Forecaster Diversity and the Benefits of Combining Forecasts

Abstract

The expected error variance of a combined forecast is necessarily lower than that of an individual forecast, but in practice there may be considerable variation around these expected values. This paper introduces a measure of the benefit from combining, the probability of a reduction in error variance, which recognizes this problem. The measure is applied to data on the forecasts and forecasting methods of a panel of U.S. economists to determine how the benefits of combining vary with the number of forecasts combined, and with the diversity in theories and techniques among the component forecasts.