Price dispersion in farmland markets: What is the role of asymmetric information?

Kahle, Christoph; Seifert, Stefan; Hüttel, Silke Working Paper Price dispersion in farmland markets: What is the role of asymmetric information? FORLand-Working Paper, No. 11 (2019) Provided in Cooperation with: DFG Research Unit 2569 FORLand "Agricultural Land Markets – Efficiency and Regulation", Humboldt-Universität Berlin Suggested Citation: Kahle, Christoph; Seifert, Stefan; Hüttel, Silke (2019) : Price dispersion in farmland markets: What is the role of asymmetric information?, FORLand-Working Paper, No. 11 (2019), Humboldt-Universität zu Berlin, DFG Research Unit 2569 FORLand "Agricultural Land Markets - Efficiency and Regulation", Berlin, https://doi.org/10.18452/20661 This Version is available at: https://hdl.handle.net/10419/213065 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/de/ Published by DFG Research Unit 2569 FORLand, Humboldt-Universität zu Berlin Unter den Linden 6, D-10099 Berlin https://www.forland.hu-berlin.de Tel +49 (30) 2093 46845, Email [email protected] Agricultural Land Markets – Efficiency and Regulation Price dispersion in farmland markets: What is the role of asymmetric information? Christoph Kahle, Stefan Seifert, Silke Hüttel FORLand-Working Paper 11 (2019) Price dispersion in farmland markets: What is the role of asymmetric information? Christoph Kahle‡, Stefan Seifert, and Silke Hüttel University of Bonn, Institute for Food and Resource Economics, Production Economics Group (ILR-PE), Germany. Abstract This article investigates the role played by informational cost in agricultural land markets to explain price dispersion. Based on a hedonic model under incomplete information, we build a two-tier stochastic frontier. By linking costs of being information deficient to agent characteristics such as degree of professionalism, we identify relative price effects of buyers and sellers related to search. We compile a comprehensive data set of more than 10,000 transactions in Saxony-Anhalt, Germany, between 2014 and 2017. We find institutional sellers to achieve the lowest losses resulting from information deficiency while tenant buyers can benefit from informational advantages. We conclude that Germany’s policy-makers can do more to support market transparency. Key words: farmland markets, hedonic pricing, information deficiency, two-tier frontier JEL classification: D82, D83, Q15, Q24 ‡Corresponding Author. Address: Meckenheimer Allee 174, 53225 Bonn, Germany. E-mail: [email protected]. Phone: +49228 732893. 0We gratefully acknowledge financial support from the German Research Foundation (DFG) through Research Unit 2596 ’Agricultural Land Markets - Efficiency and Regulation’ 2017–2020. We thank the Committee of Land Valuation Experts of Saxony-Anhalt and the State Office for Survey and Geoinformation (LVermGeo) for the fruitful collaboration, and for providing the data. We further thank the participants of the FORLand workshops, the 165th EAAE Seminar 2019 in Berlin, the 8th EAAE PhD Workshop 2019 in Uppsala, the PhD Seminar at the University of Bonn, the 59th GEWISOLA Annual Conference 2019 in Brunswick, and the 16th EWEPA 2019 in London for their helpful comments. 1 Price dispersion has been traced back to heterogeneous buyer and seller groups in markets for homogeneous goods (cf. Kaplan et al. (2019) and cited literature therein). Farmland is a heterogeneous and unique good with a limited overall supply. Given its immobility, however, suitable substitutes are often lacking, most farmland markets are narrow with a high specificity of each transaction (cf. Borchers, Ifft, and Kuethe 2014), and even though capital may be mobile, market entry remains low despite increasing demand by investors. Trading volumes range between one and two percent in many regions of the Global North (Ciaian, Kancs, and Swinnen 2010; Bigelow, Borchers, and Hubbs 2016). Thus, farmland markets share characteristics of thin markets (Kuethe and Bigelow 2018). Thin farmland markets present several problems. A seller’s maximum willingness to pay may exceed a buyer’s minimum willingness to accept. Expectations of surpluses over which they can bargain (Harding, Knight, and Sirmans 2003) may emerge. Search for the respective seller (buyer) with lowest (highest) willingness to accept (pay) is costly, as is finding substitutes. Gathering the information needed to establish an agricultural property’s true market value may be expensive and time-consuming for the seller and buyer; depending on the respective search process and bargaining position, one or the other may influence the price (King and Sinden 1994). Hence, in addition to characteristics relevant for productivity1, the information gathering proficiency and bargaining power of both seller and buyer matter (Polachek and Yoon 1987). Resulting agent-specific prices may neither send appropriate market information nor help in efficient price discovery. Agent-specific prices have been traced to the different expectations held by new owners concerning a property’s potential future returns (e.g., Brorsen, Doye, and Neal 2015). Price variations have also been explained by regional peculiarities, for instance, expectations of future land development in urban proximity (e.g., Plantinga and Miller 2001; Kolbe et al. 2015), zoning regulations in peri-urban markets (e.g., Eagle et al. 2014; Turner, Haughwout, and van der Klaaur 2014), the variety of agricultural policies (e.g., Graubner 2018), 1We refer to Nickerson and Zhang (2014) for an excellent overview on farmland price determinants. 