Extreme weather and demand for index insurance in rural India

Extreme weather and demand for index insurance in rural India Abstract Index insurance appeared recently in developing countries with the expectation to improve agricultural output and living standards in general. We investigate how experiencing extreme weather events affects farmers’ decision to purchase index insurance in India. Extreme weather events are identified from historical precipitation data and matched with a randomised household panel. Excessive rainfall in previous years during the harvest increases the insurance demand, while lack of rainfall in the planting and growing periods has no effect. The latter can be explained by access to irrigation, underscoring the importance of the local context when developing insurance products to accommodate environmental risks. 1. Introduction The monsoon observed in the summer months in South Asia is a crucial component in agricultural production as it brings more than 70 per cent of total annual rainfall. The importance for countries such as India is very high as around 60 per cent of agriculture is rain-fed, and 60 per cent of the population works in agriculture (Amrith, 2016). While a plentiful and timely monsoon boosts agricultural output, erratic rainfall and dry spells during important periods of plant growth can cause serious damage to crops. Similarly, excess rain at harvest may damage or reduce crop quality. An often cited threat to successful agriculture and consequently rural livelihoods is climate change associated with warmer temperatures, changes in rainfall patterns and increased frequency of extreme weather events. For instance, it is estimated that the summer monsoon will increase, bringing more frequent and intense precipitation days over parts of South Asia, while parts of North and Southern Africa could become drier (IPCC, 2014). To manage covariate risk from weather variability, farmers can make use of weather index insurance that calculates payouts based on readings at local rainfall stations rather than actual crop losses (Fisker, Hansen and Rand, 2015). Advocates argue that index-based insurance can help protect productive assets, improve access to credit, new technologies and modern inputs, which together can improve farm productivity and livelihood outcomes (Barnett, Barrett and Skees, 2008; Hazell and Hess, 2010; Carter, Cheng and Sarris, 2016). The demand for such insurance products is, however, found to be uneven and scarce (Binswanger-Mkhize, 2012). This observation has since given rise to a number of studies examining the determinants of insurance take-up among farmers in developing countries (McIntosh, Sarris and Papadopoulos, 2013; Mobarak and Rosenzweig, 2013; Cole et al., 2013; Cole, Stein and Tobacman, 2014a; Dercon et al., 2014; Karlan et al., 2014; Clarke et al., 2015; Cai and Song, 2017) and the welfare impacts of weather insurance (Chantarat et al., 2016; Carter, Cheng and Sarris, 2016). Recent studies have shown that the insurance demand increases with the farmer’s own receipt of insurance payouts and the receipt of payouts by others in the community in the previous year (Karlan et al., 2014; Cole, Stein and Tobacman, 2014a). But are farmers’ decisions to purchase insurance sensitive to the type of weather shocks experienced? In this article, we study how past experience of extreme rainfall and droughts affects the take-up of insurance among farmers in India. This question is important for two reasons. First, covariate risk in agriculture is particularly difficult to ensure against, as it also affects neighbouring households and limits the effectiveness of traditional risk-sharing mechanisms. Index-based insurance provides an alternative, as it minimises moral hazard and adverse selection problems normally associated with standard insurance programmes. This assumption, however, hinges on the insurance policies offered being relevant and applicable to the local context in which the product is advertised and sold. Determining the role of weather is, therefore, vital in designing and implementing weather index insurance. Second, agents react to the occurrence of recent adverse events, which tends to generate demand for protective activities (Browne and Hoyt, 2000; Gallagher, 2014; Kunreuther, Sanderson and Vetschera, 1985). Climate change is likely to increase the occurrence of extreme weather events, requiring that better instruments to insure against adverse events become available to the most vulnerable. A recent study estimates that ‘a shift in the monsoon represents the gravest threat that climate change poses to lives and livelihoods in South Asia’ (Amrith, 2016). Using historical precipitation data, we identify the occurrence of extremely dry and wet periods in different times of crop growth. We match the extreme weather events using village-level coordinates with an 8-year household panel dataset from a randomised control trial on index-based insurance covering 56 villages located in the state of Gujarat, India. The identifying assumption is that, conditional on a village’s geography, it is random whether a village experiences an extreme weather event in a particular year. Because insurance policies offered to the household distinguish between rainfall deficiency and excess rainfall, we calculate monthly anomalies normalised by the standard deviation (i.e. z-scores) and identify wet and dry events for the growing and harvesting period, respectively. We first use the extreme weather event variables to investigate the link with individual and average village-level crop loss. We then analyse what type of events influence a farmer’s decision to purchase insurance, and to what extent this effect works jointly with the community effects found by Cole, Stein and Tobacman (2014a). Finally, we consider spill-over effects on non-purchasers from rainfall deficiency and excess rainfall. Our work departs from Cole, Stein and Tobacman (2014a), who use the same data and a similar identification strategy to study community-level effects, in that it considers the importance of different types of extreme weather events for insurance take-up. While previous studies consider rainfall levels (Giné Townsend and Vickery, 2007; Mobarak and Rosenzweig, 2013; Karlan et al., 2014; Di Falco et al., 2014; Coble et al., 1996; Deschênes and Greenstone, 2007), we calculate weather anomalies in relation to long-term village-level precipitation averages. This allows us to compensate for the effect that rainfall measures over a short period of time or a narrow area typically have larger variability.1 We find that the decision to purchase insurance increases after experiencing excess rain during the harvest and not as a consequence of rainfall deficiency in the crop planting and growing period. We also find that the combination of extreme rainfall and the observed village-level payouts increases dramatically the likelihood of insurance purchase. Experiencing excess rainfall in the previous year leads to a 14–20 percentage point higher likelihood of insurance purchase, while observing a payout of Rs. 1,000 in the village and extreme rainfall leads to a 57–67 percentage point higher likelihood of insurance purchase. The missing link between rainfall deficiency and insurance purchase can be explained by widespread access to irrigation in Gujarat state. These results are robust to different measures of extreme weather events, including placebo estimations, intensity of extreme weather events, and using satellite weather data to account for measurement error from misreporting or missing values at the rain gauges. Our findings suggest that weather shocks and insurance experience at the village-level boost index insurance demand. This is in line with the earlier evidence on agricultural insurance demand (Turvey, 2001) and Giné Townsend and Vickery (2007) who found that index insurance operates like a disaster insurance by primarily insuring farmers against extreme tail events of the rainfall distribution. Our findings are also affirmative of earlier evidence that economic agents tend to overreact to recent occurrence of extreme events (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992). 2. Background 2.1. The demand for index-based insurance The recognition that households are much less successful in insuring against covariate weather risk than against idiosyncratic risk (Rosenzweig and Binswanger, 1993; Townsend, 1994) has motivated the design and implementation of weather-based index insurance for poor farmers. Advocates argue that index-based insurance is transparent, enables quick payouts and minimises moral hazard and adverse selection problems normally associated with standard insurance programmes and other risk-coping mechanisms. Moral hazard problems are irrelevant unless insurance purchasers can tamper with the measurement of rainfall at the gauge. Adverse selection problems are, in principle, bypassed by public availability of historical rainfall data, but they can occur if households purchase insurance only in years when risk is higher than normal (Khalil et al., 2007). In a setting without asymmetric information, a household’s willingness to pay for insurance is increasing in risk aversion, the expected insurance payout and the size of the insured risk. It is, however, decreasing in basis risk (i.e. the difference between the rainfall index and crop yields). Consistent with this standard model framework, Giné Townsend and Vickery (2008) find in their study of cross-sectional determinants of household insurance take-up that rainfall insurance purchase decreases with basis risk and increases with household wealth in Andhra Pradesh, India. Clarke et al. (2015) show that insurance demand decreases with price and basis risk: doubling a household’s distance to a reference weather station (a proxy for basis risk) decreases demand by 18 per cent. Highly risk-averse individuals are less likely to purchase insurance (Cole et al., 2013), particularly if they are unfamiliar with the insurance company (Giné Townsend and Vickery, 2008). It appears that such individuals give more weight to the costs of basis risk when the insurance contract does not perform than to the benefits from the increase in expected wealth when it does (Clarke, 2016). Jensen, Barrett and Mude (2016) show that with the possibility of basis risk, even at actuarially fair premium rates, risk-averse households may prefer no insurance over index insurance. Some of the characteristics found to limit insurance demand are financial illiteracy, lack of trust and understanding, availability of alternative risk-management tools and credit constraints. Lack of understanding of the insurance policy is an important barrier in the purchase decision as shown by Giné Townsend and Vickery (2008). Cole et al. (2013) test this in a randomised field experiment and find 36 per cent higher demand when the recommendation to the household comes from a trusted local agent. Consistent with this, demand is higher if households participate in village networks (Giné Townsend and Vickery, 2008), and if households already have previous experience in the insurance market, higher financial literacy and greater familiarity with probability concepts (Cole et al., 2013). Instead of purchasing insurance, farmers can diversify production, save, borrow or engage in a number of informal risk-sharing arrangements. Mobarak and Rosenzweig (2012) show that index insurance and risk-sharing are complements, especially for those for whom basis risk is higher. Dercon et al. (2014) demonstrate substantially higher insurance take-up rates among groups of individuals that can share risk. Previous experimental and non-experimental evidence also suggests that liquidity constraints reduce demand (Giné Townsend and Vickery, 2008; Cole et al., 2013), which is consistent with the predictions of the traditional model with borrowing constraints. High premiums relative to the expected payout decrease the rainfall insurance adoption (Giné Townsend and Vickery, 2007; Michler, Viens and Shively, 2015). Considering the size of the insurance premium relative to the expected payout, Giné Townsend and Vickery (2007) find that insurance premiums are on average around three times larger than expected payouts. Cole et al. (2013) investigate this at the household level by estimating the slope of the demand curve by randomly varying the price of insurance products offered to households. They find significant price sensitivity with a point estimate suggesting that insurance demand would increase 36–66 per cent if it could be priced at payout ratios similar to US retail insurance contracts. In line with this result, Michler, Viens and Shively (2015) conclude that the loading factor on these types of contracts is excessive, and that the pricing provides little incentive for smallholders to purchase insurance. However, even when farmers are given a very large price discount, few households are found to actually purchase insurance (Cole et al., 2013). Thus, lower prices alone are unlikely to be sufficient to trigger widespread index insurance adoption. A number of studies show that information generated by insurance payouts has community-wide effects on insurance demand (Browne and Hoyt, 2000; Cole et al., 2013; Cole, Stein and Tobacman, 2014a). Berhane et al. (2013) observe that ‘nothing sells insurance like insurance payouts’. Specifically, a payout of Rs. 1,000 (equivalent to 5 days of work) increases the probability of households purchasing insurance in the next year by 25–50 per cent (Cole, Stein and Tobacman, 2014a). This result holds both for insurance purchasers and non-purchasers (who did not receive a payout), and increases in the number of residents receiving the payout in the previous year. A significant fraction of households in the study by Giné Townsend and Vickery (2008) cite advice from other farmers as an important determinant of the participation decision. Past experience with disasters motivates insurance adoption (Baumann and Sims, 1978). A number of studies have shown that demand is increasing in flood damages of the prior year (Browne and Hoyt, 2000; Osberghaus, 2017; Siegrist and Gutscher, 2008; Kunreuther and Pauly, 2006). In fact, a large jump in insurance take-up occurs immediately after a flood, but the demand starts decreasing steadily after a while (Gallagher, 2014). One possible explanation for the importance of recency is availability bias, due to which the ease with which one can recall an event determines the subjective probability assigned to it (Tversky and Kahneman, 1973). A recent accident or disaster is likely to bear disproportionately more weight in decision-making than the one in the distant past and thus may generate demand for protective activities. It is also observed that people stop renewing insurance if they perceive it as an investment rather than a contingent claim, so if the payout is not collected for a prolonged period, the policy will not be renewed (Kunreuther, Sanderson and Vetschera, 1985). Another explanation is that individuals learn from experience in line with the Bayesian learning decision model where risk perceptions are updated on the basis of new information from experiencing adverse events. Risk perception will decline if no adverse events such as floods are experienced over a certain time period (Meyer, 2012; Haer et al., 2017; Kunreuther and Pauly, 2006). 2.2. Extreme weather and Gujarat state Gujarat state has a total land area of 196,024 km2 of which 4.29 million hectares are devoted to grain production, 3.08 million hectares to oil seeds and 2.69 million hectares to cotton (Ministry of Agriculture, 2015). These figures show that Gujarat state is an important agricultural region for India. Apart from cultivating grains, such as wheat, rice and maize, Gujarat is the leading producer of oil seeds in the country with 21 per cent of total oil seed production (Ministry of Agriculture, 2015). With around six million tons of groundnuts and 1.9 billion tons of cotton produced in 2014, Gujarat is ahead of all other Indian states in groundnut and cotton production. Figure 1 shows that both cotton production area and output have increased since 2007. Average output has grown by 32 per cent since 2007, when 1.4 million tons of cotton were produced. Average yields have grown by 19 per cent, from 581 kg/ha in 2007 to 692 kg/ha in 2013 (Ministry of Agriculture, 2015). This is observed against an increasing trend of non-operative cotton mills in the state. A positive trend in production is also observed for rice, wheat and local grain varieties, while maize and pulses have seen a decline in production since 2012. Fig. 1. View largeDownload slide Production area and output of most important types of crops in Gujarat. Note: Authors’ calculation based on Ministry of Agriculture (2015). Fig. 1. View largeDownload slide Production area and output of most important types of crops in Gujarat. Note: Authors’ calculation based on Ministry of Agriculture (2015). Agriculture in Gujarat is sustained by uneven rainfall. The monsoon season with high rainfall is called Kharif, and it usually lasts from June/July to September/October. Delay in the monsoon onset is a concern in Gujarat (Jain et al., 2015) and some districts, such as Saurashtra and Kutch, may not receive enough rain even during the Kharif (Hirway, Kashyap and Shah, 2002). It is estimated that only 3 per cent of the total area receives high rainfall of 1150 mm and above; 31 per cent receives medium rainfall with 750–1150 mm, and 66 per cent receives low rainfall of less than 750 mm (Hirway, Kashyap and Shah, 2002). That is why a substantial share of the production area in Gujarat is irrigated. For example, 57 per cent of cotton producing area is irrigated, and so is 46 per cent of land for grains, and 35 per cent of land for oil seed production (Ministry of Agriculture, 2015). The main Kharif crops are groundnuts and cotton, while wheat is considered as a winter crop grown from end-October to mid-December depending on the end of the monsoon period. Groundnuts are sown at the start of the rainy season, usually the third week of June, and cotton is planted around the same time, depending on the onset of the monsoon. Extreme departures from usual rainfall, such as large-scale droughts and floods, can negatively affect agricultural output and local livelihoods (Kumar, Rajagopalan and Cane, 1999). Erratic rainfall and dry spells during important periods of plant growth, such as immediately after planting and late plant growth, cause damage to the crop. Similarly, excess rain at harvest can reduce the quality of groundnuts if the crop is left to dry in the field. Wet weather also negatively impacts lint colour and cotton seed quality. Several projections on the effects of climate change warn about increased variability of rainfall in South Asia from year to year, potentially making droughts and floods more common (Challinor et al., 2009; Naidu et al., 2015). 3. Data and key measures To conduct the empirical analysis, we merge two primary data sources. The first source is a household-level panel that contains information on insurance purchasing decisions and household characteristics, including information about revenue and crop losses. The sample includes households from a random sample of 60 villages spread across three districts in Indian state Gujarat (the districts of Ahmadabad, Anand and Patan). The information on insurance purchasing decisions comes from a Gujarat-based NGO named Self-Employed Women’s Association (SEWA), which has carried out the index insurance project. They marketed rainfall insurance to a selected sample of residents from 2006 to 2013. The design of insurance policies varied by location and year. Product marketing included discounts, targeted messages and special offers on multiple purchases, which were randomly assigned each year. The data on households’ willingness to pay for insurance were obtained through the Becker–de Groot–Marschak (BDM) mechanism (Becker, de Groot and Marschak, 1964). The combination of random insurance policy offering and willingness to pay gives exogenous variation to insurance take-up and reveals households’ demand for weather insurance. Further details of the marketing interventions and experimental design can be found in Cole, Stein and Tobacman (2014a) and Cole, Xavier and Vickery (2014b). The rainfall index insurance policies provide indemnity payments according to values of an index (rainfall in this case) that serves as a proxy for losses instead of the estimated individual losses of different policyholders. The insured sum is based on production costs, determined on an agreed value basis, while the payouts follow a pre-established scale detailed in the insurance policy. The policies from our sample provided coverage against adverse rainfall events for the summer monsoon cropping season running from approximately June to September or October, depending on the village considered. The monsoon season is divided into three phases corresponding to the sowing, growing and harvest period. The duration of the first two phases is 35 days, while the last phase lasts for 40 days. The specified payout per hectare is calculated as cumulative rainfall during a fixed time period between the start and the end dates of the phases, measured at a nearby rain gauge. The start of the first phase is triggered by the monsoon rains. Specifically, the first phase starts when the accumulated rainfall since June 1 exceeds 50 mm, or on July 1, if accumulated rainfall is below 50 mm by the end of June. Insurance payouts in the first two phases are linked to low rainfall. The policy pays zero if accumulated rain during the first two phases exceeds an upper threshold or a ‘strike’. Otherwise, the policy pays Rs. 10 for each millimetre of rainfall deficiency relative to the strike until a lower threshold or ‘exit’ is reached. The policy pays a fixed amount if rainfall is below the exit value. Policies for the final, third phase is reverse but similar in structure to the first two phases, so that phase 3 insures against excess rain, which may cause damage to the crops during the harvest. The specific policy terms and payout differ by tehsil (sub-district level) and year. The payouts could occur several times per season and this is not different from index insurance products in other developing countries (e.g. Karlan et al., 2014). Every year, households must repurchase the insurance product to stay covered. Households were visited and offered the insurance each year in April or May, and were free to purchase multiple policies. In 2006, the insurance product was first introduced to households across 32 villages. The programme was extended to 20 additional villages in 2007. Within each village, 15 households were surveyed. Of these, five were randomly selected, five had previously purchased other forms of insurance and five were identified by local SEWA employees as likely to purchase insurance. Households likely to purchase insurance were purposefully over-sampled to ensure a substantial number of buyers. Finally, the programme expanded to 50 households in eight additional villages in 2009. The total sample that has been surveyed and assigned to receive insurance marketing by SEWA consists of 1,160 households in 60 villages. We use only the balanced panel of households who took part in the survey and marketing interventions each year after they have been added to the project. Four villages were dropped from the original sample as we were unable to map the location of the villages. The analysis is thus based on a sample of 905 households and 5,214 household-years for which we can observe current and lagged insurance coverage decisions. The second data source is historical gauge-based monthly precipitation data from the Global Precipitation Climate Centre (GPCC), which we merged with household data using village coordinates.2 The data on historical precipitation cover the period from 1901 to 2013. Since the survey villages might be situated on the border of the satellite precipitation grids, the rainfall information is reweighted using the four nearest precipitation grids. Particularly, precipitation for a given village at a given month is calculated as the weighted average of the measurement in the four cardinal points of the gridded cell to which the village belongs. To ensure that more weight is assigned to data points that are nearer the surveyed village, we use the inverse distance between the villages and precipitation observations. We calculate monthly anomalies from the historical monthly precipitation for the period from 1901 to 2005 by village and normalise by the standard deviation (i.e. z-scores). Based on the z-scores, we define extreme events for the summer monsoon months and, in line with insurance policies, consider droughts occurring in June, July and August, and excess rainfall in September and October. Extreme excess rainfall events are defined as the normalised precipitation anomaly larger than 1, while droughts are defined as the normalised precipitation less than −1 (Nanjundiah et al., 2013).3 Based on this normalisation, we find that farmers over the eight-year period, on average, experienced more extreme events during the harvest period compared to events occurring in the growing and planting season.4 For illustrative purposes, Figure 2 shows variation of extreme weather events in the crop-growing season aggregated across villages and time. We see that farmers, on average, experienced a range of extreme weather events between 2006 and 2013. The extreme weather measures constructed in this way are independent and different from rainfall levels used in designing insurance policies and determining payouts. In relation to the three types of insurance policies marketed by SEWA, dry events (marked in red) are relevant for phases 1 and 2, while the excess rainfall (marked in blue) is relevant for phase 3. The actual events used for the empirical analysis, however, depend on village location and time. This means that events are identified by village each year. Fig. 2. View largeDownload slide Distribution of extreme events across sampled villages in Gujarat. Fig. 2. View largeDownload slide Distribution of extreme events across sampled villages in Gujarat. Figure 3 shows village-specific experiences of dry and wet spells for each of the survey years. For instance, excess rainfall, usually occurring in September and October, was more frequent than rain deficiency in 2011, 2012 and 2013, while rain shortage was more common in 2009 and 2010. Moreover, Figure 3 shows that villages in the north experienced more dry spells in 2012 than in 2011. Fig. 3. View largeDownload slide Location of extreme weather by village and year. Fig. 3. View largeDownload slide Location of extreme weather by village and year. 3.1. Household-level descriptive statistics Around 40 per cent of households from the sample have purchased insurance at some point between 2006 and 2013 (Table A1 in the Appendix). The share of purchasers has increased from 19 to 56 per cent in that period. The average rate of repurchase was 18 per cent, growing from 5 per cent in 2007 to 32 per cent in 2013. New purchasers comprise 19 per cent of the sample, increasing from 18 per cent in 2007 to 24 per cent in 2013. Around 14 per cent of households who purchased insurance in one year decided not to repurchase in the following year. The variability of weather conditions in the observed period meant that crop loss and insurance payouts did not occur every year. Around 30 per cent of households have suffered crop loss at any point between 2006 and 2012, and around 13 per cent of households from the sample received a payout. Variation in both crop loss and payout is substantial. There were no payouts in 2006 and 2007, even though 319 and 146 households reported crop loss in those years. The proportion of payouts varied between 6 and 36 per cent, observed in 2008 and 2012, respectively. The proportion of households affected by crop loss ranged from 22 per cent in 2007 to 77 per cent in 2006. None of the households reported crop loss in 2013, and there were no payouts in that year. The insurance is designed so that a household can purchase as many policies per phase of the crop-growing cycle as wanted. Households from the sample have on average been buying two policies, starting with one in the 2006–2008 period and reaching five policies on average in 2010. Policies were subsidised in 2010 with a ‘buy one get one free’ offer, which has decreased their price and increased the number of policies purchased (Cole, Stein and Tobacman, 2014a). Hereafter, the average number of policies bought has decreased to two (2011–2013 period). The average price paid per policy was Rs. 60, while the average individual payout received was Rs. 143 (for all policies bought). On average, 16 households in a village received payouts. The average revenue lost due to crop loss was Rs. 1,274 at the village level, with standard deviation values around three times the size of the mean.5 Control variables used in all estimations are shown in Table A2, emphasising the proportion of households that randomly received different marketing offers.6 We see that 24 per cent of households participated in the BDM willingness to pay experiment, and 36 per cent of households participated in a game of bidding for four different insurance policies, which are described in Cole et al. (2013). Around 7 per cent of households received information about extreme weather on flyers, while 57 per cent received flyers about exposure risk. Discounts were offered to around 1 per cent of the sample. Households were also exposed to a series of video treatments informing about SEWA brand (3.7 per cent), peer experience (around 4 per cent), benefits (‘Peer endorsed’ and ‘Information treatment: Pays 2/10 years’, around 2 per cent) or potential difficulties if not purchasing insurance (‘Vulnerability frame,’ around 2 per cent). Around 1 per cent of the sample received flyers informing them about the risks or benefits of purchasing weather insurance. 