Payment decoupling and intra-European calf trade

1. Introduction

Since the early 1990s fundamental reforms have been made to the Common Agricultural Policy (CAP) of the European Union (EU). An increasing reorientation of production incentives from price support towards direct payments has been the focus of these reforms (Anania, 2010). The 2003 reform, in particular, aimed to sever the link between agricultural production activities and direct payments to producers (Swinnen, 2008). It replaced the existing system of coupled direct payments (CDP) with the decoupled single farm payment (SFP) scheme. Furthermore, it gave member states several options to partially retain CDP in certain sectors, in particular in the beef sector. Other sectors where such partial decoupling was possible were sheep and goats, and cereals and oilseed production. This discretion led to the coexistence of different implementation schemes among EU member states. In the beef sector, France, e.g. under the strong political influence of farmers, basically opted for keeping payments as coupled as possible, while Germany decoupled fully. Although Scandinavian member states and the Netherlands opted for a partially decoupled slaughter premium for calves and male cattle/bulls, most South European member states opted, besides retaining partially coupled slaughter premia for calves, for a suckler cow premium.

We expect this coexistence to distort production incentives. In partially decoupling member states, farmers still view the coupled payments as part of their gross margin, whereas decoupled payments do not impact gross margins. This changes willingnesses to pay for inputs between member states depending on the decoupling choices. If the inputs in questions are produced EU-wide and traded in a common intra-EU market (both criteria hold for calves), this results in additional trade flows from member states with fully decoupled payments towards those choosing to only partially decouple. If found, these additional trade flows can be viewed as artificial because they result from the discretion granted in the reform implementation. Such trade flows would represent a misallocation of production factors among member states with different decoupling regimes. Compared with the fully decoupled policy option, we expect greater factor use in those states which chose to only partially decouple. To examine these expected effects we apply a gravity trade model with heterogeneous firms to intra-EU calf trade where the bilateral trade data come from Statistical Office of the European Union (EUROSTAT). Calves are an interesting commodity because (i) calves are heavily traded within the EU, (ii) calves constitute an important input for veal- and beef-producing farms and (iii) particularly in the veal and beef sector, important discretion for the implementation of the CAP reforms was handed over to the individual member state. Thus, studying the impact of the resulting heterogeneity in the direct payment regimes across the EU on trade flows seems an interesting endeavour. Our empirical investigation of intra-EU calf trade confirms the presence of the expected trade distortions. The coexistence of different implementation schemes among EU member states led to additional calf trade flows from fully decoupling member states to only partially decoupling member states.

To our knowledge no research has addressed the economic implications of the coexistence of different policy reforms for intra-European trade with intermediate agricultural products. Our approach follows Anderson (2009) who integrates the heterogeneous firms trade model of Helpman, Melitz and Rubinstein (2008) with the Anderson and vanWincoop (2003) concept of multilateral resistance. In this way we simultaneously adjust for two sources of omitted variable bias, non-random selection of firms into trade and multilateral resistance.¹ Hence, we account for both zero and unidirectional trade flows. Following Helpman, Melitz and Rubinstein (2008) we use both a two-step nonlinear least squares (NLS) estimator and a semiparametric least squares estimator for our gravity equation. To confirm whether sample selection and/or firm heterogeneity are present, we employ the decomposition procedure proposed by Belenkiy (2009).

Our contribution is twofold; first, we address the unintended consequences of the 2003 CAP reform due to the artificial trade induced by differential decoupling; and second, we are the first to apply Anderson's (2009) heterogeneous firms trade model to an intermediate agricultural product.

In the next section we briefly review the provisions of the 2003 CAP reform for intra-EU calf trade. Following this we present a heterogeneous firms trade model for the agricultural intermediate product live calves. We continue with the empirical framework, estimation procedures and econometric results. The final section concludes and elaborates policy implications.

