Using a semisupervised topic model on 7 million New York Times articles spanning 160 years, we test whether topics of media discourse predict future stock market excess returns to test rational and behavioral hypotheses about market valuation of disaster risk. Media discourse data address the challenge of sample size even when disasters are rare. Our methodology avoids look-ahead bias and addresses semantic shifts. Our discourse topics positively predicts market excess returns, with War having an out-of-sample R^2 of 1.35%. We call this effect the war return premium. The war return premium has increased in more recent time periods.

Media tone constructed from 7000,000 articles from 2000 global media and 800 social media sites is found to be a genuine risk factor that cross-sectionally prices currencies. It can predict excess US dollar returns for up to six months and surpasses the no-change benchmark in predicting returns out of sample. Its predicted value contains information beyond those predicted by currency factors and business cycles. Evidence corroborates with the theory that Media tone increases investment returns, has pronounced predictive power for the currencies associated with hard-to-value characteristics, and that its predictive power increases with the number of news sources. Trading of rational investors, including banks, is associated with Media tone.

A war‐related factor model derived from textual analysis of media news reports explains the cross section of expected stock returns. Using a semisupervised topic model to extract discourse topics from 7,000,000 New York Times stories spanning 160 years, the war factor predicts the cross section of returns across test assets derived from both traditional and machine learning construction techniques, and spanning 138 anomalies. Our findings are consistent with assets that are good hedges for war risk receiving lower risk premia, or with assets that are more positively sensitive to war prospects being more overvalued. The return premium on the war factor is incremental to standard effects.