II. Literature review

To the best of our knowledge, the only paper on the determinants of grape-growers’ relevant choices is Corsi (Reference Corsi and Giau2003), who only considers the choice of making or not making wine on the farm by grape growers who have a wine-making plant available. Nevertheless, some literature is relevant to this issue, pertaining to different strands. One strand concerns vertical integration and, more generally, the rationale for different organizations of production chains. It mainly refers to the issue of transactional costs (Coase, Reference Coase1937; Williamson, Reference Williamson1985, Reference Williamson1991) and posits that the organizational form (contractual vs. hierarchical) is determined by the possibility or impossibility of complete contracts envisaging all possible contingencies and the related transactional costs. The second strand refers to property rights theory (Grossman and Hart, Reference Grossman and Hart1986) and considers that the costs to the firm of governing internal exchanges reduce the advantages of hierarchies. Nevertheless, the possibility of integrating the chain by acquiring ownership of downstream industries is rather limited for most grape growers, due to their small economic size. Their only possibilities of vertical integration are the production of wine from their grapes either directly, or indirectly through cooperatives.

There is a large body of literature on cooperatives, but, apart from descriptive works, the theoretical discussion focuses on different issues (Candemir, Duvaleix, and Latruffe, Reference Candemir, Duvaleix and Latruffe2021). One of them is how the different governance structures between cooperatives and investor-owned firms affect their economic objectives (Hendrikse and Bijman, Reference Hendrikse and Bijman2002; Bontems and Fulton, Reference Bontems and Fulton2009; Hueth and Marcoul, Reference Hueth and Marcoul2015; Peng, Hendrikse, and Deng, Reference Peng, Hendrikse and Deng2018), an issue further complicated when cooperative managers are considered (Fulton and Pohler, Reference Fulton and Pohler2015; Hueth and Marcoul, Reference Hueth and Marcoul2015). This literature is relevant for understanding the functioning of cooperatives but less so for analyzing grape-growers’ choices regarding the destination of their grapes. Another issue considered in the literature is the setting of markets where cooperatives and private investors coexist, and the resulting supply and price levels (Helmberger and Hoos, Reference Helmberger and Hoos1962; Albæk and Schultz, Reference Albæk and Schultz1998; Pennerstorfer and Weiss, Reference Pennerstorfer and Weiss2013; Liang and Hendrikse, Reference Liang and Hendrikse2016; Carletti et al., Reference Carletti, Hanisch, Rommel and Fulton2018). Again, this stream of literature does not analyze grape-growers’ decision-making determinants. A more relevant strand of literature for our purpose, mainly concerning developing countries, analyzes the determinants of participation in cooperatives using random utility models, according to which farmers choose the alternative providing the highest utility (Fischer and Qaim, Reference Fischer and Qaim2012, Reference Fischer and Qaim2014; Wollni and Zeller, Reference Wollni and Zeller2007; Bernard and Spielman, Reference Bernard and Spielman2009; Abebaw and Haile, Reference Abebaw and Haile2013; Mojo, Fischer, and Degefa, Reference Mojo, Fischer and Degefa2016, Reference Mojo, Fischer and Degefa2017; Shumeta and D'Haese, Reference Shumeta and D'Haese2016; Gyau, Mbugua, and Oduol, Reference Gyau, Mbugua and Oduol2016; Hao et al., Reference Hao, Bijman, Gardebroek, Heerink, Heijman and Huo2018; Wang et al., Reference Wang, Cheng, Lee, Sun and Chang2019; Lin et al., Reference Lin, Wang, Jin, Yang and Li2022). Nevertheless, this literature only considers the dichotomous choice of becoming a cooperative member or not per se. In the analyzed cases, this is not typically equivalent to deciding on a marketing outlet for farmers’ products, and cooperative membership is not an alternative to other commercial outlets (or other sources of inputs).

