Does Index-based Insurance Crowd out Credit?
Evidence from Ethiopia*
Anne B. Winkel
October 9, 2015
Abstract
Credit rationing could lead to underinvestment and difficulties to smooth consumption in developing countries. Index-based insurance might overcome credit rationing by diminishing both supply and demand constraints. This study focuses on the effects of rainfall index insurance on credit rationing for farmers in Ethiopia. By employing a direct elicitation method, it is found that more than 30% of the respondents is credit constrained. Insurance uptake could negatively affect demand constraints, whereas the effect on supply-sided constraints is mixed. As index-based insurance could serve as a substitute for credit, uptake of insurance leads to lower demand for credit. A tightened supply constraint could be the result of an increase in strategic default due to an enlarged minimum welfare level of the borrowers.
Keywords: Index-based insurance, Credit rationing, Risk JEL Classification Codes: O16, Q14, D81, G22
2
Acknowledgements
1 Introduction
Credit rationing could lead to suboptimal levels of investment and consumption-smoothing in low-income countries. Small enterprises need credit in order to increase their productive capacity, which is pivotal in boosting economic activity and growth (Freedman & Click, 2006). Credit constraints directly affect farm output (Petrick, 2004; Ali, Deininger & Duponchel, 2014) and farm profits (Foltz, 2004). Moreover, households with limited access to credit are unable to smooth their consumption, which causes irreversible damage to health and has lasting effects on education (Baez, Fuente & Santos, 2010).
The inability to obtain credit stems from various sources. A first distinction of these sources is based on whether a potential borrower is supply- or demand-sided constrained. Supply-sided constrained farmers have excess demand; the lending institution is not willing to lend (more). On the other hand, demand-sided constraints cause farmers to voluntarily withdraw from the credit market. Although their notional demand is positive, their effective demand is zero (Boucher, Guirkinger & Trivelli, 2009). Due to the large part of the risk the farmers bear or the high transaction cost involved, these potential borrowers do not apply for credit.
In recent years, the promise of index-based insurance (IBI) to farm households with limited assets has been well documented. IBI could overcome problems that are caused by credit constraints. With limited credit, farmers undertake less high-risk high-return activities and invest less in technological advancements (Hazell & Hess, 2010; McIntosh, Sarris & Papadopoulos, 2013). Moreover, insufficient credits forces farmers to opt for costly coping mechanisms, such as selling productive assets and reducing consumption (Janzen & Carter, 2013). A perfect insurance market encourages investments and enables farmers to refrain from these costly risk management strategies. This could substantially increase the welfare of poor households, especially in the presence of a poverty trap (Santos & Barrett, 2006).
In addition to solving problems caused by credit constraints, IBI could directly provide a solution to credit rationing. IBI diminishes the risk to both the borrower and the lender, thereby securing the returns on investment. Insurance could increase supply or ease the terms of borrowing. Further, IBI could limit credit rationing by increasing demand for credit, as farmers’ production risk is diminished too (Carter, Galarza & Boucher, 2007). In some circumstances, however, IBI could increase credit rationing, as insurance increases the minimum effective welfare level of the borrower and could provide incentives to more readily default (Clarke & Dercon, 2009).
4
What is the effect of index-based insurance on credit rationing?
In order to answer this question, the following objectives will be met:
1. To study the extent to which credit rationing takes place.
2. To study the effect of index-based insurance on credit rationing.
Three main contributions are made in this thesis. First, the connection between credit and index insurance is explored empirically, with specific reference to some recent normative models. This allows to explicitly test these theories. Second, a novel approach is used in order to specify how IBI can alleviate different types of credit constraints. The direct elicitation methodology as introduced by Boucher, et al. (2009) enables to classify the different types of credit rationing. Thirdly, the results from this study suggest that IBI takes on a variety of roles in relation to credit rationing. The effect of IBI uptake on supply-sided constraints is mixed. Strikingly, demand-sided constraints could be aggravated as a result of insurance uptake.
This thesis is outlined as follows. Section 2 provides a theoretical background. It includes a typology of credit rationing as well as an overview of the literature on the relation between credit and insurance. Section 3 outlines the methodology for studying the effect of IBI on credit rationing among Ethiopian farmers. Section 4 describes the data and Section 5 reports the results of the analysis. Section 6 provides a discussion of these results, while Section 7 concludes.
2 Theoretical background
In this section a theoretical background on credit rationing is provided. This section starts with a typology of credit rationing, after which some empirical evidence on the incidence and determinants of credit rationing is discussed. The major part of this section concerns the relation between IBI and credit. It ends with a description of the hypotheses that are tested.
2.1 Types of credit rationing
Several types of credit rationing can be distinguished. Boucher et al. (2009) distinguish price rationing and non-price rationing. Price rationed households are unconstrained as their notional demand is smaller than the lender’s supply. These households are either borrowers or non-borrowers. Non-price rationing occurs if a lender rations the amount of credit available due to adverse selection or moral hazard, or if a potential borrower voluntarily withdraws from the credit market.
Non-price rationing includes quantity, risk and transaction-cost rationing. Table 1 provides an overview of these types of credit rationing. Stiglitz and Weiss (1981) describe quantity rationing, which results from the refusal to lend (more), even if the borrower is willing to afford a higher interest rate. Higher interest rates would attract risky borrowers and borrowers with risky projects in the presence of asymmetric information, indicating problems of adverse selection and moral hazard. For this reason, there exists an optimal interest rate above which the lender rations credit.
Table 1: Credit rationing
Notional demand Effective demand Supply Credit rationed
Unconstrained Yes Yes Yes No
Unconstrained No No No No
Supply rationed Yes Yes No Yes
Demand rationed Yes No Yes Yes
Risk and transaction-cost rationing, on the other hand, occur if the effective demand for credit is limited. Potential borrowers voluntarily withdraw from the credit market as they find it too risky to borrow or consider the transaction costs too high. Although their notional demand is positive, their effective demand is smaller than the supply because of high transaction-costs or risk. These factors limit the utility the borrower would have in the absence of asymmetric information.
2.2 Empirical evidence of credit rationing
Research has typically only focused on quantity rationing (see for example Jaffee & Modigliani, 1969; Kochar, 1997; Foltz, 2004; Petrick, 2004; Malapit, 2012). Some of these studies find that credit constraints are not as severe as assumed, or even absent. For example, Kochar (1997) concludes that demand for credit is low and supply-sided credit constraints are not binding in India.
6
contracts (and retreating to safer, lower return activities) in order to avoid the risk of default and collateral loss’ (2007, p. 352).
Not all studies on risk rationing, however, find such constraints. Karlan, Kutsoati, McMillan and Udry (2011), for example, test whether loan take up is higher if the loan is bundled with an insurance product. They find that take up is 6% higher for the indemnified loans compared to the normal loans, but this difference is not statistically different. Neither is the effect of the type of loan on investment. Yet, the limited liability clause may already have provided some form of implicit insurance. High default rates indicate that this was a real possibility to the borrowers.
