Dairy supply chain modernisation in Poland: what about those not keeping pace?

Abstract

1. Introduction

Profound restructuring has taken place in the Polish dairy sector during the transition period from a centrally planned economy to a market economy. Three phenomena are particularly noteworthy. The most striking is the huge outflow of people from the dairy sector. Second, throughout the transition period, more than half of dairy farms have been producing milk for their own consumption and/or selling directly to local consumers and not delivering milk to the dairy industry. Third, the dairy supply chain has been thoroughly reorganised. A modern marketing channel has emerged, based on direct contact between a farm and a dairy. As a result, a number of dairy farms have had to make decisions about their market orientation and the mode of participation in transactions. This makes it very important to understand the determinants and consequences of these decisions.

Choosing to cease or to maintain a commercial dairy business depends on the endowments of a farm household. This concerns not only human capital or the farm's own financial resources but also access to external funds and information, as well as the contractual relationships in the food system. As a result, given the recent evidence of increased vertical coordination in the agrifood supply chains in transition countries (World Bank, 2005), belonging to the modern marketing channel might have a large impact on farms' relationships with the market.

Deciding whether to sell farm output or not should also be seen in the context of solving an optimisation problem. Not only do ceasing milk sales reduce the number of available income sources, it is also likely to result in a need to find an alternative to top-up the household budget. Again, given the heterogeneous impact of different modes of market participation on farm performance (Milczarek-Andrzejewska et al., 2007), it seems that the restructuring along the supply chain might play an important role in determining the optimal decision.

Although there have been a number of studies on the dairy supply chain in Poland, there has been little analysis of the factors that cause households to choose their relationship with the market. Moreover, there is scarce evidence on the resulting impact of heterogeneous market participation on farm performance. This paper, using the evidence from a unique survey conducted among randomly sampled Polish individuals, aims to fill these gaps.

The first stage of the study uses logistic regressions and investigates milk producers that fall outside the commercial dairy supply chain. It attempts to determine the main factors responsible for the decision to stop milk sales (but not necessarily to stop milk/agricultural production). The focus is on the importance of changes taking place in vertical linkages between farms and dairy processors.

The second stage of the study tries to compare the performance of those who ceased milk sales and those who stayed in the commercial dairy sector. Thus, it attempts to shed light on whether severing relations with the market could be regarded as optimal. Once again, the importance of different characteristics of the dairy supply chain is investigated. We estimate the average effect of the market participation decision on farm revenues by means of semi-parametric propensity score methods.

The paper is organised as follows. Section 2 gives a brief overview on the restructuring of the dairy supply chain in Poland during transition. Section 3 draws on the theoretical discussion of some of the most important factors that motivate farms to remain in or stop commercial production. It also reviews the findings of recent studies examining the relationship between farmers' marketing choices and farm performance. Section 4 presents the econometric approach. Section 5 describes the data and defines the variables. Section 6 reports the results of the econometric analysis and Section 7 concludes.

2. Background information

Poland's dairy supply chain has recently been the subject of several interesting studies documenting in detail the main changes that have occurred in the sector over the last two decades (see, among others, Wilkin et al., 2007; Dries et al., 2011). Therefore, we restrict ourselves to sketching some key aspects of the restructuring process here. To start with, the number of dairy farms declined by more than 1 million, from 1.8 million in 1990, to roughly 0.7 million in 2005 (GUS, var. vol.).¹ Moreover, a further decline in the number of dairy farms is expected (Tonini and Jongeneel, 2009). The share of total milk production delivered to dairy-processing companies has been continuously increasing since 2000 and in 2005 it amounted to 76 per cent (GUS, var. vol.). However, the share of milk producers delivering milk to the processing sector was only 41 per cent in 2005 (48 per cent if we include direct milk sales to the market), meaning that more than half of milk producers in Poland are outside of the commercial dairy supply chain.

The other remarkable change is the emergence of a new marketing channel through which milk is delivered to the processing industry. Traditionally, milk was delivered by farmers to collection stations operated by dairy processors. This was the only way for farmers to sell milk to the processors during the communist regime. During transition, milk was increasingly collected directly at the farm by a dairy truck and it is estimated that this new channel, hereafter referred to as the modern one, now accounts for 60–100 per cent of dairy milk supplies (depending on the dairy company) and it seems to be just a matter of time until it becomes the only way milk is delivered to dairies (Wilkin et al., 2007).

