Shepherding the Herd
62 Pages Posted: 18 Mar 2020
Date Written: February 21, 2020
This article analyzes multiple experts who forecast an underlying dynamic state based on streams of (partially overlapping) information. Each expert minimizes a convex combination of her forecasting error and deviation from the other experts’ forecasts. As a result, the experts exhibit herding behavior – a bias that has been well-recognized in the economics and psychology literature. Our first contribution is a general framework for analyzing experts’ forecasts under different levels of herding. The underlying state dynamics can be non-linear with seasonality, trends, shocks, and/or other time- series components. We illustrate our framework with an exploratory analysis, showing that negative shocks to public information have a strong and long lasting effect on welfare under high levels of herding. Our second contribution describes how models within our framework can be estimated from data. We apply our estimation procedure to two surveys of inflation forecasting and find that experts concentrate around 8% of their efforts on making similar forecasts, regardless of whether reporting is anonymous or not. Finally, we propose a simple compensation scheme that uses an estimated model to minimize the negative effects of herding.
Keywords: Bayesian statistics, Dynamic modeling, Gaussian process, Judgmental forecasting, Imperfect information game
Suggested Citation: Suggested Citation