with Ian Marsh and Yajun Xiao
Abstract: We use public news coverage about cybercrime to form a cybercrime news attention measure. This measure is consistent with the criteria for a state variable in ICAPM that is expected to forecasts economic conditions, thereby possessing the ability to predict cross-sectional equity returns. We estimate stock-level exposures to a tradeable cybercrime tracking factor. Stocks with the most positive sensitivities generate close to 10% lower annualized risk-adjusted returns than stocks with the most negative sensitivity. Our results indicate that risk-averse investors demand extra compensation to hold stocks with negative cybercrime beta and are willing to pay high prices for stocks with positive beta that hedge exposure to cybercrime. Though our main results are derived from a proprietary cybercrime series, we show that a particularly simple publicly-available alternative based on Google search trends yields very similar conclusions.
Abstract: Investors who use biased information from news media subsequently tend to make irrational decisions about acquiring firm-specific information compared to rational expectations. This model of information acquisition yields testable predictions that are verified by using a novel dataset. First, when sentiment in news articles, as a proxy for biased public information, is more optimistic, investors tend to acquire less earnings-relevant information before the earnings announcement and vice versa. Second, the return predictability from firm-specific news sentiment confirms that it contributes to variations in asset information risk due, in a biased belief equilibrium, to the proportion of informed investors deviating from rational expectations. Overall, these findings suggest that biased public information inherent in news sentiment serves to irrationalize investors’ acquisition of firm-specific information through a biased perception of uncertainties in the risky asset payoff.
with Ian Marsh
Abstract: This paper studies mood, measured by Twitter messages, which causes investors' insufficient acquisition of information about assets and the implications of asset pricing. Using a Twitter-based mood measure, we find that mood swings are negatively predictive of investors' acquisition of earnings-related information when seeking to learn about companies' performance. Therefore, we argue that this bias effect contributes to the explanation of classical (unconditional) pricing models' failures. Conducting tests on cross-sectional stock returns, we show that stocks that are more sensitive to mood earn a higher expected excess return than less mood-sensitive stocks. Sorting stocks to construct the risk factor portfolio based on mood betas as sensitivity to mood risk, we are the first to quantify the risk premium (0.56% per month) by holding stocks subject to mood risk. Our results are consistent with the theoretical prediction that investors mistakenly use mood as information rather than learning enough fundamental information about assets, thereby inducing mispricing in asset valuation.
with Ian Marsh, Paolo Mazza and Mikael Petitjean
Abstract: We use high-frequency tick data to study stylized facts on the return and volatility dynamics of the nine most liquid cryptocurrencies. Factor structures exist in both returns and volatility, but the explanatory power from the common factor is much stronger for volatility. The factor structures do not relate strongly to fundamental economic factors, and Bitcoin – which we propose is a “crypto market factor” –has only weak explanatory power. We date the bubble in Bitcoin pricing allowing us to split the sample into pre-bubble, bubble and post-bubble periods. The importance of these different periods is clear, revealing shifting relationships between the nine cryptocurrencies and Bitcoin. Model-free realized cryptocurrency betas with Bitcoin increase during the bubble period and the explained fraction of cryptocurrency variance remains at an elevated level after the bubble burst.