Farmland values and bidder behaviour in first-price land auctions

Abstract

1. Motivation

Land ownership is often transferred using auctions, given the ability of the auction mechanism to allocate land efficiently. Consensus exists that land market auctions yield higher sales prices compared to search markets due to higher transparency and competitive pressure among buyers (e.g. Bulow and Klemperer, 1996; Chow, Hafalir and Yavas, 2015). The auction mechanism appears particularly attractive if land is sold by public purse, and for this reason many Eastern European countries with a history of economic transition and land reforms privatise or redistribute land via first-price sealed bid auctions (Hartvigsen, 2014).¹ Since 1997, first price auctions with public tenders have even become obligatory – in line with European Union (EU) law – to rule out any publicly allotted allowances from the market mechanism if the public hand acts as a seller (Official Journal of the European Union No. C 209).

Maximising revenue from land sales or allotting land to owners with the greatest willingness to pay is not the only objective that land markets should pursue, at least from a policy perspective. Policy makers and other stakeholders are also interested in a ‘sound’ distribution of land property rights, that is, preventing dominant market positions and encouraging a diversity of legal forms and production systems. Indeed, the share of farmland actually owned by farmers is constantly decreasing in most developed economies (e.g. Deininger et al., 2011). Land acquisition by financial investors occurs on a broad scale not only in developing or post-Soviet Eurasian countries (e.g. Visser and Spoor, 2011; Kerven et al., 2016) but also in the global north and in the EU, as van der Ploeg, Franco and Borras (2015) show.

Not surprisingly, a heated policy debate in Europe has emerged regarding whether land markets in their present state can actually cope with these issues in a satisfactory way or whether regulations should be tightened (Kay, Peuch and Franco, 2015). Sometimes the debate about land market regulations, particularly regarding ownership restrictions, is embedded in more general discussions on how sustainable agriculture should be organised (e.g. Brady et al., 2017). Proponents of small-scale, family-based and regional agricultural production are concerned about the possible economic dominance of large-scale, potentially industrialised agricultural structures. To ensure viability of the former, these groups often propose allowing privileged access to land markets or restricting ownership; see, for instance, Lawley (2018), who investigates the relation of farmland values and the revision of The Saskatchewan Farm Security Act (Canada) relaxing ownership restrictions.

In the EU, various measures have been proposed and in some member states, they have already been imposed to achieve the aforementioned objectives. For example, since 2014, Slovakia restricts purchases by foreigners together with giving priority to farmers (Lazíková and Bandlerová, 2015). Others include relieving young farmers’ access to land by facilitating farm succession and start-ups.² At the same time, the restriction of market access for agents treating land as an investment asset without possessing any farming interests (so-called non-agricultural or financial investors) is being debated, even though this is not in line with EU treaties.

Against this background, it is desirable to understand whether using land auctions as specific market institutions facilitates the aforementioned policy goals, or whether the resulting land allocation favours or discriminates against certain farm types in terms of their legal form, their size or the provenance of the bidders. For example, are large-scale industrialised farms more successful at acquiring land from auctions than small family farms, or do foreign investors offer systematically higher bids than domestic farmers?

This begs the question as to why bids and hence valuations of land should systematically differ among bidder groups. Clearly, such differences could be related to the expected income streams generated by farmland investments. One might argue that larger farms generate higher incomes due to economies of scale. In this regard, the advantage of non-fragmented land use was found to be particularly relevant (e.g. Curtiss et al., 2013; Latruffe and Piet, 2014). Likewise, non-agricultural investors may face lower financial constraints compared to farmers, and thus lower costs of capital; these factors are often given as reasons for asymmetric bidder structures in the auction literature (Laffont, Ossard and Vuong, 1995), as are capacity constraints (Jofre-Bonet and Pesendorfer, 2003).

Local farms and farm-investors, however, may profit from better information in particular about the potential income generated from the land. This benefit could possibly lead to a different (higher or lower) valuation for given land characteristics (e.g. soil quality) compared to non-farm investors. Local farmers typically buy land with the intention of using it rather than renting it out, which could involve a higher willingness to pay (WTP), particularly when it comes to auctions for land that is already under their usage, or for a parcel surrounded by own or used parcels without any infrastructure for accessing the land. In the latter case, a better-informed estimate of the transaction costs for disentangling the field and further losses besides using the field area (such as increased production costs) might even increase the WTP of farms compared to investors.

To summarise, informational asymmetries may constitute another source of asymmetric bidders (Hendricks and Porter, 1992; Hendricks, Porter and Wilson, 1994). Besides, collusion between local farmers on auction entry might be possible. Since both parties agree to not bid against the other, this strengthens the local bidder with potential disadvantages for non-locals, including investors. While such bidding cartels have been shown to negatively affect prices, buyer asymmetries could outweigh such effects (Banerji and Meenakshi, 2004); yet to our knowledge, buyer asymmetries have thus far not been explored in land auctions, neither has the empirical validity of the aforementioned arguments. Only sparse empirical literature on the dominance of certain buyer groups in land markets exists. For instance, Curtiss et al. (2013) report higher prices if investors are involved in the transaction, a frequently observed phenomenon in transition economies in which market entry is not highly regulated. Related to that, Hüttel et al. (2013) show for Eastern Germany that realised land prices in auctions are higher when the share of agricultural bidders is low. Yet higher land prices are likewise achieved if a resident buyer wins, which might (in the view of these authors) be traced back to better knowledge of local land development plans and infrastructure, without an in-depth analysis of the consequences induced by informational asymmetries.

