Formal Insurance and Transfer Motives in Informal Risk Sharing Groups: Experimental Evidence from Iddir in Rural Ethiopia

Formal Insurance and Transfer Motives in Informal Risk Sharing Groups: Experimental Evidence from Iddir in Rural Ethiopia Karlijn Morsink a1 a University of Oxford, Centre for the Study of African Economies 31 October 2014. Please do not circulate or cite. Abstract: Formal insurance is increasingly offered to incompletely insured low-income farmers in developing countries. This leads to concerns about the complementarity or substitution of formal insurance and informal risk-sharing especially when transfers in informal risk-sharing arrangements may be the result of a variety of transfer motives such as inequality aversion or punishment. I investigate transfer motives, through applying social preference models, when index insurance or indemnity insurance is offered to farmers in artefactual field experiments with 1152 farmers from 16 different preexisting informal risk-sharing arrangements, Iddir, in rural Ethiopia. My experiments show that more than 92% of farmers do not transfer according to pure self-interest. 37% of transfer behavior can be explained by social preferences out of which the largest part punishes farmer j s decision. In a majority of cases I punishes j for not taking insurance, which is more likely for indemnity insurance and especially when farmer i is less risk averse. However, in some cases i punishes j for not taking indemnity insurance (not for index insurance) and this predominantly happens when farmer j is from the same pre-existing network. JEL classifications: C93 D03 D81 O17 1 karlijn.morsink@economics.ox.ac.uk; University of Oxford, Centre for the Study of African Economies (CSAE), Economics Department, Manor Road Building, OX1 3UQ, Oxford, Oxfordshire, United Kingdom, +447914118624 1

1. Introduction The lack of formal insurance markets in many developing countries does not mean that low-income farmers are uninsured. There is a large literature which demonstrates the prevalence of informal risksharing among low-income households in developing countries (f.e. Fafchamps and Lund, 2003; Fafchamps and Gubert, 2007; Attanasio, Barr, Cardenas, Genicot, and Meghir, 2012). However, empirical evidence also shows that informal risk-sharing does not provide full insurance against shocks (Deaton, 1992; Townsend, 1994, Udry, 1994). Several authors have attempted to explain this failure to achieve the first-best allocation by developing models of incomplete insurance as a consequence of limited commitment and imperfect information within informal risk-sharing arrangements (Coate and Ravallion, 1993; Foster and Rosenzweig, 2001; Ligon et al., 2002; Barr and Genicot, 2008) or the aggregate nature of shocks affecting all members of informal risk-sharing arrangements equally, limiting their ability to insure each other (Townsend, 1994). Incomplete insurance has serious consequences for households current consumption and future welfare, especially for low-income, rain-fed farmers in many developing countries in the world. In anticipation of shocks farmers may forego opportunities to invest in higher risk but higher return production technologies (Rosenzweig and Binswanger, 1993; Carter and Barrett, 2006). Ex-post, shocks cause serious losses in consumption, often with severe consequences for long-term welfare due to forced sale of productive assets (Rosenzweig and Wolpin, 1993) or negative consequences for health and education (Jacoby and Skoufias, 1997). Insurance against aggregate shocks such as rainfall has potential to reduce the impact of shocks on farmers development out of poverty. Despite the potential of formal agricultural insurance it has been proven difficult to create sustainable markets for crop or livestock insurance. Asymmetric information problems, such as moral hazard, adverse selection and costly state-verification have prevented the development of traditional indemnity agriculture insurance. Recently an alternative has been offered to low-income farmers: index insurance. Index insurance provides pay-outs based on independently verifiable indices (such as rainfall measures) and can overcome some of the asymmetric information problems and high transaction costs inherent to indemnity insurance. The introduction of formal insurance in developing countries is assumed to improve welfare by increasing the extent to which risk can be transferred over different states of the world. However, the prevalence of pre-existing informal risk-sharing arrangements in countries where formal insurance is 2