These changes have been driven by both private initiatives and public policies (Dries et al., 2011). The former were particularly prevalent in the second half of the 1990s. Direct contractual relationships between dairies and farms and assistance programmes in return for guaranteed supplies, facilitating adjustments at the farm level, could serve here as examples. The process of policy-driven restructuring was most visible at the beginning of transition as well as in the last decade. First public efforts were mostly directed towards market stabilisation. More recently, milk quality improvement has been given high priority. The most significant policy interventions that affected the restructuring process during our sample period included a system of preferential credits, strict sanitary and veterinary norms for milk production, special payment for ‘extra’-quality milk and, following the accession to the EU, the introduction of a milk quota system and direct payments (Wilkin et al., 2007).

3. Related literature

Given the purpose of the paper, the focus in this section is on two strands of the literature, namely the one on farms' market orientation and the one linking supply chain modernisation and farm performance.

3.1. Market orientation

In recent publications, a number of arguments are provided to explain farms' differentiated relationships with the market. In the transition context, the focus is on three aspects (Kostov and Lingard, 2004). First, decisions about production and market transactions have been linked to macroeconomic prospects and the overall economic situation (Caskie, 2000; White and Gorton, 2006). Second, relationships with the market have been closely tied to the quality of institutions that determine the scope of farmers' choice, namely the quality of output and input markets (Heidhues and Brüntrup, 2003). A number of studies indirectly support this argument, providing evidence that credit market imperfections in particular constitute an important constraint on farms' activities in several transition countries (Swinnen and Gow, 1999; Petrick, 2004). Third, an important segment of the literature focused on the role of transaction costs. They include the costs of accessing markets, associated with transportation and imperfect information, the costs of searching for trading partners, the bargaining process as well as monitoring and enforcing the contract. Transaction costs ‘raise the price effectively paid by the buyer and lower the price effectively received by sellers of a good, creating a “price band” within which some households find it unprofitable to either sell or buy’ (Key et al., 2000: 245). A number of empirical studies found that these costs are likely to affect agents' market participation decisions (e.g. Goetz, 1992; Key et al., 2000; Davidova et al., 2009).

From our perspective, the relationships between supply chain restructuring, market stability, the severity of input markets' imperfections and the level of transaction costs are important. As noted by many authors, the process of supply chain modernisation has been associated with institutional innovation in the form of contracts (e.g. White and Gorton, 2006). Further, the emergence of private capital enforcement mechanisms in vertical relationships between farms and downstream companies should also be recognised (Gow and Swinnen, 2001). It is reasonable to assume that both contracting and ensuring contract enforcement overcome hold-up problems and thus supply chain modernisation should act in favour of reducing transaction costs and price risks. Moreover, provided that the supply chain restructuring is accompanied by assistance programmes for farmers (Swinnen, 2007) it should also mitigate other risks resulting from imperfect information or other markets' imperfections. These considerations find support in the study by Milczarek-Andrzejewska et al. (2007) investigating dairy supply chain modernisation in Poland. The authors report that farmers rate the modern marketing channel more highly than the traditional one in terms of, among other factors, the price paid, security of milk collection, security of timely payments and technical and credit assistance. Similar evidence comes also from other countries from the region (White and Gorton, 2006; Swinnen, 2007). Overall, using the modern marketing channel should lower the probability of a farm to cease milk sales because of its favourable impact on the level of transaction costs and risks faced by the farmer. This is all the more plausible since joining the modern marketing channel very often requires additional investments in the farm and thus the farmer's decision may also be influenced by sunk costs.

3.2. Farm revenues and market choices

As mentioned above, we are not only interested in factors impacting household decisions (not) to cease milk sales, but also in comparing revenues of those who ceased with revenues of farms that maintained their dairy businesses. The following discussion will help to build the context for our analysis. First, several recent studies have attempted to investigate the impacts of farmers' market choices on farm performance (for an overview see World Bank, 2005; Swinnen, 2007). Although the overall picture is mixed, the evidence from transition countries seems to suggest the positive impact of supplying the modern marketing channel. Throughout the region, beneficial effects have been observed on output, productivity and product quality (Gow and Swinnen, 2001; Dries and Swinnen, 2004; White and Gorton, 2006; Dries et al., 2009). The latter, in turn, are likely to result in higher farm revenues. This has been confirmed by the evidence from the dairy sector in Poland (Milczarek-Andrzejewska et al., 2007) where choosing the modern channel had a positive impact on farms' financial situations. Second, no systematic evidence was found in previous studies in favour of the hypothesis that supply chain restructuring led to small farmers' marginalisation.² On the contrary, collected findings suggest that changes in the organisation of the supply chain (e.g. foreign direct investments, vertical coordination) brought opportunities not only to larger producers, but also to smallholders (Dries and Swinnen, 2004; Milczarek-Andrzejewska et al., 2007; Swinnen, 2007).³ This may be an argument for why we should expect additional incentives for farmers to remain in commercial production.

