Comparing the BER’s forecasts

The Bureau for Economic Research publishes annual (and quarterly) forecasts for more than 140 macroeconomic indicators, with a forecasting horizon stretching up to 6 years ahead. These forecasts are generated with the aid of a structural macro-econometric model of the South African economy. The purpose of this re-search note is to test the accuracy of the BER’s forecasts. Also to compare them with other published forecasts according to accuracy, forecast horizon and number of indicators. To determine the level of accuracy, we have calculated the mean absolute errors and the root mean squared errors of the BER’s forecasts for a selection of five economic indicators. These statistics were also calculated for the forecasts of the selected other institutions or models. From these the relative accuracy of the different forecasts were compared to each other and ranked ac-cordingly. The consensus forecast turned out to be the most accurate for the im-mediate year, followed with a narrow margin by the BER. The close proximity of these two forecasts is striking. Other conclusions are that structural forecasting models perform better than mechanical ones for the first two years, but lose their accuracy advantage from the third or fourth year onwards. They also fail to antic-ipate critical turning points in economic cycles.

C53

forecast comparison; forecast accuracy, forecast evaluation, consen-sus forecast, Theil coefficients, mean absolute error, root mean squared error, loss function