Coupled Agricultural Subsidies in the EU Undermine Climate Efforts

The CAPRI Modeling System

The present analysis was based on CAPRI Stable Release 1.3 (STAR 1.3, publicly available from www.capri-model.org), but with updated data in the area of GHG emission estimates. The CAPRI model is a partial equilibrium simulation model covering the agricultural sector (Britz and Witzke 2014). The model simulations provide results for the global impact on production and trade in the agricultural sector, aggregated to about forty trade blocks, and detailed results for NUTS2 (Nomenclature of Territorial Units for Statistics) regions within the EU. Countries outside the EU are represented in a more simplified fashion than EU countries (EU+), and therefore less detailed information on production and emissions is available for these. Trade flows between the forty regions are modeled based on the Armington assumption of product differentiation by origin. With regard to global trade, the model includes policy data on tariffs, tariff rate quotas, and the trigger price system of the EU. For EU countries, the model also contains a detailed representation of the CAP's policy measures, thus making it suitable for analyzing the impacts of agricultural policy reform scenarios. In addition, we have added VCS measures for all EU MS to the model in order to better represent the production coupling of the CAP and simulate the impact of VCS on GHG emissions.

CAPRI is a comparative static model, meaning the policy impact is inferred from a comparison of a baseline and a policy scenario at a specific point in time. In the present study, this point in time was set as 2030, after the end of the next multiannual financial framework.ⁱⁱ The CAPRI model is frequently used to assess the impact of changes in the CAP on aspects such as production, trade, and selected environmental indicators. Recent examples include: simulations of the impact of currently proposed EU free trade agreements and carbon taxes on GHG emissions (Himics et al. 2018); simulations of the impact of the so-called “greening” measures in the 2013 CAP reform (Gocht et al. 2017); and, used together with other models, simulations of the impact of climate change on agriculture (Blanco et al. 2017).

GHG Emissions in CAPRI

CAPRI's coverage of GHG emissions is global, but the method used to calculate emissions varies depending on the availability of detailed production data from the simulations. For EU+ countries,ⁱⁱⁱ more details on production are available than for other regions, allowing a bottom-up computation of emissions based on production technology. For all regions, the main direct and indirect emissions of methane (CH₄) and nitrous oxide (N₂O) from agriculture are covered^iv (representing agricultural emissions according to the UNFCCC classification). The CO₂ emissions from land use, land use change, fertilizer production, and energy use on farms are omitted from our analysis, as they are not yet covered globally in the CAPRI model. Gerber et al. (2013) estimate that about 75% of emissions from beef production are in the form of N₂O and CH_4, and about 25% are CO₂ emissions from land use and land use change, but with large uncertainties. The effect on our results of omitting emissions from land use and land use change are unclear, as the importance of omitted emissions and production methods varies across regions.

To compare emissions of different gases, Global Warming Potential (GWP) was used to convert all gases into carbon dioxide equivalents^v (CO₂-eq.). The climate change induced by the change in emissions would also have an impact on agricultural systems. That feedback is not modeled in CAPRI.

For EU+ regions, emissions are computed endogenously in the CAPRI model based on detailed input and output data. This means, for example, that changes in the feed mix for animals due to a policy change can be captured and thus result in changes in emissions. For the main emission sources, the calculation is performed using a more detailed method (Tier 2 in the 2006 IPCC 2006 guidelines), while for some sources with lower total contributions to emissions, a simplified method (Tier 1) is used. Emissions are calculated per hectare of land or per animal production activity, and then allocated to commodities associated with those agricultural activities. A more detailed description of the method is available in Leip et al. (2010), Pérez Domínguez (2005), and Pérez Domínguez et al. (2012).

The high level of detail on production technology used to compute emissions in the EU+ is not available for other regions. For these regions, computations of GHG emissions are based on estimated emission intensities (EI) per tonne (metric ton) of product, without capturing endogenous changes in the composition of inputs that may take place in simulation (Pérez Domínguez et al. 2012). This means production technology outside the EU+ is assumed not to be affected by policy changes in the EU. To calculate total emissions in each scenario, the emissions coefficients are multiplied by production level.

