Regional differences in the capitalisation of first and second pillar payments of the CAP into land rental prices


 Nearly 80 per cent of Common Agricultural Policy (CAP) expenditures are spent on three different measures: first pillar payments (FPPs), agri-environmental payments (AEPs) and less favoured area payments (LFAPs). Based on a dynamic panel model and farm accounting data for Bavaria, we find that, on average, 30 per cent of FPPs, 40–50 per cent of LFAPs, but no relevant share of AEPs are capitalised into land rental prices. The capitalisation ratio varies considerably across regions. Above average capitalisation ratios for FPPs are observed in more favourable areas with high yields, a low grassland share and large farms. The same is true for LFAPs for areas with high yields, large farms and a greater share of part-time farmers.


Introduction
Throughout its history, the Common Agricultural Policy (CAP) of the European Union (EU) has devoted a considerable budget to support farmers. The European Agricultural Guarantee Fund, or 'first pillar' of the CAP, accounts for approximately 76 per cent of total CAP expenditures and provides mainly direct (income) support to farmers. The European Agricultural Fund for Rural Development (EAFRD), or 'second pillar', accounts for the remaining 24 per cent of CAP expenditures and comprises specific aid programmes for rural 10 K. Salhofer and P. Feichtinger (2018) and Henning and Breustedt (2018). 1 Interestingly, estimated capitalisation ratios vary considerably, ranging from EUR 0 to almost EUR 1 per each additional euro paid. Most of these studies either investigate the situation in one particular country or region (Kilian et al., 2012;O'Neill and Hanrahan, 2016;Klaiber, Salhofer and Thompson, 2017;Guastella et al., 2018;Henning and Breustedt 2018) or provide an aggregated capitalisation ratio for several countries (Ciaian and Kancs, 2012) or for the entire EU (Guastella et al., 2014). A study by Michalek, Ciaian and Kancs (2014) is the only study that provides individual capitalisation ratios for several countries (all EU-15 Member States) based on the same data basis (FADN) and empirical model. They find considerable differences in the capitalisation ratio between countries, ranging from 0.04 (Greece) to 0.18 (Portugal), i.e. rental prices will increase by 4 (18) cents per hectare for each additional euro of government payments per hectare. So far, no study exists which investigates if and to what extent the capitalisation ratio can vary within a particular country, i.e. across regions which are homogenous in terms of the way in which FPPs are implemented and which share the same agri-environmental programs and less favoured area policies.
There is some empirical evidence that the degree of capitalisation of FPPs crucially depends on how the policy is implemented. Michalek, Ciaian and Kancs (2014) find that the capitalisation ratio is higher for the hybrid implementation model of the Fischler Reform of the CAP in 2003 compared to the historical model, as well as in countries which opted for full decoupling of FPPs rather than partial decoupling. Klaiber, Salhofer and Thompson (2017) provide evidence that capitalisation is stronger for the regional implementation model compared to the hybrid model. Apart from this, not much empirical evidence exists on what drives the degree of capitalisation. Michalek, Ciaian and Kancs (2014) find that the capitalisation ratio is positively correlated with farm size. However, their results are based on all EU-15 Member States. Given that these countries differ in the way they have implemented the Single Payment Scheme (SPS) and in their average farm size, these two effects on the capitalisation ratio are difficult to separate. Van Herck et al. (2014) pooled country-level data for six NMS. While they do not provide countryspecific capitalisation ratios for the six NMS, they do use some interaction dummy variables with FPPs to investigate the impact of some variables on the capitalisation ratio. They find that the capitalisation ratio is higher in more credit-constraint countries, as direct payments to farmers may reduce these credit market constraints and thus increase the demand for rented land (Van Herck et al., 2014, p. 429). Moreover, they find that the capitalisation ratio is lower in countries where a considerable share of agricultural land is used by (large) corporate farms that have more bargaining power in land rental markets. O'Neil and Hanrahan (2016) show that the capitalisation ratio varies for different farm types (dairy, cattle, sheep and tillage) without investigating the determinants behind this observation. Moreover, empirical evidence is sparse for the capitalisation ratio of second pillar payments (Patton et al., 2008;Kilian et al., 2012;Klaiber, Salhofer and Thompson, 2017;O'Neil and Hanrahan, 2016).
Given this, the aim of this paper is twofold. First, we provide empirical evidence for the capitalisation ratio of the three most important payment categories: FFPs, agri-environmental payments (AEPs) and less favoured area payments (LFAPs). We do this by examining a comprehensive dataset of more than 3,000 Bavarian farms from 2005 to 2011. Second, we explore regional differences in the capitalisation ratio, as well as what drives these differences. We do this by deriving capitalisation ratios for the total sample and for subsamples that are defined by regional differences in regard to natural conditions (yields and grassland share), farm structure (farm size, share of part-time farms and animal stocking rates) and the land market (rental share, share of agricultural area in total area and population density). Given that Bavaria is homogenous in terms of its institutional settings (e.g. customs and laws regarding land transactions) and the way the CAP is implemented, observed differences can be attributed to the factors described above.
We proceed with our analysis as follows. The next section presents background information on the implemented policies, land market regulations in Germany and Bavaria and regional differences. Moreover, it provides a short review of theoretical and empirical findings in the literature on the capitalisation of CAP payments. The third section introduces our empirical model specification, data and model estimations. The fourth section presents results. The last two sections contain the discussion and conclusions.

Background and relevant literature 2.1. CAP payments
Between 1992 and 2003, the EU converted the CAP through two reforms (the MacSharry Reform in 1992 and the AGENDA 2000 Reform) from an intervention price system, whereby farmers received a price above the world market price, to a coupled payments system, based largely on the number of hectares farmed and number of animals kept. As part of the Fischler Reform in 2003, the SPS was enacted and came into effect in 2005. As a starting point, all active farmers received SPS entitlements equal to the average number of hectares they farmed during the reference period between 2000 and 2002. To receive payments from 2005 onwards, farmers are no longer obliged to plant anything on their land or keep animals. Instead, they only have to keep their land in 'good agricultural and environmental condition' (European Union, 2013). However, to receive payments, they need entitlements and an equal number of hectares of owned or rented 'eligible area'. Therefore, payments made under the SPS are clearly decoupled from production decisions (i.e. decisions regarding if and what to plant, as well as the amount of operating inputs, such as fertiliser and pesticides), but since 1 ha of eligible area is required for each entitlement to be activated, payments are not decoupled from land.