2. The post-reform EU beef market

Perhaps the most important feature of the 2003 reform was the decoupling of direct payments from production. However, the politics of this decision turned the simple idea of full decoupling into a much more complex system; member states retained an array of options regarding the extent to which the SFP remained tied to production. In the EU beef market members could fully decouple payments or choose one of three options for partial decoupling. These options include various combinations of slaughter premia for calves, suckler cows, adult cattle and male cattle (see Table 1 for details; Deblitz, Keller and Brüggemann, 2007). In addition, member states were given discretion as to the start of the implementation process. The final policy choices are summarised in the lower part of Table 1.

New member states (accession in 2004 or later) were treated differently. They generally implemented the direct payment component of the CAP in the form of the so-called simplified area payment scheme (SAPS). For the beef and veal sector, this was equivalent to a full decoupling of all formerly coupled payments. Hence, in the empirical part we treat them like other fully decoupling member states. Slovenia was the only new member state which retained the possibility to continue using partially coupled payments.

Beef supply and demand imbalances are a common feature among EU countries. Substantial regional and farm heterogeneity, both between and within member states, characterises the structure of beef production. Differences in natural conditions and opportunity costs lead to clusters of specialisation. Veal production occurs mainly in the Netherlands, Belgium and France, while fattened bull production is largely concentrated in Italy, Spain and France. Dairy production is concentrated in Germany and France, but is also of substantial importance in Denmark, Ireland, the Netherlands, the UK and in Poland. Suckler cow production is concentrated in France and Spain.

Regional heterogeneity reflects animal genetics; male dairy calves are by-products of dairy production but not well suited for beef production. Continental or cross breeds are better suited for beef production but are less abundant in intensive dairy regions. Dairy calves in excess of replacement rates are commonly used for veal production. Demand, on the other hand, is more homogeneous and stable. Higher than average levels of beef and veal consumption are found in Denmark, France and Luxembourg, while Eastern European countries are below average.

Given this supply and demand structure, calves for beef and veal production are regularly traded within the EU. Calves are transported from excess supply regions to excess demand regions. Trade flows, however, are typically unidirectional. Calves are either exported to or imported from a trading partner but rarely exported and imported simultaneously. Usually, member states only have a few trading partners. Thus, zero trade flows between member states are common. An overview of the main exporters and importers in the EU is given in Table 2.²

During the reform phase-in period, calf trade was also affected by a major outbreak of blue tongue disease. Although usually not fatal, blue tongue reduces productivity in milk, veal and beef production. In order to combat the disease, vaccinations became mandatory and intra-EU calf trade restrictions were imposed. We expect the prevalence of blue tongue to disrupt bilateral trade (see Ihle, Brümmer and Thompson, 2012).

2.1. Microeconomics of decoupling

Differential degrees of decoupling among member states affect intra-European calf trade. Our discussion here is brief as the production effects of decoupling for international trade have been discussed elsewhere. Rude (2008) reviews the literature on ex-ante analyses of decoupling. Bhasker and Beghin (2009) summarise the literature on the potential production effects of decoupled payments. Beard and Swinbank (2001) discuss in detail how the concept of decoupling emerged in policy. Here, we review only those provisions pertinent to intra-European calf trade.

Figure 1 depicts a stylised ceteris paribus scenario where one member state fully decouples (left panel), and the other only partially decouples (right panel). $Sc$ is the aggregate supply function for dairy calves and $Dc$ the aggregate demand function by cattle farms. The superscripts indicate the extent of policy decoupling.

The introduction of CDP shifts the original aggregate demand curve $Dc$ upward to $DcCDP$⁠. This upward shift is a consequence of the headage coupling of direct payments. Cattle farmers view these payments as part of their gross margin, hence, their willingnesses to pay for calves increase.

