Smallholder preferences and willingness-to-pay measures for microcredit: Evidence from Sichuan province in China

Zhao Ding (Department of Food Economics and Consumption Studies, Christian Albrechts Universitat zu Kiel, Kiel, Germany)
Awudu Abdulai (Department of Food Economics and Consumption Studies, Christian Albrechts Universitat zu Kiel, Kiel, Germany)

China Agricultural Economic Review

ISSN: 1756-137X

Publication date: 3 September 2018

pdf (201 KB) Article view Figure view Cited (12) cite article

Abstract

Purpose

The purpose of this paper is to examine smallholders’ preferences and willingness to pay for microcredit products with varying attribute combinations, in order to contribute to the debate on the optimal design of rural microcredit.

Design/methodology/approach

Data used in this study are based on a discrete choice experiment from 552 randomly selected respondents. Mixed logit and latent class models are estimated to examine the choice probability and sources of preference heterogeneity. Endogenous attribute attendance models are applied to account for attribute non-attendance (ANA) phenomenon, focusing on separate non-attendance probability as well as joint non-attendance probability.

Findings

The results demonstrate that preference heterogeneity and ANA exist in the smallholder farmers’ microcredit choices. Averagely, smallholder farmers prefer longer credit period, smaller credit size, lower transaction costs and lower interest rate. Guarantor collateral method and installment repayment positively affect their preferences as well. Moreover, respondents are found to be willing to pay more for the attributes they consider important. The microcredit providers are able to attract new customers under the current interest rates, if the combination of attributes is appropriately adjusted.

Originality/value

This study contributes to the debate by assessing the preference trade-off of different microcredit attributes more comprehensively than in previous analyses, by taking preference heterogeneity and ANA into account.

Keywords

Citation

Ding, Z. and Abdulai, A. (2018), "Smallholder preferences and willingness-to-pay measures for microcredit", China Agricultural Economic Review, Vol. 10 No. 3, pp. 462-481. https://doi.org/10.1108/CAER-02-2017-0022

Download as .RIS

Publisher

:

Emerald Publishing Limited

Copyright © 2018, Emerald Publishing Limited

1. Introduction

Microcredit services have been expanding significantly in developing economies for several years. It aims at providing small amounts of capital to poor borrowers who have been suffering from the shortage of financial services, to enable them generate higher incomes by investing in productive activities. This approach has been regarded as an efficient pathway for poverty reduction. The literature places little emphasis on the reason why borrowers choose a certain microcredit option. However, the optimal combination of microcredit attributes for attracting the poor is strongly debated, since some combinations of the attributes constrain participation among smallholder farmers (e.g. Madajewicz, 2011; Giné and Karlan, 2014; Cheng and Ahmed, 2014). Therefore, one important issue that needs clarification is how much smallholders are willing to pay for microcredit, given the major attributes.

Participation in microcredit has been found to exert positive and significant impacts on household income and welfare in many developing countries (e.g. Ahlin and Jiang, 2008; Berhane and Gardebroek, 2011; Imai et al., 2012; Mazumder and Lu, 2014; Bruhn and Love, 2014; Maria, 2016; Ksoll et al., 2016). These findings encouraged many microcredit organizations to be mission-oriented or supply-oriented, without paying much attention to the willingness to pay by poor borrowers. However, since microcredit has been practiced for more than two decades, emphasis needs to move from mission-oriented to demand-oriented. Evidence shows that the characteristics of demand for financial services tend to influence the type of financial services and the achievement of their social and profit objective (Ritchie, 2007). A well-functioning set of credit attributes should be tailored to potential borrowers’ needs, but also considering the profitability for lenders (Tsukada et al., 2010).

Generally, the main microcredit attributes that have been used to explain the preferences of poor households are interest rate, repayment schedule, loan size and collateral method (Boucher et al., 2008; Tsukada et al., 2010; Sagamba et al., 2013; Kong et al., 2015). Interest rate, which is the price of financial service, has received the most attention (Janvry et al., 2010; Rashid et al., 2011; Khandker and Koolwal, 2016; Tan and Lin, 2016). The consensus has been that low interest rate increases lending to the poor, especially the rural poor, without lowering profits for financial intermediaries (Hermes et al., 2011; Angelucci et al., 2015). Hence, theoretically, interest rates should be set at profit-making levels, based on the notion that even poor customers favor access to credit with low interest rates (Dehejia et al., 2012). However, interest rate is not the only factor that appeals to poor customers, since some farmers are willing to obtain more credit, even at higher interest rates (Turvey et al., 2012).

Financial decisions involve complexities that individuals frequently have difficulty in understanding depending on their education, information, experiences, assets and social networks (Yesuf and Bluffstone, 2009; Cai et al., 2015). Individuals with different financial habits might prefer different types of contracts. Alternatively, lenders with different levels of sophistication may attract different client mixes and offer different contracts. Preferences for formal or informal loans, group or individual loans, and even no loans vary as well (Tsukada et al., 2010; Ayyagari et al., 2010; Attanasio et al. 2015). However, very little is known about the optimal contract structure of credit loans.

Several studies have used revealed preference method to analyze households’ preferences for microcredit (Tsukada et al., 2010; Dehejia et al., 2012; Lønborg and Rasmussen, 2014). This method is typically used for the decisions on actual alternatives. However, stated preference approach enables us to examine hypothetical choices, or ex ante strategies that allow an analysis of decision making at an early stage of the policy cycle. Very few studies have used experimental and stated preference methods to analyze the behavioral aspects of microcredit (Field et al., 2011; Sagamba et al., 2013; Bauer et al., 2012; Weber et al., 2014). In their recent study, Field et al. (2011) found that grace period repayment schedule does not contribute to decrease default and delinquency. However, Weber et al. (2014) found that farmers with flex loans and without grace periods show significantly higher delinquencies. Bauer et al. (2012) noted the significance of the structure of microfinance loans, although they did not establish the specific causal links in the structure. In the investigation of preferences of microcredit providers, using a choice experiment, Sagamba et al. (2013) found that the main determinant of microcredit for officers is the quality of applicant’s project.

Our study contributes to the literature by using a stated preference approach to examine the attributes that affect farmers’ preferences and willingness-to-pay for microcredit. Specifically, we utilize data from a survey of 552 smallholder farmers in the Sichuan province of China. The attributes considered include interest rate, credit period, loan size, collateral method, repayment schedule and transaction costs. The present study differs from the previous studies in terms of assessing the preference trade-off of different microcredit attributes more comprehensively. We employ a mixed logit model to analyze the choice probability and the existence of preference heterogeneity, and a latent class (LC) model to examine the sources of preference heterogeneity by segregating smallholders into groups with similar characteristics. We also use an endogenous attribute attendance (EAA) model to capture attribute non-attendance (ANA) phenomenon.

2. Background

Rural credit system is an important element in the financial system in China, and has been an integral part of China’s economic reform. China’s rural credit system faced serious difficulties around 2005, due to its incoherent structure, weak management and poor internal capabilities (Herd et al., 2010). As part of developing microcredit in rural areas, the Chinese Government launched many measures to strengthen rural finance in 2006. These measures were partly meant to keep with the new challenges from joining the WTO, which required the permission of foreign banks to develop their banking services by the end of 2006.

