SPATIAL CHOW-LIN METHODS: BAYESIAN AND ML FORECAST COMPARISONS

Completing data that are collected in disaggregated and heterogeneous spatial units is a quite frequent problem in spatial analyses of regional data. Chow and Lin (1971) (CL) were the rst to develop a uni ed framework for the three problems (interpolation, extrapolation and distribution) of predicting disaggregated times series by so-called indicator series. This paper develops a spatial CL procedure for disaggregating cross-sectional spatial data and compares the Maximum Likelihood and Bayesian spatial CL forecasts with the naive pro rata error distribution. We outline the error covariance structure in a spatial context, derive the BLUE for the ML estimator and the Bayesian estimation procedure by MCMC. Finally we apply the procedure to European regional GDP data and discuss the disaggregation assumptions. For the evaluation of the spatial Chow-Lin procedure we assume that only NUTS 1 GDP is known and predict it at NUTS 2 by using employment and spatial information available at NUTS 2. The spatial neighborhood is de ned by the inverse travel time by car in minutes. Finally, we present the forecast accuracy criteria comparing the predicted values with the actual observations.