If a member state opts for a fully decoupled SFP, cattle farmers view the decoupled payments as a lump sum subsidy as opposed to part of their gross margin. Accordingly, their corresponding willingnesses to pay for calves will fall. Graphically this is shown as a downward shift from $DcCDP$ to $DcSFP$(left panel).³ In the presence of a common market for calves this demand shift not only impacts the market equilibrium in the decoupled member state but spills over to those states which have retained CDP. In the absence of trade costs the new market clearing price $p∗$ equalises the marginal willingnesses to pay in both markets triggering additional exports from the decoupled to the non-decoupled member state. Because these additional trade flows are a direct consequence of differential decoupling, they are regarded as artificial side effects of the 2003 reform.

The impact of an increasing degree of decoupling in an exporting member state on trade is positive, as can be clearly seen in the left panel of Figure 1. If the importing member state (right panel) was to decouple as well, then an increasing extent of decoupling would lead to a lower level of imports, because the demand curve would shift downwards (not shown in the graph). Hence, the effect of decoupling on trade flows is asymmetric between exporters and importers. The former trade more with countries with higher degrees of decoupling, while the latter trade less.

3. An intra-European calf trade model

The EU calf trade is characterised by regional heterogeneity, determining the direction of intra-EU trade flows. However, both the size and the occurrence of trade will also be affected by farm heterogeneity, i.e. the productivity of single farms. While single farms rarely engage in direct exports to other regions (this is usually organised by marketing firms), Gopinath, Sheldon and Echeverria (2007) suggest there is an indirect link between the productivities of single farms and regional exports. Depending on the distribution of productivities, there may be some farms which are productive enough to produce exportable calves even though the average farm would not do so. Engaging in an export market is often difficult due to additional market entry costs; for calf exports to other member states, foreign language skills might be required, knowledge of local institutions in the import market has to be acquired or additional veterinary measures could be necessary. Farm productivities must be sufficiently high to generate enough revenues to cover the (usually fixed) costs of entering export markets (Kandilov and Zheng, 2011). Hence, heterogeneity in terms of farm productivity can explain why some regions export calves while others do not and why some regions export more and others less.

These characteristics of the EU beef market imply that standard gravity models (e.g. Anderson and van Wincoop, 2003) are too restrictive because they impose non-zero and symmetric trade flows between all regions. However, heterogeneous firms trade models (Melitz, 2003) are a solution. Utilising information on firm productivity distributions and fixed trade costs, these models can explicitly account for zero and unidirectional trade flows. These models are also suitable for indirect exports through trading intermediaries (Ahn, Khandelwal and Wei, 2011), the most common marketing form in agriculture. In combination with the fixed costs for engaging in agricultural trade, the heterogeneous firms trade model seems to be an ideal starting point for modelling agricultural trade flows (Gopinath, Sheldon and Echeverria, 2007).

In order to develop a suitable sectoral single product trade model⁴ of intra-European calf trade we start from the basic theoretical model proposed by Anderson (2009), which in turn is an extension of the standard gravity model of Anderson and van Wincoop (2003),⁵. Anderson combines a constant elasticity of substitution (CES) expenditure system with a monopolistic competitive and heterogeneous firm structure. A share variable $Vijt$ then explicitly accounts for the share of exporters which select to trade, thus allowing for zero and unidirectional trade flows. This idea traces back to Helpman, Melitz and Rubinstein (2008), so that the resulting model corresponds to a structural Helpman, Melitz and Rubinstein model.

The model structure (Equations (1)–(3)) extends the standard gravity model by two additional components: (i) the model is adjusted for firm heterogeneity by including a share variable $Vijt$⁠, and (ii) given a monopolistic competitive market structure with a sufficiently large number of firms, an additional standard markup parameter α is included. The main difference to Helpman, Melitz and Rubinstein is the inclusion of general equilibrium trade effects, i.e. $Πit1−σ$ and $Pjt1−σ$⁠.

