研究期間：10108~10207;This project investigates the effectiveness of combination forecasts when there are unanticipated structural changes and the forecasting model is misspecified. I develop a two-dimensional averaging procedure to combine forecasts with a variety of pooling methods. The proposed combination procedure proceeds in two steps. First, to avoid forecast breakdowns in the presence of instabilities, I propose averaging across estimation windows or weighting observations to obtain the optimal forecasts from the individual models. Then, to avoid the problem of model uncertainty, I suggest averaging across different forecasting models by using a variety of combination schemes. I will provide some analytical results to demonstrate that the proposed procedure can mitigate the impacts of parameter instability and model uncertainty on forecasting. Monte Carlo simulations will be conducted to compare the finite sample performance of the proposed procedure relative to other forecast combination methods. The two-dimensional averaging procedure can be applied in a number of empirical issues in economics and finance. To illustrate the usefulness of the proposed method, two empirical applications for predicting U.S. real GDP and employment growth rates will be presented.