Besides strengthening and reforming the existing rural financial institutions such as Agricultural Development Bank of China, Agricultural Bank of China and Rural Credit Cooperatives, China established Postal Savings Bank of China and other three types of rural banking institutions to improve and consolidate its rural financial markets. These three rural banking institutions are village and township banks, loan companies and rural fund cooperatives. In addition, in order to relax rural credit constraints for the poor and implement microcredit for purposes of poverty alleviation, China set up Poverty Village Mutual Aid Funds[1] in depressed areas where commercial banks have no interests. This program is designed to help farmers without access to credit sources from formal financial institutions and informal lenders, by providing small and short-term credits. It works as a way of public financial support, that is, associated with smallholders’ participation to enhance the poor’s access to formal credit, and has similar features as the revolving loan funds. In China, it seems that formal credits are pushed to cover most of smallholders with different income levels. But the poor are still often excluded from formal credits (Shoji et al., 2012; Yuan and Xu, 2015). The informal credit market has been supplementary to the services provided by formal markets (Cheng and Ahmed, 2014), which indicates that the credit supplies do not meet smallholders’ demands. The development and expansion of some NGO programs then stood up as a substitution for institutional lenders and informal financial networks in the rural credit market (Xiang et al., 2014).

3. Conceptual framework

3.1 Theoretical model

The conceptual model used in the analysis is based on Lancaster’s model of consumer choice and random utility theory, indicating consumers’ utility maximization. In this decision-making process, we assume smallholders are risk neutral, and credit attributes are the targeted elements dominating smallholders’ choice behavior. Smallholders are therefore assumed to choose the microcredit option that provides maximum utility.

To identify smallholders’ heterogeneous preferences for microcredit, we conduct a discrete choice experiment. In each choice set, respondents are asked to choose the most satisfactory one from distinctive options. Each option contains six attributes with different levels. These attributes are credit period, interest rate, loan size, collateral, repayment method and transaction costs. When facing a varying combination of attributes, it is preferable to test smallholders’ attitudes on attributes under different conditions.

In this framework, an individual chooses a credit alternative based on the highest utility expectation on numbers of given choice situations. It can be expressed that an individual (n) derives utility (U) from choosing an alternative (i):

(1) U n i = V n i + μ n i ,
where Vni is the deterministic component, and depends on the attributes of alternatives; μni the stochastic error term.

The probability (P) that alternative (i) will be chosen is given as:

(2) P n i = Prob ( Y = 1 | V n i + μ n i ⩾ V n j + μ n j ; ∀ i ≠ j , ∀ j ∈ C ) ,
where Y is the alternative variable, taking the value 1 when alternative (i) is chosen and 0 otherwise; j indicates another alternative; C is the finite choice set.

In line with Maddala (1983) and Train (2009), the logit model is obtained by assuming that each μ is an observed random term that is independently and identically distributed type I extreme value (Gumbel distribution). Then a succinct and closed form is given as follows:

(3) P n i = e V n i ∑ j e V n j .

3.2 Empirical specification

In this study, we assume smallholders’ preferences are heterogeneous. We use random parameter logit (RPL) model to calculate the choice probability, which, in comparison with traditional conditional logit, can detect unobserved and observed sources of heterogeneity, and also allow random preference variation (Ortega et al., 2014). It is superior to the conditional logit in terms of overall fitness and welfare estimates (Just and Gabrielyan, 2016).

In the RPL model, the deterministic component Vni takes the form Vnit=βn·Xnit, where β is a vector of random parameters and represents individual-specific tastes; X is a vector of attributes.

In line with Train (2009) and Ortega et al. (2014), the probability of the standard logit that individual (n) chooses alternative (i) from choice set (C) in situation (t) is the integral of conditional probability, which is given by:

(4) P n i t = ∫ e β n ⋅ X n i t ∑ j e β n ⋅ X n j t f ( β ) d β ,
where f(β) is the distribution function for random parameters, with its own mean and variance; the coefficient vector consists of parameters associated with individual (n), representing the individual’s preference. This model relaxes the limitation of traditional conditional logit model by allowing random preference variation within a sample according to a specified distribution, and not sensitive to the independence of irrelevant alternatives condition (Train, 2009; Ortega et al., 2014).

We then estimate an LC model to segregate smallholders into groups with similar characteristics, in order to account for heterogeneity by creating classes. The LC model is able to provide a different dimension for describing data, where farmers are expected to have different motivations and aims for their choice decisions, and as such potentially belong to discrete groups, based on their preferences and latent variables (Bello and Abdulai, 2016b). It is therefore more suitable for examining the sources of preference heterogeneity (Xie et al., 2016). RPL and LC models both relax the assumption of homogeneity from a conditional logit model, but RPL accounts for heterogeneity in the estimation, LC accounts for it by creating classes.

More recent studies show that respondents in discrete choice experiment often ignore some attributes in their decision-making processes, termed ANA (Hensher et al., 2005; Hole, 2011; Ortega and Ward, 2016; Bello and Abdulai, 2016a), resulting in biased model outputs such as masked sensitives, implausibly assigned random parameter coefficients, and over-stated taste heterogeneity. Two approaches have been proposed to account for ANA in empirical analysis. These include stated ANA and inferred ANA. Stated ANA is an experimental approach, accounting for ANA by asking respondents specific follow-up questions on which attribute was ignored when making a decision (e.g. Scarpa et al., 2010). Inferred ANA is an econometric approach. Inferred ANA provides a better model fit, while stated ANA is not consistent.

Because, first, the situation that respondents may assign low importance to some attributes which might be ignored at first will lead to overestimation (Weller et al., 2014). Besides, incorporated responses to the non-attendance questions may cause potential problems of endogeneity bias (Scarpa et al., 2012; Hole et al., 2013). LC and EAA are widely used econometric models to account for inferred ANA models (Hole, 2011; Hensher and Greene, 2010; Hensher et al., 2012; Scarpa et al., 2012). EAA can be viewed as a variant of the equality-constrained LC model, and it can comprise all possible attribute subsets and handle all combinations of ANA in comparison with conventional LC model (Hole, 2011). In this study, we apply five EAA models focusing on the variables with lower non-attendance probabilities step by step, to account for ANA.

In the LC model, β is discrete due to different β in distinct class (s) (s=1, …, S). The probability that individual (n) selects alternative (i) from choice set (C) in a given situation (t) in class (s) can be written as:

(5) P n i t = ∑ s = 1 S R n s e β s ⋅ X n i t ∑ j e β s ⋅ X n j t ,
where βs is the special parameter for class (s), and Rns is the probability that individual (n) falls into class (s) (Broch and Vedel, 2012; Ortega et al., 2014). Accordingly, Rns can be expressed as:
(6) R n s = e λ s δ n ∑ k = 1 S e λ k δ n ,
where λs(s=1, 2, …, S) is a vector of class-specific parameters to be estimated; and δn the vector of smallholder characteristics.

In the EAA model, each choice is considered as a two-step process in which the decision-maker first decides which attributes to take into account when comparing the available alternatives, and second, chooses the alternative with the best characteristics, given his or her preferences (Hole, 2011). Thus, the basic conditional logit of EAA is given as:

(7) P n i t | C k = e ∑ k ∈ C k β n k ⋅ X n i t k ∑ i = 1 I e ∑ k ∈ C k β n k ⋅ X n i t k ,
where X n i t k represents individual (n) chooses the value of attribute (k) relating to alternative (i) from choice attribute subset (Ck) on choice situation(t); β n k individual-specific coefficient for attribute (k).

As in Hole et al. (2013), the probability that individual (n) takes attribute (k) into account is specified as e γ k z n k / 1 + e γ k z n k , where z is a vector of individual characteristics and γ is a vector of parameters to be estimated. Assuming these probabilities are independent over attributes, the probability of choosing attribute subset (Ck) is given by:

(8) P n C k = ∏ k ∈ C k e γ k z n k 1 + e γ k z n k ∏ k ∉ C k 1 1 + e γ k z n k .