2 the agglomeration effects of subsidized renewable energy production (e.g., Hennig and Latacz-Lohmann 2016; Towe and Tra 2013), and the local or regional market regulations (e.g., Lawley 2018). The majority of farmland price studies have implicitly acknowledged remaining price variation by means of spatio-temporal effects (e.g., Maddison 2009), but that are hardly generalizable. Few studies have explored agent-specific prices due to thinness, such as price-sensitivity to farmer-buyer characteristics (e.g., Kuethe and Bigelow 2018) or price-effects due to the competition among potential farmer buyers (Margarian 2010). To our knowledge, only Cotteleer, Gardebroek, and Luijt (2008) have acknowledged agent-specific prices due to bargaining and market power. Typically framed within a hedonic pricing framework, these studies focus on average effects and do not consider the search process and the role of asymmetrically distributed information costs. One study has argued that such asymmetries produce different different bargaining positions with respective price-sensitivity (Curtiss et al. 2013). The results by these authors, however, lack external validity since estimation procedures have not been adjusted to acknowledge these asymmetries. Polachek and Yoon (1987) were the first to suggest a two-tier model which separates the observed prices into a hedonic part and three error components (noise, and seller- and buyer-specific price impacts) to account for the relative levels of agents’ search and information costs. While studies of the real estate market have highlighted the role of asymmetric information in price schedules (Kumbhakar and Parmeter 2010), we have not found similar studies of farmland markets. Therefore, in this article we empirically investigate the role of the search process in farmland price formation. We assume that the search process and respective additional cost can be related to a seller’s degree of professionalism, for instance, a licensed real estate agent who often relies on auctions without bargaining, versus a private seller who primarily relies on negotiating, to understand the relative price relevance of the search process with potential losses for the less professional seller. On the buyer side, we link the categories of (non-)farmer and (non- 3 tenant) farmer to asymmetries in the search process to identify the respective price effects of both parties. To differentiate seller- and buyer-specific effects, we construct a two-tier model of farmland prices within a hedonic price function with two additional one-sided error terms following Kumbhakar and Parmeter (2010). We specify the terms as functions of the observed characteristics of buyers and sellers using the scaling property (Parmeter 2018). To validate the estimation approach, we compare our two-tier model based on the theory of thin markets to a reduced form of the model, where seller and buyer characteristics linearly add to the price function. We use a data set of more than 10,000 transactions for arable land from 2014 and 2017 in, Saxony-Anhalt, a heavily agricultural state in Germany. Its history of economic transition with a professional privatization agency makes the state an ideal setting for comparing different degrees of professionalism and searches on the seller side. We expect that modeling the hedonic price function within a stochastic frontier framework combined with spatial and temporal effects will help to mitigate the omitted variable biases that typically result from the data limitations in such models (Carriazo, Ready, and Shortle 2013). We find institutional sellers relying on public tenders to achieve the lowest losses resulting from information deficiency with markups. Farmer-tenant buyers benefit from informational advantages resulting in markdowns, with the exception of very small and very large transactions. We believe that existing studies have largely underestimated the role of informational asymmetries by neglecting the explicit price-impact of buyers and sellers. Therefore, this article makes the following contributions. To our knowledge, it is the first to construct a two-tier model with a scaling property free of distributional assumptions about the error terms, and apply it to the agricultural sector.2Second, the model emphasizes the importance of making adjustments when analyzing prices in thin markets. Third, we hope it 2We refer to Bonanno et al. (2019) who applied a one-tier model to price the credence attributes of food 4 will inform the development of policy measures that support transparency and efficiency in farmland markets. The remainder of the article is organized as follows. Section 2 explains the theoretical and econometric framework used. Section 3 describes the empirical strategy, the data, and the hypotheses. Section 4 applies the two-tier and simplified models to a comprehensive data set of agricultural transactions and discusses the results. Section 5 discusses the policy implications and gives suggestions for future research. Theoretical model and estimation A hedonic pricing model with incomplete information We employ a search model with bargaining to identify the effects of asymmetric information on farmland prices. We assume that buyers and sellers enter the market with a set of beliefs about the distribution of prices, given the heterogeneity of the land. While agents may have different sets of information, they can invest in searches to improve their bargaining positions, for instance, by identifying competing offers from other sellers or buyers. We assume that all agents search optimally and that each buyer faces a trade-off between the search costs and finding a seller with a lower willingness to accept (WTA). Likewise, each seller faces a trade-off between the search costs and finding a buyer with the highest willingness to pay (WTP). Both search and informational costs may vary across agents, for instance, when local and non-local prospective buyers have different access to information, the cost variations are particularly relevant for substitutes. Similarly, an experienced professional seller may have lower search costs than a private seller with no experience. For instance, the professional seller may rely on tendering procedures that ease search finding the buyer with the highest WTP while the inexperienced private seller may rely on negotiations. In other words, when agents with higher search costs stop gathering information sooner, the buyers (sellers) with high search costs experience higher (lower) prices. 