introduced raises concerns about potential crowding-out as formal mechanisms have shown to crowd-out informal mechanisms or vice versa (Arnott and Stiglitz, 1991; Attanasio and Rios-Rull, 2000, Albaran and Atanasio, 2003). Recently however, opportunities for crowding-in are considered (Mobarak and Rosenzweig, 2012, DeJanvry, Dequiedt and Sadoulet, 2013; Dercon et al., 2014). Where traditional theoretical results for indemnity insurance show crowding out of informal risk-sharing, recent theoretical results show that index insurance and informal risk-sharing are complements to the extent that index insurance provides protection against the aggregate components of a shock while idiosyncratic components can be insured through informal risk-sharing (Mobarak and Rosenzweig, 2012, Dercon et al. 2014). This is especially interesting for index insurance which carries basis risk, which is often the result of idiosyncratic components of a shock. The potential crowding-in effect of index insurance is an especially interesting one from a policy perspective as it suggests that offering insurance to informal risksharing groups may provide a solution for low demand for index insurance which is often attributed to the existence of idiosyncratic basis risk (Clarke, 2011). For example, Hill and Robles (2011) report less than 20% take-up of individual agriculture index insurance offered to farmers in Ethiopia. Gine et al. (2008) report 5% uptake of index insurance in Andra Pradesh in India and Cole et al. (2013) 5-10% in Andra Pradesh and Gujarat. Even if premium rates are significantly below actuarially fair rates due to subsidization, more than 50% of households do not take up index insurance in India (Cole et al, 2013). However, the current evidence for the crowding-in effect of index insurance depends solely on theoretical results or indirect empirical tests. Mobarak and Rosenzweig (2013) randomly assigned index insurance and randomly placed weather stations and investigated the effect of various levels of basis risk on the extent of risk-sharing within traditional risk-sharing groups, jati, in India. They show that in communities which are better able to insurance idiosyncratic basis risk through informal risk-sharing in the jati, demand for index insurance is higher. Dercon et al. (2013) developed a theoretical model in which they show that within group risk-sharing and index insurance are complements in the sense that a marginal increase in one increases the marginal utility of the other. They provide empirical support for their model by reporting results from marketing efforts in which group leasers are trained about index insurance. By randomizing the training content they demonstrate that members of groups whose leaders had received training emphasizing risk-sharing of basis risk had considerably higher uptake. However, recent studies which have empirically investigated motives for transfers in pre-existing informal risk-sharing arrangements in developing countries find that transfers are not only based on reciprocity motives but may also reflect other regarding preferences or function as a mechanism to punish or reward the other person s behavior in the context of deviations from sharing norms (Jakiela, 2009; 3

Ligon and Schechter, 2012; Mueller, 2012; Lin, Liu and Meng, 2012). For example, Lin, Liu and Meng (2012) show that typical results from models about complementarity of formal insurance and informal risk sharing, which rely on pure reciprocity motives for sharing, may be threatened as soon as private transfers are based on other sharing motives. The continuation of the informal risk-sharing arrangement depends on sharing norms which are effectuated by threat of punishment through autarky. The introduction of formal insurance changes the ex-ante expected utility of being in the autarky state and may therefore reduce the disutility of punishment. This change in disutility may lead to changes in transfers. This is especially important to understand as recent evidence shows that small changes in transfers which deviate from the sharing norm can significantly change individuals willingness to share future risk (Hill, Maruyama and Viceisza, 2012). Social preferences may affect the value of autarky when formal insurance if offered because subjects care about others welfare (Foster and ROsenzweig, 2002; Lin, Liu and Meng, 2012). Inequality aversion may increase the risk coverage provided by informal transfers, reducing optimal demand for indemnity insurance (Lin, Liu and Meng, 2012; Dercon et al., 2014). Punishment may decrease transfers relative to a sharing norm, especially when formal insurance demand is low and heterogenous. This is especially relevant because small changes deviating from a sharing norm may significantly change individuals willingness to share future risk (Hill, Maruyana and Viceisza, 2012). Therefore I investigate motives for sharing in a context where formal insurance is introduced in a context of (pre-existing) informal risksharing arrangements, where sharing norms are institutionalized. Through the study s design I am able to distinguish between self-interested preferences, warm-glow, pure inequality aversion and inequality aversion with punishment of perceived unfair behaviour. I investigate this by applying existing social preference models (Charness and Rabin, 2002) to investigate transfer motives when formal insurance is introduced. I conduct artefactual field experiments with farmers from pre-existing informal risk-sharing groups, iddir, in rural Ethiopia in the form of a oneshot, two-person, informal risk-sharing game where subjects receive either index or indemnity insurance offers. The games were played with 1152 Ethiopian farmers from 16 different pre-existing informal risksharing groups called iddir. In a benchmark game with tw-person groups one farmer, j, experiences a shock and the other farmer, i, can transfer. In the treatment games j gets an offer to take up either indemnity (crop) insurance or index (rainfall) insurance. Survey data were collected at the iddir- and farm-level. Based on this unique data this paper makes two contributions. My experiments show that more than 92% of farmers do not transfer according to pure self-interest. 37% of transfer behavior can be explained by social preferences out of which the largest part punishes farmer j s decision. In a majority of cases I punishes j for not taking insurance, which is more likely for indemnity insurance and especially 4

