Title: Cognitive dissonance and forecaster overconfidence in a model of inflation expectations with distributional inaccuracies
Authors: Shaun Vahey - Warwick University (United Kingdom) [presenting]
Abstract: Professional forecasters and central bankers often describe and quantify the asymmetric risks associated with macroeconomic variables. In contrast, nearly all extant models of inflation expectations assume that agents know exactly the distribution of elliptical disturbances. An alternative mechanism is explored in which agents form expectations of inflation in the presence of distributional inaccuracies. Agents learn the true distribution and inflation dynamics by fitting a Gaussian copula model to the univariate time series. The resulting predictive distributions for inflation are generally asymmetric and dependence is time-varying even though the underlying inflation dynamics are linear. The Gaussian copula model produces superior forecasts to recursive least squares in large samples but not necessarily in small samples. The implications of copula learning for expectations are explored in an application using quarterly US inflation (GDP deflator) data. We examine two specific characteristics of the agents' forecasts relative to recursive least squares. Namely, large and persistent forecasting errors at the mean (cognitive dissonance) and insufficient uncertainty (overconfidence). Nevertheless, with exposure to extreme inflation events, agents better fit the tails of the unknown error distribution, consistent with the notion of adaptive learning.