Economic narratives and market outcomes: A semi-supervised topic modeling approach

Abstract

We employ sLDA to extract the narratives discussed by Shiller (2019) from 7 million NYT articles over 150 years. The estimation addresses look-ahead bias and changes in semantics. Panic and the narrative index positively predict market return and negatively predict volatility. Panic presents time-varying risk aversion. The narrative predictability increases recently at both market and portfolio and monthly and daily intervals. The narrative index constructed from 2 million WSJ articles over 130 years retains its predictive power, but Stock Bubble emerges as a negative market predictor. Media customizes their narratives to their readers, having a diverse effect on the market.

Keywords

Tags of Social Finance