In the 2013 Reform, the Basic Payment Scheme (BPS) replaced the SPS. The BPS offers a basic layer of income support to farmers, to be topped-up by other direct payments targeting specific issues or specific types of beneficiaries (European Commission, 2013). The most important top-up is 'green direct payments', which cover 30 per cent of the national direct payment envelope. Green direct payments are given as compensation for certain obligations, such as the maintenance of permanent grassland, creation of ecological focus areas and diversification of crops. Other top-ups target young famers, small farms and areas with natural constraints. However, nothing changed with regard to how farmers can claim these payments. Farmers receive a certain number of entitlements and need to own or rent an equal number of eligible hectares to activate these entitlements. They can receive green direct payments (or any other redistributive payment if it is implemented in a particular country) as a top-up on each activated entitlement. Therefore, all direct payments are linked to land. Green direct payments may capitalise to a lesser extent into land rental values than basic payments or single farm payments since they are linked to additional efforts of the farmer related to the environment. However, according to Louhichi et al. (2018), given the way that these greening measures are defined and implemented in the Member States, less than 30 per cent of farms (50 per cent of the UAA) in the EU failed to comply with the greening measures at the onset and had to change their land use to some extent.
In 2004, 12 countries joined the EU. With the exception of Slovenia and Malta, all of these countries opted to implement the Single Area Payment System (SAPS), instead of the SPS. The SAPS is a simpler version of the SPS without entitlements. Instead, under the SAPS, all farms in a Member State (or region within a Member State) receive an equal payment per hectare (i.e. a flat rate) for each eligible hectare of land. In the 2013 Reform, all Member States that applied the SAPS before 2013 were allowed to continue to do so, and all 10 Member States decided to maintain SAPS instead of introducing the BPS (European Commission, 2016).
The second pillar of the CAP is comprised of a multitude of different schemes in which Member States can choose from a 'menu' of measures detailed in the Rural Development Regulation (Regulation (EU) No 1305/2013). Second pillar schemes are 'designed to support rural areas of the Union and meet the wide range of economic, environmental and societal challenges of the 21st century' (European Parliament, 2020b). These schemes are co-financed by the EU and Member States. In most Member States, the largest shares of the second pillar budget are used for three measures: (i) investment aid (23 per cent of total second pillar expenditures); (ii) agri-environmental schemes (17 per cent); and (iii) payments for areas subject to natural constraints or other specific constraints (16 per cent) (European Parliament, 2020a). Here, we investigate the capitalisation ratio of the latter two, as investment aid takes the form of erratic payments with no clear connection to land rental prices. In Bavaria, from 2007 to 2013, AEPs and LFAPs accounted for approximately 23 per cent and 19 per cent, respectively, of total second pillar expenditures, namely EUR 4.2 billion (an average of EUR 600 million annually) (ART, 2016).

Agricultural land sales and rental market regulations in Germany and Bavaria
In Germany, agricultural land sales transactions are regulated by the 'Law on measures to improve the structure of the agricultural sector and the security of farming and forestry holdings' ('Gesetz über Maßnahmen zur Verbesserung der Agrarstruktur und zur Sicherung land-und forstwirtschaftlicher Betriebe'; in short: 'Grundstückverkehrsgesetz'), which came into force in 1961 and has only undergone a few minor changes since then (AEIAR, 2016). Sales of agricultural land must be approved by administrative authorities in the federal states ('Länder'). The only exemptions are plots under a certain size, which varies by federal state and is 2 ha in Bavaria (Ciaian et al., 2012b). Approval can be refused by the authority if: (i) the sale will lead to poor land distribution, especially if the buyer is not a farmer; (ii) the agricultural structure worsens due to a negative reduction or subdivision of land; or (iii) the sale price is disproportionate compared with that of an equivalent piece of land. However, agricultural land sales markets are historically very thin in Germany in general and in Bavaria in particular (Feichtinger and Salhofer, 2016). In the last 25 years in Germany (Bavaria), the percentage of sold agricultural area in total UAA per year has been about 0.60 per cent (0.20 per cent), on average, and has never exceeded 0.75 per cent (0.26 per cent) (own calculations based on Destatist 2020). This means that the average plot of agricultural land is sold every 167 (500) years. Therefore, given the ongoing structural change in the sector, which causes farms to steadily increase in size, land rental markets are of crucial importance. In Bavaria, between 1991 and 2016, the size of the average farm more than doubled from 15.7 to 35.9 ha, while rental shares increased from about 31 per cent to 49.5 per cent (StMELF, 2019; Destatis, 2020). The legal framework for agricultural land rentals in Germany is the Federal 'Law on land lease' ('Landpachtverkehrsgesetz') from 1985 (AEIAR, 2016). Every transaction on the rental market must be reported to the same authority, which also approves agricultural land sales transactions. Reasons for refusing to approve a rental transaction are the same as those for land sales. On the other hand, the legal framework for agricultural land rentals is comparably liberal. There are neither rental price restrictions nor requirements for the minimum/maximum duration of a rental contract. This differs considerably from some other EU countries, such as Belgium and France (Ciaian et al., 2012a). Rental agreements in Germany are almost exclusively in cash and not in-kind (Theuvsen, 2007). The usual form of a rental contract is written with a fixed price and limited duration. Nevertheless, there are still oral contracts and contracts with an unlimited duration. Salhofer et al. (2009, pp. 39-40) report that two-thirds of rental contracts are issued for a limited duration, whereas a medium-term duration (about 6 years) is most common in Southern Germany.
