The Economic Forecasts of WIFO and IHS: An Analysis for the 1980s and 1990s
In this article the accuracy of economic forecasts for Austria made by WIFO and IHS are investigated for four different time horizons (September- and December-forecasts for next year and March- and June-forecasts for the current year) for the period from 1978 to 1999. Five evaluation criteria – forecast error summary statistics, comparisons with "naive" alternative forecasts, statistical tests on the difference in the accuracy of two predictions, tests for directional forecast precision, and statistical tests for unbiasedness and efficiency – are used to assess the comparative forecast performance for ten key economic variables – growth rates of real GDP, private consumption, gross fixed investment, fixed investment in machinery and equipment, construction investment, exports, imports, employment growth, rate of unemployment and rate of inflation. For the first time, newly developed statistical methods have been applied to Austrian data which allow to test if the differences between two competing forecasts are statistically significant. It was found that for most time series under investigation the root mean squared errors (RMSE) are smaller (or only slightly greater) than one standard deviation from the actual values. In comparison with earlier studies, forecast errors tend to have decreased slightly. The forecast error for GDP growth is much smaller than that for its components. As expected, substantial errors were found for variables that are particularly sensitive to business cycle fluctuations, like investment, exports and imports. The indicators of forecast accuracy show equal or better results for WIFO than for IHS for almost all variables and forecast horizons. Such differences are even statistically significant for some variables and forecast horizons. For GDP (June forecast), fixed investment in machinery and equipment (March), construction investment (September), employment (September), unemployment (September, March, June) and inflation (June), WIFO shows significantly smaller RMSEs than IHS. Furthermore, forecasts produced by both institutes are clearly superior to simple naive forecast strategies ("no change in the rate of change" or "no change in the level"). Both institutes are also quite able to identify the direction of future economic developments. Although, the measures of directional accuracy are for all variables well above 0.5, for investment, employment and unemployment (all for both institutes) the zero hypothesis that projections and outcomes are statistically independent cannot be rejected for all time horizons. For most variables, the hypothesis of unbiased and efficient forecasts cannot be rejected at the 95 percent level of significance. However, forecasts for imports (WIFO) and employment (both institutes) show a significant bias for some forecast horizons. For some forecast horizons, the efficiency hypothesis has to be rejected for forecasts of the following variables: fixed investment in machinery and equipment (IHS), imports (both institutes), employment (IHS), unemployment (WIFO) and inflation (both institutes). (The PDF version contains a set of 24 tables with further results of the analysis which are not included in the printed issue.)