A microeconometric analysis of farmland prices

In recent decades, farmland prices in Germany have risen sharply. This has triggered a controversial public and political debate: some stakeholders attribute the price dynamics to the entry of non-agricultural investors and call for stricter regulation, while others caution that these claims rest on a limited empirical basis and that the proposed interventions may entail significant efficiency losses and societal costs. Against this backdrop, this dissertation uses detailed plot-level transaction data from Brandenburg, eastern Germany, to analyze different aspects of farmland prices and thereby contributes additional empirical evidence to the debate on agricultural land markets. It draws on a broad set of well-developed econometric methods, including linear regression, unconditional quantile regression, nonparametric regression, distributional decomposition methods, causal analysis, and conformal prediction. The first study investigates the particularly strong increases in the upper quantiles of farmland prices. To this end, we use unconditional quantile regressions in combination with propensity score reweighting to decompose price increases into two components: one reflecting changes in the valuation of land and regional characteristics, and the other reflecting changes in the distributions of these characteristics. We find that an increased valuation of soil quality and plot size is responsible for about 25\% of the surge in the upper quantiles of farmland prices. In the second study, we quantify the speculative effect of urban proximity on farmland prices. This speculative effect stems from the possibility that farmland near cities may eventually be converted to more profitable residential or commercial uses. To identify this effect, we construct a data set that combines plot-level transactions with dummy variables indicating plots exposed to speculative pressure. We find robust evidence of a substantial farmland price premium of around 18\% associated with the speculative effect. In the third study, I develop parametric and semiparametric conditional variance models to measure price dispersion for comparable types of farmland. One set of models assumes that farmland prices, conditional on observed characteristics, are lognormally distributed. Another set of models relies on a distribution-free heteroscedastic regression approach that employs a separate regression for the conditional variance. I find that the models assuming lognormality can be interpreted and estimated well and produce accurate prediction intervals for typical farmland transactions.