April 1, 2014, 14:00–15:15
Toulouse
Room MF 323
Statistics Seminar
Abstract
It is a widespread practice to estimate land use models at some aggregated scales but the consequences of such aggregations are rarely evaluated. This paper proposes an evaluation in terms of predictive accuracy, based on estimating a broad spectrum of individual and aggregated econometric models on the same dataset. Exploiting a detailed parcel-level dataset, we perform both short and long run predictions and compare them at the same 12 X 12 km aggregate scale of interest. In particular, we argue that data aggregation allows the application of spatial ***econometric*** tools. We show that modeling spatial autocorrelation can compensate for loss of information due to aggregation and, with well-designed predictors, can even outper- form individual models. We provide a detailed analysis of the available predictors in the context of spatial econometrics and show how to extend them in a context of out-of-sample and counterfactual predictions.
Keywords
Land use models; spatial econometrics; predictive accuracy; aggre- gate and individual data; JEL Classifications: Q15; Q24; R1; C21;