Title:A Quantile Approach to Evaluating Asset Pricing Models

Abstract:This paper studies the misspecification of asset pricing models by examining the local difference between the physical and risk-neutral measure. Using quantile regression, I find that the conditional physical distribution differs most from the risk-neutral distribution in the left tail, while the right tails of both distributions are almost identical. I propose a new bound on the volatility of the stochastic discount factor (SDF) to show that the local difference in the left tail implies that the SDF is highly volatile. Finally, I propose a lower bound on the difference between the physical and risk-neutral distribution in the left tail, which is approximately tight in the data. I interpret the bound as a real time measure of the Peso problem and find that it fluctuates significantly over time. At the height of the 2008 financial crisis and 2020 Covid crisis, the bound predicts that a market return of -28% or lower has a 5% probability. Both findings provide evidence for time varying disaster risk.