5 To model the search process under informational asymmetries, we use a hedonic pricing model following Kumbhakar and Parmeter (2010). We use the two-tier framework of Polachek and Yoon (1987) to incorporate information and search. Thus, the two additional one-sided error terms acknowledge the price impact of the buyer and seller characteristics related to search and informational cost. According to the standard hedonic pricing model of Rosen (1974) under full information and thick markets, the hedonic price of farmland Phis formulated as (1) Ph=h(x)+v, where xdenotes a vector of lot characteristics (e.g., lot size and soil quality), h(.)is the hedonic price function, and vcollects measurement errors and noise. In this model, price variation is only caused by heterogeneity and potential information asymmetries are disregarded. To expand this model to our setting, we adopt a two-tier frontier model and apply it to buyers and sellers separately. The maximum WTP among buyers defines an upper bound of the market price, and the lowest WTA among sellers defines the lower bound. The price a seller receives, Ps m, can be written as (2) Ps m=Pb−u, where Pbrefers to the highest WTP by a potential buyer in the market and u,u≥0 denotes a seller’s loss from information deficiency, that is, the loss caused by the inability to identify the buyer with the highest WTP. Likewise, the price a buyer pays, Pb m, can be written as (3) Pb m=Ps+w, where Psis the lowest WTA in the market, and w,w≥0 is the markup caused by being unable to identify the lowest WTA. 6 For a transaction to take place, the identical prices for buyer and seller form the market price Pmsuch that Pm=Pb m=Ps m. Using equations (2) and (3) yields Pm=Ps+w=Pb−u, which can further be rearranged such that (4) Pm+u−w=Pb−w=Ps+u, where Ps+uand Pb−ware the hedonic prices for sellers and buyers, respectively, adjusted for their information. Since Ps,Pb,u, and ware unobserved, identification requires further assumptions. Kumbhakar and Parmeter (2010, p. 10) argue that Pm+u−wcorresponds to the price under full information given by the hedonic price Ph. Thus, using equations (4) and (1), the observed market prices can be expressed as (5) Pm=h(x)+v−u+w=h(x) +ε. Equation (5) states that the observed market price of a lot consists of the implied characteristics of the lot h(x), unobserved noise v, and the costs of information deficiency of sellers (u) and buyers (w). The composite error term εcollects noise and costs of information deficiency. Note that this model collapses to the standard hedonic pricing model if buyers and sellers have identical information deficiencies (u=w), including the case of fully informed agents (u=w=0). The current setting, however, assumes identical information deficiencies for all buyers and for all sellers. To allow for potential heterogeneity across agents, we follow Parmeter (2018) and model information deficiencies as functions of agents’ characteristics. In particular, a buyer’s information deficiency wis a function of buyer characteristics zw, which may include knowledge of local market conditions. Likewise, a seller’s cost of information deficiency is a function of seller characteristics zu, which may include access to distribution channels. Extending equation (5) delivers the hedonic pricing model with incomplete information and buyer- and seller-specific costs of information deficiency, the regression 7 Table 1. Descriptive statistics for data set, 2014–2017 N = 10,778 Mean Median SD Q1 Q99 Dependent variable Price (e/m2)P1.63 1.50 0.86 0.35 4.08 Lot Characteristics Lot size (ha) xS3.08 1.02 6.40 0.03 26.93 Soil quality (Index) xQ64.11 66.00 22.65 21.00 100.00 Lot independence (1/0) xI0.86 1 0.35 0 1 Lot is leased (1/0) xL0.66 1 0.47 0 1 Wind energy area (1/0) xW0.01 0 0.08 0 1 Controls at municipal level Wind power turbines per ha mW0.002 0.001 0.003 0 0.02 Biomass capacity kW per ha mB0.33 0.1 2.44 0 2.58 Transaction share of BVVG mBVV G 0.12 0.09 0.10 0.00 0.48 Seller Characteristics BVVG (1/0) sBVVG 0.08 0 0.28 0 1 Professional seller (1/0) sPro f 0.02 0 0.13 0 1 Public seller (1/0) sPub 0.02 0 0.15 0 1 Buyer Characteristics Farmer (1/0) bF 0.74 1 0.44 0 1 Tenant (1/0) bT 0.49 0 0.50 0 1 Farmer and tenant (1/0) bFT 0.49 0 0.49 0 1 Farmer and non-tenant (1/0) bFNT 0.26 0 0.44 0 1 Note: Due to data privacy reasons, we cannot report minima and maxima. 