4. Empirical analysis Due to the experimental nature of the data and the panel structure, we first estimate a linear probability model of the probability of household insurance purchase as a function of our extreme event variables and treatment variables. The estimation follows equation (1): buyijt=αi+βDjt−1+γFjt−1+ϕ1Xijt+ϕ2Xijt−1+τt+εijt (1) where buyijt is an indicator that takes value one if household i in village j has purchased insurance in year t, Djt−1 is an indicator for experiencing rainfall deficiency and Fjt−1 is an indicator for experiencing excess rainfall in village j in the year before insurance purchase.7 Xijt and Xijt−1 are same-year and previous year’s controls, respectively, for when the household entered the experiment, for exposure to insurance marketing and for individual and village-level insurance payouts. αi are household fixed effects, τt are year dummies and εijt are random disturbances. Standard errors are clustered at the village level throughout our analysis. Crop loss is calculated as the difference between agricultural output and the mean value of output in all prior years where crop loss was not reported.8 The set of control variables includes individual or village-level insurance payouts in previous period, which could be considered endogenous with respect to the decision to purchase insurance. We therefore also estimate a two-stage least squares (2SLS) model where recent insurance payouts are instrumented with variables indicating random assignment into receiving different marketing packages (promotional videos, flyers, etc.). In this setup, only variation in recent insurance payouts due to random assignment into different marketing package treatments is used to provide variation in actual insurance payouts in the insurance purchase equation. The treatments are valid instruments for recent insurance payouts under the assumption that they affect the outcome of interest only through changing purchase status, and not through any other channel. Due to the random assignment into receiving different marketing packages, instrumental variables (IVs) are likely to be uncorrelated with the error term, as required for 2SLS to be consistent. The analysis is split into two sections. In the next section, we consider the link between our measures of extreme weather events and revenue lost due to crop loss as well as the mean village revenue lost due to crop loss. In Section 4.2, we present the results on the type of weather events that matter for insurance take-up. The findings are further discussed in Section 5 following the empirical analysis. 4.1. Crop loss and extreme events Table 1 considers the relationship between crop loss and extreme weather events. Individual-level crop loss may be caused by a variety of factors. One such factor is weather, while other factors include pest and plant diseases, shirking and sickness. Household fixed effects make it possible to account for time-invariant household-level characteristics, such as ability. We therefore explain the part of the variation in revenue lost due to crop loss as a consequence of extreme weather events only. Table 1. Extreme events and revenue lost due to crop loss Revenue lost due to crop loss (Rs ’0000) Mean village revenue lost due to crop loss (Rs ’0000) (1) (2) (3) (4) Excess rainfall 0.076 *** 0.075 *** 0.092 *** 0.091 *** (0.023) (0.023) (0.031) (0.031) Lack of rainfall 0.009 0.013 −0.003 −0.001 (0.014) (0.016) (0.030) (0.030) Purchased insurance (dummy) −0.007 0.012 (0.008) (0.011) Village payout per policy (Rs ’000s) −0.032 −0.036 (0.030) (0.054) Individual payout (Rs ’000s) 0.005 0.013 (0.018) (0.024) Constant 0.135 *** 0.010 0.134 *** 0.056 (0.016) (0.027) (0.050) (0.041) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Revenue lost due to crop loss (Rs ’0000) Mean village revenue lost due to crop loss (Rs ’0000) (1) (2) (3) (4) Excess rainfall 0.076 *** 0.075 *** 0.092 *** 0.091 *** (0.023) (0.023) (0.031) (0.031) Lack of rainfall 0.009 0.013 −0.003 −0.001 (0.014) (0.016) (0.030) (0.030) Purchased insurance (dummy) −0.007 0.012 (0.008) (0.011) Village payout per policy (Rs ’000s) −0.032 −0.036 (0.030) (0.054) Individual payout (Rs ’000s) 0.005 0.013 (0.018) (0.024) Constant 0.135 *** 0.010 0.134 *** 0.056 (0.016) (0.027) (0.050) (0.041) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Note: OLS. Standard errors clustered at the village level. ***p<0.01. View Large Table 1. Extreme events and revenue lost due to crop loss Revenue lost due to crop loss (Rs ’0000) Mean village revenue lost due to crop loss (Rs ’0000) (1) (2) (3) (4) Excess rainfall 0.076 *** 0.075 *** 0.092 *** 0.091 *** (0.023) (0.023) (0.031) (0.031) Lack of rainfall 0.009 0.013 −0.003 −0.001 (0.014) (0.016) (0.030) (0.030) Purchased insurance (dummy) −0.007 0.012 (0.008) (0.011) Village payout per policy (Rs ’000s) −0.032 −0.036 (0.030) (0.054) Individual payout (Rs ’000s) 0.005 0.013 (0.018) (0.024) Constant 0.135 *** 0.010 0.134 *** 0.056 (0.016) (0.027) (0.050) (0.041) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Revenue lost due to crop loss (Rs ’0000) Mean village revenue lost due to crop loss (Rs ’0000) (1) (2) (3) (4) Excess rainfall 0.076 *** 0.075 *** 0.092 *** 0.091 *** (0.023) (0.023) (0.031) (0.031) Lack of rainfall 0.009 0.013 −0.003 −0.001 (0.014) (0.016) (0.030) (0.030) Purchased insurance (dummy) −0.007 0.012 (0.008) (0.011) Village payout per policy (Rs ’000s) −0.032 −0.036 (0.030) (0.054) Individual payout (Rs ’000s) 0.005 0.013 (0.018) (0.024) Constant 0.135 *** 0.010 0.134 *** 0.056 (0.016) (0.027) (0.050) (0.041) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Note: OLS. Standard errors clustered at the village level. ***p<0.01. View Large The dependent variable in columns 1 and 2 is individual revenue lost due to crop loss (in ’0000s Rs), while the dependent variable in columns 3 and 4 is the mean village revenue lost due to crop loss. We find a positive effect of excess rain (i.e. extreme event above 1 SD) on crop loss in column 1, while the lack of rainfall does not seem to be important in explaining revenue lost due to crop loss. The missing link between lack of rainfall and crop loss may be explained by the widespread use of irrigation in Gujarat state.9 Inclusion of control variables in column 2 affirms the results. The effect of excess rain is even stronger in column 3, where the dependent variable is mean village revenue lost, and the coefficient size does not change when additional control variables are included in column 4. The economically small effect from weather variability on households’ experienced crop loss is consistent with Michler, Viens and Shively (2015), who found that variance in weather accounts for an important but small fraction of total variance in crop output. The reasonably low correlation may also be explained by presence of basis risk in the identified weather events (Jensen, Barrett and Mude, 2016). 4.2. Insurance purchase In this section, we present the estimates of the insurance purchase decisions with lagged measures of extreme weather and other key determinants of insurance purchase decisions. Table 2 considers the sample of insurance purchasers, while Table 3 concentrates on potential spill-over effects to non-purchasers. Table 2. What type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.143 *** 0.197 *** 0.196 *** 0.164 *** 0.181 *** 0.165 *** (0.038) (0.044) (0.051) (0.050) (0.053) (0.054) (0.053) Lack of rainfall in previous year 0.001 0.010 −0.017 −0.013 −0.008 −0.008 −0.010 (0.044) (0.056) (0.061) (0.061) (0.066) (0.064) (0.066) Village payout in previous year 0.566 *** 0.670 *** 0.576 *** 0.670 *** 0.550 *** (0.146) (0.190) (0.199) (0.192) (0.198) Individual payout in previous year −0.045 −0.048 −0.062 −0.035 (0.048) (0.049) (0.056) (0.055) Excess rainfall × village payout in previous year 0.638 * 0.757 (0.334) (0.478) Lack of rainfall × village payout in previous year −0.203 −0.183 (0.363) (0.501) Excess rainfall × individual payout in previous year 0.103 −0.048 (0.074) (0.106) Lack of rainfall × individual payout in previous year −0.031 −0.017 (0.089) (0.114) Number of insurance policies bought in previous year 0.008 0.005 0.006 0.005 (0.015) (0.015) (0.015) (0.016) Constant 0.537 *** 0.572 *** 0.409 *** 0.391 *** 0.409 *** 0.393 *** 0.412 *** (0.014) (0.081) (0.128) (0.130) (0.129) (0.130) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.143 *** 0.197 *** 0.196 *** 0.164 *** 0.181 *** 0.165 *** (0.038) (0.044) (0.051) (0.050) (0.053) (0.054) (0.053) Lack of rainfall in previous year 0.001 0.010 −0.017 −0.013 −0.008 −0.008 −0.010 (0.044) (0.056) (0.061) (0.061) (0.066) (0.064) (0.066) Village payout in previous year 0.566 *** 0.670 *** 0.576 *** 0.670 *** 0.550 *** (0.146) (0.190) (0.199) (0.192) (0.198) Individual payout in previous year −0.045 −0.048 −0.062 −0.035 (0.048) (0.049) (0.056) (0.055) Excess rainfall × village payout in previous year 0.638 * 0.757 (0.334) (0.478) Lack of rainfall × village payout in previous year −0.203 −0.183 (0.363) (0.501) Excess rainfall × individual payout in previous year 0.103 −0.048 (0.074) (0.106) Lack of rainfall × individual payout in previous year −0.031 −0.017 (0.089) (0.114) Number of insurance policies bought in previous year 0.008 0.005 0.006 0.005 (0.015) (0.015) (0.015) (0.016) Constant 0.537 *** 0.572 *** 0.409 *** 0.391 *** 0.409 *** 0.393 *** 0.412 *** (0.014) (0.081) (0.128) (0.130) (0.129) (0.130) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01*p<0.1. View Large Table 2. What type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.143 *** 0.197 *** 0.196 *** 0.164 *** 0.181 *** 0.165 *** (0.038) (0.044) (0.051) (0.050) (0.053) (0.054) (0.053) Lack of rainfall in previous year 0.001 0.010 −0.017 −0.013 −0.008 −0.008 −0.010 (0.044) (0.056) (0.061) (0.061) (0.066) (0.064) (0.066) Village payout in previous year 0.566 *** 0.670 *** 0.576 *** 0.670 *** 0.550 *** (0.146) (0.190) (0.199) (0.192) (0.198) Individual payout in previous year −0.045 −0.048 −0.062 −0.035 (0.048) (0.049) (0.056) (0.055) Excess rainfall × village payout in previous year 0.638 * 0.757 (0.334) (0.478) Lack of rainfall × village payout in previous year −0.203 −0.183 (0.363) (0.501) Excess rainfall × individual payout in previous year 0.103 −0.048 (0.074) (0.106) Lack of rainfall × individual payout in previous year −0.031 −0.017 (0.089) (0.114) Number of insurance policies bought in previous year 0.008 0.005 0.006 0.005 (0.015) (0.015) (0.015) (0.016) Constant 0.537 *** 0.572 *** 0.409 *** 0.391 *** 0.409 *** 0.393 *** 0.412 *** (0.014) (0.081) (0.128) (0.130) (0.129) (0.130) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.143 *** 0.197 *** 0.196 *** 0.164 *** 0.181 *** 0.165 *** (0.038) (0.044) (0.051) (0.050) (0.053) (0.054) (0.053) Lack of rainfall in previous year 0.001 0.010 −0.017 −0.013 −0.008 −0.008 −0.010 (0.044) (0.056) (0.061) (0.061) (0.066) (0.064) (0.066) Village payout in previous year 0.566 *** 0.670 *** 0.576 *** 0.670 *** 0.550 *** (0.146) (0.190) (0.199) (0.192) (0.198) Individual payout in previous year −0.045 −0.048 −0.062 −0.035 (0.048) (0.049) (0.056) (0.055) Excess rainfall × village payout in previous year 0.638 * 0.757 (0.334) (0.478) Lack of rainfall × village payout in previous year −0.203 −0.183 (0.363) (0.501) Excess rainfall × individual payout in previous year 0.103 −0.048 (0.074) (0.106) Lack of rainfall × individual payout in previous year −0.031 −0.017 (0.089) (0.114) Number of insurance policies bought in previous year 0.008 0.005 0.006 0.005 (0.015) (0.015) (0.015) (0.016) Constant 0.537 *** 0.572 *** 0.409 *** 0.391 *** 0.409 *** 0.393 *** 0.412 *** (0.014) (0.081) (0.128) (0.130) (0.129) (0.130) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01*p<0.1. View Large Table 3. Spill-over effects from different events on non-purchasers Insurance non-purchasers (1) (2) (3) (4) Excess rainfall in previous year 0.125 *** 0.163 *** 0.173 ** 0.140 ** (0.046) (0.038) (0.077) (0.067) Lack of rainfall in previous year 0.162 ** 0.038 −0.033 0.008 (0.062) (0.053) (0.057) (0.057) Village payout in previous year 0.284 ** 0.304 *** (0.108) (0.107) Excess rainfall × village payout in previous year 0.638 * (0.324) Lack of rainfall × village payout in previous year −0.512 (0.319) Constant 0.238 *** 0.339 *** 0.547 *** 0.520 *** (0.016) (0.038) (0.163) (0.170) Observations 3,274 3,274 3,274 3,274 Number of households 893 893 893 893 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Insurance non-purchasers (1) (2) (3) (4) Excess rainfall in previous year 0.125 *** 0.163 *** 0.173 ** 0.140 ** (0.046) (0.038) (0.077) (0.067) Lack of rainfall in previous year 0.162 ** 0.038 −0.033 0.008 (0.062) (0.053) (0.057) (0.057) Village payout in previous year 0.284 ** 0.304 *** (0.108) (0.107) Excess rainfall × village payout in previous year 0.638 * (0.324) Lack of rainfall × village payout in previous year −0.512 (0.319) Constant 0.238 *** 0.339 *** 0.547 *** 0.520 *** (0.016) (0.038) (0.163) (0.170) Observations 3,274 3,274 3,274 3,274 Number of households 893 893 893 893 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Note: OLS. Standard errors clustered at the village level. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table 3. Spill-over effects from different events on non-purchasers Insurance non-purchasers (1) (2) (3) (4) Excess rainfall in previous year 0.125 *** 0.163 *** 0.173 ** 0.140 ** (0.046) (0.038) (0.077) (0.067) Lack of rainfall in previous year 0.162 ** 0.038 −0.033 0.008 (0.062) (0.053) (0.057) (0.057) Village payout in previous year 0.284 ** 0.304 *** (0.108) (0.107) Excess rainfall × village payout in previous year 0.638 * (0.324) Lack of rainfall × village payout in previous year −0.512 (0.319) Constant 0.238 *** 0.339 *** 0.547 *** 0.520 *** (0.016) (0.038) (0.163) (0.170) Observations 3,274 3,274 3,274 3,274 Number of households 893 893 893 893 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Insurance non-purchasers (1) (2) (3) (4) Excess rainfall in previous year 0.125 *** 0.163 *** 0.173 ** 0.140 ** (0.046) (0.038) (0.077) (0.067) Lack of rainfall in previous year 0.162 ** 0.038 −0.033 0.008 (0.062) (0.053) (0.057) (0.057) Village payout in previous year 0.284 ** 0.304 *** (0.108) (0.107) Excess rainfall × village payout in previous year 0.638 * (0.324) Lack of rainfall × village payout in previous year −0.512 (0.319) Constant 0.238 *** 0.339 *** 0.547 *** 0.520 *** (0.016) (0.038) (0.163) (0.170) Observations 3,274 3,274 3,274 3,274 Number of households 893 893 893 893 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Note: OLS. Standard errors clustered at the village level. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large In Table 2, the coefficient estimate on excess rain is both statistically and economically significant, implying that excess rainfall during the harvest period causes a 11–20 percentage point increase in the probability of purchasing insurance in the next season. This result is consistent across columns 1–7 and robust to inclusion of various control variables. This effect is large, considering that 41 per cent of the villages at some point during the 8-year survey period experienced extreme excess rain. Estimation in column 3 controls for the effect of village payout per policy, while estimations in columns 4–7 control for the number of insurance policies purchased in previous year. The effect of excess rain on insurance purchase is smaller than the effect from village payout per policy, which is estimated to be 57–67 percentage points, depending on specification. This is in line with Cole, Stein and Tobacman (2014a) who find that a payout per policy of Rs. 1,000 causes a 50 percentage point increase in the probability of insurance purchase in the next season. In column 5, we interact the extreme weather event variables with the village payout per policy to allow for a varying effect from village payout between households that have experienced extreme weather in the previous year. The interaction term with excess rainfall is statistically significant, and the point estimate is of the same sign as the non-interacted variables. In comparison, the interaction with drought spells is insignificant. The joint test of the sum of the interaction with excess rain and the mean village payout per policy is significant at the 5 per cent level. This indicates that excess rain in the previous year leads to a significant increase in insurance take-up, and that it works jointly with the effect of previous village-level payouts. In fact, the combined effect size implies that experiencing excess rainfall increases the likelihood of insurance purchase by 80 percentage points for every Rs. 1,000 of village payout per policy in previous year. In column 6, we interact the extreme weather event variables with individual payout per policy. The interaction terms of both extreme event variables are insignificant. The joint test of the sum of the interaction and the individual payout variable yields an insignificant estimate. This indicates that, while depending on the extreme rainfall, the decision to purchase insurance is not driven by individual payout experience. We combine columns 5 and 6 in column 7 and find that the effect on the interaction term between excess rain and village payout per policy remains statistically significant, and that the effect size increases. The interaction term is only significant at the 10 per cent level, while the sum of the coefficients on the interaction term and the main effect from mean village payout remains statistically significant at the 1 per cent level. Number of insurance policies bought in previous year is not statistically significant in any estimation. We now consider spill-over effects from experience of extreme events on insurance non-purchasers. As shown in Table 3, we find a positive and significant spill-over effect from excess rain in the previous year, but no spill-over effect from rain deficiency on insurance non-purchasers. Consistent with Cole, Stein and Tobacman (2014a), we also find that previous insurance payouts at the village level increase the likelihood of insurance purchase the following year among the non-purchasers. Again, the interaction effect is found to be positive and statistically significant, suggesting that experience from village payout due to excess rainfall encourages non-purchasers to buy insurance the following year. Finally, we use the experimental nature of the data to conduct the IV estimation on a full sample of households. We instrument endogenous variables (the lag of village payout per policy and the lag of the number of insurance policies purchased) with variables that measure randomly assigned marketing packages and their interactions with lagged insurance payouts. The marketing package variables enter the estimation in a lagged form. The results shown in Table 4 conform with our baseline results in Table 2, with a smaller effect size. Table 4. IV estimation of the effect of extreme weather events on the full sample of households. Full sample (1) (2) (3) (4) (5) (6) Excess rainfall (t−1) 0.142 *** 0.145 *** 0.149 *** 0.117 *** 0.164 *** 0.109 ** (0.035) (0.031) (0.033) (0.033) (0.058) (0.055) Lack of rainfall (t−1) 0.014 0.013 −0.011 −0.001 −0.056 −0.110 (0.045) (0.043) (0.046) (0.050) (0.057) (0.079) Village payout (t−1) 0.293 *** 0.266 *** 0.383 *** 0.509 *** (0.091) (0.093) (0.114) (0.148) Excess rainfall (t−1)× village payout (t−1) 0.536 *** −0.070 (0.198) (0.370) Lack of rainfall (t−1) × village payout (t−1) −0.206 0.396 (0.190) (0.391) Excess rainfall (t−2) 0.132 ** 0.093 (0.060) (0.059) Lack of rainfall (t−2) 0.010 −0.032 (0.062) (0.068) Village payout (t−2) 0.473 *** (0.133) Excess rainfall (t−2)× village payout (t−2) 2.292 ** (0.982) Lack of rainfall (t−2)× village payout (t−2) 0.025 (0.352) Excess rainfall (t−3) −0.017 −0.198 (0.123) (0.129) Lack of rainfall (t−3) 0.072 0.014 (0.072) (0.063) Village payout (t−3) 0.228 ** (0.102) Excess rainfall (t−3)× village payout (t−3) 5.437 *** (1.781) Lack of rainfall (t−3)× village payout (t−3) 0.290 (0.335) Individual payout (t−1) 0.279 *** 0.277 *** 0.137 * 0.105 0.076 0.052 (0.078) (0.076) (0.077) (0.079) (0.070) (0.069) Number of insurance policies bought (t−1) −0.025 ** −0.019 ** −0.002 −0.001 −0.009 −0.006 (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) Observations 5,214 5,214 5,214 5,214 3,404 3,404 Number of households 905 905 905 905 905 905 Household FE Yes Yes Yes Yes Yes Year dummies YES Yes Yes Yes Yes Yes Controls YES Yes Yes Yes Yes Yes Cragg–Donald F-stat 24.66 23.53 23.70 23.33 17.99 17.76 Full sample (1) (2) (3) (4) (5) (6) Excess rainfall (t−1) 0.142 *** 0.145 *** 0.149 *** 0.117 *** 0.164 *** 0.109 ** (0.035) (0.031) (0.033) (0.033) (0.058) (0.055) Lack of rainfall (t−1) 0.014 0.013 −0.011 −0.001 −0.056 −0.110 (0.045) (0.043) (0.046) (0.050) (0.057) (0.079) Village payout (t−1) 0.293 *** 0.266 *** 0.383 *** 0.509 *** (0.091) (0.093) (0.114) (0.148) Excess rainfall (t−1)× village payout (t−1) 0.536 *** −0.070 (0.198) (0.370) Lack of rainfall (t−1) × village payout (t−1) −0.206 0.396 (0.190) (0.391) Excess rainfall (t−2) 0.132 ** 0.093 (0.060) (0.059) Lack of rainfall (t−2) 0.010 −0.032 (0.062) (0.068) Village payout (t−2) 0.473 *** (0.133) Excess rainfall (t−2)× village payout (t−2) 2.292 ** (0.982) Lack of rainfall (t−2)× village payout (t−2) 0.025 (0.352) Excess rainfall (t−3) −0.017 −0.198 (0.123) (0.129) Lack of rainfall (t−3) 0.072 0.014 (0.072) (0.063) Village payout (t−3) 0.228 ** (0.102) Excess rainfall (t−3)× village payout (t−3) 5.437 *** (1.781) Lack of rainfall (t−3)× village payout (t−3) 0.290 (0.335) Individual payout (t−1) 0.279 *** 0.277 *** 0.137 * 0.105 0.076 0.052 (0.078) (0.076) (0.077) (0.079) (0.070) (0.069) Number of insurance policies bought (t−1) −0.025 ** −0.019 ** −0.002 −0.001 −0.009 −0.006 (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) Observations 5,214 5,214 5,214 5,214 3,404 3,404 Number of households 905 905 905 905 905 905 Household FE Yes Yes Yes Yes Yes Year dummies YES Yes Yes Yes Yes Yes Controls YES Yes Yes Yes Yes Yes Cragg–Donald F-stat 24.66 23.53 23.70 23.33 17.99 17.76 Note: IV. Standard errors clustered at the village level. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table 4. IV estimation of the effect of extreme weather events on the full sample of households. Full sample (1) (2) (3) (4) (5) (6) Excess rainfall (t−1) 0.142 *** 0.145 *** 0.149 *** 0.117 *** 0.164 *** 0.109 ** (0.035) (0.031) (0.033) (0.033) (0.058) (0.055) Lack of rainfall (t−1) 0.014 0.013 −0.011 −0.001 −0.056 −0.110 (0.045) (0.043) (0.046) (0.050) (0.057) (0.079) Village payout (t−1) 0.293 *** 0.266 *** 0.383 *** 0.509 *** (0.091) (0.093) (0.114) (0.148) Excess rainfall (t−1)× village payout (t−1) 0.536 *** −0.070 (0.198) (0.370) Lack of rainfall (t−1) × village payout (t−1) −0.206 0.396 (0.190) (0.391) Excess rainfall (t−2) 0.132 ** 0.093 (0.060) (0.059) Lack of rainfall (t−2) 0.010 −0.032 (0.062) (0.068) Village payout (t−2) 0.473 *** (0.133) Excess rainfall (t−2)× village payout (t−2) 2.292 ** (0.982) Lack of rainfall (t−2)× village payout (t−2) 0.025 (0.352) Excess rainfall (t−3) −0.017 −0.198 (0.123) (0.129) Lack of rainfall (t−3) 0.072 0.014 (0.072) (0.063) Village payout (t−3) 0.228 ** (0.102) Excess rainfall (t−3)× village payout (t−3) 5.437 *** (1.781) Lack of rainfall (t−3)× village payout (t−3) 0.290 (0.335) Individual payout (t−1) 0.279 *** 0.277 *** 0.137 * 0.105 0.076 0.052 (0.078) (0.076) (0.077) (0.079) (0.070) (0.069) Number of insurance policies bought (t−1) −0.025 ** −0.019 ** −0.002 −0.001 −0.009 −0.006 (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) Observations 5,214 5,214 5,214 5,214 3,404 3,404 Number of households 905 905 905 905 905 905 Household FE Yes Yes Yes Yes Yes Year dummies YES Yes Yes Yes Yes Yes Controls YES Yes Yes Yes Yes Yes Cragg–Donald F-stat 24.66 23.53 23.70 23.33 17.99 17.76 Full sample (1) (2) (3) (4) (5) (6) Excess rainfall (t−1) 0.142 *** 0.145 *** 0.149 *** 0.117 *** 0.164 *** 0.109 ** (0.035) (0.031) (0.033) (0.033) (0.058) (0.055) Lack of rainfall (t−1) 0.014 0.013 −0.011 −0.001 −0.056 −0.110 (0.045) (0.043) (0.046) (0.050) (0.057) (0.079) Village payout (t−1) 0.293 *** 0.266 *** 0.383 *** 0.509 *** (0.091) (0.093) (0.114) (0.148) Excess rainfall (t−1)× village payout (t−1) 0.536 *** −0.070 (0.198) (0.370) Lack of rainfall (t−1) × village payout (t−1) −0.206 0.396 (0.190) (0.391) Excess rainfall (t−2) 0.132 ** 0.093 (0.060) (0.059) Lack of rainfall (t−2) 0.010 −0.032 (0.062) (0.068) Village payout (t−2) 0.473 *** (0.133) Excess rainfall (t−2)× village payout (t−2) 2.292 ** (0.982) Lack of rainfall (t−2)× village payout (t−2) 0.025 (0.352) Excess rainfall (t−3) −0.017 −0.198 (0.123) (0.129) Lack of rainfall (t−3) 0.072 0.014 (0.072) (0.063) Village payout (t−3) 0.228 ** (0.102) Excess rainfall (t−3)× village payout (t−3) 5.437 *** (1.781) Lack of rainfall (t−3)× village payout (t−3) 0.290 (0.335) Individual payout (t−1) 0.279 *** 0.277 *** 0.137 * 0.105 0.076 0.052 (0.078) (0.076) (0.077) (0.079) (0.070) (0.069) Number of insurance policies bought (t−1) −0.025 ** −0.019 ** −0.002 −0.001 −0.009 −0.006 (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) Observations 5,214 5,214 5,214 5,214 3,404 3,404 Number of households 905 905 905 905 905 905 Household FE Yes Yes Yes Yes Yes Year dummies YES Yes Yes Yes Yes Yes Controls YES Yes Yes Yes Yes Yes Cragg–Donald F-stat 24.66 23.53 23.70 23.33 17.99 17.76 Note: IV. Standard errors clustered at the village level. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large In columns 1–3, we show that excess rainfall increases the likelihood of insurance purchase by around 15 percentage points. In column 4, we find that excess rainfall increases the likelihood of insurance purchase by 54 percentage points for every Rs. 1,000 village payout. The decrease in the size of the effect compared to Table 2 is mainly driven by a decline in the importance of mean village payout per policy. We add the second and the third lag of our extreme event variables in column 5, as well as the interactions of lagged extreme weather events and lagged village payouts in column 6. The excess of rainfall 2 years back also matters for insurance uptake, but the size of the coefficient is smaller than the effect from excess rainfall in the previous year. We also find a persistence in the effect of payments made at the village level, but the size of coefficient estimates decreases over time. Finally, the second and the third lag of the interaction between the excess rain and the mean village payout correlate positively with the decision to purchase insurance, while the first lag loses significance. 