A further stream of literature, which is close to our methodological approach, deals with the determinants of farmers’ choices of market outlets or processing of their agricultural products, both in developing and developed countries (e.g., Fafchamps and Hill, Reference Fafchamps and Hill2005; Shilpi and Umali-Deininger, Reference Shilpi and Umali-Deininger2008; Takeshima and Winter-Nelson, Reference Takeshima and Winter-Nelson2012; Abdulai and Birachi, Reference Abdulai and Birachi2009; Arinloye et al., Reference Arinloye, Pascucci, Linnemann, Coulibaly, Hagelaar and Omta2015; Corsi, Novelli, and Pettenati, Reference Corsi, Novelli and Pettenati2018; Negi et al., Reference Negi, Birthal, Roy and Khan2018; Pham, Theuvsen, and Otter, Reference Pham, Theuvsen and Otter2019; among others). This literature assumes that the choice is determined by the comparison of the benefits (in terms of utility or income) stemming from the different alternatives, and assesses the determinants using various econometric techniques, generally logit or probit models. Most of these papers deal with dichotomous choices (adopting or not a marketing channel, processing or not the agricultural product). In the few cases when choices are multiple, no hierarchy of the choices is envisaged. One exception is Hao et al. (Reference Hao, Bijman, Gardebroek, Heerink, Heijman and Huo2018), who condition the choice of the marketing channel on cooperative membership.

III. Theoretical approach

To illustrate the theoretical framework, it is useful to present the context of our empirical exercise. The three choices grape growers have concerning their product, namely making wine on the farm, selling the grapes to an outside winery, or delivering them to a cooperative winery of which they are members, are mutually exclusive. In the Italian context, it is generally mandatory for members of cooperative wineries to deliver their whole production to the cooperative. Normally, on-farm wine-makers crash their whole production, so this choice is again an alternative to the other two.

For farmers selling their grapes to an outside winery, no further cost is incurred except for transportation. The price depends on the grape market. In the regional situation we analyzed, in some cases, grapes are sold based on long-term, sometimes informal, contracts to wine-makers who are particularly interested in the quality of the grapes and trying to ensure a secure and high-quality raw product. Another possibility is selling to industrial wine-makers who collect grapes mainly based on their price with little or no fidelity to particular grape growers.Footnote ¹ Of course, prices vary at the farm level due to the specific variety and quality.

For grape growers making their own wine on the farm, this choice implies an investment in wine-making facilities—a largely sunk cost—plus operating costs and often wine marketing costs. Operating costs comprise costs for inputs other than grapes and labor costs. Normally, in wine technology, both input and labor costs are a linear function of production. Though, since most farms are family farms, the labor costs of household members are opportunity costs if they have employment opportunities off the farm, or subjective costs if such opportunities do not exist or if on-farm and off-farm labor are not perfect substitutes in utility. The wine is either bottled and sold under the farm's label or, more rarely, sold in bulk to bottlers. The revenue depends on yields and on wine price, which is subject to a large variation based on varieties and specific quality. Nevertheless, in the area of investigation, no icon wines are produced, so the price variation is smaller than in the more prestigious Langhe area.

A third group is represented by cooperative members. In the local situation, this is a long-term choice because, generally, it is not possible to enter and exit a cooperative yearly since internal rules typically discourage exits through some constraint or penalty. The obligation to deliver the whole production to the cooperative is easily controlled by knowing the members’ vine areas and, hence, production. Cooperative membership is open to anybody who wishes to join, and the cooperatives accept the delivery of any quantity of grapes from their members. The cooperatives pay their members the price resulting from the sales minus the processing and marketing costs. Generally, this is done by making a down payment at the moment of delivery; the balance is paid when the cooperative accounts are completed. Prices paid to members are typically differentiated by the quality of the grapes. No costs are implied for the members except for a fee when joining the cooperative and transportation costs. Members’ revenues, therefore, depend on the technical and marketing efficiency of the cooperative.Footnote ²

To analyze grape-growers’ choices concerning the destination of their grapes theoretically, assume no choice has yet been made. Then grape growers are assumed to choose the grape destination that provides the highest expected utility, which includes both utility stemming from income and utility stemming from the psychic benefit of the particular choice. We assume that grape-growing costs and grape yields are identical for all three choices and that the transaction costs of each choice are reflected in the relevant prices. Differences in delivery costs between the three choices are considered negligible. Hence, income comparisons can be based on revenues, net of processing and marketing costs in the case of wine.