when farmer i is less risk averse. However, in some cases i punishes j for not taking indemnity insurance (not for index insurance) and this predominantly happens when farmer j is from the same pre-existing network. 2. Design and Predictions In this study I conducted lab-in-the-field experiments in the form of a one-shot, two-person, informal risk-sharing game. In the benchmark treatment (T1) one of the subjects (j) is randomly selected to receive an idiosyncratic shock while the other subject (i) is asked to make a transfer decision to j in case the idiosyncratic shock to j is realized. The probability of the idiosyncratic shock is conditional on the probability of an aggregate shock. In the indemnity insurance treatment (T2), before the realization of the shock, j receives an indemnity insurance offer which she can privately decide to take or reject. After the realization of the aggregate shock but before the realization of the idiosyncratic shock i is asked to make a strategic decision about transfers to j in case j experiences a shock and took insurance or in case j experiences a shock and took no insurance. The index insurance treatment (T3) is exactly the same except for the fact that the insurance provided does not cover the idiosyncratic but the aggregate shock. In the formal insurance treatment (T4) subject j is randomly offered either no insurance, indemnity insurance or index insurance and i is asked to make a strategic decision for all the possible states where j experiences a shock. The latter treatment allows for within-subject comparison of transfers. 2.1 Sample selection The games were played with 1152 Ethiopian farmers from 16 focal-iddirs and 384 farmers from nonfocal iddirs. The 16 focal-iddir were selected from 7 villages from 3 administrative regions in Tigray. Villages were randomly selected from a list of villages per administrative region in Tigray which had a minimum of six iddir per village. Per randomly selected focal-iddir 3 sessions were played with 24 farmers from that iddir in each session, covering 72 farmers per focal-iddir. The number of farmers in the 16 iddir ranged from 100-200 farmers. Per session 16 farmers were anonymously teamed up with a farmer from their own iddir and 8 farmers were teamed up with a farmer which was randomly selected from another iddir in the village. This leads to two subsamples. In sub-sample 1[G=1] farmers i from the focal iddir [i F=1 ] are paired anonymously with farmer j who is from their own pre-existing informal risksharing arrangement, the focal iddir [j F=1 ]. In sub-sample 2 farmers i from the focal iddir [i F=1 ] are paired with someone with whom they did not have a pre-existing risk-sharing arrangement, non-focal iddirs [j F=0 ]. Table 1: Sample per iddir, treatment and total 5

