Profits from participation in high value agriculture: Evidence of heterogeneous benefits in contract farming schemes in Southern India

Food Policy 44 (2014) 142–157

Contents lists available at ScienceDirect

Food Policy journal homepage: www.elsevier.com/locate/foodpol

Profits from participation in high value agriculture: Evidence of heterogeneous benefits in contract farming schemes in Southern India Sudha Narayanan ⇑ C-304, Indira Gandhi Institute of Development Research (IGIDR), Gen. A.K. Vaidya Marg, Film City Road, Goregaon (East), Mumbai 400065, India

a r t i c l e

i n f o

Article history: Received 17 June 2012 Received in revised form 23 September 2013 Accepted 29 October 2013

Keywords: Contract farming Endogenous switching model Average treatment effects India

a b s t r a c t This paper assesses the variable impact of participation in high value agriculture through contract farming arrangements in southern India. Using survey data for 474 farmers in four commodity sectors, gherkins, papaya marigold and broiler, an endogenous switching model is used to estimate net profits from participation. Findings suggest that average treatments effect vary widely across contract commodities. Papaya and broiler contracting offer clear net gains for participants whereas marigold contracting leaves participants worse off. For gherkins, while contracting holds net gains for participating farmers overall, this is true of contracts with some firms but not others. The standard deviations of point estimates of treatment effects are quite large indicating variability in profit gains even within the same commodity sectors. Thus, notwithstanding the sign of average treatment effects, contract farming arrangements have diverse impacts on income for individual farmers and these could have implications for sustained participation of farmers in high value agriculture. Ó 2013 Elsevier Ltd. All rights reserved.

Introduction The issue of income gains to small farmers from participation in agro-food supply chains in developing countries, specifically in contract farming arrangements, has acquired much significance in recent times (Minot, 2008; Swinnen, 2007; Reardon and Gulati, 2008; Barrett et al., 2012; McCullough et al., 2008). Should participation in these chains lead to net gains, there exist credible opportunities for farmers in these countries to transform their livelihoods. While existing work has been largely successful in addressing methodological issues to measure welfare impacts, most notably the profits from participation in high value agro-food supply chains, a majority of works confine themselves to assessing whether or not participant farmers benefit on average (reviewed in Barrett et al., 2012, for example). An aspect that has faced relative neglect has been the heterogeneity of impacts associated with participation, both within and across schemes. This assumes importance in the context of high mortality of contract farming schemes in developing countries and widespread prevalence of disadoption or exit from contract participation. In India, for example, the study on which this article is based recorded high farmer attrition rates in the sample villages surveyed (Narayanan, 2013). ⇑ Tel.: +91 22 28416549 (O), +91 22 28416249 (H); fax: +91 22 28416399, +91 22 28402752. E-mail address: [email protected] 0306-9192/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.foodpol.2013.10.010

Further, among attrition farmers who were interviewed, as many as 20% of them stated economic losses from contracting as the reason for exiting the system and this was the single most important reason for exit. Thus, while, on average, participating farmers benefit, the heterogeneity of farmer experiences bears ingredients of churning and attrition in these schemes. This study uses unique survey data of farmers in multiple commodity schemes to answer the following questions: Do contracting farmers in high value supply chains do better than those who do not participate, on average? How much do they stand to gain relative to their counterparts who do not participate? How do these treatment effects vary for participating farmers within a commodity group? Do these patterns differ across contract commodities? This study tackles a particular difficulty where sometimes the decision to contract coincides with a decision to grow the contract commodity, so that all production of the high value commodity is contract-based and a domestic spot market is absent or too small to offer a credible comparison group. This makes it impossible to identify the impact of contracting separately from that associated with growing a high value commodity. This is not the case with most of the previous literature on welfare impacts from contract farming, where typically there exists a spot market for the contracted commodity or traditional marketing channels for the commodity in question. The presence of an appropriate counterfactual and a close comparison group in those cases enables use of techniques such as propensity score matching (Maertens and Swinnen,

S. Narayanan / Food Policy 44 (2014) 142–157

2009), Heckman’s selection models (Miyata et al., 2009), instrumental variable approaches.1 In the somewhat exceptional context described in this study, there is no precise counterfactual for contract participation per se. This study maintains that it is nevertheless possible to assess the impact of participation in contracting arrangements for high value commodity chains in totality relative to the counterfactual of persisting with a status quo of cultivating ’traditional’ crops and/or marketing channels. The counterfactual in these cases would have to be defined as not participating at all in the high value commodity chain in question. Alternatively, where there are a number of firms with whom farmers can contract for the high value commodity, it might be possible to assess the impact of contracting to a specific firm versus supplying to other firms also procuring the same commodity on contract or not being part of the supply chain altogether, as the case might be. Whereas the former assesses impact of growing the high value commodity under contracts versus the status quo of growing persisting with the traditional cropping pattern, the latter assesses impact of contracting with the subject firm, relative to other options, including contracting for the same commodity with another firm(s) or growing another crop altogether. To address this issue of coincidence of cropping choice and contracting choice, the study adopts an endogenous switching model where farmers sort themselves into two very different but comparable regimes, contracting for growing the high value commodity and not contracting (and therefore not growing the high value commodity). The sorting is based, in part, by the perceived differential welfare gains between the two regimes. While this enables assessment of profitability of participation in the two distinct regimes, it also allows me to comment on the differential returns to factors across these regimes and see if different regimes reward key factors of production differently. I then explore the variation in estimated treatment effects across schemes and across farmers, the treatment here referring to participation in high value agriculture through contract farming. Following this introduction is a description the survey data, its empirical context and the estimation strategy adopted. The next section describes the variables used and presents key results from the estimation of the endogenous switching model, focusing on incremental net profit associated with contracting. I then discuss the structure of costs and returns to highlight the sources of gains and comment on the returns to key factors of production under contracting and not contracting, before concluding the paper.

The commodities and their contexts The data for this study come from a survey of 474 farmers covering four commodity sectors, gherkins, marigold, papaya, and broiler chickens, in the southern state of Tamil Nadu and was conducted between 2009 and 2010. The list of contracting farmers for the year of the survey was obtained from one contracting firm (henceforth the subject or sample firm) in each of the commodities 1 The efficacy of these approaches invariably depends on the choice of an instrument that enables identification of the parameters of the model. Miyata et al. (2009) treat the distance between a respondents farm and the farm of the village chief as an instrument. Rao and Qaim (2011) use farmer group membership to serve as an instrument and Simmons et al. (2005) choose number of organizations farmers are members of as an instrument. Other instruments include the number of female laborers in the respondents household as well as a dummy for whether a female in the household is a member of a womens organization (Maertens and Swinnen, 2009), farmer willingness to pay (WTP) for a certain return from a randomly drawn level of investment (Bellemare, 2012). Across methods, the central challenge is to find an appropriate instrument that can break any correlation between selection and the unexplained variation in welfare outcomes. Panels using difference-in-differences have also been used (Michelson, 2013).

143

studied.2 Based on this list, all the hamlets in the sample area were divided into contracting and non-contracting hamlets and their corresponding villages into contracting villages or non-contracting villages. A similar exercise was carried out for the larger administrative units called blocks and then districts. Starting from the largest administrative unit for the study area, contracting districts were sampled, within which contract and non-contract blocks were randomly sampled and then further on, within sampled blocks, contract and non-contract villages were sampled and so too with hamlets. In the hamlets sampled, a census of all households identified four key types of farmers: those currently contracting with the subject firm (Contract farmers); those who were growing the contract crop but for the open market or contracting for other firms (Other Contract farmers); those who had given up contracting with the subject firm and no longer grew the contract commodity (Attrition farmers); and those who had never contracted the commodity with any firm (Never Contract farmers). The sample respondents were randomly selected from each type. If a farmer grew the contract crop for some other firm and quit, they were not sampled at all. All the contract farming schemes studied operate in rainfed agricultural areas and have diverse arrangements with farmers. Gherkins are a non-traditional export crop with no domestic market, but there are several firms that procure, mostly through contract farming and sometimes through informal procurement by agents. The crop is procured from farmers and processed at small-scale plants by washing, rinsing and preserving in brine, acetic acid or vinegar. These are either bottled and labeled for international clients or shipped out in barrels for bottling. Papaya was introduced in the region in the 1990s for extracting papain, which has wideranging industrial uses. The variety is appropriate, but not ideal, for table consumption, and the fruit is a by-product that is used to make candied fruit or for pureeing. Papaya for papein is procured through contracts but papaya for direct consumption is not. The subject firm is the lone processor of papein. Marigold contracting was initiated by firms for oleoresin extraction for export, mainly as coloring agent for poultry feed. Marigold has a thriving local market, however, for fresh cut flowers that are used for a number of occasions, religious and otherwise. Although three firms procure marigold, in the sample area there were no farmers who contracted with other firms and only a few who grew specifically for the fresh flower market. The broiler industry in the study region is almost completely vertically coordinated, a process that began in the mid-1990s. Day-old chicks are provided by the firm and bought back by the contracting firm. The firm acts as an aggregrator-intermediary, but also has its own brand of chicken in various processed forms. In many ways, the four schemes are fairly typical of contract production arrangements elsewhere in the developing world. All contract commodities are cash crops and involve production processes that require farmers to respond continuously to the need to maintain quality. Firms engaged in contract farming thus engage actively in the production process, not only providing critical inputs but also maintaining close supervision from sowing through to harvest and post-harvest handling. The commodities and firms selected for study represent varying degrees of involvement by the firm in the production process or intensity of contractual relationship, and this varies even across firms within the same commodity complex. Broiler represents high relationship intensity, with the firm’s officials visiting contract growers every day to monitor health and status of the birds. These firms provide day old chicks to the farm and have detailed protocols for the feed mix and vaccination schedules. For papaya, the 2 All firms were approached, who were contracting for the particular commodity in the study area. The firms selected as the subject or sample firms were those that were contracting that year and were willing to share the complete list of contract farmers. The study firms were the first to share these lists.