A number of studies have focused on the determinants of credit rationing. Assets function as collateral and therefore tend to relate negatively to credit rationing (Zeller, 1994; Steijvers & Voordeckers, 2009), whereas the current amount of borrowing (in the form of land or credit) is positively related to credit rationing (Petrick, 2004). Age, education and yields seem to be negatively related to credit rationing (Konchar, 1997; Jia, Heidhues & Zeller, 2010). Studying the relationship between gender and credit rationing, Zeller (1994) and Malapit (2012) both find that women more often apply for informal credit for consumption, whereas man more often apply for formal credit. While Zeller concludes that men are more likely to be rejected, Malapit finds that women are more often credit rationed.
Although Ali et al. (2014) do not find large differences in determinants between quantity rationing and other constraints, some studies have explicitly tested for the determinants of risk rationing. For example, the current amount of borrowing is negatively related to risk rationing (Cheng, 2014). In addition, some studies indicate a negative relationships between land and credit demand (Barslund, 2007), but not all. Furthermore, it is found that risk aversion is positively related to risk rationing (Ali et al., 2014).
2.3 Insurance and credit rationing
In this subsection, the effect of insurance on credit is elaborated upon. The model provided by Carter, Cheng and Sarris (2011) provides a useful starting point for this analysis. Their study demonstrates how index insurance can alleviate credit constraints and improve productivity. Other articles that shine some light on the complex relationship between credit and insurance are discussed thereafter.
This risk constraint could alter in the presence of formal insurance. If insurance is available, the asset-rich farmers do borrow. Index insurance decreases the risk they face, which leads to lower default rates as well as a lower probability of losing one’s collateral. Thus, farmers with high collateral gain from the stand-alone insurance as they bear all the risk of losing the collateral themselves.
On the other hand, farmers with little collateral choose to borrow if the interest rate is not too high, as high interest rates serve as a substitute for collateral. These farmers are price rationed if they cannot afford the high interest rate. As these farmers have little collateral, they do not bear the risk of losing it either. Purchasing an insurance contract is thus beneficial to their lenders, but provides little additional benefits to the borrowers. Providing the farmers with a stand-alone insurance contract does not alter the terms of borrowing.
Other studies focusing on insurance and credit help understand this relationship in more detail. For instance, Carter et al. (2007) consider that IBI could limit the risk of insolvency of the lender due to covariate risk or might improve the repayment rate and variation in repayment of the borrowers (Mishra, 1994). Although a stand-alone insurance contract does not by definition lead to lower interest rate, lenders could choose to do so. Indeed, Mahul and Skees (2007) observe that lenders lowered interest rate and were willing to lend more. IBI seems to be likely to relax supply-sided constraints, i.e. quantity rationing, too (Giné & Yang, 2009).
Two studies, however, point at negative rather than a positive effect of IBI on credit supply. Clarke and Dercon (2009) suggest that insurance could possibly crowd-out credit if IBI changes the behavior of the borrower. A stand-alone insurance product could increase the minimum welfare level that can be reached in case of strategic default by the borrower. In this way, quantity rationing could become more severe. Farrin and Miranda (2013) confirm that the possibility exists that banks are willing to lend less if the borrower has acquired a non-interlinked insurance contract, because insured borrowers have higher default rates than uninsured borrowers.
Little empirical evidence on the effect of IBI on credit rationing exists, but some studies provide more insights. In an experimental game conducted in China, Cheng (2014) studies the effect of offering IBI to risk rationed households. He finds more than half of the farmers that are risk rationed decide to apply for credit when IBI is made available to them. Additionally, roughly two thirds of the credit diverters choose to use their loan for productive investment rather than consumption when IBI is made available. The rationale behind this finding is that IBI reduces production risk, while the risk associated with using the loan for consumption stays constant or might even increase as the households pays for the premium but does not invest.
8
they expected that farmers would prefer the insured loan over the basic loan, demand for the basic loan was 13% higher than for the insured loan. Insurance did not lead to higher demand for credit. However, the authors suggest that the limited liability clause of the basic loan provided some implicit form of insurance. As suggested by Carter et al. (2011), a stand-alone insurance product (which effectively was the product studied by Giné and Yang) does not provide additional benefits to farmers with low collateral.
Khantachavana et al. (2012) conduct a study on the determinants of risk rationing. Although they too expected that insurance would have a negative effect on risk rationing, they find that risk rationing takes place despite the presence of insurance markets. One explanation for this surprising finding is that risk averse individuals are more likely to purchase insurance and to be risk rationed in the first place. Alternatively, one could reason that the type of insurance needs to match the purpose of the loan. For example, medical insurance is not likely to help one in obtaining a loan for farm inputs, whereas a drought insurance would.
Groh and McKenzie (2014) focus on risk rationing too by randomly offering current borrowers the opportunity to purchase an insurance product to hedge the risk of political uncertainty and macroeconomic instability. They tested whether the offer of insurance leads to an increase in in the likelihood of renewing one’s loan. Although they encountered a large demand for insurance, the insurance offer did not affect borrowing behavior. As an explanation, it is suggested that the insurance does not cover the type of risk that hinders the investments made by small and medium enterprises.
Lastly, Karlan, Osei, Osei-Akoto and Udry (2014) study the effects of insurance and capital grants on investment and welfare among Ghanaian farmers. Comparing farmers who only received the insurance grant with farmers who received both an insurance and a capital grant, they find that the former do not invest less than the latter. Apparently, farmers with solely insurance are able to find credit to increase investments. This indicates that insurance is able to overcome investment obstacles, whereas a lack of credit is not the bottleneck to investment. These findings suggest that supply-sided constraints might not be as severe as initially assumed, while risk constraints dampen investment.
2.4 Hypotheses
The literature discussed above provides a starting point for the hypotheses that are described below. Insurance could fulfill different roles in situations with credit rationing. How does insurance help overcome these types of credit rationing specifically? A number of hypotheses on the effect of non-interlinked index insurance on credit rationing are formulated.
households, insurance can overcome the risk of losing one’s collateral in case of default. Insurance could lower transaction-cost rationing too, as households that have gone through the process of purchasing insurance might be more familiar with the paperwork that has to be done. Although the empirical literature suggests that insurance does not increase demand for credit, the insurance products offered in the aforementioned studies might not have corresponded to the risk purchasers were confronted with. Together, these notions bring us to the first hypothesis:
Hypothesis 1: Insured households are less demand-sided constrained than uninsured households.
Secondly, the potential effect of index insurance on the quantity rationed is hypothesized. Although one would intuitively argue for a negative effect of IBI on quantity rationing, a number of studies have addressed the possibility that IBI aggravates supply constraints. As insurance increases the minimum welfare level of borrowers, borrowers are more likely to strategically default. Consequently, lenders might supply less or no credit. It is thus expected that IBI tightens supply constraints:
Hypothesis 2: Insured households are more supply-sided constrained than uninsured households.
In total, it is expected that IBI has an ambiguous effect on credit rationing. An increase in supply-sided constraints and a decrease in demand-sided constraints can neutralize the total effect of IBI on credit rationing. This serves as the basis of the third and last hypothesis:
Hypothesis 3: Insured households are equally credit constrained as uninsured households.
3 Methodology
This section describes how the effect of IBI on credit rationing is measured. The insurance was sold as a new product to farmers in Ethiopia. Data of both adopters and non-adopters has been gathered thereafter. An empirical strategy is developed to study the impact of insurance on credit rationing on the basis of this data.
