Research trying to explain tourism flows and expenditures for different destinations has so far adopted either a tourism-demand or a tourism-supply approach. Whereas on the one hand the former ignores the product specificities (Papatheodorou, 2001), the latter, on the other, fails to take into account the characteristics of the tourist origin markets. In recent years attempts to merge the two views have come from scholars using spatial econometric techniques, i.e. origin-destination models (O-D), which have been able to consider both effects simultaneously (Marrocu and Paci, 2013; Massidda and Etzo, 2012). This paper contributes to this literature by investigating the determinants of the expenditures of foreign tourists in 103 Italian provinces (NUTS 3). We depart from the previous literature in that our dependent variable is not tourist flows but foreign tourist expenditures. This variable, recently made available by the Bank of Italy for the years 1997-2012, is more informative than tourist flows in that it captures not only the number of arrivals but also their contribution to a destination's GDP. The observations of our cross-section database reflect the tourist expenditure for each Italian province from each of the 20 highest spending countries of origin, accounting for 85% of total receipts. Without having to use O-D models, we will disentangle the effects of both demand and supply variables on a province's tourism exports. Among the former ones, per capita GDP levels at origin and a measure of relative price will be considered. Among the latter ones: per capita GDP levels at destination and supply variables such as capacity constraints of tourist accomodations; tourism and transport infrastructures; crime, cultural and environmental capital, climate, settlement structure typology, etc. Moreover, we will take into account the role of the distance between origin and destination, which is also a proxy of transportation costs, and the possible spillover effects originating by the supply variables in contiguous provinces. Following Halleck Vega and Elhorst (2013), spatial effects will be analysed using the spatial lag of some of the independent variables and by parameterizing the spatial matrix W. Moreover, we will use a Poisson pseudo-maximum-likelihood method as suggested by Santos Silva and Tenreyro (2006) since this method is robust to different patterns of heteroskedasticity and provides a natural way to deal with zeros in trade data.