If a contract reaches maturity without recent arrangements, it is converted automatically to a contract with an unlimited duration, which can be cancelled without a specific reason each year either by the tenant or the landowner with a notice period of 6 months. The duration of rental contracts for arable land and grassland are the same (Ciaian et al., 2012a).
Swinnen, Van Herck and Vranken (2014) calculated a 'Total Land Regulation Index' (LRI) for 24 EU Member States. The LRI is comprised of different aspects of land market regulations to provide a measure of the degree of land market regulation for each EU Member State. The most regulated land markets are in France (LRI = 9) and Hungary (LRI = 8), while Germany (LRI = 1.5) belongs to the group of countries with relatively liberal land rental and sales markets.

Regional differences in land market determinants
In general, one can observe considerable regional variations within Bavaria regarding natural production conditions (e.g. yields and grassland share), farm organisation (e.g. farm size, share of part-time farms and stocking rates) and land rental market characteristics (e.g. rental shares, population density and share of agricultural area in total area). All of these characteristics are clearly interrelated to some extent as illustrated by the correlation matrix in the appendix (Table A1) and will have an impact on the supply of and demand for rented land, rental prices and subsequently the extent to which payments are handed over to landowners. Most likely, factors increasing competition between farmers for renting land, which is the scarcest resource in agricultural production, and factors increasing the bargaining power of landowners will increase rental prices and the capitalisation ratio.
To identify regions with different characteristics, we use the Nomenclature of Territorial Units for Statistics (NUTS). Bavaria is a NUTS-1 region, subdivided into 7 NUTS-2 regions ('Bezirke'), 96 NUTS-3 regions which are equivalent to districts ('Landkreise und kreisfreie Städte') and 2,056 local administration units (LAUs) which are equivalent to municipalities ('Gemeinden'). Here, we concentrate on NUTS-3 regions, as our farm-level data does not include information about the LAU where a farm is located. Table 1 depicts the heterogeneity of agriculture in Bavaria, based on average values for the 96 NUTS-3 regions.
First, we use a relative yield index and the share of grassland in total UAA to account for differences in natural production conditions, particularly land quality. Assuming that high-quality land is only available in a limited amount, we can hypothesise that competition for land and the capitalisation ratio will increase with land quality. We measure relative yields as an index of the unweighted averages of the relative yields of the two most important crops (wheat and silage maize). Relative yields are the yields of each NUTS-3 region that are normalised by the mean value of all 96 NUTS-3 regions. Table 1 shows that relative yields for the 96 NUTS-3 regions vary between 80 per cent and 116 per cent. Most productive NUTS-3 regions are in the middle of Bavaria  and to some extent in the northeast part of Bavaria. Since grassland is predominantly found on plots with lower land quality, we assume that the capitalisation ratio decreases with the share of grassland in a NUTS-3 region. Grassland shares vary between 5 per cent and nearly 100 per cent. Although grassland shares are high in the Alpine regions in southern Bavaria, there is not a clear north-south divide. There are also regions with high grassland shares in eastern and northern Bavaria, while the areas with the lowest grassland shares are located in the middle and in some regions in the northeast (Dorfner and Zenger, 2017). Second, we describe farm organisation by farm size, animal stocking rates and the share of part-time farms. The average farm size over the 96 NUTS-3 regions also varies considerably, from about 9-64 ha. Although there is not a clear north-south divide in terms of farm size, the largest average farm sizes are observed in the NUTS-3 regions of northern Bavaria, while farms are much smaller in the Alpine regions in southern and especially in southwestern Bavaria (StMELF, 2019). On the one hand, it could be argued that large farms are more business oriented, more profitable and more obliged to rent land. This would increase competition for land in areas with larger farms, On the other hand, as the average size of farms increases, this should decrease the number of competitors a farm has for a specific plot and decrease landowners' bargaining power. Stocking rates vary between 0.1 and 1.9 animal units per hectare and are highest in the milk-producing regions in southeastern and southwestern Bavaria. A high stocking rate increases farmers' demand for land to produce feedstock and apply manure. Hence, we would expect increasing capitalisation ratios with increasing stocking rates. Furthermore, we consider the share of part-time farmers. In Bavaria, this share is lowest in the southwest and south and largest in the northeast, varying between 23 per cent and 72 per cent (StMELF, 2019). We hypothesise that part-time farmers have less need to increase their farm size, as agriculture provides only part of their family income. Therefore, their pressure to rent land is lower, which should lead to a lower capitalisation ratio.
Third, we use rental shares, population density and the share of agricultural area in total land area as land rental market characteristics. With regard to rental shares, we observe a clear north-south divide, with rental shares ranging from slightly above 20 per cent in some NUTS-3 regions in the southwest to more than 81 per cent in the northeast of Bavaria (LfL, 2013). Rental shares might be an indication of the size of the rental market. Therefore, higher rental shares may lead to less competition for a particular plot between farmers and lower bargaining power of landowners. However, higher rental shares may also indicate a less liquid land sales market and a stronger dependency on rented land. Therefore, the effect on the capitalisation ratio remains ambiguous. Population density is a proxy for pressure on land markets from alternative uses of land, in particular housing (Capozza and Helsley, 1989). While plenty empirical evidence exists that urban pressure influences land sales prices (e.g. Cavailhés and Wavresky, 2003;Delcbecq et al., 2014;Feichtinger and Salhofer, 2016), the effect on rental prices is not so clear. Finally, a higher share of agricultural area in total land, ranging from 18 per cent to 73 per cent in Bavaria, means that more agricultural land is available, putting less pressure on land rental markets. However, very high shares of agricultural areas can be found in the productive areas in the middle of Bavaria, while these shares are low in the less productive areas in the south, as well as some areas in the east and the northwest (StMELF 2019). This is related to the fact that the share of forest area is higher in these less favoured areas.