Furthermore, the underlying farm heterogeneity, in terms of productivities, allows for the possibility that in exporting country i none of its $ni$ firms is productive enough to profitably export to country j. In particular, additional fixed bilateral trade costs $fijt$⁹ might render exports unprofitable when revenues are too low to cover both normal costs (i.e. production costs and variable bilateral trade costs) and the additional fixed bilateral trade costs (Anderson, 2011).

The relation between fixed bilateral trade costs and firm productivities determines both the occurrence of trade and the proportion of i's exporting firms. Depending on the distribution of farm productivities the number of i's $ni$ firms which can export profitably to j will vary. According to Helpman, Melitz and Rubinstein this selection mechanism can be implemented by the share variable $Vijt$ (⁠$Vijt∈[0,1]$⁠); where, $Vijt=0$ indicates zero trade (i.e. $Mijt=0$⁠) and $Vijt=1$ exports of all $ni$ firms to j. The framework also allows for asymmetric trade flows $Vijt≠Vjit$ (i.e. $Mijt≠Mjit$⁠), which also include unidirectional trade flows, with $Mijt>0$ and $Mjit=0$⁠, or $Mijt=0$ and $Mjit>0$ (Helpman, Melitz and Rubinstein, 2008).

4. Empirical framework

We now consider the econometric implementation of our sectoral single product trade model of intra-European calf trade. The applied econometric methodology is discussed in detail in Helpman, Melitz and Rubinstein (2008). In principle, the methodology mimics a sample selection model with the exception that the probit estimates are also used to proxy the unobserved share variable $Vijt$⁠.

For the econometric estimation of Equation (5) a number of additional steps are required. First, a proxy for the unobserved share variable $vijt$ is required. Therefore, we utilise the zero-profit condition (Equation (4)) to construct a latent variable model. Based on this model a probit estimation is done. The probit estimates then are used to construct the proxy for $vijt$⁠. Second, the variable distance $dijt$ must be replaced by a suitable set of observed variables. In addition to the common proxies for distance, variables which describe the implementation of decoupling will be used. Third, the time-varying exporter (i) and importer (j) specific terms will be subsumed in exporter- and importer-fixed effects¹⁰ and the constant terms in Equation (5) will be subsumed in a single intercept.

$Φ$ is the cdf of a standard normal distribution.¹¹

Helpman, Melitz and Rubinstein (2008) suggest to use the probit estimates $ρˆijt$ to construct a consistent proxy for $vijt$⁠. The proxy is defined as $v¯ˆijt∗(δ)≡ln⁡(exp[δ(zˆijt∗+λ¯ˆijt∗)]−1)$⁠. $δ$ is a parameter combining both the elasticity of substitution and the shape parameter of the assumed Pareto distribution (details are given in the Technical Appendix in supplementary data at ERAE online). $zˆijt∗$ is estimated as $zˆijt∗=Φ−1(ρˆijt)$⁠, and the inverse Mills Ratio $λ¯ˆijt∗$ as $λ¯ˆijt∗=ϕ(zˆijt∗)/Φ(zˆijt∗)$⁠.

Second, a refined specification for the distance term $−γdijt$ is required. In addition to physical distance $DISTijt$ we use a binary indicator for neighbouring country pairs $ADJijt$⁠. Because some of the countries in our sample became member states of the EU during the observation period, we use a dummy variable $NMSit$ to control for differences in trade costs before these countries became member states of the EU. For the European calf market over the observation period, variables capturing the impact of blue tongue on trade costs should also be added because the trade restrictions during disease outbreaks act as additional barriers to trade. For the exporting country, we use the numbers of outbreaks $BTi1t$⁠, of suspected cases $BTi2t$⁠, and of confirmed cases $BTi3t$ to capture various aspects of the disease dynamics.