3.1 Sale of rainfall index insurance in Ethiopia
10
production to food security, and the likelihood of index insurance to limit variance in agricultural production.
The details of the insurance product are as follows. OIC sells insurance products for a premium of ETB 100 per hectare.1 If rainfall in the nearest meteorological station is below the trigger point, a partial pay-out will be made. If rainfall is below the exit level, a full pay-out of ETB 500 is provided to the policy holders. The exact trigger and exit level differ per kebele and growing phase. Additionally, a cap was set at a maximum of 10 mm per day, to prevent households from being withheld indemnity payments due to one large cloudburst. The insurance was sold at a price 20% above the actuarially fair value to cover administration costs. The insurance could be purchased in order to cover the planting or flowering season. The planting phase runs from early April to the end of June while the flowering season covers early August to late October.
OIC makes use of satellite rainfall data, which has been estimated for areas of ten squared kilometer since 1983. With this measure, however, basis risk is present. As basis risk can be proxied by distance to the nearest weather station, it can partially be controlled for in the subsequent analysis.
Prior to the start of the sales, representatives from cooperatives and kebeles were invited to a training of trainers explaining the working of index insurance. Next, all farmers were invited to an information meeting given by these trainers, in which awareness about IBI was raised by providing information on the working of index insurance, the (dis)advantages of the product and the importance of risk management. The farmers were given the opportunity to purchase insurance up to 2-4 weeks prior to the start of each phase.
In total, around 12,000 farmers attended the awareness-raising meetings. As 1,286 farmers decided to take up the insurance product, the take-up rate of 10.7% is somewhat low compared to other field experiments in Ethiopia. For example, others report average take-up rates of 23% and 25% in Ethiopia (McIntosh et al., 2013; Dercon, Hill, Clarke, Outes-Leon & Taffesse, 2014). In framed field experiments and willingness-to-pay studies, take-up rates are even higher, ranging to 65% in for example Clarke and Kalani (2011). These studies, however, might misrepresent actual demand as households do not really have to inflict on their own financial resources in these situations.
The insurance product was sold in both 2013 and 2014. Indemnity payments were provided in both 2013 and 2014, depending on the location.
3.2 Data gathering and the direct elicitation method
The total sample consists of households from two different areas: the treatment area and the expansion (control) area. The treatment area consists of 788 households that had access to insurance. They had the
possibility to purchase insurance in 2013 and 2014. The expansion area consists of 337 households to whom OIC has not sold insurance yet, they do not have access to insurance.
Respondents in both areas were randomly sampled. A survey is drafted and conducted in the treatment and expansion areas. The survey aims to measure a number of variables that capture household characteristics as well as variables directly related to this study, including credit rationing, income and awareness about insurance.
Credit rationing is measured by using the direct elicitation methodology as described in detail by Boucher et al. (2009). It can be used to disentangle different forms of credit rationing. Although certain randomization strategies based on policy and field experiments have successfully been applied to measure credit constraints, the opportunities to conduct these experiments are rare. Alternatively, asking households step-wise about their credit rationing experience helps one to identify the different types of credit rationing. The answers to questions concerning applications for a bank loan, rejection and believes about acceptation of one’s potential application enable one to categorize the respondents into the different groups. Figure 1 provides an overview of the questions that are required to attain these responses.
12
The quantity or supply-sided constrained households are those that applied for a loan and have been rejected, who would actually have preferred a larger loan, or who did not apply for a loan because they thought the bank would not approve their application.
On the demand-sided constrained households, risk rationed and transaction-cost rationed households can be identified. Risk rationed households answer ‘I don’t want to put my land (or other collateral) at risk’, ‘I do not want to be worried; I am afraid’ or ‘Formal lenders are too strict; they are not as flexible as informal ones’ when asked why they did not apply for a loan. Transaction-cost rationed households answer ‘The branch is too far away’ or ‘There is too much paperwork; the costs associated with loan application are too high’ to the question. Unconstrained households are identified by answers such as ‘I do not need a loan’, ‘The interest rate is too high’, ‘Farming does not give me enough to repay a debt’ and ‘I prefer working with my own liquidity’.
DEM has proven to adequately indicate underlying motives for non-participation in the credit market. Boucher et al. (2009) have demonstrated this using data from Peru by showing that the observed rationing categories are strongly correlated with factors that influence the credit demand and supply. Other uses of DEM include research by Carter and Olinto (2003), Foltz (2004) and Petrick (2004).
3.3 Empirical strategy
This section describes how the effect of IBI on credit rationing is tested. The effect of insurance uptake on credit rationing is approximated using a difference-in-differences (DD) analysis. Next, this section describes the estimation techniques.
3.3.1 Estimating impact
In order to estimate the impact of insurance uptake, one should compare the outcome of a group receiving the treatment (i.e. insurance purchase) to the outcome of a similar control group. Ideally, participants in the study should randomly be assigned to either the control or treatment group to ensure that the two groups are as similar as possible (Armendáriz & Morduch, 2005). Then, the effect of insurance uptake can be measured by focusing on the differences in outcome (i.e. credit rationing) between the participants in the control and treatment group.
between the purchasers and the non-purchasers could explain differences in loan application or loan status. Respondents who purchased insurance might have been more willing to purchase insurance too, for example because their loan application might have been rejected due to the high risk their production faces.
In order to correct for the biases arising from endogenous program placement and the possibility that respondents self-selected into purchasing insurance, a difference-in-differences analysis is conducted to estimate the true effect of IBI uptake more accurately. The approach of Coleman (1999) is followed. Coleman aimed to study the impact of participation in a microfinance program in Northeast Thailand. As the microfinance institute selected certain target areas and the household could choose whether or not they would like to participate, problems with causal interference arising from endogenous program placement and self-selection were present. A special feature of the roll-out of the program, however, enabled Coleman to overcome these problems. The program did not start at the same time in all villages: in some villages, banks were active from the beginning of the program onwards, while banks in other areas would not open for another year. In both areas, however, the households already had the opportunity to choose to participate at the beginning of the program. By including a dummy variable indicating (future) bank membership and a variable indicating the number of months that credit was available to the households, self-selection and endogenous program placement could be controlled for.
In the subsequent analysis, I follow the same approach as Coleman. The sample is divided among two characteristics: treatment status and willingness to purchase IBI. Treatment status is similar to the observation whether the households had access to microfinance in Coleman’s study, whereas willingness to purchase IBI corresponds with the households’ decision to participate in the microfinance program or not.
Willingness to purchase IBI is known for those in the treatment area, as data about insurance uptake is available. In order to calculate the willingness to purchase insurance for the respondents in the expansion area, however, a different strategy is used. A propensity score, which is based on a number of variables that determine insurance uptake, is calculated for the respondents that purchased insurance in the treatment area. These variables are selected on the basis of existing literature and can be classified into the following categories: factors related to risk, to understanding and education, and to the respondent's financial situation.
The factors concerning risk include basis risk (Clarke, 2011), price (Elabed & Carter, 2014), risk aversion (Hill, Robles & Ceballos, 2013; Cheng, 2014) and risk mitigation strategies (Giné, Menand, Townsend & Vickery, 2010; Dercon et al. 2014). Basis risk and price are expected to limit demand, whereas the effect of risk aversion and the availability of risk mitigation strategies are less clearly related to insurance demand.