14 We hypothesize that information deficiencies for private sellers will be higher than for professional sellers, but higher for professional sellers than for BVVG. BVVG as the major player in Saxony Anhalt’s farmland market has around 20 % market share by acreage on average, and up to 60 % in some regions (LVermGeo 2018a). BVVG relies on tendering procedures, where auction rules, bidding requirements, and auction results are publicly available on the BVVG website and published in local media and farmers’ magazines. BVVG’s professionalism and level of specialization is hypothesized to ease the search process and finding potential buyers with the highest WTP. Potential buyers may further perceive lower risks concerning a transaction failure when considering BVVG. This may even attract potential buyers and further ease search. Therefore, BVVG may benefit, and fostered by the auction mechanism we hypothesize higher prices compared to other sellers. That is, this group is expected to incur lower losses of information deficiency than other sellers. Although private owners continued to transact on their own, others began to use licensed real estate agents. These professional sellers use, for example, procedures comparable to public tenders, and advertise and target potential buyers efficiently. Due to this professionalism, we expect lower costs of information deficiency for these sellers compared to private sellers without experience. However, compared to BVVG, real estate agents have a lower turnover rate and thus we expect the markup to be lower for this group. As a third group, public authorities such as municipalities or local governments may benefit from experience but at a lower extent compared to real estate agents. This advantage may further be off-set by costs caused by a potential principal agent problem: public sellers’ goal may not primarily be selling at profit maximizing prices, and lower prices might be accepted due to time limitations and missing incentives to invest in search (cf. Attkinson and Halvorsen 1986). Hypothesis 2: Buyer information deficiency We hypothesize that informational deficiencies for farmers and tenant buyers will be lower than for non-farmers, but higher for non-tenant farmers than for tenant farmers. 15 On the buyer side, we can distinguish by farmers and non-farmer buyers, and whether the buyer was the former tenant. Thus, we cannot clearly rank by professionalism, we can rather rely on asymmetric knowledge: farmers and tenants in particular are hypothesized to be better informed about potential returns from land-use and the local land market conditions. This informational advantage may in particular be relevant for expected alternative supply offers at the time of the bidding. This may result in lower costs for information acquisition for these groups and might offer to form more realistic expectations about the returns reducing the likelihood of overpaying (e.g., winners’ curse). In our second hypothesis, we expect a price decreasing effect of tenancy: as a result of the existing relation between sellers and tenants prior to the transaction, social capital on both sides might influence the price by reducing search cost (e.g., Kostov 2010; Robinson, Myers, and Siles 2002). This social capital may result in a reduction of cost of being information deficient for tenant farmers compared to non-tenant farmers and non-farmers. Hypothesis 3: Information advantage based on lot size We hypothesize that farmers and/or tenants have a price advantage over non-farmers and/or non-tenants that is increasing in plot size. Identification of a pure farmer effect may be challenging (cf. Croonenbroeck, Odening, and Hüttel 2019), in particular since both groups, farmers and non-farmers, are rather heterogeneous. Both groups could contain investors and it may rather be the intention on how to use the land after purchase that determines willingness to invest in search (e.g., Magnan and Sunley 2017). This is why we consider that informational advantages of tenants and potentially farmers with the intention to use that land could vary in plot size. For instance, for larger plots, the group of non-farmer buyers may be less heterogeneous since these lots do not reasonably allow alternative land use apart from farm operation such as horse keeping, gardening or real estate (Brorsen, Doye, and Neal 2015). We further expect larger transactions to be less heterogeneous in terms of lot constitution and valuation.4Since val- 4We refer to Yiu, Wong, and Chau (2009) for evidence in real estate market 16 0 1 2 3 4 5 6 0.0 0.2 0.4 0.6 0.8 1.0 e/m2 ˆ Fz(e/m2) Private, n = 9431 BVVG, n = 910 Professional, n = 181 Public, n = 256 0 1 2 3 4 5 6 0.0 0.2 0.4 0.6 0.8 1.0 e/m2 Farmer & tenant, n = 5236 Farmer & non-tenant, n = 2770 Non-farmer, n = 2772 Figure 2. Unconditional cumulative density functions of prices paid by seller type (left) and buyer type (right) uation is mainly based on the conventional and observable determinants of farmland prices, this allows tenants and farmers to better use their knowledge about the expected returns. Therefore, under hypothesis 3 we suspect larger plots to be sold with price markdowns to farmers and/or tenants that increase with lot size. Descriptive evidence: seller and buyer specific price differentials We use the data set to systematically investigate the unconditional cumulative distribution function of the raw prices by buyer and seller type. As shown in figure 2, there are differences in prices between professional, public and private sellers in the raw data. There are, however, only small differences between the three potential combinations of (non-)farmers and (non-)tenants types in the raw data, with the exception of slightly higher prices for non-tenant farmers. Model specification To test the three hypotheses, we use log-linear regression equations consisting of a hedonic part h(x)and the combined error term εthat collects noise and the information deficiency costs for buyers and sellers. Based on a Box-Cox transformation for the continuous variables, lot size and soil quality, we enter them into the model in power transformations 17 and in interaction, and enter other regional