The expected overall utility of choice “sell the grapes” (choice G) is:

(1)$${\rm U}( {\rm G} ) = {\rm E}[ {\rm U}( {\,p_{gm}y \,- \,c_p\,- \,c_t} ) + {\rm U}( {{\rm s}G( H ) } ) ] , \;$$

where y is grape production; p_gm denotes the market price of grapes; c_p and c_t are production and transportation costs. U(sG) is the subjective utility from a particular choice (e.g., users’ appreciation of the quality of the grapes; lack of confidence in their personal winemaking skills; risk aversion toward the investments in wine-making), a function of personal and household characteristics H.

The expected overall utility of choice “making wine” (choice W) is:

(2)$${\rm U}( {\rm W} ) = {\rm E}[ {\rm U}( {\,p_way \,- \,c_p\,- \,c_w( {y, \;p_i, \;l} ) \,- \,c_t} ) + {\rm U}( {{\rm s}W( H ) } ) ] , \;$$

where y is grape production; a denotes the transformation coefficient of grapes into wine (usually 0.7 in weight); p_w is the wine price; c_w are wine-making and marketing costs, function of the quantity, input prices p_{i ,} and labor unitary cost l (either market wage or subjective labor cost); U(sW) is the expected subjective utility from the choice of wine-making (e.g., pride of being a wine-maker; wine quality appreciation by consumers; risk-loving attitudes), again a function of personal and household characteristics H.

The expected overall utility of choice “delivering to a cooperative” (choice C) is:

(3)$${\rm U}( {\rm C} ) = {\rm E}[ {\rm U}( {\,p_{gc}y \,- \,c_p\,- \,c_t} ) + {\rm U}( {{\rm s}C( H ) } ) ], $$

where y is the grape production; p_gc is the grape price paid by the cooperative to its members; c_p and c_t are production and transportation costs; and U(sC) is the subjective utility from the specific choice (e.g., psychic benefits from the participation in a collective endeavor, risk attitudes toward engaging in a long-term commitment), again a function of personal and household characteristics H.

Since the quantity of grapes is the same and the quantity of wine is a fixed ratio of the quantity of grapes, the most relevant variables for the economic benefit are the prices fetched in the three choices and wine-making and marketing costs. Variable wine-making input costs are assumed to linearly depend on observable characteristics of wine-making farms, with an unobservable random component. Family labor costs can depend on the opportunity cost of labor and, hence, on farmers’ observable personal characteristics determining the potential wage and on an unobservable random component. In addition to these, if labor cost is subjective, either because of the different utility of on-farm and off-farm labor or because no off-farm employment opportunity is available, it also depends on personal and household characteristics. The subjective components of utility for the different choices are also assumed to depend linearly on personal and household characteristics, plus a random component.

In summary, choice j is made (j = wine, grapes, co-op) if:

(4)$${\rm U}_j[ {{\rm I}( {{\rm p}_j, \;{\rm F}, \;{\rm H}} ) , \;{\rm \!H }} ] + \varepsilon _j > {\rm U}_k[ {{\rm I}( {{\rm p}_k, \;{\rm F}, \;{\rm \!H}} ) , \;{\rm \!H}} ] + \varepsilon _k\forall \;k\ne j, \;$$

where income I is a function of p_j (p_k), the relevant price from choice j(k); of F, a vector of farm characteristics; of H, a vector of farmer's and household's characteristics; personal and household characteristics H also determines the psychic benefits of the different choices; ɛ is the random component.