EI for non-EU regions are estimated to follow the overall agricultural emissions reported in FAOSTAT GHG inventories as closely as possible over time. The estimation, carried out for each non-EU region and emission category individually, is based on time series data of regional GHG inventories and production of agricultural commodities. Data on production quantities come from the CAPRI database, and the GHG inventories come from FAOSTAT (FAO 2010–2018). In most cases the data cover the period 1990–2009, while in some cases fewer years are available. In many cases, we have many commodities compared to the number of years of GHG inventory and production data, and thus the degrees of freedom might end up being small or even negative. In order to improve the robustness of the estimates, we include prior distributions for the emission intensities in a Bayesian estimation framework (e.g. Koop 2003, p. 15). To capture the possible change in emission intensities over time, the estimations also contain a trend component.

Bayesian prior distributions for the EI are derived from various sources, such as the expert estimates in Leip et al. (2010). Additionally, we construct priors for many commodities and emission categories with data on activity levels and production levels from the 2014 version of the AGLINK-COSIMO model (OECD 2015). Emissions per activity are computed following the Tier 1 methodology in the IPCC Guidelines (IPCC 1997; IPCC 2006), and then converted to emissions per product. Also, average EU emission coefficients computed in the CAPRI model are used as priors when the previous sources are not available.

Decomposition of Emission Changes

Emissions leakage is influenced by changes in the level of production, but also by its reallocation to regions with different emission intensities. When production is reallocated to regions with higher emission intensities, the total emissions will increase for a given level of production and vice versa. In order to disentangle the impacts of production changes and changes in average EIs, we made an additional computation of emission changes: First, we set all the EIs equal to the global average in the reference scenario for all countries, and thereafter we calculated the emissions using the production changes in the policy scenario. This computation captures only the effect of changing global production levels. Those calculated changes (i.e., changes due to changed production levels) were subtracted from the global changes in GHG emissions computed using regionally specific emission factors, giving the emissions changes caused by reallocation of production to regions with different EIs as a residual.

Baseline for Agriculture and Policy in the EU

The CAPRI baseline projects agricultural production and emissions to the year 2030 under a business-as-usual scenario. Trends for factors exogenous to the model such as population growth and consumer preferences are set based on external projections. The development of agriculture in the EU is based on the Agricultural Outlook published by the European Commission. The CAP is assumed to be fully implemented up to 2021 and then unchanged.

Within the CAP, the largest part is Pillar I measures, which mainly involve support and some market intervention schemes. Pillar II covers support to certain agricultural production, environmental measures, and rural development. Within Pillar I, most support (75%) consists of direct payments to farmers on a per-hectare basis for all qualifying agricultural land. The largest proportion of these payments is the Basic Payment Scheme (or the Single Area Payment Scheme in some regions), with support allocated to all agricultural land with entitlements. This support is considered to be decoupled from production, and member states are obliged to harmonize per-hectare rates across regions (European Union 2013). The greening payment is another large part, and it comes with associated constraints on crop diversification, grassland maintenance, and keeping ecological focus areas. A smaller part of Pillar I is dedicated to payments to young farmers and smaller farms, and areas with natural constraints. In addition, there is crop-specific coupled support for cotton in some countries, and complementary National Direct Payments in some countries.

VCS, the focus of the present study, permits MS to use up to 13%^vi of the Pillar I payments for coupled support to sectors undergoing economic, social, or environmental difficulties in maintaining/increasing production (European Commission 2017b). The measure is used by most MS and mainly targets cattle^vii and other ruminants^viii (European Commission 2019). In total, we modeled 278 different VCS measures across the EU member states. Thirteen percent of the total budget might be considered a minor share of the budget, but it has a potentially strong impact on emissions: Most VCS, about 43% of the total in our data set, is linked to the production of beef and veal, and another 12% to sheep and goat production. These sectors together cause large emissions of the GHG methane and N₂O, either directly or via fertilizer used for producing fodder. The dairy sector also receives much VCS, about 20% of total VCS payments, but in dairy it generally constitutes a smaller proportion of the revenues than in beef production.^ix Among the crop sectors, the production of protein receives notable amounts (8.5% of the total) of VCS in many member states, followed by fruit and vegetables (at 5%), but these sectors are less interesting from a GHG emissions perspective.

