We propose two new testing methods for detecting changes in forecast accuracy by using pseudo out-of-sample procedures. One is a test that doubly takes supremum of Wald statistics over possible change dates and possible in-sample window sizes. We also consider investigating changes in a joint sequence of in- and out-of-sample forecast losses. Our Monte Carlo simulation shows that the proposed methods significantly improve power over the existing methods. We provide empirical examples of forecasting exercises for crude oil price and U.S. inflation rate.