continuous variables such as renewable energy sources linearly. To capture the implicit spatio-temporal effects, we use twelve location indicators LCkbased on regional classes provided by the Committee of Land Valuation Experts (LVermGeo 2018b).5We model the time effects by using linear-quadratic trend, τand τ2, as well as a locally differing trends by interacting the trend variables with the location indicators. The regression equation can be formulated as (8) log(P) = βS√xS+βQ√xQ+βSQ(xS·xQ)+βIxI+βLxL+βWxW +γWmW+γBmB+γBVVGmBVV G +γττ+γτ2τ2 + 12 ∑ k=1 γLC,kLCk+ 12 ∑ k=1 γLC,τ,k(LCk·τ)+ε, where the β’s denote the respective hedonic parameters to be estimated and γ’s denote the parameters for regional variables at the municipal level and the time- and spatial effects. We consider two models, TT1 and TT2 using the two-tier approach with differing error term specifications εby model (cf. table 2). Both models obey an identical specification of the seller side: dummy variables for BVVG, other professional, and public sellers (hypothesis 1). On the buyer side, in TT1 we assess whether information asymmetries for tenant and non-tenant farmers compared to non-farmers exist and if they are price influencing (hypothesis 2). We also assess whether the informational advantages are more pronounced depending on the size of the transaction (hypothesis 3). To test these hypotheses, indicators of tenant and non-tenant farmers enter the error term interacted with lot size.6To test hypothesis 3 we enhance model T T 1 by adding interaction terms of lot size and its square with buyer characteristics. This gives model TT2 that allows us to test whether buyer-specific price effects are sensitive to lot size. As shown in table 2, δdenote the parameters to be estimated and capture the impact of buyer and seller 5Each location class represents a geographically compact area with similar characteristics, such as connection to infrastructure. Refer to figure 7 in the Appendix for a map. 6A simpler model specification failed to converge probably caused by the high overlap of the two buyer indicators and missing variation over transactions. 18 Table 2. Specifications of the error term ε Model ε=−u(zu,i,δu)+ ω(zω,i,δω)+ v LIN α+δsBVVGsBVV G +δsPubsPub +δsPro f sPro f +δbFT,xS(bFT ·xS)+δbFNT,xS(bFNT ·xS) +v TT1 -exp[µS+δsBVV GsBVV G +δsPubsPub +δsPro f sPro f ] +exp[µB+δbFT,xS(bFT ·xS)+δbFNT,xS(bFNT ·xS)]+v TT2 -exp[µS+δsBVV GsBVV G +δsPubsPub +δsPro f sPro f ] +exp[µB+δbFT,xS(bFT ·xS)+δbFNT,xS(bFNT ·xS) +δbFT,x2 S(bFT ·x2 S)+ δbFNT,x2 S(bFNT ·x2 S)]+ v characteristics, where µBand µSdenote the baseline cost of information deficiency for buyers and sellers (µ∗ wand µ∗ u). We estimate them as intercepts of the exponential functions to ensure that the sign of the effect of the costs of being information deficient on the price is consistent with theory. The baseline cost of information deficiency is required to scale the impact of the respective buyer and seller characteristics, but will not be interpreted directly. Finally, we compare the findings of the two-tier model and a linear benchmark model LIN, where the seller and buyer characteristics linearly add to the hedonic part. Model LIN includes an intercept α, which is omitted in the two-tier model, to identify baseline inefficiency terms µBand µS, respectively. Both models share an additional noise term v. We estimate models T T1 and T T 2 using the NLS procedure and estimate the linear benchmark model LIN using OLS. To account for heteroscedasticity induced by the composed error term, we refer to multiway clustered standard errors with clusters corresponding to combinations of thirty quantiles of lot size and the squared soil quality.7 7All calculations are performed with R (R Core Team 2019). NLS estimation uses the nls function from the stats package. Estimation of robust standard errors uses the sandwich package (Zeileis 2004). To ensure convergence, we run estimations for the non-linear models 5000 times with random starting values. R codes are available upon request. 19 Results Table 3 lists the models’ parameter estimates for the hedonic variables and the buyer and seller characteristics; see table 6 in the appendix for the estimates of spatial controls and time trends. Overall, the models show satisfactory goodness of fit as indicated by the squared correlation coefficient of about 0.67 across all models. The hedonic estimates are strikingly similar across the different specifications. As noted by Kumbhakar and Parmeter (2010), the intercept of LIN corresponds to the sum of the baseline cost of being information deficient if E[w−u] = 0, otherwise OLS-estimates would be biased. A comparison with the sum of the baseline terms in TT1 (−2.142 ≈−e0.925 +e−0.990) does not indicate such bias. In this case, the estimates of the benchmark model LIN represent average effects for sellers and buyers. In line with previous studies, we find positive coefficients of soil quality and lot size (e.g., Lehn and Bahrs 2018), although non-linear (e.g., Maddison 2000; Sheng, Jackson, and Lawson 2018). Including the negative interaction coefficients, the price effect of additional size decreases in size, whereas the price effect for higher quality soils may be too costly, that is, the effect of additional size can even reverse (cf. figure 3 for T T 1). In the latter case, capital or borrowing constraints may also increase (Brorsen, Doye, and Neal 2015). Interestingly, we note that whether a lot can be independently used and leased out is irrelevant for the price vector. Other studies, however, have indicated a potential relationship between the price effect of lot size and an independent use (e.g., Gluszak and Zygmunt 2018). A recent study by Haan and Simmler (2018) using aggregated data found significant land-owner effects in other regions of Germany, whereas we find that regional renewable energy production does not influence the price of individual farmlands. Other studies using transaction data, however, found significant effects of biomass based energy production on rental prices in boom years (e.g., Hennig and Latacz-Lohmann 2016). 