Received 14 December 2015; accepted 23 February 2016; published 26 February 2016
An extensive literature on the political economy of agricultural policies has developed over the past 20 years. Papers in this literature have attempted to provide an explanation for the stylized facts on agricultural protection, such as the widely observed increase in agricultural protection when an economy grows―see de Gorter and Swinnen (2002)  , Swinnen (2010)  and Anderson et al. (2013)  for reviews. Studies have attempted to provide an explanation for the stylized facts on agricultural protection, such as the widely observed increase in agricultural protection when an economy grows (Anderson and Hayami, 1986)  . Theoretical studies attempting to explain these and other facts have stressed the implications of organization costs on the political decision- making process (Gardner, 1987; Olson, 1990)   , structural factors affecting the distributional effects of agricultural protection (Anderson and Tyers, 1988; Honma and Hayami, 1986; Swinnen, 1994)  - , the relative income position of agriculture (Bullock, 1992; de Gorter and Tsur, 1991; Swinnen and de Gorter, 1993)  - and political institutions like differences in electoral rules and the degree of democracy (Beghin and Kerallah, 1994; Swinnen et al., 2001; Olper, 2001; 2007; Olper and Raimondi, 2010)  - .
These political economy studies of agricultural policy have focused primarily on explaining the level of policy intervention and less attention is paid to the explanation of the instruments used for intervention (de Gorter and Swinnen, 2002)  . This bias in focus is an important shortcoming of the literature. From a welfare perspective the key question should be why governments have introduced so many market distortions through agricultural policies. The distortionary effects of government interventions are equally dependent on the choice of the instrument as on the level of the intervention. Therefore the choice of instrument should be at least of equal concern as the intervention level. In fact, both the trade and agricultural policy literature and the policy debates have reflected this importance (e.g. Gardner 1983)  . In the policy world, the debate on the choice of instruments has been a very important element of policy discussions. The differences in distortionary effects is recognized by the WTO in the classification of agricultural policy instruments in green, blue and amber boxes―with the green box for non-trade distorting policies instruments. This distinction between the level of support and the extent of market and trade distortions is at the heart of some important policy reforms, such as those of the EU’s Common Agricultural Policy (CAP) over the past two decades. In fact, one could argue that the issue of instrument choice was the key element of the CAP reform, more so than the level of support (Swinnen, 2008)  .
Surprisingly, this attention to instrument choice in the literature and the policy debate has not translated in similar attention in studies on the political economy of agricultural and food policies where most of the focus has been on explaining the level of intervention rather than its form. A possible explanation for this bias in focus is differences in the availability of good empirical data, resulting in some well-known and puzzling stylized facts on policy level but not on instruments.1
This paper contributed to filling this gap, first of all presenting such stylized facts, drawing on OECD data. The paper then proceeds to provide an explanation of these stylized facts.
The paper is organized as follows. In the next section we present stylized facts on instrument choice in agricultural and food policies. Then we develop a theoretical model and derive some key hypotheses. Afterwards we empirically test these hypotheses using an econometric study. The final section concludes.
2. Instrument Choice in Agricultural Policy
We first present some stylized facts on agricultural and food policy instrument choice in OECD countries over the past 25 years.2
Since 1986 the OECD calculates policy support given to agriculture. The total amount of support to agriculture is referred to as Producer Support Estimate (PSE).3 The PSE data cover 28 countries, 12 of which are not OECD members, over the period 1986-2009 (Table A1). The OECD’s calculation of policy support distinguishes between several instruments. For the purpose of our analysis it is convenient to combine the instruments into “market price support” (mps), “input subsidies” (is) and “direct payments” (dp). Their share in total support (PSE) is represented by mpsh, ish, and dpsh, respectively.
The first instrument, mps, includes all transfers through tariffs, price support and subsidies directly linked to agricultural production. These instruments are typically considered as being the most distortive. The second instrument, is, are input subsidies and cover a very heterogeneous set of measures, spanning from investment aids and labor subsidies to land protection programs. Finally, the third instrument, dp, includes fully decoupled and partially decoupled agricultural payments. These instruments are generally considered the least distortive.
In the 1980s, the most important instrument was mps. The share of market price support in total support was 82%, whereas the share of direct payments made up only 9%, and of inputs subsidies 9%. In the next two decades the share of market price support has declined and that of direct payments increased substantially (Table 1). By the late 2000s the former had decreased to 49% and the later increased to 38%. In contrast, the share of input subsidies remained about the same.