144

S. Narayanan / Food Policy 44 (2014) 142–157

involvement of the firm varies over the life cycle of the crop. In the nursery stage, field officials monitor the crop closely with daily visits and once the plant matures into the flowering stage, there is limited oversight, unless the situation demands it. In papaya, an interesting feature is that labor for latex extraction is organized and trained by the firm, with the wages being borne by the farmer. Latex extraction requires great skill and the firm believes it can ensure quality and supply of latex for the plant by maintaining a pool of trained workers, who extract latex on contract farms. Marigold represents the least participation of the firm in the production process, related partly to fewer quality requirements that need only modest supervision. In fact, the marigold firm suggests that monitoring is required more for contract enforcement rather than for production under contract. The marigold firm thus restricts itself to providing high quality seeds at subsidized prices and training new contract farmers in the cultivation practice for marigold. The firm’s field officials advise farmers periodically on pest and disease control. Across the schemes there is heterogeneity in the way risks are distributed between firm and farmers, although they do share many features, such as provision of some critical inputs, technical advice and an agreement to buy back at the end of the season. In general, contracts in the same commodity sector, irrespective of the identity of the firm, tend to be similar. That said, it could vary in practice many respects even across farmers even within the same contracting scheme, the most important being the timing of input delivery and its quantity as also supervision.

Empirical approach Given the specific context of the study, it is imperative to elaborate the nature of comparisons for each of the schemes. For gherkins, farmers who participate in contracts are committing to both a mode of production and/or transaction and simultaneously to growing a new crop. Here, three kinds of comparisons are made. First, the treatment group is defined to include all contract gherkins growers (irrespective of the client firm) and the comparison group to comprise growers who do not grow gherkins at all. This comparison assesses the impact on profits from participation in the gherkins supply chain versus the status quo. A second comparison redefines the treatment group to include those who contract for the subject firm (for which the farmer sample is representative) and a comparison group of farmers consisting of those contracting for other firms and those not growing gherkins at all are selected. This is because if the farmer did not contract with the subject firm, (s)he would have possibly contracted with another gherkins processor or not grown gherkins at all and grown some other crop instead. This case evaluates the net profits of contracting for gherkins with the subject firm relative to all other alternatives available to the farmer. A third possibility is to compare contract farmers for the subject firm with either exclusively those who do not grow gherkins at all or exclusively with those who grow gherkins on contract for firms other than the subject firm. In the context of the survey the former is possible, but the latter is not, since the sample is not representative for those who contract for these other firms and they are fairly small in number. Even when there is an alternate domestic market, as in the case of papaya, broiler and marigold, the definition of the counterfactual is challenging. For papaya, farmers commit to growing a different variety under contract. If farmers were not contracting, they would opt for other crops or other table varieties of papaya. The latter is uncommon in the region and only a couple of farmers in the sample belonged to this group. The counterfactual for papaya therefore are farmers who not only supply other varieties of papaya to the market but those who might be growing other crops instead. Unlike gherkins, there exists no comparison group of farmers who

contract for papaya with other firms. For marigold, the treatment group comprises farmers contracting with the subject firm and the control group those who are supplying to the local market or growing competing crops. As with papaya, there were very few, only six farmers, supplying to the local markets, since most marigold in the study region was under contract with the subject firm. Comparisons for broiler deserve special attention. The broiler comparison or control group comprises farmers who do not grow poultry at all while the treatment group is made of farmers who grow poultry under contract either for the subject firm or one of many other firms. Since growing a field crop and growing poultry are very different choices, the control group is very heterogeneous. Here, the monthly net profits for those who do not grow poultry has been scaled down to an area of 5000 square feet for comparability.3 Despite the fact that the comparison group that includes non-growers of poultry makes for comparisons across very different categories, in the context of a regime switching model, this yields important insight. Broiler farmers often require heavy investment in fixed assets (sheds to house the birds, feeders, drinkers, etc.) that work as barriers to entry. Most farmers convert farmland to broiler sheds. Those who are unable to do this are invariably resigned to continuing cultivation of field crops. In this context, the particular nature of comparison makes sense. The treatment group is defined as contract broiler growers for any firm and compared with farmers who have chosen not to participate at all in the broiler supply chains. A few issues however merit attention. Ideally, the treatment and comparison groups should supply to precisely the same endconsumer and for the same purpose. In the case of marigold for instance, the firm contracts for extraction and the non-contract farmers sell in the fresh flower market. Conflating these two implies that it is no longer clear if the impact measures the effect of distinguishing use or destination or whether it measures incremental benefits from contracting versus not contracting. It is essential therefore to distinguish between concluding that contracting benefits small farmers and concluding that supplying to export markets is more lucrative than to domestic markets. Recent literature on supermarket participation makes this distinction clear by framing the question differently, asking if farmers benefit from participating in modern supply chains rather than traditional channels (Minten et al., 2009; Maertens and Swinnen, 2009).4 This study is similar in spirit to this stream of literature. Furthermore, even with comparisons for the same crop between contracting and a traditional channel, given the crop has the same use across channels, the character of the local market can be transformed by the presence of contracting on a large scale and this leads to different kinds of empirical problems, so that the general equilibrium effect on the local market can alter the returns to non-participants as well. Thus, whenever contracting appears jointly with some other distinguishing characteristic, either in terms of destination, end use or varietal difference, the challenge is to measure of welfare impact of contracting per se, delinked from other coincidental attributes. In general, it is extremely difficult to isolate these impacts in survey data. This problem permeates all four commodity sectors chosen for this study.

3 The underlying assumption is that farmers who do not grow poultry would use the farmland to grow other field crops. 4 For India, there are several examples that suggest supplying for export markets yield higher returns than serving domestic markets. A study of Mahagrapes showed that profits earned per acre per annum by contract growers were nearly 38% higher than that for non-contract growers mainly because Mahagrapes serves global markets, and hence prices received are almost three times higher than in the local markets (Narrod et al., 2009). A similar case study of contract grape growers in Andhra Pradesh, also supplying the export market, showed that contract growers received 55% higher net returns than supplying to the domestic markets (Dev and Rao, 2005). For gherkins growers in Andhra Pradesh in 2004–2005, returns over variable costs were 30% higher than for other vegetable crops (Dev and Rao, 2005).

S. Narayanan / Food Policy 44 (2014) 142–157

An endogenous switching regression model (under known sample separation) offers a way to negotiate this difficulty because it enables comparison across distinct alternatives or regimes (Maddala, 1983; Dutoit, 2007). The rationale for this is that selection into contracting for high value commodities puts farmers in different groups, associated with different profit streams. The survey data on net profits per acre collected for both contract and non-contract farmers implies that I observe sample outcomes, or draws, from both profit streams or both regimes, presumably coming from different distributions. This is analogous to sorting into different occupations or choosing public or private sector employment or whether to migrate or not, which are typical applications of the model. In this study, the sorting is potentially endogenous, driven at least in part by the difference in the perceived net profits per acre when contracting and not contracting for a high value commodity.5 This separates the sample into two streams, those that contract for high value commodities and those who do not. This separation into two streams cannot isolate perfectly the cropping pattern effect from the contracting effect. Yet it is now possible to account for specific elements of the contractual relationship to assess impacts of specific contractual features or firm-specific effects on the net profit for those who are sorted into the ‘treatment’ group. For example, controlling for inputs or supervision provided by the firm, when there are multiple firms, for those in the contracting regime, allows identification of the impact of contractual elements. This approach is one way to overcome the difficulties in separating the crop switching effect from the contracting effect. Where there is a unique firm, however, such separation is rendered impossible. The switching model also draws attention to an important aspect that is often neglected in empirical work in contract farming. In estimating outcome metrics as functions of farmer characteristics, it is possible to compare the returns to particular factors of production across the two regimes, however different. In other approaches that account for selection, since regimes are pooled by construction, they mask the differential structure in returns to factors of production and other covariates. In contrast, the switching regression approach allows for structural differences in the relative determinants of profitability. This approach has been used in a variety of contexts (Lee, 1978; Adamchik and Bedi, 2000; Fuglie and Bosch, 1995; Manrique and Ojah, 2003). In the context of agriculture, Cadot et al. (2005) and Dutoit (2007) use this approach to assess the impact and ability of farmers to switch from subsistence farming to commercial agriculture in Madagascar, Cai et al. (2008) to evaluate rice contract farming in Thailand and Rao and Qaim (2011) supermarket participation on welfare of Kenyan farmers. A Full Information Maximum Likelihood (FIML) that esti mates the entire set of equations at once is an efficient way to estimate the model.6 Using this model it is possible to compute the average treatment effect among the treated and the average treatment effect for those not treated. In addition, it is also possible to assess the pattern of sorting. Suppose the correlation coefficients of the unexplained component between the first stage selection and outcomes from regimes 1 and 2 are q1v and q2v respectively. Whenever the signs of the estimated correlation alternate across the regimes, it implies that farmers are in regimes that offer them comparative advantage, so that, say, if q1v < 0 and q2v > 0, farmers with above average net income in regime 1 (contracting) are associated with a higher likelihood of being in regime 1 (contracting) and those earning a higher net profit not contracting are less likely

5 Exogenous factors drive participation too in terms of the selection process adopted by the firm that rations out farmers who might be willing to contract but are not offered contracts by the firm. 6 For a discussion see Dutoit (2007) and Lokshin and Sajaia (2004) for its implementation in STATA.

145

to participate in contracting. Alternatively, if both coefficients are negative, i.e., q1v < 0 and q2v < 0, there is what Fuglie and Bosch (1995) refer to as hierarchical sorting, so that those in regime 1 have better than average net profits in both regimes, but are better off in regime 1. Those in regime 2 face below average profits in both regimes, but are better off in regime 1. If, on the other hand, q1v > 0 and q2v < 0 it represents a situation where, regime 1 farmers would actually have below average profitability with their status quo but would have above average gains in regime 2 and those in regime 2 have above average performance in regime 1 but below average performance in regime 2. The last possibility, when q1v > 0 and q2v > 0, implies that farmers in regime 1 have higher than average profitability whether they contract or not, whether they are in regime 1 or 2, and hence have absolute advantage. The endogenous switching model comes with a few caveats. First, it assumes joint normality of errors. Second, the identification of the model comes through variables in the participation equation that influences participation but not the welfare outcome. The next section discusses the nature of comparisons for each of the commodities and describes the variables used in the estimation of the model in detail.

