Model confidence sets and forecast combination: An application to mortality forecasting Seminar

Event details

Averaging the forecasts from a range of models often improves upon forecasts based on a single model. We analyze the effects of trimming the set of models before averaging, with equal weight averaging working well. We analyze the effects of trimming the set of models before averaging, with equal weight averaging working well. We propose an approach based on model confidence sets that incorporate the statistical significance of the in-sample forecasting performance. Using Japanese age-specific mortality data, we find robust out-of-sample forecast accuracy from the proposed trimming method. (Joint work with Steven Haberman, Cass Business School) After the talk, refreshments will be served in the staff reading room on level 4 of Building 54.