The probability that an individual (n) chooses alternative (i) from choice set (C) in a given situation (t) through choosing attribute subset (Ck) can be written as:

(9) P n i t E A A = ∑ k = 1 K P n C k ∏ t = 1 T ∏ i = 1 I ( P n i t | C k ) Y n i t ,
where Ynit takes the value 1 when alternative (i) is chosen and, 0 otherwise; f(βn|θ) denotes the density for βn in which θ is the parameter of distribution.

4. Survey design and data description

The framework for understanding smallholders’ heterogeneous preferences for microcredit was implemented in China. Questionnaires were filled via face-to-face household interview, and conducted between October and December 2015 in Sichuan province, which is one of the major agricultural production provinces in China. Several types of agricultural products and a distinctive economic status coupled with different pilot projects on microcredits make this province a reasonable area for studying heterogeneous preferences.

Data were drawn from six regions according to a multistage random sampling approach, in which the mentioned characteristics of the field were purposively taken into account. The regions included Ya’an, Guangyuan, Guang’an, Nanchong, Mianyang and Leshan. The percentage of respondents in each region, which were sampled in relation to population size, are 15.94, 18.30, 17.94, 17.75, 18.48 and 11.59 percent, respectively. Questionnaires were administered to 552 randomly selected respondents in 27 villages. Our survey focused on three areas of variables: households’ social demographic data, choice experiment and follow-up questions on attributes.

Table I presents the descriptive statistics of the variables used in the analysis. According to the figures, the vast majority of household heads in the sample are males. Average farm size per household is around 3.388 mu (1 mu=1/5 hectare). Education is captured by the educational level of household heads, since they are the decision makers in a family. Family economic situation is measured by annual total income and annual liquid balance. Net returns is measured as the difference between the agricultural income and per unit costs of inputs, including seeds, chemical fertilizer, pesticides, herbicides, plastic sheets, hired labor costs and equipment rents. As can be seen from Table I, the average agricultural net income was 2.086 thousand yuan. Credit history indicates that most respondents had loans within the last five years. Distance to the nearest financial institution is about 3.6 km, indicating that microfinance institutions are widely distributed. Demand for loans shows that around 38.6 percent of respondents needed to borrow for some reasons at the time of field survey. Loan purpose for agricultural investment and for non-agricultural investment is captured differently. The former captures the productive investments in agriculture, such as buying machinery, livestock and expanding the scale of production, the latter indicates the other productive investments besides agriculture, such as running business. The mean value of village mutual aid funds indicates that a large majority of the sampled households live in less depressed villages. More smallholders prefer there credit from informal financial institutions such as relatives and friends.

Choice experiment in this study provided information on how smallholders value the characteristics of microcredit and their willingness to pay for its attributes. Currently, various microcredit services are provided to smallholders by formal institutions and informal individuals in China. Different traits such as interest rate, loan size, collateral method and repayment schedule are designed in each product, aimed at seeking their own utility maximization. These factors are the main determinants of rural microloan disbursements (Tsukada et al., 2010; Turvey et al., 2010; Sagamba et al., 2013). In our CE, the credit attributes we consider are six vital loaning components: credit period, interest rate, loan size, collateral method, repayment schedule and transaction costs.

Levels were given on the basis of existing and associated microcredit regulations. Attributes and their levels are shown in Table II. Specifically, interest rate was given level at 0 percent for the cases of post-disaster reconstruction loans and informal individual loans. (For example, farmers were able to obtain a three-year interest-free loan after the Sichuan earthquake in 2008, and most of the loans coming from relatives and friends have no interest (Cheng and Ahmed, 2014). The 5.1 and 8.25 percent were the lower and upper limits of loan interests set by the People’s Bank of China in May 2015. In total, 8 percent is the guidance interest rate of the Poverty Village Mutual Aid Funds, and it is fixed during the credit period. Transaction cost is included to capture the costs involved in the credit transactions. In particular, asymmetric information and lower educational level result in high research costs, negotiation costs, as well as monitoring costs in the transacting process. These costs, which are normally higher for rural residents, compared to their urban counterparts, tend to influence the decisions of farmers.

A full-factorial design for our CE would require 576 profiles, which would be too large for a survey to handle. Therefore, D-optimal and blocked design were used via JMP 10 (SAS). Given 2nd interactions and powers, 72 set was calculated and could be composed as 3 blocks of 24 sets (Table AI). This method can scientifically narrow down choice sets within a reasonable scale. Even then, some attributes might be the same for alternative options in one choice set, varying combinations of different attributes are able to effectively capture respondents’ preferences. Each respondent saw only one of the randomly assigned blocks with eight choice situations instead of the entire design. So three versions of the questionnaire were used, and total respondents answered 13,248 choice sets. Respondents made decisions among three alternatives in a choice situation. A sample choice situation that consisted of three alternatives is presented in Table III.

5. Empirical results

5.1 RPL estimates

The results of the RPL model estimations are presented in Table IV, where we use first level as base for each attribute. All the attributes except the loan size 2 at 50,000 yuan are significant. According to the p-value, the model is statistically significant. ASC is an alternative-specific constant, defined as a situation with 0 percent interest rate option that equals to 1, otherwise 0. According to the mean value, on average, smallholders prefer longer credit period, smaller loan size, lower transaction costs and lower interest rate. Guarantor and installment would be more attractive collateral and repayment method, respectively. The statistical significance of the standard deviations for credit period, collateral method, repayment schedule and interest rate indicate preference heterogeneity for these attributes.

From the magnitudes of the standard deviation related to the mean coefficients, 70 percent of smallholders prefer microcredit with interest rates. In terms of the two credit periods, the mean value indicates that smallholders prefer longer credit period in comparison with the base option, with 78.1 percent of them preferring the credit period of three years, while only 11.2 percent prefer five years. Although smaller credit size is more attractive for smallholders, the loan size of 50 and 100,000 yuan satisfy them all. The largest amount here only attracts 12.6 percent smallholders. The difference exhibiting on the collateral method and repayment schedule shows that only 35 percent of respondents prefer to find pledge of assets as the way to guarantee their credits, 19 percent of them prefer lump sum as the repayment way. Even transaction costs and interest rate indicate that the lower the better, 5.4 and 0.2 percent of respondents accept higher transaction costs, and prefer to pay higher interest rates, respectively.

5.2 LC estimates

Four LCs were obtained through comparison, using the Akaike information criterion (AIC) and Bayesian information criterion (BIC). According to the results presented in Table V, AIC and BIC are minimized with six and four classes, respectively. Changes in AIC is smaller from class 5 to class 6 than from class 3 to class 4, indicating that adding another class probably does not improve the model markedly. Our analysis therefore focuses on the four class specification to economize on space. The four class LC model is substantially better than RPL model with regards to the goodness of model fit. This indicates that different customer groups take on different preferences for attributes, and also verifies the necessity of using the LC model.

Table VI presents the results of the LC model. It shows that four classes of the total samples account for 5.8, 29.3, 18.5 and 46.4 percent, respectively. The coefficients of ASC and interest rate are significantly negative for all classes, suggesting that all the consumers prefer lower interest rate. The difference in preferences is mainly affected by attributes such as credit period, collateral method and repayment schedule.

For example, the coefficients of credit period 2 are positive and significantly different from 0 for classes 2, 3 and 4, indicating that credit period at three years is significantly attractive for these classes. Respondents in class 1 appear to obtain lower utility from choosing credit period of five years, since the coefficient of credit period 3 is significantly negative in this class. Guarantor mortgage contributes more utility than pledging of assets for respondents in class 1. Likewise, members in classes 1, 2 and 4 significantly prefer installment repayment, while respondents in class 3 prefer the lump sum option. Notably, larger amount of loan is significantly and invariably less attractive than smaller sizes at different levels for the decision makers in all classes. This may result from the consideration of profitability and repayment pressure.