5. Discussion We have shown that the likelihood of insurance purchase is driven by village payouts caused by wet spells in the previous year. Experiencing excess rainfall in the previous year leads to a 14–20 percentage point higher likelihood of insurance purchase. We also find that the effect works jointly with the effect of previous village-level payouts. This is in line with Giné Townsend and Vickery (2007) who find that index insurance operates like a disaster insurance by primarily insuring farmers against extreme tail events of the rainfall distribution. The results on excess rain, corresponding to phase 3 of contracts that insure against heavy rainfall during the harvesting period, are not consistent with the pricing problem related to three hypothetical contracts considered by Michler, Viens and Shively (2015) who find that there is essentially no chance of any payout in their data for phase 3. They are also in contrast to a recent study in which 89 per cent of surveyed rural landowners in Andhra Pradesh cite drought as the most important risk faced by the household (Giné Townsend and Vickery 2008). What may explain our finding? One possible explanation could be the widespread use of irrigation in Gujrat state that can limit the threat to yields from rainfall deficiency.10 If areas are highly irrigated, we would expect farmers to react less to extreme weather events, particularly to rainfall deficiency while excess rainfall may still cause considerable damage, leaving farmers unaffected regarding their decision to purchase insurance against excess rainfall. To investigate the importance of irrigation, we combine our two main data sources with information about the percentage of the specific district area that is irrigated. We use the latest version of the Global Map of Irrigation Areas produced by the Land and Water Division of the Food and Agriculture Organization (FAO) and the Rheinische Friedrich-Wilhelms-Universitt in Bonn, Germany (Siebert et al., 2013). The irrigation information is at the district level as we believe it is more reliable than the village-level information, which contains many missing observations. The main disadvantage is that the information is constant over time by district, which leads us to estimate a pooled Ordinary Least Squares (OLS) excluding household fixed effects. On average, 84 per cent of the study area is irrigated. Estimation results are presented in Tables A3 and A4. Column 1 in Table A3 corresponds to column 2 in Table 2, while column 2 presents estimation results on the pooled sample, excluding household fixed effects. Column 3 controls for the share of the area irrigated by district. The coefficient estimate is negative and significant at the 10 per cent level, suggesting that insurance uptake decreases as the irrigation share increases. The result on irrigation, however, disappears when additional controls are included in columns 4 and 5. The pooled OLS results also show that farmers who experienced rainfall deficiency in the previous year are less likely to purchase insurance. Including interactions between the extreme event variables and the share of the area irrigated (column 3 in Table A4), we get that the interaction with excess rain is negative and statistically significant, while the interaction with rainfall deficiency is positive and statistically significant. Considering the joint effect for rainfall deficiency, we find that farmers in more irrigated areas, who experienced rainfall deficiency in the previous year, are less likely to purchase index insurance the following year. This implies that irrigation may be efficient in dealing with droughts. The joint effect for excess rain, on the other hand, suggests that farmers in more irrigated areas are more likely to purchase insurance, possibly because they could be wealthier and more likely to overweight the risk from excess rain. Taken together, these results confirm that the lack of evidence linked to rainfall deficiency is likely to be explained by widespread access to irrigation in Gujarat state. In our primary empirical analysis, we use a binary measure to indicate whether a household was hit by an event in the months covered by the insurance product offered. Another possibility is to consider the intensity of extreme weather events by including the actual z-scores for events identified as extreme (i.e. above 1 and below −1) in order to look at the intensive margin of insurance demand. Results using the intensity of weather events are reported in Table A6. Again, we find that excess rainfall in the previous year matters for insurance uptake, while lack of rainfall has no effect on farmers’ demand for insurance. The positive and significant coefficient on excess rain indicates that farmers are more likely to purchase insurance the more extreme the event experienced during the harvest. To further validate our results, we perform a placebo estimation by using future events to predict current uptake of index-based insurance. We construct the placebo-event variables based on the events taking place the following year. We create a binary variable for excess rain (rainfall deficiency), taking the value one if the household experienced excess rainfall (drought) in the next period, given that they did not experience an event in the present year, and zero otherwise. We expect to find no significant effect from the placebo-event variables if uptake is truly determined by previous events. The results are shown in Table A5. Across all estimations, we find no significant effect from future events on current insurance uptake. Hence, our main results remain: farmers in Gujarat state are more likely to purchase insurance if they experienced excess rain in the previous period. A potential threat to the definition of extreme weather events is measurement error, which may occur in the presence of misreporting or missing values and could lead to an incomplete rainfall picture for individual villages. This type of measurement error is relevant as insurance companies currently use rainfall station data to design policies and to calculate payouts and as our measure of extreme events may not include the accurate amount of rainfall actually experienced by farmers. If the measurement error is random, which is not unlikely, then the OLS underestimates the true effect (i.e. attenuation bias). Thus, the presence of a random measurement error in our extreme weather event variable means that the true coefficient is larger in absolute value. To ascertain how sensitive our findings are to the rainfall data used to define extreme weather events, we re-estimate equation (1) using a different source of data. We replace the historical GPCC data exclusively based on rainfall station measurement with monthly satellite-based Climate Hazards Group Infrared Precipitation (CHIRP) data to identify extreme events, while the z-score distribution is still based on the historical data. The CHIRP precipitation data incorporate 0.05° resolution satellite imagery to create 30+ year gridded rainfall time series, primarily used for trend analysis and seasonal drought monitoring (Funk et al., 2015). The advantage of GPCC data is that we can create the z-score distribution over a very long period. The literature suggests that satellite data could be a more precise indicator of rainfall, which in turn may help address the basis risk coverage (Hazell and Hess, 2010; Chantarat et al., 2013). In that case, we would expect to see a stronger correlation between revenue lost due to crop loss and recorded precipitation. The estimation results using CHIRP are shown in Tables A7 andA8. The coefficient estimates for the impact of excess rain on individual and village-level revenue lost due to crop loss, respectively, are slightly smaller than the reported estimates in Table 1. However, the estimates remain statistically significant at the 1 per cent level, and only excess rainfall is found to be an important determinant of farmers’ lost revenue due to crop loss (see Table A7). The results for insurance uptake are almost identical to the results reported in our baseline estimation presented in Table 2, suggesting that the importance of excess rainfall is not sensitive to the measure of rainfall used. 6. Conclusion It is widely acknowledged that weather insurance products could safeguard agricultural households against climate variability and the associated exposure to extreme weather events. Agricultural households, however, appear reluctant to invest in such products (Giné Townsend and Vickery, 2008; Cole et al., 2013). This article has investigated how extreme weather affects the take-up of weather index insurance among farmers in India. The analysis is based on an 8-year panel dataset of weather insurance purchase from a randomised control trial and historical rainfall data used to identify extreme weather events. We show that index insurance operates like a disaster insurance by protecting farmers against adverse effects of extreme tail events of the rainfall distribution. Specifically, the results show a consistently positive effect of experiencing excess rainfall on the insurance purchase decision, and that this effect increases the importance of other determinants of insurance demand, such as previous insurance payouts in the same location. This result holds for different definitions of extreme events, using more spatially precise satellite data and evaluating against placebo events. We do not find any significant relationship between dry weather spells and insurance purchase. The missing link between drought spells and insurance demand is explained by access to irrigation: farmers in more irrigated areas, who experienced rainfall deficiency in the previous year, are less likely to purchase index insurance the following year. This implies that irrigation may be efficient in dealing with droughts in Gujarat state. Overall, this result underscores the importance of taking the local context into consideration when developing policies to spark insurance demand while lowering the environmental covariate risk faced by farmers. In our study, we have focused on a single risk: the one from extreme weather events. Agricultural households face other risks related to, for example, input costs, commodity prices or pests, and they can employ other coping strategies, such as, distress sale of assets, migration or diverting investment from production to consumption. To the extent that these other risks are uncorrelated with precipitation, our results would not qualitatively change. We were not able to see from the dataset the exact type of policies households bought, that is, whether they bought phase 1, 2 or 3 policies or if farmers purchase other types of insurance. Access to this information would be beneficial for more precise estimation of the effects of extreme weather. The study could also be extended by estimating productivity and livelihood outcomes for households that have purchased index-based insurance. Acknowledgements The authors would like to thank two anonymous referees and the editor for their helpful comments, the participants of Development Economics Research Group (DERG) seminar and the participants of the Nordic Conference in Development Economics in Oslo. We are also grateful to Jeremy Tobacman for the help with household data, Sharissa Devina Funk and Tobias Harboe Haenschke for research assistance and Aleksandar Božinović for satellite data scraping code. Footnotes 1 Hill, Robles and Ceballos (2016) use historical weather data not to estimate insurance demand, but the actuarially fair price of insurance contracts. 2 The household data do not contain geo-referenced locations, so we searched location coordinates by village name. Four villages could not be geo-referenced possibly because of the use of local or alternative names, so we could not include them in the analysis. 3 Other methods of identifying extreme precipitation focus on a shorter time-frame. For example, Kunkel et al. (2013) define extreme precipitation events in the USA as those occurring once in 5 years and conclude that the number of such events has significantly increased in recent years. Hill et al. (2017) study the demand for weather index insurance in Bangladesh, where dry spells are identified from a 30-year average. 4 The frequency of extreme events is taken into consideration when deciding about payouts and premium rates. The index insurance in case was marketed as giving payouts once in 5 years, which could be considered very frequent for some insurance products, but probably not for index insurance, where several payouts can occur per season (Karlan et al., 2014; Cole et al., 2013; Cole, Stein and Tobacman, 2014a) and where premium rates may not differ for events with different probabilities of occurrence. For example, Hill et al. (2017) do not report difference in premium rates for insuring against a drought in Bangladesh that lasts at least 14 days (which would payout 600 Bangladesh Taka and which occurs roughly once every 10 years) and a drought lasting 12–13 days (which would pay half of the amount and occurs once every 5 years on average). 5 Compared to the average payout, this number at first seems high, however, the average revenue lost due to crop loss also includes not only losses encountered due to extreme weather but also other factors such as plant diseases, soil conditions, farmer illness and labour effort. 6 Control variables are also interacted with year 2010 when insurance policies were subsidised. 7 We also include 2 and 3 year lags of extreme weather variables as a robustness check. 8 Using quantity of output and a base year price that is kept constant over time and product could be an alternative approach to calculating crop loss, which we were unable to pursue due to the lack of per product price data over time. 9 We further test the importance of irrigation in Section 5. 10 Irrigation use in Gujarat is linked with delayed monsoon onset. Jain et al. (2015) describe that farmers in Gujarat adopt a variety of strategies to cope with delayed monsoon, such as increasing irrigation use, switching to more drought-tolerant crops, or delaying sowing. References Amrith , S. ( 2016 ). Risk and the South Asian monsoon . Climatic Change 1 – 12 . https://doi.org/10.1007/s10584-016-1629-x Barnett , B. J. , Barrett , C. B. and Skees , J. R. ( 2008 ). Poverty traps and index-based risk transfer products . World Development 36 ( 10 ): 1766 – 1785 . Google Scholar CrossRef Search ADS Baumann , D. D. and Sims , J. H. ( 1978 ). Flood insurance: some determinants of adoption . Economic Geography 54 ( 3 ): 189 – 196 . Google Scholar CrossRef Search ADS Becker , G. M. , de Groot , M. H. and Marschak , J. ( 1964 ). Measuring utility by a single-response sequential method . Behavioral Science 9 ( 3 ): 226 – 232 . Google Scholar CrossRef Search ADS Berhane , G. , Clarke , D. , Dercon , S. , Hill , R. and Taffesse , S. ( 2013 ). Insuring Against the Weather. ESSP Research Note 20, International Food Policy Research Institute (IFPRI). Binswanger-Mkhize , H. P. ( 2012 ). Is there too much hype about index-based agricultural insurance? Journal of Development Studies 48 ( 2 ): 187 – 200 . Google Scholar CrossRef Search ADS Browne , M. J. and Hoyt , R. E. ( 2000 ). The demand for flood insurance: empirical evidence . Journal of Risk and Uncertainty 20 ( 3 ): 291 – 306 . Google Scholar CrossRef Search ADS Cai , J. and Song , C. ( 2017 ). Do disaster experience and knowledge affect insurance take-up decisions? Journal of Development Economics 124 : 83 – 94 . Google Scholar CrossRef Search ADS Carter , M. R. , Cheng , L. and Sarris , A. ( 2016 ). Where and how index insurance can boost the adoption of improved agricultural technologies . Journal of Development Economics 118 : 59 – 71 . Google Scholar CrossRef Search ADS Challinor , A. J. , Ewert , F. , Arnold , S. , Simelton , E. and Fraser , E. ( 2009 ). Crops and climate change: progress, trends, and challenges in simulating impacts and informing adaptation . Journal of Experimental Botany 60 ( 10 ): 2775 – 2789 . Google Scholar CrossRef Search ADS Chantarat , S. , Mude , A. G. , Barrett , C. B. and Carter , M. R. ( 2013 ). Designing index-based livestock insurance for managing asset risk in Northern Kenya . Journal of Risk and Insurance 80 ( 1 ): 205 – 237 . Google Scholar CrossRef Search ADS Chantarat , S. , Mude , A. , Barrett , C. and Turvey , C. ( 2016 ). Welfare Impacts of Index Insurance in the Presence of a Poverty Trap. PIER Discussion Paper 24. Puey Ungphakorn Institute for Economic Research. Clarke , D. J. ( 2016 ). A theory of rational demand for index insurance . American Economic Journal: Microeconomics 8 ( 1 ): 283 – 306 . Google Scholar CrossRef Search ADS Clarke , D. , Hill , R. V. , de Nicola , F. , Kumar , N. and Mehta , P. ( 2015 ). A chat about insurance: experimental results from rural Bangladesh . Applied Economic Perspectives and Policy 37 ( 3 ): 477 – 501 . Google Scholar CrossRef Search ADS Coble , K. H. , Knight , T. O. , Pope , R. D. and Williams , J. R. ( 1996 ). Modeling farm-level crop insurance demand with panel data . American Journal of Agricultural Economics 78 ( 2 ): 439 – 447 . Google Scholar CrossRef Search ADS Cole , S. , Giné , X. , Tobacman , J. , Topalova , P. , Townsend , R. and Vickery , J. ( 2013 ). Barriers to household risk management: evidence from India . American Economic Journal: Applied Economics 5 ( 1 ): 104 – 135 . Google Scholar CrossRef Search ADS Cole , S. , Stein , D. and Tobacman , J. ( 2014 a). Dynamics of demand for index insurance: evidence from a long-run field experiment . American Economic Review 104 ( 5 ): 284 – 290 . Google Scholar CrossRef Search ADS Cole , S. , Xavier , G. and Vickery , J. ( 2014 b). How does risk management influence production decisions? Evidence from a field experiment. Harvard Business School Working Paper 13–80, 5 pp. Dercon , S. , Hill , R. V. , 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 . Google Scholar CrossRef Search ADS Deschênes , O. and Greenstone , M. ( 2007 ). The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather . American Economic Review 97 ( 1 ): 354 – 385 . Google Scholar CrossRef Search ADS Di Falco , S. , Adinolfi , F. , Bozzola , M. and Capitanio , F. ( 2014 ). Crop insurance as a strategy for adapting to climate change . Journal of Agricultural Economics 65 ( 2 ): 485 – 504 . Google Scholar CrossRef Search ADS Fisker , P. K. , Hansen , H. and Rand , J. ( 2015 ). Disaster financing in a developing country context. In: R. Dahlberg , O. Rubin and M. T. Vendeloe (eds) , Disaster Research: Multidisciplinary and International Perspectives . Abingdon : Routledge , 209 – 223 . Funk , C. , Peterson , P. , Landsfeld , M. , Pedreros , D. , Verdin , J. , Shukla , S. , Husak , G. , Rowland , J. , Harrison , L. , Hoell , A. and Michaelsen , J. ( 2015 ). The climate hazards infrared precipitation with stations–a new environmental record for monitoring extremes . Scientific Data 2 : 150066 . Google Scholar CrossRef Search ADS Gallagher , J. ( 2014 ). Learning about an infrequent event: evidence from flood insurance take-up in the United States . American Economic Journal: Applied Economics 6 ( 3 ): 206 – 233 . Google Scholar CrossRef Search ADS Giné , X. , Townsend , R. and Vickery , J. ( 2007 ). Statistical analysis of rainfall insurance payouts in southern India . American Journal of Agricultural Economics 89 ( 5 ): 1248 – 1254 . Google Scholar CrossRef Search ADS Giné , X. , Townsend , R. and Vickery , J. ( 2008 ). Patterns of rainfall insurance participation in rural India . World Bank Economic Review 22 ( 3 ): 539 – 566 . Google Scholar CrossRef Search ADS Haer , T. , Botzen , W. J. W. , de Moel , H. and Aerts , J. C. J. H. ( 2017 ). Integrating household risk mitigation behavior in flood risk analysis: an agent-based model approach . Risk Analysis 37 ( 10 ): 1977 – 1992 . Google Scholar CrossRef Search ADS Hazell , P. B. R. and Hess , U. ( 2010 ). Drought insurance for agricultural development and food security in dryland areas . Food Security 2 ( 4 ): 395 – 405 . Google Scholar CrossRef Search ADS Hill , R. V. , Kumar , N. , Magnan , N. , Makhija , S. , de Nicola , F. , Spielman , D. J. and Ward , P. S. ( 2017 ). Insuring against droughts: evidence on agricultural intensification and index insurance demand from a randomized evaluation in rural Bangladesh. Technical Report 1630, International Food Policy Research Institute (IFPRI). Hill , R. V. , Robles , M. and Ceballos , F. ( 2016 ). Demand for a simple weather insurance product in India: theory and evidence . American Journal of Agricultural Economics 98 ( 4 ): 1250 – 1270 . Google Scholar CrossRef Search ADS Hirway , I. , Kashyap , S. and Shah , A. ( 2002 ). Dynamics of Development in Gujarat . New Dehli : Concept Publishing Company . IPCC ( 2014 ). Part B: regional aspects. Contribution of Working Group II to the fifth assessment report of the intergovernmental panel on climate change. In: V. Barros , C. Field , D. Dokken , M. Mastrandrea , K. Mach , T. Bilir , M. Chatterjee , K. Ebi , Y. Estrada , R. Genova , B. Girma , E. Kissel , A. Levy , S. MacCracken , P. Mastrandrea and L. White (eds) , Climate Change 2014: Impacts, Adaptation and Vulnerability . Cambridge, UK and New York, NY, USA : Cambridge University Press . Jain , M. , Naeem , S. , Orlove , B. , Modi , V. and DeFries , R. S. ( 2015 ). Understanding the causes and consequences of differential decision-making in adaptation research: adapting to a delayed monsoon onset in Gujarat, India . Global Environmental Change 31 : 98 – 109 . Google Scholar CrossRef Search ADS Jensen , N. D. , Barrett , C. B. and Mude , A. G. ( 2016 ). Index insurance quality and basis risk: evidence from northern Kenya . American Journal of Agricultural Economics 98 ( 5 ): 1450 – 1469 . Google Scholar CrossRef Search ADS Kahneman , D. and Tversky , A. ( 1979 ). Prospect theory: an analysis of decision under risk . Econometrica: Journal of the Econometric Society 47 ( 2 ): 263 – 291 . Google Scholar CrossRef Search ADS Karlan , D. , Osei , R. , Osei-Akoto , I. and Udry , C. ( 2014 ). Agricultural decisions after relaxing credit and risk constraints . Quarterly Journal of Economics 129 ( 2 ): 597 – 652 . Google Scholar CrossRef Search ADS Khalil , A. F. , Kwon , H.-H. , Lall , U. , Miranda , M. J. and Skees , J. ( 2007 ). El Niño-southern oscillation-based index insurance for floods: statistical risk analyses and application to Peru . Water Resources Research 43 ( 10 ): W10416 . Google Scholar CrossRef Search ADS Kumar , K. K. , Rajagopalan , B. and Cane , M. A. ( 1999 ). On the weakening relationship between the Indian monsoon and ENSO . Science (New York, NY) 284 ( 5423 ): 2156 – 2159 . Google Scholar CrossRef Search ADS Kunkel , K. E. , Karl , T. R. , Brooks , H. , Kossin , J. , Lawrimore , J. H. , Arndt , D. , Bosart , L. , Changnon , D. , Cutter , S. L. , Doesken , N. , Emanuel , K. , Groisman , P. Y. , Katz , R. W. , Knutson , T. , O’Brien , J. , Paciorek , C. J. , Peterson , T. C. , Redmond , K. , Robinson , D. , Trapp , J. , Vose , R. , Weaver , S. , Wehner , M. , Wolter , K. and Wuebbles , D. ( 2013 ). Monitoring and understanding trends in extreme storms: state of knowledge . Bulletin of the American Meteorological Society 94 ( 4 ): 499 – 514 . Google Scholar CrossRef Search ADS Kunreuther , H. and Pauly , M. ( 2006 ). Rules rather than discretion: lessons from Hurricane Katrina . Journal of Risk and Uncertainty 33 ( 1–2 ): 101 – 116 . Google Scholar CrossRef Search ADS Kunreuther , H. , Sanderson , W. and Vetschera , R. ( 1985 ). A behavioral model of the adoption of protective activities . Journal of Economic Behavior & Organization 6 ( 1 ): 1 – 15 . Google Scholar CrossRef Search ADS McIntosh , C. , Sarris , A. and Papadopoulos , F. ( 2013 ). Productivity, credit, risk, and the demand for weather index insurance in smallholder agriculture in Ethiopia . Agricultural Economics 44 ( 4–5 ): 399 – 417 . Google Scholar CrossRef Search ADS Meyer , R. J. ( 2012 ). Failing to learn from experience about catastrophes: the case of hurricane preparedness . Journal of Risk and Uncertainty 45 ( 1 ): 25 – 50 . Google Scholar CrossRef Search ADS Michler , J. D. , Viens , F. G. and Shively , G. E. ( 2015 ). Risk, agricultural production, and weather index insurance in village South Asia. Selected Paper Prepared for Presentation at the 2015 Agricultural & Applied Economics Association and Western Agricultural Economics Association Annual Meeting, San Francisco, CA, 4. Ministry of Agriculture ( 2015 ). Agricultural Statistics at a Glance 2014 . Government of India, New Delhi : Oxford University Press . Mobarak , A. M. and Rosenzweig , M. R. ( 2012 ). Selling Formal Insurance to the Informally Insured. SSRN Electronic Journal, Economic Growth Center Discussion Paper No. 1007, 50. Mobarak , A. M. and Rosenzweig , M. R. ( 2013 ). Informal risk sharing, index insurance, and risk taking in developing countries . American Economic Review 103 ( 3 ): 375 – 380 . Google Scholar CrossRef Search ADS Naidu , C. V. , Satyanarayana , G. C. , Malleswara Rao , L. , Durgalakshmi , K. , Dharma Raju , A. , Vinay Kumar , P. and Jeevana Mounika , G. ( 2015 ). Anomalous behavior of Indian summer monsoon in the warming environment . Earth-Science Reviews 150 : 243 – 255 . Google Scholar CrossRef Search ADS Nanjundiah , R. S. , Francis , P. A. , Ved , M. and Gadgil , S. ( 2013 ). Predicting the extremes of Indian summer monsoon rainfall with coupled ocean-atmosphere models . Current Science 104 ( 10 ): 1380 – 1393 . Osberghaus , D. ( 2017 ). The effect of flood experience on household mitigation – evidence from longitudinal and insurance data . Global Environmental Change 43 ( Supplement C ): 126 – 136 . Google Scholar CrossRef Search ADS Rosenzweig , M. R. and Binswanger , H. P. ( 1993 ). Wealth, weather risk and the composition and profitability of agricultural investments . The Economic Journal 103 ( 416 ): 56 – 78 . Google Scholar CrossRef Search ADS Siebert , S. , Henrich , V. , Frenken , K. and Burke , J. ( 2013 ). Update of the Global Map of Irrigation Areas to Version 5 . Rome, Italy : Rheinische Friedrich-Wilhelms-University, Bonn, Germany/Food and Agriculture Organization of the United Nations . Siegrist , M. and Gutscher , H. ( 2008 ). Natural hazards and motivation for mitigation behavior: people cannot predict the affect evoked by a severe flood . Risk Analysis 28 ( 3 ): 771 – 778 . Google Scholar CrossRef Search ADS Townsend , R. M. ( 1994 ). Risk and insurance in village India . Econometrica: Journal of the Econometric Society 62 ( 3 ): 539 – 591 . Google Scholar CrossRef Search ADS Turvey , C. G. ( 2001 ). Weather derivatives for specific event risks in agriculture . Applied Economic Perspectives and Policy 23 ( 2 ): 333 – 351 . Tversky , A. and Kahneman , D. ( 1973 ). Availability: a heuristic for judging frequency and probability . Cognitive Psychology 5 ( 2 ): 207 – 232 . Google Scholar CrossRef Search ADS Tversky , A. and Kahneman , D. ( 1992 ). Advances in prospect theory: cumulative representation of uncertainty . Journal of Risk and Uncertainty 5 ( 4 ): 297 – 323 . Google Scholar CrossRef Search ADS Table A1. Summary statistics 2006 2007 2008 2009 2010 2011 2012 2013 Total SD Purchased insurance (%) 19.05 40.30 21.05 17.13 54.70 45.75 47.62 56.42 40.27 49.05 Repurchasers (%) 0.00 5.26 14.14 5.75 10.94 31.27 27.85 32.25 18.10 38.51 New purchasers (%) 0.00 17.60 6.91 7.96 43.76 14.48 19.78 24.17 18.80 39.07 Quitters (have not purchased again) (%) 0.00 6.58 26.15 8.40 6.19 23.43 17.90 15.07 13.77 34.46 Received payout (yes/no) (%) 0.00 0.00 6.25 6.96 33.15 7.07 35.03 – 12.61 33.20 Suffered crop loss (yes/no) (%) 77.51 22.04 30.43 49.72 29.72 23.20 27.96 – 28.92 45.34 Number of insurance policies bought 1.03 1.02 1.07 2.33 4.51 2.14 1.95 1.99 2.36 2.09 Price paid per policy 103 69 140 59 21 62 63 63 60 56 Number of households in village who received a payout (if village payout per policy≥Rs.1) 0 0 10 13 27 11 14 – 16 12 Mean village revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,400 259 1,917 473 1,359 – 1,274 3,206 Individual payout received (Rs., if payout≥Rs.1) 0 0 169 87 321 24 354 – 143 384 Individual payout per policy received (Rs., if payout≥Rs.1) 0 0 169 38 78 14 175 – 59 124 Revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,726 311 2,076 421 1,274 – 1,505 8,559 Insurance premium 215 68 190 152 75 195 200 200 161 54 Average price paid per policy/market price 0.49 1.00 0.74 0.37 0.28 0.32 0.31 0.32 0.41 0.35 Number of households 378 608 608 905 905 905 905 989 6,203 2006 2007 2008 2009 2010 2011 2012 2013 Total SD Purchased insurance (%) 19.05 40.30 21.05 17.13 54.70 45.75 47.62 56.42 40.27 49.05 Repurchasers (%) 0.00 5.26 14.14 5.75 10.94 31.27 27.85 32.25 18.10 38.51 New purchasers (%) 0.00 17.60 6.91 7.96 43.76 14.48 19.78 24.17 18.80 39.07 Quitters (have not purchased again) (%) 0.00 6.58 26.15 8.40 6.19 23.43 17.90 15.07 13.77 34.46 Received payout (yes/no) (%) 0.00 0.00 6.25 6.96 33.15 7.07 35.03 – 12.61 33.20 Suffered crop loss (yes/no) (%) 77.51 22.04 30.43 49.72 29.72 23.20 27.96 – 28.92 45.34 Number of insurance policies bought 1.03 1.02 1.07 2.33 4.51 2.14 1.95 1.99 2.36 2.09 Price paid per policy 103 69 140 59 21 62 63 63 60 56 Number of households in village who received a payout (if village payout per policy≥Rs.1) 0 0 10 13 27 11 14 – 16 12 Mean village revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,400 259 1,917 473 1,359 – 1,274 3,206 Individual payout received (Rs., if