Literature review
The effects of FPPs on land rental prices have recently been examined by various authors based on either regionally aggregated data or farm-level accounting data (Table 2). 2 In terms of the former, Kilian et al. (2012) use cross-sectional data at the municipality level for Bavaria and find a relatively high capitalisation ratio between 0.44 and 0.98. A similar strong capitalisation is revealed by Henning and Breustedt (2018) Roberts, Kirwan and Hopkins, 2003;Lence and Mishra, 2003;Kirwan, 2009;Qiu, Gervais and Goodwin, 2010;Goodwin, Mishra and Ortalo-Magné, 2011;Hendricks, Janzen and Dhuyvetter, 2012). Moreover, there is also literature addressing the effects of agricultural policies on land sales prices. Latruffe and Le Mouël (2009) and Feichtinger and Salhofer (2013) provide comprehensive reviews of this issue.  Patton et al. (2008), Kilian et al. (2012) and Klaiber, Salhofer and Thompson (2017). Estimates range from about 0.20 (Kilian et al., 2012) to full capitalisation (Patton et al., 2008). Only Kilian et al. (2012) and Klaiber, Salhofer and Thompson (2017) consider AEPs of the CAP, and neither find the capitalisation of these payments into land rental prices to be statistically significantly different from zero.

Model specification
In estimating a rental price equation, five main econometric challenges have been identified in the literature: unobserved firm heterogeneity, sample selection bias, spatial interdependency, endogeneity (expectation error) and the dynamic nature of the problem. The previous literature has approached these problems to a different extent, depending on the focus of the study and availability of data. No study exists which accounts for all of these five challenges. For instance, unobserved firm heterogeneity is taken into account in different ways by nearly all studies with panel data (see Table 2). The sample selection problem caused by using only those farms in the sample which actually rent land is tackled by Ciaian and Kancs (2012), Michalek, Ciaian and Kancs (2014), Guastella et al. (2018) and Klaiber, Salhofer and Thompson (2017). Spatial independency of rental prices and/or spatial correlation is considered by Breustedt and Habermann (2011) and Guastella et al. (2014). Moreover, endogeneity issues are taken into account by Ciaian and Kancs (2012), O'Neill and Hanrahan (2016) and Guastella et al. (2018), while so far only O'Neill and Hanrahan (2016) have modelled the dynamic nature of rental prices. Since our data neither includes the exact coordinates nor municipality of farms, we cannot account for spatial interdependency. Moreover, since Klaiber, Salhofer and Thompson (2017), who use a different sample of the same dataset for Bavaria, do not find a sample selection bias in their land rental price analyses, we do not account for a possible selection bias. Therefore, our study accounts for unobserved firm heterogeneity, the dynamic nature of rental prices and endogeneity (the expectation error).
We model rental prices (r) determined from expected net returns from the market and different expected government payments. It can be argued that changes in returns and government payments do not instantaneously change the rental price r because of multiple-year rental contracts and/or other transaction costs that cause inertia (Hendricks, Janzen and Dhuyvetter, 2012). This can be modelled in a simple way by Nerlove's (1958) partial adjustment model: where r * t is the equilibrium rental price and is the adjustment coefficient. Here, ρ = 1 represents a full (instantaneous) adjustment, whereas 0 < ρ < 1 implies a partial adjustment. With this in mind, we define a dynamic panel data model as: where r it is the average per hectare rental price farm i pays during time period t, r it−1 is the rental price paid in the previous time period t−1, m it denotes expected net returns from selling their products, s lit denotes expected per hectare averages of L different government payments (e.g. FPPs, AEPs and LFAPs) a farm receives, x eit denotes E covariates that control for observed farm-specific conditions, d t denotes T−1 time dummy variables absorbing year specific shocks (e.g. weather, interest rates, etc.) that affect all farm operations and u it is an error term. The two components of this error term, the unobserved time-invariant heterogeneity v i ∼ IID (0, δ 2 υ ) and the idiosyncratic portion ε it ∼ IID (0, δ 2 ε ) are assumed to be independent of one another and among themselves (Baltagi, 2013). θ as well as all γs, λs and µs are coefficients to be estimated. Whereas these coefficients capture short-run effects, long-run effects are estimated by dividing the respective coefficient by the estimate of the adjustment coefficient: . For example, if the estimated short-run capitalisation coefficient of FPPs isλ FPP the long-run coefficient is given byλ L FPP =λ FPP (1−ρ) . The inclusion of the lagged rental price (r it−1 ) in combination with unobserved heterogeneity (v i ) introduces two problems. The first problem is autocorrelation since r it−1 is correlated with ε it−1 , ε it−2 , etc. The second problem is the 'dynamic panel bias' caused by endogeneity since r it−1 is correlated with fixed effects in the error term (Nickell, 1981). Therefore, the OLS estimator is inconsistent and upward biased (Bond, 2002), whereas the fixed-effects estimator is typically downward biased (Nickell, 1981). These problems can be solved by transforming the variables into first differences (Anderson and Hsiao, 1982;Holtz-Eakin, Newey and Rosen, 1988;Arellano and Bond, 1991) 3 : (2) where, for example, ∆r it = r it − r it−1 . The lagged dependent variable and error term remain correlated. As suggested by Anderson and Hsiao (1982), we can use r it−2 as instruments for ∆r it−1 . While r it−2 is the only instrument available for period t = 3, the number of available instruments increases with increasing t, with r it−2 , . . . , r iT−2 as instruments for period T.
Apart from those discussed above, other potential sources of endogeneity remain. In particular, Kirwan (2009), Breustedt and Habermann (2011), Hendricks, Janzen and Dhuyvetter (2012 and others stress the importance of expectation errors when conducting land rental price analyses. Essentially, as one form of measurement error, an expectation error arises when rental prices are negotiated before the growing season. Tenants must form some expectations about future market returns and government payments. As expectations are not observable, the researcher must use actual values in their estimations. When actual values differ from expected values, biased coefficient estimates result. As the value of CAP payments was precisely given and known upfront to farmers, we do not anticipate an expectation error in this regard. However, the result may differ for market returns. Therefore, we use the same dynamic instrument procedure applied for ∆r it−1 in the case of expected market returns ∆m it .

