3.1 Data collection

The dataset used in this analysis is based on a household survey. A survey was conducted by Frederik Noack and Marie-Catherine Riekhof, together with an interviewer team, in 17 fishing villages around Chilika Lagoon, Odisha, India, from 21 February to 12 April 2011. Chilika Lagoon is located in the Bay of Bengal. It is the largest coastal wetland ecosystem on the Indian sub-continent (Mohapatra et al., Reference Mohapatra, Mohanty, Mohanty, Bhatta and Das2007). Our survey was part of the ‘Integrated Coastal Zone Management’ Program, funded by the World Bank.

To collect the information, we stratified the entire fishing community according to ecological regions and village size. Within villages, we interviewed a total of 508 randomly-chosen heads of households. In addition to the interviews, we conducted group discussions and spoke to local stakeholders like money-lenders, micro-finance organizations and teachers.

3.2 Description of the sample

This study focuses on interest rates. From the 508 interviewed households, 436 households were indebted. In total, 555 loans were reported, but information on the interest rate was only available for 430 loans. As the data set only provides information on current debt and not on loans already repaid, information was only used from loans that were taken out in 2009, 2010 and the first months of 2011. This gives 319 loans held by a total of 234 households.Footnote ¹⁴ Including loans from earlier years could lead to a selection bias, as information on loans taken out during that period, but already repaid, is missing. Solely considering these relatively recent loans reduces the probability of missing loans.

3.3 Chilika Lagoon fisheries

Chilika Lagoon is around 65 km long and 18 km wide (Sahu et al., Reference Sahu, Pati and Panigrahy2014). It is a brackish water body with saltwater inflows from the sea and freshwater inflows from rivers. This generates different ecological conditions within the lagoon. They are reflected in the four sectors of the lagoon, namely the Northern, the Central, the Southern and the Outer Channel Sector. The Outer Channel Sector encompasses the lagoon's main connection to the sea. Villages in this sector are more difficult to reach. The Chilika Development Authority reports 32,530 active fishermen in the lagoon in 2010/11. Total annual catches in 2010/11 of fish, crab, shrimp and prawn were 13 thousand metric tons (Directorate of Fisheries, Government of Odisha, India, 2013).

Predominantly, the male household members go fishing. They fish in groups and share the catch. On average, these so-called ‘fishing units’ have three members. If not all members belong to the same household, the groups use different remuneration systems. In most cases, the catch is divided into equal shares. The number of shares depends on the remuneration of capital: each member receives one share, but often the boat owner receives an additional share. Sometimes, the net owner also receives an additional share. This implies that the catch is divided into n, n + 1 or n + 2 equal shares, with n denoting the number of fishing unit members. Paying wages to collaborators is not very common. Table 1 gives an overview of the frequency of the different remuneration systems. If the members of a fishing unit do not belong to the same family, the most important sharing mechanism includes that an additional share is given to the boat owner. Accordingly, the income share of capital is 25 per cent for an average fishing unit with three members.

Table 1. Remuneration system of fishing units, n = 234

Fishermen use different methods to target different species. Fishing methods used to be related to the subcastes of the fishermen, but that system is disappearing. Over 80 per cent of the households go fishing by boat. Fishing trips last two days on average. The fish caught is usually brought to so-called ‘landing centres’, where fish traders buy the fish and transport it to national or international markets or re-sell it on local markets.

Fishing incomes are low and vary across the three fishing seasons – summer, monsoon and winter – as shown in table 2. There is no lean season as e.g., in agriculture, but interannual income variability, as in many fisheries (Kasperski and Holland, Reference Kasperski and Holland2013). Data points are based on the fishermen's ability to recall fishing income for a typical day in each season for 2010/2011, leaving us with one observation per fisherman per season. While more data points would clearly be preferable, the available data still give an idea of the order of magnitude of volatility the fishermen have to deal with over the seasons.

Table 2. Averagemonthly fishing income per fishing unit for the three seasons, n = 234

Rs = Indian Rupees

The average of the coefficients of variation of all households' fishing incomes in the different seasons is 0.51, with a standard deviation of 0.33 across households. Some households face a rather stable fishing income over the seasons – e.g., for a coefficient of variation of 0.51 − 0.33 = 0.18 – while others face quite a lot of variation – e.g., for a coefficient of variation close to one (0.51 + 0.33 = 0.84).

Few households have an additional income source besides fishing. If they do, fishermen usually work as unskilled laborers to supplement fishing income.