20 Table 3. Parameter estimates for hedonic variables and buyer and seller characteristics N = 10,778 LIN TT1T T 2 Lot characteristics Intercept −2.142∗∗∗ (0.053) √Size 0.125∗∗∗ (0.010)0.134∗∗∗ (0.010)0.107∗∗∗ (0.010) √Quality 0.202∗∗∗ (0.005)0.203∗∗∗ (0.006)0.200∗∗∗ (0.005) Quality ·Size −0.0001∗∗∗ (0.00002)−0.0001∗∗∗ (0.00002)−0.00003 (0.00003) Independence −0.001 (0.011)−0.004 (0.011)−0.001 (0.011) Wind energy area −0.007 (0.048)−0.006 (0.047)−0.001 (0.046) Lot is leased out −0.015∗(0.009)−0.014 (0.009)−0.012 (0.008) Regional characteristics Wind power turbines −0.435 (0.964)−0.431 (0.965)−0.321 (0.958) Biomass capacity 0.002 (0.001)0.002 (0.001)0.002 (0.001) BVVG share 0.155∗∗∗ (0.055)0.150∗∗∗ (0.054)0.149∗∗∗ (0.054) Seller characteristics Baseline cost 0.925∗∗∗ (0.055)0.905∗∗∗ (0.058) BVVG 0.390∗∗∗ (0.016)−0.167∗∗∗ (0.009)−0.169∗∗∗ (0.011) Public seller 0.070∗(0.037)−0.028∗(0.015)−0.028∗(0.015) Professional seller 0.185∗∗∗ (0.024)−0.077∗∗∗ (0.012)−0.075∗∗∗ (0.011) Buyer characteristics Baseline cost −0.990∗∗∗ (0.381)−1.055∗∗ (0.441) Farmer ·tenant ·size −0.006∗∗∗ (0.002)−0.027∗∗ (0.011)−0.008∗∗∗ (0.001) Farmer ·non-tenant ·size −0.0001 (0.001)−0.001 (0.003)0.040∗∗ (0.016) Farmer ·tenant ·size2−0.001∗∗∗ (0.0002) Farmer ·non-tenant ·size2−0.002∗∗∗ (0.0005) Note: Clustered standard errors (at thirty quantiles of lot size and the squared soil quality) in parentheses. Asterisks indicate the following: ∗= p<0.1; ∗∗ = p<0.05; ∗∗∗ = p<0.01. Parameter estimates for location classes, time dummies and interactions are reported the Appendix. 21 0.00 0.02 0.04 0.06 0.08 0.10 Size [ha] 0 20 40 60 80 100 Marginal effect of size on the price [e/m2] Mean quality Mean quality + 1 SD Mean quality - 1 SD Figure 3. Marginal effect of lot size on the price for different soil qualities based on TT1 22 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10 Estimates Farmer & non-tenant δbFNT,x2 s Farmer & tenant δbFT,x2 s Farmer & non-tenant δbFNT,xs Farmer & tenant δbFT,xs Public seller δsPub Professional seller δsPro f BVVG δsBVVG TT1 TT2 Figure 4. Parameter estimates and 95 % confidence intervals for buyer and seller variables Results underline relevance of BVVG’s regional activities: a 1 % increase in the share of transactions corresponds to a 0.15 % price increase, a finding which is contrary to Hüttel, Wildermann, and Croonenbroeck (2016). In regions and times when BVVG has a higher share, overall market transparency may be higher. This is because BVVG’s tendering procedures offer more information on the market volume and their policy to publish transaction data may ease forming a bid. As a result, prices could be closer to the competitive price. BVVG’s dominance over space and time, however, may also be evidence of market power. Figure 4 summarizes the estimates of seller and buyer variables for models TT1 and TT2. Respective positive δcoefficients indicate increasing information deficiency with the respective variable, and negative signs indicate the opposite. Regarding price effects, a positive δparameter for one buyer type indicates a higher price for this group compared to the reference (non-farmer, non-tenant). A positive parameter on the seller side, however, indicates lower prices for the respective seller group compared to the reference group (nonprofessional sellers). Models TT1 and TT2, which both show statistically significant negative parameter estimates for the three seller types, reveal that professional seller groups obtain higher prices 23 Appendix Figure 7. Map of location classes 30 Table 5. Descriptive statistics by seller and buyer types Mean Median St. Dev. Q1 Q99 Seller: BVVG Price (e/m2) 2.43 2.42 1.00 0.60 4.77 Lot Size (ha) 8.17 3.38 14.73 0.03 90.66 Soil Quality (Index) 63.91 65.00 21.99 21.00 99.00 Lot Independence (1/0) 0.86 1.00 0.35 0.00 1.00 Wind energy area (1/0) 0.00 0.00 0.07 0.00 0.00 Lot is leased (1/0) 0.67 1.00 0.47 0.00 1.00 Seller: Professional Seller Price (e/m2) 2.37 2.42 1.06 0.57 4.61 Lot Size (ha) 5.65 4.58 7.61 0.06 23.00 Soil Quality (Index) 70.71 75.00 22.29 22.00 99.00 Lot Independence (1/0) 0.94 1.00 0.23 0.00 1.00 Wind energy area (1/0) 0.01 0.00 0.07 0.00 0.00 Lot is leased (1/0) 0.69 1.00 0.46 0.00 1.00 Seller: Public Seller Price (e/m2) 1.58 1.40 0.89 0.41 4.13 Lot Size (ha) 4.39 0.82 10.94 0.02 68.37 Soil Quality (Index) 59.18 58.50 23.13 20.10 100.00 Lot Independence (1/0) 0.72 1.00 0.45 0.00 1.00 Wind energy area (1/0) 0.00 0.00 0.00 0.00 0.00 Lot is leased (1/0) 0.48 0.00 0.50 0.00 1.00 Buyer: Farmer Price (e/m2) 1.65 1.52 0.86 0.35 4.18 Lot Size (ha) 3.36 1.18 6.68 0.06 29.06 Soil Quality (Index) 64.93 67.00 22.42 21.00 100.00 Lot Independence (1/0) 0.89 1.00 0.31 0.00 1.00 Wind energy area (1/0) 0.01 0.00 0.09 0.00 0.00 Lot is leased (1/0) 0.78 1.00 0.41 0.00 1.00 Buyer: Tenant Price (e/m2) 1.59 1.50 0.81 0.35 3.96 Lot Size (ha) 3.02 1.00 6.43 0.06 28.21 Soil Quality (Index) 65.50 68.00 22.32 21.00 100.00 Lot Independence (1/0) 0.89 1.00 0.31 0.00 1.00 Wind energy area (1/0) 0.01 0.00 0.09 0.00 0.00 Lot is leased (1/0) 0.88 1.00 0.33 0.00 1.00 Note: Due to data privacy reasons, we cannot report minima and maxima. 