After a short description of the econometric strategy used to address the research questions in the following section, we will attempt to collate the theoretical considerations presented above with the Polish dairy farm data.

4. Econometric strategy

The first research question, i.e. what factors impact the decision to cease milk sales, is addressed by using a logit model estimated by maximum likelihood. We assume that our binary-dependent variable, equal to one for all households ceasing milk sales at some time within the sample and zero for those that do not, depends on an unobserved continuous propensity of a farmer not to participate in a dairy market. We specify this unobserved variable as a linear function of a number of explanatory variables.

It is important to note that our sample contains information on each household at two points in time, namely for the years 2001 and 2006. In 2001, all the surveyed households maintained a commercial dairy business. Those who decided to cease milk sales did it at some point after 2001. All explanatory variables used in the logistic regressions refer to 2001, in order to avoid a potential reverse-causality problem and minimise the endogeneity issue.⁴

As far as the second research question is concerned, i.e. how do the farms that ceased their participation in the dairy market perform compared to those that still sell their milk, we employed propensity score matching methods (Rosenbaum and Rubin, 1983; Heckman et al., 1997, 1998; Smith and Todd, 2005). The main idea behind the matching approach is to mimic a controlled experiment. In our context, the treatment is ceasing milk sales. The outcome of interest is farm revenues.⁵ We use two measures of farm performance, namely annual agricultural revenues and total revenues per capita.

The matching technique reproduces the treatment group among the non-treated by pairing each participant with members of the non-treated group, controlling for observable characteristics. This is done in order to get the missing information on the treated, had they not participated in the experiment. The focus of propensity score matching is to rule out the impact of systematic differences between the treated and non-treated that might affect the outcome of interest and thus to overcome the potential selection bias problem. Note that using a simple dummy indicator for treatment (ceasing milk sales in our case) in a conventional regression approach, would yield the difference between the average revenues of all treated and the average revenues of all controls. The propensity score matching method instead compares each treated farm with a non-treated farm having a set of similar controls. Moreover, matching relaxes the linearity assumption, allowing for any heterogeneity in the effect of ceasing milk sales, as long as it is related to the observable factors (Blundell and Costa Dias, 2008). Applying this method reduces the risk that the effect of ceasing milk sales is confounded with the effect of the factors determining this decision. These advantages are of particular importance, since one may assume that the decision to cease milk sales is not randomly made, but is rather a result of ‘self-selection into treatment’.

To put it more formally, let Q be an indicator of ceasing milk sales (Q = 1) or remaining in commercial milk production (Q = 0). Let Y denote the outcome variable, i.e. farm revenues, such that Y¹ and Y⁰ denote outcomes depending on the relationship to the market Q. Finally, let p(X) be the probability of ceasing milk sales, the propensity score. Our parameter of interest is the average effect of treatment on the treated (ATT), namely

(1)

which measures the effect of ceasing milk sales on farm revenues for the farms that actually withdrew from milk sales, relative to what would have happened had they maintained their relationship with the market. Since one does not observe what would have happened if the farms that ceased had remained in commercial production (or the converse), an estimate of the counterfactual is constructed, i.e. E(Y⁰|p(X), Q = 1). The estimation procedure consists of two steps: first, conditional on the number of observable characteristics, the probability of ceasing milk sales is calculated for each farm i, the propensity score p(X_i). Based on this estimate, the next step involves evaluating the difference in farm revenues between the farms that decided to withdraw from the dairy market (treated) and those that did not (control).⁶ To evaluate the ATT, we used a kernel-matching estimator with calliper (0.005), which performed better than other estimators in terms of assuring that the distribution of observed covariates is balanced between the farms in the treated and control groups.⁷

The matching approach relies on two crucial assumptions. First, the so-called conditional independence assumption (CIA) assumes that observed characteristics contain all the necessary information about the potential outcome in the absence of treatment (Y⁰Q|X). The second matching assumption, the so-called common support assumption, states that the propensity score is bounded away from 0 and 1 (0 < p(Xi) < 1). The plausibility of these assumptions depends on what variables enter the vector X. Any omitted variables uncorrelated with X that affect the decision to cease milk sales and the resultant consequences would violate the CIA and bias the estimation results (Becker and Ichino, 2002). On the other hand, including too many covariates might result in predicting the treatment too well and thus violating the common support assumption. This would result in the treated observations having no counterparts in the control group. Our strategy, therefore, was to select a limited number of covariates that are likely to influence both the decision to cease/maintain the dairy business and farm revenues to ensure appropriate similarity between treated and controls without violating the common support assumption. Using a limited number of covariates is also advisable in our case given our relatively small sample.

