HYPOTHESIS TESTING USING NUMEROUS APPROXIMATING FUNCTIONAL FORMS

While the combination of several or more models is often found to improve forecasts (Brandt and Bessler, Min and Zellner, Norwood and Schroeder), hypothesis tests are typically conducted using a single model approach 1 . Hypothesis tests and forecasts have similar goals; they seek to define a range over which a parameter should lie within a degree of confidence. If it is true that, on average, composite forecasts are more accurate than a single model's forecast, it might also be true that hypothesis tests using information from numerous models are, on average, more accurate in the sense of lower Type I and Type II errors than hypothesis tests using a single model.