Ideally, to analyze these choices, one would like to have panel data allowing one to identify the choices as they are made. Unfortunately, such data are not available, in part because some decisions were made a long time ago, which would require data for a very long time span. Hence, our analysis will be based on cross-sectional data under the implicit assumption that the choices are affected by the explanatory variables in their values at the time of observation (except for prices, which were included with their lagged values).

IV. Econometric approaches

For the empirical analysis of the determinants of the three choices, a first possibility is adopting a MNL model:

(5)$${\rm Prob}( {{\rm choice\;is\;j}} ) = \displaystyle{{e^{\beta _jX_i + \gamma _jZ_{ij}}} \over {\mathop \sum \nolimits_j^J e^{\beta _jX_i + \gamma _jZ_{ij}}}}{\rm \;}, \;$$

where X_i are individual-specific explanatory variables; Z_ij are choice-specific explanatory variables; and β and γ are parameters to be estimated. Due to identification restrictions, only the relative probabilities across choices can be estimated. The MNL model assumes identical and independent extreme value distributions of the error terms in the utility functions of the different choices and implies the independence of the irrelevant alternatives (IIA) property, whereby the ratio between the probability of two choices is independent of the remaining probabilities. A Hausmann-McFadden test can be used to verify the appropriateness of IIA in the specific case.

The MNL model places the different choices “on the same ground.” An alternative model is Nested Logit (NL), whereby the choices are grouped into so-called nests so that the IIA property holds within nests but the variance is allowed to differ across nests. In practical terms, this allows considering that there is some conceptual ordering of the choices, so that in our case, for example, a first choice is whether to sell the grapes to an outside winery or not and, if not, whether to become a cooperative member and deliver them to the cooperative or make wine from them. In statistical terms, this approach would imply that some choices have some unobservable determinants in common, so that a tree can be envisaged, with “branches” and “twigs” (Greene, Reference Greene2012). In our case of three choices, a nest can comprise two choices belonging to the same branch, while the third branch provides just one choice. Namely, there are three possibilities that will be tested econometrically. The first choice is either: (1) selling the grapes to an outside winery or not; (2) making wine from one's grapes or not; or (3) becoming a member of a cooperative and delivering one's grapes to it or not. If the first choice is not adopted, the next choice is between the remaining alternatives. A priori, there can be considerations favoring each alternative choice structure. The NL model allows for testing this empirically.

If the J choices are divided into L nests, and the (individual and choice-specific) attributes of choice j in nest l are C_j|l and those of nest l are N_l, the probability of choosing a specific twig j in branch l is (Greene, Reference Greene2012):

(6)$$Prob( {twig{\rm \;}j, \;{\rm \;}branch{\rm \;}l} ) = \displaystyle{{e^{\beta C_{\,j\vert l} + \gamma N_l}} \over {\mathop \sum \nolimits_{l = 1}^L \mathop \sum \nolimits_{\,j = 1}^{J_l} e^{\beta C_{\,j\vert l} + \gamma N_l}}}$$

from which it can be shown that

(7)$$Prob( {\,j{\rm \vert }l} ) = \displaystyle{{e^{\beta C_{\,j\vert l}}} \over {\mathop \sum \nolimits_{\,j = 1}^{Jl} e^{\beta C_{\,j\vert l}}}}$$

and

(8)$$Prob( l ) = \displaystyle{{e^{\gamma _jN_l + \tau _iI_l}} \over {\mathop \sum \nolimits_{l = 1}^L e^{\gamma _jN_l + \tau _iI_l}}}, \;$$

where $I_l = ln\left({\mathop \sum_{j = 1}^{J_l} e^{\beta C_{j\vert l}}} \right)$, the Inclusive Value for the lth branch, is a function of the underlying correlation between the unobserved components for pairs of alternatives in the nest (the lower the Inclusive Value, the higher the correlation). For the model to be consistent with utility maximization, the Inclusive Value must be bound between 0 and 1.