Our objective is to analyse whether bidder groups differ systematically in their valuation distributions of the land. Understanding buyer asymmetries in land auctions is relevant for two reasons. First, one can verify whether bidders representing certain farm types are dominant and have the potential to crowd out competitors, which would undermine the aforementioned diversity goal, at least on a local scale. Second, bidder asymmetries may entail market inefficiencies with non-competitive prices (Klemperer, 1999). As such, first price auctions might then not lead to an efficient allocation of the land. That is, the land will not end up in the hand of those with the highest valuation. The reason is that weakness of bidders leads to a more aggressive bidding strategy to compensate for the weakness (Flambard and Perrigne, 2006). This may even prevent other bidders from participating, and the resulting prices will deviate from those that would prevail if all bidders were to submit their true demand curve. In addition, the revenue equivalence principle – stating that under symmetry and other conditions any mechanism gives the same results – no longer holds under asymmetry. Therefore, the question arises, which auction format will maximise sellers’ profits?

To identify potential bidder asymmetries in land auctions we use a comprehensive data set from the major land privatising agency in Eastern Germany (Bodenverwertungs- und -verwaltungs GmbH) over the period 2007–2015. We analyse these auction data using a structural estimation approach. The structural estimation of auction data (SEAD) relies on the idea that the observed bids are the equilibrium bids of the auction model considered, in our case a first-price sealed-bid auction (cf. Hong and Shum, 2000; Paarsch and Hong, 2006). Compared to reduced-form models, the SEAD approach shows several advantages. First and foremost, it allows recovering the (unobserved) distribution of the bidders’ valuations from the observed distribution of bids. That is, the aim is to investigate the data-generating process of bids directly to estimate the characteristics of the market (Hendricks and Porter, 2007). In contrast, reduced-form models can only provide an average mark-up in bids possibly attributed to a specific bidder group. Reduced form models would only allow testing the implications from the theory, where we would have to acknowledge that other behavioural models would provide the same reduced-form prediction. This procedure would not be sufficient for our research question. Another advantage of SEAD is that the theoretical auction model imposes restrictions that can be tested, such as monotonously increasing bid functions or a markdown of valuations in first-price sealed-bid auctions. Hence, theoretical results serve as a basis for testing the validity of the auction model under consideration (Athey and Haile, 2007). To the best of our knowledge, this is the first application of the SEAD approach to land auctions.

In our empirical analysis, we explore asymmetries within three pairs of bidder groups in land auctions in Eastern Germany. We classify the groups by their legal form, their tenancy status and the involvement of foreign investors, respectively. Regarding legal status, we differentiate between natural persons (single farms, private partnerships) and legal entities (cooperatives, limited liability companies and joint stock companies). The distinction between natural persons and legal entities is more than a formal one. While corporations and cooperatives are in most cases direct successors of the formerly agricultural cooperatives of the German Democratic Republic,³ newly established and re-established family farms were typically founded as natural persons. In a simplistic manner, these two groups represent opposing principles of organising agricultural production. Legal entities stand for industrialised agriculture; they are, on average, much larger than single farms, rely exclusively on workers via contracting and operate with a larger share of leased land. In contrast, natural persons are smaller, centred on a family (a household) and typically strive to transfer the farm to the next generation. The second distinction refers to former tenants and other bidders, and is motivated by informational gains that the former group may have when bidding for land. Practical experience that former tenants may have gained facilitates a realistic and more accurate valuation of the tendered land plots. It is, however, unclear whether bids between former tenants and non-tenants systematically differ. The third classification aims at identifying buyers who plan to rent out land instead of actively using it for agricultural production. Although we cannot clearly measure the motivations that drive bidders in land auctions, it seems very likely that land is considered as a financial asset if foreigners are involved in the buyer consortium. In this view, foreign bidders fall into the heterogeneous class of financial investors and it is therefore interesting to test whether their willingness to pay for land is different from that of domestic farmers.⁴ Our findings reveal statistically different valuations by buyer groups, but do not give evidence on strong asymmetric structures and thus challenge the hypothesis that land privatisation via auctions discriminates specific buyer groups. On these grounds, we cannot support the justification of stricter regulation of agricultural land markets because of bidder asymmetries.

The remainder of the article is structured as follows: we start by presenting the theoretical background; that is, the independent private value auction model, from which we derive optimal bidding strategies for asymmetric bidders. Next, we explain the empirical identification strategy of the SEAD approach. In Section 3, we describe the data and provide background information about the land market under consideration. In Section 4, we detail our empirical strategy and explain the non-parametric structural estimation. In Section 5, we discuss our empirical results and in the final section, we conclude.