31 Table 6. Regional and time control variable estimates N = 10,778 LIN TT1T T 2 Location classes Altmark-Mitte 0.347∗∗∗ (0.077)0.345∗∗∗ (0.077)0.346∗∗∗ (0.077) Altmark-Ost 0.098 (0.092)0.098 (0.092)0.095 (0.092) Altmark-West 0.273∗∗∗ (0.070)0.273∗∗∗ (0.071)0.278∗∗∗ (0.071) Boerde 0.673∗∗∗ (0.045)0.672∗∗∗ (0.046)0.673∗∗∗ (0.046) HAL-Sued 0.474∗∗∗ (0.045)0.474∗∗∗ (0.045)0.470∗∗∗ (0.045) Harz 0.078∗∗ (0.039)0.077∗∗ (0.039)0.077∗∗ (0.039) MD-HAL 0.515∗∗∗ (0.033)0.516∗∗∗ (0.034)0.516∗∗∗ (0.034) MD-Nord 0.180∗∗∗ (0.039)0.179∗∗∗ (0.039)0.183∗∗∗ (0.038) MD-Ost 0.462∗∗∗ (0.049)0.460∗∗∗ (0.049)0.460∗∗∗ (0.049) ST-Sued 0.563∗∗∗ (0.045)0.561∗∗∗ (0.045)0.555∗∗∗ (0.046) Vorharz 0.428∗∗∗ (0.046)0.427∗∗∗ (0.046)0.427∗∗∗ (0.045) Time trend Trend 0.161∗∗∗ (0.020)0.160∗∗∗ (0.020)0.161∗∗∗ (0.020) Trend2−0.012∗∗∗ (0.003)−0.012∗∗∗ (0.003)−0.012∗∗∗ (0.003) Interactions Altmark-Mitte ·trend 0.009 (0.023)0.008 (0.023)0.009 (0.023) Altmark-Ost ·trend 0.038 (0.030)0.037 (0.030)0.04 (0.030) Altmark-West ·trend 0.013 (0.025)0.012 (0.026)0.012 (0.025) Boerde ·trend −0.021∗∗ (0.009)−0.021∗∗ (0.009)−0.021∗∗ (0.009) HAL-Sued ·trend −0.024∗(0.013)−0.024∗(0.013)−0.022∗(0.012) Harz ·trend 0.011 (0.011)0.011 (0.012)0.012 (0.011) MD-HAL ·trend −0.022∗∗∗ (0.007)−0.022∗∗∗ (0.007)−0.022∗∗∗ (0.007) MD-Nord ·trend 0.008 (0.014)0.008 (0.014)0.007 (0.013) MD-Ost ·trend −0.022∗(0.013)−0.022∗(0.013)−0.021 (0.013) ST-Sued ·trend −0.016 (0.012)−0.016 (0.012)−0.013 (0.012) Vorharz ·trend −0.006 (0.012)−0.006 (0.012)−0.005 (0.012) Location classes yes yes yes Observations 10,778 10,778 10,778 CorP, ˆ P0.676 0.676 0.677 Residual Std. Error 0.326 (df = 10,739) 0.326 (df = 10,738) 0.325 (df = 10,736) Note: Clustered standard errors (at thirty quantiles of lot size and the squared soil quality) in parentheses. Asterisks indicate the following: ∗= p<0.1; ∗∗ = p<0.05; ∗∗∗ = p<0.01. 32 References Alvarez, A., C. Amsler, L. Orea, and P.S. Schmidt. 2006. “Interpreting and Testing the Scaling Property in Models where Inefficiency Depends on Firm Characteristics.” Journal of Productivity Analysis 25:201–212. Attkinson, S.E., and R. Halvorsen. 1986. “The relative efficiency of public and private firms in a regulated environment: The case of U.S. electric utilities.” Journal of Public Economics 29:281–294. Bigelow, D., A. Borchers, and T. Hubbs. 2016. U.S. Farmland Ownership, Tenure, and Transfer. Washington, D.C.: U.S. Department of Agriculture, Economic Research Service Economic Information Bulletin 161, August. BMEL. 2018a. Daten und Fakten. Land-, Forst- und Ernährungswirtschaft mit Fischerei und Wein- und Gartenbau. Berlin: Bundesministerium für Ernährung und Landwirtschaft. —. 2018b. Statistisches Jahrbuch über Ernährung, Landwirtschaft und Forsten 2017. Bonn: Bundesministerium für Ernährung und Landwirtschaft. BNetzA. 2018. “Marktstammdatenregister (MaStR) [Core energy market data register].” Retrieved from https://www.marktstammdatenregister.de/MaStR/Einheit/Lokationen/ LokationenOeffentlichUebersicht. Bonanno, A., F. Bimbo, M. Costanigro, A. Oude Lansink, and R. Viscecchia. 2019. “Credence attributes and the quest for a higher price – a hedonic stochastic frontier approach.” European Review of Agricultural Economics 46:163–192. Borchers, A., J. Ifft, and T. Kuethe. 2014. “Linking the Price of Agricultural Land to Use Values and Amenities.” American Journal of Agricultural Economics 96:1307–1320. Brorsen, B.W., D. Doye, and K.B. Neal. 2015. “Agricultural Land and the Small Parcel Size Premium Puzzle.” Land Economics 91:572–585. 33 Bulow, J., and P. Klemperer. 1996. “Auctions versus Negotiations.” American Economic Review 86:180–194. Carriazo, F., R. Ready, and J. Shortle. 2013. “Using stochastic frontier models to mitigate omitted variable bias in hedonic pricing models: A case study for air quality in Bogotá, Colombia.” Ecological Economics 91:80–88. Chow, Y.L., I.E. Hafalir, and A. Yavas. 2015. “Auction versus Negotiated Sale: Evidence from Real Estate Sales.” Real Estate Economics 43:432–470. Ciaian, P., d. Kancs, and J.F.M. Swinnen. 2010. EU Land Markets and the Common Agricultural Policy. Brüssel: European Union and Centre for European Policy Studies. Cotteleer, G., C. Gardebroek, and J. Luijt. 2008. “Market Power in a GIS-Based Hedonic Price Model of Local Farmland Markets.” Land Economics 84:573–592. Croonenbroeck, C., M. Odening, and S. Hüttel. 2019. “Farmland values and bidder behaviour in first-price land auctions.” European Review of Agricultural Economics, in press. Curtiss, J., L. Lelínek, M. Hruska, T. Medonos, and V. Vilhelm. 2013. “The Effect of Heterogeneous Buyers on Agricultural Land Prices: The Case of the Czech Land Market.” German Journal of Agricultural Economics 62:116–133. Destatis. 2018. Land- und Forstwirtschaft, Fischerei: Kaufwerte für landwirtschaftliche Grundstücke 2017. Wiesbaden: Statistisches Bundesamt. Eagle, A.J., D.E. Eagle, T.E. Stobbe, and G.C. van Kooten. 2014. “Farmland Protection and Agricultural Land Values at the Urban-Rural Fringe: British Columbia’s Agricultural Land Reserve.” American Journal of Agricultural Economics 97:282–298. German Bundestag. 2007. “Gesetz zur Schätzung des landwirtschaftlichen Kulturbodens: Bodenschätzungsgesetz - BodSchätzG.” Bundesgesetzblatt I 2007:3176–3184. Gluszak, M., and R. Zygmunt. 2018. “Development density, administrative decisions, and land values: An empirical investigation.” Land Use Policy 70:153–161. 34 Graubner, M. 2018. “Lost in space? The effect of direct payments on land rental prices.” European Review of Agricultural Economics 45:143–171. Haan, P., and M. Simmler. 2018. “Wind electricity subsidies — A windfall for landowners? Evidence from a feed-in tariff in Germany.” Journal of Public Economics 159:16–32. Harding, J.P., J.R. Knight, and C.F. Sirmans. 