4. Variables used and identification strategy The metric used in the analysis is net profit per acre per month (henceforth, net profit) for field crops and net profit per 5000 birds per month for broiler growers.7 For all contract farmers, this is the net profit per acre derived from growing the contract commodity and for non-contract farmers, it is the net profit per acre either under the contract crop for an alternate market or for the crop they have chosen to be the closest substitute for the contract commodity. Net profits refer to income earned from all main and by-products of cultivation minus all paid out costs.8 Farmers were thus simply asked for the net income they were left with per unit area of production at the end of the season, after paying out all production and transactions costs for the entire season, including multiple harvests. The expression of net profits in terms of monthly earnings is merely to facilitate the discussion of diverse commodities. The net profit per acre per month was obtained by dividing the net profits per acre for the entire cropping season by the duration of the season in months. The aggregation over the season would account for multiple pickings or harvests and smoothen biases introduced by price volatility over the season. While this contributes to aggregation bias, it also makes inter-farmer comparisons more reliable. For perennials like papaya, data was obtained for the most recent month and for broiler, for the most recent completed cycle (comprising six weeks from chick to broiler). While farmers were typically encouraged to refer to any written accounts they had, in most cases, farmers relied on recall. The detailed costs and returns were obtained according to established protocols for collecting such data.9 7 This is for a 5000 square feet shed space that is considered a standard scale for broiler farmers in the region, by broiler contracting firms. 8 Net profit refers to the sum of income from sale of fruit or flower minus, farm yard manure, seeds, micronutrients, plant protection chemicals, fertilizer application, weeding, land preparation, seedbed preparation, seed treatment, intercropping expenses, if any. Transactions costs associated with input purchase and output sale are also subtracted, included transport and commissions. Labour costs comprising hired labor (male and female) are also accounted for. Family labour is recorded but not accounted for. For broiler, net profit represents income from sale of birds and poultry manure minus medicines, electricity, water charges, charcoal, cleaning costs, supplementary feed, hired labor, transport and commissions, if any. For the other crops, cited by farmers, as the next best alternative, the same components of costs are used and depending on the nature of the crop, the income is either from sale of flower, fruit or vegetable. 9 I also had access to cost of cultivation studies that provided benchmarks as crosschecks.

146

S. Narayanan / Food Policy 44 (2014) 142–157

Table 1 Kolomogorv–Smirnoff tests of equality of distributions of net profit. Commodity

Nature of comparison

D

p-Value

Gherkins

 (1) Contract with any gherkins firm versus farmers who are not part of the gherkins supply chain.  (2) Contract with subject gherkins firm versus farmers who are not part of the gherkins supply chain.  (3) Contract with subject gherkins firm versus farmers who contract with other firms or are not part of the gherkins supply chain.

0.275 0.315 0.22

0.04** 0.02*** 0.01***

Marigold

 Contract with subject (and the only) firm versus farmers who grow marigold for the spot market or an alternate crop.

0.273

0.01***

Papaya

 Contract with subject (and the only) firm versus farmers who grow table varieties of papaya or are not part of the papain supply chain

0.491

0.00***

Broiler

 Contract with any broiler firm versus growers are not part of the supply chain. Contract with subject broiler firm versus growers who are not part of the supply chain. Contract with subject broiler firm versus growers who contract for other firms or are not part of the supply chain.

1.000 1.000 1.000

0.00*** 0.00*** 0.00***

*

Notes: [1] Significance levels 10%. [2] p-Values reported up to two decimal places. [3] The distribution are for net profit per acre in Rs./month for gherkins, papaya and marigold and net profit in Rs./month for 5000 birds (or 5000 square feet of space). [4]  Denotes the cases for which detailed results are presented. The results for remaining cases are available with the author. ** Notes: Significance levels 5%. *** Notes: Significance levels 1%.

The treatment effect is then computed as the change in this variable associated with a change in contracting status, conditional on covariates. An exclusive focus on net profit, i.e. the assumption of separability, can be faulted for not taking the entire context of household decisions, the particular place of the contract commodity in a portfolio of crops or of its impact on other aspects of welfare. However, given that eliciting reliable data on incomes from households is notoriously difficult in the context of the study area, this was not pursued. Further, it is typical for farmers to treat the contract commodity as a cash crop substitute so that they allocate acreage either to the contract commodity or to an alternate cash crop. The assumption that the contracting crop does not alter the essential nature of the entire portfolio of crops is therefore reasonable in this case. The other caveat, which holds for most studies of this nature, is that the net profit per acre recorded in the year of survey is one draw from a distribution. It is therefore natural that these can vary greatly over different years, depending on a host of exogenous conditions such as weather, pest pressure and external market conditions. There is also a large variation across farmers. The results on outcomes are therefore to be interpreted with caution. In the context of the larger work, this is in fact illustrative of heterogeneity of farmer experiences across time, contributing to the dynamics or life cycle of contracting schemes. Table 1 presents the Kolomogorov–Smirnov for comparing net profit distributions for different categories of farmer. Consistently, the unconditional distributions of net profits for contract farmers (whether they contract exclusively for the subject firm or for any firm) is statistically significantly different from those of the control groups. To achieve identification, I use instruments that reflect farmers’ relative perceptions of contracting for high value commodity relative to its next best alternative as identified by the farmer. In the Farmer Survey, I elicit subjective distributions of net profits per acre that farmers associate with contracting versus the next best alternative that they have nominated. Relative moments of these subjective distributions and stochastic dominance of one over the other presumably influences whether they want to contract or not. Expressed as they are in relative terms, the link between risk perception levels and net profits is precluded. In addition, risk scores from psychometric mapping of relative benefits and risks that farmers associate with contract participation in high value agriculture and a self-identified the status quo are obtained. These risk scores are also relative measures and include aspects of participation in contracting arrangements for high

value commodities that are typically hard to express in monetary terms (for example, impact on health, the notion of self-respect, etc.).10 These two variables are catch-all measures that incorporate aspects like risk attitudes, assessments of farmers’ abilities and expectations with regard to uncertainties of nature, among other things. In particular, it represents the net incremental risk a farmer associates with contracting (net, because it factors in both the risk and benefits to contracting, and incremental, because the net risk from not participating is subtracted from the score) and hence indicates a farmer’s inclination towards contract participation in high value commodity chains. Clearly, these perceptions of relative benefits and risks of contracting over alternatives potentially drives selection, but cannot determine net returns per acre and can be used as instruments for identification. The survey data confirms that the correlation between outcome and the instruments are weak (with an absolute value below 0.3 for all the commodities and close to 0 for several). There is a case to be made however of potential endogeneity, that the most recent outcome influences perceptions, given that the survey collects information on the most recent outcome as well as perceptions after realization of net profits that season. I would argue that this endogeneity is weak at best. First, subjective distributions of profits were elicited for a twenty season (about six years) time frame, and for a typical year, so that while the most recent experience is surely incorporated, it is unlikely to drive the responses overwhelmingly. Also, the measures used in the selection equation are relative measures, of contracting and not contracting, so that the influence of the most recent experience is further muted. Fig. 1 plots the range of subjective net profits per acre per month elicited in the survey alongside the actual net returns, for contract and non-contract farmers. Whenever the range bar straddles the line of equality, the actual or realized net profit that season falls within the range of subjective expectations. It is evident that in a number of cases, observations lie outside the ‘typical’ range expected by farmers, indicating that the most recent actual outcome has not overwhelmingly influenced the range of typical 10 A Combined Risk Score is computed where a complete listing of all risk reducing and risk enhancing attributes the farmer associates with contracting and the next best alternative is used. The farmer assigns values between one and ten to each relevant attribute denoting the stated frequency of occurrence and another value to represent the criticality of the attribute to the farmer’s sense of well-being. The score uses these two values as weights to compute a summary number that denotes the net incremental risk of the farmer from choosing contracting over his or her self-declared next best alternative (2012).

S. Narayanan / Food Policy 44 (2014) 142–157

147

Fig. 1. Comparing subjective assessments with actual outcomes. Note: The dashed line reflects the line of equality.

subjective returns. Had that been the case, one would expect the actual outcomes to fall within the range for most farmers. In other words, if the season of the survey happened to be an extraordinarily good or bad year for the farmer, the figure suggests that farmers have not incorporated this in their assessments of subjective distributions of typical returns, suggesting that the instruments for identification are reliable. Owing to the specific problem of an absence of a precise counterfactual group, the approach of using the ‘next best alternative’ helps define the counterfactual for each farmer, should a farmer choose to persist with the stays quo ‘traditional’ cropping pattern or grow the contract commodity for other firms or the spot market. Rather than imposing a presumed cropping pattern, which would have weak empirical foundation and constitute a generalization, the farmer-defined alternative narrows down the crop that they are most likely to replace in order to contract or regard as the closest substitute for the high value commodity. This offers the strongest empirical foundation for creating a counterfactual outcome measure. Each farmer was able to single out a unique next best alternative and across the schemes, maize, sorghum, banana, paddy and tomato were named most frequently as the relevant alternative. The counterfactual outcomes pertain to net profits from growing these commodities. In order to get better estimates of the counterfactual net profits with lower error and a more precise representation of the traditional cropping pattern in the study area, the study uses a pooled comparison group that includes all farmers not participating any of the four contract commodities to increase the number of observations growing these traditional crops to serve as comparison groups for each of the contract commodities in question. Also, while the sampling approach was designed to be representative of the contract farmers for the subject firm and those who do not contract with the subject firm, the sample is neither representative of farmers who choose to grow each of these alternate crops nor of those who contract for the chosen contract commodities with other firms. Using all the data on non-participants from all four commodity schemes for estimation of the switching model is therefore desirable. While the parameters of the model are estimated using observations from a pooled group, the treatment effects are computed only for the comparison group selected specifically (or the commodity-specific control group) for each contract commodity. This is to ensure tighter comparison of outcomes. I use an additional instrument to enable pooling that is constructed by an interaction of an individual farmer’s coefficient of absolute risk aversion elicited though experiments recording the