2. Theoretical framework

In the land market under study, the Bodenverwertungs- und-verwaltungs GmbH (BVVG, see Section 3 for details), privatises the former state-owned land in Eastern Germany by means of first-price sealed-bid auctions with public tenders without binding and without reported reservation prices. To model optimal bidding strategies in first-price auctions under potential asymmetries we make use of the independent private value (IPV) paradigm. Bidders’ expected development of future income streams, from either using the land or renting it out, depends on expected returns from farming but also on individual knowledge, ability, bidders’ overall net income flow, wealth status, as well as their time and risk preferences. Private land values can thus be justified because they are based on private information. This equally holds for potential costs of not winning the auction, which also remains private. Nonetheless, relying on the IPV paradigm as the adequate informational structure of land auctions can be questioned. If land (with similar attributes) is rented out or resold for the same price after the auction, this would rather correspond to a common value environment in which the value of the auctioned item is the same (but unknown) for all bidders. However, we argue that even if land is considered as a financial investment and is rented out after purchase, the return is not the same for all investors. The information on expected returns depends on potential tenants’ behaviour, in particular on their expectations, both of which remain private for the bidders. In a recent analysis of the US land rental market, Kuethe and Bigelow (2018) show that land rental contracts (ceteris paribus) differ widely among market participants. A main reason is the thinness of local land markets that enables considerable bargaining power, which, in turn, is determined by the characteristics of land owners and tenants (e.g. Harding, Rosenthal and Sirmans, 2003). Moreover, land investments are typically long-term investments made with the intention to increase the production base or to securely store and increase wealth by realising capital gains through renting it out to farmers (e.g. Magnan and Sunley, 2017, for evidence in Saskatchewan, Canada). This argument is supported by the low overall mobility of farmland in Germany, where 0.7 per cent of the total utilised agricultural area was transacted in 2015 (Destatis, 2015). Although we cannot rule out the premise that land markets might obey a common component or affiliated values, we proceed with the assumptions of the IPV model.⁵

2.1. Independent private value model with asymmetric bidders

In the independent private value setting, $n$ bidders are assumed to be risk-neutral and compete for a plot of land with the goal of maximising utility, denoted by $Ui=U(vi)$⁠.⁶ This function is assumed to increase in $vi$⁠, with $vi$ denoting bidder $i$’s private information on the valuation of the land. Under independent private values, environment uncertainty arises from the fact that the value for competing bidders is unknown. That is, the other bidders’ valuation remains uninformative for the bidder and thus utility reduces to $Ui(vi)=vi$⁠. Each bidder is assumed to have an individual expectation of the value of the land, $V$⁠, which can be interpreted as the present value of future returns from utilising the land, or renting it out as would be the case for investors. The exact present value is typically unknown and hence $V$ is modelled as a random variable, with realisations $vi$ reflecting the valuations of the bidders. That is, bidders draw their private values independently from the common distribution function, which reduces to $n$ dimensions in $FV(v)$ and density $fV(v).$ For further details, we refer to Milgrom and Weber (1982).

The bidders are further assumed to estimate the relationship between their bid and the probability of winning the auction, and to behave incentive compatible (bidder rationality). Since BVVG proceeds on non-binding and non-reported reserve prices, which means that auction participants will behave as if there was no reservation price at all, we use $v̲$ to denote the lower bound of the distribution domain.

The second term on the right-hand side of equation (2) is strictly positive and reflects ‘bid shading’, that is, a discount on their actual value that bidders in first-price auctions make to maximise their expected revenues (e.g. Krishna, 2009: 16). This theoretical property is useful to check the consistency of our estimation.

While the symmetry assumption seems to constitute a reasonable starting point, we argue that the competitors among land within BVVG auctions are not ex ante identical per se. We thus follow up on a more general model setup that allows us to acknowledge the asymmetries among bidders. Symbol $nj$ now denotes the number of potential bidders by type, with $j∈{0,1}$ and $n=n0+n1$ indicating the respective bidder groups: first, legal entities versus private persons including partnerships; second, tenants versus non-tenants and third, domestic buyers versus buyer groups with foreigner participation.

Based on these equilibrium strategies, we expect bid shading to differ systematically by bidder group. More specifically, ’weaker’ bidders are expected to bid more aggressively, that is, closer to their valuation, to compensate for their disadvantage compared to ‘stronger’ bidders. Note that this system of differential equations is complex and intractable for simulation-based empirical investigation. Thus, we follow the indirect structural empirical approach introduced by Guerre, Perrigne and Vuong (2000), which relies on non-parametric estimation techniques. Applying this empirical approach has the advantage that the equilibrium strategy does not need to be computed, but must rely on a sound identification strategy.

2.2. Non-parametric identification of the asymmetric IPV model

Identification in the estimation of auction models can be reduced to the question of whether, for a given equilibrium, a bijective relationship exists between the unobserved distributions of the bidders’ valuations and the observed bids (Hendricks and Porter, 2007). That is, the question is whether $F1(⋅)$⁠, $F0(⋅)$⁠, $n0$ and $n1$ can be identified from the observed bids and the number of actual bidders. The core assumption on which non-parametric identification relies is that the observed winning bids are assumed to be the equilibrium bids, and the number of potential bidders is equal to the number of actual bidders.

