Forecasting international growth rates using Bayesian shrinkage and other procedures

Abstract

Forecasts of eighteen countries' annual output growth rates, 1974–1984, are computed using an autoregressive model containing leading indicator variables and various forecasting procedures. It is found that Bayesian shrinkage forecasting techniques produced improved forecasts in terms of an out-of-sample root-mean-square criterion relative to those provided by three naive models and by autoregressive models with and without leading indicator variables. The precision of our forecasts compares favorably with that of OECD forecasts derived from complicated econometric ‘structural’ macroeconometric models and subjected to judgmental adjustments. Work to rationalize our forecasting models in terms of macroeconomic theory is in progress.