3.4 Interlinked loans

Fish traders provide loans without asking for fixed interest payments, but with special agreements about buying the fish at a lower price from the borrower. Price differences are fixed at the beginning of the contract. The indebted fisherman has to offer his entire catch to the fish trader. If the fisherman catches a lot in terms of weight, the income lost because of the interest payments is larger than if he catches little. The principal is repaid separately. One could say that interest is paid in fish and the principal is paid in money. The principal is usually repaid at once.

As the interlinked loans normally do not have a fixed maturity, the repayment of the principal terminates the contract. This set-up is also reported by Platteau and Abraham (Reference Platteau and Abraham1987). The fishermen have an incentive to quickly repay the loan to keep the additional income. Still, due to their low income, it takes some time.Footnote ¹⁵ Traders do not necessarily favor quick repayment, as they have preferential access to the fish catch as long as the fisherman is indebted. As the data does not report a lot more old interlinked loans compared to standard loans, interlinked loans do not seem to create life-long indebtedness. Also, it is usually possible to take out another loan to repay an existing one.

The set-up of interlinked loans implies that each household can only have one interlinked loan, because it involves offering the entire catch to the fish trader the household is indebted to. An additional standard loan is possible.

To distinguish a trader that lends money from a trader that does not, we will call the former a ‘trader-lender’. Some people around Chilika Lagoon reported that trader-lenders often deduct some baseline amount from the price they give to an unknown fisherman, as they first have to ensure that this fisherman is not indebted to another trader-lender.

The interest rates from interlinked loans are calculated based on the borrower's income loss from selling the fish at a lower price to the trader-lender. These calculations of the interlinked interest rates are either based on income forgone per day due to selling to the trader-lender instead of another trader, or on the fisherman's lost number of Indian Rupees per kg per day due to selling to the trader-lender instead of another trader. In the first case, we multiply the income forgone per day by the fishing days per month, differentiated for the three seasons, and then sum over all months to attain the yearly interest payments. In the second case, we multiply the amount lost by the catch in kilograms. The catch in kilograms means the catch of the total fishing unit times the share the household receives according to the sharing mechanism (see table 1). Then again, we multiply the amount lost per day by the fishing days per month, differentiated for the three seasons, and then sum over all months to attain the yearly interest payments.Footnote ¹⁶

Relating the resulting total yearly interest payments to the yearly fishing income of the household yields the share the trader-lender receives from the fisherman's income as interest payment. On average, the trader-lender receives 26.1 per cent. This is very close to the average income share of capital in fishing income for an average fishing unit with three members, based on the sharing mechanism used most widely if the members of the fishing unit do not belong to the same family (see table 1).

Relating the yearly interest payments to the loan amount yields the yearly interest rate. The average interest rate for a loan from a trader-lender is 49 per cent p.a. In the next section, we connect the information on interlinked loan contracts with other credit contracts.

3.5 Informal credit markets around Chilika Lagoon

The description of the credit market is based on the collected data as well as on information from group discussions in the surveyed villages. It seems to be rather representative for the fishing-related part of the local population, but may not carry over to households mainly active in agriculture.

Several lender types are available in the credit markets around Chilika Lagoon. This is typical for rural credit markets in less developed economies (see e.g., Menkhoff et al. (Reference Menkhoff, Neuberger and Rungruxsirivorn2012)). Besides trader-lenders, who provide 16 per cent of all loans, other informal sources – i.e., money-lenders, family, friends, neighbors – provide 34 per cent. Formal sources – i.e., banks, micro-finance institutions, cooperatives – provide 50 per cent. From the formal sources, 83 per cent of the loans are loans from micro-finance institutions. Interlinked loans become more important if one considers the amount lent instead of the number of loans (see table 3).

Interest rates are high and loan amounts comparably small, which is also typical for many rural credit markets in less developed economies. Table 3 depicts the average yearly interest rates as well as the average loan amounts. Interest rates are nominal and calculated as of 2011. The inflation rate is around 10 per cent p.a., based on the consumer price index for agricultural laborers in Orissa (Government of Orissa, India, 2011). On average, interest rates from formal sources are lower than from informal sources, and interest rates from interlinked loans are highest. The difference between interest rates from interlinked loans compared to interest rates from all other loans – 37 per cent p.a. – is significant at the 5 per cent level.Footnote ¹⁷ The result is similar to the findings from Crow and Murshid (Reference Crow and Murshid1992, Reference Crow and Murshid1994 and Bell et al. (Reference Bell, Srinivasan and Udry1997). When comparing interest rates from interlinked loans to interest rates from other informal loans, the null hypothesis of equality cannot be rejected.Footnote ¹⁸ This result is in line with the results from Minten et al. (Reference Minten, Vandeplas and Swinnen2012) and Platteau and Abraham (Reference Platteau and Abraham1987).

