Predicting Gross Domestic Product Components through Tsallis Entropy Econometrics

This article proposes the Tsallis non-extensive entropy econometric approach to forecast components of the country gross domestic product based on the knowledge of time series macroeconomic aggregates of the past period, plus some sparse and imperfect information of the current period. Non-extensive entropy technique has proved to remain a good modelling device not only in the case of high frequency series, but also in the case of aggregated series. To predict the missing GDP components, we set up a q-generalized Kullback-Leibler information divergence criterion function with a priori consistency, GDP related macroeconomic constraints and regular conditions. The model forecasts are compared to the official Polish GDP components of the corresponding period. The proposed Tsallis entropy approach leads to high predictive performance and shows a stronger estimation stability through different model simulations than the traditional Shannon model. Furthermore, as expected this Tsallis related approach seems to reflect a higher stability through parameter computation and simulation in comparison with the traditional Shannon-Gibbs entropy technique.