Time Varying Risk Aversion and Asset Returns

Prices and investors' behavior are heavily influenced by risk aversion. As it is unobservable, estimating risk aversion has been challenging for a long time. This paper proposes using a Machine Learning approach (a combination of Autoencoder and Long-Short Term Memory) to estimate the time-varying risk aversion. I show that credit spread, dividend yield, next-period recession probability, and expectation of inflation change are the most influential variables for estimating risk aversion. The estimated risk aversion is positively (negatively) correlated with market uncertainty (consumption growth).This paper extends the long run risk model by considering a stochastic risk aversion process. By solving the model numerically, I show that a time varying risk aversion helps us to get closer to matching the price moments