Economic Narratives and Market Outcomes: A Semi-supervised Topic Modeling Approach
Abstract: I employ the seeded Latent Dirichlet Process (sLDA) model in natural language processing to extract the narratives discussed by Shiller (2019) from nearly seven million New York Times articles over the past 150 years. The estimation scheme is designed to avoid any look-ahead bias in constructing the monthly narrative weights. Among the narratives considered, the most important one is Panic, which encompasses various stress- and anxiety-related themes including economic downturns, wars, political tensions, and epidemics. I find that Panic and a narrative index that loads heavily on Panic are strong positive predictors of excess U.S. market return and negative predictors of both realized and implied market volatility, consistent with a version of the conditional CAPM. The predictability of narratives over market outcomes holds at both market and portfolio level, and at both monthly and daily interval. Importantly, the predictive power of Panic is increasing over time. Over the past 20 years, in monthly predictive regression of excess market return, Panic produces an in-sample R-squared of 4.2% and an out-of-sample R-squared of 2.6%.
(with Kuntara Pukthuanthong, 2020)
Abstract: We conduct empirical tests of a simplified version of the ratio habit model developed in Abel (1990), in which habit is extended beyond the preceding period. We show that change in four-year consumption growth—the measure of consumption resulting from our ratio habit preference—explains the joint equity premium–risk-free rate puzzle with a risk aversion coefficient much lower than any existing consumption measures under the standard consumption model. This outperformance of our ratio habit model over the standard model is robust across 18 non-U.S. countries. From 1928-2017, change in four-year consumption growth encompasses other consumption measures in explaining the cross-sectional variation of expected returns on various portfolios and it is the only consumption measure that passes the robust tests of the factor risk premium proposed by Kleibergen and Zhan (2020). While our measure constructed from nondurables does better at pricing the equity premium and risk-free rate, our service-based measure outperforms in explaining the cross-sectional variation of stock returns.
WORKS IN PROGRESS
Evolution of News Sentiment over the Past 200 Years