payout≥Rs.1) 0 0 169 87 321 24 354 – 143 384 Individual payout per policy received (Rs., if payout≥Rs.1) 0 0 169 38 78 14 175 – 59 124 Revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,726 311 2,076 421 1,274 – 1,505 8,559 Insurance premium 215 68 190 152 75 195 200 200 161 54 Average price paid per policy/market price 0.49 1.00 0.74 0.37 0.28 0.32 0.31 0.32 0.41 0.35 Number of households 378 608 608 905 905 905 905 989 6,203 View Large Table A1. Summary statistics 2006 2007 2008 2009 2010 2011 2012 2013 Total SD Purchased insurance (%) 19.05 40.30 21.05 17.13 54.70 45.75 47.62 56.42 40.27 49.05 Repurchasers (%) 0.00 5.26 14.14 5.75 10.94 31.27 27.85 32.25 18.10 38.51 New purchasers (%) 0.00 17.60 6.91 7.96 43.76 14.48 19.78 24.17 18.80 39.07 Quitters (have not purchased again) (%) 0.00 6.58 26.15 8.40 6.19 23.43 17.90 15.07 13.77 34.46 Received payout (yes/no) (%) 0.00 0.00 6.25 6.96 33.15 7.07 35.03 – 12.61 33.20 Suffered crop loss (yes/no) (%) 77.51 22.04 30.43 49.72 29.72 23.20 27.96 – 28.92 45.34 Number of insurance policies bought 1.03 1.02 1.07 2.33 4.51 2.14 1.95 1.99 2.36 2.09 Price paid per policy 103 69 140 59 21 62 63 63 60 56 Number of households in village who received a payout (if village payout per policy≥Rs.1) 0 0 10 13 27 11 14 – 16 12 Mean village revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,400 259 1,917 473 1,359 – 1,274 3,206 Individual payout received (Rs., if payout≥Rs.1) 0 0 169 87 321 24 354 – 143 384 Individual payout per policy received (Rs., if payout≥Rs.1) 0 0 169 38 78 14 175 – 59 124 Revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,726 311 2,076 421 1,274 – 1,505 8,559 Insurance premium 215 68 190 152 75 195 200 200 161 54 Average price paid per policy/market price 0.49 1.00 0.74 0.37 0.28 0.32 0.31 0.32 0.41 0.35 Number of households 378 608 608 905 905 905 905 989 6,203 2006 2007 2008 2009 2010 2011 2012 2013 Total SD Purchased insurance (%) 19.05 40.30 21.05 17.13 54.70 45.75 47.62 56.42 40.27 49.05 Repurchasers (%) 0.00 5.26 14.14 5.75 10.94 31.27 27.85 32.25 18.10 38.51 New purchasers (%) 0.00 17.60 6.91 7.96 43.76 14.48 19.78 24.17 18.80 39.07 Quitters (have not purchased again) (%) 0.00 6.58 26.15 8.40 6.19 23.43 17.90 15.07 13.77 34.46 Received payout (yes/no) (%) 0.00 0.00 6.25 6.96 33.15 7.07 35.03 – 12.61 33.20 Suffered crop loss (yes/no) (%) 77.51 22.04 30.43 49.72 29.72 23.20 27.96 – 28.92 45.34 Number of insurance policies bought 1.03 1.02 1.07 2.33 4.51 2.14 1.95 1.99 2.36 2.09 Price paid per policy 103 69 140 59 21 62 63 63 60 56 Number of households in village who received a payout (if village payout per policy≥Rs.1) 0 0 10 13 27 11 14 – 16 12 Mean village revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,400 259 1,917 473 1,359 – 1,274 3,206 Individual payout received (Rs., if payout≥Rs.1) 0 0 169 87 321 24 354 – 143 384 Individual payout per policy received (Rs., if payout≥Rs.1) 0 0 169 38 78 14 175 – 59 124 Revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,726 311 2,076 421 1,274 – 1,505 8,559 Insurance premium 215 68 190 152 75 195 200 200 161 54 Average price paid per policy/market price 0.49 1.00 0.74 0.37 0.28 0.32 0.31 0.32 0.41 0.35 Number of households 378 608 608 905 905 905 905 989 6,203 View Large Table A2. Summary statistics of marketing interventions Mean SD BDM offer/premium 24.23 24.92 Game for four policies 36.39 48.11 Risk exposure flyer 57.28 49.47 Assigned video testimonial 7.17 25.81 Assigned drought flyer 7.14 25.75 Assigned subsidies flyer 7.19 25.83 Assigned loan 7.01 25.54 Rebates: Buy one get 50% off 0.97 9.79 Rebates: Buy two get one free 0.84 9.12 Rebates: Buy three get one free 0.98 9.87 HYV flyer 2.29 14.96 Group treatment (flyer) 2.90 16.79 Muslim treatment (flyer) 2.11 14.38 Hindu treatment (flyer) 2.06 14.22 Strong SEWA brand 3.71 18.90 Peer endorsed 3.71 18.90 Information treatment: pays 2/10 years 1.95 13.83 Positive frame 1.79 13.26 Vulnerability frame 1.92 13.72 Negative flyer 0.74 8.58 Flyer positive language 0.53 7.27 Flyer negative language 0.84 9.12 Discount amount (%) 3.92 15.13 Discount squared 244.10 1,214.75 Observations 6,203 Mean SD BDM offer/premium 24.23 24.92 Game for four policies 36.39 48.11 Risk exposure flyer 57.28 49.47 Assigned video testimonial 7.17 25.81 Assigned drought flyer 7.14 25.75 Assigned subsidies flyer 7.19 25.83 Assigned loan 7.01 25.54 Rebates: Buy one get 50% off 0.97 9.79 Rebates: Buy two get one free 0.84 9.12 Rebates: Buy three get one free 0.98 9.87 HYV flyer 2.29 14.96 Group treatment (flyer) 2.90 16.79 Muslim treatment (flyer) 2.11 14.38 Hindu treatment (flyer) 2.06 14.22 Strong SEWA brand 3.71 18.90 Peer endorsed 3.71 18.90 Information treatment: pays 2/10 years 1.95 13.83 Positive frame 1.79 13.26 Vulnerability frame 1.92 13.72 Negative flyer 0.74 8.58 Flyer positive language 0.53 7.27 Flyer negative language 0.84 9.12 Discount amount (%) 3.92 15.13 Discount squared 244.10 1,214.75 Observations 6,203 Note: All values apart from discount amount and discount squared show the proportion of the sample that has been subject to specific market intervention. Discount amount is the average discount rate for the whole sample. View Large Table A2. Summary statistics of marketing interventions Mean SD BDM offer/premium 24.23 24.92 Game for four policies 36.39 48.11 Risk exposure flyer 57.28 49.47 Assigned video testimonial 7.17 25.81 Assigned drought flyer 7.14 25.75 Assigned subsidies flyer 7.19 25.83 Assigned loan 7.01 25.54 Rebates: Buy one get 50% off 0.97 9.79 Rebates: Buy two get one free 0.84 9.12 Rebates: Buy three get one free 0.98 9.87 HYV flyer 2.29 14.96 Group treatment (flyer) 2.90 16.79 Muslim treatment (flyer) 2.11 14.38 Hindu treatment (flyer) 2.06 14.22 Strong SEWA brand 3.71 18.90 Peer endorsed 3.71 18.90 Information treatment: pays 2/10 years 1.95 13.83 Positive frame 1.79 13.26 Vulnerability frame 1.92 13.72 Negative flyer 0.74 8.58 Flyer positive language 0.53 7.27 Flyer negative language 0.84 9.12 Discount amount (%) 3.92 15.13 Discount squared 244.10 1,214.75 Observations 6,203 Mean SD BDM offer/premium 24.23 24.92 Game for four policies 36.39 48.11 Risk exposure flyer 57.28 49.47 Assigned video testimonial 7.17 25.81 Assigned drought flyer 7.14 25.75 Assigned subsidies flyer 7.19 25.83 Assigned loan 7.01 25.54 Rebates: Buy one get 50% off 0.97 9.79 Rebates: Buy two get one free 0.84 9.12 Rebates: Buy three get one free 0.98 9.87 HYV flyer 2.29 14.96 Group treatment (flyer) 2.90 16.79 Muslim treatment (flyer) 2.11 14.38 Hindu treatment (flyer) 2.06 14.22 Strong SEWA brand 3.71 18.90 Peer endorsed 3.71 18.90 Information treatment: pays 2/10 years 1.95 13.83 Positive frame 1.79 13.26 Vulnerability frame 1.92 13.72 Negative flyer 0.74 8.58 Flyer positive language 0.53 7.27 Flyer negative language 0.84 9.12 Discount amount (%) 3.92 15.13 Discount squared 244.10 1,214.75 Observations 6,203 Note: All values apart from discount amount and discount squared show the proportion of the sample that has been subject to specific market intervention. Discount amount is the average discount rate for the whole sample. View Large Table A3. Is irrigation important for the decision to purchase insurance? (1) (2) (3) (4) (5) Excess rainfall in previous year 0.164 *** 0.151 *** 0.176 *** 0.157 *** 0.150 *** (0.053) (0.042) (0.039) (0.040) (0.042) Lack of rainfall in previous year −0.008 −0.101 ** −0.050 −0.071 * −0.101 ** (0.066) (0.050) (0.036) (0.037) (0.050) Percentage irrigated −0.149 * −0.015 0.004 (0.083) (0.083) (0.091) Village payout in previous year 0.576 *** 0.696 *** 0.792 *** 0.697 *** (0.199) (0.172) (0.152) (0.169) Individual payout in previous year −0.048 0.020 0.018 0.020 (0.049) (0.049) (0.049) (0.050) Number of insurance policies bought in previous year 0.005 0.012 0.012 0.012 (0.015) (0.013) (0.013) (0.013) Excess rainfall × village payout in previous year 0.638 * −0.061 −0.059 (0.334) (0.358) (0.365) Lack of rainfall × village payout in previous year −0.203 0.379 0.380 (0.363) (0.330) (0.331) Constant 0.409 *** 0.306 ** 0.780 *** 0.323 ** 0.303 * (0.129) (0.134) (0.104) (0.159) (0.161) Observations 1,940 1,940 1,940 1,940 1,940 Household FE Yes No No No No Year dummies Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes (1) (2) (3) (4) (5) Excess rainfall in previous year 0.164 *** 0.151 *** 0.176 *** 0.157 *** 0.150 *** (0.053) (0.042) (0.039) (0.040) (0.042) Lack of rainfall in previous year −0.008 −0.101 ** −0.050 −0.071 * −0.101 ** (0.066) (0.050) (0.036) (0.037) (0.050) Percentage irrigated −0.149 * −0.015 0.004 (0.083) (0.083) (0.091) Village payout in previous year 0.576 *** 0.696 *** 0.792 *** 0.697 *** (0.199) (0.172) (0.152) (0.169) Individual payout in previous year −0.048 0.020 0.018 0.020 (0.049) (0.049) (0.049) (0.050) Number of insurance policies bought in previous year 0.005 0.012 0.012 0.012 (0.015) (0.013) (0.013) (0.013) Excess rainfall × village payout in previous year 0.638 * −0.061 −0.059 (0.334) (0.358) (0.365) Lack of rainfall × village payout in previous year −0.203 0.379 0.380 (0.363) (0.330) (0.331) Constant 0.409 *** 0.306 ** 0.780 *** 0.323 ** 0.303 * (0.129) (0.134) (0.104) (0.159) (0.161) Observations 1,940 1,940 1,940 1,940 1,940 Household FE Yes No No No No Year dummies Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05. View Large Table A3. Is irrigation important for the decision to purchase insurance? (1) (2) (3) (4) (5) Excess rainfall in previous year 0.164 *** 0.151 *** 0.176 *** 0.157 *** 0.150 *** (0.053) (0.042) (0.039) (0.040) (0.042) Lack of rainfall in previous year −0.008 −0.101 ** −0.050 −0.071 * −0.101 ** (0.066) (0.050) (0.036) (0.037) (0.050) Percentage irrigated −0.149 * −0.015 0.004 (0.083) (0.083) (0.091) Village payout in previous year 0.576 *** 0.696 *** 0.792 *** 0.697 *** (0.199) (0.172) (0.152) (0.169) Individual payout in previous year −0.048 0.020 0.018 0.020 (0.049) (0.049) (0.049) (0.050) Number of insurance policies bought in previous year 0.005 0.012 0.012 0.012 (0.015) (0.013) (0.013) (0.013) Excess rainfall × village payout in previous year 0.638 * −0.061 −0.059 (0.334) (0.358) (0.365) Lack of rainfall × village payout in previous year −0.203 0.379 0.380 (0.363) (0.330) (0.331) Constant 0.409 *** 0.306 ** 0.780 *** 0.323 ** 0.303 * (0.129) (0.134) (0.104) (0.159) (0.161) Observations 1,940 1,940 1,940 1,940 1,940 Household FE Yes No No No No Year dummies Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes (1) (2) (3) (4) (5) Excess rainfall in previous year 0.164 *** 0.151 *** 0.176 *** 0.157 *** 0.150 *** (0.053) (0.042) (0.039) (0.040) (0.042) Lack of rainfall in previous year −0.008 −0.101 ** −0.050 −0.071 * −0.101 ** (0.066) (0.050) (0.036) (0.037) (0.050) Percentage irrigated −0.149 * −0.015 0.004 (0.083) (0.083) (0.091) Village payout in previous year 0.576 *** 0.696 *** 0.792 *** 0.697 *** (0.199) (0.172) (0.152) (0.169) Individual payout in previous year −0.048 0.020 0.018 0.020 (0.049) (0.049) (0.049) (0.050) Number of insurance policies bought in previous year 0.005 0.012 0.012 0.012 (0.015) (0.013) (0.013) (0.013) Excess rainfall × village payout in previous year 0.638 * −0.061 −0.059 (0.334) (0.358) (0.365) Lack of rainfall × village payout in previous year −0.203 0.379 0.380 (0.363) (0.330) (0.331) Constant 0.409 *** 0.306 ** 0.780 *** 0.323 ** 0.303 * (0.129) (0.134) (0.104) (0.159) (0.161) Observations 1,940 1,940 1,940 1,940 1,940 Household FE Yes No No No No Year dummies Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05. View Large Table A4. Does irrigation explain our baseline result? (1) (2) (3) (4) Excess rainfall in previous year 0.407 ** 0.467 *** 0.413 *** (0.171) (0.137) (0.132) Lack of rainfall in previous year 0.079 −0.352 *** −0.416 *** (0.186) (0.124) (0.129) Percentage irrigated −0.062 −0.097 −0.004 −0.010 (0.126) (0.114) (0.111) (0.065) Excess rainfall × percentage irrigated −0.294 −0.362 ** −0.303 ** (0.190) (0.157) (0.147) Lack of rainfall × percentage irrigated −0.095 0.307 ** 0.350 ** (0.208) (0.145) (0.140) Village payout in previous year 0.781 *** 0.692 *** (0.162) (0.144) Individual payout in previous year 0.016 0.020 (0.051) (0.040) Number of insurance policies bought in previous year 0.011 0.010 (0.013) (0.015) Excess rainfall × village payout in previous year −0.051 (0.300) Lack of rainfall × village payout in previous year 0.389 (0.287) Constant 0.701 *** 0.745 *** 0.744 *** 0.745 *** (0.139) (0.127) (0.147) (0.112) Observations 1,940 1,940 1,940 1,940 Household FE No No No No Year dummies Yes Yes Yes Yes Controls Yes Yes Yes Yes (1) (2) (3) (4) Excess rainfall in previous year 0.407 ** 0.467 *** 0.413 *** (0.171) (0.137) (0.132) Lack of rainfall in previous year 0.079 −0.352 *** −0.416 *** (0.186) (0.124) (0.129) Percentage irrigated −0.062 −0.097 −0.004 −0.010 (0.126) (0.114) (0.111) (0.065) Excess rainfall × percentage irrigated −0.294 −0.362 ** −0.303 ** (0.190) (0.157) (0.147) Lack of rainfall × percentage irrigated −0.095 0.307 ** 0.350 ** (0.208) (0.145) (0.140) Village payout in previous year 0.781 *** 0.692 *** (0.162) (0.144) Individual payout in previous year 0.016 0.020 (0.051) (0.040) Number of insurance policies bought in previous year 0.011 0.010 (0.013) (0.015) Excess rainfall × village payout in previous year −0.051 (0.300) Lack of rainfall × village payout in previous year 0.389 (0.287) Constant 0.701 *** 0.745 *** 0.744 *** 0.745 *** (0.139) (0.127) (0.147) (0.112) Observations 1,940 1,940 1,940 1,940 Household FE No No No No Year dummies Yes Yes Yes Yes Controls Yes Yes Yes Yes Note: OLS. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05. View Large Table A4. Does irrigation explain our baseline result? (1) (2) (3) (4) Excess rainfall in previous year 0.407 ** 0.467 *** 0.413 *** (0.171) (0.137) (0.132) Lack of rainfall in previous year 0.079 −0.352 *** −0.416 *** (0.186) (0.124) (0.129) Percentage irrigated −0.062 −0.097 −0.004 −0.010 (0.126) (0.114) (0.111) (0.065) Excess rainfall × percentage irrigated −0.294 −0.362 ** −0.303 ** (0.190) (0.157) (0.147) Lack of rainfall × percentage irrigated −0.095 0.307 ** 0.350 ** (0.208) (0.145) (0.140) Village payout in previous year 0.781 *** 0.692 *** (0.162) (0.144) Individual payout in previous year 0.016 0.020 (0.051) (0.040) Number of insurance policies bought in previous year 0.011 0.010 (0.013) (0.015) Excess rainfall × village payout in previous year −0.051 (0.300) Lack of rainfall × village payout in previous year 0.389 (0.287) Constant 0.701 *** 0.745 *** 0.744 *** 0.745 *** (0.139) (0.127) (0.147) (0.112) Observations 1,940 1,940 1,940 1,940 Household FE No No No No Year dummies Yes Yes Yes Yes Controls Yes Yes Yes Yes (1) (2) (3) (4) Excess rainfall in previous year 0.407 ** 0.467 *** 0.413 *** (0.171) (0.137) (0.132) Lack of rainfall in previous year 0.079 −0.352 *** −0.416 *** (0.186) (0.124) (0.129) Percentage irrigated −0.062 −0.097 −0.004 −0.010 (0.126) (0.114) (0.111) (0.065) Excess rainfall × percentage irrigated −0.294 −0.362 ** −0.303 ** (0.190) (0.157) (0.147) Lack of rainfall × percentage irrigated −0.095 0.307 ** 0.350 ** (0.208) (0.145) (0.140) Village payout in previous year 0.781 *** 0.692 *** (0.162) (0.144) Individual payout in previous year 0.016 0.020 (0.051) (0.040) Number of insurance policies bought in previous year 0.011 0.010 (0.013) (0.015) Excess rainfall × village payout in previous year −0.051 (0.300) Lack of rainfall × village payout in previous year 0.389 (0.287) Constant 0.701 *** 0.745 *** 0.744 *** 0.745 *** (0.139) (0.127) (0.147) (0.112) Observations 1,940 1,940 1,940 1,940 Household FE No No No No Year dummies Yes Yes Yes Yes Controls Yes Yes Yes Yes Note: OLS. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05. View Large Table A5. Placebo estimation: Do future events matter for take-up? Insurance purchasers (1) (2) (3) Excess rainfall in next year −0.050 −0.067 −0.069 (0.080) (0.078) (0.079) Lack of rainfall in next year 0.094 0.082 0.083 (0.080) (0.086) (0.086) Village payout in previous year 0.359 ** 0.468 ** (0.157) (0.196) Individual payout in previous year −0.050 (0.059) Number of insurance policies bought in previous year −0.012 (0.014) Constant 0.466 *** 0.441 * 0.449 ** (0.088) (0.221) (0.219) Observations 1,509 1,509 1,509 Number of households 749 749 749 Household FE Yes Yes Yes Year dummies No Yes Yes Controls Yes Yes Yes Insurance purchasers (1) (2) (3) Excess rainfall in next year −0.050 −0.067 −0.069 (0.080) (0.078) (0.079) Lack of rainfall in next year 0.094 0.082 0.083 (0.080) (0.086) (0.086) Village payout in previous year 0.359 ** 0.468 ** (0.157) (0.196) Individual payout in previous year −0.050 (0.059) Number of insurance policies bought in previous year −0.012 (0.014) Constant 0.466 *** 0.441 * 0.449 ** (0.088) (0.221) (0.219) Observations 1,509 1,509 1,509 Number of households 749 749 749 Household FE Yes Yes Yes Year dummies No Yes Yes Controls Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A5. Placebo estimation: Do future events matter for take-up? Insurance purchasers (1) (2) (3) Excess rainfall in next year −0.050 −0.067 −0.069 (0.080) (0.078) (0.079) Lack of rainfall in next year 0.094 0.082 0.083 (0.080) (0.086) (0.086) Village payout in previous year 0.359 ** 0.468 ** (0.157) (0.196) Individual payout in previous year −0.050 (0.059) Number of insurance policies bought in previous year −0.012 (0.014) Constant 0.466 *** 0.441 * 0.449 ** (0.088) (0.221) (0.219) Observations 1,509 1,509 1,509 Number of households 749 749 749 Household FE Yes Yes Yes Year dummies No Yes Yes Controls Yes Yes Yes Insurance purchasers (1) (2) (3) Excess rainfall in next year −0.050 −0.067 −0.069 (0.080) (0.078) (0.079) Lack of rainfall in next year 0.094 0.082 0.083 (0.080) (0.086) (0.086) Village payout in previous year 0.359 ** 0.468 ** (0.157) (0.196) Individual payout in previous year −0.050 (0.059) Number of insurance policies bought in previous year −0.012 (0.014) Constant 0.466 *** 0.441 * 0.449 ** (0.088) (0.221) (0.219) Observations 1,509 1,509 1,509 Number of households 749 749 749 Household FE Yes Yes Yes Year dummies No Yes Yes Controls Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A6. Intensity of events: What type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) Excess rainfall in previous year (z-score) 0.079 *** 0.098 *** 0.152 *** 0.151 *** (0.024) (0.028) (0.035) (0.035) Lack of rainfall in previous year (z-score) 0.002 −0.008 0.012 0.009 (0.043) (0.051) (0.055) (0.055) Village payout in previous year 0.569 *** 0.654 *** (0.149) (0.192) Individual payout in previous year −0.036 (0.049) Number of insurance policies bought in previous year 0.009 (0.015) Constant 0.538 *** 0.571 *** 0.404 *** 0.385 *** (0.014) (0.081) (0.129) (0.131) Observations 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Insurance purchasers (1) (2) (3) (4) Excess rainfall in previous year (z-score) 0.079 *** 0.098 *** 0.152 *** 0.151 *** (0.024) (0.028) (0.035) (0.035) Lack of rainfall in previous year (z-score) 0.002 −0.008 0.012 0.009 (0.043) (0.051) (0.055) (0.055) Village payout in previous year 0.569 *** 0.654 *** (0.149) (0.192) Individual payout in previous year −0.036 (0.049) Number of insurance policies bought in previous year 0.009 (0.015) Constant 0.538 *** 0.571 *** 0.404 *** 0.385 *** (0.014) (0.081) (0.129) (0.131) Observations 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01. View Large Table A6. Intensity of events: What type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) Excess rainfall in previous year (z-score) 0.079 *** 0.098 *** 0.152 *** 0.151 *** (0.024) (0.028) (0.035) (0.035) Lack of rainfall in previous year (z-score) 0.002 −0.008 0.012 0.009 (0.043) (0.051) (0.055) (0.055) Village payout in previous year 0.569 *** 0.654 *** (0.149) (0.192) Individual payout in previous year −0.036 (0.049) Number of insurance policies bought in previous year 0.009 (0.015) Constant 0.538 *** 0.571 *** 0.404 *** 0.385 *** (0.014) (0.081) (0.129) (0.131) Observations 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Insurance purchasers (1) (2) (3) (4) Excess rainfall in previous year (z-score) 0.079 *** 0.098 *** 0.152 *** 0.151 *** (0.024) (0.028) (0.035) (0.035) Lack of rainfall in previous year (z-score) 0.002 −0.008 0.012 0.009 (0.043) (0.051) (0.055) (0.055) Village payout in previous year 0.569 *** 0.654 *** (0.149) (0.192) Individual payout in previous year −0.036 (0.049) Number of insurance policies bought in previous year 0.009 (0.015) Constant 0.538 *** 0.571 *** 0.404 *** 0.385 *** (0.014) (0.081) (0.129) (0.131) Observations 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01. View Large Table A7. Robustness using CHIRP: extreme events and revenue lost due to crop loss Revenue lost due to crop loss (Rs ’0000s Mean village revenue lost due to crop loss (Rs ’0000s) (1) (2) (3) (4) Excess rainfall 0.060 *** 0.059 *** 0.084 *** 0.083 *** (0.019) (0.019) (0.024) (0.024) Lack of rainfall −0.003 0.001 −0.016 −0.015 (0.014) (0.015) (0.023) (0.023) Purchased insurance (dummy) −0.006 0.013 (0.008) (0.011) Village payout per policy (Rs. ’000s) −0.028 −0.030 (0.028) (0.052) Individual payout in previous year 0.004 0.012 (0.018) (0.024) Constant 0.128 *** 0.024 0.123 ** 0.063 * (0.017) (0.023) (0.050) (0.036) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Revenue lost due to crop loss (Rs ’0000s Mean village revenue lost due to crop loss (Rs ’0000s) (1) (2) (3) (4) Excess rainfall 0.060 *** 0.059 *** 0.084 *** 0.083 *** (0.019) (0.019) (0.024) (0.024) Lack of rainfall −0.003 0.001 −0.016 −0.015 (0.014) (0.015) (0.023) (0.023) Purchased insurance (dummy) −0.006 0.013 (0.008) (0.011) Village payout per policy (Rs. ’000s) −0.028 −0.030 (0.028) (0.052) Individual payout in previous year 0.004 0.012 (0.018) (0.024) Constant 0.128 *** 0.024 0.123 ** 0.063 * (0.017) (0.023) (0.050) (0.036) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Note: OLS. Extreme events are calculated using the CHIRP data rather than the GPCC data. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A7. Robustness using CHIRP: extreme events and revenue lost due to crop loss Revenue lost due to crop loss (Rs ’0000s Mean village revenue lost due to crop loss (Rs ’0000s) (1) (2) (3) (4) Excess rainfall 0.060 *** 0.059 *** 0.084 *** 0.083 *** (0.019) (0.019) (0.024) (0.024) Lack of rainfall −0.003 0.001 −0.016 −0.015 (0.014) (0.015) (0.023) (0.023) Purchased insurance (dummy) −0.006 0.013 (0.008) (0.011) Village payout per policy (Rs. ’000s) −0.028 −0.030 (0.028) (0.052) Individual payout in previous year 0.004 0.012 (0.018) (0.024) Constant 0.128 *** 0.024 0.123 ** 0.063 * (0.017) (0.023) (0.050) (0.036) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Revenue lost due to crop loss (Rs ’0000s Mean village revenue lost due to crop loss (Rs ’0000s) (1) (2) (3) (4) Excess rainfall 0.060 *** 0.059 *** 0.084 *** 0.083 *** (0.019) (0.019) (0.024) (0.024) Lack of rainfall −0.003 0.001 −0.016 −0.015 (0.014) (0.015) (0.023) (0.023) Purchased insurance (dummy) −0.006 0.013 (0.008) (0.011) Village payout per policy (Rs. ’000s) −0.028 −0.030 (0.028) (0.052) Individual payout in previous year 0.004 0.012 (0.018) (0.024) Constant 0.128 *** 0.024 0.123 ** 0.063 * (0.017) (0.023) (0.050) (0.036) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Note: OLS. Extreme events are calculated using the CHIRP data rather than the GPCC data. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A8. Robustness using CHIRP: what type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.128 *** 0.174 *** 0.172 *** 0.159 *** 0.164 *** 0.160 *** (0.033) (0.038) (0.044) (0.043) (0.043) (0.044) (0.043) Lack of rainfall in previous year −0.003 0.013 −0.007 −0.004 −0.016 −0.005 −0.017 (0.032) (0.039) (0.046) (0.046) (0.058) (0.053) (0.058) Village payout in previous year 0.571 *** 0.669 *** 0.548 *** 0.661 *** 0.527 ** (0.144) (0.189) (0.196) (0.189) (0.200) Individual payout in previous year −0.042 −0.043 −0.061 −0.032 (0.049) (0.049) (0.055) (0.056) Excess rainfall × village payout in previous year 0.327 ** 0.362 * (0.126) (0.186) Lack of rainfall × village payout in previous year 0.065 0.124 (0.144) (0.175) Excess rainfall × individual payout in previous year 0.064 −0.015 (0.051) (0.073) Lack of rainfall × individual payout in previous year 0.007 −0.032 (0.054) (0.062) Number of insurance policies bought in previous year 0.008 0.006 0.006 0.006 (0.015) (0.015) (0.015) (0.015) Constant 0.520 *** 0.552 *** 0.382 *** 0.364 *** 0.388 *** 0.370 *** 0.390 *** (0.017) (0.082) (0.130) (0.132) (0.129) (0.131) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.128 *** 0.174 *** 0.172 *** 0.159 *** 0.164 *** 0.160 *** (0.033) (0.038) (0.044) (0.043) (0.043) (0.044) (0.043) Lack of rainfall in previous year −0.003 0.013 −0.007 −0.004 −0.016 −0.005 −0.017 (0.032) (0.039) (0.046) (0.046) (0.058) (0.053) (0.058) Village payout in previous year 0.571 *** 0.669 *** 0.548 *** 0.661 *** 0.527 ** (0.144) (0.189) (0.196) (0.189) (0.200) Individual payout in previous year −0.042 −0.043 −0.061 −0.032 (0.049) (0.049) (0.055) (0.056) Excess rainfall × village payout in previous year 0.327 ** 0.362 * (0.126) (0.186) Lack of rainfall × village payout in previous year 0.065 0.124 (0.144) (0.175) Excess rainfall × individual payout in previous year 0.064 −0.015 (0.051) (0.073) Lack of rainfall × individual payout in previous year 0.007 −0.032 (0.054) (0.062) Number of insurance policies bought in previous year 0.008 0.006 0.006 0.006 (0.015) (0.015) (0.015) (0.015) Constant 0.520 *** 0.552 *** 0.382 *** 0.364 *** 0.388 *** 0.370 *** 0.390 *** (0.017) (0.082) (0.130) (0.132) (0.129) (0.131) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Note: OLS. Extreme events are calculated using the CHIRP data rather than the GPCC data. Standard errors clustered at the village level are in parentheses. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 % level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A8. Robustness using CHIRP: what type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.128 *** 0.174 *** 0.172 *** 0.159 *** 0.164 *** 0.160 *** (0.033) (0.038) (0.044) (0.043) (0.043) (0.044) (0.043) Lack of rainfall in previous year −0.003 0.013 −0.007 −0.004 −0.016 −0.005 −0.017 (0.032) (0.039) (0.046) (0.046) (0.058) (0.053) (0.058) Village payout in previous year 0.571 *** 0.669 *** 0.548 *** 0.661 *** 0.527 ** (0.144) (0.189) (0.196) (0.189) (0.200) Individual payout in previous year −0.042 −0.043 −0.061 −0.032 (0.049) (0.049) (0.055) (0.056) Excess rainfall × village payout in previous year 0.327 ** 0.362 * (0.126) (0.186) Lack of rainfall × village payout in previous year 0.065 0.124 (0.144) (0.175) Excess rainfall × individual payout in previous year 0.064 −0.015 (0.051) (0.073) Lack of rainfall × individual payout in previous year 0.007 −0.032 (0.054) (0.062) Number of insurance policies bought in previous year 0.008 0.006 0.006 0.006 (0.015) (0.015) (0.015) (0.015) Constant 0.520 *** 0.552 *** 0.382 *** 0.364 *** 0.388 *** 0.370 *** 0.390 *** (0.017) (0.082) (0.130) (0.132) (0.129) (0.131) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.128 *** 0.174 *** 0.172 *** 0.159 *** 0.164 *** 0.160 *** (0.033) (0.038) (0.044) (0.043) (0.043) (0.044) (0.043) Lack of rainfall in previous year −0.003 0.013 −0.007 −0.004 −0.016 −0.005 −0.017 (0.032) (0.039) (0.046) (0.046) (0.058) (0.053) (0.058) Village payout in previous year 0.571 *** 0.669 *** 0.548 *** 0.661 *** 0.527 ** (0.144) (0.189) (0.196) (0.189) (0.200) Individual payout in previous year −0.042 −0.043 −0.061 −0.032 (0.049) (0.049) (0.055) (0.056) Excess rainfall × village payout in previous year 0.327 ** 0.362 * (0.126) (0.186) Lack of rainfall × village payout in previous year 0.065 0.124 (0.144) (0.175) Excess rainfall × individual payout in previous year 0.064 −0.015 (0.051) (0.073) Lack of rainfall × individual payout in previous year 0.007 −0.032 (0.054) (0.062) Number of insurance policies bought in previous year 0.008 0.006 0.006 0.006 (0.015) (0.015) (0.015) (0.015) Constant 0.520 *** 0.552 *** 0.382 *** 0.364 *** 0.388 *** 0.370 *** 0.390 *** (0.017) (0.082) (0.130) (0.132) (0.129) (0.131) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Note: OLS. Extreme events are calculated using the CHIRP data rather than the GPCC data. Standard errors clustered at the village level are in parentheses. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 % level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Author notes Review coordinated by Iain Fraser © Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png European Review of Agricultural Economics Oxford University Press

Extreme weather and demand for index insurance in rural India

Loading next page...
 
/lp/ou_press/extreme-weather-and-demand-for-index-insurance-in-rural-india-bFvrepFkPH
Publisher
Oxford University Press
Copyright
© Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
0165-1587
eISSN
1464-3618
D.O.I.
