Growth forecasts using time series and growth models

The authors consider two alternative methods of forecasting real per capita GDP at various horizons: 1) univariate time series models estimated country by country; and 2) cross-country growth regressions. They evaluate the out-of-sample forecasting performance of both approaches for a large sample of industrial and developing countries. They find only modest differences between the two approaches. In almost all cases, differences in median (across countries) forecast performance are small relative to the large discrepancies between forecasts and actual outcomes. Interestingly, the performance of both models is similar to that of forecasts generated by the World Bank's Unified Survey. The results do not provide a compelling case for one approach over another, but they do indicate that there are potential gains from combining time series and growth-regression-based forecasting approaches.