Sub-sample [i,j] T1 T2 T3 T4 Total/iddir Iddir sample Sub-sample 1 [G= 1] [i F=1 ] 6 6 6 6 24 [j G=1 ] 6 6 6 6 24 Sub-sample 2 [G=0] [i G=1 ] 6 6 6 6 24 [j G=0 ] 6 6 6 6 24 24 24 24 24 96 Total sample Sub-sample 1 [G= 1] Total [i F=1,G=1 ] 96 96 96 96 384 Total [j F=1,G=1 ] 96 96 96 96 384 Sub-sample 2 [G=0] Total [i F=1,G=0 ] 96 96 96 96 384 Total [j F=1,G=0 ] 96 96 96 96 384 384 384 384 384 1536 During the recruitment phase, subjects were informed that they were eligible to participate in an experiment in which they would be asked to play two games and participate in a survey. They were informed that in the second game they would make individual decisions between risky or safe lotteries. In the first game they would be teamed up with a farmer either from their own or another iddir and would be asked to make decisions about taking up insurance and sharing risk. They were informed that they would receive a base-payment of 50 Ethiopian Birr irrespective of the outcomes of their own or other farmers decisions in the games. They were also informed that they would be able to win an additional 0 110 ETB depending on the decisions they and others would make in the games. They were told that the additional payment would be based on the outcome in either the first or the second game, based on a random draw at the end of the session. Subjects were also informed that the total participation time including the two games, the survey and the payment would not be more than three hours. The incentives in the games reflect a daily wage for unskilled labour, which ranges between 50 and 150 ETB during the timing of the experiments. 2.2 The benchmark risk-sharing treatment At the beginning of the instruction phase subjects learned that there would be two types of farmers in their session, farmer i and farmer j, and that they would be randomly assigned to be either one of these in the course of the game. They also learned that ¾ of the farmers were members of the focal iddir and ¼ of the farmers were members of another iddir. Farmers were instructed that, after they had been assigned to be farmer i or j, they would be randomly matched with another farmer to form a pair of (i, j). 6

Farmer i and j have initial wealth 0 ETB. Both farmer i and j can earn income of 100 ETB from agricultural production. Farmer j s agricultural production is subject to an idiosyncratic loss [Sj=0,1] with binary loss which can take values {28,100}. The probability of idiosyncratic loss is determined by the realization of an aggregate weather shock A. The aggregate weather shock A takes a value of 1 with probability ¼ and value of 0 with probability ¾. The state of A determines the probability of the statedependent idiosyncratic losses to j, Sj. If A=0 the probability of idiosyncratic loss is 1/3 and provides income of 28 ETB and the probability of no loss is 2/3 and provides income of 100 ETB. If A=1 the probability of idiosyncratic loss is 2/3 and provides income of 28 ETB and the probability of no loss is 1/3 and provides income of 100 ETB. The game consisted of five parts: 1. The realization of the aggregate shock; 2. The certain income received by i.; 3. The strategic risk-sharing decision by i; 4. The realization of the idiosyncratic loss to j s income; 5. The actual risk-sharing from i to j. The realization of the income was simulated in the game as the realization of either a good crop or a bad crop. Farmers were informed that both farmer i and j could have a good crop which would lead to income received by the farmer of 100 ETB. Farmers also learned that farmer j s crop would be subject to an idiosyncratic loss which could lead the farmer to have a bad crop which would reduce his income to 28 ETB. Farmers were informed that farmer i would be allowed to make a risk-sharing decision in which she would get a choice to share part of her income (100 ETB) with farmer j in case farmer j would receive the idiosyncratic shock with loss to income of 72 ETB resulting in an income of 28 ETB. In the first part of the game, the realization of the aggregate shock, they were informed that both farmers would receive an aggregate shock [A=0,1] to their income with probability ¼. In the game this was simulated by a draw from an envelope which contained four tokens, three blue ones, representing good weather and one yellow one representing bad weather. After the realization of the aggregate shock farmer i would have a good crop with certain income of 100 ETB. After i received 100 ETB i was asked to make a strategic decision about risk-sharing where she was first asked if she wanted to transfer part of her income to j, and second how much, in case j would experience an idiosyncratic loss [Sj=1] and would receive only 28 ETB. In the fourth part j s crop was realized. In case of idiosyncratic loss j would have a bad crop and receive 28 ETB while in case of no shock j would have a good crop and receive 100 ETB. The probability of the idiosyncratic loss was conditional on the realization of the aggregate shock. In case of an aggregate shock the probability of idiosyncratic loss was 2/3 while in case of no aggregate shock the probability of idiosyncratic loss was 1/3. The conditional probabilities for farmer j are presented in table 2. The conditional probability of idiosyncratic loss is thus 5/12. In the game a red dice and a white dice with 7