14
on uptake is not as large as expected and it appears that basic literacy might be more important than high levels of education (Clarke & Kalani, 2011).
Lastly, a number of factors indicating the household’s financial situation are of importance too. Although insurance demand should be independent from one’s wealth, in the presence of basis risk wealth and demand could be related to each other (Clarke, 2011). De Nicola and Hill (2013) therefore predict that demand for IBI will be hump-shaped in wealth. In addition to wealth, liquidity is found to be a driver of insurance demand. Giné, Menand, Townsend Vickery (2010) indicate that 80% of the people who did not purchase insurance mentioned insufficient funds as the most important reason for remaining uninsured. Income, finally, mainly affects insurance demand through the wealth effect (Chantarat, Mude & Barrett, 2009).
Not all of the variables described above have been measured quantitatively. On the basis of the available data, the co-variates that are used to measure the propensity score are selected: basis risk, production risk, risk mitigation strategies, understanding, education, wealth, and income. Next, this score is applied to all respondents in the expansion area. 2 By selecting a threshold for respondents’ likelihood to purchase IBI, the sample can be divided among this dimension too.
These two dimensions effectively split the total sample into four groups. Let Ti, i ∈ {0, 1} denote whether the household is part of the treatment or expansion area. Let wi, i ∈ {0, 1} denote whether the household would be willing to purchase insurance or not. W1 indicates the households in the treatment area that actually purchased insurance and the households in the expansion area that have a propensity score higher than the threshold. W0 represents all remaining households. In this way, four groups can be identified and compared. The first group, T1w1, consists of the households in the treatment area who have actually purchased insurance. The second group, T1w0, consists of households in the treatment area who did not purchase insurance. The third group, T0w1, consists of households in the expansion are who would like to purchase insurance but are not offered the opportunity yet. They would certainly like to purchase insurance if it would be made available to them. The last group, T0w0, consists of households in the expansion are that would not purchase insurance if it would be offered to them.
The DD technique can be described as follows:
DD = E(Yw=1 − Yw=0) | T = 1) − E(Yw=1 − Yw=0 | T = 0) (1)
In this way, the difference between households that purchased insurance and did not purchase insurance is adapted to take into account that the participating household were more willing to purchase insurance in the
2
first place. Interpreted alternatively, the differences between the households of the two areas that are willing to purchase insurance is adjusted for by pre-existing differences between these areas.
The effect of insurance purchase on credit rationing can more formally described as:
Yn,i = α + βTiw + γTi + δw + ζXi + εi, (2)
in which Yn,i indicates the vector of dependent variables with n ∈ {1, 2, 3}. T is a dummy indicating whether the households is part of the treatment or expansion area, whereas w indicates the respondent’s willingness to purchase insurance. β measures the effect of purchasing insurance for those in the treatment area, which is in line with the effect of the treatment on the treated. γ captures the differences between those in the treatment area and the expansion area, whereas δ measures the differences between those willing to purchase index-based insurance and those who are not willing. It is expected that β is negative and significant for demand-sided rationing, positive and significant for supply-sided rationing, and insignificant for credit rationing in general. Xi, lastly, represents a vector of control variables, whereas ε indicates the error term.
The most critical assumption the DD method hinges on is the one of the parallel trend. If differences are taken over time, this means that unobserved characteristics should change equally in both groups. It is described as follows:
Cov(εip, T1w) = 0 (3)
This assumption takes somewhat of a different form in this context. Here it means that unobserved characteristics should be the same in the group of household that are willing to purchase IBI and those who are not. This assumption is only partially met as not all characteristics that determine demand for insurance can be controlled for. Any unobservable factors driving demand for insurance purchase limit the validity of this study. Despite these weaknesses, this method reveals most accurately the true effect of insurance uptake on credit rationing compared to the aforementioned methods.
3.3.2 Estimation techniques
This last part of Section 3 describes the techniques to estimate the Equation 2. This equation is estimated using both linear and logistic regressions. Although linear models have the benefit of allowing the direct interpretation of the regression coefficients, they encounter a number of errors when estimating non-linear models. For this reason, logistic regressions are ran in addition (Verbeek, 2012).
16
matrix and standard errors are likely to be biased (Petersen, 2009). Clustering the standard errors at the village level enables the adjustment of the standard errors.
Still, the results obtained with clustered standard errors can be biased if there are too few clusters. With only 23 clusters, this number is insufficient indeed. Cameron, Gelbach and Miller suggest to apply bootstrapping in order to produce unbiased statistics (2008). The bootstrapping technique recommended for this type of data is wild cluster bootstrap-t, because it allows for heteroscedasticity across clusters as well as clusters of unequal sizes. For this reason, the wild cluster boostrap-t is used to estimate the linear models. As the wild cluster boostrap-t option is only available for linear models in Stata, the non-linear models are estimated using the pairs cluster boostrap-t procedure. This procedure could nonetheless over-reject, whereas the wild cluster boostrap-t is less likely to do so.
The wild cluster bootstrap-t procedure computes a cluster-robust adjusted Wald t-statistic for each bootstrap replication by resampling the clusters. By calculating the proportion of Wald statistics that is larger than the intended size, the associated p-value can correctly be estimated (Cameron & Miller, 2013; Esarey & Menger, 2015).
In addition, a number of control variables are added. These controls, that allow to estimate the regressors of interest more precisely, include household characteristics, such as gender, age, education and the number of family members. Additionally, sets of variables related to household income, household wealth and production risk. A group of variables indicating insurance awareness and financial understanding are added too. Characteristics such as distance to a main road and market as well as access to off-farm income, piped water and electricity control for regional effects.3A full overview and description of these variables is given in Table 1 in the Appendix. These variables are selected as the literature discussed above indicates that they are determinants of either credit rationing or insurance uptake.
4 Data
The previous section described how the hypotheses of this research are tested. This section details the data. I begin with an overview of the summary statistics after which I focus on the extent to which credit rationing occurs.
4.1 Description data
Data has been obtained from 1,125 households in Ethiopia. Table 2 provides key summary statistics of the data set. Additional summary statistics are provided in the Appendix in Table 2. The majority of the
3
respondents is male (84.1%), the average age is 39 years and the average household size consists of 7 members. Most respondents are sampled from kebeles in the Adami Tulu region (65.2%), while 16.7% and 18.0% of the respondents live in the Bora and Arsi Negele region respectively.
From Table 2 information on the uptake of insurance can be assessed. In total, roughly 11% of the households had purchased insurance in 2013. In 2014, this number doubled. About 7% of the households purchased insurance in both 2013 and 2014. In the regions in which the company is already active and households have access to insurance, 122 households purchased insurance in 2013, 247 households purchased insurance in 2014 and 83 households purchased insurance in both years. No households in the expansion area purchased insurance.