In Equation (9), the parameter $βuη≡corr(uijt,ηijt)(σu/ση)$ associated with $λ¯ˆijt∗$ corrects for a potential upward bias caused by ignoring non-random selection into trade. Not to account for this bias would imply that small and far distant positive trade flows are overvalued. The switch from the i.i.d. error $uijt$ in Equation (5) to $ϵijt$ reflects this Heckman-type correction; thus, $ϵijt$ is an i.i.d. error term satisfying $E[ϵijt|,Tijt=1]=0$⁠. On the other hand, the term $ln⁡(exp[δ(zˆijt∗+λ¯ˆijt∗)]−1)$ corrects for a potential downward bias caused by non-consideration of firm heterogeneity (Larch et al., 2012). On average no firm might be able to export profitably to a remote destination; nevertheless, given firm heterogeneity, some firms might be able to export with positive profits. Hence, non-consideration of firm heterogeneity would cause an omitted variable bias.

In order to avoid the Pareto assumption on the productivity distribution, Helpman, Melitz and Rubinstein (2008) propose a semiparametric alternative for modelling firm heterogeneity. They propose to substitute a polynomial approximation of degree 2 in $z¯ˆijt∗=zˆijt∗+λ¯ˆijt∗$ for $ln⁡(exp⁡[δ(zˆijt∗+λ¯ˆijt∗)]−1)$⁠. This polynomial approximation adds additional flexibility because a specific distributional assumption is no longer necessary.

To deal appropriately with multilateral resistance in panel data setting, a time-varying fixed effects approach would be the first choice. A time-varying fixed effects approach, however, has one disadvantage; it is not applicable to exporter- and importer-invariant variables, i.e. variables which do not vary within an exporter or importer. The lack of within-group variation makes it impossible to identify these variables. Decoupling policy variables, however, are by nature exporter- and importer-invariant; a member state can only choose one policy option. An exporter/importer/time fixed effects approach is an alternative, which in our specific application should be sufficient to deal with multilateral resistance for the following reasons: first, exporter- and importer-fixed effects appropriately deal with the cross-sectional variation. The problem is the remaining time-series variation with which exporter- and importer-fixed effects cannot deal. In order to mitigate this problem, additional time fixed effects are included which should at least partially capture the time-series variation, i.e. the general market development should be controlled for by including time fixed effects. There might still be an issue with the residual member state specific time-series variation. This variation, however, should be highly correlated with decoupling policy (we also take care for the other major event in European calf markets, blue tongue).

4.1. Endogeneity of policy

If policy endogeneity is present parameter estimates could be biased. This can be due to omitted relevant variables, simultaneity, or measurement error (Baier and Bergstrand, 2007; Grant and Lambert, 2008). In our case however, we can exclude simultaneity and measurement error. Simultaneity could be an issue if the extent of decoupling and export decisions were determined within the same time period. This situation can be ruled out as member states make a one-time adoption implementation decision. Neither are measurement errors problematic as policy variables $ΔitSFP$ and $ΔjtSFP$ are defined as continuous interval variables so that attenuation bias is sufficiently handled (Baier and Bergstrand, 2007).

On the other hand, omitted relevant variables can be a source of endogeneity. Unobservable components, such as managerial skills or farm structure, included in the error term may be correlated with both trade flows and policy. Baier and Bergstrand (2007) argue that ‘instrumental variable and cost function approaches do not adjust for [omitted variable] bias well’ (p. 1). Alternatively, they suggest a panel data fixed effects approach. In the absence of these fixed effects $ΔitSFP$ and $ΔjtSFP$ could misleadingly explain variation in trade flows due to unobservables; thus biasing the policy effect. As we include importer and exporter fixed effects in all specifications policy endogeneity should be appropriately handled.

5. Data and model specification

Our sample consists of 15 importing and 18 exporting member states¹² for which intra-EU trade flows for live calves are observed: 5 (7) new member states, and 10 (11) old member states.¹³ For the sample period from 2003 to 2007 we use the annual value of bilateral live calf trade flows. Therefore, there are 1,275 ((15 × 17) × 5 years) observations. The number of non-zero trade observations is 412, which is nearly 32 per cent of the sample (i.e. 68 per cent zeros).