Simulated Scenarios

In the reference scenario, the current CAP was assumed to continue until 2030, thus including VCS as described above.

The policy scenario was identical to the reference scenario, except that VCS for ruminants was removed. In the CAPRI model, these subsidies are implemented as a direct subsidy per head, with budgetary ceilings as reported by EU countries. The budget that was released when VCS was removed was allocated to the other farm payments (the Basic Payment Scheme) in each MS, so that the total budget for farm payments in each MS remained unchanged in the reference and policy scenarios. The redistribution of support in the policy scenario resulted in an average increase in per-hectare payments for agricultural land of 6.5% in the EU, while support linked to beef cattle decreased by 69% per head, support for dairy cows by 41% per head and for sheep and goats by 36% per head. The remaining coupled support consisted of payments that are not part of VCS: national payments such as Nordic Aid and environmental and rural development support. The impact of a policy change in 2030 was derived by comparing the two scenarios.

Sensitivity Analyses

We do not know the covariance of the uncertain parameters across regions and products. In order to avoid running a very large number of simulation experiments, we chose to vary the parameters for all products and regions in concert by setting all parameters of the same type to lo/ML/hi simultaneously. For instance, we set the demand elasticities of products in all countries simultaneously to hi, ML or lo, giving just 3 demand settings instead of thousands, and similar for the other parameters in the sensitivity analysis. We thus obtain 3 × 3 × 3 × 3 = 81 result sets; this should span the extremes of the result space.

Global Changes in Emissions of GHG from Agriculture

When VCS for ruminants was removed, emissions in the EU, but also outside the EU, were affected. Figure 1 shows differences in agricultural GHG emissions in thousand ton (kt) CO₂-eq. between the policy scenario and the reference scenario (i.e., the simulated impacts of removing VCS) for 2030. The asterisks (*) in the top panel show the results with standard (ML) parameter settings. With the policy change, the GHG emissions in the EU decreased by 2,354 kt. However, there was an emissions leakage effect, as emissions in the rest of the world increased by 1,738 kt. This resulted in a net decrease on a global basis of 616 kt, or approximately 26% of the emissions decrease in the EU.

The boxes in figure 1 indicate sensitivity with respect to the four groups of parameters: supply elasticities (SupElas), demand elasticities (DemElas), import substitution elasticities (CES), and emission factors of non-EU regions (EF). The sensitivity analyses in Panel A show that the emissions in the major regions analyzed (EU, non-EU, World) depend strongly on parameters of the model, so that our results on global emissions change could be larger or smaller, with the extreme outcomes for the “World” region ranging from −2,956 kt to +1,465 kt. There are more outcomes in the lower range than in the higher range, as indicated by the median line being to the left of the asterisk.

The results seem about equally sensitive to variations in the four parameters, yet the disaggregation in Panel B allows some general conclusions. Each box in Panel B shows the variation of global emissions (i.e. the shaded box in Panel A) if each group of parameters in turn is fixed at one of the three levels: If the emission factors of the Non-EU regions are at EF.lo (20–60% lower than standard), or demand elasticities are at DemElas.hi (50% higher than standard), or the import substitution parameters are at CES.lo (50% less than standard), the global emissions change is negative, regardless of how the other parameters are set within the ranges analyzed. The bottom three boxes, showing dependence on supply elasticities within the EU, illustrate how these parameters merely scale the total results, and thus are of importance to the absolute size of the impact, but not to the qualitative results.

Studying the main results in more detail, we find that about 90% of the emissions reduction in the EU derived from production of beef, with an absolute decrease in emissions of 2,088 kt CO₂-eq (Table 1). This was a result of less production, as production in relative terms decreased by 1.1% (see Table 2). As can be seen in Table 1, milk was the largest source of emissions in the EU, but the change in emissions for milk—where VCS is less important—was much smaller than for beef. Emissions from pork and poultry increase due to consumers replacing some of the more expensive beef with relatively less expensive pork or poultry. Since emission intensities for poultry and pork are significantly lower than for beef, the emission increase associated with pork and poultry production was small. For crop products, emissions barely changed. The slight increases in emissions associated with arable crops were caused by a larger crop area combined with lower average yields. Feed demand went down and exports increased, leading to the small net increases in emissions in the EU shown in Table 1.