Herein, $ξ1$ and $ξ0$ are the inverse bid functions that map bids into values. Since bids, bid distribution, bid densities, the number and the identity of bidders are observable, equation (4) enables the identification of latent private values. In fact, Laffont and Vuong (1996) show that the asymmetric IPV model is identified if the functions $ξ1(si)$ and $ξ0(si)$ are strictly increasing in the bids $(si)$⁠. The intuition is that $F(⋅)$ is identified from $G(⋅)$ since observed bids are linked to private values by a strictly increasing equilibrium strategy.

For empirical identification of the asymmetric IPV model we refer to Athey and Haile (2002: 2116) who assert that identifying the asymmetric IPV model is possible from a single bid as long as the identity of the bidder is observed in first-price auctions. The core idea is to consider bids as order statistics, then the winning bid can be understood as the highest-order statistic. Since we observe the transaction price as well as the legal form of the winning bidder, the status as former tenant as well as bidders’ origin (domestic or not), our data set provides sufficient information to empirically identify the auction model and to assess potential asymmetries among those bidder groups. We detail these steps after the data description in Section 4.

3. Background, data and preliminary assessment of asymmetries

3.1. Background and Data

In 1992, BVVG was founded with the target of privatising the former state-owned agricultural and forestland in Eastern Germany on behalf of the Federal Ministry of Finance. BVVG privatises according to the Privatisation Principles in Germany using first-price sealed-bid auctions with public tenders, as well as special transactions for reparation reasons. BVVG auctions make up a considerable local market share in Eastern Germany, with regionally observed shares of up to 60 per cent. Not surprisingly, BVVG as a notable seller has been under debate to contribute to price increases because higher prices can be achieved in auctions compared to those from negotiated sales (e.g. Hüttel, Wildermann and Croonenbroeck, 2016). BVVG’s insinuated target is to maximise revenues of the public household rather than supporting farms and public tenders have been negatively judged to ease investors’ market entry, particularly foreigners (e.g. Wolz, 2013). In summary, the question remains whether this market type discriminates against certain buyer groups in the sense that they are forced out of the market or dominated by other bidder types. We will explore this empirically, where our data set covers all BVVG auctions from 3 January 2007 to 17 July 2015 for all Eastern Federal States of Germany. For each transaction the winning bid, the total number of bidders, the exact date of the auction, the location, plot-specific land characteristics including a soil quality index⁷, lot size, number of parcels and type of usage are available. Most important for our research question on the analysis of bidder asymmetries, the data set contains bidder characteristics that allow us to differentiate by legal status, whether the buyer was a former tenant (not necessarily on the same plot) and whether a foreigner was involved in the buyer consortium.

We exclude transactions where churches, municipalities and others such as non-profit organisations were involved since these are uninformative for our questions (in sum, 455 transactions were excluded in this step), while tenders with only forest, recreation and areas for natural reserve were already excluded by BVVG. This data selection does not entail a bias: the shares of the bidder types remain in the same range before and after data cleaning. For readability reasons, we present further details in the A ppendix, Table A1. From the remaining 12,250 transactions, we further exclude incomplete observations, those that BVVG identified as outliers, and those with only minor holdings or building land (arable and grassland amounts to zero). Finally, we exclude auctions with only one bidder (2,525), because there is virtually no competition under the assumption that the number of actual bidders equals the number of potential bidders. Moreover, we are interested in investigating potential asymmetries and it is evident from equation (4) that the notion of asymmetry does not make sense in the case of only one bidder. Summary statistics (cf.  Table A2 in the Appendix for details) support our argumentation: winning bids for auctions with more than one bidder show a mean of 1.18 Euros per square metre, while this value amounts only to 0.796 Euros for auctions with one bidder. The final sample contains 9,684 observations with at least two bidders that are rather uniformly distributed by years (cf.  Table A3 in the Appendix for details). Relating these auctions to reported market volumes, the analysed auctions amount to roughly 20 per cent of the traded land in Eastern Germany over the period studied.

To investigate potential asymmetries, we classify bidder groups by their legal form, their tenancy status and the involvement of foreign investors, respectively. Regarding legal status, we differentiate between natural persons (single farms, private partnerships) and legal entities (cooperatives, limited liability companies and joint stock companies). Distinguishing by legal status allows us to investigate whether potential bidders with legal forms typically chosen by newly and re-established farms (mainly privately organised) value land differently than potential bidders from the group of legal entities and cooperatives, typical forms of (larger) successor-farms. The latter group typically operates at a larger size and employs mainly external workers. Legal entities may benefit from economies of scale and scope, have greater financial resources and may thus appear stronger. In contrast, farms in the group of natural persons typically run at a smaller size, rely mainly on family labour and are often locally centred on a family or a household. In the sample, cooperatives win in 9 per cent of the transactions considered, incorporated enterprises win in 22 per cent of the transactions and civil law associations together with single firms are victorious in 68 per cent of the transactions (cf. Appendix  Table A1). In terms of auctioned area, the relation changes slightly: corporations and cooperatives buy 34 per cent of the land, while single firms and partnerships buy 64 per cent. In relation to their occurrence, legal entities seem to be overrepresented. However, given that legal entities operate approximately 50 per cent of the farmland in Eastern German, we cannot conclude that they dominate BVVG auctions.