The alternative-specific constant, transaction costs and interest rate exhibit significantly negative preference among the four classes, indicating that lower utility will be obtained when paying for higher interest rate and transaction costs. However, respondents in class 1 appear to be showing higher utility for higher transaction costs. If we link class membership estimates to this irrational preference, one possible reason might be the preference for loans from formal financial institutions, but with less credit history. This is probably because it could be hard for these inexperienced consumers to be fully informed and to prepare all the application materials during the initial transaction. They may be willing to pay higher costs, such as the costs of information collection, material preparation, negotiation with financial institutions, etc., in order to borrow money successfully and accumulate experience.

Socio-demographic variables are included in the LC model to explain the class probability. Negative and significant class membership estimates for family size and annual liquid balance reveal that members in classes 1, 2 and 3 have smaller family size and less liquid balance, compared to class 4. Nevertheless, these three classes present better total income situation than the base class. Respondents in these three classes are more willing to accept credit from formal financial institutions than those in the fourth class. With regards to specific differences in each class, females are more likely to fall into class 1, since the gender is significantly negative. Members in class 1 are more likely to live in poor villages, as the coefficient of village mutual aid funds is positive, and also significantly different from other groups. Interestingly, joint guarantee is the collateral method of VMAF, and respondents in the first group just prefer guarantor collateral method. This confirms the rationality of the classification. Members in classes 2 and 3 have similar characteristics. For example, their household heads are both younger and more likely to be male. Other estimates, such as educational level, credit history, distance to financial institutions and loan purpose cannot be used to differentiate respondents into different classes according to their preferences.

5.3 EAA estimates

Results from the EAA model are presented in Table VII. The values of overall coefficients in a separate non-attendance probability (model 1) are larger than that of the RPL model, with qualitatively similar signs. The variables display different significant levels from the RPL and LC models, when taking ANA into consideration. Since the non-attendance probabilities are captured by attributes, each attribute has only one ANA probability. According to the AIC and BIC tests, the first EAA model fits the data better than the RPL model. This finding can be taken as evidence of preference heterogeneity and ANA. Models 2–5 focus on the attributes with lower non-attendance probability.

In the model 1, interest rate and repayment schedule significantly present the lowest separate non-attendance probability, denoting that the probability when interest rate is ignored in one choice situation is 11.4 percent, and the probability when repayment schedule is ignored in one choice situation is 39.7 percent. The most frequently ignored attribute is collateral method, accounting for 94.9 percent, followed by transaction costs, with 90.1 percent, which then are excluded in model 2. Loan size and credit period are insignificant in explaining the ANA probability.

Models 2–5 are EAA estimates with jointly estimated ANA probabilities. For example, in model 2, transaction costs and collateral method are the two excluded variables, which are jointly ignored by 87.1 percent. Credit period is now significant, and tend to explain the non-attendance probability by 62.5 percent, which is larger than that of interest rate and repayment schedule. So credit period is excluded in the following model. In the model 3, the excluded attributes, credit period, transaction costs and collateral method will be ignored by 86.2 percent in one choice situation. The ANA probability of interest rate is higher, while that of repayment schedule is lower. However, loan size is still insignificant in explaining the non-attendance. The ANA probability is the lowest according to the value, indicating that this attribute is less likely to play an important role in the decision-making process. Models 4 and 5 treat interest rate and repayment schedule as two separate attributes and one combination, respectively, in order to compare with other excluded variables. The results reveal that the probability that loan size, credit period, transaction costs and collateral method would be ignored in one choice set is 49.1 percent, and the probability that people merely take interest rate and repayment schedule into consideration is 71.5 percent. In comparison with ASC and interest rate, the estimate shows that people normally pay more attention to how much they can save, instead of whether they have to pay when engaging in microcredit choices.

5.4 Willingness to pay estimates

Table VIII presents smallholder farmers’ willingness to pay estimates for microcredit. The high proportion of respondents take interest rate into consideration, indicating this attribute should be regarded as price in terms of willingness to pay. This standard WTP results are shown in price 1 column. In order to compare the different WTP estimates, price 2 takes ASC as the price, indicating the willingness to involve in interest rates-cost credits. Given the definition of ASC in the previous section, the signs of the variables in price 2 rightly contrast with those in price 1.

According to the RPL model, respondents are willing to pay 0.704, 0.567, 0.629 and 2.699 percent higher interest rates when a credit is characterized by three years period, five years period, guarantor and installment repayment, respectively. However, they are willing to pay less when the loan size becomes larger and the transaction costs have to be taken into account. In comparison with price 1, the absolute values of attributes in price 2 category are smaller, indicating that when facing only one prior choice issue whether to pay interest rate or not (price 2), such as the issue that borrowing from formal financial institutions or relatives and friends, the amount of willingness or unwillingness to pay for each attribute is smaller. It reveals that respondents are more sensitive to attributes on how much to pay, since the cost is always inevitable. Repayment schedule and interest rate display the largest two positive WTP coefficients. And they are also the attributes that take up the highest probability of attribute attendance in the EAA estimation.

Distinct willingness to pay estimates are displayed in different classes based on the LC model. In the price 1 column, for example, respondents in class 1 are significantly willing to pay 4.930 percent less interest rate for a five-year credit, while members in classes 2, 3 and 4 would like to pay more for a three-year credit. Smallholders in class 3 will be willing to pay 0.955 more for lump sum credits, while individuals in classes 1, 2 and 4 are able to accept 1.803, 14.512 and 1.107 percent higher interest rates for installment repayment credits, respectively. When the attributes work in combination, for example, more than 75 percent of the respondents (classes 2 and 4) would like to pay higher interest rates, if a three-year microcredit adopts installment repayment schedule. However, the magnitude of willingness to pay is incongruent with the size of the coefficient. For example, with regard to the loan size, smallholder farmers in class 4 are significantly less unwilling to pay (Table VIII), although the coefficient is higher, compare with the other classes, as shown in Table VI. This is because WTP estimation only investigates people’s attitude to price, which is just one of the factors in the utility of each choice. It therefore shows evidence why some elements are replaceable and why higher interest rates are acceptable for some borrowers.

Willingness to pay estimation of EAA only used the separate non-attendance probability model, which exhibits a better model fit. Results indicate coefficients on the basis of price 1 are greatly larger than that of the RPL model, while coefficients in price 2 are smaller. It suggests that when taking ANA phenomenon into consideration, people will show stronger attitudes on willingness to pay for factors they consider important.

6. Discussion and conclusion

This study used RPL, LC and EAA models to analyze smallholders’ preferences and willingness to pay for microcredit with the consideration of preference heterogeneity and ANA, using data from a discrete choice experiment in China.

The results demonstrated that preference heterogeneity and ANA exist in the smallholder farmers’ microcredit choices, indicating that microcredit products cannot be optimally designed without targeting different groups and considering the relevant attributes. In particular, the estimates from the RPL model indicate that, on average, smallholder farmers prefer longer credit period, smaller credit size, lower transaction costs and lower interest rates. Guarantor and installment would be more attractive collateral and repayment methods, respectively. The findings for the LC model indicate that preference heterogeneity is related to socio-demographic features. These various preferences can be classified into four groups. Interest rate and transaction costs were found to be negatively and significantly influencing individuals’ utility. Although there is an exception with regards to the transaction costs that proportion is quite small. The results from the EAA model showed that when taking ANA phenomenon into consideration, people will show stronger attitudes on willingness to pay for factors they consider important.

The findings also revealed that high transaction costs tend to lower utility and the willingness to pay. Potential ways to reduce the transaction costs should include clean loan terms and simplifying application procedures, since distance to financial institutions is insignificant to smallholders. The results further showed that smallholders characterized by higher educational level, larger family size and better annual liquid balance prefer to use guarantor as collateral method. Because these traits are able to support the assumption that these respondents are impossible to side-step a responsibility of repayment and bear the blame from relatives or friends of breaking a credit contract.

