Testing for Equal Average Forecast Accuracy
Prof. Yang ZU
Associate Professor of Economics
Faculty of Social Sciences, UM
Date: 21 November 2023 (Tuesday)
Time: 13:00pm to 2:00pm
Venue: FBA Lobby
Abstract
Classical forecast evaluation tests assuming the performance of candidate forecasts to be constant along time, an assumption might not be satisfied in practical situations. We consider the issue of evaluating forecasts when their forecast performance is possibly varying with time and testing a new concept of average forecast performance equality. We find the classical Diebold and Mariano (1995) test has an asymptotic size of zero and reduced power in such cases. We therefore suggest a simple modified DM statistic which recovers the asymptotic size and power properties associated with the original test in the constant mean case. The new test is applied to evaluate the relative performance between the Survey of Professional Forecasters’s forecasts and some commonly used model-based ones.
All are welcome!
