Investors’ attention and information losses under market stress

The paper proposes a novel point-wise entropy approach to measure the time-varying losses in the value of information that investors associate with market signals, financial and economic indicators, and news. We cast our approach in a Bayesian framework and assume that market agents update their beliefs to incoming signals based on a prior in- formation set. By exploiting the distribution rather than the time-series properties of in- formation signals, our method is able to construct univariate signal-specific, but also com- posite proxies of information loss, with the latter being more efficient in reducing mis- leading effects and interpretation errors. As an empirical illustration, we construct infor- mation loss proxies for the US equity market from several mainstream information signals and find that the majority of information loss indicators can influence investors’ atten- tion, which then intermediates the impact of information signals on market outcomes. Fi- nally, we show that, by relying on composites rather than univariate proxies, market agents can diversify and thus reduce their information losses when interpreting signals associated with the same underlying event.