The choice of policy instrument is correlated with 1) the level of development; 2) the trade status, and 3) the GATT Uruguay Round Agreement on Agriculture (URAA). Figure 1 (left panel) illustrates a positive correlation between economic development and the use of direct payments and Figure 1 (right panel) shows a negative correlation with the export share. In addition, Figure 2 indicates that the shift from market price support to direct payments started in the early 1990s, which was the time of the conclusion of the URAA and has continued during the Doha WTO negotiations.
GATT/WTO rules distinguish between instruments according to their distortionary impact and limit the use of distorting measures while non-distorting measures are not regulated. More specifically, the WTO classifies agricultural policy instruments in green, blue and amber boxes―with the green box for non-trade distorting policies instruments (see Josling and Tangermann, 1999; Tangermann, 1999; Josling, 2000 for more details)  - .
In summary, these empirical indications suggest that the choice of instruments is non-random. As stylized facts, we find that the choice of instruments is correlated with three factors: 1) a country’s level of development; 2) the URAA and the Doha WTO negotiations; and 3) a country’s trade status. We now develop a theoretical model to explain these stylized observations.
We use the same static framework as most models in the literature and consider the choice of governments
Table 1. Support by policy instrument based on OECD PSE database (Million US $).
Figure 1. Relation between the share of decoupled (dpsh) and coupled (mpsh) transfers with the economic development (left panel) and the trade export share (right panel) in OECD.
Figure 2. Evolution of total PSE (billions of US $) and the shares of coupled and decoupled transfers in OECD. (Source: own computation on PSE/CSE database).
between instruments in the absence of existing policies (see, e.g., Hillman and Ursprung, 1988; Foster and Rausser, 1993; Kono, 2006)  - . We assume that governments have perfect information on the impact of the various policy instruments, so there is no room for policy obfuscation. Consider that for some reason, e.g. a dramatic decline in world market prices for agricultural products, the government introduces policies to support producers’ incomes.
We assume that the government has two different policy instruments at its disposal (see Hillman and Ursprung, 1988; Rodrik, 1986; Coate and Morris, 1995)    to transfer income to producers: instruments t and s, which are assumed to have the following characteristics:
Policy t has low transaction costs but high costs of market distortions, and has a positive (negative) impact on government revenue if the country is a net importer (exporter). Policy s causes fewer distortions but is characterized by high transaction costs, and has a strongly negative impact on government revenue, independent of the country’s trade status. Even if the country is a net exporter, the impact of instrument s on government revenue is more negative than instrument t’s impact. One could think of tariffs vis-à-vis lump-sum transfers, or market price support vis-à-vis direct income support as examples of policies t and s, respectively.
As in Kono (2006)  , we assume that governments need both voter support and money to stay in power. Money can be raised both through interest-group contributions and through revenues from the implementation of policy instruments. Our assumptions imply a modified Grossman and Helpman (1994)  model of government decision-making where, in line with Maggi and Rodríguez-Clare (2000)  , the government maximizes a weighted sum of interest group contributions, policy revenues, and total voter support:
where G is government utility, C are the interest-group contributions, R measures the budgetary costs or revenues of the policy instruments, and V is total voter support. t and s are the income transfers of the two policy instruments, and and are the weights that the government gives to respectively revenue considerations and total voter support. β represents the trade balance of that country, and α is an inverse measure for a country’s institutional development.
In developing countries―with relatively underdeveloped institutions―raising revenue through foreign trade taxes constitutes the single largest source of public revenue (Burgess and Stern, 1993; Rodrik, 1995; Bates and Block, 2010)  - . The revenue motive is therefore more imperative in countries with less developed institutions. Since α inversely measures the country’s institutional development, we assume that the weight attached
by the government to the revenue function increases with less developed institutions.
