Spatial interactions in location decisions: Empirical evidence from a Bayesian spatial probit model

In the past few decades spatial econometric models have become a standard tool in empirical research. Nevertheless applications in binary-choice models remain scarce. This paper makes use of Bayesian Spatial Probit Models to model and estimate spatial interactions in location decisions. For this purpose, we focus on the Austrian retail gasoline market, which is going through a process of remarkable structural changes. A short analysis shows that, during the last decade 10.9% of the stations had left the market and a percentage of 29.6% had either left the market or had changed the brand. This paper aims at investigating this process. A special characteristic of this market is the local competition structure which is characterized by spatial dependencies along local competitors. To capture these spatial dependencies and since the dependent variable is binary in nature (an exit had taken place or not), we apply a Bayesian spatial probit model using MCMC estimation on station level data for the whole Austrian retail gasoline market. Our results suggest, that the decision to leave the market, does not only depend on own characteristics, but also on competitors. In particular, we find the exit decisions to exhibit a negative spatial correlation. Moreover, our model allows to quantify spatial spillover effects of this market.