The findings of this study provide some policy implications for adjusting rural microcredit strategies and improving microcredit development. Generally, they suggest that microcredit providers are able to attract new customers under the current interest rates, if the combination of interest rate, repayment schedule, loan size, credit period and collateral method is appropriately adjusted. For example, repayment schedule exhibits the lowest non-attendance probability besides interest rate, and most smallholders are willing to pay more when a credit product includes installment repayment method. Formal microcredit suppliers who intend to expand services for smallholders in rural areas may need to combine more small-scale credit products with installment repayment schedule. In particular, for some non-profit credit service organizations, such as Poverty Village Mutual Aid Funds, clear target group is needed to ensure effective operation and goal implementation. Furthermore, government ought to take the responsibility to improve the identification of assets in rural areas, in order to reduce financial institutions’ bad debt risk and transaction costs.

This study focused on microcredit in rural areas. Further research can consider comparing the difference between the rural and urban individuals on this issue, given that China is in a process of rapid urbanization.

Sample descriptive statistics

Variable Description Mean SD
Age Physical age of household head 59.071 11.423
Gender 1 if the household head is male, 0 otherwise 0.911 0.285
Education Educational level of household head: 0=no schooling, 1=primary (1–6 years), 2=junior middle (7–9 years), 3=senior middle (10–12 years), 4=training school (13–15 years), 5=bachelor (13–16 years), 6=master or higher 1.172 0.791
Family size Number of persons live in the family and share meals 3.542 1.529
Farm size Arable land, including the rent and cultivated land (mu) 3.388 2.014
Computer 1 if family owns computer, 0 otherwise 0.167 0.373
Automobile 1 if family owns automobile, 0 otherwise 0.069 0.253
Motorcycle 1 if family owns motorcycle, 0 otherwise 0.299 0.458
Total income Total family income (thousand yuan/year) 50.653 100.734
Annual liquid balance The difference between total income and total expenditure (thousand yuan/year) 25.096 83.368
Net return Average agricultural net income per unit (thousand yuan) 2.086 12.321
Credit history Times of loan in recent 5 years 0.708 0.695
Credit constraint 1 if an application for a loan was rejected in recent 5 years, 0 otherwise 0.165 0.371
Family debt situation Positive for the amount of claim, negative for the amount of debt −8.235 −28.144
Distance Distance to the nearest financial institution (km) 3.587 3.280
Demand 1 if family has demand for loan, 0 otherwise 0.386 0.487
Loan purpose for agricultural investment 1 if loan purpose is agricultural productive investment, 0 otherwise 0.286 0.452
Loan purpose for non-agricultural investment 1 if loan purpose is non-agricultural productive investment, 0 otherwise 0.112 0.316
VMAFs 1 if the village runs village mutual aid funds, 0 otherwise 0.391 0.489
Formal finance 1 if respondent prefers to credit from formal financial institutions 0.348 0.477

Attribute descriptions and attribute levels in the choice experiment

Attributes Description Attribute levels
Credit period The time before repaying off a loan 1 year, 3 years, 5 years
Interest rate Annual interest rate 0%, 5.10%, 8%, 8.25%
Loan size Maximum limitation of a loan (RMB yuan) 10,000, 50,000, 100,000, 200,000
Collateral method Security method against the possibility of repayment default Pledge assets, guarantor
Repayment schedule The way of repaying Lump sum, installment
Transaction costs Cumbersome degree of applying for a loan Low, medium, high

Sample choice scenario

Option 1 Option 2 Option 3
Credit period (years) 3 1 1
Interest rate (%) 8 8.25 8.25
Loan size 50,000 50,000 200,000
Collateral method Guarantor Guarantor Assets pledge
Repayment method Installment Lump sum Installment
Transaction costs Medium Medium High
I would prefer

Estimates of random parameter logit model

Mean SD
Variable Coef. SE Coef. SE Prob of coef. (negative)
ASC −0.957*** 0.267 1.820*** 0.204 0.700
Credit period 2 0.486*** 0.094 0.626*** 0.152 0.219
Credit period 3 0.392*** 0.092 −0.324** 0.144 0.887
Loan size 2 −0.131 0.116 −0.004 0.188 0.000
Loan size 3 −0.447*** 0.128 −0.029 0.158 0.000
Loan size 4 −0.303*** 0.117 0.264 0.171 0.874
Transaction costs 2 −0.535*** 0.101 0.074 0.158 0.767
Transaction costs 3 −1.100*** 0.129 0.683*** 0.166 0.946
Collateral method 0.435*** 0.082 1.180*** 0.106 0.356
Repayment schedule 1.865*** 0.107 2.125*** 0.120 0.190
Interest rate −0.691*** 0.036 0.236*** 0.024 0.998
Number of obs 13,248
Log likelihood −2,633.647
ρ2 0.138
LR χ2(11) 841.57
Prob>χ2 0.000

Notes: Probability of negative coefficient is calculated by 100×Φ(−mean/SD), where Φ is the cumulative standard normal distribution. In the empirical estimation, credit period 1 (1 year), loan size 1 (10,000), transaction costs 1 (low), pledge of asserts and lump sum repayment way are the base levels. *,**,***Statistical significance at 10, 5 and 1 percent levels, respectively

AIC and BIC values for different numbers of classes

Classes LL Nparam AIC BIC LL0 ρ2
2 −2,706.938 38 5,489.876 5,653.790 −3,054.433 0.114
3 −2,499.774 65 5,129.547 5,409.928 −3,054.433 0.182
4 −2,356.615 92 4,897.230 5,294.077 −3,054.433 0.228
5 −2,286.637 119 4,811.274 5,324.586 −3,054.433 0.251
6 −2,243.937 146 4,779.874 5,409.652 −3,054.433 0.265
RPL model −2,633.647 11 5,289.295 5,336.744 −3,054.433 0.138

Notes: ρ2=1−(LL)/LL0; AIC=−2(LL−P); BIC=−2LL+[P×InN]

Estimates of latent class model

Class 1 Class 2 Class 3 Class 4
Variable Coef. SE Coef. SE Coef. SE Coef. SE
ASC 1.935 1.354 0.423 0.597 −0.801** 0.407 −1.945** 0.767
Interest rate 0.003 0.154 −0.243*** 0.077 −0.327*** 0.055 −1.319*** 0.108
Credit period 2 −0.155 0.405 0.669*** 0.197 0.532*** 0.148 1.433*** 0.369
Credit period 3 −1.652** 0.673 0.007 0.210 0.141 0.142 1.119*** 0.309
Loan size 2 −0.406 0.606 −0.298 0.208 −0.344* 0.176 0.397 0.304
Loan size 3 −1.601** 0.626 −0.348 0.225 −0.488** 0.219 −1.232*** 0.410
Loan size 4 −0.984** 0.554 −0.107 0.198 −0.735*** 0.189 −0.852** 0.358
Transaction costs 2 0.589 0.512 −0.576*** 0.210 −1.274*** 0.175 0.366 0.344
Transaction costs 3 1.437** 0.580 −0.588*** 0.198 −2.236*** 0.245 0.049 0.438
Collateral method 6.364*** 1.005 −0.045 0.120 0.047 0.116 −0.432** 0.179
Repayment schedule 0.604* 0.314 3.533*** 0.264 −0.312*** 0.114 1.461*** 0.248
Class membership estimates
_cons 11.277*** 2.003 0.921 0.977 0.481 1.220
Age 0.008 0.026 −0.025** 0.011 −0.025* 0.014
Gender −15.077*** 1.988 0.393 0.392 0.073 0.500
Education −0.006 0.235 −0.100 0.115 −0.158 0.133
Family size −0.454** 0.206 −0.341*** 0.105 −0.226* 0.119
Total income 0.054*** 0.014 0.034*** 0.011 0.033*** 0.011
Credit history −0.077 0.430 0.032 0.206 0.344 0.238
Village mutual aid funds 1.911** 0.804 −0.112 0.318 −0.847** 0.399
Credit constraint −21.713 1.004 0.428 0.313 −0.018 0.424
Family debt situation 0.034 0.026 0.004 0.006 0.018** 0.008
Distance to financial institution −0.087 0.158 0.0001 0.037 −0.075 0.053
Loan purpose for agricultural investment 0.246 0.507 0.287 0.297 0.996*** 0.338
Loan purpose for non-agricultural investment −21.531 1.052 −0.092 0.422 0.476 0.486
Annual liquid balance −0.049*** 0.016 −0.034*** 0.013 −0.026** 0.013
Formal financial institution preference 1.878*** 0.635 0.894*** 0.330 1.468*** 0.384
Demand for loan 0.101 0.485 0.319 0.269 −0.075 0.322
Probability of class 5.8% 29.3% 18.5% 46.4%
Number of obs 13,248
Log likelihood −2,356.615
ρ2 0.228
Prob>χ2 0.000