10.1093/erae/jbx037
Publisher site
See Article on Publisher Site

Abstract

Abstract Index insurance appeared recently in developing countries with the expectation to improve agricultural output and living standards in general. We investigate how experiencing extreme weather events affects farmers’ decision to purchase index insurance in India. Extreme weather events are identified from historical precipitation data and matched with a randomised household panel. Excessive rainfall in previous years during the harvest increases the insurance demand, while lack of rainfall in the planting and growing periods has no effect. The latter can be explained by access to irrigation, underscoring the importance of the local context when developing insurance products to accommodate environmental risks. 1. Introduction The monsoon observed in the summer months in South Asia is a crucial component in agricultural production as it brings more than 70 per cent of total annual rainfall. The importance for countries such as India is very high as around 60 per cent of agriculture is rain-fed, and 60 per cent of the population works in agriculture (Amrith, 2016). While a plentiful and timely monsoon boosts agricultural output, erratic rainfall and dry spells during important periods of plant growth can cause serious damage to crops. Similarly, excess rain at harvest may damage or reduce crop quality. An often cited threat to successful agriculture and consequently rural livelihoods is climate change associated with warmer temperatures, changes in rainfall patterns and increased frequency of extreme weather events. For instance, it is estimated that the summer monsoon will increase, bringing more frequent and intense precipitation days over parts of South Asia, while parts of North and Southern Africa could become drier (IPCC, 2014). To manage covariate risk from weather variability, farmers can make use of weather index insurance that calculates payouts based on readings at local rainfall stations rather than actual crop losses (Fisker, Hansen and Rand, 2015). Advocates argue that index-based insurance can help protect productive assets, improve access to credit, new technologies and modern inputs, which together can improve farm productivity and livelihood outcomes (Barnett, Barrett and Skees, 2008; Hazell and Hess, 2010; Carter, Cheng and Sarris, 2016). The demand for such insurance products is, however, found to be uneven and scarce (Binswanger-Mkhize, 2012). This observation has since given rise to a number of studies examining the determinants of insurance take-up among farmers in developing countries (McIntosh, Sarris and Papadopoulos, 2013; Mobarak and Rosenzweig, 2013; Cole et al., 2013; Cole, Stein and Tobacman, 2014a; Dercon et al., 2014; Karlan et al., 2014; Clarke et al., 2015; Cai and Song, 2017) and the welfare impacts of weather insurance (Chantarat et al., 2016; Carter, Cheng and Sarris, 2016). Recent studies have shown that the insurance demand increases with the farmer’s own receipt of insurance payouts and the receipt of payouts by others in the community in the previous year (Karlan et al., 2014; Cole, Stein and Tobacman, 2014a). But are farmers’ decisions to purchase insurance sensitive to the type of weather shocks experienced? In this article, we study how past experience of extreme rainfall and droughts affects the take-up of insurance among farmers in India. This question is important for two reasons. First, covariate risk in agriculture is particularly difficult to ensure against, as it also affects neighbouring households and limits the effectiveness of traditional risk-sharing mechanisms. Index-based insurance provides an alternative, as it minimises moral hazard and adverse selection problems normally associated with standard insurance programmes. This assumption, however, hinges on the insurance policies offered being relevant and applicable to the local context in which the product is advertised and sold. Determining the role of weather is, therefore, vital in designing and implementing weather index insurance. Second, agents react to the occurrence of recent adverse events, which tends to generate demand for protective activities (Browne and Hoyt, 2000; Gallagher, 2014; Kunreuther, Sanderson and Vetschera, 1985). Climate change is likely to increase the occurrence of extreme weather events, requiring that better instruments to insure against adverse events become available to the most vulnerable. A recent study estimates that ‘a shift in the monsoon represents the gravest threat that climate change poses to lives and livelihoods in South Asia’ (Amrith, 2016). Using historical precipitation data, we identify the occurrence of extremely dry and wet periods in different times of crop growth. We match the extreme weather events using village-level coordinates with an 8-year household panel dataset from a randomised control trial on index-based insurance covering 56 villages located in the state of Gujarat, India. The identifying assumption is that, conditional on a village’s geography, it is random whether a village experiences an extreme weather event in a particular year. Because insurance policies offered to the household distinguish between rainfall deficiency and excess rainfall, we calculate monthly anomalies normalised by the standard deviation (i.e. z-scores) and identify wet and dry events for the growing and harvesting period, respectively. We first use the extreme weather event variables to investigate the link with individual and average village-level crop loss. We then analyse what type of events influence a farmer’s decision to purchase insurance, and to what extent this effect works jointly with the community effects found by Cole, Stein and Tobacman (2014a). Finally, we consider spill-over effects on non-purchasers from rainfall deficiency and excess rainfall. Our work departs from Cole, Stein and Tobacman (2014a), who use the same data and a similar identification strategy to study community-level effects, in that it considers the importance of different types of extreme weather events for insurance take-up. While previous studies consider rainfall levels (Giné Townsend and Vickery, 2007; Mobarak and Rosenzweig, 2013; Karlan et al., 2014; Di Falco et al., 2014; Coble et al., 1996; Deschênes and Greenstone, 2007), we calculate weather anomalies in relation to long-term village-level precipitation averages. This allows us to compensate for the effect that rainfall measures over a short period of time or a narrow area typically have larger variability.1 We find that the decision to purchase insurance increases after experiencing excess rain during the harvest and not as a consequence of rainfall deficiency in the crop planting and growing period. We also find that the combination of extreme rainfall and the observed village-level payouts increases dramatically the likelihood of insurance purchase. Experiencing excess rainfall in the previous year leads to a 14–20 percentage point higher likelihood of insurance purchase, while observing a payout of Rs. 1,000 in the village and extreme rainfall leads to a 57–67 percentage point higher likelihood of insurance purchase. The missing link between rainfall deficiency and insurance purchase can be explained by widespread access to irrigation in Gujarat state. These results are robust to different measures of extreme weather events, including placebo estimations, intensity of extreme weather events, and using satellite weather data to account for measurement error from misreporting or missing values at the rain gauges. Our findings suggest that weather shocks and insurance experience at the village-level boost index insurance demand. This is in line with the earlier evidence on agricultural insurance demand (Turvey, 2001) and Giné Townsend and Vickery (2007) who found that index insurance operates like a disaster insurance by primarily insuring farmers against extreme tail events of the rainfall distribution. Our findings are also affirmative of earlier evidence that economic agents tend to overreact to recent occurrence of extreme events (Kahneman and Tversky, 1979; Tversky and Kahneman, 1992). 2. Background 2.1. The demand for index-based insurance The recognition that households are much less successful in insuring against covariate weather risk than against idiosyncratic risk (Rosenzweig and Binswanger, 1993; Townsend, 1994) has motivated the design and implementation of weather-based index insurance for poor farmers. Advocates argue that index-based insurance is transparent, enables quick payouts and minimises moral hazard and adverse selection problems normally associated with standard insurance programmes and other risk-coping mechanisms. Moral hazard problems are irrelevant unless insurance purchasers can tamper with the measurement of rainfall at the gauge. Adverse selection problems are, in principle, bypassed by public availability of historical rainfall data, but they can occur if households purchase insurance only in years when risk is higher than normal (Khalil et al., 2007). In a setting without asymmetric information, a household’s willingness to pay for insurance is increasing in risk aversion, the expected insurance payout and the size of the insured risk. It is, however, decreasing in basis risk (i.e. the difference between the rainfall index and crop yields). Consistent with this standard model framework, Giné Townsend and Vickery (2008) find in their study of cross-sectional determinants of household insurance take-up that rainfall insurance purchase decreases with basis risk and increases with household wealth in Andhra Pradesh, India. Clarke et al. (2015) show that insurance demand decreases with price and basis risk: doubling a household’s distance to a reference weather station (a proxy for basis risk) decreases demand by 18 per cent. Highly risk-averse individuals are less likely to purchase insurance (Cole et al., 2013), particularly if they are unfamiliar with the insurance company (Giné Townsend and Vickery, 2008). It appears that such individuals give more weight to the costs of basis risk when the insurance contract does not perform than to the benefits from the increase in expected wealth when it does (Clarke, 2016). Jensen, Barrett and Mude (2016) show that with the possibility of basis risk, even at actuarially fair premium rates, risk-averse households may prefer no insurance over index insurance. Some of the characteristics found to limit insurance demand are financial illiteracy, lack of trust and understanding, availability of alternative risk-management tools and credit constraints. Lack of understanding of the insurance policy is an important barrier in the purchase decision as shown by Giné Townsend and Vickery (2008). Cole et al. (2013) test this in a randomised field experiment and find 36 per cent higher demand when the recommendation to the household comes from a trusted local agent. Consistent with this, demand is higher if households participate in village networks (Giné Townsend and Vickery, 2008), and if households already have previous experience in the insurance market, higher financial literacy and greater familiarity with probability concepts (Cole et al., 2013). Instead of purchasing insurance, farmers can diversify production, save, borrow or engage in a number of informal risk-sharing arrangements. Mobarak and Rosenzweig (2012) show that index insurance and risk-sharing are complements, especially for those for whom basis risk is higher. Dercon et al. (2014) demonstrate substantially higher insurance take-up rates among groups of individuals that can share risk. Previous experimental and non-experimental evidence also suggests that liquidity constraints reduce demand (Giné Townsend and Vickery, 2008; Cole et al., 2013), which is consistent with the predictions of the traditional model with borrowing constraints. High premiums relative to the expected payout decrease the rainfall insurance adoption (Giné Townsend and Vickery, 2007; Michler, Viens and Shively, 2015). Considering the size of the insurance premium relative to the expected payout, Giné Townsend and Vickery (2007) find that insurance premiums are on average around three times larger than expected payouts. Cole et al. (2013) investigate this at the household level by estimating the slope of the demand curve by randomly varying the price of insurance products offered to households. They find significant price sensitivity with a point estimate suggesting that insurance demand would increase 36–66 per cent if it could be priced at payout ratios similar to US retail insurance contracts. In line with this result, Michler, Viens and Shively (2015) conclude that the loading factor on these types of contracts is excessive, and that the pricing provides little incentive for smallholders to purchase insurance. However, even when farmers are given a very large price discount, few households are found to actually purchase insurance (Cole et al., 2013). Thus, lower prices alone are unlikely to be sufficient to trigger widespread index insurance adoption. A number of studies show that information generated by insurance payouts has community-wide effects on insurance demand (Browne and Hoyt, 2000; Cole et al., 2013; Cole, Stein and Tobacman, 2014a). Berhane et al. (2013) observe that ‘nothing sells insurance like insurance payouts’. Specifically, a payout of Rs. 1,000 (equivalent to 5 days of work) increases the probability of households purchasing insurance in the next year by 25–50 per cent (Cole, Stein and Tobacman, 2014a). This result holds both for insurance purchasers and non-purchasers (who did not receive a payout), and increases in the number of residents receiving the payout in the previous year. A significant fraction of households in the study by Giné Townsend and Vickery (2008) cite advice from other farmers as an important determinant of the participation decision. Past experience with disasters motivates insurance adoption (Baumann and Sims, 1978). A number of studies have shown that demand is increasing in flood damages of the prior year (Browne and Hoyt, 2000; Osberghaus, 2017; Siegrist and Gutscher, 2008; Kunreuther and Pauly, 2006). In fact, a large jump in insurance take-up occurs immediately after a flood, but the demand starts decreasing steadily after a while (Gallagher, 2014). One possible explanation for the importance of recency is availability bias, due to which the ease with which one can recall an event determines the subjective probability assigned to it (Tversky and Kahneman, 1973). A recent accident or disaster is likely to bear disproportionately more weight in decision-making than the one in the distant past and thus may generate demand for protective activities. It is also observed that people stop renewing insurance if they perceive it as an investment rather than a contingent claim, so if the payout is not collected for a prolonged period, the policy will not be renewed (Kunreuther, Sanderson and Vetschera, 1985). Another explanation is that individuals learn from experience in line with the Bayesian learning decision model where risk perceptions are updated on the basis of new information from experiencing adverse events. Risk perception will decline if no adverse events such as floods are experienced over a certain time period (Meyer, 2012; Haer et al., 2017; Kunreuther and Pauly, 2006). 2.2. Extreme weather and Gujarat state Gujarat state has a total land area of 196,024 km2 of which 4.29 million hectares are devoted to grain production, 3.08 million hectares to oil seeds and 2.69 million hectares to cotton (Ministry of Agriculture, 2015). These figures show that Gujarat state is an important agricultural region for India. Apart from cultivating grains, such as wheat, rice and maize, Gujarat is the leading producer of oil seeds in the country with 21 per cent of total oil seed production (Ministry of Agriculture, 2015). With around six million tons of groundnuts and 1.9 billion tons of cotton produced in 2014, Gujarat is ahead of all other Indian states in groundnut and cotton production. Figure 1 shows that both cotton production area and output have increased since 2007. Average output has grown by 32 per cent since 2007, when 1.4 million tons of cotton were produced. Average yields have grown by 19 per cent, from 581 kg/ha in 2007 to 692 kg/ha in 2013 (Ministry of Agriculture, 2015). This is observed against an increasing trend of non-operative cotton mills in the state. A positive trend in production is also observed for rice, wheat and local grain varieties, while maize and pulses have seen a decline in production since 2012. Fig. 1. View largeDownload slide Production area and output of most important types of crops in Gujarat. Note: Authors’ calculation based on Ministry of Agriculture (2015). Fig. 1. View largeDownload slide Production area and output of most important types of crops in Gujarat. Note: Authors’ calculation based on Ministry of Agriculture (2015). Agriculture in Gujarat is sustained by uneven rainfall. The monsoon season with high rainfall is called Kharif, and it usually lasts from June/July to September/October. Delay in the monsoon onset is a concern in Gujarat (Jain et al., 2015) and some districts, such as Saurashtra and Kutch, may not receive enough rain even during the Kharif (Hirway, Kashyap and Shah, 2002). It is estimated that only 3 per cent of the total area receives high rainfall of 1150 mm and above; 31 per cent receives medium rainfall with 750–1150 mm, and 66 per cent receives low rainfall of less than 750 mm (Hirway, Kashyap and Shah, 2002). That is why a substantial share of the production area in Gujarat is irrigated. For example, 57 per cent of cotton producing area is irrigated, and so is 46 per cent of land for grains, and 35 per cent of land for oil seed production (Ministry of Agriculture, 2015). The main Kharif crops are groundnuts and cotton, while wheat is considered as a winter crop grown from end-October to mid-December depending on the end of the monsoon period. Groundnuts are sown at the start of the rainy season, usually the third week of June, and cotton is planted around the same time, depending on the onset of the monsoon. Extreme departures from usual rainfall, such as large-scale droughts and floods, can negatively affect agricultural output and local livelihoods (Kumar, Rajagopalan and Cane, 1999). Erratic rainfall and dry spells during important periods of plant growth, such as immediately after planting and late plant growth, cause damage to the crop. Similarly, excess rain at harvest can reduce the quality of groundnuts if the crop is left to dry in the field. Wet weather also negatively impacts lint colour and cotton seed quality. Several projections on the effects of climate change warn about increased variability of rainfall in South Asia from year to year, potentially making droughts and floods more common (Challinor et al., 2009; Naidu et al., 2015). 3. Data and key measures To conduct the empirical analysis, we merge two primary data sources. The first source is a household-level panel that contains information on insurance purchasing decisions and household characteristics, including information about revenue and crop losses. The sample includes households from a random sample of 60 villages spread across three districts in Indian state Gujarat (the districts of Ahmadabad, Anand and Patan). The information on insurance purchasing decisions comes from a Gujarat-based NGO named Self-Employed Women’s Association (SEWA), which has carried out the index insurance project. They marketed rainfall insurance to a selected sample of residents from 2006 to 2013. The design of insurance policies varied by location and year. Product marketing included discounts, targeted messages and special offers on multiple purchases, which were randomly assigned each year. The data on households’ willingness to pay for insurance were obtained through the Becker–de Groot–Marschak (BDM) mechanism (Becker, de Groot and Marschak, 1964). The combination of random insurance policy offering and willingness to pay gives exogenous variation to insurance take-up and reveals households’ demand for weather insurance. Further details of the marketing interventions and experimental design can be found in Cole, Stein and Tobacman (2014a) and Cole, Xavier and Vickery (2014b). The rainfall index insurance policies provide indemnity payments according to values of an index (rainfall in this case) that serves as a proxy for losses instead of the estimated individual losses of different policyholders. The insured sum is based on production costs, determined on an agreed value basis, while the payouts follow a pre-established scale detailed in the insurance policy. The policies from our sample provided coverage against adverse rainfall events for the summer monsoon cropping season running from approximately June to September or October, depending on the village considered. The monsoon season is divided into three phases corresponding to the sowing, growing and harvest period. The duration of the first two phases is 35 days, while the last phase lasts for 40 days. The specified payout per hectare is calculated as cumulative rainfall during a fixed time period between the start and the end dates of the phases, measured at a nearby rain gauge. The start of the first phase is triggered by the monsoon rains. Specifically, the first phase starts when the accumulated rainfall since June 1 exceeds 50 mm, or on July 1, if accumulated rainfall is below 50 mm by the end of June. Insurance payouts in the first two phases are linked to low rainfall. The policy pays zero if accumulated rain during the first two phases exceeds an upper threshold or a ‘strike’. Otherwise, the policy pays Rs. 10 for each millimetre of rainfall deficiency relative to the strike until a lower threshold or ‘exit’ is reached. The policy pays a fixed amount if rainfall is below the exit value. Policies for the final, third phase is reverse but similar in structure to the first two phases, so that phase 3 insures against excess rain, which may cause damage to the crops during the harvest. The specific policy terms and payout differ by tehsil (sub-district level) and year. The payouts could occur several times per season and this is not different from index insurance products in other developing countries (e.g. Karlan et al., 2014). Every year, households must repurchase the insurance product to stay covered. Households were visited and offered the insurance each year in April or May, and were free to purchase multiple policies. In 2006, the insurance product was first introduced to households across 32 villages. The programme was extended to 20 additional villages in 2007. Within each village, 15 households were surveyed. Of these, five were randomly selected, five had previously purchased other forms of insurance and five were identified by local SEWA employees as likely to purchase insurance. Households likely to purchase insurance were purposefully over-sampled to ensure a substantial number of buyers. Finally, the programme expanded to 50 households in eight additional villages in 2009. The total sample that has been surveyed and assigned to receive insurance marketing by SEWA consists of 1,160 households in 60 villages. We use only the balanced panel of households who took part in the survey and marketing interventions each year after they have been added to the project. Four villages were dropped from the original sample as we were unable to map the location of the villages. The analysis is thus based on a sample of 905 households and 5,214 household-years for which we can observe current and lagged insurance coverage decisions. The second data source is historical gauge-based monthly precipitation data from the Global Precipitation Climate Centre (GPCC), which we merged with household data using village coordinates.2 The data on historical precipitation cover the period from 1901 to 2013. Since the survey villages might be situated on the border of the satellite precipitation grids, the rainfall information is reweighted using the four nearest precipitation grids. Particularly, precipitation for a given village at a given month is calculated as the weighted average of the measurement in the four cardinal points of the gridded cell to which the village belongs. To ensure that more weight is assigned to data points that are nearer the surveyed village, we use the inverse distance between the villages and precipitation observations. We calculate monthly anomalies from the historical monthly precipitation for the period from 1901 to 2005 by village and normalise by the standard deviation (i.e. z-scores). Based on the z-scores, we define extreme events for the summer monsoon months and, in line with insurance policies, consider droughts occurring in June, July and August, and excess rainfall in September and October. Extreme excess rainfall events are defined as the normalised precipitation anomaly larger than 1, while droughts are defined as the normalised precipitation less than −1 (Nanjundiah et al., 2013).3 Based on this normalisation, we find that farmers over the eight-year period, on average, experienced more extreme events during the harvest period compared to events occurring in the growing and planting season.4 For illustrative purposes, Figure 2 shows variation of extreme weather events in the crop-growing season aggregated across villages and time. We see that farmers, on average, experienced a range of extreme weather events between 2006 and 2013. The extreme weather measures constructed in this way are independent and different from rainfall levels used in designing insurance policies and determining payouts. In relation to the three types of insurance policies marketed by SEWA, dry events (marked in red) are relevant for phases 1 and 2, while the excess rainfall (marked in blue) is relevant for phase 3. The actual events used for the empirical analysis, however, depend on village location and time. This means that events are identified by village each year. Fig. 2. View largeDownload slide Distribution of extreme events across sampled villages in Gujarat. Fig. 2. View largeDownload slide Distribution of extreme events across sampled villages in Gujarat. Figure 3 shows village-specific experiences of dry and wet spells for each of the survey years. For instance, excess rainfall, usually occurring in September and October, was more frequent than rain deficiency in 2011, 2012 and 2013, while rain shortage was more common in 2009 and 2010. Moreover, Figure 3 shows that villages in the north experienced more dry spells in 2012 than in 2011. Fig. 3. View largeDownload slide Location of extreme weather by village and year. Fig. 3. View largeDownload slide Location of extreme weather by village and year. 3.1. Household-level descriptive statistics Around 40 per cent of households from the sample have purchased insurance at some point between 2006 and 2013 (Table A1 in the Appendix). The share of purchasers has increased from 19 to 56 per cent in that period. The average rate of repurchase was 18 per cent, growing from 5 per cent in 2007 to 32 per cent in 2013. New purchasers comprise 19 per cent of the sample, increasing from 18 per cent in 2007 to 24 per cent in 2013. Around 14 per cent of households who purchased insurance in one year decided not to repurchase in the following year. The variability of weather conditions in the observed period meant that crop loss and insurance payouts did not occur every year. Around 30 per cent of households have suffered crop loss at any point between 2006 and 2012, and around 13 per cent of households from the sample received a payout. Variation in both crop loss and payout is substantial. There were no payouts in 2006 and 2007, even though 319 and 146 households reported crop loss in those years. The proportion of payouts varied between 6 and 36 per cent, observed in 2008 and 2012, respectively. The proportion of households affected by crop loss ranged from 22 per cent in 2007 to 77 per cent in 2006. None of the households reported crop loss in 2013, and there were no payouts in that year. The insurance is designed so that a household can purchase as many policies per phase of the crop-growing cycle as wanted. Households from the sample have on average been buying two policies, starting with one in the 2006–2008 period and reaching five policies on average in 2010. Policies were subsidised in 2010 with a ‘buy one get one free’ offer, which has decreased their price and increased the number of policies purchased (Cole, Stein and Tobacman, 2014a). Hereafter, the average number of policies bought has decreased to two (2011–2013 period). The average price paid per policy was Rs. 60, while the average individual payout received was Rs. 143 (for all policies bought). On average, 16 households in a village received payouts. The average revenue lost due to crop loss was Rs. 1,274 at the village level, with standard deviation values around three times the size of the mean.5 Control variables used in all estimations are shown in Table A2, emphasising the proportion of households that randomly received different marketing offers.6 We see that 24 per cent of households participated in the BDM willingness to pay experiment, and 36 per cent of households participated in a game of bidding for four different insurance policies, which are described in Cole et al. (2013). Around 7 per cent of households received information about extreme weather on flyers, while 57 per cent received flyers about exposure risk. Discounts were offered to around 1 per cent of the sample. Households were also exposed to a series of video treatments informing about SEWA brand (3.7 per cent), peer experience (around 4 per cent), benefits (‘Peer endorsed’ and ‘Information treatment: Pays 2/10 years’, around 2 per cent) or potential difficulties if not purchasing insurance (‘Vulnerability frame,’ around 2 per cent). Around 1 per cent of the sample received flyers informing them about the risks or benefits of purchasing weather insurance. 