rd largest rice paddy in Vietnam and represents the agricultural picture of the Mekong Delta district (Dong Thap Portal, 2013). In the next sampling step, the districts of Dong Thap province were divided into two strata: one under the government s pilot rice insurance programme, which includes 3 districts Tan Hong, Thap Muoi and Chau Thanh districts; one outside the government s pilot programme, which consists of the other 9 districts. In consultation with Dong Thap Bao Viet insurance company, the sole provider of the rice insurance product in the province, for each stratum, Tan Hong district pilot district and Thanh Binh district non-pilot district were selected to conduct interviews. The socioeconomic conditions of the two study sites are summarized below in Table 4, and the locations of the two districts in Dong Thap province are included in Figure 2. Table 4: Socio-economic conditions of sampled districts Socio-economic and demographic conditions Thanh Binh Tan Hong Population 154,580 91,534 Land area (ha) 34,100 31,100 Poverty (%) 12.56% 19.54% Source: (Dong Thap Portal, 2011; Quang, 2013) In each district, officers from each Department of Agriculture and Rural development advised on the 2 representative communes in each district: Tan Cong Chi and Tan Thanh A in Tan Hong district; Phu Loi and Binh Thanh in Thanh Binh. Samples were drawn based on the localized landbased wealth classification. This classification was obtained from focus group discussions and interviews with local leaders, including officers of district s Office of Agriculture and Rural Development in each commune. In rural communities, land ownership is a good indicator for economic status. In this classification, people considered small farmers are those who own less than 0.6 ha of cultivation land, medium farmers are those who own from ha of cultivation land, and better-off farmers own more than 2.5 ha of cultivation land. The numbers of household samples from these 4 communes are proportionate to the population of the two strata districts (108/188), as these 2 districts have similar average household sizes. In each 24

35 communes, a quota sampling procedure was followed to match the proportion of poor/medium/better-off farmers in the districts, which was advised to be Figure 2: Map of Dong Thap province with 2 sample districts 4.2. Contingent valuation survey Source: (Investinvietnam.com, 2013) The CV data was generated through CV questionnaires which were filled out by numerators during face-to-face interviews with farmers. Respondents were household heads as the major decision makers and they were allowed to consult other household members in making decisions. I personally interviewed over 40 farmers, while the rest of the questionnaires and interviews were conducted by a trained group of enumerators, which includes senior students from Can Tho University and Ho Chi Minh university of Economics, as well as a lecturer-cum-ph.d student. Before 2 Although the province-level reported poverty rate in Tan Hong is higher than that in Thanh Binh, these poverty rates do not take into the fact that Thanh Binh holds a considerable proportion of the population who work in industry instead of rice-farming. Thanh Binh, however, inhabits more small-sized farmers and thus the poverty rate among farming households is higher than that in Tan Hong (interview with local leaders). 25