bid price of a risky, fair bet11 with the coefficient of variation of the spot or alternate market price of the contract commodity for which the model is estimated.12 The higher the farmer’s risk aversion or the coefficient of variation of spot market price for a commodity, the greater would be the propensity of the farmer to opt to contract. This is an absolute number that allows for pooling farmers who might grow different crops. Despite existing evidence on the relationship between returns and risk coefficients or variability of prices (Foster and Rausser, 1991; Rao et al., 2012), there is nevertheless an empirical case for using the above instrument. Neither component term is strongly correlated with outcome and the interaction of price variability in the regional markets with the risk coefficients has a similar relationship. The correlation with the outcome is weak at best (an absolute value of no more than 0.2) and nearly 0 for marigold and papaya. This is true for each type of farmer represented in the survey. Also, the correlation of the instrument with inputs, be it family or hired labor, fertilizers, chemicals or sunk investments, are all comparably weak. This addresses to an extent the concern that the instrument could affect allocation of resources. That said, this instrument is not relied upon exclusively to achieve identification and is only used for gherkins where the competitive market is constituted by other contracting firms whose prices might influence the propensity of the sample farmer to contract with the subject firm. In the case of poultry, I use as an additional instrument, fixed costs on infrastructure as a driver of selection. As mentioned, the sunk cost in often an entry barrier and hence impacts selection. Also, total investment in shed only sets the scale of operations. There are no scale economies associated with the range of shed space farmers in the region can possibly achieve, implying that scale of investment cannot contribute to net profit per area operated. Furthermore, there is not much difference in the quality of these fixed investments that might affect net profits. It is difficult too to argue that it might be endogenous since the investment has already taken place, and is influenced by perceptions of incremental risk at the time of the investment and not influenced by current welfare outcomes. By the same logic, I use investment in

11 This was obtained from maximum price farmers would be willing to pay for a lottery that would fetch them, with equal probability, an amount equivalent to two days of wages for a male unskilled agricultural laborer (Rs.300) or one day’s wage for an unskilled male worker (Rs.150). 12 The prices for marigold and papaya were collected from secondary data and for gherkins and broiler, the distribution of actual realized prices obtained by the nonsubject firm growers in the survey season was used.

148

S. Narayanan / Food Policy 44 (2014) 142–157

irrigation facility in the selection equation for papaya, controlling for land owned that influences the ability to make such investments.13 Together, the instruments capture essential elements for a specific non-contract farmer to serve as a comparison group for any of the contract commodities. In the selection equation that sorts farmers into different regimes, apart from these instruments, I include specific farm and household characteristics, when it is of particular relevance to the contract commodity. Implicitly, these capture variables that are associated with attributes firms might be concerned about when choosing farmers and those that represent farmer willingness to contract. The attempt in this paper is not to model the selection process but rather to account for its correlates. To enable estimation that includes non-participant farmers across schemes, I control for the possibility that farmers are not afforded the opportunity to contract with the contracting firm by using the number of contract hamlets in the block where the respondent is located. This cannot serve as an instrument since firms would choose regions that are more suitable for growing the contract commodity and hence does influence outcomes, but it is important to control for spatial selection in order validate a pooled comparison group. If firms select spatially, as they often do, this would control for the sample contracting firm’s locational preference for sourcing supplies. Region fixed effects were ineffective since often the choice of contract regions coincides with schemes, or social group. Hence the number of hamlets was chosen over region dummy variables. In fact, this scores over the use of region dummy variable, since it captures the variation in the intensity of a firm’s presence within a particular region. For the outcome equation, I use a hybrid of a traditional production/profit function approach with those more commonly used in studies in contract farming. This is in part to judge the relative strength of association between inputs and profits across the two regimes, and to account fully for the fact that in the commodities studied contract growing almost always involves higher intensity of input use, be it family labor or fertilizers and pesticides. The outcome, net profit per acre, is treated as a function of the total area under contract cultivation, application of human labor, use of fertilizers (both chemical and farm yard manure), plant protection chemicals. It is also a function of farmer and farm characteristics that might be associated with entrepreneurial abilities, experience and so on. The estimated model varies across commodities in terms of the set of explanatory variables used. This was driven, in part, by what seemed relevant to the particular commodity. Farmer characteristics include age, social group, some indicator of education, either of the farmer or of the most educated member of the household, land owned, and distance from the nearest road. The availability of irrigation is represented by either the proportion of land irrigated or by an indicator variable for whether the farmer is dependent on rain. In the equation for outcomes, depending on the crop, input use, labor days over the season, of both hired and family labor is invariably included. Supervision enters in some cases as a binary variable. This is derived from the number of visits over the entire cropping season or growing cycle, since the use of the latter yielded unstable coefficients. If there has been any supervision at all in the past season, the variable carries a value of one and zero otherwise. Not all the variables are represented in all the equations. The models correct for heteroscedasticity. The standard errors are, on occasion, clustered at the village level to account for correlation in the errors. The inclusion of variables other than the instruments, to account for selection has broad relevance but is not meant to be a rigorous specification of the

13 Farmers were asked if they undertook any investment specifically to be enable to contract for papaya, and is hence not a generic investment.

covariates of selection. The summary statistics and results of the estimation are presented for each commodity in Tables 2–5. In this paper I present on type of comparison for marigold, papaya and broiler and estimate three different models illustrating a richer set of possible comparisons for gherkins (as shown in Table 1).

Estimated treatment effects Table 6 shows the average treatment effect for both the treated as well as for the untreated, depending on the nature of comparisons made. The treatment effect is measured as the incremental net profit per month in rupees (Rs.). For the former, it represents the average difference between the expected net profit for the treatment group and what they would have earned had they been in the alternate regime. For the latter, it represents the average difference in expected net profit for the comparison group of farmers had they been treated and the expected net profit when accruing in the comparison regime.14 These are computed for the commodityspecific sample to ensure tight comparisons. The table also shows the standard deviation of the distribution of the estimated treatment effects to emphasize that notwithstanding the sign of the average, particular farmers might gain significantly from contracting whereas others might be significantly worse off with contracting. The variation is in part reflective of the large variations in the predicted net profits earned in the season surveyed and partly from similar variations in the estimated counterfactuals. Participating in contracting arrangements for both papaya and broiler is very profitable relative to the status quo. For broiler and papaya, contracting increases net profits on average, for both contract farmers and those not currently contracting. Papaya contract farmers would have foregone 37% of their current net profits had they chosen not to contract for papaya, broiler growers would have lost one and a half times their average net earnings had they not opted to contract. For those not participating, entering papaya contracting would enhance a non-participating farmer’s net profit by 47% and if non-contract farmers were to take up contracting with the sample firm for broiler, they would earn net profits that are more 123% higher. In the case of marigold, the treatment effect on both the treated and untreated is negative, implying that contracting leaves both contract and non-contract farmers worse off in terms of net profit per acre per month. Marigold farmers could have earned a return that was fifty percent higher than their net profit from contracting had they not grown for the sample firm. Gherkins contract farmers (irrespective of which firm they supplied to) earned, on average, a higher net profit by virtue of choosing to contract. Had they not, they would have, on average, lost 21% of their net profit from contracting. At the same time, those who do not contract for the subject firm are better off not doing so. The other comparisons for gherkins illustrate the sensitivity of findings to the nature of comparisons. In particular, these results suggest that the identity of the firms matters. When the treatment group is redefined to include only those contracting with the subject firm, the treatment effects for both the treated and the comparison groups are consistently negative. This suggests that the sample firm selected for the study does not particularly offer a lucrative opportunity for farmers. On the other hand, competing firms that also contract for gherkins might represent much higher earnings for farmers. To put these results in perspective, those who did not grow gherkins often grew tomato or other horticultural crops, which fetched the farmers particularly good returns that season. As for 14 As mentioned, for gherkins and broiler, the treatment group represents contract farmers who contract with either the subject firm or any other firm.

149

S. Narayanan / Food Policy 44 (2014) 142–157 Table 2 Summary statistics and endogenous switching model results: marigold. Variable

Summary statistics

Estimation results

Non-contract farmers

Contract farmers

Mean or proportion

Standard deviation

Mean or proportion

Standard deviation

3565.1

6783.6

1705.1

1701.7

Coefficient

Robust standard error

z-Statistic

Regime 1: Contracting with subject or sample firm (D) Rainfed farm Fertilizer application (kgs.) Land owned (acres) Distance from surfaced road (kms.) Total hired labor (days) Total family labor (days) Age (years) (D) Family Education (1 = Illiterate) Constant

160.53 15.16 45.97 53.33 0.44 122.96 19 32.46 4274.66

381.13 5.07 42.43 106.89 3.25 52.25 17.95 350.2 1741.75

0.42 2.99*** 1.08 0.5 0.14 2.35** 1.06 0.09 2.45**

Regime 2: Not contracting with subject firm Land owned (acres) (D) Rainfed Fertilizer application (kgs.) Distance from surfaced road (kms.) Total hired labor (days) Total family labor (days) (D) Family illiterate Age (years) Constant

17.32 1544.48 0.6 71.52 10.63 7.32 750.48 28.89 2562.74

17.04 1040.3 1.09 139.22 10.28 7.28 2689.73 35.39 2144.39

1.02 1.48 0.55 0.51 1.03 1.01 0.28 0.82 1.2

1.38 0.003 0.05 0.02 0 0.03 0.01 0.02 0.01 0.1 0.06 1.13 1.19 0.15 1.21 7.89

0.77 0.001 0.02 0.01 0 0.01 0.01 0.33 0 0.16 0.19 0.66 0.59 0.16 0.36 2.75

1.79* 1.93* 2.5*** 2* 1.07 3*** 1 0.06 3.31*** 0.63 0.32 1.71* 2.02** 0.94 3.36*** 2.87***

Dependent variable: Net profit per acre per month (Rs.)

