Dynamic Forecasting Behavior by Analysts : Theory and Evidence

We examine the dynamic forecasting behavior of security analysts in response to their prior performance relative to their peers. In a multi-period Bayesian learning model where an analyst and her employer (the investment bank or brokerage house) bargain over the surplus that she generates in each period, we show that the analyst's compensation at any date is convex in her perceived forecasting ability or reputation. Further, she faces employment risk as a result of the variation of her perceived ability with her relative forecasting performance. The interplay between the convexity of her payoff structure and the presence of employment risk leads to a non-monotonic U-shaped relationship between the deviation of the analyst's forecast from the consensus, and her prior relative performance. Hence, analysts who significantly out-perform or under-perform their peers issue bolder forecasts than intermediate performers. Consistent with existing empirical evidence, our theory also predicts that, ceteris paribus, the incentive to herd decreases with the analyst's experience. In our empirical analysis, we document significant support for the predicted U-shaped relation between the boldness of an analyst's forecast and her prior relative forecasting performance. We also document a positive relation between boldness and experience controlling for the predicted non-monotonic relation between boldness and prior performance that previous empirical studies do not incorporate. Finally, we directly test the importance of employment risk in driving variations in forecasting behavior. Consistent with the theory, we document that analysts with either high or low probabilities of being terminated in the future are more likely to issue bolder forecasts than those with intermediate termination probabilities