Table 2: Summary statistics
Variable Obs Mean Std. Dev. Min Max
sex 1125 0.841 0.366 0 1 age 1125 38.568 12.273 18 100 d1 1125 0.180 0.385 0 1 d2 1125 0.652 0.476 0 1 d3 1125 0.167 0.373 0 1 familysize 1125 7.020 2.821 1 27 ins13 1125 0.108 0.311 0 1 ins14 1125 0.220 0.414 0 1 ins1314 1125 0.074 0.262 0 1 w13 1125 0.153 0.360 0 1 w14 1125 0.331 0.471 0 1 qr 1125 0.262 0.440 0 1 rr 1125 0.047 0.212 0. 1 tcr 1125 0.012 0.111 0 1 dr 1125 0.060 0.237 0 1 cr 1125 0.322 0.467 0 1
In addition to actual uptake, the respondents’ willingness to purchase insurance is estimated. For the respondents in the treatment area, ‘willingness to purchase’ (w13) equals 1 if they did purchase insurance. For respondents in the expansion area, ‘willingness to purchase’ is set at 1 if their estimated propensity score to take up insurance in 2013 is larger than a certain threshold. This propensity score is estimated as follows. A logistic regression is ran to estimate demand for insurance purchase in the treatment area. Based on a set of observable characteristics, that are grounded on literature about insurance demand, the likelihood that a household would purchase insurance is estimated. Then, the predicted values from this regressions are applied to the expansion area. Table 3 demonstrates the results of this regression. It can be observed that knowing and trusting the insurance company and basis risk are positively related to insurance purchase. The amount of land rented and the degree of basis risk and risk expectation, on the other hand, are negatively related to insurance purchase.
18
’willingness to purchase insurance’ is set at 1. This threshold of 0.414 is chosen so that the proportion of households that is willing to purchase IBI is equal in both areas, as can be seen in Table 4.
Table 3: Propensity to purchase insurance
Variables Insurance purchase
Coefficient Robust standard error
sex -0.065 0.330 age -0.001 0.011 familysize -0.056 0.049 basisrisk -0.005** 0.002 shortrain 0.028 0.562 replanting -0.024 0.434 riskprone 0.227 0.176 riskexpectation -0.659** 0.308 accesspipedwater 0.203 0.314 knowsinsurance 1.478** 0.706 attendcampaign 1.202** 0.554 peerknowsinsurance 1.073* 0.589 education -0.068 0.139 finlit 0.238 0.162 inslit -0.004 0.086 ofincome 0.000 0.000 incomecrops 0.000 0.000 incomelivestock 0.000* 0.000 assets 0.000** 0.000 land 0.035** 0.015 landrented -0.101*** 0.034 implements 0.000*** 0.000 Constant -3.718*** 0.895 Observations 650 Pseudo R-squared 0.171
Note: This table report results from a logistic regression.
*** p<0.01, ** p<0.05, * p<0.1
Table 4: Willingness to purchase insurance by area
Treatment area Percentage Expansion area Percentage
Willing to purchase insurance 122 15.327 50 15.198
Not willing to purchase insurance 674 84.673 279 84.812
Total 796 100 329 100
4.2 Credit rationing
Using the direct elicitation methodology, data on credit constraints has been obtained for 1,125 households in Ethiopia. As described above, the DEM enables to categorize the households into different types of credit rationing.
be rejected. Of all credit rationed households, 14.6% is risk rationed while 3.9% is transaction-cost rationed. As the absolute number of risk and transaction-cost rationed households is relatively small, these two categories are put together in the subsequent analysis.
Table 5 gives further insights in the type of credit rationing. For the respondents that are quantity rationed, most respondents did not apply for a loan because they expected that they would certainly be rejected. Most of the risk rationed do not apply for a loan because they fear to put their collateral at risk. The majority of the transaction-cost rationed, lastly, do not apply because the lending procedure involves too much paperwork.
Table 5: Classification of credit rationing
Amount Percentage
Quantity rationing 295 26.222
Unsatisfied borrowers 33 2.933
Rejected applicants 7 0.622
‘Certainly rejected non-applicants’ 266 23.644
Risk rationed 53 4.711
I do not want to put my collateral at risk 43 3.822
I do not want to be worried; I am afraid 4 0.356
Formal lenders are too strict; they are not as flexible as informal ones 5 0.444
Formal lenders do not offer refinancing 1 0.089
Transaction-cost rationed 14 1.244
The branch is too far away 5 0.444
There is too much paperwork 9 0.008
Credit rationed 362 32.178
Although the survey was designed to specify in which one category the household belongs, a mistake in the administration of the survey has caused households to answer questions they were supposed to omit. This has led to some errors with the measurement, causing some households to fall in more than one category. For example, although a respondent could have answered ’yes’ to the question whether he would lend from a commercial bank if he would apply, he is nonetheless also asked the question whether he would apply if he would be certain that a commercial bank would approve his application. This has in some instances led to different stated reasons, which is the reason some respondents belong to two categories. The solution to this error is to interpret the answers is a conservative way. By going through the answers in the correct order of the survey and ignoring questions that should not have been asked, the answers can be considered as if the survey was administered in the correct way.
20
Table 6: Credit rationing by access to insurance
Type of rationing Treatment area mean Control area mean Difference P-value
Price 0.443 0.435 0.008 0.787 Credit 0.362 0.225 0.137 0.000 Risk 0.049 0.043 0.006 0.643 Transaction-cost 0.016 0.003 0.013 0.067 Quantity 0.296 0.179 0.117 0.000
5 Results
In this section, the results on the analysis of the relation between index-based insurance and credit rationing are presented. It is studied whether insurance uptake explains the differences in credit rationing between the treatment and the expansion area. Although households living in the treatment area are more likely to be credit rationed, this is not likely to be caused by access to insurance itself. Could insurance purchase explain the difference with the expansion area? I aim to control for endogenous program placement and the possibility that households self-selected into purchasing insurance by employing a difference-in-differences analysis.
5.1 Results difference-in-differences analysis
The results of this analysis are reported in Table 7A and 7B. In both tables, column 1-3 present the OLS results for credit rationing, demand-sided rationing, and supply-sided rationing respectively. Column 4-6 present the same results for logistic regressions. Demand-sided rationing includes both risk and transaction-cost rationing. Significance levels are displayed with asterisks on the basis of clustered standard errors. The values in squared brackets indicate adjusted significance levels for the bootstrapped results.
Table 7A: Insurance uptake and credit rationing
(1) (2) (3) (4) (5) (6)
Variables Credit Demand Supply Credit Demand Supply
T * Willingness to purchase insurance -0.098 0.025 -0.123 -0.537* 13.069*** -0.762* (0.084) (0.029) (0.100) (0.317) (0.880) (0.389) [0.346] [0.482] [0.310] [0.381] [0.000] [0.381] T 0.152** 0.016 0.136*** 0.767*** 0.277 0.817*** (0.053) (0.028) (0.042) (0.292) (0.491) (0.241) [0.026] [0.760] [0.006] [0.305] [0.760] [0.029] Willingness to purchase insurance 0.136** -0.054** 0.189** 0.696*** -13.632*** 1.067*** (0.062) (0.020) (0.082) (0.207) (0.755) (0.295) [0.000] [0.002] [0.000] [0.514] [0.000] [0.485] Constant 0.204*** 0.054** 0.151*** -1.360*** -2.868*** -1.730*** (0.041) (0.020) (0.021) (0.250) (0.392) (0.161) Observations 1125 1125 1125 1125 1125 1125 R-squared 0.022 0.005 0.025 Pseudo R-squared 0.018 0.016 0.022
Note: Column 1-3 report results from linear regressions; Column 4-6 report results from logistic regressions. Clustered standard
By focusing on the interaction between access to insurance and willingness to purchase insurance, the actual effect of insurance uptake on credit rationing can be estimated. This effect can more clearly be indicated once the effects of program placement and willingness to purchase insurance are controlled for. The results suggest that IBI uptakes results in demand-sided rationing. An increase in one’s willingness to purchase IBI diminishes demand-sided constraints, indicating that households that are willing to purchase insurance are more likely to apply for a loan. However, actual purchase of insurance aggravates demand-sided constraints. An increase in supply-sided rationing is mainly related to access to insurance and not to insurance uptake or one’s willingness to purchase insurance.