Regime selection (D) Rainfed Fertilizer application (kgs.) Land owned (acres) Distance from surfaced road (kms.) Total hired labor (days) Total family labor (days) Age (years) (D) Family illiterate Combined risk score X Marigold Scheme dummy Risk aversion to open market prices Difference in skewness (D) FOSD (D) SOSD Ratio of mean returns Number of contract hamlets in block Constant Number of observations Mills’ Ratio Regime 1 (mean) Mill’s Ratio Regime 2 (mean) q1v q2v Wald Test of Independence of Equations (chi2(1)) Log pseudo-likelihood Wald chi2(8) Prob > chi2

0 217 6 1 73 55 46 16 117 9 0 8 7 0.35 1

N.A. 260 11 7 78 62 11 N.A. 193 2 0.6 N.A. N.A. 0.9 1.6

208

0 201.2 5.2 1.6 142.7 14.5 45.3 53 601.7 6.4 0.3 25 14 0.8 4

N.A. 50.71 3.84 1.76 63.06 4.52 12.33 N.A. 169.1 1.29 1.02 N.A. N.A. 0.74 0.95

59

267 3.83 0.31 0.21 0.14 2.53*** 2675.51 23.51 0.003***

0.21 0.11

[1] (D) means dummy variable taking on the value 1 when the variable is true and 0 otherwise. [2] FOSD means contracting first order stochastic dominates next best alternative. [3] SOSD means contracting second order stochastic dominates next best alternative. [4] N.A. means not applicable. * 10% Significance level. ** 5% Significance level. *** 1% Significance level.

marigold, the price in the fresh flower market often shoots up and is typically higher than the contract price. Perhaps, the negative treatment effect reflects this effect. It is important to note that this is an average across farmers for just one season, so that it only represents a snapshot view that might not necessarily be robust to variations across years or seasons. In the gherkins contracting scheme, q1v is positive and statistically significant when comparisons are made with reference to the subject firm, i.e., models (2) and (3) (Table 6). This indicates that farmers who contract for gherkins have an absolute advantage in participating. They tend to have a higher than average net profit

whether or not they are contracting. In the broiler contracting scheme, there is clear evidence of hierarchical sorting. Both q1v and q2v are negative and statistically significant. This indicates that those who contract have better than average profits, irrespective of whether they contract or not, but are better off when contracting. Those who do not contract face below average profits in both regimes, and would be better off not contracting. This is indicative of exclusion of ‘low ability’ farmers. Papaya is similar to broiler in that both q1v and q2v are negative but neither is statistically significant indicating that selection is potentially exogenous. The coefficient of correlation is not statistically significant for marigold

150

S. Narayanan / Food Policy 44 (2014) 142–157

Table 3 Summary statistics and endogenous switching model results: papaya. Variable

Summary statistics

Estimation results

Contract farmers

Non-contract farmers

Mean or proportion

Standard deviation

Mean or proportion

Standard deviation

9289

4724.8

3565.1

6783.6

Coefficient

Robust standard error

z-Statistic

Regime 1: Contracting with sample firm Land owned (acres) Percentage of cultivated land that is irrigated Distance from surfaced road (kms.) (D) Supervision Total hired labor (days) Total family labor (days) Age (years) Plant protection (liters) (D) Family illiterate (D) Scheduled Caste/Tribe Number of crops grown annually per acre Constant

219.0 9.4 902.3 1146.0 0.5 8.0 27.0 1.8 5703.9 1957.7 454.3 9690.1

120.7 24.4 385.1 1081.6 8.2 16.2 48.8 0.6 4803.0 2462.8 1112.3 4282.9

1.8* 0.4 2.3** 1.1 0.1 0.5 0.6 3.2*** 1.2 0.8 0.4 2.3**

Regime 2: Not contracting with sample firm Land owned (acres) Percentage of cultivated land that is irrigated Distance from surfaced road (kms.) Total hired labor (days) Plant protection (liters) Total family labor (days) (D) Scheduled Caste/Tribe (D) Family member not studied beyond primary school Age (years) (D) District 1 (D) District 2 Number of crops grown annually per acre Constant

16.9 55.9 48.1 2.9 1.5 20.2 6100.4 1661.1 13.0 1951.2 5290.6 343.0 4061.2

49.5 15.4 69.1 7.6 1.3 11.7 6386.3 1187.4 45.2 1723.6 1575.0 357.6 3043.1

0.3 3.6*** 0.7 0.4 1.2 1.7* 1.0 1.4 0.3 1.1 3.4*** 1.0 1.3

0.01 0.01 1.83 0.00 0.01 0.01 0.00 0.25 0.66 0.91 0.79 0.16 20.22 19.36 0.00 0.23 4.96 0.19 1.03 0.15 0.76 0.00 0.17 21.97

0.0 0.0 0.5 0.0 0.0 0.0 0.0 1.5 0.5 1.2 0.3 0.1 115.0 115.0 0.0 0.2 1.7 0.2 0.8 0.8 0.3 0.0 0.1 115.1

1.1 0.2 3.6*** 1.2 2.5** 0.5 3.4*** 0.2 1.4 0.8 2.8*** 2.5** 0.2 0.2 1.1 1.3 2.9*** 0.8 1.2 0.2 2.6** 1.5* 1.5 0.2

Dependent variable: Net profit per acre per month (Rs.)

Regime selection Percentage of cultivated land that is irrigated Distance from surfaced road (kms.) (D) Supervision Total hired labor (days) Total family labor (days) Age (years) Plant protection (D) Family illiterate (D) Family member not studied beyond primary school (D) Scheduled Caste/Tribe Number of crops grown annually per acre Land owned (acres) (D) District 1 (D) District 2 Combined risk score X Papaya dummy Risk aversion to open market prices Ratio of coefficient of variation in returns Difference in Skewness (D) FOSD (D) SOSD Ratio of mean returns Sunk Cost (Rs.‘0000) Number of contract hamlets in block Constant Number of observations Mills’ Ratio Regime 1 (mean) Mill’s Ratio Regime 2 (mean) q1v q2v Log pseudo-likelihood Wald chi2(11) Prob > chi2 LR Test of independent equations

88.30 0.80 2.80 63.20 96.20 45.10 962.20 1.00 11.11 6.00 0.64 5.70 80.56 19.44 96.10 4.10 1.06 0.00 7.00 36.00 1.00 6708.33 3.00 72

20.10 1.40 N.A 82.40 44.80 9.50 1072.80 N.A N.A N.A 0.53 5.00 N.A N.A 186.20 1.30 0.82 0.30 N.A N.A 1.00 27135.26 2.60

76.20 1.40 3.80 72.70 55.40 46.30 212.30 8.00 24.04 0.00 1.34 6.20 39.42 33.17 13.90 4.70 0.18 0.00 7.00 7.00 1.80 701.92 0.80

30.30 6.70 N.A 77.70 62.20 11.10 502.90 N.A N.A N.A 1.72 10.80 N.A N.A 63.20 1.00 0.53 1.10 N.A N.A 2.70 4103.45 1.30

208

[1] (D) means dummy variable taking on the value 1 when the variable is true and 0 otherwise. [2] FOSD means contracting first order stochastic dominates next best alternative. [3] SOSD means contracting second order stochastic dominates next best alternative. [4] N.A. means not applicable. * 10% Significance level. ** 5% Significance level. *** 1% Significance level.

267 6.14 0.52 0.15 0.03 2717 47.64 0.00*** 252.99***

0.40 0.34

151

S. Narayanan / Food Policy 44 (2014) 142–157 Table 4 Summary statistics and endogenous regime switching model: broiler. Variable

Dependent variable: Net profit per acre per month (Rs.)

Summary statistics

Estimation results

Non-broiler farmers

All broiler contract farmers (any firm)

Mean or proportion

Standard deviation

Mean or proportion

Standard deviation

690

1032

11,729

1958

Regime 1: Contracting for broiler with any firm Land owned (acres) Percentage of cultivated land that is irrigated Distance from surfaced road (kms.) Age (years) Total hired labor (days) Total family labor (days) Constant Regime 2: Not contracting with any firm, not growing broiler Age (years) Land owned (acres) Percentage of cultivated land that is irrigated Total hired labor (days) Total family labor (days) Distance from surfaced road (kms.) Constant Regime selection Land owned (acres) Percentage of cultivated land that is irrigated Distance from surfaced road (kms.) Age (years) Total hired labor (days) Total family labor (days) Combined risk score X Broiler dummy Sunk cost (Rs. ‘0000) Number of contract hamlets Constant Number of observations Mills’ Ratio Regime 1 (mean) Mill’s Ratio Regime 2 (mean) q1v q2v Wald Test of Independence of equations chi2(1) Log likelihood Wald chi2(5) Prob > chi2