5. Data and variables definition

5.1 Data

The analysis presented below uses primary survey data collected in 2007 in two regions (NUTS 2) located in northeast Poland, namely Warmińsko-Mazurskie and Podlaskie. Both of them rank among the best-developed dairy regions after experiencing considerable restructuring. They may serve as textbook cases of the sector's transformation and provide valuable insights for other parts of the country that are behind in the adjustment process. In order to capture the dynamics of the restructuring process, information for 2 years, namely 2006 and 2001, was collected.

The sample was selected in accordance with a stratified random sampling methodology. From each of the two-selected regions, three powiats (subregions, NUTS 4) were randomly selected (from 14 in the Podlaskie region and 19 in the Warmińsko-Mazurskie region). From each of the selected powiats, a random choice was made of three or four gminas (local communities, NUTS 5) (from 35 in the six chosen powiats). Villages were randomly selected from each gmina. Overall, the obtained sample contained observations from 20 gminas and 108 villages.

The questionnaire aimed at capturing the main features of changes in dairy supply chain with the focus on adjustments taking place at the farm level. The sample was composed to have sufficient observations for each identified relationship with the market. Accordingly, it includes information on farms in both marketing channels, modern and traditional, as well as on farms that withdrew from milk sales over the analysed period. The distribution of observations by channel was based on the qualitative study that preceded the survey (see Milczarek-Andrzejewska et al., 2007; Wilkin et al., 2007 for details).

In total, 397 dairy farm households were surveyed representing 0.5 (0.93) per cent of the total number of (commercial) dairy producers in both surveyed regions. After cleaning, 389 observations remained for analysis. Of those, 323 farms were still delivering milk to the market in 2006 and 66 had stopped at some point after 2001.⁸ Table 1 presents the distribution of observations in the sample by farm orientation and marketing channel.

In 2001, 146 farms were in the modern marketing channel (milk collected at the farm) and 243 were in the traditional one (milk delivered to collection points). Taking into account the 66 farms that ceased milk sales, in 2006 these numbers were 218 and 105, respectively. As such, we classify farmers into two groups: modern channel farmers and traditional channel ones. Within those two categories, farmers that ceased milk production, farmers that switched between marketing channels and farmers that did neither can be distinguished.

5.2. Variables definition – logit model

Since we are interested in identifying the causal effect of supply chain modernisation on market orientation, the key variable of interest in the first stage of the analysis is a dummy variable modern_channel. It takes on a value of 1 for farms belonging to the modern marketing channel and 0 otherwise. As mentioned in the literature review, it is reasonable to assume that belonging to the modern marketing channel should encourage farmers to maintain dairy businesses. Therefore, given that the dependent variable is a dummy variable distinguishing farms that stopped milk sales, we expect the modern_channel to have a negative sign.

We also include some other covariates to control for household, farm and neighbourhood characteristics. The variables age, education and agric_education refer to farm operators and represent age in years, a dummy variable for operators with only elementary education,⁹ and a dummy variable for operators with agricultural education, respectively. The variables herd-size and yields represent number of cows and log of annual milk yields (in litres) per cow, respectively. The variables dairy-assets and two dummy variables, milking parlour and manure storage, aim at capturing farm equipment used specifically for milk production. The former is a general index of dairy-specific assets based on the Mokken scale procedure (Sijtsma and Molenaar, 2002) and factor analysis (Bartholomew and Knott, 1999).¹⁰ To have a milking parlour and manure storage area, separate scales are needed. Therefore, these two items are not captured in the index but are represented with appropriate dummies. Further, to control for milk quality, milk-price, which represents the milk price (in PLN¹¹/l) paid to farmers is also included. In addition, the variables coop-member, share-milk, labour-endowments, revenues and off-farm refer to household characteristics. The first is a dummy variable and informs whether a farmer belonged to a dairy cooperative. Share-milk represents the share of milk sales (per cent) out of total agricultural sales, thus capturing the importance of the dairy business in terms of revenues when compared to other agricultural activities. Labour-endowment is defined as a weighted sum of household members over 15 years old.¹²Revenues represent the log of annual farm revenues per capita (in PLN) and off-farm is a dummy variable, which distinguishes households having access to off-farm wages. Both of these variables aim at capturing the potential effect of overall economic prosperity. Finally, to control for the regional differences, the estimated logistic regressions include regional dummies. In order to control for the fact that farmers' decisions to cease/maintain milk sales could be affected by neighbourhood influences, we use data for the most disaggregated NUTS five regions (gmina).