On average, winning bids of legal entities are 3 per cent higher compared to natural persons (cf. Table 1). Interestingly, legal entities seem to self-select into auctions with larger plots (9.498 versus 7.566 ha) but also into auctions with a higher number of parcels (8 versus 5 at the mean) with significant differences based on the Wilcoxon rank sum test (p-value 0.000). Comparing the means, the 10 per cent and 90 per cent quantiles of the lot size by legal status reveals larger lots for legal entities by roughly 25, 95 and 23 per cent, respectively. The number of bids is around four in both groups while the average distance between bidder and plot is larger in the group of natural persons, which seems to contradict the idea that natural persons mainly buy locally.

The second classification is between former tenants of the BVVG and non-tenants. We conjecture that former tenants can rely on previous experience when estimating returns and thus the value of the auctioned land. These tenants may have an advantage in assessing transaction costs that may come along with the acquisition of the tendered land plot. These costs may involve mapping the auctioned plot from a potentially larger lot under usage, but also to ensure access, some infrastructure (paths, roads) might need to be (re-)established. Furthermore, former tenants may benefit from learning gains by repeated transactions with BVVG. In view of these informational advantages, former tenants may be considered the stronger group compared to non-tenants. Former tenants win 27 per cent of the auctions, which corresponds to a share of 30 per cent of the auctioned land in the sample (we refer to  Table A1 in the Appendix for further details). Table 2 shows that average winning bids are significantly lower for tenants (1.131 Euros per square metre) than for non-tenants (1.201 Euros per square metre). In addition, the 10 per cent and 90 per cent quantile is lower (0.377 versus 0.38 and 2.279 versus 2.445). Size, the number of parcels and the share of arable land are significantly different between tenants and non-tenants, whereas average soil quality is similar.

Table 3 compares domestic and foreign bidders. A priori, it is unclear if one of these groups is dominant and if so, which one. On the one hand, foreign investors may have an advantage in financial ability. On the other hand, domestic buyers (including locals, farmers and former tenants) may have informational advantages. Our sample contains 99 transactions with international bidders (1 per cent of the cases and the land auctioned). We cannot reject the null hypothesis of no price difference by provenance. Foreign bidders buy larger lots on average, which makes economic sense. This self-selection raises the question of whether we face an endogenous entry problem in the land auction or not. The implications are important, because in the presence of endogenous entry, the number of potential bidders is larger than the number of actual bidders. This would violate our assumption that we can approximate the (unobserved) number of potential bidders by the (observed) number of actual bidders. In fact, if bidders face a sunk fixed cost before they get a signal about the land value, then it would follow that the selection of bidders in specific plots constitutes an endogenous entry problem. However, the situation is different here. Foreign investors most likely have an idea about the value of the land and they simply decide not to participate in auctions of small parcels because transaction costs for bidding, such as providing a proof of finance and time submitting the bid, notary expenses and finding a tenant exceed the value in these cases. Hence, these investors will not be potential bidders at all and since the structure of transaction costs in BVVG auctions is transparent, competitors can anticipate this decision and adjust their bids accordingly.

3.2. Preliminary assessment of asymmetries

Asymmetry among bidder groups implies differences in the empirical distribution functions of the winning bids (given in Euros per square metre). Thus, we proceed in two steps: first, we compare estimates of the cumulative distribution functions of the winning bids and second, we test for differences in reduced-form hedonic estimates following Flambard and Perrigne (2006).

Inspecting panel (a) of Figure 1 shows that the cumulative distribution functions (CDFs) of natural persons and legal entities are quite similar. Only in the range between 1 and 2 Euros per square metre the CDF of natural persons exceeds the CDF of legal entities to some extent. Similarly, as apparent from panel (b) of Figure 1, almost no differences can be observed in the lower range of the winning bids between tenants and non-tenants. In the upper range, the tenants’ distributions is slightly shifted upwards compared to non-tenants. A similar finding applies to the comparison of domestic and foreign bidders: the CDF of German buyers is located below the CDF of foreign buyers between 2 and 3 Euros per square metre.

Fig. 1.

Cumulative distribution functions (CDF) of the winning bids in €/m² by (a) legal form, (b) tenancy status and (c) foreigner participation.

