We investigate consumer welfare in transalpine freight transport using micro-data on individual route choice, tackling two core questions. First, to what extent does the way we model unobserved heterogeneity matter for welfare estimates in discrete choice models? Second, what is the loss in consumer surplus per year from shutting down a transalpine road infrastructure such as the Mont Blanc tunnel? Closing this tunnel has been proposed in the political debate following several fatal road accidents in large alpine tunnels. The most severe accident, in the Mont Blanc in 1999, led to a full closure of the tunnel for a period of 3 years. We model route choice as a discrete choice among a number of mutually exclusive alternatives. Due to our rich data, we can exibly model unobserved heterogeneity of decision makers in their valuation of money and time. Decision makers may be heterogenous for a number of reasons. For example, the value of alternatives may depend on the weight or type of commodity a truck is transporting. These are examples for observable heterogeneity which are easily controlled for. However, the value of money and time is also likely to depend on unobservable truck characteristics. These could be en route pick-ups of goods, special logistic needs, or truck drivers' personal tastes that favor one route over another. Modeling such unobserved heterogeneity in the discrete choice framework has been at the heart of research analyzing economic choices during the last two decades. Only recently, researchers have started asking how the way we model unobserved heterogeneity affects policy-relevant measures of consumer welfare. We contribute to this literature by applying a recently proposed flexible nonparametric estimator of unobserved heterogeneity to a random coefficients logit model and investigating the impact of parametric assumptions on a measure of consumer welfare. In a nutshell, the idea of the estimator is to approximate the true underlying taste distribution by a finite grid in the preference space. To our knowledge, we are the first to apply this estimator to real-world data in a static discrete choice model with random coefficients. To identify the underlying structural parameters, we use a large scale individual choice data set from the 2004 Cross-Alpine Freight Transport survey. We exploit exogenous variation in travel cost and time arising from the fixed geographic locations of origin, destination, and alpine crossing points. While endogeneity concerns are less important with individual-level data, we discuss several potential sources of endogeneity bias such as congestion or weather conditions. We find that parametric assumptions and the dimensionality of modeled unobserved heterogeneity have a significant impact on welfare results. Our nonparametric estimates predict economically significantly higher annual losses in user surplus due to the Mont- Blanc tunnel closure. The latter implies a loss of e5.39 millions and the parametric random coefficients logit model a loss of e2.97 millions in specifications where both price and time are assumed to have random coefficients. With one random coefficient, the nonparametric estimate is almost double that of the parametric random coefficients logit estimate, €7.09 millions versus €3.62 millions. Compared to the logit with fixed coefficients and the nonparametric estimates, both parametric random coefficient specifications underestimate the loss in consumer surplus.

Keywords

Discrete Choice, Consumer Surplus, Nonparametric Estimation, Transalpine Freight