6.2 76.2 1.4 46.3 72.7 55.4 7.5 702 2.1

10.8 30.3 6.7 11.1 77.7 62.2 50.2 4103 3.6 208

6.9 59.5 0.3 46.1 107.5 111.6 194.6 83,369 6.8

6.3 34.7 0.7 11 87.4 66.7 97.3 38,842 4

Coefficient

Robust standard error

z-Statistic

48.87 10.38 153.95 13.15 4.38 21.41 7935.9

25.49 5.65 129.87 14.86 2.15 3.57 981.34

1.92* 1.84* 1.19 0.88 2.04** 6*** 8.09***

5.02 10.57 9.7 1.56 0.77 3.83 507.39

5.87 3.5 2.04 1.29 0.89 13.05 299.88

0.86 3.02*** 4.75*** 1.21 0.87 0.29 1.69*

0.08 0.04 1.38 0.01 0.02 0.01 0.02 0.00 0.26 2.48

0.66 0.10 4.17 0.72 0.01 0.02 0.14 0.00 4.56 41.44

0.13 0.42 0.33 0.01 2.84** 0.70 0.15 2.32** 0.06 0.06

81

289 6.77 0.58 0.99 0.98 0.03 2431.53 55.76 0***

*

10% Significance level. 5% Significance level. *** 1% Significance level. **

and for model (1) in gherkins and hence, here too selection is possibly exogenous. This is broadly consistent with the current operational status of the schemes, as evident from interviews with agribusinesses. For example, for the marigold scheme, the contracting arrangements are not tight in the sense that sidesale to the spot market is very high. This muddies any evidence of sorting. Again for papaya, this year saw a catastrophic loss of the papaya crop to mealybug infestation, again rendering inferences regarding sorting murky.15 Apart from the fact that this is evidence for only one season and that these average out substantial heterogeneity across farmers, it might reflect the prices in the markets for the other crops that farmers in the comparison group grew, tomato, for instance. The 15 The coexistence of the beneficial treatment effect and the catastrophic loss is partly on account of the timing of the survey. The actual net profits recorded for papaya were for the preceding year, which was then converted to net profit per acre per year. At the time of the survey, the mealybug epidemic had been affecting crops for about three months. Since papaya latex extraction is a continuous process, the outcome of interest captures a mixture of high yields and low yields. Without the mealybug attack, papaya contract farmers are likely to have benefited much more from papaya contracting, and there might have been stronger evidence on sorting.

treatment effect in this case can easily switch signs depending on market conditions. Treatment effects that measure an average impact on the set of farmers can potentially mask the heterogeneity of farmer experiences. The large standard deviations in the point estimates of the average treatment effects across the commodity schemes reflect on the one hand variation in net profits accruing to the farmers in the survey season but also the range of potential impact of contracting for the various farmers. They are key to understanding the origins of the dynamics of contract farming arrangements. The large deviations suggests that some contract farmer might make losses from participation in high value commodity chains (even as others make profits) and might be better off opting out of the regime, while there are some non-contract farmers who might have an incentive to participate in contracting arrangements. This opens up the possibility of farmers reassessing their decisions to contract or taking specific actions to enter into transactional arrangements with firms. Indeed, evidence from the same survey suggests that not only is there significant farmer attrition, but farmers also choose to opt in and out of the contracting schemes so that episodes of participation are wedged between episodes of non-participation (Narayanan, 2013).

Variable

152

Table 5 Summary statistics and endogenous switching model results: gherkins. Summary statistics (for model 1)

Estimation results

Non-gherkins contract farmers

All gherkins contract farmers (any firm)

(1) All gherkins contract farmers (any firm) versus non-gherkins contract farmers

Mean or proportion

Standard deviation

Mean or proportion

Standard deviation

Coefficient

Robust standard error

Coefficient Robust zStatistic standard error

zCoefficient Robust Statistic standard error

zStatistic

6945.7

3282.8

5058.3

1128.29 3849.17 102.26 1920.44 79.28 75.66

970.42 1120.09 107.05 1022.73 45.41 199.91

1.16 3.44*** 0.96 1.88* 1.75* 0.38

Dependent variable: Net profit per acre per 3453.5 month (Rs.)

Regime 2 (D) Contract with a non-subject firm Land owned (acres) Distance from surfaced road (kms.) Age (years) Number of years of education of the most educated family member Proportion of land owned under irrigation (%) Fertilizer application (kgs.) Plant protection Total hired labor (days) Total family labor (days) Constant Regime Sorting (D) Contract with a non-subject firm (D) Contract with Subject Firm (D) Supervision Land owned (acres) Distance from surfaced road (kms.) Age (years) Number of years of education of the most educated family member Proportion of land owned under irrigation (%) Fertilizer application (kgs.) Plant protection Total hired labor (days)

5.68

35.45

(3) Gherkins contract farmers with subject firm versus other farmers (both non-gherkins contract farmers and those who contract for gherkins with other firms)

145.33 1617.51 56.64 274.36

143.39 754.97 130.73 385.60

1.01 2.14** 0.43 0.71

143.64 1627.66 56.22 267.36

115.41 751.01 126.58 359.79

1.24 2.17** 0.44 0.74

0.16

1.44 3.22 12.94 35.83 5114.41

4.46 4.81 49.78 14.31 3572.92

0.32 0.67 0.26 2.50** 1.43

5.21 2.64 66.01 45.44 4639.63

8.55 3.97 46.16 26.20 5811.19

0.61 0.67 1.43 1.73* 0.80

5.18 2.67 65.93 45.53 4676.58

8.47 3.99 46.47 25.09 5772.46

0.61 0.67 1.42 1.81* 0.81

47.68 65.46 49.61 125.19

46.64 72.81 47.85 132.71

1.02 0.90 1.04 0.94

42.02 60.84 75.54 81.82

26.06 136.85 75.64 164.05

1.61 0.44 1.00 0.50

4889.53 41.72 67.32 62.64 77.17

3624.17 24.15 137.69 58.08 154.16

1.35 1.73* 0.49 1.08 0.50

33.82759

36.61658

0.92

0.92 2.52 6.15 15.21 2919.05

2.17 1.16 7.18 13.70 2801.27

0.42 2.17** 0.86 1.11 1.04

3.21 1.95 10.46 18.78 4604.00

2.81 1.03 11.29 28.57 3211.77

1.14 1.89* 0.93 0.66 1.43

2.97 1.83 12.10 22.97 3754.29

2.08 0.98 11.13 18.20 2953.60

1.43* 1.86 1.09 1.26 1.27

N.A N.A N.A 6.74 1.51 46.91 10.66

N.A N.A N.A 11.36 7.07 11.11 4.16

29.87 70.13 36.36 2.25 0.39 38.45 8.92

N.A N.A N.A 4.58 0.49 9.28 2.59

6.507 1.813 0.158 1.614 0.037 0.170

3.654 3.299 0.088 0.352 0.012 0.066

1.78* 0.55 1.80* 0.008 4.58*** 0.067 3.05*** 0.020 2.58** 0.180

11.0224

36.37

14.24

33.20

14.41

0.190

0.021

9.25***

172.78 157.33 75.19

231.59 496.88 81.80

535.37 600.92 46.40

166.67 257.86 19.57

0.002 0.001 0.028

0.001 0.000 0.005

2.84*** 1.96* 5.28***

0.002 0.000 0.016

8.8505

1.25

0.092 0.080 0.053 0.183

0.09 0.83 0.38 0.99

0.007 0.067 0.021 0.180

0.069 0.071 0.042 0.147

0.10 0.94 0.49 1.23

0.001 0.001 0.005

1.64 0.25 3.38***

0.002 0.000 0.016

0.001 0.001 0.005

1.93* 0.28 3.34***

S. Narayanan / Food Policy 44 (2014) 142–157

Regime 1 (D) Contract with Subject Firm (D) Supervision Land owned (acres) Distance from surfaced road (kms.) Age (years) Number of years of education of the most educated family member Proportion of land owned under irrigation (%) Fertilizer application (kgs.) Plant protection Total hired labor (days) Total family labor (days) Constant

(2) Gherkins farmers contracting with subject firm versus nongherkins contract farmers

Total family labor (days) Combined risk score X Dummy for Gherkins Risk aversion to alternate prices of firms Ratio of mean returns from contracting over next best alternative Number of contract hamlets in block Constant Number Mills’ ratio regime 1 (mean) Mill’s ratio regime 2 (mean) q1 q2 Wald chi2(8) Prob > chi2 Log pseudo-likelihood LR test of independence of equations

40.90 142.37 1.12 0.69

46.20 221.65 0.25 2.44

155.40 33.75 1.33 3.16

48.64 167.94 0.11 4.50

2.05

4.15

8.09

3.92

185

0.026 0.000

0.003 0.001

7.42*** 0.25

0.149

0.034

4.44***

0.766 17.328

0.090 1.941

8.53*** 0.133 8.93*** 10.574

77

261 3.88 1.39 0.98 0.99 48.91 0 2634.94 chi 2(1) = 40.76***

0.009 0.001 2.705 0.043

2.27** 2.37** 2.36** 1.24

0.020 0.002 6.466 0.056

0.008 0.001 2.701 0.042

2.59** 2.29** 2.39** 1.34

0.050 6.080

2.69*** 1.74*

0.136 10.688

0.049 5.652

2.77*** 1.89*

238 3.71 0.40 0.68 0.88

0.37 0.74 14.61 0.07 2445.63 1.78

261 4.42 0.31 0.68 0.86

0.37 0.63 14.8 0.0631 2674.46 1.81

S. Narayanan / Food Policy 44 (2014) 142–157

[1] (D) means dummy variable taking on the value 1 when the variable is true and 0 otherwise. [2] FOSD means contracting first order stochastic dominates next best alternative. [3] SOSD means contracting second order stochastic dominates next best alternative. [4] N.A. means not applicable. * 10% Significance level. ** 5% Significance level. *** 1% Significance level.