Table 3. Data on credit markets, n = 319

Rs = Indian Rupees, in 2011, 50 Rs≈1 US$. The ‘share in total number of loans’ does not add up to 100% due to rounding errors.

The average loan amount is lowest from formal sources and highest in interlinked loans (which corresponds to the results in Bell et al. (Reference Bell, Srinivasan and Udry1997)). The maximum loan amount, in turn, is from a standard informal loan.

The loans are usually not secured by formal collateral. Furthermore, only some loans – mainly loans from formal sources – have a fixed maturity.

Fishermen are self-employed so households need credit for productive as well as consumptive purposes. In the case of productive needs, one sometimes differentiates between fixed capital and working capital (Ray, Reference Ray1998). The need to finance fixed capital, such as boats and nets, is relatively more important around Chilika Lagoon because fishermen normally sell the fish on the day they catch it. The need to finance working capital is only relevant if they go on longer fishing trips. Demand for consumptive credit arises if the income falls, for example, due to seasonality in fish catches, a decrease in prices or fish stocks, or if consumption needs increase, for example due to weddings, illness or death. The main purposes for taking out a loan for the fishery households around Chilika Lagoon are fishing activities as well as consumption needs (see table 4).

Table 4. Importance of loan purposes, n = 319

One single loan can have several purposes.

4.1 Credit supply

Consider a small open economy in which two credit-contract types may be offered. The loan types differ only with respect to their interest payment rule. One contract is the standard credit contract that implies constant interest payments α, the α -contract. It encompasses all different kinds of ‘business models’ and does not distinguish e.g., between a money lender and a micro-finance organization. The other contract is the β -contract, an interlinked contract. Here, interest payments occur in the form of the income share 1-β, with β > 0 such that the borrower keeps a positive income.

Assume that the α -contract is always offered and that its interest rate is fixed in the sense that it is fully determined by household and credit characteristics. For the α -contracts, one can think of contestable monopolies with free entry and zero expected profits as market structure, as in Bell et al. (Reference Bell, Srinivasan and Udry1997).Footnote ²¹

For the β -contract, two aspects have to be taken into account. First, with lenders acting risk-neutral, e.g., because they can diversify between different borrowers, they will only offer the β -contract if interest payments are at least α. Otherwise, they will also offer an α -contract. Second, the β -contract calls for a monitoring of the borrower's income, as the borrower has an incentive to hide parts of his income to reduce interest payments. Not all lenders are in the position to do so. Only traders can easily observe the borrower's income,Footnote ²² but even they have to cooperate with each other to ensure that an indebted fisherman is not selling part of his catch before reporting to the trader he is indebted to. Together, the fish traders have a monopoly on β -contracts. As the α -contracts represent the exit option for the borrower, the traders will choose a limit-price policy. They will set β in such a way that the household just prefers the β - to the α -contract.

4.2 Credit demand

Consider a household that demands loan amount A to increase its income. Income g(A) is stochastic, with expected income E(g(A)) = μ and variance σ ² = E(g(A) ²) − μ ². The loan amount A is taken as exogenous. Thus, the household only chooses between the two credit-contracts, the α -contract and the β -contract.

Let

(1)$$r_\beta : = (1 - \beta )\mu /A$$

denote the interest rate in the β-contract and

$$r_\alpha : = \alpha /A$$

the interest rate in the α-contract.

Consider a household with mean-variance preferences.Footnote ²³ Mean-variance preferences are compatible with expected utility maximization if the utility function is a quadratic function or if all distributions of the choice set of the household belong to the same linear classFootnote ²⁴ (Sinn, Reference Sinn1990). Let the quadratic utility function u(g) = g − λg ² represent preferences over disposable income. Its form is problematic because the marginal utility turns negative as incomes become large. We presume that marginal utility turns negative only beyond the range in which g varies: we assume that $u^{\prime} \lpar \tilde{g}\rpar >0$ with $\tilde{g}=\beta \mu \lpar 1+\sigma^{2}/\mu^{2}$), i.e.,

$$\Delta : = 1 - 2\lambda \beta \mu (1 + \sigma ^2/\mu ^2) > 0.$$

Then, u′(g) > 0 for all $g < \tilde g$.

