Weighted Spatial Adaptive Filtering: Monte Carlo Studies and Application to Illicit Drug Market Modeling

Abstract

This paper introduces a pattern recognizer, similar to weighting schemes used in combining time series forecasts, for use in spatial adaptive filtering applied to estimating multivariate cross‐sectional models. The pattern recognizer enhances the ability to automatically detect and estimate parameters with discontinuous or sharp gradient changes over geographic contexts. Results from Monte Carlo studies suggest that the weighted spatial adaptive filter is at least as accurate as the unweighted filter for cases having smoothly changing parameters, but superior for cases having discontinuous, step‐jump parameters. A case study on illicit drug‐market activities using census tract‐level data from Pittsburgh, Pennsylvania, further illustrates the advantages of the weighted filter.