Bilateral live calf trade data $Mijt$ (CN8-Code 01029005 and 01029029) come from EUROSTAT. The codes refer to live bovine animals not for slaughter with a weight of less than 80 kg (CN8-Code 01029005) and more than 80 kg and less than 160 kg (CN8-Code 01029029) and are given in monetary values. Because decoupling policy is not specific for weight category, data for both weight categories are aggregated per country pair. Therefore, there are 1,275 ((15 × 17) × 5 years) observations in total. Data on physical distance $DISTijt$ and common border $ADJijt$ come from the Centre d'Etudes Prospectives et d'Informations Internationales (CEPII). Data on the number of blue tongue outbreaks $BTi1t$⁠, of blue tongue suspected cases $BTi2t$⁠, and of confirmed blue tongue cases $BTi3t$ come from World Animal Health Information Database (WAHID).

A binary variable is used for country accession to the EU. The dummy variable $NMSit$ takes the value one when the exporting partner was not yet an EU country at the time of observation, otherwise it takes the value zero.

The decoupling indices $ΔitSFP$ and $ΔjtSFP$ are constructed from official figures of the Directorate-General for Agriculture and Rural Development (DG Agri). For each year after the start of the reform we compute the total value of CDP for beef and veal production and divide it by the average annual value of CDP for the beef and veal sector in the 3 years before 2005. Subtracting the resulting figure from one gives our index of the extent of decoupling, which is a continuous variable in the unit interval.¹⁴

For the estimation of the exporter selection model additional variables which determine the probability of exporting but do not affect the actual value of bilateral trade are required. The inclusion of these variables is necessary since otherwise the parameters of both the exporter selection model and the gravity model could not be identified. In accordance with Gomez-Herrera, Martens and Turlea (2013) we select country-level governance indicators for this purpose because they likely capture the extent to which a specific import market is easily accessible. Hence, these indicators approximate bilateral fixed costs of trading. We use indices for quality of regulations $REG.QUAL.jt$⁠, governmental efficiency $GOV.EFF.jt$ and rule of law $RoLjt$ from the World Bank, Worldwide Governance Indicators (World Bank). For the countries in our sample, these indicators range from −2.5 to 2.5.

We now evaluate different model specifications to deal with issues of sample selection, firm heterogeneity and endogeneity. Sample selection bias is likely in the presence of zero trade flows (Haq, Meilke and Cranfield, 2013). Our data set includes zero trade flow observations, thus contributing to a problem of non-random selection into trade. The presence of firm heterogeneity bias is not readily apparent. Belenkiy (2009) argues that the importance of the firm heterogeneity term depends on the elasticity of product substitution and is expected to be significant only in the presence of low elasticities of substitution. As agricultural products typically have high elasticities of substitution, firm heterogeneity is not expected to be important. On the other hand, we analyse disaggregate trade for an intermediate input. Because the varieties of calves in the EU show marked differences in their suitability for beef production both between and within the countries, the elasticity of factor substitution is not likely to be high. Therefore, for intra-EU calf trade firm heterogeneity could be an important issue. Furthermore, if there is a high correlation between the sample selection term and the firm heterogeneity term, Larch, Norbäck and Urban (2010) caution that the Heckman estimator may be biased. We statistically test for firm heterogeneity using Belenkiy's decomposition procedure.

Empirically we estimate the standard gravity model (Anderson and van Wincoop, 2003) and the ‘heterogeneous firms’ model. The former is estimated by ordinary least squares (OLS) and the latter is estimated in two variants. First, we use a two-step NLS procedure based on the Pareto distribution for productivities. Second, we use a polynomial approximation for an unknown distribution. All models except the standard gravity model are estimated in two stages, where the first stage is a standard probit regression. Identification is facilitated by variable exclusion restrictions.¹⁵ Here, fixed bilateral trade costs $fijt$ (proxied by $REG.QUALjt$⁠, $GOV.EFF.jt$ and $RoLjt$⁠) are excluded from the second-stage estimation.¹⁶ The probit estimates are used to construct correction terms for sample selection and firm heterogeneity.