different probabilities of idiosyncratic loss simulated the conditionality on the aggregate shock. In case a blue token was drawn (A=0) farmers would play the red dice (p=1/3) which had four blue sides, representing a good crop (100 ETB), and two yellow sides, representing a bad crop (28 ETB). In case a yellow token was drawn (A=1) farmer would play the white dice (p=2/3) which had two blue sides and four yellow sides. In the final part of the benchmark risk-sharing game the actual transfer was made based on the strategic decision by farmer i in part 3 of the game. The payoffs for i and j are presented in Table 3. 8

Table 2: Income structure benchmark risk-sharing treatment (T1) Aggregate shock A [0,1] Idiosyncratic loss Sj [0,1] Probability p Crop income 1 (p=1/4) 1 2/3 28 2/12 1 (p=1/4) 0 1/3 100 1/12 0 (p=3/4) 1 1/3 28 3/12 0 (p=3/4) 0 2/3 100 6/12 Conditional probability Table 3: Pay-offs i and j in case of transfer from i to j (Tij) in benchmark risk-sharing treatment (T1) Aggregate shock Idiosyncratic Conditional Payoff i Payoff j A [0,1] loss Sj [0,1] probability 1 1 (p=1/4) 1 (p=2/3) 2/12 100-Tij 28+Tij 2 1 (p=1/4) 0 (p=1/3) 1/12 100 100 3 0 (p=3/4) 1 (p=1/3) 3/12 100-Tij 28+Tij 4 0 (p=3/4) 0 (p=2/3) 6/12 100 100 The expected value if Tij=0 for i is 100 ETB and for individual j is 70 ETB. With full risk-sharing (Tij=36) the expected value of T1 for both i and j is 85 ETB. Figure 2: Benchmark game: Informal risk sharing 9

1.3 The indemnity insurance treatment In the indemnity insurance treatment, before the realization of the aggregate shock, farmer j receives the option to take up actuarially fair indemnity insurance with premium m Im = 5/12*P Im where m Im is 30 ETB and P Im in case Sj=1 of 72 ETB. The indemnity insurance reduces the variance of the idiosyncratic loss (p=5/12) to farmer j. The premium and insurance payout are chosen to provide full insurance implying smooth income over different states of the world. The endowment is chosen such that subjects with the insurance treatment have the same non-insurance income as subjects in the benchmark treatment if they choose not to take up insurance. The endowment provided reflects the insurance premium of 30 ETB. Care was taken to explain to farmers that in case they chose to take up insurance the premium would be deducted from their income in the game while if they chose not to take up the insurance the premium would not be deducted. Farmers j s insurance decisions were made privately. After the realization of the aggregate shock and farmer i s reception of income, farmers i were first asked to make an unconditional strategic decision about risk-sharing in case farmer j would experience a loss. In the fourth part j s crop was realized in the same way as the benchmark treatment. In the final part of the game the insurance payout was given in case j had taken insurance and experienced a loss. After the determination of the potential insurance payout the actual transfer was made based on the strategic decision by farmer i in part 3 of the game. Table 4: Income structure indemnity insurance treatment (T2) Insurance premium m Im Aggregate shock A [0,1] Idiosyncratic loss Sj [0,1] Probability p Crop income Insurance payout PIm Conditional probability p 30 1 (p=1/4) 1 2/3 28 72 2/12 70 30 1 (p=1/4) 0 1/3 100 0 1/12 70 30 0 (p=3/4) 1 2/3 28 72 3/12 70 30 0 (p=3/4) 0 1/3 100 0 6/12 70 Net income π Table 5: Pay-offs i and j in case of transfer from i to j (Tij) in indemnity insurance treatment (T2) Aggregate shock A [0,1] Idiosyncratic loss Sj [0,1] Conditional probability J insurance Imj [0,1] Payoff i Payoff j 1 1 (p=1/4) 1 (p=2/3) 2/12 1 100-Tij 70 + Tij 2 1 (p=1/4) 1 (p=2/3) 2/12 0 100-Tij 28 + Tij 3 1 (p=1/4) 0 (p=1/3) 1/12 1 100 70 4 1 (p=1/4) 0 (p=1/3) 1/12 0 100 100 5 0 (p=3/4) 1 (p=1/3) 3/12 1 100-Tij 70 + Tij 10