If interest payments were the same for both contracts, i.e., α = (1 − β)μ, choosing the interlinked contract would lead to a higher expected utility, as − λσ ² < −λβ ²σ ² based on comparing

$$E[u(g - \alpha )] = \mu - \alpha - \lambda (\sigma ^2 + (\mu - \alpha )^2)\quad \hbox{and}\quad E[u(\beta \hbox{g})]{\rm = }\beta \mu - \lambda \beta ^{\rm 2}(\sigma ^{\rm 2}{\rm + }\mu ^{\rm 2}).$$

Thus, a risk-averse borrower who has mean-variance preferences will strictly prefer the β-contract. The trader can increase total payments in the β-contract such that they are higher than in the α-contract. This gives the result that the borrower is willing to pay an insurance premium in addition to the pure credit costs α, r _β > r _α.

In case of var(g) = 0, the trader has to set β such that interest payments are equal to α. This implies ∂(1 − β)/∂σ > 0|_var(g)=0. The insurance premium is increasing in the variance for variance levels close to zero.

The household is indifferent between the two contracts if

(2)$$E[u(\beta g)] = E[u(g - \alpha )] \Leftrightarrow \,\,\,\,\mu - \alpha - \lambda (\sigma ^2 + (\mu - \alpha )^2) = \beta \mu - \lambda \beta ^2(\sigma ^2 + \mu ^2).$$

For the comparison among several households, the income variance relative to the income level is a more appropriate measure. It can be depicted by the coefficient of variation V: = σ/μ. Another advantage is that the coefficient of variation is dimensionless.

Rearranging (2) and rewriting in terms of r _α, r _β and V ² gives

(3)$$r_\beta - r_\alpha - \lambda ((V^2 + 1)r_\beta (2\mu - r_\beta \hbox{A}) - r_\alpha (2\mu - r_\alpha \hbox{A})) = 0.$$

Based on equation (3), one can derive the following testable hypotheses for a risk-averse household:

• H-V The interest rates from interlinked loans are increasing in the coefficient of variation:

  $$\displaystyle{{\partial r_\beta } \over {\partial V}} = - \displaystyle{{ - \lambda r_\beta (2\mu - r_\beta A)} \over {\underbrace{{1 - \lambda (V^2 + 1)2(\mu - r_\beta A)}}_{{ = \Delta > 0}}}} > 0.$$
  The nominator in H-V is < 0 as 2μ − ((1 − β)μ/A) A = 2μ − μ + βμ > 0.

• H-μ The interest rates from interlinked loans are increasing in the mean income:

  $$\displaystyle{{\partial r_\beta } \over {\partial \mu }} = - \displaystyle{{ - \lambda ((V^2 + 1)r_\beta ^2 - r_\alpha ^2 )} \over \Delta } > 0.$$

• H-A The interest rates from interlinked loans are decreasing in the loan amount:

  $$\displaystyle{{\partial r_\beta } \over {\partial A}} = - \displaystyle{{\lambda ((V^2 + 1)r_\beta ^2 - r_\alpha ^2 )} \over \Delta } < 0.$$

The impacts from mean income and loan amount on the interest rate from interlinked loans are driven by two counter-acting effects. The first effect is related to the definition of the interest rate for interlinked loans for a constant β. A higher mean income increases the interest rate from interlinked loans. A larger loan amount, in turn, decreases the interest rate, as interest payments are distributed over a larger amount. The second effect relates to the constraint of equal expected utilities from an interlinked and a standard loan contract. The increase in interest rates from interlinked loans due to a higher mean income, for example, is bounded by the interest rate for standard loans such that the share given to the fish trader may need to be reduced when mean income increases. For H-A and H-μ, the first effect dominates. For H-V, there is no counteracting effect through the constraint of equal expected utilities from both contracts, because volatility only matters in interlinked loans.

Two remarks are in order. First, the results also hold with free entry into the β-segment as long as lending traders are risk-averse and ask for a compensation of their risk-taking.Footnote ²⁵ If the traders act risk-neutrally, free entry would drive the mark-up down to zero. Second, the model illustrates the importance of the α -segment in limiting the interest rates in the β -segment. Related to this, if the fishing income is permanently reduced, e.g., because of a long-run decrease in the resource stock, a lending trader will adapt β in new contracts to earn again at least the same expected interest payments as with an α -contract.

