A0692
Title: Macroeconomic survey forecasting in times of crises
Authors: Robinson Kruse-Becher - FernUniversität in Hagen (Germany) [presenting]
Philip Letixerant - FernUniversität Hagen (Germany)
Abstract: Survey-based forecasts such as the SPF are generally of great importance. The aim is to investigate the potential for improvement by exploiting historical information that is similar to the current situation in which the expert forecasts are made. The real-time adjustment applied to the SPF is a similarity-based intercept correction. Periods of similarity are found by a nearest neighbor matching procedure. A large number of aspects in the matching algorithm are investigated by considering, e.g., matching variables (levels versus forecast errors), matching quality measures, and weighting based on the degree of similarity, a recency filter, etc. Hyperparameters such as the block length and the number of matched periods to be averaged are selected by cross-validation. The focus is on the great financial crisis and Covid-19 recession, and the potential to reduce the relative out-of-sample MSFE is analyzed in such times of crises. The analysis is complemented by considering the performance of the adjustment in non-crisis periods.