6 0 (p=3/4) 1 (p=1/3) 3/12 0 100-Tij 28 + Tij 7 0 (p=3/4) 0 (p=2/3) 6/12 1 100 70 8 0 (p=3/4) 0 (p=2/3) 6/12 0 100 100 11

Figure 3: Indemnity insurance offer to farmer j 1.4 The index insurance treatment In the index insurance treatment, before the realization of the aggregate shock, farmer j receives the option to take index insurance. The index insurance is actuarially fair, m Ix = ¼*P Ix, and insures farmer j against the aggregate shock (p=1/4). The premium is 18 ETB and provides a payout to farmer j in case A=1 of 72 ETB. Table 6: Income structure index insurance treatment (T3) Insurance premium Aggregate shock A [0,1] Idiosyncrat ic loss Sj [0,1] Probabili ty Crop income Insurance payout Condition al probabilit y 18 1 (p=1/4) 1 2/3 28 72 2/12 82 18 1 (p=1/4) 0 1/3 100 72 1/12 154 18 0 (p=3/4) 1 2/3 28 0 3/12 10 18 0 (p=3/4) 0 1/3 100 0 6/12 82 Net income 12

Table 7: Pay-offs i and j in case of transfer from i to j (Tij) in index insurance treatment (T3) Aggregate Idiosyncratic Conditional J insurance Payoff i Payoff j shock A [0,1] loss Sj [0,1] probability Ixj [0,1] 1 1 (p=1/4) 1 (p=2/3) 2/12 1 100-Tij 82 + Tij 2 1 (p=1/4) 1 (p=2/3) 2/12 0 100-Tij 28 + Tij 3 1 (p=1/4) 0 (p=1/3) 1/12 1 100 154 4 1 (p=1/4) 0 (p=1/3) 1/12 0 100 100 5 0 (p=3/4) 1 (p=1/3) 3/12 1 100-Tij 10 + Tij 6 0 (p=3/4) 1 (p=1/3) 3/12 0 100-Tij 28 + Tij 7 0 (p=3/4) 0 (p=2/3) 6/12 1 100 82 8 0 (p=3/4) 0 (p=2/3) 6/12 0 100 100 Figure 4: Index insurance offer to farmer j 13