Data
Our empirical model is applied to a comprehensive dataset of bookkeeping records of Bavarian farms. The sample is stratified with respect to legal form, farm type (agriculture, viniculture, horticulture and forestry), farm size and geographic region. However, very small farms and part-time farms are underrepresented. The reporting period is the financial year, which begins on 1st July and ends on 30th June of the following year. We refer to the financial To control for data problems and outliers, we exclude observations indicating specific farming or rental situations and implausible values. In particular, we exclude farms specialised in viniculture, horticulture and forestry (2.8 per cent of initial observations) and those with negative rental prices and average rental prices of more than EUR 3,000/ha (2.2 per cent and 1.9 per cent). We do 3 An alternative transformation is forward orthogonal deviations,r it = r it − 1 T−t ∑ T t+1 r it (Arellano 1988;Arellano and Bover, 1995). We focus here on first-differences, but find that the results for forward orthogonal deviations are not very different. not consider smallholder farms with less than 5 ha of agricultural land (0.7 per cent) or farms renting less than 1 ha (8.2 per cent). To avoid gaps in our panel dataset, we exclude any farm if one criterion listed above is met for any year from 2000 to 2011. For these exclusion criteria, we also include the years prior to 2005 since we use rental prices as instruments in the estimations. We also exclude farms with revenues exceeding EUR 12,000/ha (3.4 per cent), FPPs above EUR 1,000/ha (0.7 per cent), negative AEPs (1.4 per cent), AEPs above EUR 2,500/ha (0.1 per cent) and LFAPs above EUR 250 (0.7 per cent). Finally, we exclude farms for which our data report implausible values for our covariates, i.e. farms for which the ratio of corn, wheat or rapeseed area to total crop area was greater than 1 (0.5 per cent). All these criteria have to be met for any year from 2005 to 2011, which is the period under investigation. Overall, we exclude 2,878 observations, which is 15.8 per cent of the initial dataset. 4 Our final dataset is an unbalanced panel of 15,343 observations for 3,036 farms. The average farm reports data for 5.1 out of 6 years. Table 3 presents descriptive statistics for 2006-2011. Here, we describe how the dependent variable is constructed. Our dataset includes information on farmed and owned land for each farm. We first subtract owned land from farmed hectares to obtain a measure of net rented land. We then divide the total expenditures of a farm dedicated to renting land by net rented hectares. The average rental price is EUR 251/ha. As a proxy for market returns, we use market revenues, which are an average of EUR 3,003/ha. 5 Farms considered in our sample received an average of EUR 357/ha in FPPs, EUR 60/ha in AEPs and EUR 38/ha in LFAPs.
4 Please be aware that the sum of excluded observations for each criterion is larger than the total amount of excluded variables, as some observations (farms) fail to fulfill more than one criterion. 5 Market revenues are only an appropriate proxy for market returns, defined as market revenues minus costs of all production factors besides rents, as long as the ratio between returns and costs is the same across all revenue levels. However, calculating the market returns in the case of agriculture is not straight forward, as labor and capital are owned by farmers to a different extent which may cause biases in cost estimates.
We consider two types of variables to account for farm heterogeneity. Ratios of cash crops (corn, wheat and rapeseed) are used to describe the specific natural production conditions of the farm. We expect a positive relationship between cash crop area and rental prices. In addition, we use the UAA and its squared value to account for different farm sizes. The average number of hectares farmed in our sample is 59. One may assume that large farms are able to pay higher rental prices because of economies of scale. However, as a farm gets larger, transportation costs increase. Therefore, we expect UAA to be positive and its squared term to be negative. Finally, we have no definite expectations regarding the sign of the average plot size of rented land. On the one hand, one may hypothesise that larger plots are more cost-efficient and thus farmers are able to pay a higher rental price. On the other hand, one may expect a discount for renting large plots.
We also test other covariates, including milk production in tons per hectare UAA and the share of revenues from animal production in total revenues, the share of family labour in total farm labour, age of the farm manager and dummy variables for part-time farming, gender of the farm manager and level of education of the farm manager. However, none of these were statistically significantly different from zero for various reasons, including low variance across farms (e.g. a very high share of family labour for most farms), near time-invariance (gender, age and level of education of the farm manager) and multicollinearity (part-time dummy and farm size).
In the estimations, we first use the full sample and then we split our sample of farms in NUTS-3 regions between farms with above median and below median values of our eight characteristics (relative yields, grassland shares, farm size, share of part-time farmers, stocking rates, rental share, population density and share of agricultural area) in Table 1.

Model estimation
The model in equation (2) is estimated by utilising the feasible efficient Generalized method of moments (GMM) estimator with Windmeijer (2005) corrections for two-step standard errors as implemented in the Xtabond2 package by Roodman (2009b). The first column of Table 4 (model 1) shows the results for the standard Arellano-Bond estimator in first differences with no limitations on lags of the instrumented variables. Given that we have data from 2000 to 2011, endogenous variables are instrumented by up to 10 lags. Validity of the estimator depends on no serial correlation of the disturbances and on exogeneity of instruments. The Arellano and Bond (1991) test fails to reject the null hypothesis of no serial correlation in the first difference regression. The Hansen (1982) test fails to reject the null hypothesis that the over-identifying restrictions are valid. However, the Hansen (1982) test grows weaker with an increased number of instruments (105 in this case). For this reason and because a large instrument collection can overfit endogenous variables (Rodman, 2009b), model 2 in Table 4 restricts the instruments to the second to seventh lagged levels and uses a collapsed instrument matrix as suggested by  Roodman (2009a). This decreases the number of instruments to 27. Model 3 utilises the same instruments as model 2, but applies forward orthogonal deviations transformation instead of first differences. This preserves the sample size in panels with gaps (Arellano and Bover, 1995). Finally, model 4 is equivalent to model 2, but utilises a balanced panel of 2,044 farms (instead of 3,036 farms in the unbalanced panel), for which complete information for at least the period from 2004 to 2011 is available.

