A0957
Title: How far can we forecast the economy?
Authors: Marc-Oliver Pohle - Heidelberg Institute for Theoretical Studies (Germany)
Tanja Zahn - Goethe University Frankfurt (Germany) [presenting]
Abstract: Forecasts are usually issued over multiple forecast horizons. Often, they range quite far into the future, giving rise to questions on how useful long-horizon forecasts are and up to which maximum horizon forecasting is really sensible. A methodology is developed to answer this question for arbitrary types of forecasts, including univariate and multivariate mean, quantile and probabilistic forecasts. The key tool is a nicely interpretable measure of predictive power, which amounts to the ratio of variation explained by the forecasts to overall variation in the variable of interest. This measure nicely complements the standard toolkit of forecast evaluation, informing about the usefulness of the forecasts in the first place. Measures for information content of forecasts and predictability of the variable of interest are introduced and their relationship to predictive power is discussed. The methodology is applied to macroeconomic forecasts. The analysis shows that the predictive power of state-of-the-art forecasting methods for inflation and GDP growth can be very limited even for short forecast horizons, which hints at a lack of predictability of the economy.