4. Empirical analysis Due to the experimental nature of the data and the panel structure, we first estimate a linear probability model of the probability of household insurance purchase as a function of our extreme event variables and treatment variables. The estimation follows equation (1): buyijt=αi+βDjt−1+γFjt−1+ϕ1Xijt+ϕ2Xijt−1+τt+εijt (1) where buyijt is an indicator that takes value one if household i in village j has purchased insurance in year t, Djt−1 is an indicator for experiencing rainfall deficiency and Fjt−1 is an indicator for experiencing excess rainfall in village j in the year before insurance purchase.7 Xijt and Xijt−1 are same-year and previous year’s controls, respectively, for when the household entered the experiment, for exposure to insurance marketing and for individual and village-level insurance payouts. αi are household fixed effects, τt are year dummies and εijt are random disturbances. Standard errors are clustered at the village level throughout our analysis. Crop loss is calculated as the difference between agricultural output and the mean value of output in all prior years where crop loss was not reported.8 The set of control variables includes individual or village-level insurance payouts in previous period, which could be considered endogenous with respect to the decision to purchase insurance. We therefore also estimate a two-stage least squares (2SLS) model where recent insurance payouts are instrumented with variables indicating random assignment into receiving different marketing packages (promotional videos, flyers, etc.). In this setup, only variation in recent insurance payouts due to random assignment into different marketing package treatments is used to provide variation in actual insurance payouts in the insurance purchase equation. The treatments are valid instruments for recent insurance payouts under the assumption that they affect the outcome of interest only through changing purchase status, and not through any other channel. Due to the random assignment into receiving different marketing packages, instrumental variables (IVs) are likely to be uncorrelated with the error term, as required for 2SLS to be consistent. The analysis is split into two sections. In the next section, we consider the link between our measures of extreme weather events and revenue lost due to crop loss as well as the mean village revenue lost due to crop loss. In Section 4.2, we present the results on the type of weather events that matter for insurance take-up. The findings are further discussed in Section 5 following the empirical analysis. 4.1. Crop loss and extreme events Table 1 considers the relationship between crop loss and extreme weather events. Individual-level crop loss may be caused by a variety of factors. One such factor is weather, while other factors include pest and plant diseases, shirking and sickness. Household fixed effects make it possible to account for time-invariant household-level characteristics, such as ability. We therefore explain the part of the variation in revenue lost due to crop loss as a consequence of extreme weather events only. Table 1. Extreme events and revenue lost due to crop loss Revenue lost due to crop loss (Rs ’0000) Mean village revenue lost due to crop loss (Rs ’0000) (1) (2) (3) (4) Excess rainfall 0.076 *** 0.075 *** 0.092 *** 0.091 *** (0.023) (0.023) (0.031) (0.031) Lack of rainfall 0.009 0.013 −0.003 −0.001 (0.014) (0.016) (0.030) (0.030) Purchased insurance (dummy) −0.007 0.012 (0.008) (0.011) Village payout per policy (Rs ’000s) −0.032 −0.036 (0.030) (0.054) Individual payout (Rs ’000s) 0.005 0.013 (0.018) (0.024) Constant 0.135 *** 0.010 0.134 *** 0.056 (0.016) (0.027) (0.050) (0.041) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Revenue lost due to crop loss (Rs ’0000) Mean village revenue lost due to crop loss (Rs ’0000) (1) (2) (3) (4) Excess rainfall 0.076 *** 0.075 *** 0.092 *** 0.091 *** (0.023) (0.023) (0.031) (0.031) Lack of rainfall 0.009 0.013 −0.003 −0.001 (0.014) (0.016) (0.030) (0.030) Purchased insurance (dummy) −0.007 0.012 (0.008) (0.011) Village payout per policy (Rs ’000s) −0.032 −0.036 (0.030) (0.054) Individual payout (Rs ’000s) 0.005 0.013 (0.018) (0.024) Constant 0.135 *** 0.010 0.134 *** 0.056 (0.016) (0.027) (0.050) (0.041) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Note: OLS. Standard errors clustered at the village level. ***p<0.01. View Large Table 1. Extreme events and revenue lost due to crop loss Revenue lost due to crop loss (Rs ’0000) Mean village revenue lost due to crop loss (Rs ’0000) (1) (2) (3) (4) Excess rainfall 0.076 *** 0.075 *** 0.092 *** 0.091 *** (0.023) (0.023) (0.031) (0.031) Lack of rainfall 0.009 0.013 −0.003 −0.001 (0.014) (0.016) (0.030) (0.030) Purchased insurance (dummy) −0.007 0.012 (0.008) (0.011) Village payout per policy (Rs ’000s) −0.032 −0.036 (0.030) (0.054) Individual payout (Rs ’000s) 0.005 0.013 (0.018) (0.024) Constant 0.135 *** 0.010 0.134 *** 0.056 (0.016) (0.027) (0.050) (0.041) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Revenue lost due to crop loss (Rs ’0000) Mean village revenue lost due to crop loss (Rs ’0000) (1) (2) (3) (4) Excess rainfall 0.076 *** 0.075 *** 0.092 *** 0.091 *** (0.023) (0.023) (0.031) (0.031) Lack of rainfall 0.009 0.013 −0.003 −0.001 (0.014) (0.016) (0.030) (0.030) Purchased insurance (dummy) −0.007 0.012 (0.008) (0.011) Village payout per policy (Rs ’000s) −0.032 −0.036 (0.030) (0.054) Individual payout (Rs ’000s) 0.005 0.013 (0.018) (0.024) Constant 0.135 *** 0.010 0.134 *** 0.056 (0.016) (0.027) (0.050) (0.041) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Note: OLS. Standard errors clustered at the village level. ***p<0.01. View Large The dependent variable in columns 1 and 2 is individual revenue lost due to crop loss (in ’0000s Rs), while the dependent variable in columns 3 and 4 is the mean village revenue lost due to crop loss. We find a positive effect of excess rain (i.e. extreme event above 1 SD) on crop loss in column 1, while the lack of rainfall does not seem to be important in explaining revenue lost due to crop loss. The missing link between lack of rainfall and crop loss may be explained by the widespread use of irrigation in Gujarat state.9 Inclusion of control variables in column 2 affirms the results. The effect of excess rain is even stronger in column 3, where the dependent variable is mean village revenue lost, and the coefficient size does not change when additional control variables are included in column 4. The economically small effect from weather variability on households’ experienced crop loss is consistent with Michler, Viens and Shively (2015), who found that variance in weather accounts for an important but small fraction of total variance in crop output. The reasonably low correlation may also be explained by presence of basis risk in the identified weather events (Jensen, Barrett and Mude, 2016). 4.2. Insurance purchase In this section, we present the estimates of the insurance purchase decisions with lagged measures of extreme weather and other key determinants of insurance purchase decisions. Table 2 considers the sample of insurance purchasers, while Table 3 concentrates on potential spill-over effects to non-purchasers. Table 2. What type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.143 *** 0.197 *** 0.196 *** 0.164 *** 0.181 *** 0.165 *** (0.038) (0.044) (0.051) (0.050) (0.053) (0.054) (0.053) Lack of rainfall in previous year 0.001 0.010 −0.017 −0.013 −0.008 −0.008 −0.010 (0.044) (0.056) (0.061) (0.061) (0.066) (0.064) (0.066) Village payout in previous year 0.566 *** 0.670 *** 0.576 *** 0.670 *** 0.550 *** (0.146) (0.190) (0.199) (0.192) (0.198) Individual payout in previous year −0.045 −0.048 −0.062 −0.035 (0.048) (0.049) (0.056) (0.055) Excess rainfall × village payout in previous year 0.638 * 0.757 (0.334) (0.478) Lack of rainfall × village payout in previous year −0.203 −0.183 (0.363) (0.501) Excess rainfall × individual payout in previous year 0.103 −0.048 (0.074) (0.106) Lack of rainfall × individual payout in previous year −0.031 −0.017 (0.089) (0.114) Number of insurance policies bought in previous year 0.008 0.005 0.006 0.005 (0.015) (0.015) (0.015) (0.016) Constant 0.537 *** 0.572 *** 0.409 *** 0.391 *** 0.409 *** 0.393 *** 0.412 *** (0.014) (0.081) (0.128) (0.130) (0.129) (0.130) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.143 *** 0.197 *** 0.196 *** 0.164 *** 0.181 *** 0.165 *** (0.038) (0.044) (0.051) (0.050) (0.053) (0.054) (0.053) Lack of rainfall in previous year 0.001 0.010 −0.017 −0.013 −0.008 −0.008 −0.010 (0.044) (0.056) (0.061) (0.061) (0.066) (0.064) (0.066) Village payout in previous year 0.566 *** 0.670 *** 0.576 *** 0.670 *** 0.550 *** (0.146) (0.190) (0.199) (0.192) (0.198) Individual payout in previous year −0.045 −0.048 −0.062 −0.035 (0.048) (0.049) (0.056) (0.055) Excess rainfall × village payout in previous year 0.638 * 0.757 (0.334) (0.478) Lack of rainfall × village payout in previous year −0.203 −0.183 (0.363) (0.501) Excess rainfall × individual payout in previous year 0.103 −0.048 (0.074) (0.106) Lack of rainfall × individual payout in previous year −0.031 −0.017 (0.089) (0.114) Number of insurance policies bought in previous year 0.008 0.005 0.006 0.005 (0.015) (0.015) (0.015) (0.016) Constant 0.537 *** 0.572 *** 0.409 *** 0.391 *** 0.409 *** 0.393 *** 0.412 *** (0.014) (0.081) (0.128) (0.130) (0.129) (0.130) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01*p<0.1. View Large Table 2. What type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.143 *** 0.197 *** 0.196 *** 0.164 *** 0.181 *** 0.165 *** (0.038) (0.044) (0.051) (0.050) (0.053) (0.054) (0.053) Lack of rainfall in previous year 0.001 0.010 −0.017 −0.013 −0.008 −0.008 −0.010 (0.044) (0.056) (0.061) (0.061) (0.066) (0.064) (0.066) Village payout in previous year 0.566 *** 0.670 *** 0.576 *** 0.670 *** 0.550 *** (0.146) (0.190) (0.199) (0.192) (0.198) Individual payout in previous year −0.045 −0.048 −0.062 −0.035 (0.048) (0.049) (0.056) (0.055) Excess rainfall × village payout in previous year 0.638 * 0.757 (0.334) (0.478) Lack of rainfall × village payout in previous year −0.203 −0.183 (0.363) (0.501) Excess rainfall × individual payout in previous year 0.103 −0.048 (0.074) (0.106) Lack of rainfall × individual payout in previous year −0.031 −0.017 (0.089) (0.114) Number of insurance policies bought in previous year 0.008 0.005 0.006 0.005 (0.015) (0.015) (0.015) (0.016) Constant 0.537 *** 0.572 *** 0.409 *** 0.391 *** 0.409 *** 0.393 *** 0.412 *** (0.014) (0.081) (0.128) (0.130) (0.129) (0.130) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.143 *** 0.197 *** 0.196 *** 0.164 *** 0.181 *** 0.165 *** (0.038) (0.044) (0.051) (0.050) (0.053) (0.054) (0.053) Lack of rainfall in previous year 0.001 0.010 −0.017 −0.013 −0.008 −0.008 −0.010 (0.044) (0.056) (0.061) (0.061) (0.066) (0.064) (0.066) Village payout in previous year 0.566 *** 0.670 *** 0.576 *** 0.670 *** 0.550 *** (0.146) (0.190) (0.199) (0.192) (0.198) Individual payout in previous year −0.045 −0.048 −0.062 −0.035 (0.048) (0.049) (0.056) (0.055) Excess rainfall × village payout in previous year 0.638 * 0.757 (0.334) (0.478) Lack of rainfall × village payout in previous year −0.203 −0.183 (0.363) (0.501) Excess rainfall × individual payout in previous year 0.103 −0.048 (0.074) (0.106) Lack of rainfall × individual payout in previous year −0.031 −0.017 (0.089) (0.114) Number of insurance policies bought in previous year 0.008 0.005 0.006 0.005 (0.015) (0.015) (0.015) (0.016) Constant 0.537 *** 0.572 *** 0.409 *** 0.391 *** 0.409 *** 0.393 *** 0.412 *** (0.014) (0.081) (0.128) (0.130) (0.129) (0.130) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01*p<0.1. View Large Table 3. Spill-over effects from different events on non-purchasers Insurance non-purchasers (1) (2) (3) (4) Excess rainfall in previous year 0.125 *** 0.163 *** 0.173 ** 0.140 ** (0.046) (0.038) (0.077) (0.067) Lack of rainfall in previous year 0.162 ** 0.038 −0.033 0.008 (0.062) (0.053) (0.057) (0.057) Village payout in previous year 0.284 ** 0.304 *** (0.108) (0.107) Excess rainfall × village payout in previous year 0.638 * (0.324) Lack of rainfall × village payout in previous year −0.512 (0.319) Constant 0.238 *** 0.339 *** 0.547 *** 0.520 *** (0.016) (0.038) (0.163) (0.170) Observations 3,274 3,274 3,274 3,274 Number of households 893 893 893 893 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Insurance non-purchasers (1) (2) (3) (4) Excess rainfall in previous year 0.125 *** 0.163 *** 0.173 ** 0.140 ** (0.046) (0.038) (0.077) (0.067) Lack of rainfall in previous year 0.162 ** 0.038 −0.033 0.008 (0.062) (0.053) (0.057) (0.057) Village payout in previous year 0.284 ** 0.304 *** (0.108) (0.107) Excess rainfall × village payout in previous year 0.638 * (0.324) Lack of rainfall × village payout in previous year −0.512 (0.319) Constant 0.238 *** 0.339 *** 0.547 *** 0.520 *** (0.016) (0.038) (0.163) (0.170) Observations 3,274 3,274 3,274 3,274 Number of households 893 893 893 893 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Note: OLS. Standard errors clustered at the village level. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table 3. Spill-over effects from different events on non-purchasers Insurance non-purchasers (1) (2) (3) (4) Excess rainfall in previous year 0.125 *** 0.163 *** 0.173 ** 0.140 ** (0.046) (0.038) (0.077) (0.067) Lack of rainfall in previous year 0.162 ** 0.038 −0.033 0.008 (0.062) (0.053) (0.057) (0.057) Village payout in previous year 0.284 ** 0.304 *** (0.108) (0.107) Excess rainfall × village payout in previous year 0.638 * (0.324) Lack of rainfall × village payout in previous year −0.512 (0.319) Constant 0.238 *** 0.339 *** 0.547 *** 0.520 *** (0.016) (0.038) (0.163) (0.170) Observations 3,274 3,274 3,274 3,274 Number of households 893 893 893 893 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Insurance non-purchasers (1) (2) (3) (4) Excess rainfall in previous year 0.125 *** 0.163 *** 0.173 ** 0.140 ** (0.046) (0.038) (0.077) (0.067) Lack of rainfall in previous year 0.162 ** 0.038 −0.033 0.008 (0.062) (0.053) (0.057) (0.057) Village payout in previous year 0.284 ** 0.304 *** (0.108) (0.107) Excess rainfall × village payout in previous year 0.638 * (0.324) Lack of rainfall × village payout in previous year −0.512 (0.319) Constant 0.238 *** 0.339 *** 0.547 *** 0.520 *** (0.016) (0.038) (0.163) (0.170) Observations 3,274 3,274 3,274 3,274 Number of households 893 893 893 893 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Note: OLS. Standard errors clustered at the village level. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large In Table 2, the coefficient estimate on excess rain is both statistically and economically significant, implying that excess rainfall during the harvest period causes a 11–20 percentage point increase in the probability of purchasing insurance in the next season. This result is consistent across columns 1–7 and robust to inclusion of various control variables. This effect is large, considering that 41 per cent of the villages at some point during the 8-year survey period experienced extreme excess rain. Estimation in column 3 controls for the effect of village payout per policy, while estimations in columns 4–7 control for the number of insurance policies purchased in previous year. The effect of excess rain on insurance purchase is smaller than the effect from village payout per policy, which is estimated to be 57–67 percentage points, depending on specification. This is in line with Cole, Stein and Tobacman (2014a) who find that a payout per policy of Rs. 1,000 causes a 50 percentage point increase in the probability of insurance purchase in the next season. In column 5, we interact the extreme weather event variables with the village payout per policy to allow for a varying effect from village payout between households that have experienced extreme weather in the previous year. The interaction term with excess rainfall is statistically significant, and the point estimate is of the same sign as the non-interacted variables. In comparison, the interaction with drought spells is insignificant. The joint test of the sum of the interaction with excess rain and the mean village payout per policy is significant at the 5 per cent level. This indicates that excess rain in the previous year leads to a significant increase in insurance take-up, and that it works jointly with the effect of previous village-level payouts. In fact, the combined effect size implies that experiencing excess rainfall increases the likelihood of insurance purchase by 80 percentage points for every Rs. 1,000 of village payout per policy in previous year. In column 6, we interact the extreme weather event variables with individual payout per policy. The interaction terms of both extreme event variables are insignificant. The joint test of the sum of the interaction and the individual payout variable yields an insignificant estimate. This indicates that, while depending on the extreme rainfall, the decision to purchase insurance is not driven by individual payout experience. We combine columns 5 and 6 in column 7 and find that the effect on the interaction term between excess rain and village payout per policy remains statistically significant, and that the effect size increases. The interaction term is only significant at the 10 per cent level, while the sum of the coefficients on the interaction term and the main effect from mean village payout remains statistically significant at the 1 per cent level. Number of insurance policies bought in previous year is not statistically significant in any estimation. We now consider spill-over effects from experience of extreme events on insurance non-purchasers. As shown in Table 3, we find a positive and significant spill-over effect from excess rain in the previous year, but no spill-over effect from rain deficiency on insurance non-purchasers. Consistent with Cole, Stein and Tobacman (2014a), we also find that previous insurance payouts at the village level increase the likelihood of insurance purchase the following year among the non-purchasers. Again, the interaction effect is found to be positive and statistically significant, suggesting that experience from village payout due to excess rainfall encourages non-purchasers to buy insurance the following year. Finally, we use the experimental nature of the data to conduct the IV estimation on a full sample of households. We instrument endogenous variables (the lag of village payout per policy and the lag of the number of insurance policies purchased) with variables that measure randomly assigned marketing packages and their interactions with lagged insurance payouts. The marketing package variables enter the estimation in a lagged form. The results shown in Table 4 conform with our baseline results in Table 2, with a smaller effect size. Table 4. IV estimation of the effect of extreme weather events on the full sample of households. Full sample (1) (2) (3) (4) (5) (6) Excess rainfall (t−1) 0.142 *** 0.145 *** 0.149 *** 0.117 *** 0.164 *** 0.109 ** (0.035) (0.031) (0.033) (0.033) (0.058) (0.055) Lack of rainfall (t−1) 0.014 0.013 −0.011 −0.001 −0.056 −0.110 (0.045) (0.043) (0.046) (0.050) (0.057) (0.079) Village payout (t−1) 0.293 *** 0.266 *** 0.383 *** 0.509 *** (0.091) (0.093) (0.114) (0.148) Excess rainfall (t−1)× village payout (t−1) 0.536 *** −0.070 (0.198) (0.370) Lack of rainfall (t−1) × village payout (t−1) −0.206 0.396 (0.190) (0.391) Excess rainfall (t−2) 0.132 ** 0.093 (0.060) (0.059) Lack of rainfall (t−2) 0.010 −0.032 (0.062) (0.068) Village payout (t−2) 0.473 *** (0.133) Excess rainfall (t−2)× village payout (t−2) 2.292 ** (0.982) Lack of rainfall (t−2)× village payout (t−2) 0.025 (0.352) Excess rainfall (t−3) −0.017 −0.198 (0.123) (0.129) Lack of rainfall (t−3) 0.072 0.014 (0.072) (0.063) Village payout (t−3) 0.228 ** (0.102) Excess rainfall (t−3)× village payout (t−3) 5.437 *** (1.781) Lack of rainfall (t−3)× village payout (t−3) 0.290 (0.335) Individual payout (t−1) 0.279 *** 0.277 *** 0.137 * 0.105 0.076 0.052 (0.078) (0.076) (0.077) (0.079) (0.070) (0.069) Number of insurance policies bought (t−1) −0.025 ** −0.019 ** −0.002 −0.001 −0.009 −0.006 (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) Observations 5,214 5,214 5,214 5,214 3,404 3,404 Number of households 905 905 905 905 905 905 Household FE Yes Yes Yes Yes Yes Year dummies YES Yes Yes Yes Yes Yes Controls YES Yes Yes Yes Yes Yes Cragg–Donald F-stat 24.66 23.53 23.70 23.33 17.99 17.76 Full sample (1) (2) (3) (4) (5) (6) Excess rainfall (t−1) 0.142 *** 0.145 *** 0.149 *** 0.117 *** 0.164 *** 0.109 ** (0.035) (0.031) (0.033) (0.033) (0.058) (0.055) Lack of rainfall (t−1) 0.014 0.013 −0.011 −0.001 −0.056 −0.110 (0.045) (0.043) (0.046) (0.050) (0.057) (0.079) Village payout (t−1) 0.293 *** 0.266 *** 0.383 *** 0.509 *** (0.091) (0.093) (0.114) (0.148) Excess rainfall (t−1)× village payout (t−1) 0.536 *** −0.070 (0.198) (0.370) Lack of rainfall (t−1) × village payout (t−1) −0.206 0.396 (0.190) (0.391) Excess rainfall (t−2) 0.132 ** 0.093 (0.060) (0.059) Lack of rainfall (t−2) 0.010 −0.032 (0.062) (0.068) Village payout (t−2) 0.473 *** (0.133) Excess rainfall (t−2)× village payout (t−2) 2.292 ** (0.982) Lack of rainfall (t−2)× village payout (t−2) 0.025 (0.352) Excess rainfall (t−3) −0.017 −0.198 (0.123) (0.129) Lack of rainfall (t−3) 0.072 0.014 (0.072) (0.063) Village payout (t−3) 0.228 ** (0.102) Excess rainfall (t−3)× village payout (t−3) 5.437 *** (1.781) Lack of rainfall (t−3)× village payout (t−3) 0.290 (0.335) Individual payout (t−1) 0.279 *** 0.277 *** 0.137 * 0.105 0.076 0.052 (0.078) (0.076) (0.077) (0.079) (0.070) (0.069) Number of insurance policies bought (t−1) −0.025 ** −0.019 ** −0.002 −0.001 −0.009 −0.006 (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) Observations 5,214 5,214 5,214 5,214 3,404 3,404 Number of households 905 905 905 905 905 905 Household FE Yes Yes Yes Yes Yes Year dummies YES Yes Yes Yes Yes Yes Controls YES Yes Yes Yes Yes Yes Cragg–Donald F-stat 24.66 23.53 23.70 23.33 17.99 17.76 Note: IV. Standard errors clustered at the village level. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table 4. IV estimation of the effect of extreme weather events on the full sample of households. Full sample (1) (2) (3) (4) (5) (6) Excess rainfall (t−1) 0.142 *** 0.145 *** 0.149 *** 0.117 *** 0.164 *** 0.109 ** (0.035) (0.031) (0.033) (0.033) (0.058) (0.055) Lack of rainfall (t−1) 0.014 0.013 −0.011 −0.001 −0.056 −0.110 (0.045) (0.043) (0.046) (0.050) (0.057) (0.079) Village payout (t−1) 0.293 *** 0.266 *** 0.383 *** 0.509 *** (0.091) (0.093) (0.114) (0.148) Excess rainfall (t−1)× village payout (t−1) 0.536 *** −0.070 (0.198) (0.370) Lack of rainfall (t−1) × village payout (t−1) −0.206 0.396 (0.190) (0.391) Excess rainfall (t−2) 0.132 ** 0.093 (0.060) (0.059) Lack of rainfall (t−2) 0.010 −0.032 (0.062) (0.068) Village payout (t−2) 0.473 *** (0.133) Excess rainfall (t−2)× village payout (t−2) 2.292 ** (0.982) Lack of rainfall (t−2)× village payout (t−2) 0.025 (0.352) Excess rainfall (t−3) −0.017 −0.198 (0.123) (0.129) Lack of rainfall (t−3) 0.072 0.014 (0.072) (0.063) Village payout (t−3) 0.228 ** (0.102) Excess rainfall (t−3)× village payout (t−3) 5.437 *** (1.781) Lack of rainfall (t−3)× village payout (t−3) 0.290 (0.335) Individual payout (t−1) 0.279 *** 0.277 *** 0.137 * 0.105 0.076 0.052 (0.078) (0.076) (0.077) (0.079) (0.070) (0.069) Number of insurance policies bought (t−1) −0.025 ** −0.019 ** −0.002 −0.001 −0.009 −0.006 (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) Observations 5,214 5,214 5,214 5,214 3,404 3,404 Number of households 905 905 905 905 905 905 Household FE Yes Yes Yes Yes Yes Year dummies YES Yes Yes Yes Yes Yes Controls YES Yes Yes Yes Yes Yes Cragg–Donald F-stat 24.66 23.53 23.70 23.33 17.99 17.76 Full sample (1) (2) (3) (4) (5) (6) Excess rainfall (t−1) 0.142 *** 0.145 *** 0.149 *** 0.117 *** 0.164 *** 0.109 ** (0.035) (0.031) (0.033) (0.033) (0.058) (0.055) Lack of rainfall (t−1) 0.014 0.013 −0.011 −0.001 −0.056 −0.110 (0.045) (0.043) (0.046) (0.050) (0.057) (0.079) Village payout (t−1) 0.293 *** 0.266 *** 0.383 *** 0.509 *** (0.091) (0.093) (0.114) (0.148) Excess rainfall (t−1)× village payout (t−1) 0.536 *** −0.070 (0.198) (0.370) Lack of rainfall (t−1) × village payout (t−1) −0.206 0.396 (0.190) (0.391) Excess rainfall (t−2) 0.132 ** 0.093 (0.060) (0.059) Lack of rainfall (t−2) 0.010 −0.032 (0.062) (0.068) Village payout (t−2) 0.473 *** (0.133) Excess rainfall (t−2)× village payout (t−2) 2.292 ** (0.982) Lack of rainfall (t−2)× village payout (t−2) 0.025 (0.352) Excess rainfall (t−3) −0.017 −0.198 (0.123) (0.129) Lack of rainfall (t−3) 0.072 0.014 (0.072) (0.063) Village payout (t−3) 0.228 ** (0.102) Excess rainfall (t−3)× village payout (t−3) 5.437 *** (1.781) Lack of rainfall (t−3)× village payout (t−3) 0.290 (0.335) Individual payout (t−1) 0.279 *** 0.277 *** 0.137 * 0.105 0.076 0.052 (0.078) (0.076) (0.077) (0.079) (0.070) (0.069) Number of insurance policies bought (t−1) −0.025 ** −0.019 ** −0.002 −0.001 −0.009 −0.006 (0.009) (0.009) (0.010) (0.010) (0.010) (0.010) Observations 5,214 5,214 5,214 5,214 3,404 3,404 Number of households 905 905 905 905 905 905 Household FE Yes Yes Yes Yes Yes Year dummies YES Yes Yes Yes Yes Yes Controls YES Yes Yes Yes Yes Yes Cragg–Donald F-stat 24.66 23.53 23.70 23.33 17.99 17.76 Note: IV. Standard errors clustered at the village level. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large In columns 1–3, we show that excess rainfall increases the likelihood of insurance purchase by around 15 percentage points. In column 4, we find that excess rainfall increases the likelihood of insurance purchase by 54 percentage points for every Rs. 1,000 village payout. The decrease in the size of the effect compared to Table 2 is mainly driven by a decline in the importance of mean village payout per policy. We add the second and the third lag of our extreme event variables in column 5, as well as the interactions of lagged extreme weather events and lagged village payouts in column 6. The excess of rainfall 2 years back also matters for insurance uptake, but the size of the coefficient is smaller than the effect from excess rainfall in the previous year. We also find a persistence in the effect of payments made at the village level, but the size of coefficient estimates decreases over time. Finally, the second and the third lag of the interaction between the excess rain and the mean village payout correlate positively with the decision to purchase insurance, while the first lag loses significance. 