Open in new tabDownload slide

When comparing CDFs, an important special situation is first-order stochastic dominance: for each probability level, the winning bid of the stronger group is equal or exceeds the winning bid of the weaker group (with at least one strong inequality). In this case, the stronger bidders are more likely to win the auction over the full range of values. Since the presence of stochastic dominance of these empirical distributions cannot be verified by visual inspection, we consider Dunn’s test (1964). Based on this test we reject the null hypotheses of no first-order stochastic dominance for all three pairs of bidder groups with p-values of about 0.000 (legal status), 0.003 (tenancy status) and 0.045 (foreign bidders), respectively. The outcome of this test, however, must be interpreted with prudence, since it rests on the assumption that the (true) distributions do not intersect.⁹

Since winning bids relate to hedonic lot characteristics, in a second step within this preliminary assessment we investigate potentially different relations of the hedonic variables to the winning bids by each of the groups. Therefore, we use an auxiliary reduced form model and regress the log of winning bids on plot and auction characteristics. We consider lot size, soil quality, number of parcels, distance lot-buyer, number of bidders and time dummy variables. We run these auxiliary regressions for each bidder type except the distinction by foreigner participation because of the low number of observations. A Chow test reveals that the estimated coefficients for number of bids, distance and soil quality differ significantly by natural persons and legal entities. The same holds for the periods 2007, 2012, 2013 and 2015. Quality and dummy variables for 2008, 2013 and 2014 differ significantly when distinguishing between former tenants and non-tenants (cf. Table S1 in the Appendix in supplementary data at ERAE online).

Taken together, regarding the winning bids’ distributions, the potentially stronger legal entities and the potentially weaker non-tenants and domestic buyers seem to dominate (cf. panels (a–c) in Figure 1). However, given the strong heterogeneity of the tendered land plots and the considerable price impact of land characteristics on winning bids, differences among the respective winning bid’s distributions must be interpreted with caution. This assessment underlines the necessity of the structural investigation whether relevant asymmetries in valuations exist.

4. Estimation procedure

The SEAD approach uses the structure of the auction model as presented in Section 2 to map the theoretical equilibrium bid functions and their distribution into econometric models. Hence, the probability law of the valuations of potential bidders, that is, the primitives of the underlying theoretical auction model, can be identified. The non-parametric approach proposed by Guerre, Perrigne and Vuong (2000) evaluates the sample analogue of equation (4). The estimation proceeds in two steps: first, estimate the distributions $G(⋅)$ and densities $g(⋅)$ non-parametrically using kernel estimation. Second, based on equation (4), calculate pseudo valuations $vˆi$ and estimate the empirical distributions $Fˆ(v)$ from these pseudo values. This approach, al though data-intensive, exhibits two main advantages: it is computationally simple as no explicit expression of the optimal bid function is required. Moreover, it avoids parametric assumptions about $F$⁠, which is convenient because theory offers little guidance on the functional form of the distributions of the valuations. In our application, we have to take into account three peculiarities that require some modifications of the aforementioned procedure, namely missing information on the composition of bidders types in an individual auction, incomplete bid data and heterogeneity of the sold land plots. In what follows, we discuss how we deal with these issues.

Our data set covers a large geographical region (all Eastern German federal states) and reveals heterogeneous patterns, where these might be due to differences in local infrastructure or farming structure. Such spatio-temporal effects will not be captured by variables describing lot characteristics or the number of bids, and must be disentangled from effects attributed to differences valuations that may also locally differ. To keep the specification $Γ(⋅)$ parsimonious, we acknowledge this heterogeneity by time dummy variables interacted with region dummy variables at the NUTS¹⁰ one level (i.e. the five Eastern Federal States of Germany) in addition to $zt$ in $Γ(⋅)$ and $γ$⁠.

Calculation of valuations via equation (4) requires information on the composition of bidder groups in an auction, that is, the number of potential bidders in each group $n0$⁠. and $n1$⁠. Moreover, a sufficiently large number of auctions with the same absolute number of bidders is required for a reliable non-parametric estimation of the bid distributions and densities. As Tables 1–3 illustrate, the number of bidders, and thus the competitive pressure, varies considerably between auctions, with an average value of 4. We account for this variability similarly to the heterogeneity of land characteristics. That is, we include the number of bids in the hedonic regression equation (7) and calculate homogenised bids for an auction with an average number of observed bids (4). The subsequent SEAD estimation is based on the simplifying assumption that the number of actual bidders equals the number of potential bidders and that the relative size of bidder groups is the same for all auctions. Estimating the relative size by the share of won auctions leads to $n0=3$ for the group of natural persons, non-tenants, and domestic buyers, respectively, while $n1=1$ for legal entities, former tenants and foreign buyers, respectively.¹¹

These calculations are carried out for each pair of bidder groups, that is, legal versus natural persons, tenants versus non-tenants and German versus foreign buyers. We finally compare the densities by buyer groups and test whether different bidder groups have different distributions of their valuations by pairwise two-sample Kolmogorov–Smirnov tests (e.g. Marsaglia, Tsang and Wang, 2003; Chernomaz and Yoshimoto, 2017) and Dunn’s test for investigating the special situation of stochastic dominance of one value distribution over the other.