Notes: *,**,***Statistical significance at 10, 5 and 1 percent levels, respectively

Results of endogenous attribute attendance model

Model 1 Model 2 Model 3 Model 4 Model 5
Variable Coef. ANA Coef. ANA Coef. ANA Coef. ANA Coef. ANA
ASC −4.131*** 0.772*** −3.727*** 0.818*** −3.634*** 0.841*** −3.887*** 0.783*** −3.671*** 0.793***
(0.320) (0.042) (0.260) (0.090) (0.242) (0.061) (0.244) (0.035) (0.196) (0.024)
Interest rate −0.820*** 0.114*** −0.754*** 0.152** −0.730*** 0.166*** −0.778*** 0.157*** −0.672*** 0.117***
(−0.055) (0.032) (0.043) (0.060) (0.040) (0.044) (0.033) (0.026) (0.027) (0.017)
Repayment schedule 3.387*** 0.397*** 3.070*** 0.377*** 2.989*** 0.344*** 3.081*** 0.340*** 2.252***
(0.260) (0.033) (0.235) (0.050) (0.185) (0.329) (0.154) (0.030) (0.103)
Loan size 2 −0.080 0.186 0.016 0.159 0.097 0.109 −0.068 0.076
(0.141) (0.319) (0.134) (0.345) (0.102) (0.300) (0.186) (0.236)
Loan size 3 −0.744** −0.580** −0.508** −0.501** −0.200
(0.292) (0.256) (0.199) (0.197) (0.243)
Loan size 4 −0.744* −0.475* −0.396* −0.337* −0.069
(0.383) (0.272) (0.207) (0.183) (0.265)
Credit period 2 0.629** 0.201 1.276 0.625** −0.516 −0.001 −0.465**
(−0.25) (0.322) (0.993) (0.276) (0.563) (0.175) (0.187)
Credit period 3 0.597** 0.974*** −0.274 0.217 −0.291
(0.303) (0.267) (0.324) (0.147) (0.215)
Transaction costs 2 −4.080*** 0.901*** −3.638*** −3.527*** −1.329*** −2.295***
(0.700) (0.017) (0.912) (0.865) (0.345) (0.337)
Transaction costs 3 −21.250** −7.142*** −7.126*** −2.530*** −4.007***
(9.630) (2.234) (1.853) (0.582) (0.566)
Collateral method 38.118 0.949*** 2.394** 2.148*** 0.663*** 1.085***
(104.223) (0.012) (1.099) (0.725) (0.191) (0.212)
Excluded attributes 0.871*** 0.862*** 0.491*** 0.715***
(0.043) (0.041) (0.136) (0.059)
Number of obs 13,248 13,248 13,248 13,248 13,248
Log likelihood −2,545.154 −2,668.623 −2,685.746 −2,698.726 −2,786.767
Wald chi2(11) 613.650 416.710 493.740 677.870 1,005.620
Prob>χ2 0.000 0.000 0.000 0.000 0.000
AIC 5,112.307 5,359.246 5,393.492 5,419.452 5,595.533
BIC 5,159.756 5,406.695 5,440.941 5,466.901 5,642.982

Notes: Standard errors are in parentheses. *,**,***Statistical significance at 10, 5 and 1 percent levels, respectively

Willingness to pay estimates

LC model
RPL model Class 1 Class 2 Class 3 Class 4 EAA model
Price 1 Price 2 Price 1 Price 2 Price 1 Price 2 Price 1 Price 2 Price 1 Price 2 Price 1 Price 2
ASC −1.385*** 5.775 1.739 −2.449** −1.474** −5.039***
Interest rate 0.722*** −0.002 0.575*** −0.408*** −0.678*** 0.198***
Credit period 2 0.704*** −0.508*** 0.464 −0.080 2.748*** −1.580*** 1.627*** −0.665*** 1.086*** −0.737*** 0.767*** −0.152***
Credit period 3 0.567*** −0.410*** −4.930** 0.854** 0.028 −0.016 0.431 −0.176 0.848*** −0.575*** 0.728** −0.144**
Loan size 2 −0.189 0.137 −1.211 0.210 −1.225 0.705 −1.052* 0.430* 0.301 −0.204 −0.097 0.019
Loan size 3 −0.647*** 0.468*** −4.778** 0.827** −1.430 0.822 −1.493** 0.610** −0.934*** 0.633*** −0.908*** 0.181***
Loan size 4 −0.438*** 0.316** −2.935* 0.508* −0.441 0.253 −2.247*** 0.918*** −0.646** 0.438** −0.908** 0.180**
Transaction costs 2 −0.775*** 0.560*** 1.759 −0.305 −2.367*** 1.361*** −3.893*** 1.590*** 0.277 −0.188 −4.976*** 0.988***
Transaction costs 3 −1.592*** 1.15*** 4.289** −0.743** −2.417*** 1.390*** −6.834*** 2.791*** 0.037 −0.025 −25.921** 5.144**
Collateral method 0.629*** −0.454*** 18.991*** −3.288*** −0.185 0.107 0.145 −0.059 −0.327** 0.222** 46.497 −9.227
Repayment schedule 2.699*** −1.949*** 1.803* −0.312* 14.512*** −8.346*** −0.955*** 0.390*** 1.107*** −0.751*** 4.131*** −0.820***

Notes: *,**,***Statistical significance at 10, 5 and 1 percent levels, respectively