Different products have different sensitivities to emissions leakage. For beef, much of the reduction in the EU was canceled out by increased emissions outside the EU. For sheep and goat meat, there was even an increase in emissions globally, despite the 75kt CO₂-eq. reduction in the EU in the policy scenario. In contrast, the reduction in emissions from milk production in the EU was accompanied by an additional small emissions reduction outside of the EU, caused mostly by a reallocation of production among world regions. For crops, increased exports from the EU replaced production abroad, leading to reduced emissions there and a small net reduction associated with crops globally.

Beef markets merit extra attention, because beef meat was the largest contributor to the change in GHG emissions following the removal of VCS. Table 2 shows changes in the EU beef market. In the policy scenario, beef production in the EU decreased, leading to higher producer and consumer prices for beef meat in the EU. The higher prices dampened the negative impact on production. Production decreased by 89kt, while consumption was rather inelastic and decreased by only 50kt. The balance between decreased production and consumption of beef was maintained by a reduction in exports (−22kt) from the EU, and by increased imports to the EU (+17 kt). This caused production changes in countries outside the EU, driving the results on emissions leakage.

Table 3 shows impacts on production in and trade with the non-EU regions of CAPRI that are most strongly affected. Imports of beef to the EU increased most from the US, while exports from the EU decreased, in particular for Mediterranean countries and Russia. The latter was met by a production increase in Russia. Argentina and Brazil remained the main trading partners, but their exports to the EU did not change greatly. Instead, changing world market prices affected their trade with other parts of the world, resulting in large production increases in Brazil. Other regions outside the EU also changed their production and trade relations. India's production and exports increased slightly, which had a large effect on global emissions, since Indian production is relatively emissions intensive.

Within the EU, the largest decreases in GHG emissions in absolute terms were found in France, Spain, and Poland. In contrast, in the United Kingdom, Ireland, and Germany, where ruminant production receives little or no VCS, GHG emissions increased, since in these regions ruminant herds slightly increased in response to the higher prices. Table 4 shows the changes in the suckler cow^xi herd, beef production, methane, and total non-CO₂ GHG emissions in the EU countries with the largest absolute changes in the latter. The increase in emissions in the countries with expanding ruminant sectors might be considered a case of “intracommunity leakage” where the GHG-saving effects in the majority of countries are counteracted by emission increases in others.

In our computations, removing VCS to ruminants increases agricultural incomes in the EU by about €1,400 million annually.^xii The income increase is due to two things: Firstly, the VCS funds are transferred to the basic farm payment, where it tops up income without requiring additional variable costs, i.e., animals that were unprofitable without subsidies are no longer produced, while the subsidy is still obtained. Secondly, the prices of some animal products rise and thus raise farm incomes. There is reduction in consumer welfare of €868 million annually due to the higher prices. The impact on tax payers is negligible, €40+ million spending, since the total CAP budget is unchanged. Thus, we might expect a gain in welfare in the EU across these three groups totaling €492 million euro annually.

Decomposition of Emissions Leakage

The results show that abolishing VCS to ruminants would reduce global agricultural GHG emissions due to the reallocation of production. To gain insights into this process, we decomposed the changes in emissions. The obvious reason for increases in emissions outside the EU is increased production of beef in countries outside the EU. Another reason is that production is more or less intense in terms of GHG emissions in different locations, which means that reallocation of production has an impact on emissions. In addition, changing conditions may alter production technology, which could affect the emission intensity of a product. In our simulations, these technological changes were only modeled endogenously for EU+ countries.

The disaggregation of emissions changes for beef resulting from production volume and reallocation effects are presented in figure 2. The bar to the left shows the emissions changes that would have occurred if the average emission intensity in the world (from the reference scenario) applied to all regions, while the production changes remained the same. This emissions change can be attributed to the change in global production volume. The reduction in production would thus have reduced global emissions by 1,666 kt CO₂-eq. However, the actual emissions reduction globally was 616 kt CO₂-eq., which is 1,050 kt less than the emissions reduction brought about by production level changes. This discrepancy is explained by the reallocation of production to locations with higher emission intensity than the EU.