5. Results

Before we turn to the results of the SEAD model, that is, the bidder-specific distributions of land valuations, we start by presenting the results of the homogenisation step used to remove heterogeneity in the tendered land plots according to equation (6). The adjusted R² of 0.616 documents that the specified hedonic regression with regional time trends captures the overall variability in winning bids fairly well (for details, we refer to  Table A4 in the Appendix). The signs of the estimated coefficients are plausible and in line with other empirical studies dealing with the impact of land attributes on land prices (e.g. Nickerson and Zhang, 2014; Lehn and Bahrs, 2018). As expected, land quality, the share of arable land and plot size have a positive effect on the transaction price. Size matters, although with decreasing impact (positive linear and negative quadratic coefficient), while the number of plots has a negative impact (cf. Maddison, 2000). The positive coefficient of the number of bidders is in line with theoretical predictions for the IPV auction model. The regionally interacted year dummies reflect the steady increase of the general land price level observed in Germany between 2008 and 2015 with regional differences. Residual diagnostics reveal no differences in the variance when comparing each of the group-pairs, which supports our homogenisation procedure.

Turning to the SEAD model, the estimated value distributions $FˆVj(vi)$ for each pair of the buyer groups are depicted in Figure 2 and the summary statistics of the pseudo values in Table 4. For readability reasons, we refer to the  Appendix for details: the density plots (Appendix,  Figure A1), as well as the CDFs of homogeneous winning bids (Appendix,  Figure A2). To verify the consistency of our results, we first compare the distributions of winning bids as presented in Figure 1 and the distributions of the homogenised winning bids as shown in the Appendix,  Figure A2 to visualise the effect of the bid adjustment equation (7) in the homogenisation step. The CDFs of homogenised bids are steeper and show less variability since they refer to a hypothetical land plot with average values of attributes. Next, we compare CDFs of homogenised winning bids and bidders’ maximum valuations (Appendix,  Figure A3). Bid shading implies a left shift of the bid distribution and thus, value distributions should stochastically dominate bid distribution. Based on a two-sample Kolmogorov–Smirnov and Dunn’s test, we reject the null of no difference and no stochastic dominance between these distributions, respectively. From this, we conclude that our data support bid shading for each bidder type (all p-values are 0.000).

Figure 2.

CDF of valuations in €/m² by (a) legal form, (b) tenancy status and (c) foreigner participation.

Open in new tabDownload slide

To investigate the bidder asymmetries in detail, we first contrast the value distributions by legal forms of bidders. Figure 2a shows that the value distribution of legal entities has somewhat larger quantiles compared with that one of natural persons over a large range of values. A two-sample Kolmogorov–Smirnov test reveals that the value distributions differ significantly at any level (p-value: 0.000) and based on Dunn’s test, we reject the null of no stochastic dominance. The difference between the means of the CDFs, however, is rather modest with a mark-up of about 7.38 per cent at the mean (cf. Table 4), which amounts to 770 Euros per hectare. This finding may indicate that legal entities in Eastern Germany yield higher marginal returns from utilising land compared to natural persons. Higher returns may stem from benefits from size (economies of scale) since legal entities including cooperatives are, on average, considerably larger than private (mainly family) farms. In 2014/15, the average size of legal entities was about 1,125 ha in Eastern Germany, while private farms classified as larger according to Federal Ministry statistics used an average of 178 ha (entire Germany). Moreover, legal entities may face lower financial restrictions, which might also relate to the opportunities of investors’ participation. However, single farms and private persons’ partnerships draw from distributions with slightly lower valuations, and this group wins 74 per cent of the auctions (64 per cent of the auctioned land). To compensate for such disadvantages, natural persons bid more aggressively, on average. Table 4 compares homogenised winning bids and valuations for both groups and shows that the discount on mean valuations amounts to 38 per cent (1.120–0.811)/0.811) for legal entities, but only 24 per cent for natural persons.

The estimated distribution functions of the valuations for former tenants and non-tenants are displayed in Figure 2b. Both CDFs appear strikingly similar over a broad range of value apart from the lower left and upper right tail. Still, a two-sample Kolmogorov–Smirnov rejects the null of equal distributions (p-value: 0.000) and Dunn’s test rejects the null hypothesis of no dominance (p-value 0.000). Thus, tenants seem to have a slightly higher valuation of land than non-tenants with a mark-up of about 4.72 per cent, that is, 497 Euros per hectare evaluated at the mean (cf. Table 4). However, former tenants win in 28 per cent of the cases (30 per cent of the auctioned land in hectares), while non-tenants show bids closer to their valuation. Their discount comes to 23 per cent while former tenants bid almost 30 per cent below the valuations for the homogenised lot.

Finally, we analyse whether bidder groups with foreign participation have a different valuation of land compared to domestic bidders. The often alleged preponderance of foreign (financial) investors is not apparent from visual inspection of Figure 2c, except for the upper right tail. However, a two-sample Kolmogorov–Smirnov rejects the null of equal distributions (p-value: 0.000) and Dunn’s test rejects the null hypothesis of no dominance. The mark-up amounts to 5.6 per cent for foreign buyers. This difference in mean valuations by about 586 Euros per hectare is rather moderate in light of the heated debate. In addition, bid shading is similar: domestic buyers discount their mean valuations by 30.62 per cent while foreign buyers reduce valuations by 35.8 per cent (cf. Table 4). These results suggest that potential advantages of foreign bidders (i.e. lower financial restrictions, lower cost of capital and benefits from risk diversification by including land into their financial portfolios) are offset by other factors, such as informational disadvantages or the fact that returns from using land have to be shared with tenants. Clearly, these results must be interpreted with care. First, the number of auctions won by foreigners is low (99 cases) and this weakens the reliability of the non-parametric estimate of CDFs for bids and valuations. Second, the composition of the group of domestic buyers is diverse. For example, domestic bidders do not represent solely farm-owners, but can also involve financial investors. Finally, recall from Section 3.1 that foreign buyers self-select into auctions particularly suitable for investors (minimum size to ease renting it out). That is, in practice, foreign and domestic buyers bid for different items rather than competing for a hypothetical land plot with average attributes.