5.3 Variables definition – propensity score matching

In accordance with the earlier discussion, we decided to base the estimation of propensity scores on a limited number of covariates: modern_channel, age, herd-size, milk-price, revenues, share-milk, labour-endowments and off-farm. As before, all these covariates refer to the year 2001. The two outcome variables that we use are annual agricultural revenues per capita and total farm revenues per capita (both in PLN). The data on outcome variables refer to 2006.

Table 2 displays means and standard deviations for the main variables of interest in the full sample as well as in the two sub-samples: farms that ceased milk sales and farms that maintained a commercial dairy business.

6. Empirical analysis

We discuss the empirical results in two parts. First, we investigate the determinants of ceasing a commercial dairy business. Second, we estimate the effect of ceasing milk sales on the treated farms.

6.1 Logit model

The results of estimation analysing the binary decision to cease milk sales/maintain commercial production are presented in Table 3. Our main results are presented in column 1.

Five key points should be noted: first and most importantly, a negative and significant effect of the market choice is found. This in turn indicates that development opportunities in the dairy sector might have been restricted to farmers already included in the modern marketing channel. This result corresponds well to the fact that ceasing milk sales is less likely among larger producers in terms of herd size, which we consider a second important finding. It should be noted, however, that this result might simply reflect the fact that larger farms are more profitable. Third, the variables describing farm productivity provide evidence of an important dichotomy: while higher milk yields per cow appear to be insignificant, the variable dairy-assets is negatively associated with ceasing milk sales. The latter could be a reflection of sunk investments likely to bind dairy producers to the market. The same can be said about the negative impact of manure storage and milking parlour. Fourth, the decision to cease milk sales is positively affected by the farmer's age. Provided that older age is closely related to retirement considerations and poor health, this result is in accordance with expectations. It is also corroborated by the literature investigating the dairy farm development/exit (Kimhi and Rubin, 2007; Peerlings and Ooms, 2008). Fifth, farms with access to off-farm jobs are more likely to cease milk sales. This would indicate that off-farm jobs could provide a substitute for earnings in the dairy sector. Furthermore, it might be worth noting that other variables, although being statistically insignificant, have the expected sign. Importantly, our results are consistent with those obtained by Dries and Swinnen (2004) who analysed Polish dairy farms' survival based on data from 1995 to 2000, i.e. the 5-year period preceding our study.

To assess the robustness of our results, we estimate additional specification without variables representing farms' dairy equipment (column 2). This is done to control for potential multicollinearity, since having dairy-specific assets is highly correlated with variables modern_channel and herd-size.¹³ Our findings are robust to this exclusion.

6.2. Propensity score matching

Having established the main determinants of a farm's decision to cease milk sales, we move to addressing the second research question: Was that decision optimal from a farm revenues perspective?¹⁴ In order to answer this question, we compare revenues generated by farms that ceased milk sales, but remained in agricultural production, with revenues of farms that decided to maintain commercial milk production. Accordingly, we exclude farms that ceased all agricultural enterprises. In effect, we are working with 377 observations, 54 observations for farmers who ceased milk sales and 323 observations for farmers who remained in the commercial dairy business.

To gain more insight into the heterogeneity of farms' development and to take into account that the majority of farmers that ceased milk sales were in the traditional channel, we consider here three cases that differ from each other in terms of the control group. The full sample specification compares farms that ceased milk sales with all farms that kept commercial dairy businesses. The limited sample A specification, on the other hand, compares farms that ceased milk sales with farms that maintained commercial production but remained in the traditional marketing channel. Finally, the limited sample B specification compares farms that ceased milk sales with farms that remained in the dairy market and made a shift from the traditional to the modern marketing channel.¹⁵ This strategy allows for a more detailed investigation of the supply chain's modernisation effect on farm revenues and helps to control not only for selection into ceased/remained in milk sales samples, but also for selection into different modes of participation in transactions. In order to avoid strong assumptions of multinomial logit and computational difficulties with multinomial probit, we run a series of binomial probit models to calculate the propensity scores (Lechner, 2001; Greene, 2003; Caliendo and Kopeinig, 2005).