As in Grossman and Helpman (1994)  , we assume that the interest group consists of active lobbyists that solicit income transfers from the government. For this purpose the interest group offers the government a schedule that lists the interest group’s contributions as a function of the income transfers. The interest-group contributions
rise with the level of the income transfers, but at a decreasing rate.4
The policy revenue function, is assumed to be decreasing in policy instrument s, whereas the revenue impact of instrument t can be either positive or negative, depending on the trade status of the country (respectively net importing or net exporting). We assume that to represent that instrument s has a highly negative impact on government revenue, even more negative than instrument t in the case of a net-exporting country. is concave in the income transfers. In line with the marginal impact of instrument t being respectively positive and negative for a net-importing and net- exporting country, the impact of an increase in the trade balance β, on instrument t’s marginal revenue impact is negative, i.e.. The negative revenue impact of instrument t is independent of the trade balance:.
The function for total voter support, , is given by
where the first term, , represents total voter welfare, and the second and third terms, and, measure the total transaction costs related to each instrument. As before, α is an inverse measure for the country’s institutional development. In an unfavourable institutional environment where institutions are underdeveloped and the administrative capacity is low, transaction costs are higher for the same amount of income transfer (Burgess and Stern, 1993)  . Since policy t involves lower transaction costs than policy s for the same amount of income transfer, we assume that. Hence and are the average transaction costs per unit of the income transfers t and s.
As both policy instruments are distortionary measures, is decreasing in the income transfers at an increasing rate. Instrument t is more distorting than instrument s, so.
The equilibrium pair of income transfers is determined by the first order conditions (FOCs):5
Define as the share of policy t in the total income transfer. Comparative statics on the equilibrium for changes in the institutional development of a country, α, and the country’s trade balance, β, yield the following two results.
Result 1: If and:.
Proof: See Appendix 1.
Result 1 implies that in countries with lower institutional capacity, where policy transaction costs are higher and the revenue motive is more important, ceteris paribus, the relative share of income transfer is higher in equilibrium. Hence countries with less developed institutions (α higher) will apply relatively more distorting policies (larger), provided that the transaction costs of the more distorting policy are sufficiently lower than
that of the other policy, and that the less distorting policy has a sufficiently more negative impact on government revenue. The latter condition is always fulfilled if the more distorting policy has a positive impact on government revenue.
To illustrate this result, take the specific case of a net-importing country. In that case, , and the second condition is fulfilled. If in addition the more distorting instrument (t) involves no transaction costs, i.e., the first condition holds as well. It is clear from Result 1 and the proof in Appendix 1 that in this specific case, an increase of the institutional capacity of a country (α lower) will always result in a higher relative share of the less distorting policy instrument in equilibrium (smaller). The result also holds under less strict conditions, which are discussed in Appendix 1.
Proof: See Appendix 1.
Result 2 implies that if the trade balance of a country increases (increase), the relative share of the more distorting policy in the total income transfer decreases (decreases). For example, if for some exogenous reason a country’s imports decrease, ceteris paribus, the country will shift to using the less distorting policy relatively more, although it involves relatively higher transaction costs.
4. Empirical Evidence
To formally test whether our theoretical hypotheses are consistent with the observed evidence on instrument choices we use the share of market price support in total support (mpsh) as a proxy for the instrument t and the share of direct payments in total support (dpsh) as a proxy for the instrument s. As we explained in Section 2, the OECD data on instrument choice cover 28 countries over the 1986-2009 period.6
We proxy the institutional development and administrative capacity of a country by the real GDP per capita (gdppc), taken from the World Development Indicators. As an indicator of the trade status we use the net export share in total production (exsh), based on FAO data.7 To capture the effect of international agreement we include a dummy variable, d_gatt. This dummy takes the value of 1 since 1995 (0 otherwise). 1995 was the first year of the GATT URAA implementation, which has introduced more constraints on the use of highly distortionary policy instruments like mps, than on lower distortionary instruments, like dp. In fact, fully decoupled policies which are not trade distorting are allowed under WTO principles.
As control variable and to account for path dependency and the persistence of policies, we include the level of the dependent variable in the previous period.
Finally, one may argue that from a conceptual point of view, the empirical model should also include the level of support (PSE). By including PSE as explanatory variable, one can analyze the relation between the policy level and instrument choice. However, there are two econometric reasons that render the inclusion of the level of support in our instrument choice equations problematic. First, PSE is endogenous, as the level of support is likely to depend itself on the policy instrument. Second, our explanatory variables, exsh, gdppc, and d_gatt, are also important determinants of the overall protection level. While the first problem could be solved potentially by using a simultaneous equation model, the second problem precludes finding good instruments for PSE in the mpsh and dpsh equations. We therefore do not include PSE in the regressions.