6. Estimation results

The parameter estimates are shown in Table 3. The second column reports the probit estimates for the probability of exports from member state i to member state j. In addition to most of the time fixed effects, we find that the common border variable (ADJ_ijt) and the dummy variable for the exporting partner being one of the new member states before accession (NMS_it) significantly affect the probability of non-zero trade flows. A common border increases this probability while NMS exporters have lower export participation probabilities before accession. The rule of law index of the importing partner (RoL_jt) significantly decreases the probability of non-zero calf exports to this country.

Small differences in the estimates are observed among the standard gravity model (column 3) and the alternative specifications of the full gravity model with exporter selection and firm heterogeneity (columns 4 and 5). The parameter estimates from the extended models are generally smaller in magnitude than those from the standard model. This suggests that the OLS estimates of the standard gravity model are biased upward. Moreover, the coefficient of determination of the heterogeneous firms models (0.59) is noticeably higher than the standard model (0.54). Among the two variants of the full specification, the Pareto-based NLS estimate shows an insignificant estimate for δ, the parameter associated with firm heterogeneity. However, the results from the semiparametric approximation show statistically significant estimates for the polynomial terms $z¯ˆijt∗$ and $z¯ˆijt ∗2$⁠. The underlying distributional assumption might be causing the non-significant estimate for δ. Hence, the semiparametric model in the last column of Table 3 is our preferred model.¹⁷

As will be shown henceforth, all parameter estimates in the preferred model conform to our theoretical expectations and are consistent with those of other researchers (e.g. de Frahan and Vancauteren, 2006; Olper and Raimondi, 2008; Sun and Reed, 2010). The estimates of physical distance $DISTijt$ and common border $ADJijt$ show the usual signs (negative and positive, respectively). As expected, calf imports decrease with trading partner distance while a common border favours trade. The negative estimate for the dummy variable NMS_it indicates that export opportunities expanded after the Central and Eastern European Countries joined the EU.

We find too that the signs and magnitudes of the time fixed effects reflect market developments over the sample period. For instance, beginning in 2003 (and especially in 2004) young bull calf prices followed a strong secular increase. This trend lasted until animal disease blue tongue outbreak in mid-2006. As a result of weaker prices bull fattening became more attractive in less competitive member states, thus increasing production and reducing exports. This development is reflected in negative signs on the time fixed effects variables for 2003–2005, with the relatively large negative value in 2004. However, in 2006 and 2007 bull prices stagnated and the negative effects of blue tongue became more apparent. This is reflected by positive signs in 2006 and 2007, implying greater exports. We also control for blue tongue disease. Yet, we cannot find a significant impact.¹⁸

Before we turn to the policy variables, we assess the relative importance of firm heterogeneity and exporter selection. This issue is investigated in more detail by using Belenkiy's decomposition procedure. The principal idea of this decomposition procedure is to estimate the standard gravity model twice: first, extended by a sample selection term, and second, by a firm heterogeneity term. The former corrects for a downward bias whereas the latter corrects for an upward bias. The significance levels of the terms then indicate if the correction is required or not. The results of the decomposition procedure are shown in Table 4. The estimation results in the last column for the pure firm heterogeneity model, i.e. the standard gravity model extended by a linear firm heterogeneity term $z¯ˆijt∗$⁠, (last column, OLS – $z¯ˆijt∗$⁠) suggest that firm heterogeneity is highly significant. Nevertheless, when comparing the results for the ‘Sample Selection’ and the pure firm heterogeneity models, we find that the upward correction of the ‘Sample Selection’ approach is more important than the downward correction of the firm heterogeneity approach. This is evident as the parameter estimates of the ‘Sample Selection’ model are much closer to those of the preferred model (column 3).