5. Discussion We have shown that the likelihood of insurance purchase is driven by village payouts caused by wet spells in the previous year. Experiencing excess rainfall in the previous year leads to a 14–20 percentage point higher likelihood of insurance purchase. We also find that the effect works jointly with the effect of previous village-level payouts. This is in line with Giné Townsend and Vickery (2007) who find that index insurance operates like a disaster insurance by primarily insuring farmers against extreme tail events of the rainfall distribution. The results on excess rain, corresponding to phase 3 of contracts that insure against heavy rainfall during the harvesting period, are not consistent with the pricing problem related to three hypothetical contracts considered by Michler, Viens and Shively (2015) who find that there is essentially no chance of any payout in their data for phase 3. They are also in contrast to a recent study in which 89 per cent of surveyed rural landowners in Andhra Pradesh cite drought as the most important risk faced by the household (Giné Townsend and Vickery 2008). What may explain our finding? One possible explanation could be the widespread use of irrigation in Gujrat state that can limit the threat to yields from rainfall deficiency.10 If areas are highly irrigated, we would expect farmers to react less to extreme weather events, particularly to rainfall deficiency while excess rainfall may still cause considerable damage, leaving farmers unaffected regarding their decision to purchase insurance against excess rainfall. To investigate the importance of irrigation, we combine our two main data sources with information about the percentage of the specific district area that is irrigated. We use the latest version of the Global Map of Irrigation Areas produced by the Land and Water Division of the Food and Agriculture Organization (FAO) and the Rheinische Friedrich-Wilhelms-Universitt in Bonn, Germany (Siebert et al., 2013). The irrigation information is at the district level as we believe it is more reliable than the village-level information, which contains many missing observations. The main disadvantage is that the information is constant over time by district, which leads us to estimate a pooled Ordinary Least Squares (OLS) excluding household fixed effects. On average, 84 per cent of the study area is irrigated. Estimation results are presented in Tables A3 and A4. Column 1 in Table A3 corresponds to column 2 in Table 2, while column 2 presents estimation results on the pooled sample, excluding household fixed effects. Column 3 controls for the share of the area irrigated by district. The coefficient estimate is negative and significant at the 10 per cent level, suggesting that insurance uptake decreases as the irrigation share increases. The result on irrigation, however, disappears when additional controls are included in columns 4 and 5. The pooled OLS results also show that farmers who experienced rainfall deficiency in the previous year are less likely to purchase insurance. Including interactions between the extreme event variables and the share of the area irrigated (column 3 in Table A4), we get that the interaction with excess rain is negative and statistically significant, while the interaction with rainfall deficiency is positive and statistically significant. Considering the joint effect for rainfall deficiency, we find that farmers in more irrigated areas, who experienced rainfall deficiency in the previous year, are less likely to purchase index insurance the following year. This implies that irrigation may be efficient in dealing with droughts. The joint effect for excess rain, on the other hand, suggests that farmers in more irrigated areas are more likely to purchase insurance, possibly because they could be wealthier and more likely to overweight the risk from excess rain. Taken together, these results confirm that the lack of evidence linked to rainfall deficiency is likely to be explained by widespread access to irrigation in Gujarat state. In our primary empirical analysis, we use a binary measure to indicate whether a household was hit by an event in the months covered by the insurance product offered. Another possibility is to consider the intensity of extreme weather events by including the actual z-scores for events identified as extreme (i.e. above 1 and below −1) in order to look at the intensive margin of insurance demand. Results using the intensity of weather events are reported in Table A6. Again, we find that excess rainfall in the previous year matters for insurance uptake, while lack of rainfall has no effect on farmers’ demand for insurance. The positive and significant coefficient on excess rain indicates that farmers are more likely to purchase insurance the more extreme the event experienced during the harvest. To further validate our results, we perform a placebo estimation by using future events to predict current uptake of index-based insurance. We construct the placebo-event variables based on the events taking place the following year. We create a binary variable for excess rain (rainfall deficiency), taking the value one if the household experienced excess rainfall (drought) in the next period, given that they did not experience an event in the present year, and zero otherwise. We expect to find no significant effect from the placebo-event variables if uptake is truly determined by previous events. The results are shown in Table A5. Across all estimations, we find no significant effect from future events on current insurance uptake. Hence, our main results remain: farmers in Gujarat state are more likely to purchase insurance if they experienced excess rain in the previous period. A potential threat to the definition of extreme weather events is measurement error, which may occur in the presence of misreporting or missing values and could lead to an incomplete rainfall picture for individual villages. This type of measurement error is relevant as insurance companies currently use rainfall station data to design policies and to calculate payouts and as our measure of extreme events may not include the accurate amount of rainfall actually experienced by farmers. If the measurement error is random, which is not unlikely, then the OLS underestimates the true effect (i.e. attenuation bias). Thus, the presence of a random measurement error in our extreme weather event variable means that the true coefficient is larger in absolute value. To ascertain how sensitive our findings are to the rainfall data used to define extreme weather events, we re-estimate equation (1) using a different source of data. We replace the historical GPCC data exclusively based on rainfall station measurement with monthly satellite-based Climate Hazards Group Infrared Precipitation (CHIRP) data to identify extreme events, while the z-score distribution is still based on the historical data. The CHIRP precipitation data incorporate 0.05° resolution satellite imagery to create 30+ year gridded rainfall time series, primarily used for trend analysis and seasonal drought monitoring (Funk et al., 2015). The advantage of GPCC data is that we can create the z-score distribution over a very long period. The literature suggests that satellite data could be a more precise indicator of rainfall, which in turn may help address the basis risk coverage (Hazell and Hess, 2010; Chantarat et al., 2013). In that case, we would expect to see a stronger correlation between revenue lost due to crop loss and recorded precipitation. The estimation results using CHIRP are shown in Tables A7 andA8. The coefficient estimates for the impact of excess rain on individual and village-level revenue lost due to crop loss, respectively, are slightly smaller than the reported estimates in Table 1. However, the estimates remain statistically significant at the 1 per cent level, and only excess rainfall is found to be an important determinant of farmers’ lost revenue due to crop loss (see Table A7). The results for insurance uptake are almost identical to the results reported in our baseline estimation presented in Table 2, suggesting that the importance of excess rainfall is not sensitive to the measure of rainfall used. 6. Conclusion It is widely acknowledged that weather insurance products could safeguard agricultural households against climate variability and the associated exposure to extreme weather events. Agricultural households, however, appear reluctant to invest in such products (Giné Townsend and Vickery, 2008; Cole et al., 2013). This article has investigated how extreme weather affects the take-up of weather index insurance among farmers in India. The analysis is based on an 8-year panel dataset of weather insurance purchase from a randomised control trial and historical rainfall data used to identify extreme weather events. We show that index insurance operates like a disaster insurance by protecting farmers against adverse effects of extreme tail events of the rainfall distribution. Specifically, the results show a consistently positive effect of experiencing excess rainfall on the insurance purchase decision, and that this effect increases the importance of other determinants of insurance demand, such as previous insurance payouts in the same location. This result holds for different definitions of extreme events, using more spatially precise satellite data and evaluating against placebo events. We do not find any significant relationship between dry weather spells and insurance purchase. The missing link between drought spells and insurance demand is explained by access to irrigation: farmers in more irrigated areas, who experienced rainfall deficiency in the previous year, are less likely to purchase index insurance the following year. This implies that irrigation may be efficient in dealing with droughts in Gujarat state. Overall, this result underscores the importance of taking the local context into consideration when developing policies to spark insurance demand while lowering the environmental covariate risk faced by farmers. In our study, we have focused on a single risk: the one from extreme weather events. Agricultural households face other risks related to, for example, input costs, commodity prices or pests, and they can employ other coping strategies, such as, distress sale of assets, migration or diverting investment from production to consumption. To the extent that these other risks are uncorrelated with precipitation, our results would not qualitatively change. We were not able to see from the dataset the exact type of policies households bought, that is, whether they bought phase 1, 2 or 3 policies or if farmers purchase other types of insurance. Access to this information would be beneficial for more precise estimation of the effects of extreme weather. The study could also be extended by estimating productivity and livelihood outcomes for households that have purchased index-based insurance. Acknowledgements The authors would like to thank two anonymous referees and the editor for their helpful comments, the participants of Development Economics Research Group (DERG) seminar and the participants of the Nordic Conference in Development Economics in Oslo. We are also grateful to Jeremy Tobacman for the help with household data, Sharissa Devina Funk and Tobias Harboe Haenschke for research assistance and Aleksandar Božinović for satellite data scraping code. Footnotes 1 Hill, Robles and Ceballos (2016) use historical weather data not to estimate insurance demand, but the actuarially fair price of insurance contracts. 2 The household data do not contain geo-referenced locations, so we searched location coordinates by village name. Four villages could not be geo-referenced possibly because of the use of local or alternative names, so we could not include them in the analysis. 3 Other methods of identifying extreme precipitation focus on a shorter time-frame. For example, Kunkel et al. (2013) define extreme precipitation events in the USA as those occurring once in 5 years and conclude that the number of such events has significantly increased in recent years. Hill et al. (2017) study the demand for weather index insurance in Bangladesh, where dry spells are identified from a 30-year average. 4 The frequency of extreme events is taken into consideration when deciding about payouts and premium rates. The index insurance in case was marketed as giving payouts once in 5 years, which could be considered very frequent for some insurance products, but probably not for index insurance, where several payouts can occur per season (Karlan et al., 2014; Cole et al., 2013; Cole, Stein and Tobacman, 2014a) and where premium rates may not differ for events with different probabilities of occurrence. For example, Hill et al. (2017) do not report difference in premium rates for insuring against a drought in Bangladesh that lasts at least 14 days (which would payout 600 Bangladesh Taka and which occurs roughly once every 10 years) and a drought lasting 12–13 days (which would pay half of the amount and occurs once every 5 years on average). 5 Compared to the average payout, this number at first seems high, however, the average revenue lost due to crop loss also includes not only losses encountered due to extreme weather but also other factors such as plant diseases, soil conditions, farmer illness and labour effort. 6 Control variables are also interacted with year 2010 when insurance policies were subsidised. 7 We also include 2 and 3 year lags of extreme weather variables as a robustness check. 8 Using quantity of output and a base year price that is kept constant over time and product could be an alternative approach to calculating crop loss, which we were unable to pursue due to the lack of per product price data over time. 9 We further test the importance of irrigation in Section 5. 10 Irrigation use in Gujarat is linked with delayed monsoon onset. Jain et al. (2015) describe that farmers in Gujarat adopt a variety of strategies to cope with delayed monsoon, such as increasing irrigation use, switching to more drought-tolerant crops, or delaying sowing. References Amrith , S. ( 2016 ). Risk and the South Asian monsoon . Climatic Change 1 – 12 . https://doi.org/10.1007/s10584-016-1629-x Barnett , B. J. , Barrett , C. B. and Skees , J. R. ( 2008 ). Poverty traps and index-based risk transfer products . World Development 36 ( 10 ): 1766 – 1785 . Google Scholar CrossRef Search ADS Baumann , D. D. and Sims , J. H. ( 1978 ). Flood insurance: some determinants of adoption . Economic Geography 54 ( 3 ): 189 – 196 . Google Scholar CrossRef Search ADS Becker , G. M. , de Groot , M. H. and Marschak , J. ( 1964 ). Measuring utility by a single-response sequential method . Behavioral Science 9 ( 3 ): 226 – 232 . Google Scholar CrossRef Search ADS Berhane , G. , Clarke , D. , Dercon , S. , Hill , R. and Taffesse , S. ( 2013 ). Insuring Against the Weather. ESSP Research Note 20, International Food Policy Research Institute (IFPRI). Binswanger-Mkhize , H. P. ( 2012 ). Is there too much hype about index-based agricultural insurance? Journal of Development Studies 48 ( 2 ): 187 – 200 . Google Scholar CrossRef Search ADS Browne , M. J. and Hoyt , R. E. ( 2000 ). The demand for flood insurance: empirical evidence . Journal of Risk and Uncertainty 20 ( 3 ): 291 – 306 . Google Scholar CrossRef Search ADS Cai , J. and Song , C. ( 2017 ). Do disaster experience and knowledge affect insurance take-up decisions? Journal of Development Economics 124 : 83 – 94 . Google Scholar CrossRef Search ADS Carter , M. R. , Cheng , L. and Sarris , A. ( 2016 ). Where and how index insurance can boost the adoption of improved agricultural technologies . Journal of Development Economics 118 : 59 – 71 . Google Scholar CrossRef Search ADS Challinor , A. J. , Ewert , F. , Arnold , S. , Simelton , E. and Fraser , E. ( 2009 ). Crops and climate change: progress, trends, and challenges in simulating impacts and informing adaptation . Journal of Experimental Botany 60 ( 10 ): 2775 – 2789 . Google Scholar CrossRef Search ADS Chantarat , S. , Mude , A. G. , Barrett , C. B. and Carter , M. R. ( 2013 ). Designing index-based livestock insurance for managing asset risk in Northern Kenya . Journal of Risk and Insurance 80 ( 1 ): 205 – 237 . Google Scholar CrossRef Search ADS Chantarat , S. , Mude , A. , Barrett , C. and Turvey , C. ( 2016 ). Welfare Impacts of Index Insurance in the Presence of a Poverty Trap. PIER Discussion Paper 24. Puey Ungphakorn Institute for Economic Research. Clarke , D. J. ( 2016 ). A theory of rational demand for index insurance . American Economic Journal: Microeconomics 8 ( 1 ): 283 – 306 . Google Scholar CrossRef Search ADS Clarke , D. , Hill , R. V. , de Nicola , F. , Kumar , N. and Mehta , P. ( 2015 ). A chat about insurance: experimental results from rural Bangladesh . Applied Economic Perspectives and Policy 37 ( 3 ): 477 – 501 . Google Scholar CrossRef Search ADS Coble , K. H. , Knight , T. O. , Pope , R. D. and Williams , J. R. ( 1996 ). Modeling farm-level crop insurance demand with panel data . American Journal of Agricultural Economics 78 ( 2 ): 439 – 447 . Google Scholar CrossRef Search ADS Cole , S. , Giné , X. , Tobacman , J. , Topalova , P. , Townsend , R. and Vickery , J. ( 2013 ). Barriers to household risk management: evidence from India . American Economic Journal: Applied Economics 5 ( 1 ): 104 – 135 . Google Scholar CrossRef Search ADS Cole , S. , Stein , D. and Tobacman , J. ( 2014 a). Dynamics of demand for index insurance: evidence from a long-run field experiment . American Economic Review 104 ( 5 ): 284 – 290 . Google Scholar CrossRef Search ADS Cole , S. , Xavier , G. and Vickery , J. ( 2014 b). How does risk management influence production decisions? Evidence from a field experiment. Harvard Business School Working Paper 13–80, 5 pp. Dercon , S. , Hill , R. V. , 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 . Google Scholar CrossRef Search ADS Deschênes , O. and Greenstone , M. ( 2007 ). The economic impacts of climate change: evidence from agricultural output and random fluctuations in weather . American Economic Review 97 ( 1 ): 354 – 385 . Google Scholar CrossRef Search ADS Di Falco , S. , Adinolfi , F. , Bozzola , M. and Capitanio , F. ( 2014 ). Crop insurance as a strategy for adapting to climate change . Journal of Agricultural Economics 65 ( 2 ): 485 – 504 . Google Scholar CrossRef Search ADS Fisker , P. K. , Hansen , H. and Rand , J. ( 2015 ). Disaster financing in a developing country context. In: R. Dahlberg , O. Rubin and M. T. Vendeloe (eds) , Disaster Research: Multidisciplinary and International Perspectives . Abingdon : Routledge , 209 – 223 . Funk , C. , Peterson , P. , Landsfeld , M. , Pedreros , D. , Verdin , J. , Shukla , S. , Husak , G. , Rowland , J. , Harrison , L. , Hoell , A. and Michaelsen , J. ( 2015 ). The climate hazards infrared precipitation with stations–a new environmental record for monitoring extremes . Scientific Data 2 : 150066 . Google Scholar CrossRef Search ADS Gallagher , J. ( 2014 ). Learning about an infrequent event: evidence from flood insurance take-up in the United States . American Economic Journal: Applied Economics 6 ( 3 ): 206 – 233 . Google Scholar CrossRef Search ADS Giné , X. , Townsend , R. and Vickery , J. ( 2007 ). Statistical analysis of rainfall insurance payouts in southern India . American Journal of Agricultural Economics 89 ( 5 ): 1248 – 1254 . Google Scholar CrossRef Search ADS Giné , X. , Townsend , R. and Vickery , J. ( 2008 ). Patterns of rainfall insurance participation in rural India . World Bank Economic Review 22 ( 3 ): 539 – 566 . Google Scholar CrossRef Search ADS Haer , T. , Botzen , W. J. W. , de Moel , H. and Aerts , J. C. J. H. ( 2017 ). Integrating household risk mitigation behavior in flood risk analysis: an agent-based model approach . Risk Analysis 37 ( 10 ): 1977 – 1992 . Google Scholar CrossRef Search ADS Hazell , P. B. R. and Hess , U. ( 2010 ). Drought insurance for agricultural development and food security in dryland areas . Food Security 2 ( 4 ): 395 – 405 . Google Scholar CrossRef Search ADS Hill , R. V. , Kumar , N. , Magnan , N. , Makhija , S. , de Nicola , F. , Spielman , D. J. and Ward , P. S. ( 2017 ). Insuring against droughts: evidence on agricultural intensification and index insurance demand from a randomized evaluation in rural Bangladesh. Technical Report 1630, International Food Policy Research Institute (IFPRI). Hill , R. V. , Robles , M. and Ceballos , F. ( 2016 ). Demand for a simple weather insurance product in India: theory and evidence . American Journal of Agricultural Economics 98 ( 4 ): 1250 – 1270 . Google Scholar CrossRef Search ADS Hirway , I. , Kashyap , S. and Shah , A. ( 2002 ). Dynamics of Development in Gujarat . New Dehli : Concept Publishing Company . IPCC ( 2014 ). Part B: regional aspects. Contribution of Working Group II to the fifth assessment report of the intergovernmental panel on climate change. In: V. Barros , C. Field , D. Dokken , M. Mastrandrea , K. Mach , T. Bilir , M. Chatterjee , K. Ebi , Y. Estrada , R. Genova , B. Girma , E. Kissel , A. Levy , S. MacCracken , P. Mastrandrea and L. White (eds) , Climate Change 2014: Impacts, Adaptation and Vulnerability . Cambridge, UK and New York, NY, USA : Cambridge University Press . Jain , M. , Naeem , S. , Orlove , B. , Modi , V. and DeFries , R. S. ( 2015 ). Understanding the causes and consequences of differential decision-making in adaptation research: adapting to a delayed monsoon onset in Gujarat, India . Global Environmental Change 31 : 98 – 109 . Google Scholar CrossRef Search ADS Jensen , N. D. , Barrett , C. B. and Mude , A. G. ( 2016 ). Index insurance quality and basis risk: evidence from northern Kenya . American Journal of Agricultural Economics 98 ( 5 ): 1450 – 1469 . Google Scholar CrossRef Search ADS Kahneman , D. and Tversky , A. ( 1979 ). Prospect theory: an analysis of decision under risk . Econometrica: Journal of the Econometric Society 47 ( 2 ): 263 – 291 . Google Scholar CrossRef Search ADS Karlan , D. , Osei , R. , Osei-Akoto , I. and Udry , C. ( 2014 ). Agricultural decisions after relaxing credit and risk constraints . Quarterly Journal of Economics 129 ( 2 ): 597 – 652 . Google Scholar CrossRef Search ADS Khalil , A. F. , Kwon , H.-H. , Lall , U. , Miranda , M. J. and Skees , J. ( 2007 ). El Niño-southern oscillation-based index insurance for floods: statistical risk analyses and application to Peru . Water Resources Research 43 ( 10 ): W10416 . Google Scholar CrossRef Search ADS Kumar , K. K. , Rajagopalan , B. and Cane , M. A. ( 1999 ). On the weakening relationship between the Indian monsoon and ENSO . Science (New York, NY) 284 ( 5423 ): 2156 – 2159 . Google Scholar CrossRef Search ADS Kunkel , K. E. , Karl , T. R. , Brooks , H. , Kossin , J. , Lawrimore , J. H. , Arndt , D. , Bosart , L. , Changnon , D. , Cutter , S. L. , Doesken , N. , Emanuel , K. , Groisman , P. Y. , Katz , R. W. , Knutson , T. , O’Brien , J. , Paciorek , C. J. , Peterson , T. C. , Redmond , K. , Robinson , D. , Trapp , J. , Vose , R. , Weaver , S. , Wehner , M. , Wolter , K. and Wuebbles , D. ( 2013 ). Monitoring and understanding trends in extreme storms: state of knowledge . Bulletin of the American Meteorological Society 94 ( 4 ): 499 – 514 . Google Scholar CrossRef Search ADS Kunreuther , H. and Pauly , M. ( 2006 ). Rules rather than discretion: lessons from Hurricane Katrina . Journal of Risk and Uncertainty 33 ( 1–2 ): 101 – 116 . Google Scholar CrossRef Search ADS Kunreuther , H. , Sanderson , W. and Vetschera , R. ( 1985 ). A behavioral model of the adoption of protective activities . Journal of Economic Behavior & Organization 6 ( 1 ): 1 – 15 . Google Scholar CrossRef Search ADS McIntosh , C. , Sarris , A. and Papadopoulos , F. ( 2013 ). Productivity, credit, risk, and the demand for weather index insurance in smallholder agriculture in Ethiopia . Agricultural Economics 44 ( 4–5 ): 399 – 417 . Google Scholar CrossRef Search ADS Meyer , R. J. ( 2012 ). Failing to learn from experience about catastrophes: the case of hurricane preparedness . Journal of Risk and Uncertainty 45 ( 1 ): 25 – 50 . Google Scholar CrossRef Search ADS Michler , J. D. , Viens , F. G. and Shively , G. E. ( 2015 ). Risk, agricultural production, and weather index insurance in village South Asia. Selected Paper Prepared for Presentation at the 2015 Agricultural & Applied Economics Association and Western Agricultural Economics Association Annual Meeting, San Francisco, CA, 4. Ministry of Agriculture ( 2015 ). Agricultural Statistics at a Glance 2014 . Government of India, New Delhi : Oxford University Press . Mobarak , A. M. and Rosenzweig , M. R. ( 2012 ). Selling Formal Insurance to the Informally Insured. SSRN Electronic Journal, Economic Growth Center Discussion Paper No. 1007, 50. Mobarak , A. M. and Rosenzweig , M. R. ( 2013 ). Informal risk sharing, index insurance, and risk taking in developing countries . American Economic Review 103 ( 3 ): 375 – 380 . Google Scholar CrossRef Search ADS Naidu , C. V. , Satyanarayana , G. C. , Malleswara Rao , L. , Durgalakshmi , K. , Dharma Raju , A. , Vinay Kumar , P. and Jeevana Mounika , G. ( 2015 ). Anomalous behavior of Indian summer monsoon in the warming environment . Earth-Science Reviews 150 : 243 – 255 . Google Scholar CrossRef Search ADS Nanjundiah , R. S. , Francis , P. A. , Ved , M. and Gadgil , S. ( 2013 ). Predicting the extremes of Indian summer monsoon rainfall with coupled ocean-atmosphere models . Current Science 104 ( 10 ): 1380 – 1393 . Osberghaus , D. ( 2017 ). The effect of flood experience on household mitigation – evidence from longitudinal and insurance data . Global Environmental Change 43 ( Supplement C ): 126 – 136 . Google Scholar CrossRef Search ADS Rosenzweig , M. R. and Binswanger , H. P. ( 1993 ). Wealth, weather risk and the composition and profitability of agricultural investments . The Economic Journal 103 ( 416 ): 56 – 78 . Google Scholar CrossRef Search ADS Siebert , S. , Henrich , V. , Frenken , K. and Burke , J. ( 2013 ). Update of the Global Map of Irrigation Areas to Version 5 . Rome, Italy : Rheinische Friedrich-Wilhelms-University, Bonn, Germany/Food and Agriculture Organization of the United Nations . Siegrist , M. and Gutscher , H. ( 2008 ). Natural hazards and motivation for mitigation behavior: people cannot predict the affect evoked by a severe flood . Risk Analysis 28 ( 3 ): 771 – 778 . Google Scholar CrossRef Search ADS Townsend , R. M. ( 1994 ). Risk and insurance in village India . Econometrica: Journal of the Econometric Society 62 ( 3 ): 539 – 591 . Google Scholar CrossRef Search ADS Turvey , C. G. ( 2001 ). Weather derivatives for specific event risks in agriculture . Applied Economic Perspectives and Policy 23 ( 2 ): 333 – 351 . Tversky , A. and Kahneman , D. ( 1973 ). Availability: a heuristic for judging frequency and probability . Cognitive Psychology 5 ( 2 ): 207 – 232 . Google Scholar CrossRef Search ADS Tversky , A. and Kahneman , D. ( 1992 ). Advances in prospect theory: cumulative representation of uncertainty . Journal of Risk and Uncertainty 5 ( 4 ): 297 – 323 . Google Scholar CrossRef Search ADS Table A1. Summary statistics 2006 2007 2008 2009 2010 2011 2012 2013 Total SD Purchased insurance (%) 19.05 40.30 21.05 17.13 54.70 45.75 47.62 56.42 40.27 49.05 Repurchasers (%) 0.00 5.26 14.14 5.75 10.94 31.27 27.85 32.25 18.10 38.51 New purchasers (%) 0.00 17.60 6.91 7.96 43.76 14.48 19.78 24.17 18.80 39.07 Quitters (have not purchased again) (%) 0.00 6.58 26.15 8.40 6.19 23.43 17.90 15.07 13.77 34.46 Received payout (yes/no) (%) 0.00 0.00 6.25 6.96 33.15 7.07 35.03 – 12.61 33.20 Suffered crop loss (yes/no) (%) 77.51 22.04 30.43 49.72 29.72 23.20 27.96 – 28.92 45.34 Number of insurance policies bought 1.03 1.02 1.07 2.33 4.51 2.14 1.95 1.99 2.36 2.09 Price paid per policy 103 69 140 59 21 62 63 63 60 56 Number of households in village who received a payout (if village payout per policy≥Rs.1) 0 0 10 13 27 11 14 – 16 12 Mean village revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,400 259 1,917 473 1,359 – 1,274 3,206 Individual payout received (Rs., if payout≥Rs.1) 0 0 169 87 321 24 354 – 143 384 Individual payout per policy received (Rs., if payout≥Rs.1) 0 0 169 38 78 14 175 – 59 124 Revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,726 311 2,076 421 1,274 – 1,505 8,559 Insurance premium 215 68 190 152 75 195 200 200 161 54 Average price paid per policy/market price 0.49 1.00 0.74 0.37 0.28 0.32 0.31 0.32 0.41 0.35 Number of households 378 608 608 905 905 905 905 989 6,203 2006 2007 2008 2009 2010 2011 2012 2013 Total SD Purchased insurance (%) 19.05 40.30 21.05 17.13 54.70 45.75 47.62 56.42 40.27 49.05 Repurchasers (%) 0.00 5.26 14.14 5.75 10.94 31.27 27.85 32.25 18.10 38.51 New purchasers (%) 0.00 17.60 6.91 7.96 43.76 14.48 19.78 24.17 18.80 39.07 Quitters (have not purchased again) (%) 0.00 6.58 26.15 8.40 6.19 23.43 17.90 15.07 13.77 34.46 Received payout (yes/no) (%) 0.00 0.00 6.25 6.96 33.15 7.07 35.03 – 12.61 33.20 Suffered crop loss (yes/no) (%) 77.51 22.04 30.43 49.72 29.72 23.20 27.96 – 28.92 45.34 Number of insurance policies bought 1.03 1.02 1.07 2.33 4.51 2.14 1.95 1.99 2.36 2.09 Price paid per policy 103 69 140 59 21 62 63 63 60 56 Number of households in village who received a payout (if village payout per policy≥Rs.1) 0 0 10 13 27 11 14 – 16 12 Mean village revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,400 259 1,917 473 1,359 – 1,274 3,206 Individual payout received (Rs., if payout≥Rs.1) 0 0 169 87 321 24 354 – 143 384 Individual payout per policy received (Rs., if payout≥Rs.1) 0 0 169 38 78 14 175 – 59 124 Revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,726 311 2,076 421 1,274 – 1,505 8,559 Insurance premium 215 68 190 152 75 195 200 200 161 54 Average price paid per policy/market price 0.49 1.00 0.74 0.37 0.28 0.32 0.31 0.32 0.41 0.35 Number of households 378 608 608 905 