1.5 The no insurance, indemnity insurance and index insurance treatment In the fourth treatment, the combined treatment, before the realization of the aggregate shock, farmers j are randomly assigned to receive either no insurance, indemnity insurance or index insurance. Farmers i, when asked to make a strategic decision are asked to indicate how much they want to share with j in case j has experienced a shocks and 1.was not offered insurance; 2. Was offered indemnity insurance and took indemnity insurance; 3. Was offered indemnity insurance and did not take indemnity insurance; 4. Was offered index insurance and took index insurance; 5. Was offered index insurance and did not take index insurance. farmers j are randomly In the index insurance treatment, before the realization of the aggregate shock, farmer j receives the option to take index insurance. The index insurance is actuarially fair, m Ix = ¼*P Ix, and insures farmer j against the aggregate shock (p=1/4). The premium is 18 ETB and provides a payout to farmer j in case A=1 of 72 ETB. To prevent wealth effects from affecting a comparison between transfers in the indemnity insurance treatment, Tij Imj=1, and transfers in the index insurance treatment, Tij Ixj=1, the pay-offs for j need to be the same, in expectation. This implies that: m Ix - ¼*P Ix = m Im - 5/12*P Im and, assuming that a rational risk averse decision-maker will purchase acturially fair indemnity insurance: 140/12 < m Im < 500/12 A solution is m Im = 30; P Im = 72; m Ix = 18 and P Ix = 72 Table 8 Game protocol risk-sharing treatments 1. Endowment given v v v 2. Farmers informed about aggregate and idiosyncratic shock v v v v v v v v 3. Farmers informed about indemnity insurance v v v v 4. Farmers informed about index insurance v v v v 5. Farmers informed about risk-sharing v v v v v v v v 6. Public random assignment of farmers to be i or j. v v v v v v v v 7. Private insurance take-up decision with premium payment by j v v v v v v 8. Aggregate shock realized v v v v v v v v 9. Certain agriculture income (100 ETB) given to i v v v v 10. Private strategic unconditional risk-sharing decision by i v v v v 11. Private strategic conditional risk-sharing decision by i v v v 12. Private idiosyncratic loss realization by j v v v v 13. Payment of agriculture income to j v v v v 14. (based on state realization in 11) Payment of insurance payout to j v v v 15. i informed about realization of idiosyncratic loss by j v v v v T1 i T1 j T2 i T2 j T3 i T3 j T4 i T4 j 14

16. (based on strategic decision in 10) Transfer made from i to j v v v v v v v v Table 9 Predictions based on self-interested and social preferences 3. Results 15

Table 10: Balance test Table 12: Consistency of behaviour with different models Table 13: Probability of behaviour coinciding with preference models 16

Table 14: Probability of behaviour coinciding with pure inequality aversion 17

Table 15: Probability of behaviour coinciding with punishment of not taking insurance 18

Table 16: Interaction of treatment with risk preferences Table 17: Probability of behaviour coinciding with punishment of taking insurance References 19

Attanasio, O., Barr, A., Cardenas, J.C., Genicot, G. and Meghir, C. (2012) Risk Pooling, Risk Preferences, and Social Networks. American Economic Journal: Applied Economics, 4(2): 134 167 Cole S.A, Gine, X., Tobacman, J., Topalova, P.B., Townsend, R.M. and Vickery, J.I. (2013). Barriers to household risk management: Evidence from India. American Economic Journal: Applied Economics 5(1): 104-35 De Janvry A., Dequiedt V., Sadoulet E. (2014) The demand for insurance against common shocks. Journal of Development Economics. 106: 227-238 Dercon, S. Vargas Hill, R., Clarke, D., Outes-Leon, I. and Seyoum Taffesse, A. (2014) Offering rainfall insurance to informal insurance groups: Evidence from a field experiment in Ethiopia. Journal of Development Economics, 106: 132 143 Hill, R.V. and Robles, M. (2011) Flexible insurance for heterogenous farmersl Results from a smallscale pilot in Ethiopia. IFPRI Discussion Paper. Mobarak, A.M. and Rosenzweig, M. R. (2012) Selling Formal Insurance to the Informally Insured. Yale Economics Department Working Paper No. 97. Mueller, A.M. (2012) In the Public Eye Norms of Distributive Justice and Sharing Behavior under Asymmetric Information: Evidence from Rural Malawi. Harvard. Job Market Paper Rosenzweig, M.R. and Binswanger, H. (1993) Wealth, Weather Risk and the Composition and Profitability of Agricultural Investments. The Economic Journal 103: 56-78 Rosenzweig, M. (1988) Risk, Implicit Contracts and the Family in Rural Areas of Low Income Countries. Economic Journal 98 (4) 1148-70 Townsend, R.M. (1994) Risk and Insurance in a Rural Credit Market: An Empirical Investigation in Northern Nigeria. Review of Economic Studies 63: 495-526 20