Ross recovery with time series information and economic constraints

Abstract

I propose a new type of Ross recovery informed from time series, and subject to shape restrictions from economic theory. Within my data-driven and nonparametric framework, I find that decreasing marginal utility, or a monotonic stochastic discount factor, strongly dominates the more flexible pure no-arbitrage model out-of-sample. Both specifications, with or without monotonicity imposed, imply an equity premium the cyclicality of which changes conditionally with the state of the world. Both generate sizable out-of-sample predictability of realized returns beyond extant conditional predictors and recovery frameworks.

Keywords: Ross recovery, monotonic SDF, options, dimension reduction, factor model, scenario analysis, machine learning.

JEL Classification: G11, G12, G15