Forecast Combination with Entry and Exit of Experts

Combination of forecasts from survey data is complicated by the frequent entry and exit in real time of individual forecasters which renders conventional least squares regression approaches to estimation of the combination weights infeasible. We explore the consequences of this for a variety of forecast combination methods in common use and propose a new method that projects actual outcomes on the equal-weighted forecast as a means of adjusting for biases and noise in the underlying forecasts. Through simulations and an empirical application to inflation forecasts we show that the entry and exit of individual forecasters can have a large effect on the real time performance of conventional forecast combination methods. We also find that the proposed projection on the equal-weighted forecast works well in practice.