Results for the total sample
The results are very similar for all four models. The estimate of the lagged dependent variable is between 0.052 and 0.061. This indicates relatively fast rental price adjustments to changes in market conditions or subsidies and an adjustment coefficientρ = (1 − θ) between 0.939 and 0.948. Therefore, if we take model 2 as the standard, rental prices per hectare increase in the short-run by 5.4 cents, on average, when market revenues increase by EUR 1. The longrun equilibrium effect is only slightly higher, 5.67 cents. For example, from 2005 to 2011, revenues per hectare increased by EUR 827, from EUR 2,598 to EUR 3,425. This increases rental prices by, on average, about EUR 47/ha. We find that land rental prices are influenced by the different CAP payments to varying degrees. Every additional euro given as an FPP to a farmer increases land rental prices from 27 to 31 cents in the short-run and from 29 to 32 cents in the long-run. The capitalisation ratio for LFAPs is estimated to be from 40 to 49 cents in the short-run and from 43 to 52 cents in the long-run. In contrast, the capitalisation ratio of AEPs is relatively small, ranging from 2 to 8 cents in the short-run and 2-9 cents in the long-run. Three out of four estimated coefficients are statistically not significantly different from zero. Figure 1 depicts the estimated short-run capitalisation ratios and their corresponding 95 per cent Downloaded from https://academic.oup.com/erae/article/48/1/8/5991791 by guest on 29 November 2021 Regional differences in the capitalisation of CAP payments into land rental prices 25 confidence interval for market revenues, FPPs, AEPs and LFAPs. 6 We see that the results from the different models are relatively similar and have relatively narrow confidence intervals, especially in the case of market revenues and AEPs (with the exception of model 3), but also in the case of FPPs. Results are more dispersed and more uncertain for LFAPs. We use several crop ratios to describe differences in land productivity. As expected, all of these variables have positive coefficients because these crops are typically planted in relatively productive soil in climatically and topographically advantaged regions. Apart from that, we derive a positive coefficient for farm size and a negative coefficient for farm size squared. This may reflect economies of scale (large farms are more profitable and can pay higher rents), as well as increasing transportation costs as farms increase in size. Moreover, we derive a negative effect for plot size, i.e. larger plots are less expensive to rent per hectare.

Regional analysis
We investigate the extent to which capitalisation ratios vary between NUTS-3 regions with different characteristics in terms of natural production conditions (relative yields and grassland share), farm organisation (farm size, share of part-time farms and stocking rates) and rental markets (rental share, population density and share of agricultural area). To accomplish this, we estimate model 2 using subsamples defined by the median of the eight characteristics in Table 1. For example, we put all the farms located in NUTS-3 regions with average yields below the median of 99.9 into the subsample 'relative low yields' and the rest of the farms into the subsample 'relative high yields'. We proceed in the same way for all other seven characteristics. Therefore, all subsamples consist of approximately 1,500 farms with approximately 7,500 observations. 7 We provide some descriptive statistics of these 16 subsamples in Table A2 in the appendix. We derive estimates of the capitalisation ratio for the 2 × 8 different subsamples for the three different payment categories (FPPs, AEPs and LFAPs) and compare them in Table 5, Figures 2 and 3. With regard to FPPs, we find that the capitalisation ratio of the subsample of NUTS-3 regions with low yields is 0.26, whereas the subsample of NUTS-3 regions with average yields above the median has a capitalisation ratio of 0.46. The difference in the estimated FPP capitalisation ratios is also depicted in Figure 2, where the black (grey) line in the left-hand panel shows the 95 per cent confidence interval of the capitalisation ratio of the subsample with low (high) yields and the dots depict the estimated coefficients. The dashed line depicts the point estimate for the total sample, i.e. 0.31 in Table 4, model 2.
In order to test whether the coefficients of the pairwise subsamples are different from each other, we also use a Z-Test (Paternoster et al., 1998): SEδ 1 2 − SEδ 2 2 where δ 1 and δ 2 are the coefficients to be 26 K. Salhofer and P. Feichtinger    compared and SEδ 1 2 and SEδ 1 2 are their respective variances. Under the null hypothesis that δ 1 = δ 2 the test statistic is Z standard normally distributed. For example, when splitting the sample into subsamples that are below and above the median relative yields, we find a Z-value of −1.887 and a p-value of 0.03, which indicates that the capitalisation ratios of the two subsamples are statistically significantly different from each other.
The capitalisation ratio of FPPs is also much lower in NUTS-3 regions where the average grassland share is above the median (λ FPP = 0.4) compared with the FPPs in NUTS-3 regions where the average grassland share is below the median (λ FPP = 0.25). In terms of farm organisation, we find that the capitalisation ratio of FPPs is considerably higher in NUTS-3 regions where farms are on average larger (λ FPP = 0.41) compared with those regions where farms are on average smaller (λ FPP = 0.26). The capitalisation ratios are very similar and not statistically different in terms of the share of parttime farms (0.30 for the below median subsample versus 0.30 for the above median subsample) and animal stocking rates (0.35 versus 0.31). In regard to rental market characteristics, we find some differences for areas with different average rental shares. The capitalisation ratio is higher in areas where more land is rented (0.39) compared with areas where less land is rented (0.31). We do not find differences in regard to population density (0.33 versus 0.32). However, we find that areas with a higher share of agricultural area in total land area have a higher capitalisation ratio of FFPs (0.19 versus 0.41). Overall, the capitalisation ratio for our different subsamples varies from 0.19 to 0.46, with an average of 0.33. The results for the long-run capitalisation ratios are qualitatively the same (Table A3). The average value for the long-run capitalisation ratio over all samples is 0.36.
