Makroøkonomiske forudsigelser baseret på diffusionsindeks

Abstract:
 

This paper presents simulated out-of-sample forecasts based on diffusion indexes for monthly Danish macroeconomic data. The diffusion indexes are derived from 246 series (172 monthly and 74 quarterly series). The primary focus is on forecasts of unemployment, industrial turnover, and inflation at horizons of 1, 6, and 12 months over the period 1995-2003. The use of diffusion indexes is primarily inspired by Stock and Watson (1998 and 2002a, b).

Using unfiltered data it is shown, that it is possible to obtain MSFE (mean squared forecast error) of diffusion index forecasts for Danish data that are slightly smaller than the MSFE of a standard autoregressive model. However, the gain in forecasting accuracy is not robust to the specification of the forecasting equation – that is the maximum number of diffusion indexes and the maximum number of lags included in the general specification. If the specification is not chosen “correctly”, forecasts based on the diffusion indexes could very well be worse, than the autoregressive forecasts. As it is difficult to establish general rules with respect to the parameterization of the forecast-ing equation this poses a serious problem in real time forecasts.

One explanation for the disappointing results could be that the Danish data are rather noisy, leading to very volatile estimates of the diffusion indexes. Fore-casts based on diffusion indexes calculated on filtered data – that is using only the business cycle component of the data – result in a reduction of MSFE relative to autoregressive forecasts of 10-15 per cent. Even though some of the problems with specifying the forecasting equation are similar to the case with the unfiltered data, the gain in forecasting accuracy using with filtered data seem relative robust.