Total choice sets of the choice experiment

Choice sets Credit period (years) Interest rate (%) Loan size Collateral method Repayment method Transaction costs
 1 3 8 50,000 Guarantor Installment Medium
 1 1 8.25 50,000 Guarantor Lump sum Medium
 1 1 8.25 200,000 Assets pledge Installment High
 2 5 8 200,000 Assets pledge Installment Low
 2 1 5.10 200,000 Guarantor Installment High
 2 5 8 10,000 Guarantor Lump sum Medium
 3 1 8 100,000 Assets pledge Installment High
 3 1 0 100,000 Guarantor Installment Medium
 3 5 8 100,000 Assets pledge Lump sum Medium
 4 3 8 50,000 Assets pledge Lump sum High
 4 5 0 10,000 Guarantor Lump sum High
 4 5 5.10 10,000 Assets pledge Installment High
 5 5 8 100,000 Guarantor Installment Low
 5 5 5.10 200,000 Guarantor Lump sum Low
 5 3 5.10 200,000 Assets pledge Installment Low
 6 3 0 10,000 Assets pledge Lump sum Medium
 6 5 8.25 100,000 Guarantor Installment Medium
 6 1 8 10,000 Assets pledge Installment Medium
 7 5 0 200,000 Assets pledge Installment High
 7 3 8.25 50,000 Assets pledge Installment Low
 7 3 0 100,000 Assets pledge Lump sum Low
 8 1 0 50,000 Assets pledge Lump sum High
 8 3 8.25 200,000 Assets pledge Lump sum Medium
 8 5 8.25 100,000 Assets pledge Lump sum High
 9 3 8 200,000 Guarantor Lump sum Low
 9 5 0 200,000 Guarantor Installment Low
 9 1 5.10 100,000 Guarantor Installment Low
10 1 0 200,000 Assets pledge Installment Medium
10 5 5.10 100,000 Assets pledge Lump sum Low
10 5 8.25 200,000 Guarantor Lump sum High
11 5 5.10 200,000 Assets pledge Installment Medium
11 5 5.10 100,000 Guarantor Installment High
11 1 5.10 100,000 Assets pledge Lump sum Medium
12 3 0 200,000 Guarantor Lump sum High
12 5 8.25 10,000 Assets pledge Installment Low
12 1 8 200,000 Guarantor Lump sum Medium
13 1 0 10,000 Assets pledge Installment Low
13 3 5.10 10,000 Guarantor Lump sum Medium
13 5 8.25 10,000 Guarantor Lump sum Low
14 5 0 50,000 Assets pledge Lump sum Low
14 5 0 10,000 Guarantor Installment Medium
14 5 0 200,000 Guarantor Lump sum Medium
15 1 8.25 10,000 Assets pledge Lump sum Medium
15 1 8 50,000 Guarantor Installment High
15 1 0 10,000 Guarantor Lump sum Medium
16 3 8.25 50,000 Guarantor Installment High
16 1 0 100,000 Guarantor Lump sum High
16 3 8 200,000 Guarantor Installment High
17 1 5.10 10,000 Assets pledge Lump sum Low
17 1 8 50,000 Assets pledge Lump sum Low
17 3 8 10,000 Guarantor Installment Low
18 3 8 100,000 Guarantor Lump sum Medium
18 3 5.10 50,000 Guarantor Lump sum Low
18 5 5.10 50,000 Guarantor Installment Medium
19 3 8.25 100,000 Guarantor Installment Low
19 3 0 50,000 Assets pledge Installment Medium
19 1 8.25 100,000 Assets pledge Lump sum Low
20 3 0 10,000 Assets pledge Installment High
20 3 5.10 200,000 Assets pledge Lump sum High
20 5 8 50,000 Guarantor Lump sum High
21 3 8.25 10,000 Assets pledge Lump sum High
21 1 5.10 50,000 Assets pledge Installment Medium
21 1 0 50,000 Guarantor Installment Low
22 1 0 200,000 Assets pledge Lump sum Low
22 1 8.25 200,000 Guarantor Installment Low
22 5 0 100,000 Assets pledge Installment Medium
23 3 5.10 100,000 Assets pledge Installment Medium
23 5 8.25 50,000 Assets pledge Lump sum Medium
23 1 8.25 10,000 Guarantor Installment High
24 5 8 50,000 Assets pledge Installment High
24 1 5.10 50,000 Guarantor Lump sum High
24 1 8 10,000 Assets pledge Lump sum High

Note

1.

Many terms refer to this program. This study uses the term “Village Mutual Aid Funds” according to the report Access to Finance, Microfinance Innovations in the People’s Republic of China, Asian Develop Bank, 2014.

Appendix

Table AI

References

Ahlin, C. and Jiang, N. (2008), “Can micro-credit bring development?”, Journal of Development Economics, Vol. 86 No. 1, pp. 1-21.

Angelucci, M., Karlan, D. and Zinman, J. (2015), “Microcredit impacts: evidence from a randomized microcredit program placement experiment by Compartamos Banco”, American Economic Journal: Applied Economics, Vol. 7 No. 1, pp. 151-182.

Attanasio, O., Augsburg, B., De Haas, R., Fitzsimons, E. and Harmgart, H. (2015), “The impacts of microcredit: evidence from joint-liability lending in Mongolia”, American Economic Journal: Applied Economics, Vol. 7 No. 1, pp. 90-122.

Ayyagari, M., Demirgüç-Kunt, A. and Maksimovic, V. (2010), “Formal versus informal finance: evidence from China”, The Review of Financial Studies, Vol. 23 No. 8, pp. 3048-3097.

Bauer, M., Chytilová, J. and Morduch, J. (2012), “Behavioral foundations of microcredit: experimental and survey evidence from rural India”, American Economic Review, Vol. 102 No. 2, pp. 1118-1139.

Bello, M. and Abdulai, A. (2016a), “Impact of ex-ante hypothetical bias mitigation methods on attribute non-attendance in choice experiments”, American Journal of Agricultural Economics, Vol. 98 No. 2, pp. 1-21.

Bello, M. and Abdulai, A. (2016b), “Measuring heterogeneity, survey engagement and response quality in preferences for organic products in Nigeria”, Applied Economics, Vol. 48 No. 13, pp. 1159-1178.

Berhane, G. and Gardebroek, C. (2011), “Does microfinance reduce rural poverty? Evidence based on household panel data from Northern Ethiopia”, American Journal of Agricultural Economics, Vol. 93 No. 1, pp. 43-55.

Boucher, S.R., Carter, M.R. and Guirkinger, C. (2008), “Risk rationing and wealth effects in credit markets: theory and implications for agricultural development”, American Journal of Agricultural Economics, Vol. 90 No. 2, pp. 409-423.

Broch, S.W. and Vedel, S.E. (2012), “Using choice experiments to investigate the policy relevance of heterogeneity in farmer agri-environmental contract preferences”, Environmental and Resource Economics, Vol. 54 No. 4, pp. 561-581.

Bruhn, M. and Love, I. (2014), “The real impact of improved access to finance: evidence from Mexico”, The Journal of Finance, Vol. 69 No. 3, pp. 1347-1376.

Cai, J., Janvry, A.D. and Sadoulet, E. (2015), “Social networks and the decision to insure”, American Economic Journal: Applied Economics, Vol. 7 No. 2, pp. 81-108.

Cheng, E.J. and Ahmed, A.D. (2014), “The demand for credit, credit rationing and the role of microfinance”, China Agricultural Economic Review, Vol. 6 No. 2, pp. 295-315.

Dehejia, R., Montgomery, H. and Morduch, J. (2012), “Do interest rates matter? Credit demand in the Dhaka slums”, Journal of Development Economics, Vol. 97 No. 2, pp. 437-449.

Field, E., Pande, R., Papp, J. and Rigol, N. (2011), “Term structure of debt and entrepreneurship: experiment evidence from microfinance”, working paper, International Growth Centre, London.

Giné, X. and Karlan, D.S. (2014), “Group versus individual liability: short and long term evidence from Philippine microcredit lending groups”, Journal of Development Economics, Vol. 107 No. C, pp. 65-83.

Hensher, D.A. and Greene, W.H. (2010), “Non-attendance and dual processing of common-metric attributes in choice analysis: a latent class specification”, Empirical Economics, Vol. 39 No. 2, pp. 413-426.

Hensher, D.A., Rose, J. and Greene, W.H. (2005), “The implications on willingness to pay of respondent ignoring specific attributes”, Transportation, Vol. 32 No. 3, pp. 203-222.

Hensher, D.A., Rose, J.M. and Greene, W.H. (2012), “Inferring attribute non-attendance from stated choice data: implications for willingness to pay estimates and a warning for stated choice experiment design”, Transportation, Vol. 39 No. 2, pp. 235-245.