We do not depict and discuss the results for the subsamples in regard to AEPs since we find only very low capitalisation ratios in general (between −0.02 and 0.06), which are statistically not significantly different from zero and there are no statistically significant differences between the subsamples. However, we do depict the results for LFAPs (Table 5 and Figure 3). The capitalisation ratio varies across the different subsamples, from 0.214 to 0.830, with an average of 0.456. The results are partly in line with the results for FPPs, but also reveal some differences. In line with FPPs, we find higher capitalisation ratios in areas with higher yields (0.45 versus 0.83), larger farm sizes (0.32 versus 0.73) and higher rental shares (0.34 versus 0.59). However, the capitalisation ratio is larger in areas with a higher grassland share with regard to LFAPs (0.52 versus 0.43), while the opposite is true for FPPs (0.25 versus 0.40). We also find reversed relations for stocking rates and the share of agricultural area. Moreover, we find a considerable higher capitalisation ratio for areas with a below median share of part-time farming (0.58 versus 0.214), what was not the case for FPPs. The results for the long-run capitalisation ratios are qualitatively the same (Table A3).

Discussion
Recently, a few empirical studies investigated the extent to which CAP payments capitalise into rental prices. The results do not provide a clear answer, but rather, estimate capitalisation ratios ranging from zero to almost full capitalisation (Ciaian and Kancs, 2012;Kilian et al., 2012;Guastella et al., 2014;Michalek, Ciaian and Kancs, 2014;Van Herck et al., 2014;O'Neill and Hanrahan, 2016;Klaiber, Salhofer and Thompson, 2017;Guastella et al., 2018;Henning and Breustedt 2018). Compared to these recent studies, except for Klaiber, Salhofer and Thompson (2017), we have a more comprehensive data set available, i.e. more years and observations for a relatively small and homogenous agricultural area. With regard to FPPs, our results are in line with two studies employing similar estimation methods and comparable datasets, but for different countries. For Ireland, O'Neill and Hanrahan (2016) estimate a range of 21-53 cents in the long-run based on farm type. In reference to decoupled payments made in Kansas, Hendricks, Janzen and Dhuyvetter (2012) estimate a magnitude of 37 cents. Our results also correspond to estimates of 38 cents per additional euro dedicated to coupled area payments made prior to the Fischler Reform in 2013 by Breustedt and Habermann (2011) for cropland in Germany and with estimates from Klaiber, Salhofer and Thompson (2017) of 37-53 cents per additional FPP euro. The latter is based on a similar dataset; however, the model is different because it accounts for self-selection bias, but does not account for price dynamics and endogeneity of market revenues.
Less empirical evidence exists for LFAPs. Our results are considerably higher than the 19-29 cent range found by Kilian et al. (2012), but are much lower than those of Patton et al. (2008), who report full capitalisation of LFAPs. We attribute this high capitalisation ratio for LFAPs to the fact that no substantial obligations are tied to such payments. Moreover, the exact value of payments is not only evident to farmers, but also to landowners, potentially improving their bargaining position.
Although our results are for FPPs as implemented from 2005 to 2013, we do not expect very different results for the policy in place today with the BPS and additional payments (green direct payments, redistributive payments, payments for areas with natural or other specific constraints and payments for young farmers). All of these payments are still linked to land in the same way as payments were before. This is especially true for the most important top-up, green direct payments. Though conditional on farmers' compliance with some mandatory practices, which potentially benefit the environment, the amount of payments does not depend on the actual level of environmentalfriendly activities or the environmental performance of the farm, but instead is a top-up (30 per cent) proportionate to the BPS received. Moreover, in the way these mandatory practices are defined and implemented in most countries, it has been argued that only a small proportion of farms actually have to change their land use behaviour (Louhichi et al., 2018) and their environmental effects are relatively minor (Gocht et al., 2017). In regard to the 'CAP beyond 2020', it is not clear yet what the Member States will actually propose as to the 'New Delivery Model' of the CAP. Its start will be postponed at least until 2022 (Matthews, 2020). Based on the proposal of the European Commission (2018), the fundamental structure of the direct payments system (payments granted on a per hectare basis) is not likely to be changed (Jongeneel and Silvis, 2018). Hence, capitalisation of payments into land rental prices will most likely remain.
We do not find any substantial capitalisation ratio for AEPs. This is consistent with results obtained in other studies. For example, Lence and Mishra (2003) did not find that the US Conservation Reserve Program payments have a statistically significant influence on cash rental prices in Iowa. Kilian et al. (2012) found a negative effect of AEPs on land rental prices in Bavaria. 8 O'Neill and Hanrahan (2016) obtained coefficient estimates of close to zero for EU second pillar payments in general. Our results suggest that zero to very moderate windfall profits for landowners are generated from these programmes. However, this does not necessarily mean that agri-environmental schemes do not create windfall profits for farmers, such as if they are compensated for practices they would have implemented anyway or if payments exceed additional costs of participation (Salhofer and Streicher, 2005;Chabé-Ferret and Subervie, 2013;Menning and Sauer, 2019). Rather, what we find here is that even if there are windfall profits, they are not passed to landowners through rental prices. A possible explanation for the difference in AEPs compared to FPPs and LFAPs is the difference in information asymmetry. Landowners are usually very aware that farmers receive FPPs and if owned land is in an area eligible for LFAPs. They usually also, at least approximately, know the amount of payments per hectare this implies. This is not the case for AEPs. Landowners usually do not know if and in which agri-environmental scheme the farmer will participate. Even if landowners know, they probably neither know the amount of payments nor the additional costs of participation. Therefore, they cannot take advantage of this in rental price negotiations.
While there is some empirical evidence that the degree of capitalisation crucially depends on how the policy is implemented (Michalek, Ciaian and Kancs, 2014;Klaiber, Salhofer and Thompson, 2017), not much is known about other determinates. We try to fill this gap by dividing our total sample into subsamples based on the NUTS-3 regions in which the farm is located and different characteristics of these regions. We find considerable variation between these subsamples for FPPs (0.21-0.52) and even more for LFAPs (0.23-0.94). One has to keep in mind that these results are based by dividing the sample into two equally large parts. Hence, the estimated capitalisation ratios are probably not minima and maxima, but rather just give an indication of how average capitalisation ratios might be different in different regions with different agricultural and land market characteristics. Comparing the capitalisation ratios of belowand above-average subsamples, we find strong evidence that the capitalisation ratio for FPPs is higher in areas with higher yields, a lower grassland share, a larger average farm size, a higher share of agricultural area and to some extent a higher rental share. With regard to yields and grassland share, the results may indicate that competition for high-quality land is higher and implies that more of the payments are passed down to landowners. This is also in line with results from O'Neill and Hanrahan (2016) for Ireland, which find that short-run and long-run capitalisation ratios of FPPs are higher for tillage farms compared to dairy, cattle and sheep farms.
