Impact of Cognitive Biases on Forecasting Models

Abstract

The impact of the cognitive biases of overconfidence, underconfidence and anchoring on the distribution of errors of forecasting models is analyzed using an analytical framework based on a flexible two-piece generalized distribution. The total forecasting bias, measured by the expected value of a model’s errors, is decomposed to an anchoring bias and a skewness bias. An examination of BEA’s preliminary estimates of the final GDP growth rates reveals that the underprediction present is to a large extent the result of negative skewness bias and to a lesser extent of negative anchoring bias. The latter are attributes of underconfident forecasters.