Can the Variance After-Effect Distort Stock Returns?

Abstract

Variance after-effect is a perceptual bias in the dynamic assessment of variance. Experimental evidence shows that perceived variance is decreased after prolonged exposure to high variance and increased after exposure to low variance. We introduce this effect in an otherwise standard financial model where information about variance is incomplete and updated sequentially. We introduce a variance after- effect adjustment factor in a bayesian learning model and derive the associated predictive variance. We show theoretically how this adjustment factor affects both average and volatility of excess returns. We construct a proxy of the adjustment factor using the sequence of dispersion of analysts earnings forecast. We provide empirical evidence using US stock data over the sample 1982 - 2019, that fluctuations in this measure are significantly and positively related to excess volatility as predicted by the model. Further confirming the model's implications, we also show how stock returns are positively impacted by the adjustment factor and construct long short strategies that generate significant positive alpha with respect to the Fama-French 5 factor model.

Keywords: Variance after-effect, learning, turnover, volatility, earnings forecasts

JEL Classification: G11, G12, G41