Summarizing, in what follows we will run the following empirical specifications:
where and are expected to be positive;, and are expected to be negative; and, and are expected to be positive. We first run the regressions using a simple OLS estimator, and later will do robustness tests with alternative estimation techniques.
OLS regressions are reported in Table 2, with Columns (1) and (5) for mpsh and dpsh regressions, respectively. All the relevant variables have their expected signs and are statistically significant at the 95% level or more. Moreover, the adjusted R2 of the models, ranging from 0.58 to 0.80, indicates a good explanatory power of the selected variables.
gdppc has a significant negative effect on mpsh, the share of market price support, a relation that turn into positive when the share of direct income support, dpsh, is considered. These results are consistent with our theoretical argument that countries with lower administrative capacity and lower institutional development have a preference for price support. Also in line with our hypothesis, the net export share has a significant negative effect on mpsh, and a positive and significant effect on dpsh.
The 1994 GATT Agreement as captured by the dummy d_gatt is significantly negatively correlated with mpsh, and positively with dpsh. These results are consistent with the argument that the GATT constraints exert an effect on instrument choices: mpsh declined on average after the implementation of the 1994 URAA, and dpsh increased. In all regressions the coefficients of the lagged value of the dependent variable are positive and strongly significant. The magnitude of the lagged coefficients, ranging from 0.66 to 0.86, confirm a strong level of persistency in instrument choice.
Table 2. Regression results.
Notes: t-statistics based on clustered standard errors under coefficients. All regressions include also a constant term. The System GMM estimator is implemented in STATA using the xtabound2 routine, with the option collapse to limit the instruments proliferation. ***, ** and * p-value < 0.01, 0.05, 0.10 respectively.
We performed a series of additional robustness test. First, a potential problem in applying OLS to our specifi- cation is that the lagged dependent variable can be endogenous to the fixed effects in the error term, which gives the well know dynamic panel bias (see Roodman, 2009)  . A first step to deal with this is by removing the fixed effects from the error term, running the standard Least Square with Dummy Variables (LSDV) estimator. In doing so, we also control for any unobserved heterogeneity that are correlated with our explanatory variables.
The LSDV regression results are reported in Columns (2) and (6) for the mpsh and dpsh equation, respectively. As expected, the estimated coefficient of the lagged dependent variable is now lower in magnitude, but still strongly significant and has its expected sign in the two regressions. Generally speaking, all the OLS results are confirmed by the LSDV estimator, the only different being the effect of the trade status that now is estimated with less precision.
A potential problem with the (dynamic) LSDV estimator is that, when applied to a panel structure where N > T, namely the year dimension (T) is lower than the number of countries (N), it suffers of the so called Nickell bias, due to the endogeneity of the lagged dependent variable. To address this potential source of bias, the system GMM estimator proposed by Blundell and Bond (1998)  is used. This means estimating a system with the first-differences and the level equations, where the endogenous variables are instrumented by their level in the first-differenced equation and first-differenced instruments for the equation in level. Columns (3-4) and (7-8) report the results, considering both the one step and two step GMM option. As can be seen at the bottom of the columns, the Arellano-Bond tests for autocorrelation (AR1 and AR2), confirmed the presence of first order, but no second order, serial correlation, suggesting that the model dynamic is correctly specified. Moreover, the standard Hansen test confirms that in all cases our set of instruments is valid.8
The system GMM regressions strongly confirm our previous results, showing that the trade status and the level of development affected negatively the share of market price support and positively the share of direct payments, as well as the GATT dummy causes a shift from market price towards direct income support. Hence we can conclude that our results are very stable to different estimators.
In this paper we developed a theoretical political economy model to explain how various factors affect policy instruments choices. The theoretical model provides hypotheses on policy instrument use which are consistent with three stylized facts, i.e., 1) their correlation with the level of a country’s institutional development; 2) their correlation with a country’s net trade position; and 3) the impact of GATT/WTO rules. Moreover, the model explains these key observations with a rational choice political economy model without having to rely on imperfect information of policy effects or on theories of bureaucratic inertia and obstruction.