905 905 905 989 6,203 View Large Table A1. Summary statistics 2006 2007 2008 2009 2010 2011 2012 2013 Total SD Purchased insurance (%) 19.05 40.30 21.05 17.13 54.70 45.75 47.62 56.42 40.27 49.05 Repurchasers (%) 0.00 5.26 14.14 5.75 10.94 31.27 27.85 32.25 18.10 38.51 New purchasers (%) 0.00 17.60 6.91 7.96 43.76 14.48 19.78 24.17 18.80 39.07 Quitters (have not purchased again) (%) 0.00 6.58 26.15 8.40 6.19 23.43 17.90 15.07 13.77 34.46 Received payout (yes/no) (%) 0.00 0.00 6.25 6.96 33.15 7.07 35.03 – 12.61 33.20 Suffered crop loss (yes/no) (%) 77.51 22.04 30.43 49.72 29.72 23.20 27.96 – 28.92 45.34 Number of insurance policies bought 1.03 1.02 1.07 2.33 4.51 2.14 1.95 1.99 2.36 2.09 Price paid per policy 103 69 140 59 21 62 63 63 60 56 Number of households in village who received a payout (if village payout per policy≥Rs.1) 0 0 10 13 27 11 14 – 16 12 Mean village revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,400 259 1,917 473 1,359 – 1,274 3,206 Individual payout received (Rs., if payout≥Rs.1) 0 0 169 87 321 24 354 – 143 384 Individual payout per policy received (Rs., if payout≥Rs.1) 0 0 169 38 78 14 175 – 59 124 Revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,726 311 2,076 421 1,274 – 1,505 8,559 Insurance premium 215 68 190 152 75 195 200 200 161 54 Average price paid per policy/market price 0.49 1.00 0.74 0.37 0.28 0.32 0.31 0.32 0.41 0.35 Number of households 378 608 608 905 905 905 905 989 6,203 2006 2007 2008 2009 2010 2011 2012 2013 Total SD Purchased insurance (%) 19.05 40.30 21.05 17.13 54.70 45.75 47.62 56.42 40.27 49.05 Repurchasers (%) 0.00 5.26 14.14 5.75 10.94 31.27 27.85 32.25 18.10 38.51 New purchasers (%) 0.00 17.60 6.91 7.96 43.76 14.48 19.78 24.17 18.80 39.07 Quitters (have not purchased again) (%) 0.00 6.58 26.15 8.40 6.19 23.43 17.90 15.07 13.77 34.46 Received payout (yes/no) (%) 0.00 0.00 6.25 6.96 33.15 7.07 35.03 – 12.61 33.20 Suffered crop loss (yes/no) (%) 77.51 22.04 30.43 49.72 29.72 23.20 27.96 – 28.92 45.34 Number of insurance policies bought 1.03 1.02 1.07 2.33 4.51 2.14 1.95 1.99 2.36 2.09 Price paid per policy 103 69 140 59 21 62 63 63 60 56 Number of households in village who received a payout (if village payout per policy≥Rs.1) 0 0 10 13 27 11 14 – 16 12 Mean village revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,400 259 1,917 473 1,359 – 1,274 3,206 Individual payout received (Rs., if payout≥Rs.1) 0 0 169 87 321 24 354 – 143 384 Individual payout per policy received (Rs., if payout≥Rs.1) 0 0 169 38 78 14 175 – 59 124 Revenue lost due to crop loss (Rs., if payout≥Rs.1) 0 0 2,726 311 2,076 421 1,274 – 1,505 8,559 Insurance premium 215 68 190 152 75 195 200 200 161 54 Average price paid per policy/market price 0.49 1.00 0.74 0.37 0.28 0.32 0.31 0.32 0.41 0.35 Number of households 378 608 608 905 905 905 905 989 6,203 View Large Table A2. Summary statistics of marketing interventions Mean SD BDM offer/premium 24.23 24.92 Game for four policies 36.39 48.11 Risk exposure flyer 57.28 49.47 Assigned video testimonial 7.17 25.81 Assigned drought flyer 7.14 25.75 Assigned subsidies flyer 7.19 25.83 Assigned loan 7.01 25.54 Rebates: Buy one get 50% off 0.97 9.79 Rebates: Buy two get one free 0.84 9.12 Rebates: Buy three get one free 0.98 9.87 HYV flyer 2.29 14.96 Group treatment (flyer) 2.90 16.79 Muslim treatment (flyer) 2.11 14.38 Hindu treatment (flyer) 2.06 14.22 Strong SEWA brand 3.71 18.90 Peer endorsed 3.71 18.90 Information treatment: pays 2/10 years 1.95 13.83 Positive frame 1.79 13.26 Vulnerability frame 1.92 13.72 Negative flyer 0.74 8.58 Flyer positive language 0.53 7.27 Flyer negative language 0.84 9.12 Discount amount (%) 3.92 15.13 Discount squared 244.10 1,214.75 Observations 6,203 Mean SD BDM offer/premium 24.23 24.92 Game for four policies 36.39 48.11 Risk exposure flyer 57.28 49.47 Assigned video testimonial 7.17 25.81 Assigned drought flyer 7.14 25.75 Assigned subsidies flyer 7.19 25.83 Assigned loan 7.01 25.54 Rebates: Buy one get 50% off 0.97 9.79 Rebates: Buy two get one free 0.84 9.12 Rebates: Buy three get one free 0.98 9.87 HYV flyer 2.29 14.96 Group treatment (flyer) 2.90 16.79 Muslim treatment (flyer) 2.11 14.38 Hindu treatment (flyer) 2.06 14.22 Strong SEWA brand 3.71 18.90 Peer endorsed 3.71 18.90 Information treatment: pays 2/10 years 1.95 13.83 Positive frame 1.79 13.26 Vulnerability frame 1.92 13.72 Negative flyer 0.74 8.58 Flyer positive language 0.53 7.27 Flyer negative language 0.84 9.12 Discount amount (%) 3.92 15.13 Discount squared 244.10 1,214.75 Observations 6,203 Note: All values apart from discount amount and discount squared show the proportion of the sample that has been subject to specific market intervention. Discount amount is the average discount rate for the whole sample. View Large Table A2. Summary statistics of marketing interventions Mean SD BDM offer/premium 24.23 24.92 Game for four policies 36.39 48.11 Risk exposure flyer 57.28 49.47 Assigned video testimonial 7.17 25.81 Assigned drought flyer 7.14 25.75 Assigned subsidies flyer 7.19 25.83 Assigned loan 7.01 25.54 Rebates: Buy one get 50% off 0.97 9.79 Rebates: Buy two get one free 0.84 9.12 Rebates: Buy three get one free 0.98 9.87 HYV flyer 2.29 14.96 Group treatment (flyer) 2.90 16.79 Muslim treatment (flyer) 2.11 14.38 Hindu treatment (flyer) 2.06 14.22 Strong SEWA brand 3.71 18.90 Peer endorsed 3.71 18.90 Information treatment: pays 2/10 years 1.95 13.83 Positive frame 1.79 13.26 Vulnerability frame 1.92 13.72 Negative flyer 0.74 8.58 Flyer positive language 0.53 7.27 Flyer negative language 0.84 9.12 Discount amount (%) 3.92 15.13 Discount squared 244.10 1,214.75 Observations 6,203 Mean SD BDM offer/premium 24.23 24.92 Game for four policies 36.39 48.11 Risk exposure flyer 57.28 49.47 Assigned video testimonial 7.17 25.81 Assigned drought flyer 7.14 25.75 Assigned subsidies flyer 7.19 25.83 Assigned loan 7.01 25.54 Rebates: Buy one get 50% off 0.97 9.79 Rebates: Buy two get one free 0.84 9.12 Rebates: Buy three get one free 0.98 9.87 HYV flyer 2.29 14.96 Group treatment (flyer) 2.90 16.79 Muslim treatment (flyer) 2.11 14.38 Hindu treatment (flyer) 2.06 14.22 Strong SEWA brand 3.71 18.90 Peer endorsed 3.71 18.90 Information treatment: pays 2/10 years 1.95 13.83 Positive frame 1.79 13.26 Vulnerability frame 1.92 13.72 Negative flyer 0.74 8.58 Flyer positive language 0.53 7.27 Flyer negative language 0.84 9.12 Discount amount (%) 3.92 15.13 Discount squared 244.10 1,214.75 Observations 6,203 Note: All values apart from discount amount and discount squared show the proportion of the sample that has been subject to specific market intervention. Discount amount is the average discount rate for the whole sample. View Large Table A3. Is irrigation important for the decision to purchase insurance? (1) (2) (3) (4) (5) Excess rainfall in previous year 0.164 *** 0.151 *** 0.176 *** 0.157 *** 0.150 *** (0.053) (0.042) (0.039) (0.040) (0.042) Lack of rainfall in previous year −0.008 −0.101 ** −0.050 −0.071 * −0.101 ** (0.066) (0.050) (0.036) (0.037) (0.050) Percentage irrigated −0.149 * −0.015 0.004 (0.083) (0.083) (0.091) Village payout in previous year 0.576 *** 0.696 *** 0.792 *** 0.697 *** (0.199) (0.172) (0.152) (0.169) Individual payout in previous year −0.048 0.020 0.018 0.020 (0.049) (0.049) (0.049) (0.050) Number of insurance policies bought in previous year 0.005 0.012 0.012 0.012 (0.015) (0.013) (0.013) (0.013) Excess rainfall × village payout in previous year 0.638 * −0.061 −0.059 (0.334) (0.358) (0.365) Lack of rainfall × village payout in previous year −0.203 0.379 0.380 (0.363) (0.330) (0.331) Constant 0.409 *** 0.306 ** 0.780 *** 0.323 ** 0.303 * (0.129) (0.134) (0.104) (0.159) (0.161) Observations 1,940 1,940 1,940 1,940 1,940 Household FE Yes No No No No Year dummies Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes (1) (2) (3) (4) (5) Excess rainfall in previous year 0.164 *** 0.151 *** 0.176 *** 0.157 *** 0.150 *** (0.053) (0.042) (0.039) (0.040) (0.042) Lack of rainfall in previous year −0.008 −0.101 ** −0.050 −0.071 * −0.101 ** (0.066) (0.050) (0.036) (0.037) (0.050) Percentage irrigated −0.149 * −0.015 0.004 (0.083) (0.083) (0.091) Village payout in previous year 0.576 *** 0.696 *** 0.792 *** 0.697 *** (0.199) (0.172) (0.152) (0.169) Individual payout in previous year −0.048 0.020 0.018 0.020 (0.049) (0.049) (0.049) (0.050) Number of insurance policies bought in previous year 0.005 0.012 0.012 0.012 (0.015) (0.013) (0.013) (0.013) Excess rainfall × village payout in previous year 0.638 * −0.061 −0.059 (0.334) (0.358) (0.365) Lack of rainfall × village payout in previous year −0.203 0.379 0.380 (0.363) (0.330) (0.331) Constant 0.409 *** 0.306 ** 0.780 *** 0.323 ** 0.303 * (0.129) (0.134) (0.104) (0.159) (0.161) Observations 1,940 1,940 1,940 1,940 1,940 Household FE Yes No No No No Year dummies Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05. View Large Table A3. Is irrigation important for the decision to purchase insurance? (1) (2) (3) (4) (5) Excess rainfall in previous year 0.164 *** 0.151 *** 0.176 *** 0.157 *** 0.150 *** (0.053) (0.042) (0.039) (0.040) (0.042) Lack of rainfall in previous year −0.008 −0.101 ** −0.050 −0.071 * −0.101 ** (0.066) (0.050) (0.036) (0.037) (0.050) Percentage irrigated −0.149 * −0.015 0.004 (0.083) (0.083) (0.091) Village payout in previous year 0.576 *** 0.696 *** 0.792 *** 0.697 *** (0.199) (0.172) (0.152) (0.169) Individual payout in previous year −0.048 0.020 0.018 0.020 (0.049) (0.049) (0.049) (0.050) Number of insurance policies bought in previous year 0.005 0.012 0.012 0.012 (0.015) (0.013) (0.013) (0.013) Excess rainfall × village payout in previous year 0.638 * −0.061 −0.059 (0.334) (0.358) (0.365) Lack of rainfall × village payout in previous year −0.203 0.379 0.380 (0.363) (0.330) (0.331) Constant 0.409 *** 0.306 ** 0.780 *** 0.323 ** 0.303 * (0.129) (0.134) (0.104) (0.159) (0.161) Observations 1,940 1,940 1,940 1,940 1,940 Household FE Yes No No No No Year dummies Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes (1) (2) (3) (4) (5) Excess rainfall in previous year 0.164 *** 0.151 *** 0.176 *** 0.157 *** 0.150 *** (0.053) (0.042) (0.039) (0.040) (0.042) Lack of rainfall in previous year −0.008 −0.101 ** −0.050 −0.071 * −0.101 ** (0.066) (0.050) (0.036) (0.037) (0.050) Percentage irrigated −0.149 * −0.015 0.004 (0.083) (0.083) (0.091) Village payout in previous year 0.576 *** 0.696 *** 0.792 *** 0.697 *** (0.199) (0.172) (0.152) (0.169) Individual payout in previous year −0.048 0.020 0.018 0.020 (0.049) (0.049) (0.049) (0.050) Number of insurance policies bought in previous year 0.005 0.012 0.012 0.012 (0.015) (0.013) (0.013) (0.013) Excess rainfall × village payout in previous year 0.638 * −0.061 −0.059 (0.334) (0.358) (0.365) Lack of rainfall × village payout in previous year −0.203 0.379 0.380 (0.363) (0.330) (0.331) Constant 0.409 *** 0.306 ** 0.780 *** 0.323 ** 0.303 * (0.129) (0.134) (0.104) (0.159) (0.161) Observations 1,940 1,940 1,940 1,940 1,940 Household FE Yes No No No No Year dummies Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05. View Large Table A4. Does irrigation explain our baseline result? (1) (2) (3) (4) Excess rainfall in previous year 0.407 ** 0.467 *** 0.413 *** (0.171) (0.137) (0.132) Lack of rainfall in previous year 0.079 −0.352 *** −0.416 *** (0.186) (0.124) (0.129) Percentage irrigated −0.062 −0.097 −0.004 −0.010 (0.126) (0.114) (0.111) (0.065) Excess rainfall × percentage irrigated −0.294 −0.362 ** −0.303 ** (0.190) (0.157) (0.147) Lack of rainfall × percentage irrigated −0.095 0.307 ** 0.350 ** (0.208) (0.145) (0.140) Village payout in previous year 0.781 *** 0.692 *** (0.162) (0.144) Individual payout in previous year 0.016 0.020 (0.051) (0.040) Number of insurance policies bought in previous year 0.011 0.010 (0.013) (0.015) Excess rainfall × village payout in previous year −0.051 (0.300) Lack of rainfall × village payout in previous year 0.389 (0.287) Constant 0.701 *** 0.745 *** 0.744 *** 0.745 *** (0.139) (0.127) (0.147) (0.112) Observations 1,940 1,940 1,940 1,940 Household FE No No No No Year dummies Yes Yes Yes Yes Controls Yes Yes Yes Yes (1) (2) (3) (4) Excess rainfall in previous year 0.407 ** 0.467 *** 0.413 *** (0.171) (0.137) (0.132) Lack of rainfall in previous year 0.079 −0.352 *** −0.416 *** (0.186) (0.124) (0.129) Percentage irrigated −0.062 −0.097 −0.004 −0.010 (0.126) (0.114) (0.111) (0.065) Excess rainfall × percentage irrigated −0.294 −0.362 ** −0.303 ** (0.190) (0.157) (0.147) Lack of rainfall × percentage irrigated −0.095 0.307 ** 0.350 ** (0.208) (0.145) (0.140) Village payout in previous year 0.781 *** 0.692 *** (0.162) (0.144) Individual payout in previous year 0.016 0.020 (0.051) (0.040) Number of insurance policies bought in previous year 0.011 0.010 (0.013) (0.015) Excess rainfall × village payout in previous year −0.051 (0.300) Lack of rainfall × village payout in previous year 0.389 (0.287) Constant 0.701 *** 0.745 *** 0.744 *** 0.745 *** (0.139) (0.127) (0.147) (0.112) Observations 1,940 1,940 1,940 1,940 Household FE No No No No Year dummies Yes Yes Yes Yes Controls Yes Yes Yes Yes Note: OLS. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05. View Large Table A4. Does irrigation explain our baseline result? (1) (2) (3) (4) Excess rainfall in previous year 0.407 ** 0.467 *** 0.413 *** (0.171) (0.137) (0.132) Lack of rainfall in previous year 0.079 −0.352 *** −0.416 *** (0.186) (0.124) (0.129) Percentage irrigated −0.062 −0.097 −0.004 −0.010 (0.126) (0.114) (0.111) (0.065) Excess rainfall × percentage irrigated −0.294 −0.362 ** −0.303 ** (0.190) (0.157) (0.147) Lack of rainfall × percentage irrigated −0.095 0.307 ** 0.350 ** (0.208) (0.145) (0.140) Village payout in previous year 0.781 *** 0.692 *** (0.162) (0.144) Individual payout in previous year 0.016 0.020 (0.051) (0.040) Number of insurance policies bought in previous year 0.011 0.010 (0.013) (0.015) Excess rainfall × village payout in previous year −0.051 (0.300) Lack of rainfall × village payout in previous year 0.389 (0.287) Constant 0.701 *** 0.745 *** 0.744 *** 0.745 *** (0.139) (0.127) (0.147) (0.112) Observations 1,940 1,940 1,940 1,940 Household FE No No No No Year dummies Yes Yes Yes Yes Controls Yes Yes Yes Yes (1) (2) (3) (4) Excess rainfall in previous year 0.407 ** 0.467 *** 0.413 *** (0.171) (0.137) (0.132) Lack of rainfall in previous year 0.079 −0.352 *** −0.416 *** (0.186) (0.124) (0.129) Percentage irrigated −0.062 −0.097 −0.004 −0.010 (0.126) (0.114) (0.111) (0.065) Excess rainfall × percentage irrigated −0.294 −0.362 ** −0.303 ** (0.190) (0.157) (0.147) Lack of rainfall × percentage irrigated −0.095 0.307 ** 0.350 ** (0.208) (0.145) (0.140) Village payout in previous year 0.781 *** 0.692 *** (0.162) (0.144) Individual payout in previous year 0.016 0.020 (0.051) (0.040) Number of insurance policies bought in previous year 0.011 0.010 (0.013) (0.015) Excess rainfall × village payout in previous year −0.051 (0.300) Lack of rainfall × village payout in previous year 0.389 (0.287) Constant 0.701 *** 0.745 *** 0.744 *** 0.745 *** (0.139) (0.127) (0.147) (0.112) Observations 1,940 1,940 1,940 1,940 Household FE No No No No Year dummies Yes Yes Yes Yes Controls Yes Yes Yes Yes Note: OLS. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 per cent level. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05. View Large Table A5. Placebo estimation: Do future events matter for take-up? Insurance purchasers (1) (2) (3) Excess rainfall in next year −0.050 −0.067 −0.069 (0.080) (0.078) (0.079) Lack of rainfall in next year 0.094 0.082 0.083 (0.080) (0.086) (0.086) Village payout in previous year 0.359 ** 0.468 ** (0.157) (0.196) Individual payout in previous year −0.050 (0.059) Number of insurance policies bought in previous year −0.012 (0.014) Constant 0.466 *** 0.441 * 0.449 ** (0.088) (0.221) (0.219) Observations 1,509 1,509 1,509 Number of households 749 749 749 Household FE Yes Yes Yes Year dummies No Yes Yes Controls Yes Yes Yes Insurance purchasers (1) (2) (3) Excess rainfall in next year −0.050 −0.067 −0.069 (0.080) (0.078) (0.079) Lack of rainfall in next year 0.094 0.082 0.083 (0.080) (0.086) (0.086) Village payout in previous year 0.359 ** 0.468 ** (0.157) (0.196) Individual payout in previous year −0.050 (0.059) Number of insurance policies bought in previous year −0.012 (0.014) Constant 0.466 *** 0.441 * 0.449 ** (0.088) (0.221) (0.219) Observations 1,509 1,509 1,509 Number of households 749 749 749 Household FE Yes Yes Yes Year dummies No Yes Yes Controls Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A5. Placebo estimation: Do future events matter for take-up? Insurance purchasers (1) (2) (3) Excess rainfall in next year −0.050 −0.067 −0.069 (0.080) (0.078) (0.079) Lack of rainfall in next year 0.094 0.082 0.083 (0.080) (0.086) (0.086) Village payout in previous year 0.359 ** 0.468 ** (0.157) (0.196) Individual payout in previous year −0.050 (0.059) Number of insurance policies bought in previous year −0.012 (0.014) Constant 0.466 *** 0.441 * 0.449 ** (0.088) (0.221) (0.219) Observations 1,509 1,509 1,509 Number of households 749 749 749 Household FE Yes Yes Yes Year dummies No Yes Yes Controls Yes Yes Yes Insurance purchasers (1) (2) (3) Excess rainfall in next year −0.050 −0.067 −0.069 (0.080) (0.078) (0.079) Lack of rainfall in next year 0.094 0.082 0.083 (0.080) (0.086) (0.086) Village payout in previous year 0.359 ** 0.468 ** (0.157) (0.196) Individual payout in previous year −0.050 (0.059) Number of insurance policies bought in previous year −0.012 (0.014) Constant 0.466 *** 0.441 * 0.449 ** (0.088) (0.221) (0.219) Observations 1,509 1,509 1,509 Number of households 749 749 749 Household FE Yes Yes Yes Year dummies No Yes Yes Controls Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A6. Intensity of events: What type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) Excess rainfall in previous year (z-score) 0.079 *** 0.098 *** 0.152 *** 0.151 *** (0.024) (0.028) (0.035) (0.035) Lack of rainfall in previous year (z-score) 0.002 −0.008 0.012 0.009 (0.043) (0.051) (0.055) (0.055) Village payout in previous year 0.569 *** 0.654 *** (0.149) (0.192) Individual payout in previous year −0.036 (0.049) Number of insurance policies bought in previous year 0.009 (0.015) Constant 0.538 *** 0.571 *** 0.404 *** 0.385 *** (0.014) (0.081) (0.129) (0.131) Observations 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Insurance purchasers (1) (2) (3) (4) Excess rainfall in previous year (z-score) 0.079 *** 0.098 *** 0.152 *** 0.151 *** (0.024) (0.028) (0.035) (0.035) Lack of rainfall in previous year (z-score) 0.002 −0.008 0.012 0.009 (0.043) (0.051) (0.055) (0.055) Village payout in previous year 0.569 *** 0.654 *** (0.149) (0.192) Individual payout in previous year −0.036 (0.049) Number of insurance policies bought in previous year 0.009 (0.015) Constant 0.538 *** 0.571 *** 0.404 *** 0.385 *** (0.014) (0.081) (0.129) (0.131) Observations 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01. View Large Table A6. Intensity of events: What type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) Excess rainfall in previous year (z-score) 0.079 *** 0.098 *** 0.152 *** 0.151 *** (0.024) (0.028) (0.035) (0.035) Lack of rainfall in previous year (z-score) 0.002 −0.008 0.012 0.009 (0.043) (0.051) (0.055) (0.055) Village payout in previous year 0.569 *** 0.654 *** (0.149) (0.192) Individual payout in previous year −0.036 (0.049) Number of insurance policies bought in previous year 0.009 (0.015) Constant 0.538 *** 0.571 *** 0.404 *** 0.385 *** (0.014) (0.081) (0.129) (0.131) Observations 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Insurance purchasers (1) (2) (3) (4) Excess rainfall in previous year (z-score) 0.079 *** 0.098 *** 0.152 *** 0.151 *** (0.024) (0.028) (0.035) (0.035) Lack of rainfall in previous year (z-score) 0.002 −0.008 0.012 0.009 (0.043) (0.051) (0.055) (0.055) Village payout in previous year 0.569 *** 0.654 *** (0.149) (0.192) Individual payout in previous year −0.036 (0.049) Number of insurance policies bought in previous year 0.009 (0.015) Constant 0.538 *** 0.571 *** 0.404 *** 0.385 *** (0.014) (0.081) (0.129) (0.131) Observations 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 Household FE Yes Yes Yes Yes Year dummies No No Yes Yes Controls No Yes Yes Yes Note: OLS. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01. View Large Table A7. Robustness using CHIRP: extreme events and revenue lost due to crop loss Revenue lost due to crop loss (Rs ’0000s Mean village revenue lost due to crop loss (Rs ’0000s) (1) (2) (3) (4) Excess rainfall 0.060 *** 0.059 *** 0.084 *** 0.083 *** (0.019) (0.019) (0.024) (0.024) Lack of rainfall −0.003 0.001 −0.016 −0.015 (0.014) (0.015) (0.023) (0.023) Purchased insurance (dummy) −0.006 0.013 (0.008) (0.011) Village payout per policy (Rs. ’000s) −0.028 −0.030 (0.028) (0.052) Individual payout in previous year 0.004 0.012 (0.018) (0.024) Constant 0.128 *** 0.024 0.123 ** 0.063 * (0.017) (0.023) (0.050) (0.036) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Revenue lost due to crop loss (Rs ’0000s Mean village revenue lost due to crop loss (Rs ’0000s) (1) (2) (3) (4) Excess rainfall 0.060 *** 0.059 *** 0.084 *** 0.083 *** (0.019) (0.019) (0.024) (0.024) Lack of rainfall −0.003 0.001 −0.016 −0.015 (0.014) (0.015) (0.023) (0.023) Purchased insurance (dummy) −0.006 0.013 (0.008) (0.011) Village payout per policy (Rs. ’000s) −0.028 −0.030 (0.028) (0.052) Individual payout in previous year 0.004 0.012 (0.018) (0.024) Constant 0.128 *** 0.024 0.123 ** 0.063 * (0.017) (0.023) (0.050) (0.036) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Note: OLS. Extreme events are calculated using the CHIRP data rather than the GPCC data. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A7. Robustness using CHIRP: extreme events and revenue lost due to crop loss Revenue lost due to crop loss (Rs ’0000s Mean village revenue lost due to crop loss (Rs ’0000s) (1) (2) (3) (4) Excess rainfall 0.060 *** 0.059 *** 0.084 *** 0.083 *** (0.019) (0.019) (0.024) (0.024) Lack of rainfall −0.003 0.001 −0.016 −0.015 (0.014) (0.015) (0.023) (0.023) Purchased insurance (dummy) −0.006 0.013 (0.008) (0.011) Village payout per policy (Rs. ’000s) −0.028 −0.030 (0.028) (0.052) Individual payout in previous year 0.004 0.012 (0.018) (0.024) Constant 0.128 *** 0.024 0.123 ** 0.063 * (0.017) (0.023) (0.050) (0.036) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Revenue lost due to crop loss (Rs ’0000s Mean village revenue lost due to crop loss (Rs ’0000s) (1) (2) (3) (4) Excess rainfall 0.060 *** 0.059 *** 0.084 *** 0.083 *** (0.019) (0.019) (0.024) (0.024) Lack of rainfall −0.003 0.001 −0.016 −0.015 (0.014) (0.015) (0.023) (0.023) Purchased insurance (dummy) −0.006 0.013 (0.008) (0.011) Village payout per policy (Rs. ’000s) −0.028 −0.030 (0.028) (0.052) Individual payout in previous year 0.004 0.012 (0.018) (0.024) Constant 0.128 *** 0.024 0.123 ** 0.063 * (0.017) (0.023) (0.050) (0.036) Observations 5,214 5,214 5,214 5,214 Number of households 905 905 905 905 Household FE Yes Yes Yes Yes Time dummies Yes Yes Yes Yes Controls No Yes No Yes Note: OLS. Extreme events are calculated using the CHIRP data rather than the GPCC data. Standard errors clustered at the village level are in parentheses. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A8. Robustness using CHIRP: what type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.128 *** 0.174 *** 0.172 *** 0.159 *** 0.164 *** 0.160 *** (0.033) (0.038) (0.044) (0.043) (0.043) (0.044) (0.043) Lack of rainfall in previous year −0.003 0.013 −0.007 −0.004 −0.016 −0.005 −0.017 (0.032) (0.039) (0.046) (0.046) (0.058) (0.053) (0.058) Village payout in previous year 0.571 *** 0.669 *** 0.548 *** 0.661 *** 0.527 ** (0.144) (0.189) (0.196) (0.189) (0.200) Individual payout in previous year −0.042 −0.043 −0.061 −0.032 (0.049) (0.049) (0.055) (0.056) Excess rainfall × village payout in previous year 0.327 ** 0.362 * (0.126) (0.186) Lack of rainfall × village payout in previous year 0.065 0.124 (0.144) (0.175) Excess rainfall × individual payout in previous year 0.064 −0.015 (0.051) (0.073) Lack of rainfall × individual payout in previous year 0.007 −0.032 (0.054) (0.062) Number of insurance policies bought in previous year 0.008 0.006 0.006 0.006 (0.015) (0.015) (0.015) (0.015) Constant 0.520 *** 0.552 *** 0.382 *** 0.364 *** 0.388 *** 0.370 *** 0.390 *** (0.017) (0.082) (0.130) (0.132) (0.129) (0.131) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.128 *** 0.174 *** 0.172 *** 0.159 *** 0.164 *** 0.160 *** (0.033) (0.038) (0.044) (0.043) (0.043) (0.044) (0.043) Lack of rainfall in previous year −0.003 0.013 −0.007 −0.004 −0.016 −0.005 −0.017 (0.032) (0.039) (0.046) (0.046) (0.058) (0.053) (0.058) Village payout in previous year 0.571 *** 0.669 *** 0.548 *** 0.661 *** 0.527 ** (0.144) (0.189) (0.196) (0.189) (0.200) Individual payout in previous year −0.042 −0.043 −0.061 −0.032 (0.049) (0.049) (0.055) (0.056) Excess rainfall × village payout in previous year 0.327 ** 0.362 * (0.126) (0.186) Lack of rainfall × village payout in previous year 0.065 0.124 (0.144) (0.175) Excess rainfall × individual payout in previous year 0.064 −0.015 (0.051) (0.073) Lack of rainfall × individual payout in previous year 0.007 −0.032 (0.054) (0.062) Number of insurance policies bought in previous year 0.008 0.006 0.006 0.006 (0.015) (0.015) (0.015) (0.015) Constant 0.520 *** 0.552 *** 0.382 *** 0.364 *** 0.388 *** 0.370 *** 0.390 *** (0.017) (0.082) (0.130) (0.132) (0.129) (0.131) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Note: OLS. Extreme events are calculated using the CHIRP data rather than the GPCC data. Standard errors clustered at the village level are in parentheses. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 % level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Table A8. Robustness using CHIRP: what type of events matter for insurance take-up? Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.128 *** 0.174 *** 0.172 *** 0.159 *** 0.164 *** 0.160 *** (0.033) (0.038) (0.044) (0.043) (0.043) (0.044) (0.043) Lack of rainfall in previous year −0.003 0.013 −0.007 −0.004 −0.016 −0.005 −0.017 (0.032) (0.039) (0.046) (0.046) (0.058) (0.053) (0.058) Village payout in previous year 0.571 *** 0.669 *** 0.548 *** 0.661 *** 0.527 ** (0.144) (0.189) (0.196) (0.189) (0.200) Individual payout in previous year −0.042 −0.043 −0.061 −0.032 (0.049) (0.049) (0.055) (0.056) Excess rainfall × village payout in previous year 0.327 ** 0.362 * (0.126) (0.186) Lack of rainfall × village payout in previous year 0.065 0.124 (0.144) (0.175) Excess rainfall × individual payout in previous year 0.064 −0.015 (0.051) (0.073) Lack of rainfall × individual payout in previous year 0.007 −0.032 (0.054) (0.062) Number of insurance policies bought in previous year 0.008 0.006 0.006 0.006 (0.015) (0.015) (0.015) (0.015) Constant 0.520 *** 0.552 *** 0.382 *** 0.364 *** 0.388 *** 0.370 *** 0.390 *** (0.017) (0.082) (0.130) (0.132) (0.129) (0.131) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Insurance purchasers (1) (2) (3) (4) (5) (6) (7) Excess rainfall in previous year 0.114 *** 0.128 *** 0.174 *** 0.172 *** 0.159 *** 0.164 *** 0.160 *** (0.033) (0.038) (0.044) (0.043) (0.043) (0.044) (0.043) Lack of rainfall in previous year −0.003 0.013 −0.007 −0.004 −0.016 −0.005 −0.017 (0.032) (0.039) (0.046) (0.046) (0.058) (0.053) (0.058) Village payout in previous year 0.571 *** 0.669 *** 0.548 *** 0.661 *** 0.527 ** (0.144) (0.189) (0.196) (0.189) (0.200) Individual payout in previous year −0.042 −0.043 −0.061 −0.032 (0.049) (0.049) (0.055) (0.056) Excess rainfall × village payout in previous year 0.327 ** 0.362 * (0.126) (0.186) Lack of rainfall × village payout in previous year 0.065 0.124 (0.144) (0.175) Excess rainfall × individual payout in previous year 0.064 −0.015 (0.051) (0.073) Lack of rainfall × individual payout in previous year 0.007 −0.032 (0.054) (0.062) Number of insurance policies bought in previous year 0.008 0.006 0.006 0.006 (0.015) (0.015) (0.015) (0.015) Constant 0.520 *** 0.552 *** 0.382 *** 0.364 *** 0.388 *** 0.370 *** 0.390 *** (0.017) (0.082) (0.130) (0.132) (0.129) (0.131) (0.129) Observations 1,940 1,940 1,940 1,940 1,940 1,940 1,940 Number of households 807 807 807 807 807 807 807 Household FE Yes Yes Yes Yes Yes Yes Yes Year dummies No No Yes Yes Yes Yes Yes Controls No Yes Yes Yes Yes Yes Yes Note: OLS. Extreme events are calculated using the CHIRP data rather than the GPCC data. Standard errors clustered at the village level are in parentheses. A bold number indicates that the main effect plus the interaction effect is statistically significant at least at the 5 % level. Payout values are in Rs. 1,000. ***p<0.01**p<0.05*p<0.1. View Large Author notes Review coordinated by Iain Fraser © Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

European Review of Agricultural EconomicsOxford University Press

Published: Mar 15, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off