V. Data

We chose to run our analysis in the Province of Asti in the Piedmont Region, northwest of Italy, where grapes are grown almost entirely on hilly terrain. This province has a good number of cooperative and industrial wineries, together with numerous farms making their own wines. Unlike other wine areas in Piedmont like Langhe, where icon wines are produced, the Asti province presents less heterogeneity in grape quality, wine refining, and aging techniques. Wine or grape prices are less variable across areas and over time, making them a good example for analyzing grape-growers’ choices.

The data for farm and farmer characteristics were drawn from the last available official Agricultural Census, in 2010. Of the original individual records in the Province of Asti of farms with vines of any type (5,339 farms), only family farms with at least 60% of the Utilized Agricultural Area (UAA) devoted to vinesFootnote ³ were retained (3,606 farms). Furthermore, we excluded farms where more than 40% of the vine area was covered by varieties for which prices were not available or by Moscato vines. Moscato grapes need some industrial processing for winemaking that is not usually performed on the farms; hence, their destination is almost predetermined. After finally dropping farms with incomplete records, a total of 2,927 farms were retained.

The classification of farms into the three outlets was based on the responses to Census questionnaire questions, which revealed the destination of grapes to cooperatives or others, as well as the production of wine.Footnote ⁴

Among farm characteristics, we included: the total vine area, which can affect economies of scale and hence average costs for wine-making; the share of the UAA devoted to grapes, as an index of specialization; and the share of vine area fit for appellation wines over the total vine area, as an index of quality that might affect the wine-making choice.Footnote ⁵ Information on operators’ characteristics provided by the Agricultural Census includes age, gender, and education levels. Education levels were transformed into education years, so as to have a discrete variable. A dichotomous variable further indicates if the respective high school or university specialized in agricultural studies. These variables are proxies for the technical and marketing skills of the operators assumed to affect wine and grape quality and/or prices at the farm level, relative to average prices. Nevertheless, they also measure the level of human capital, affecting the potential wage that might be earned on the labor market, which raises the opportunity cost of a farmer's labor. In addition, they may affect individual preferences toward a particular choice. Hence, we do not have precise expectations of their signs. We also included dichotomous variables indicating whether the operator had an off-farm job but mainly worked on a farm (“Secondary part-time”) or if he/she had an off-farm job as his/her main occupation (“Principal part-time”). Farm characteristics, together with household characteristics, might influence subjective labor costs. Hence, choosing variables relating to farm and household characteristics seems appropriate. The labor burden of household members (the ratio of adult household members to vine area) is expected to increase the negative utility of work and discourage further use of family labor in wine-making. This variable and the number of children of different ages (under 6 and 6–13 years old) were also included to check if household characteristics affect production choices, as indicators of non-separability between production and individual utility maximization choices (Singh, Squire, and Strauss, Reference Singh, Squire and Strauss1986).Footnote ⁶

The other relevant explanatory variables concerned prices for wine and grapes and prices paid by cooperatives to their members. Prices actually received by each farm for making a particular choice were not available, and in any case, the potential price for an alternative choice to the one actually made would not be observable. Our strategy was, therefore, to estimate the potential price each farm could fetch when making the different choices. Of course, actual prices also differ by grape quality. Unfortunately, there is no information on the quality of grapes from individual farms except for the grape variety. We then wanted to obtain the potential price that a specific farm could obtain when selling the grapes, making wine, or delivering the grapes to a cooperative, conditional on its mix of varieties. To reduce the variability due to harvest effects, all prices were taken as averages of the 2008 and 2009 harvests, that is, the two years before the Agricultural Census. Taking backward-lagged prices also reduces the risk of endogeneity, whereby the destination choices affect the relevant prices.Footnote ⁷ As all Piedmont wines are basically made from single grape varieties, grape and wine prices can be compared. With regard to the sale of the grapes on the market, the provincial prices from the weekly or fortnightly price lists of the Asti Chamber of Commerce were used. Listed grape prices only included the main varieties cultivated in the Province, namely Barbera, Freisa, Grignolino, Brachetto, Dolcetto, Cortese, and Chardonnay. We took the average of the prices between the beginning and the end of the harvestFootnote ⁸ and averaged it between 2008 and 2009. We obtained the potential price of each farm for wine grapes sold on the market as the average of the provincial prices of the different varieties obtained as noted previously, weighted by the relevant vine areas of the different varieties in that farm.