Herd, R., Hill, S. and Pigott, C. (2010), “China’s financial sector reforms”, OECD Economics Department Working Papers No. 747, OECD Publishing, Paris.

Hermes, N., Lensink, R. and Meesters, A. (2011), “Outreach and efficiency of microfinance institutions”, World Development, Vol. 39 No. 6, pp. 938-948.

Hole, A.R. (2011), “A discrete choice model with endogenous attribute attendance”, Economics Letters, Vol. 110 No. 3, pp. 203-205.

Hole, A.R., Kolstad, J.R. and Gyrd-Hansen, D. (2013), “Inferred vs stated attribute non-attendance in choice experiments: a study of doctors’ prescription behavior”, Journal of Economic Behavior & Organization, Vol. 96 No. C, pp. 21-31.

Imai, K.S., Gaiha, R., Thapa, G. and Annim, S.K. (2012), “Microfinance and poverty-a macro perspective”, World Development, Vol. 40 No. 8, pp. 1675-1689.

Janvry, A.D., Mclntosh, C. and Sadoulet, E. (2010), “The supply- and demand-side impacts of credit market information”, Journal of Development Economics, Vol. 93 No. 2, pp. 173-188.

Just, D.R. and Gabrielyan, G. (2016), “Food and consumer behavior: why the details matter”, Agricultural Economics, Vol. 47 No. S1, pp. 73-83.

Khandker, S.R. and Koolwal, G.B. (2016), “How has microcredit supported agriculture? Evidence using panel data from Bangladesh”, Agricultural Economics, Vol. 47 No. 2, pp. 157-168.

Kong, R., Turvey, C.G., Channa, H. and Peng, Y. (2015), “Factors affecting farmers’ participation in China’s group guarantee lending program”, China Agricultural Economic Review, Vol. 7 No. 1, pp. 45-64.

Ksoll, C., Lilleoer, H.B., Lønborg, J.H. and Rasmussen, O.D. (2016), “Impact of village savings and loan associations: evidence from a cluster randomized trial”, Journal of Development Economics, Vol. 120 No. C, pp. 70-85.

Lønborg, J.H. and Rasmussen, O.D. (2014), “Can microfinance reach the poorest: evidence from a community-managed microfinance intervention”, World Development, Vol. 64 No. C, pp. 460-472.

Madajewicz, M. (2011), “Joint liability versus individual liability in credit contracts”, Journal of Economic Behavior & Organization, Vol. 77 No. 2, pp. 107-123.

Maddala, G.S. (1983), Limited-Dependent and Qualitative Variables in Econometrics, Cambridge University Press, Cambridge, p. 60.

Maria, P. (2016), “Effects of microcredit and other loans on female empowerment in Bangladesh: the borrower’s gender influences intra-household resource allocation”, Agricultural Economics, Vol. 47 No. 2, pp. 235-245.

Mazumder, M.S.U. and Lu, W.C. (2014), “What impact does microfinance have on rural livelihood? A comparison of governmental and non-governmental microfinance programs in Bangladesh”, World Development, Vol. 68 No. C, pp. 336-354.

Ortega, D.L. and Ward, P.S. (2016), “Information processing strategies and framing effects in developing country choice experiments: results from rice farmers in India”, Agricultural Economics, Vol. 47 No. 5, pp. 493-504.

Ortega, D.L., Wang, H.H., Olynk Widmar, N.J. and Wu, L. (2014), “Reprint of Chinese producer behavior: aquaculture farmers in Southern China”, China Economic Review, Vol. 30 No. C, pp. 540-547.

Rashid, S., Yoon, Y. and Kashem, S.B. (2011), “Assessing the potential impact of microfinance with agent-based modeling”, Economic Modelling, Vol. 28 No. 4, pp. 1907-1913.

Ritchie, A. (2007), “Community-based financial organizations: a solution to access in remote rural areas?”, Agriculture and Rural Development Discussion Paper No. 34, The World Bank, Washington, DC.

Sagamba, M., Shchetinin, O. and Yusupov, N. (2013), “Do microloan officers want to lend to the less advantaged? Evidence from a choice experiment”, World Development, Vol. 42 No. C, pp. 182-198.

Scarpa, R., Thiene, M. and Hensher, D.A. (2010), “Monitoring choice task attribute attendance in nonmarket valuation of multiple park management services: does it matter?”, Land Economics, Vol. 86 No. 4, pp. 817-839.

Scarpa, R., Zanoli, R., Bruschi, V. and Naspetti, S. (2012), “Inferred and stated attribute non-attendance in food choice experiment”, American Journal of Agricultural Economics, Vol. 95 No. 1, pp. 165-180.

Shoji, M., Aoyagi, K., Kasahara, R.J., Sawada, Y. and Ueyama, M. (2012), “Social capital formation and credit access: evidence from Sri Lanka”, World Development, Vol. 40 No. 12, pp. 2522-2536.

Tan, X. and Lin, W. (2016), “Can poor farmers afford higher micro-credit interest rates than the un-poor?”, China Agricultural Economic Review, Vol. 8 No. 1, pp. 100-111.

Train, K.E. (2009), Discrete Choice Methods with Simulation, 2nd ed., Cambridge University Press, New York, NY, p. 34.

Tsukada, K., Higashikata, T. and Takahashi, K. (2010), “Microfinance penetration and its influence on credit choice in Indonesia: evidence from a household panel survey”, The Developing Economies, Vol. 48 No. 1, pp. 102-127.

Turvey, C.G., Kong, R. and Huo, X.X. (2010), “Borrowing amongst friends: the economics of informal credit in rural China”, China Agricultural Economic Review, Vol. 2 No. 2, pp. 133-147.

Turvey, C.G., He, G., Ma, J., Kong, R. and Meagher, P. (2012), “Farm credit and credit demand elasticities in Shaanxi and Gansu”, China Economic Review, Vol. 23 No. 4, pp. 1020-1035.

Weber, R., Musshoff, O. and Petrick, M. (2014), “How flexible repayment schedules affect credit risk in agricultural microfinance”, discussion paper, Department for Agricultural Economics and Rural Development, Georg-August-Universität Göttingen, Göttingen.

Weller, P., Oehlmann, M., Mariel, P. and Meyerhoff, J. (2014), “Stated and inferred attribute non-attendance in a design of designs approach”, Journal of Choice Modelling, Vol. 11 No. C, pp. 43-56.

Xiang, C., Jia, X. and Huang, J. (2014), “Microfinance through non-governmental organizations and its effects on formal and informal credit: evidence form rural China”, China Agricultural Economic Review, Vol. 6 No. 2, pp. 182-197.

Xie, J., Gao, Z., Swisher, M. and Zhao, X. (2016), “Consumers’ preferences for fresh broccolis: interactive effects between country of origin and organic labels”, Agricultural Economics, Vol. 47 No. 2, pp. 181-191.

Yesuf, M. and Bluffstone, R.A. (2009), “Poverty, risk aversion, and path dependence in low-income countries: experimental evidence from Ethiopia”, American Journal of Agricultural Economics, Vol. 91 No. 4, pp. 1022-1037.

Yuan, Y. and Xu, L.H. (2015), “Are poor able to access the informal credit market? Evidence form rural households in China”, China Economic Review, Vol. 33 No. C, pp. 232-246.

Further reading

Hensher, D.A. (2010), “Hypothetical bias, choice experiments and willingness to pay”, Transportation Research Part B, Vol. 44 No. 6, pp. 735-752.

Ortega, D.L., Wang, H.H., Wu, L. and Olynk, N.J. (2011), “Modeling heterogeneity in consumer preferences for select food safety attributes in China”, Food Policy, Vol. 36 No. 2, pp. 318-324.

Corresponding author

Awudu Abdulai can be contacted at: aabdula@food-econ.uni-kiel.de