In the last part of the paper we econometrically tested these theoretical predictions using OECD data on instrument choice in agricultural policy. Our empirical analysis confirms the hypotheses and provides strong evidence that the shift from distortionary to less distortionary instruments is positively influenced by institutional development, the net trade status, and the GATT/WTO framework. Moreover, we also find evidence of strong persistency of policy instruments. The main empirical limitation of the paper lies in the difficulty to directly isolate the effect of institutions from the one of economic development, due to the high correlation between the two dimensions. Hence, future development should pay particular attention to this issue, for example by establishing the extent to which different political institutions, such as forms of government or electoral rules, affect differently the patterns of policy instrument choice.
The authors thank Jason Hartell and Goedele De Nolf for research assistance. This work was financially supported by the Research Foundation-Flanders (FWO), the European Commission and KU Leuven (OT, EF, and Methusalem projects).
Proof of Result 1:
Using Cramer’s rule, it follows from (3) that
Under the assumptions that and, it follows that
From the assumptions to secure uniqueness of the equilibrium and reaction function stability, , and (see footnote 5), it follows that
Performing comparative statics on gives
Using (6) and (7), it immediately follows that.
The conditions imposed under Result 1 are stricter than necessary. Substituting Equations ((4) and (5)) into Equation (8) and rearranging gives the following condition for:
It is clear that condition (9) also holds, such that, for several cases that violate the conditions imposed in Result 1. For example, condition (9) is satisfied when one of the two conditions is violated, but the other condition is sufficiently non-binding. Also, at, condition (9) holds when. Hence Result is more general than what the two conditions seem to suggest―the attractiveness of these two conditions is in their intuitive interpretation.
Proof of Result 2:
Using Cramer’s rule, it follows that
Since and, it follows unambiguously that
Performing comparative statics on gives
Using (13), it follows that.
Table A1. Support by policy instrument based on OECD PSE database (Million US $).
Notes: The policy instruments considered are based on the following items of the PSE database: “market price support” refers to support based on commodity outputs (items A1 and A2, of the PSE database); “input subsidies” is the sum of payments based on input use and miscellaneous payments (items B and G); “direct payments” refer to different payments decoupled or partially decoupled from production (items from C to F). mpsh, ish and dpsh are the share of market price support, inputs subsidies and direct payments on total support, respectively. Source: own computation based on OECD PSE/CSE database (2010).
1This argument may apply more widely, while there have been some studies in the general literature on explaining instrument choice, in particular why governments chose inefficient policies to redistribute income or protect certain groups (e.g. Cassing and Hillman, 1985; Rodrik, 1986; Coate and Morris, 1995; Acemoglu and Robinson, 2001)  - , these studies are almost exclusively theoretical. Only recently have there been a few empirical studies on the determinants of instrument choice, including Kono (2006)  and Ederington and Minier (2006)  .
2For stylized facts on the historical evolution of agricultural policy instruments, see e.g. Tracy (1989)  , Josling (2007)  , Swinnen (2009)  ; on the historical evolution of trade policy, see e.g. Irwin (2008)  ; and Williamson (2003)  .
3Initially the PSE calculations were only for OECD member states but more recently also some other countries, such as China and Brazil, are covered. For countries not belonging to the OECD, the time coverage is not complete: the first year observation is around 1990-1992 and the last is 2007.
4Subscripts t and s denote partial derivatives.
5The Hessian matrix of the government’s objective function is. In order to obtain a global maximum and to perform comparative statics, this matrix must be negative definite. Since all the Hessian’s elements are negative, i.e., , and, the matrix is negative definite if (Winston, 2004). To secure uniqueness of the equilibrium and reaction function stability, in line with Brander and Spencer (1983)  and Dixit (1984)  , we assume that the own effects of the income transfers on marginal contributions, revenue, and total voter welfare exceed cross effects such that and.
6In some cases, the PSE and the instruments share are negative: 13.3% of the observations have negative values for mpsh and 7.2% for dpsh. We dealt with this problem in three different ways. First, we recalculated mpsh and dpsh variables using absolute values of each instrument. Second, we ran the regressions excluding the negative values for mpsh or dpsh. Third, we ran the regressions with the subsample of the OECD member countries only, where there are no negative values. The model results are robust across these different samples.
7More specifically, exsh = (export value − import value)/production value.
8According to Roodman (2009), the instrument count does not exceed the number of groups and, to control for instrument proliferation that cause a weak Hansen test, we used the xtabond2 collapse option in STATA, instead of all available lags for instruments.