In the case of the cooperatives’ provisions, the average prices paid by the local cooperatives for each variety were directly asked by cooperative managers.Footnote ⁹ In this case, when average prices paid differed according to the different qualities of the grapes, prices were also calculated as a weighted average of the different qualities in each cooperative. In 2008–2009, in the Asti province, 13 cooperative wineries were active, but only from 9 of them could we obtain data on the prices paid to their members. The 9 cooperatives overall stated they harvested grapes from 54 of the 118 municipalities in the province, representing 79.4% of the total vineyard area in the province. The potential cooperative price for wine grapes on each individual farm was calculated as the average of the prices of the different varieties paid by the closest cooperative,Footnote ¹⁰ weighted by the relevant vine areas of the different varieties on that farm.

Wine prices were also drawn from the Chamber of Commerce price lists in 2008 and 2009. Since grape varieties can potentially produce different wines, the calculation of wine prices was more complex than the one for grape prices. Selling prices may vary by quality, location (depending on the boundaries of the appellations), level of refinement, and the expected aging potential (e.g., the “Riserva” quality has a different compulsory level of aging). Therefore, we calculated an average provincial price weighted by the quantities produced in those years for each type of wine and as the mean of the different monthly prices of 2008 and 2009. The potential wine price for each farm was then calculated as the weighted mean of the provincial wine prices, the weight given by the shares of vine areas dedicated to the different varieties originating the specific wines.

The descriptive statistics of the variables, for the total and the three choices, are presented in Table 1. The table also reports the test results for the differences in group averages. Of the total number of farms in the sample, 41% took their grapes to a cooperative (Cooperative group, C), 39% sold their grapes on the market (Grape group, G), and 20% made wine (Wine group, W). The average vine area was 2.77 hectares and did not differ much in absolute terms across groups, though in statistical terms the average was significantly different between C and G. The average share of vines over the UAA, overall equal to 77%, was significantly different across choices, with the lowest in the W group and the largest in the C group. The shares of vine area for appellation winesFootnote ¹¹ over total vine area (57% of the total) were also significantly different across groups, with W at the lowest level and C at the largest. There are only small differences in the operator mean age, which is about 58 years for the full sample. The share of male operators (73%) is not significantly different across groups. The education level is slightly but significantly lower for C relative to W and G, and the same applies to agricultural education. Part-time off-farm jobs as the principal occupation concerned a little less than 10 percent of operators, more for C than for W. Secondary part-time employment concerned 6.3 percent of the farms, with no significant difference across groups. Only 0.9 percent of the operators were foreigners, a share higher for W than for G, probably reflecting investments in wine farms by foreign personalities. The labor intensity (overall 1.95 household members per hectare) was significantly higher for W (2.50) than for G (1.87) and C (1.76). There were no significant differences in the average number of children, which was very low anyway (0.05 and 0.10 for the two age groups).

Table 1.

Descriptive statistics of the variables by choice

Note: Same superscript letters a, b, and c indicate not significant mean differences at the 5% level between the relevant means.

The most interesting data concern prices. The average price received by cooperative members, as calculated earlier, is significantly higher than what would be received by farmers who sell their grapes or make wine. On the contrary, the average price received by wine-makers is not significantly different from the one potentially received by the farmers that sell grapes to other market outlets.