2003. “Estimating Bargaining Effects in Hedonic Models: Evidence from the Housing Market.” Real Estate Economics 31:601–622. Hennig, S., and U. Latacz-Lohmann. 2016. “The incidence of biogas feed-in tariffs on farmland rental rates – evidence from northern Germany.” European Review of Agricultural Economics 13:221. Hüttel, S., L. Wildermann, and C. Croonenbroeck. 2016. “How Do Institutional Market Players Matter in Farmland Pricing?” Land Use Policy 59:154–167. Kaplan, G., G. Menzio, L. Rudanko, and N. Trachter. 2019. “Relative Price Dispersion: Evidence and Theory.” American Economic Journal: Microeconomics 11(3):68–124. King, D.A., and J.A. Sinden. 1994. “Price Formation in Farm Land Markets.” Land Economics 70:38–52. Kolbe, J., R. Schulz, M. Wersing, and A. Werwatz. 2015. “Identifying Berlin’s Land Value Map Using Adaptive Weights Smoothing.” Computational Statistics 30:767–790. Kostov, P. 2010. “Do Buyers’ Characteristics and Personal Relationships Affect Agricultural Land Prices?” Land Economics 86:48–65. Kuethe, T.H., and D.P. Bigelow. 2018. “Bargaining Power in Farmland Rental Markets: Bargaining Power in Farmland Rental Markets.” Selected Paper prepared for presentation at the 2018 Agricultural and Applied Economics Association Annual Meeting, Washington, D.C., 5–7 August. Kumbhakar, S.C., and C.F. Parmeter. 2009. “The effects of match uncertainty and bargaining on labor market outcomes: Evidence from firm and worker specific estimates.” Journal of Productivity Analysis 31:1–14. 35 —. 2010. “Estimation of hedonic price functions with incomplete information.” Empirical Economics 39:1–25. Lawley, C. 2018. “Ownership Restrictions and Farmland Values: Evidence from the 2003 Saskatchewan Farm Security Act Amendment.” American Journal of Agricultural Economics 100:311–337. Lehn, F., and E. Bahrs. 2018. “Analysis of factors influencing standard farmland values with regard to stronger interventions in the German farmland market.” Land Use Policy 73:138–146. LVermGeo. 2018a. Auskunft aus der Kaufpreissammlung: Bereitstellung für wissenschaftliche Zwecke 17.05.2018. Magdeburg: Landesamt für Vermessung und Geoinformation Sachsen-Anhalt. —. 2018b. Lageunterschiede. Magdeburg: Landesamt für Vermessung und Geoinformation Sachsen-Anhalt. —. 2018c. Windvorranggebiete. Magdeburg: Landesamt für Vermessung und Geoinformation Sachsen-Anhalt. Maddison, D. 2000. “A Hedonic Analysis of Agricultural Land Prices in England and Wales.” European Review of Agricultural Economics 27:519–532. —. 2009. “A Spatio-Temporal Model of Farmland Values.” Journal of Agricultural Economics 60:171–189. Magnan, A., and S. Sunley. 2017. “Farmland investment and financialization in Saskatchewan, 2003–2014: An empirical analysis of farmland transactions.” Journal of Rural Studies 49:92–103. Margarian, A. 2010. “Coordination and Differentiation of Strategies: The Impact on Farm Growth of Strategic Interaction on the Rental Market for Land.” German Journal of Agricultural Economics 59:202–216. Nickerson, C.J., and W. Zhang. 2014. “Modeling the Determinants of Farmland Values in the United States.” In J. M. Duke and J. Wu, eds. The Oxford Handbook of Land 36 Economics. New York: Oxford University Press, pp. 111–138. Parmeter, C.F. 2018. “Estimation of the two-tiered stochastic frontier model with the scaling property.” Journal of Productivity Analysis 49:37–47. Plantinga, A.J., and D.J. Miller. 2001. “Agricultural Land Values and the Value of Rights to Future Land Development.” Land Economics 77:56–67. Polachek, S.W., and B.J. Yoon. 1987. “A Two-Tiered Earnings Frontier Estimation of Employer and Employee Information in the Labor Market.” The Review of Economics and Statistics 69:296–302. R Core Team. 2019. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. Robinson, L.J., R.J. Myers, and M.E. Siles. 2002. “Social Capital and the Terms of Trade for Farmland.” Review of Agricultural Economics 24:44–58. Rosen, S. 1974. “Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition.” Journal of Political Economy 82:34–55. Rousseeuw, P.J., and K. van Driessen. 1999. “A fast algorithm for the minimum covariance determinant estimator.” Technometrics 41:212–223. Sheng, Y., T. Jackson, and K. Lawson. 2018. “Evaluating the benefits from transport infrastructure in agriculture: a hedonic analysis of farmland prices.” Australian Journal of Agricultural and Resource Economics 62:237–255. StaLa. 2017. Agrarstrukturerhebung. Teil 3: Landwirtschaftliche Betriebe mit eigener und/oder gepachteter landwirtschaftlich genutzter Fläche 2016. Arbeitskräfte in landwirtschaftlichen Betrieben 2016 nach Art der Beschäftigung, Geschlecht, Rechtsformen sozialökonomischen Betriebstypen, Berufsbildung. Halle (Saale): Statistisches Landesamt Sachsen-Anhalt. —. 2018. “Bodenfläche nach Art der tatsächlichen Nutzung 2017 [Land area by type of use].” Retrieved from https://statistik.sachsen-anhalt.de/themen/gebiet-und-wahlen/ gebiet/tabellen-gebiet/. 37 Towe, C., and C.I. Tra. 2013. “Vegetable Spirits and Energy Policy.” American Journal of Agricultural Economics 95:1–16. Turner, M.A., A. Haughwout, and W. van der Klaaur. 2014. “Land use Regulation and Welfare.” Econometrica 82:1341–1403. Wang, H.J., and P.S. Schmidt. 2002. “One-Step and Two-Step Estimation of the Effects of Exogenous Variables on Technical Efficiency Levels.” Journal of Productivity Analysis 18:129–144. Yiu, C.Y., S.K. Wong, and K.W. Chau. 2009. “Transaction Volume and Price Dispersion in the Presale and Spot Real Estate Markets.” The Journal of Real Estate Finance and Economics 38:241–253. Zeileis, A. 2004. “Econometric Computing with HC and HAC Covariance Matrix Estimators.” Journal of Statistical Software 11(10):1–17. 38