The non-normality of some macroeconomic forecast errors

This paper investigates the distributional properties of individual and consensus time series macroeconomic forecast errors, using data from the Survey of Professional Forecasters. The degree of autocorrelation and the presence of ARCH in the consensus errors is also determined. We find strong evidence of leptokurtic forecast errors as well as evidence of skewness, suggesting that an assumption of error normality is inappropriate; many of the forecast error series are found to have non-zero mean, and we find widespread evidence of consensus error ARCH. Properties of the distribution of cross-sectional forecast errors are also examined.