6. Concluding remarks

This study was motivated by the question of whether auctions facilitate policy goals such as a ‘sound’ distribution of land property rights, a prevention of dominant market positions and a diversity of legal forms, ownership and production systems. The current debate across Europe and beyond questions that land markets and, in particular auctions, can cope with these issues in a satisfactory way and proposes tightening market regulations. Most prominently discussed is restricted market access for agents treating land as an investment asset without farming interests, so-called non-agricultural or financial investors. Likewise, privileged access to land markets has been proposed for small family farms, particularly younger farmers. Using a unique and representative data set, we empirically analyse the outcomes of agricultural land auctions in Eastern Germany and investigate the behaviour and the performance of specific bidder groups, which we differentiate regarding their legal status, their tenancy status and their nationality. We apply a structural econometric auction model to estimate the distribution of latent land values for these bidder groups, and to test for differences in their land valuations between bidder groups. Value distributions are derived from the Bayes–Nash equilibrium rationale and can be estimated using non-parametric techniques.

Although earlier studies report that higher transaction prices for agricultural land are realised in auctions compared to search markets, we cannot find empirical evidence for the claim that this market type discriminates against certain buyer groups in the sense that they are forced out of the market or dominated by other bidder types. The 9,684 BVVG land auctions that we analyse are won by a diversity of legal forms, as well as former tenants and non-tenants. Only a small fraction of tendered land plots has been sold to foreign bidders. The distributions of winning bids, after controlling for heterogeneity in land amenities, are quite similar for all pairs of bidder types. Also, the underlying distributions of land valuations, from which bids are derived, do not show pronounced differences. Foreign investors have a different valuation of land than other participants but only at a smaller magnitude. This supports the argumentation that foreign investors’ expectations about potential returns from investing in land is similar to other bidders. We also find slightly higher private values of land for legal entities, which most likely reflect the economic strength of large-scale farms compared to small-scale farms (e.g. Lissitsa and Odening, 2005; Hansson, 2008; Woodhouse, 2010; Bojnec and Fertő, 2013). Nonetheless, ascertained differences in the valuation of land between natural persons and legal entities remain modest. Overall, our findings support the argumentation that increasing land prices are rather the result of fierce competition among all bidder groups, including operating farmers and domestic investors. Our results are relevant for the current policy debate on land market regulations. In many EU member states, policy-makers are concerned that the acquisition of land by non-locals and non-farmers crowds family farmers out of land markets. Our results do not support this hypothesis, at least for the considered market segment. An apparent implication is that a tightening of existing market regulations cannot be justified by this argument. While many concerns about land acquisition by large holdings or foreign investors are based on case studies or anecdotal evidence, our empirical study refers to a comprehensive data set that covers a major segment of the agricultural land market in Eastern Germany. Our analysis does not reveal reasons for not using auctions as an instrument for selling or privatising land. Thus, proposed policy interventions into land markets must be justified by other arguments, such as local market power or social preferences regarding certain farm types.

From a methodological perspective, this study contributes to the understanding of price formation in land auctions. The micro-structural approach offers the advantage of analysing the distribution of latent valuations behind observed bidding behaviour and investigating potential asymmetries among bidder groups. Identification of the model, however, relies on the independent private value paradigm. While we provide arguments for the plausibility of this assumption, a more formal testing of this assumption against the alternative of common values would be desirable.

Regarding the validity of the implications of our empirical results, we have to acknowledge that we consider only one segment of the German land market. For example, land is also auctioned by state-owned land trusts other than BVVG. These land trusts offer preferential conditions for local farmers and former tenants and it cannot be ruled out that certain bidder types select themselves into particular auction formats. Moreover, land acquisition not only takes place via land markets but also through the purchase of entire farms or shares of agricultural cooperatives and joint stock companies (so called ‘share deals’). This kind of transaction, which we disregard in our analysis, is typically linked with financial investors. Finally, the extent to which our empirical findings apply to agricultural land markets in other regions with different institutional settings and regulatory frameworks is an empirical question that we propose for further research.

Supplementary data

Supplementary data are available at European Review of Agricultural Economics online.

Acknowledgements

The authors gratefully acknowledge financial support from the Deutsche Forschungsgemeinschaft (DFG) through Research Unit 2596 ‘Agricultural Land Markets – Efficiency and Regulation’. (http://gepris.dfg.de/gepris/projekt/317374551). We further thank the Bodenverwertungs- und -verwaltungs GmbH (BVVG) for providing the data.

Footnotes

Review coordinated by Carl Johan Lagerkvist

References