Table 7B: Insurance uptake and credit rationing
(1) (2) (3) (4) (5) (6)
Variables Credit Demand Supply Credit Demand Supply
22
Table 7B Continued: Insurance uptake and credit rationing
(1) (2) (3) (4) (5) (6)
Variables Credit Demand Supply Credit Demand Supply
incomecrops -0.000 0.000 -0.000 -0.000 0.000 -0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) incomelivestock -0.000 -0.000 -0.000 -0.000*** -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) ofincome 0.000* 0.000 0.000 0.000** 0.000*** 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) shortrain 0.120** 0.0124 0.108** 0.664** 0.208 0.652* (0.050) (0.043) (0.050) (0.281) (0.773) (0.361) replanting 0.022 0.012 0.011 0.101 0.171 0.057 (0.041) (0.023) (0.041) (0.212) (0.384) (0.263) riskprone 0.020 0.003 0.017 0.130 0.152 0.146 (0.026) (0.020) (0.024) (0.136) (0.348) (0.145) riskexpectation -0.089*** -0.007 -0.082*** -0.476*** -0.136 -0.524*** (0.023) (0.019) (0.028) (0.107) (0.263) (0.176) accesspipedwater -0.087* -0.006 -0.081* -0.466 0.026 -0.509* (0.050) (0.023) (0.045) (0.284) (0.439) (0.300) basisrisk 0.000 0.000 0.000 0.001 -0.004 0.002 (0.000) (0.000) (0.000) (0.002) (0.003) (0.002) implements -0.000 0.000 -0.000 -0.000 0.000 -0.000** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) distanceroad 0.000 0.000* 0.000 0.001 0.005** -0.001 (0.000) (0.000) (0.000) (0.001) (0.002) (0.001) distanceworeda 0.000 -0.000** 0.000 -0.002 -0.010** 0.000 (0.000) (0.000) (0.000) (0.002) (0.004) (0.002) distancemarket 0.000 0.000 0.000 -0.001 -0.005 0.001 (0.000) (0.000) (0.000) (0.002) (0.003) (0.002) accessofincome 0.038 0.001 0.037 0.186 0.047 0.276 (0.052) (0.028) (0.045) (0.243) (0.360) (0.235) accesselectricity 0.143** -0.0103 0.153*** 0.701*** -0.352 0.843*** (0.054) (0.034) (0.052) (0.259) (0.729) (0.275) Constant 0.427** 0.131 0.296* -0.578 -1.431 -1.373 (0.156) (0.103) (0.154) (0.801) (1.454) (0.918) Observations 922 922 922 922 922 922 R-squared 0.146 0.062 0.184 Pseudo R-squared 0.129 0.145 0.185
Note: Column 1-3 report results from linear regressions; Column 4-6 report results from logistic regressions. Clustered standard
errors in parentheses. P-values in squared brackets are generated using the wild cluster boostrap-t procedure for column 1-3 and the pairs cluster bootstrap-t for column 4-6. *** p<0.01, ** p<0.05, * p<0.1
could increase one’s literacy. It is surprising, however, that insurance literacy is positively related to demand-sided rationing. Although the coefficient is insignificant, it could indicate that a high degree of insurance literacy increases one’s understanding of risk, which in turn leads to risk rationing.
Factors related to risk, such as risk expectation and experiencing a shortage of rain, are significantly related to quantity rationing. Lenders seem less likely to lend to farmers that experience a shortage of rain while they are more likely to lend to farmers with a higher expectation of risk and farmers that use more inputs for their farm. This could signal that lenders are less willing to lend to farmers confronted with high risk, although this could be countered if the farmer is aware of the risk he faces. The farmer might show that he exercises high effort by purchasing inputs.
In sum, these findings indicate a positive relationship between insurance uptake and demand-sided rationing and no relationship between insurance uptake and supply-sided rationing. This is in contrast with both the first and the second hypothesis. The third hypothesis is confirmed, albeit in a different way than expected. In the next section, the robustness of these results is tested.
5.2 Robustness analysis
In order to ensure that insurance take up and no other factors are driving the differences in credit rationing, a number of robustness checks are performed. This section reports on a number of additional checks.
First of all, the variable ’willingness to purchase insurance’ is based on a somewhat arbitrary threshold. For the respondents in the treatment area, willingness to purchase insurance is set at one if the respondents actually purchased insurance. For the respondents in the expansion area, however, this variable was set at 1 if the estimated propensity to purchase insurance was larger or equal to 0.414. Do the results change if a higher boundary level is chosen? As a higher threshold for setting one’s willingness to purchase equal to one is more intuitive, I set the level at 0.5. Regression analyses demonstrate that this alters the results.
Table 3A and 3B in the Appendix demonstrate the results for a threshold of 0.50. Although Table 3A is relatively similar to Table 7A, Table 3B indicates that IBI purchase could not only affect demand-sided constraints, but also aggravate quantity constraints. Actual IBI uptake is associated with a large and significant decrease in both supply and demand constraints. As the threshold for ‘willingness to purchase insurance’ is debatable, this effect on quantity rationing is to be considered too. Section 6 provides a potential explanation for this finding.
24
way in which IBI uptake affects credit rationing. For this reason, the results of propensity score matching are added to the Appendix. The results of this analysis suggest that insurance uptake could have a small and negative effect on demand-sided rationing, and a positive effect on supply-sided rationing. However, as these results are limited to the respondents in the treatment area only, these results should be interpreted with caution.
Third, the results for insurance uptake and willingness to purchase insurance in 2014 could be different from the results over 2013. Such a difference could indicate that the findings of 2013 are unlikely to last. Indeed, the effect of insurance uptake in 2014 is no longer significantly related to quantity rationing. The effect of insurance uptake on risk and transaction-cost rationing, however, is positive and significant. Although beyond the scope of this study, how exactly insurance uptake alter credit constraints over time is worth investigating.
Lastly, it is analyzed whether outliers drive the differences in credit rationing. For this reason, variables with outliers are trimmed at 1, 5 and 10% from the top. These include variables related to income, land, assets, implements, basis risk as well as distance to the market, main road and district office. Including these trimmed variables in the regression analysis does not alter the results of the key independent variables, whereas it does emphasize the effects of other variables.
In total, the evidence for the relation between IBI purchase and supply-sided rationing is mixed, while the results do indicate a positive relationship between IBI purchase and demand-sided rationing. Purchasing IBI could thus aggravate demand constraints, whereas the effect on supply-sided constraints is ambiguous. These results might alter over time, but are not likely to be caused by outliers in the data.
6 Discussion
It has been demonstrated that insurance uptake might be strongly related to credit rationing. Insurance could increase both demand and supply constraints. This section seeks to elaborate on these findings. First, I provide a number of reasons for the increase in demand rationing. Then, I discuss why insurance uptake could lead to an increase in supply-sided rationing.