The finding that farm size and the capitalisation ratio are positively correlated is in line with similar results by Michalek, Ciaian and Kancs (2014) based on farm-level data for EU-15 Member States. However, this finding is not in line with theoretical findings in Graubner (2018), which argues, based on a model of spatial competition for land, that in areas with larger farms, on average, the number of farms is lower and farmers are more likely to cooperate (collude) and depress land rental prices. However, one has to keep in mind that average farm sizes are still relatively small, even in favoured crop areas in Bavaria compared to Eastern Germany or some NMS. Farm size and rental shares are found to be positively correlated (Table A1 in the appendix). Therefore, one could argue that once a farmer decides to work full-time and is solely dependent on farm income, s/he is probably more obliged to rent land and willing to pay a higher rental price. However, this is also not completely separated from natural conditions and the farm type. Specialised crop farms need more land to employ a full-time farmer as do dairy farms. Given that the land sales market in Bavaria is very thin, crop farmers are more obliged to rent land than for example dairy farms. Our result that the capitalisation ratio is higher in areas with a higher share of agricultural area in total area may appear counterintuitive at first, as we would expect less pressure on the rental market in more rural areas. However, as mentioned before, areas with a large share of agricultural area are not in the less favoured, mountainous areas of the south where forests cover a large share of total area, but rather in the more favoured areas in the middle of Bavaria. This is also confirmed by a positive correlation coefficient of the share of agricultural area with yields and a negative correlation coefficient with the grassland share (Table A1 in the appendix). Hence, the conclusion of all of these partial results is clearly that the capitalisation ratio of FPPs is higher in favoured agricultural areas, characterised by high yields, a low share of grassland, large farms and high rental shares. We do not find much difference in the capitalisation ratio of FPPs in regard to the share of part-time farming, the population density and the animal stocking rate.
The interpretation of our results in regard to LFAPs is less clear. In general, the results are more dispersed compared to FPPs. We find capitalisation rates for LFAPs to be higher than for FPPs for 13 out of 16 subsamples. Similar to FPPs, for LFAPs, we find a higher capitalisation ratio for areas with higher yields, larger farms and higher rental shares. Together with the observation that the capitalisation ratio is higher in areas with fewer part-time farms, this again points at higher capitalisation rates in more favourable areas. However, there are also characteristics for which we find reversed correlations. Contrary to FPPs, capitalisation ratios for LFAPs are higher in areas with a higher grassland share, higher stocking rates and a lower share of agricultural area. A possible interpretation of these results is, that while the capitalisation of FPPs is lower in less-favoured areas, this is not the case for LFAPs. However, these results remain puzzling to some extent and highlight that the capitalisation ratio of LFAPs may be influenced by many factors. The results for the long-run capitalisation ratios are qualitatively the same (Table A3).

Conclusions
The CAP supports EU farmers with more than EUR 58 billion per year. About 80 per cent of these payments are spent on three different measures: first pillar payments (FPPs), payments for participation in agri-environmental programs (AEPs) and payments for less-favoured areas (LFAPs). Since all of these payments are tied to land and more than 50 per cent of UAA in the EU is rented, the question of whether these payments actually support the income of active farmers or are windfall profits for landowners through higher rental prices is crucial. Moreover, if the degree of capitalisation varies between countries and regions, this may cause biases in income, competitiveness and economic sustainability of farms.
To investigate these issues, we analyse a comprehensive data set of more than 3,000 farms in Bavaria over 6 years. Examining this data set has the advantage that all farms face exactly the same regulations in regard to first and second pillar policies of the CAP. Hence, while the differences in the results of past studies could probably be explained, at least to some extent, by national differences in regard to the implementation of the CAP and customs and regulations related to land markets, here we can investigate differences in the capitalisation ratio based on natural, farm organisation and rental market characteristics. We do so by estimating dynamic panel models for the total sample and subsamples defined along these characteristics.
We find that in general, FPPs and LFAPs capitalise into rental prices to a considerable extent, while AEPs do not. In particular, based on estimations with our total sample and different subsamples, our best guess is that, on average, about 30 per cent of FPPs and 40-50 per cent of LFAPs are capitalised into land rental prices. Bearing in mind that about 45 per cent of all Bavarian agricultural land was rented in 2010 (StMELF, 2019), a considerable share of support is realized by landowners rather than by active farmers. This clearly contradicts the objective of the CAP to direct 'support exclusively to active farmers' (European Commission 2010, p. 3). We find very low point estimates for the capitalisation ratio of AEPs, which are statistically not significantly different from zero in almost all of our models and subsamples.
We also show that even in a relatively small region like Bavaria, the capitalisation ratio can vary considerably across regions. We find strong evidence that the capitalisation ratio of FPPs is higher in favoured areas with high-quality land, larger farms and a higher rental share. This makes sense if high-quality land is more limited and more competition for such land exists. It also implies that the income effect of FPPs is higher in a less favoured area with smaller farms. The results in regard to LFAPs are harder to interpret. Capitalisation ratios are in general higher and also more dispersed. In line with FPPs, we also find higher capitalisation rates in areas with higher yields, larger farms and higher rental rates. However, we also find higher capitalisation rates in areas with high grassland shares, high stocking rates and a smaller share of agricultural area in total area, i.e. less favoured areas.
Our findings have important policy implications. First, they question the usefulness of FPPs to transfer income to farmers in general, as well as the usefulness of LFAPs to account for a natural disadvantage, since high shares of these payments are passed to landowners. Second, the effects of the CAP payments are very heterogeneous, even within a region sharing the same policies and land market institutions. Therefore, CAP payments can cause biases in farm income, competitiveness and economic sustainability of farms.