I introduce a novel approach for the empirical analysis of asset price comovement that relates the inter-firm textual similarity of news reports to their equity return correlation. I find that this measure of news similarity is just as important for predicting future cross-firm comovement as contemporaneous return correlation. This predictability remains after controlling for industry correlation, size, book-to-market, momentum, and price-decile correlation, index membership, and headquarters location, as well as institutional holding and analyst coverage. These results contribute to the growing literature examining the role of the media in financial markets, and provide empirical support for an alternative description of return comovementthat does not depend on friction-based explanations such as “category,” “habitat,” or “information diffusion.”
The prediction of return covariance matrices is necessary for minimizing portfolio risk. By decomposing covariances into correlations and standard deviations, I am able to measure each parameter estimate’s sensitivity to measurement error and time series variation in underlying fundamentals separately. For a large sample of individual stocks, I find that the susceptibility of correlation estimates to measurement error can be mitigated by longer estimation windows, while the vulnerability of standard deviations to fundamental time series variations can be overcome by shorter estimation periods. I also introduce a new approach for predicting return correlations that, when used in the formation of minimum variance portfolios, can generate lower out-of-sample volatility and higher mean returns than a variety of traditional and passive strategies.