6.1 Demand-sided rationing
borrowing (2014). However, as Table 7B indicates that one’s risk expectation is not related to demand constraints, this explanation in unlikely to hold.
Second, if a potential borrower incorrectly assumes that borrowing is a requirement for purchasing insurance, borrowing rates might decline (Grosh & McKenzie, 2014). This reason is not capable of explaining the findings of this paper either, as insurance and credit are not linked to each other because they are offered by separate institutions.
Thirdly, if insurance could serve as a substitute of credit, purchasing insurance could diminish one’s demand for credit. Insurance then functions as an alternative buffer against shocks. Prior to a shock, insurance could restrain farmers from precautionary borrowing as the insurance product hedges some of the risk. After a shock occurs, the indemnity provided could withhold farmers from opting for a loan.
6.2 Supply-sided rationing
In addition to an increase in demand-sided rationing resulting from IBI uptake, there could be instances in which IBI uptake aggravates quantity constraints as well. Here I provide an explanation for this finding by elaborating on Banerjee (2000), Clarke and Dercon (2009), and Farrin and Miranda (2013).
Banerjee (2000) holds that credit constraints are larger for the poor than the rich, as the rich can invest part of their own wealth in the project while the poor have no wealth to invest. If one finances a large part of the project by himself, strategic default is unlikely. For the poor, however, strategic default can be a favorable option. As the lender has difficulties distinguishing between default as a result of bad luck and strategic default and the borrower is only caught with some probability, the poor can gain much from strategic default. If one is caught, the lender punishes the borrower by forcing repayment until the point the borrower reaches his minimum welfare level. By definition, this minimum welfare level is the wealth the lender cannot confiscate. With an increase in this minimum welfare, strategic default becomes less painful.
26
In addition, insurance could lead to a higher probability of default as the borrower takes more risk (moral hazard). For example, if production risk is hedged through an insurance product, farmers could decide to take on more risk in other activities. In this way, IBI uptake could encourage risk taking behavior, which subsequently limits the possibilities to borrow. However, insurance products that limit moral hazard, such as rainfall insurance and other types of index-based insurance, should limit the probability of default simultaneously.
7 Conclusion
In this thesis, the effect of index-based insurance on credit rationing is assessed. This paper is a first attempt to study the relation between index-based insurance and credit rationing empirically. A novel methodology is used to elicit the type of credit rationing that occurs among both borrowers and non-borrowers. In this way, it is found that almost 6% of the respondents is demand-sided constrained, whereas more than 26% of the respondents is supply-sided constrained.
More specifically, this paper tests the hypotheses that index-based insurance leads to a decrease in risk and transaction-cost rationing and an increase in quantity rationing. Some evidence for a negative effect of IBI on supply-sided rationing could be detected. Contrary to expectations, a positive impact of IBI uptake on sided rationing is observed. Instead of alleviating demand constraints, IBI could aggravate demand-sided rationing if it serves as a substitute for credit.
In addition to examining the degree to which insurance and credit are substitutes, future research could focus on a number of alternative issues. First of all, this study is limited to formal financial services only. In order to fully understand the relation between insurance and credit rationing, informal services should be taken into account too. Second, in this paper the severity of credit constraints are not assessed. Analyzing the degree of credit rationing that results from the uptake of IBI could shed more light on this issue. Thirdly, it could be investigated how insurance uptake affects credit constraints over time. Is there a learning-effect among lenders or borrowers? Future research might help understand how credit constraints can be resolved to spur economic growth and development.
References
Ali, D. A., Deininger, K., and Duponchel, M. (2014). Credit Constraints and Agricultural Productivity: Evidence from Rural Rwanda. Journal of Development Studies, 50(5): 649–665.
Armendáriz, B., and Morduch, J. (2005). The Economics of Microfinance. The MIT Press, Cambridge. Baez, J., Fuente, A. D., and Santos, I. (2010). Do Natural Disasters Affect Human Capital? An Assessment
Based on Existing Empirical Evidence. IZA, Discussion Paper No. 5163, Bonn.
Banerjee, A. (2000). The Two Poverties. Nordic Journal of Political Economy, 26(2): 129-141.
Barslund, M. (2007). Microeconometric Applications in Development Economics. PhD thesis, University of Copenhagen.
Berhane, G., Clarke, D., Dercon, S., Hill, R. V., & Taffesse, A. S. (2013). Insuring Against the Weather. International Food Policy Research Institute, Washington, DC.
Binswanger-Mkhize, H. P. (2012). Is There Too Much Hype about Index-based Agricultural Insurance?
Journal of Development Studies, 48(2): 187–200.
Boucher, S. R., Guirkinger, C., and Trivelli, C. (2009). Direct Elicitation of Credit Constraints: Conceptual and Practical Issues with an Application to Peruvian Agriculture. Economic Development and
Cultural Change, 57(4): 609-640.
Caliendo, M., & Kopeinig, S. (2008). Some Practical Guidance for the Implementation of Propensity Score Matching. Journal of Economic Surveys, 22(1): 31-72.
Cameron, A. C., Gelbach, J. B., and Miller, D. L. (2008). Bootstrap-Based Improvements for Inference with Clustered Errors. Review of Economics and Statistics, 90(3): 414–427.
Cameron, A. C. and Miller, D. L. (2013). A Practitioner’s Guide to Cluster-Robust Inference. Journal of
Human Resources, 50(2): 317-372.
Carter, M. R., Cheng, L., and Sarris, A. (2011). The Impact of Interlinked Index Insurance and
Credit Contracts on Financial Market Deepening and Small Farm Productivity. In Annual Meeting of
the American Applied Economics Association, Pittsburgh.
Carter, M. R., Galarza, F., and Boucher, S. (2007). Underwriting Area-based Yield Insurance to Crowd-in Credit Supply and Demand. Savings and Development, 31(3): 335–360.
Carter, M. and Olinto, P. (2003). Getting Institutions ‘Right’ for Whom? Credit Constraints and the Impact of Property Rights on the Quantity and Composition of Investment. American Journal of
Agricultural Economics, 85(1): 173–186.
Chantarat, S., Mude, A. G., & Barrett, C. B. (2009). Willingness to Pay for Index Based Livestock Insurance:
Results from a Field Experiment in northern Kenya. Cornell University, Ithaca.
Cheng, L. (2014). The Impact of Index Insurance on Borrower’s Moral Hazard Behavior in Rural Credit
Markets. University of California, Davis.
Clarke, D. J. (2011). A Theory of Rational Demand for Index Insurance. University of Oxford.
28
Clarke, D. and Kalani, G. (2011). Microinsurance Decisions: Evidence from Ethiopia. International Labour Office and Microinsurance Innovation Facility, Geneva.
Cole, S., Stein, D., & Tobacman, J. (2011). What is Rainfall Index Insurance Worth? A Comparison of
Valuation Techniques. Harvard University Business School. Working Paper.
Coleman, B. E. (1999). The Impact of Group Lending in Northeast Thailand. Journal of Development
Economics, 60(1): 105–141.
