Out-of-sample equity premium prediction: The role of option-implied constraints

We study a new constrained equity premium forecasting approach which employs the option-implied lower bounds for the conditional market premium from Martin (2017) and Chabi-Yo and Loudis (2020), respectively, as forecast constraints. This constrained approach delivers considerable out-of-sample gains in both statistical and economic criteria relative to the unconstrained predictive regression forecasts, and outperforms the prevailing non-negativity constraint at longer forecast horizons. We provide two explanations for the improvements: (1) option-implied bounds help to eliminate erratic return predictions and stabilize the unconstrained forecasts; (2) constrained forecasts integrate the information in conventional economic predictors and the forward-looking information about the expected returns implied by option prices. In addition, we find substantial differences in the forecasting performance of constrained forecasts. The information complementarity between the bounds and economic predictors helps to explain such heterogeneity.