A1183
Title: Nowcasting and forecasting with heavy tails in macroeconomics
Authors: Nicholas Fawcett - Bank of England and Centre for Macroeconomics (United Kingdom) [presenting]
Andrew Harvey - University of Cambridge (United Kingdom)
Martin Weale - Bank of England (United Kingdom)
Abstract: Although most macroeconomic forecasting models assume that errors follow a Gaussian distribution, in many cases this is invalid: large outliers are much more likely than the models allow. We use a new class of state space models in which errors can have heavy-tailed distributions to produce better nowcasts and short-term forecasts of macroeconomic series. These models offer a natural way of down-weighting large outliers when they occur, which provides robustness to noise, allowing us to distinguish better between structural breaks and outliers. We apply the framework to a range of series, including nominal wages, GDP, and a small macroeconomic system including oil prices, GDP, inflation and interest rates. The framework allows for variation in the underlying location and scale variables over time.