De Nicola, F., & Hill, R. V. (2012). Interplay Among Credit, Weather Insurance and Savings for Farmers in Ethiopia. In The 2013 North American Winter Meeting of the Econometric Society, San Diego. Dercon, S., Hill, R. V., Clarke, D., Outes-Leon, I., and Taffesse, A. (2014). Offering Rainfall Insurance to
Informal Insurance Groups: Evidence from a Field Experiment in Ethiopia. Journal of Development
Economics, 106(0): 132-143.
Elabed, G., & Carter, M. (2014). Ex-ante Impacts of Agricultural Insurance: Evidence from a Field
Experiment in Mali. University of California, Davis.
Esarey, J. and Menger, A. (2015). Practical and Effective Approaches to Dealing with Clustered Data. Rice University, Houston.
Farrin, K. and Miranda, M. J. (2013). Premium Benefits? A Heterogeneous Agent Model of Credit-Linked Index Insurance and Farm Technology Adoption. In Annual Meeting of the Agricultural and Applied
Economics Association, Washington, DC.
Foltz, J. D. (2004). Credit Market Access and Profitability in Tunisian Agriculture. Agricultural Economics,
30(3): 229–240.
Freedman, P. L. and Click, R. W. (2006). Banks That Don’t Lend? Unlocking Credit to Spur Growth in Developing Countries. Development Policy Review, 24(3): 279–302.
Gaurav, S., Cole, S., & Tobacman, J. (2011). Marketing Complex Financial Products in Emerging Markets: Evidence from Rainfall Insurance in India. Journal of Marketing Research, 48(SPL): S150-S162. Giné, X., Menand, L., Townsend, R. M., & Vickery, J. I. (2012). Microinsurance: A Case Study of the
Indian Rainfall Index Insurance Market. World Bank Working Paper. In Chetan. G. (ed.):
Handbook of the Indian Economy, Oxford University Press.
Giné, X. and Yang, D. (2009). Insurance, Credit, and Technology Adoption: Field Experimental Evidence from Malawi. Journal of Development Economics, 89(1): 1-11.
Groh, M., and McKenzie, D. (2015). Macroinsurance for Microenterprises: A Randomized Experiment in
Post-Revolution Egypt. World Bank Working Paper No. 7048.
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.
Hill, R. V., Robles, M., & Ceballos, F. (2013). Demand for Weather Hedges in India: An Empirical
Exploration of Theoretical Predictions. International Food Policy Research Institute, Washington,
DC.
Jaffee, D. M. and Modigliani, F. (1969). A Theory and Test of Credit Rationing. The American Economic
Janzen, S. A. and Carter, M. R. (2013). The Impact of Microinsurance on Asset Accumulation and Human
Capital investments: Evidence from a Drought in Kenya. International Labour Office and
Microinsurance Innovation Facility, Geneva.
Jia, X., Heidhues, F., and Zeller, M. (2010). Credit Rationing of Rural Households in China. Agricultural
Finance Review, 70(1): 37–54.
Karlan, D., Kutsoati, E., McMillan, M., and Udry, C. (2011). Crop Price Indemnified Loans for Farmers: A Pilot Experiment in Rural Ghana. Journal of Risk and Insurance, 78(1): 37-55.
Karlan, D., Osei, R., Osei-Akoto, I., and Udry, C. (2014). Agricultural Decisions After Relaxing Credit Constraints. Quarterly Journal of Economics, 129(2): 597–652.
Khantachavana, S. V., Verteramo Chiu, L. J., Turvey, C. G., and Kong, R. (2012). Risk Rationing and the
Demand for Agricultural Credit. SSRN Working Paper No. 2191297.
Kochar, A. (1997). An Empirical Investigation of Rationing Constraints in Rural Credit Markets in India.
Journal of Development Economics, 53(2): 339–371.
Mahul, O. and Skees, J. (2007). Managing Agricultural Risk at the Country Level: The Case of Index-Based
Livestock Insurance in Mongolia. World Bank Working Paper No. 4325.
Malapit, H. J. L. (2012). Are Women More Likely to Be Credit Constrained? Evidence from Low-income Urban Households in the Philippines. Feminist Economics, 18(3): 1–34.
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.
Mishra, P. K. (1994). Crop Insurance and Crop Credit: Impact of the Comprehensive Crop Insurance Scheme on Cooperative Credit in Gujarat. Journal of International Development, 6(5): 529–567. Petersen, M. A. (2009). Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches.
Review of Financial Studies, 22(1): 435-480.
Petrick, M. (2004). A Microeconometric Analysis of Credit Rationing in the Polish Farm Sector.
European Review of Agricultural Economics, 31(1): 77 –101.
Rosenzweig, M. R. and Wolpin, K. I. (1993). Credit Market Constraints, Consumption Smoothing, and the Accumulation of Durable Production Assets in Low-Income Countries: Investments in Bullocks in India. Journal of Political Economy, 101(2): 223-244.
Santos, P. and Barrett, C. B. (2006). Informal Insurance in the Presence of Poverty Traps: Evidence from
Southern Ethiopia. SSRN Working Paper No. 998541.
Steijvers, T. and Voordeckers, W. (2009). Collateral and Credit Rationing: A Review of Recent Empirical Studies as a Guide for Future Research. Journal of Economic Surveys, 23(5): 924–946.
Stiglitz, J. E. and Weiss, A. (1981). Credit Rationing in Markets with Imperfect Information. The American
Economic Review, 71(3): 393–410.
Verbeek, M. (2012). A Guide to Modern Econometrics. John Wiley & Sons, Chichester.
30
Appendix
Table 1: Description variables
Variable Description Unit
sex Gender respondent 1=male
age Age respondent age in years
d1 Arsi Negele district 1=resident
d2 Adami Tulu district 1=resident
d3 Bora district 1=resident
family size Family size of the household Number of family members education Level of education of the
respondent
1=no education, 2=grade 1-3, 3=grade 4-8, 4=grade 9-12, 5=diploma/certificate, 6=degree or above
ins13 Whether the household purchased insurance in 2013
1=yes ins14 Whether the household purchased
insurance in 2014
1=yes ins1314 Whether the household purchased
insurance in both 2013 and 2014
1=yes w13 Whether the household is willing
to purchase insurance in 2013
1=yes w14 Whether the household is willing
to purchase insurance in 2014
1=yes qr Whether the household is quantity
rationed
1=yes rr Whether the household is risk
rationed
1=yes tcr Whether the household is
transaction-cost rationed
1=yes dr Whether the household is demand
rationed
1=yes cr Whether the household is credit
rationed
1=yes basisrisk Household’s distance to weather
station
Walking distance in minutes inslit Insurance literacy Number of correct answers finlit Financial literacy Number of correct answers attendcampaign Whether the respondent
participated in the marketing campaign
1=yes
peerknowsinsurance Whether the household has friends, relatives or
neighbors who bought insurance 1=yes knowsinsurance Whether the household knows
and trusts JICA/OIC
1=yes incomecrops Total value of sold crops Value in ETB ofincome Total value of off-farm income Value in ETB incomelivestock Total value of livestock sales Value in ETB applied Whether the household applied
for a loan in the past five years
1=yes accepted Whether the household’s
application was accepted
1=yes largerloan Whether the household would
have wanted a larger loan at the same interest rate
1=yes
canborrow Whether a bank would lend to the household if it would apply