Estimation with household fixed effects

From the 319 loans in the sample, 168 loans are held by households with only one loan in total, while 151 loans are held by households with more than one loan in total (see table 5). The 151 loans are held by 67 households. When using households FEs, we lose the information from the households that only took out one loan. To be transparent on this, we exclude those loans from the sample used for the FE-estimations. Including those loans will not change the results, as their impacts are captured by the household FEs.

Table 5. Descriptive statistics loans, n = 319

Interpretation: 124 standard loans are held by households that – in total – have more than one loan.

The baseline specification with household FEs takes the form

(4)$$\eqalign{ {\rm interest\,rat}{\rm e}_{l,h} =& \xi _1{\rm interlinked\,dumm}{\rm y}_l + \xi _2({\rm interlinked\,dumm}{\rm y}_l \times V_h) \cr & + \xi _3({\rm interlinked\,dumm}{\rm y}_l \times \log (\mu _h)) \cr & + \xi _4({\rm interlinked\,dumm}{\rm y}_l \times \log (A_l)) + \xi _5\log (A_l) \cr & + \Gamma L_l^T + \Lambda _h + c + \Theta D_y^T + {\rm \epsilon }_{l,h}.} $$

The interest rate relates to the loan l from household h. The interlinked dummy is one if the interest rate belongs to an interlinked loan and zero otherwise. The coefficient ξ₂ is the effect of income volatility V on interest rates from interlinked contracts and relates to hypothesis H-V. Mean income μ denotes the household's average fishing income per year. Mean income and loan amount A are included in logs. Both are interacted with the interlinked dummy to test H-μ and H-A. The ‘log loan amount’ is also included on its own to control for its general influence on interest rates. The general impact of mean income and income volatility is controlled for through the FEs. Further loan characteristics are summarized in the vector L _l. In line with the literature on credit markets, the vector L _l includes a dummy that is one if the loan is from a formal source, as well as a dummy that is one if a repayment date is fixed. This maturity dummy is always zero for interlinked loans, but especially for formal loans, repayment dates are specified. Table A1 in the online appendix gives more information on the credit market variables. The superscript T in equation (4) indicates that the row vector is transposed to a column vector. The vector Λ_h represents the household FEs. The constant c captures the conditions in 2011 such that the yearly dummies in the vector D _y capture the difference of a specific year compared to 2011. The error term is ε _l,h.

Ordinary least square estimation

The specification is similar to (4). Instead of the household FEs, several controls on the household level are included. Table 6 gives an overview of the variables and their definitions. It also gives some descriptive statistics. We briefly introduce the variables below.

Table 6. Definition and summary statistics of key variables on the household level, n = 234

*The household head's attitude is measured by the answers to the following questions:

‘Imagine: Two farmers keep goats on community land. Now, the land should be privatized and divided among the two farmers. After the division and privatization, each farmer can sell the land to get some money or he can keep it and use it for whatever he likes. The two farmers are of the same age, are healthy and have similar families. Both farmers have exactly the same amount of goats.

(a) The first farmer has always grazed his goats on the land. The second farmer has started to graze his goats on this land only one year ago.

(b) The first farmer, however, has sold most of his goats to buy luxury consumption goods. Now he is poor and only few of his goats remain. The second farmer has led a modest life. Now he is rich and has increased the size of his goat herd.

Which division of land would you think is fair? (A) The first farmer gets more land. (B) The second farmer gets more land. (C) Both get the same amount.’

We create one dummy called ‘attitude I’ that equals one if the answer in the first case was (B), i.e., the newly-arrived should receive more land, and a second dummy, ‘attitute II’ that equals one if the answer to the second question was (A).

The vector of household variables includes the log-income as well as the coefficient of variation. It also includes dummies that indicate whether the household owns a boat, a motor for the boat, or a cellphone, respectively, and dummies that indicate whether the household head has an additional income generating activity, whether he is literate and whether he ever attended vocational training. Another variable gives the number of working-age (age 12–60) male household members in the household. The vector also inlcudes dummies that control for the geographical sector the household lives in and a dummy that indicates whether the household belongs to the subcaste Khartia.Footnote ²⁶ Two additional dummies measure the attitude of the household head. One attitude-dummy measures whether the household head is open to newcomers while the other measures his attitude towards supporting people with self-inflicted problems (see table 6 for details). In the OLS specification, the constant c captures the situation in 2011 in the Outer Channel sector.