0.020 0.002 6.392 0.053

153

154

S. Narayanan / Food Policy 44 (2014) 142–157

As mentioned earlier in the paper, the switching model offers a way to compare the above structural differences, in a limited way, between contract growers and non-contract growers in different schemes and also if there exist firm specific differences when different farmers contract with different firms. All contracting firms offer technical advice and supply inputs and. variables representing these aspects reflect the contribution of contractual inputs provided by the firm. Not only does this vary across firms but even within the same scheme, the extent of support could vary across farmers. The data in this study is not adequate to make detailed comparisons of contracts of the subject firm versus other firms, other than for gherkins (Table 5). For gherkins, farmers located farther from the road are associated with greater net profit per acre; so do inputs of family labor. Perhaps the chief attribute that differs across firms is the supervision and extension aspect. Results for model (1) in Table 5 suggest

that controlling contracting with the subject firm is negative but not significant. However, supervision is associated with higher net profits. Monitoring the production process is a very important aspect in the gherkins contracting scheme and is valued highly by the farmers and the results are therefore consistent with qualitative material from the interviews. For other commodities, although this separation of impact of contractual attributes from conjoint attributes is difficult, it is still possible to comment on the relationships between some of the inputs and the outcomes across the different regimes. Returns to fertilizer use (an input provided by the firm) and family labor are statistically significant for marigold contract farmers, whereas these do not seem to matter to farmers who do not contract for marigold. In the case of papaya, contract farmers earn higher net profit with plant protection. This is most likely owing to the mealybug

Table 6 Treatment effects and regime sorting. Variable

Nature of comparisons

Treatment effect on the treated Marigold Contract with subject (and the only) firm versus farmers who grow marigold for the spot market or an alternate crop. Papaya Contract with subject (and the only) firm versus farmers who grow table varieties of papaya or are not part of the papain supply chain Broiler Contract with any broiler firm versus growers are not part of the supply chain. Gherkin (1) All gherkins contract farmers (any firm) versus non-gherkins contract farmers (2) Gherkins farmers contracting with subject firm versus non-gherkins contract farmers (3) Gherkins contract farmers with subject firm versus other farmers (both non-gherkins contract farmers and those who contract for gherkins with other firms) Treatment effect on the untreated Marigold Contract with subject (and the only) firm versus farmers who grow marigold for the spot market or an alternate crop. Papaya Contract with subject (and the only) firm versus farmers who grow table varieties of papaya or are not part of the papain supply chain Broiler Contract with any broiler firm versus growers are not part of the supply chain. Gherkin (1) All gherkins contract farmers (any firm) versus non-gherkins contract farmers (2) Gherkins farmers contracting with subject firm versus non-gherkins contract farmers (3) Gherkins contract farmers with subject firm versus other farmers (both non-gherkins contract farmers and those who contract for gherkins with other firms) Correlation coefficients in the switching models

Marigold Papaya Broiler Gherkin

Contract with subject (and the only) firm versus farmers who grow marigold for the spot market or an alternate crop. Contract with subject (and the only) firm versus farmers who grow table varieties of papaya or are not part of the papain supply chain Contract with any broiler firm versus growers are not part of the supply chain. (1) All gherkins contract farmers (any firm) versus non-gherkins contract farmers (2) Gherkins farmers contracting with subject firm versus non-gherkins contract farmers (3) Gherkins contract farmers with subject firm versus other farmers (both non-gherkins contract farmers and those who contract for gherkins with other firms)

Mean incremental income (Rs.) 1577

Standard deviation of distribution of point estimates 1334

Average treatment effect as a proportion of average actual net profit 0.49

Number of observations

59

2802

3497

0.32

71

11,064

1241

1.5

81

703

3199.5

0.21

77

8056

2353

2.9

54

7979

2141

2.87

54

4167

2558

1.29

62

4948

2081

0.57

27

9032

1339

1.23

57

1420

2388

0.58

38

6016

2929

2.47

38

4570

3335

1.43

61

Regime 1:q1v

Regime 2: q2v

Estimate

Standard error

0.21

0.21

Estimate 0.14

0.11

0.15

0.4

0.99

0.42**

0.98

0.7

0.99

0.97

0.98

0.98

0.68 0.68

0.03

Standard error

0.34

**

0.88

0.74

**

0.86

0.63

0.37 0.37

[1] For gherkins and marigold all costs are for the most recent season completed, for one acre, which spans three months. [2] For papaya, this is an annual figure that has been converted to an equivalent per month per acre. [3] For broiler, this is a monthly figure per 5000 birds or 5000 square feet of shed space. [4] All treatment effects are computed and averaged over the sample within commodities. ** 5% Significance level.

S. Narayanan / Food Policy 44 (2014) 142–157

infestation during the season surveyed. As with gherkins, farmers located farther away from the road tend to earn higher net profit per acre than those who are closer to a road. Larger farms are associated with less net profit per acre from the contract farm. This is presumably due to the managerial demands made as the scale of operation increases. For those who do not contract for papaya, family labor inputs and the proportion of land that is irrigated are associated with higher returns. Sources of welfare gains: The structure of costs and returns The heterogeneity of levels in treatment effects goes hand in hand with the sources of these gains (or losses) from contracting. This section undertakes a simple decomposition of costs and returns to identify whether incremental incomes for contract farmers come from higher prices for the produce (being high-value crops as compared to the substitute) or via savings in transactions costs. It then assesses the returns to key factors of production across regimes that come from the estimation of the endogenous switching models. Tables 7 and 8 indicate that other than for marigold, contract farmers, irrespective of whether they contract with the sample firm or any other firm, earn higher net returns on average. This is despite higher costs associated with contract growing. This conforms to several previous studies that examine returns and cost structures in India. Singh (2007) and Gulati et al. (2008) review these in some detail. In general, findings suggest that the contract growing is associated with much higher costs of cultivation, 17–24% in potato contract farming in Haryana (Tripathi and Singh, 2005) and for tomato in the Punjab (Kumar, 2007; Dileep et al., 2002), but also higher gross and net returns driven in part Table 7 Cost of cultivation: comparisons across schemes and farmer groups for broiler (Indian Rupees). Variable Returns Net profit per month per 5000 birdsa Net return (per cycle of six weeks) per 5000 birds Gross Return (per cycle of six weeks) per 5000 birds Recurring costs per six week cycle Total costs Total labor cost (as % of total cost) Total other costs (as % of total cost) Fixed Costs Fixed Costs Labour per six week cycle Male hired labor(days) Male family labor(days) Female hired labor (days) Female family labor (days)

Subject contract farmers

Other contract farmers

11602 23205

12117** 24234**

34839

34684

11635 3189 27

10451 2061*** 20

8446 73

8390 80

82700

85411

38 14 16 6 2

33 9* 17 2** 4*

[1] All costs and returns are in Indian Rupees. [2] Feed and chick costs are excluded from estimate of costs since the costs are borne by the contracting firm. The return is computed using grower charges, net of the costs of feed and chick for all farmer categories. [3] The never contract farmers here grow other crops and are not presented in this table. a This is obtained by dividing the per cycle figure by 2, since the cycle of six weeks is combined with two weeks to prepare for the next cycle. * 10% Significance level. ** 5% Significance level. *** 1% Significance level.

155

by higher yields and in savings in transactions costs. Examples are gherkins (hybrid cucumber) in Andhra Pradesh (Haque, 2000; Dev and Rao, 2005), tomato in Punjab (Haque, 2000; Rangi and Sidhu, 2000) and Haryana (Dileep et al., 2002). Contract farming, when it involved a switch from traditional crops, gave much higher (almost three times) gross returns compared with that from the traditional crops of wheat, paddy in a study of tomato (Rangi and Sidhu, 2000). Studies show too that transactions costs were over 20% lower for contract milk and vegetable producers (Birthal et al., 2005). In several cases, contract farmers emerged with larger net returns per unit area of contract crop relative to those who were not contracting or grew traditional crops. A notable feature in the gherkins sample is that those who contract with the subject firm do worse than those who contract for other firms. The decomposition exercise suggests that the cost structures are not statistically significantly different between these two groups. Yet, the average gross returns is significantly higher among farmers associated with firms other than the subject firm. Given that the contract prices offered by these firms are all comparable, this is best explained by the timeliness of harvests or the preferences of the contracting firm. In general, the smallest gherkins fetch the highest price and contract firm offer a schedule of prices for five grades based on size with increasingly larger gherkins fetching lower and lower prices. One plausible explanation for the source of this difference is that the subject firm picked up a larger proportion of bigger gherkins (because of the nature of their client’s orders) or that the farmers contracting with other firms managed timely harvests and were able to deliver a larger proportion of smaller gherkins. As for fixed investments, broiler contracting requires large fixed investments in sheds to house birds, drinkers and feeders and so on (Table 7). The chicks and feed are provided by the firm. In the schemes studied, typically, the farmers procure medicines and take care of the maintenance expenses. Typically, women are far less involved in broiler production than men. The other commodities do not require much fixed investment, although for gherkins and papaya, most invested in either irrigation facilities or in spraying machines (Table 8). Input costs of gherkins tend to be high, owing to heavy use of fertilizer, pesticides and micronutrients. This is not the case for marigold and papaya where contract farmers make do with farm yard manure and minimal fertilizers. An interesting contrast is the use of labor (Table 8). Gherkins farmers rely heavily on family labor, and it is clear that relative to farmers who do not contract, they use far greater labor per three months, owing to the demands of harvesting in time and in the application of inputs, trellising and so on. In the case of marigold, there is much greater reliance on hired labor. Here too contracting implies a greater need for labor, mainly for harvesting. Papaya and broiler require very little labor in general. As is to be expected the costs associated with transactions, marketing, transport, commissions is typically zero for contracting farmers, while non-contracting farmers do incur these expenses. These differences emphasize the heterogeneity across crops just as the distribution of treatment effects pointed to differences in farmer experiences. Both emphasize a need to acknowledge these differences in studying the instrumentality of contract farming in transforming smallholder livelihoods.

