Title: Simulation-based selection of prediction models in development-economics panels
Authors: Robert Kunst - Institute for Advanced Studies (Austria) [presenting]
Adusei Jumah - Central University Accra (Ghana)
Abstract: Basing model selection decisions in a forecasting context on simulations from estimated rival structures can be an attractive albeit time-consuming tool in macro-economic forecasting. The simulations fuse data information and the structure hypothesized by tentative rival models. We explicitly focus on applying the idea to a panel of macro-economic data from African countries. The declared target is the optimization of predictions for economic growth in individual countries, based on their growth history and on sector shares. Data on sector shares are conveniently available for most African countries over extended time spans and are representative of the ongoing process of structural transformation. Our procedure chooses among few tentative forecast models in the presence of data, in this example univariate and multivariate models with a varying degree of panel homogeneity assumptions. From models fitted to the data, pseudo-data are generated. Again, the models are applied to the pseudo-data and their out-of-sample performance is evaluated. The ultimate choice of the forecasting model is based on the relative performance of rival models in predicting their own data and those of the rival model.