Table 4 provides information on how well the matching procedure performs in our case.

The presented results refer to the full sample specification. As shown there, farms that ceased milk sales and farms that maintained their dairy business differ systematically in terms of a number of characteristics. This is illustrated by the fact that the null hypothesis of equal means of unmatched treated and control is rejected for the majority of variables. What is evident, though, is that matching performs well and removes all significant differences.¹⁶ Overall we conclude that matching is needed to achieve a reliable comparisons between the groups of farms we are interested in.

Matching estimates are presented in Table 5. In case of the full sample specification, we find that farms that ceased milk sales generated lower agricultural revenues per capita than farms that maintained commercial milk production. This result is also found for the limited sample B specification, where farms shifting from the traditional to the modern marketing channel serve as a reference point. A systematic difference is also observed between agricultural revenues per capita for farms that decided to cease and those supplying the traditional marketing channel, although here the finding is only significant at the 10 per cent level. These results robustly suggest that farms no longer participating in dairy markets did not manage to substitute potential milk revenues from the traditional channel with other agricultural enterprises.

A different picture emerges from comparisons based on total farm revenues per capita. Again, when comparing farms that ceased milk sales with all farms that remained in the dairy business (full sample), the former group comes out poorly. However, when compared only to the group of farms supplying the traditional marketing channel, total revenues of farms that ceased milk sales do not lag behind (limited sample A). This seems to indicate that farms that ceased milk sales succeeded in finding non-agricultural income sources so that their total revenues per capita are comparable with farms delivering milk to collection points. These two findings are complemented by the result obtained for limited sample B, namely that total revenues of farms that ceased milk sales do systematically differ from total revenues of farms that joined the modern channel at some point after 2001.

Overall we conclude that the decision to cease milk sales could be regarded as optimal, but only for farms not willing to modernise in the future. Whether this decision is a voluntary one remains an open question. While interpreting these results, two things should be kept in mind: first, our outcome variables were based on farm revenues, which are at best imperfect proxies for farm performance. Second, due to data limitations, the ATT in our matching procedure was estimated as a difference between farm revenues of the treated and untreated farms observed in 2006. As noted by Heckman et al. (1997), however, treated and controls may still differ even after the conditioning of observables. This may be due to some unobservable characteristics. A potential solution to this problem is to combine matching with difference-in-difference estimates applied to a new data set.¹⁷

7. Conclusions

In response to rapid and profound changes taking place in the Polish dairy sector, this paper aimed at investigating factors impacting on a farmer's decision to cease milk sales, taking into account the fact that almost half of milk producers in Poland are outside of the commercial dairy supply chain. In order to understand the consequences of this decision, the paper further attempted to check whether terminating the relationship with the market could be regarded as optimal. In order to do so, semi-parametric methods were employed to compare revenues of farms that ceased milk sales with revenues of farms that maintained commercial milk production. The analysis was based on a unique data set on individuals with different relationships to the market and belonging to different dairy supply chains.

The key finding is that farms that ceased milk sales compare unfavourably with farms that participated in the dairy market in terms of total (agricultural) revenue per capita. However, important differences depending on dairy supply chain characteristics were revealed among farms that maintained dairy businesses. Compared to farms that ceased milk sales, farms from the modern marketing channel have significantly higher revenues, whereas farms from the traditional marketing channel have similar revenues. Consequently, ceasing milk sales could be regarded as optimal for farms not willing to modernise. For those who wanted to modernise but failed to do so, irrespective of the reasons, this decision resulted in a worse financial situation.

The decision to cease milk sales negatively depends on belonging to the modern marketing channel, herd size, access to production-specific assets and age and positively depends on having off-farm job opportunities. The first three results could be of importance for the discussion about the relationship between supply chain modernisation and smallholders' exclusion. It should be stressed, though, that they do not allow for direct verification of the exclusion/inclusion hypothesis and further research is needed to better understand this phenomenon.

Acknowledgements

This research was conducted within the Regoverning Markets project. For further information, see the website www.regoverningmarkets.org. The author would like to thank Thomas Heckelei, Gábor Körösi, Beata Łopaciuk-Gonczaryk, Agata Malak-Rawlikowska, Dominika Milczarek-Andrzejewska, three anonymous referees and participants from seminars in Kent and Budapest for their helpful comments. The usual disclaimer applies.

References