6. Concluding remarks Assessing the extent of profit gains to participation from contract farming arrangements in high value agriculture is important to be able to make a case for promoting contract smallholder inclusion in high value supply chains, whether for retailing or processing. Empirical accounts of contract farming schemes in developing

156

S. Narayanan / Food Policy 44 (2014) 142–157

Table 8 Cost of cultivation: comparisons across schemes and farmer groups for gherkins, marigold and papaya (Indian rupees) Variable

Gherkins

Marigold

Papaya

Subject contract farmers

Contract farmers with other firms

Attrition farmers (not growing gherkins anymore)

Never contract farmers

Subject contract farmers

Other marigold farmers

Attrition farmers (not growing marigold anymore)

Returns Net profit per acre (per month) Net profit per acre (per season) Gross Return per acre

2780 8340 28919

4463* 13388* 34706*

2623 7870 15296***

2308 6924 12981***

1705 5115 25495

5168*** 15504*** 44514***

Recurring costs Total cost Input costs (as % of total cost) Operations cost (as % of total cost) Marketing Costs Total labor cost (as % of total cost) Fixed costs

20578 10597 51 9284 45 0 696 3 10685

21317 10704 50 9781 46 0 833 4 6348

7427*** 3872*** 52 2274*** 31 1228*** 53*** 1 0

6057*** 2584*** 43 2193*** 36 1138*** 142*** 2 0

20380 8932 44 1320 6 0 10127 50 719

192 64 1 84 43

224 76** 1 96* ** 52

110 38*** 2 70** *** 0

90 43*** 2** 44*** 0

157 4 72 10 70

Fixed cost Labor Female family labor (days) Female hired labor (days) Male family labor(days) Male hired labor(days)

Never contract farmers

Subject contract farmers

Never contract 2farmers

5630** 16890** 35190***

3243 9729 31497

9289

6934*

13215

10602*

29010*** 12462*** 43 1822*** 6 0 *** 14726 51 0

18300*** 5788*** 32 5189*** 28 0 7324*** 40 0

21768 9270 43 3486*** 16 0 9011* 41 0

3926 1028 26 147 4 33 2718 69 6708

3618 1596** 44 265** 7 624*** *** 1134 31 0

173 7*** 73 8*** 85**

137 0*** 51** 0*** 86*

133 0*** 47** 0*** 85*

159 8 32 88 31

103 4* 28 47*** 23

[1] For gherkins and marigold all costs are seasonal figures adjusted to per month per acre [2] For papaya, this is an annual figure that has been converted to an equivalent per month per acre. [3] All costs and returns are in Indian Rupees. * 10% Significance level. ** 5% Significance level. *** 1% Significance level.

countries however not only suggest high mortality rates but also that schemes have high farmer exit or attrition rates, indicating that farmer experiences might be variable. The findings of the paper underscore the variability in the treatment effects not only across contract farming schemes but also between farmers within a particular scheme. The presence of contract farmers who in fact might be better off not contracting or of those not contracting but who have much to gain from contracting holds the possibility of attrition and expansion (or more broadly, churning) in a contracting firm’s portfolio of supplier farmers. This is key to understanding the dynamics of farmer participation and selection in the broader context of modern agro-food chains. The net gains or losses that are associated with the participation suggest a complex pattern of sorting into schemes. While in broiler, it seems clear that farmers who do better than the average are selected, for gherkins it is evident that farmers who opt out of contracting do not have a comparative advantage participating. The diversity and heterogeneity in sorting is valuable from a public policy perspective. Considerable policy attention has been devoted to addressing the challenge of making markets work for the poor smallholders or identifying interventions that would enable inclusive high value agricultural supply chains. While this focus is justified, it is important to bear in mind that it is not self-evident that inclusion will improve farmer incomes unequivocally let alone to the same extent across all farmers. It is important to recognize that there are diverse groups of farmers. Only a subset of them fare well participating in modern supply chains, others are likely to fare poorly irrespective of whether they participate or not. It is when farmers can do better with contracting, but are rationed out by the firm, that exclusion of farmers from contracting arrangements becomes a policy concern. As long as farmers opt out voluntarily, on account of perceived risks or because they fare better when not participating, there is less cause for concern with regard to farmer capacity to participate in contract farming arrangements. An additional dimension is that of time. Owing to

stochastic nature of the profit stream, even for the same farmer, the profits from participation could vary considerably over a number of seasons. One needs to be cautious about generalizing experiences from cross sectional data. But it also suggests policy focus should incorporate more effectively concerns of farmers’ staying power within these contract arrangements in ways that they are able to withstand momentary shocks for the sake of long term gains or gain in the short term without bearing the possible negative consequences over the longer term. These questions deserve attention from both academics and policy makers.

References Adamchik, V.A., Bedi, A.S., 2000. Wage differentials between the public and the private sectors: evidence from an economy in transition. Labour Econ. 7, 203– 224. Barrett, C., Bellemare, M.F., Bachke, M., Michelson, H., Walker, T., 2012. Smallholder market participation in evolving agricultural value chains: comparative evidence from three continents. World Dev. 40 (4), 715–730. Bellemare, M.F., 2012. As you sow, so shall you reap: the welfare impacts of contract farming. World Dev. 40 (7), 1418–1434. Birthal, P., Joshi, P., Gulati, A., 2005. Vertical coordination in high-value commodities. In: Markets, Trade and Institutions Division (MTID) Discussion Paper 85, International Food Policy Research Institute (IFPRI), Washington, DC. Cadot, O., Olarreaga, M., Dutoit, L., 2005. How costly is it for poor farmers to lift themselves out of subsistence. In: CEPR Discussion Paper. 5392. Centre for Economic Policy Research, London. Cai, J., Ung, L., Setboonsarng, S., Leung, P., Rice contract farming in cambodia: empower ing farmers to move beyond the contract towards independence. Discussion Paper No.109, Asian Development Bank Institute. Dev, S.M., Rao, C., 2005. Food processing and contract farming in Andhra Pradesh: a small farmer perspective. review of agriculture. Econ. Polit. Weekly 40 (26), 2705–2713. Dileep, B.K., Grover, R.K., Rai, K.N., 2002. Contract farming in tomato: an economic analysis. Indian J. Agric. Econ. 57 (2), 179–210. Dutoit, L.C., 2007. Heckman’s selection model, endogenous and exogenous switching models: a survey. . (accessed 12.12.10). Foster, W.E., Rausser, G.C., 1991. Farmer behavior under risk of failure. Am. J. Agric. Econ. 73 (2), 276–288.

S. Narayanan / Food Policy 44 (2014) 142–157 Fuglie, K.O., Bosch, D.J., 1995. Economic and environmental implications of soil nitrogen testing: a switching-regression analysis. Am. J. Agric. Econ. 77, 891– 900. Gulati, A., Ganguly, K., Landes, M.R., 2008. Toward contract farming in a changing agri-food sys tem. In: Contract Farming in India: A Resource Book. ICAR, IFPRI, USDA, New Delhi. Haque, T., 2000. Contractual arrangements in land and labour markets in rural India. Indian J. Agric. Econ. 55 (3), 233–252. Kumar, P., 2007. Contract farming through agribusiness firms and state corporation: a case study in Punjab. Econ. Polit. Weekly 41 (52), 5367–5375. Lee, L.F., 1978. Unionism and wage rates: a simultaneous equations model with qualitative and limited dependent variables. Int. Econ. Rev. 19, 415–433. Lokshin, M., Sajaia, Z., 2004. Maximum likelihood estimation of endogenous switching regression models. Stata J. 4, 282–289. Maddala, G.S., 1983. Limited-dependent and Qualitative Variables in Econometrics. Cambridge University Press, Cambridge, UK, New York. Maertens, M., Swinnen, J.F.M., 2009. Trade, standards, and poverty: evidence from Senegal. World Dev. 37, 161–178. Manrique, J., Ojah, K., 2003. The demand for housing in Spain: an endogenous switching regression analysis. Appl. Econ. 35, 323–336. McCullough, E., Pingali, P.L., Stamoulis, K.G. (Eds.), 2008. The Transformation of the Agri–Food Systems: Globalization, Supply Chains and Smallholder Farmers. Earthscan, London. Michelson, H.C., 2013. Small farmers, NGOs, and a Walmart World: Welfare effects of supermarkets operating in Nicaragua. Am. J. Agric. Econ. 95 (3), 628–649. Minot, N., 2008. Contract farming in developing countries: patterns, impact, and policy implications. In: Pinstrup-Andersen, P., Cheng, F. (Eds.), Case Studies in Food Policy for Developing Countries, vol. 6–3. Cornell University Press, Ithaca, NY. Minten, B., Randrianarison, L., Swinnen, J.F.M., 2009. Global retail chains and poor farmers: evidence from Madagascar. World Dev. 37, 1728–1741.

157

Miyata, S., Minot, N., Hu, D., 2009. Impact of contract farming on income: linking small farmers, packers, and supermarkets in China. World Dev. 37, 1781–1790. Narayanan, Sudha., 2013. Smallholder attrition in contract farming schemes in India: extent, causes and concerns. Food Chain 3 (3), 155–170. Narrod, C., Roy, D., Okello, J., Avendao, B., Rich, K., Thorat, A., 2009. Public-private partnerships and collective action in high value fruit and vegetable supply chains. Food Policy 34, 8–15. Rangi, P.S., Sidhu, M.S., 2000. A study on contract farming of tomato in Punjab. Agric. Market. 42 (4), 15–23. Rao, E.J., Qaim, M., 2011. Supermarkets, farm household income, and poverty: insights from Kenya. World Dev. 39 (5), 784–796. Rao, E.O.J., Bruemmer, B., Qaim, M., 2012. Farmer participation in supermarket channels, production technology, and efficiency: the case of vegetables in Kenya. Am. J. Agric. Econ. 94 (4), 891–912. Reardon, T., Gulati, A., 2008. The supermarket revolution in developing countries. Policies for Competitiveness with Inclusiveness. IFPRI Policy Brief 2. International Food Policy Research Institute, Washington, DC. Simmons, P., Winters, P., Patrick, I., 2005. An analysis of contract farming in East Java, Bali, and Lombok, Indonesia. Agric. Econ. 33, 513–525. Singh, S., 2007. Leveraging contract farming for improving supply Chain efficiency in India: some innovative and successful models. In: Batt, P.J. (Ed.), Proceedings of the Second International Symposium on Improving the Performance of Supply Chains in the Transition Economies, Acta Horticulturae, No. 794, ISHS, Belgium, August 2008, pp. 317–323. Swinnen, J.F.M., 2007. Global Supply Chains, Standards and the Poor: How the Globalization of Food Systems and Standards Affects Rural Development and Poverty. CABI, Wallingford, UK and Cambridge. Tripathi, R.S., Singh, S., 2005. Contract farming in potato production: an alternative for managing risk and uncertainty. Agric. Econ. Res. Rev. 18, 47–60.