Working Papers

• Statistical Properties of News Sentiment Inhibit Return Predictability [PDF] [SSRN]
  

  Revision Requested at Critical Finance Review

  Using text-based variables with unknown statistical properties alongside conventional numeric variables carries a high risk of distorting economic inference. I find that daily news-based variables, regardless of their construction methodology, are often nonstationary and therefore violate common assumptions of time series analysis. To demonstrate the empirical relevance of these findings, I use robust methods to reexamine the conclusions of Garcia (2013). Contrary to Garcia (2013), I find that daily news sentiment forecasts stock market returns equally well during both recessions and expansions, with some evidence suggesting better predictability during expansions.

• The Value of Trade Secrets: Evidence from Economic Espionage [PDF] [SSRN]
  

  with Alexander Michaelides, Andreas Milidonis, and Yupana Wiwattanakantang

  We estimate the value of trade secrets, a key component of intangible capital, by hand-collecting criminal cases involving trade secret theft filed under the Economic Espionage Act of 1996 between 1996 and 2019. The cumulative average abnormal return due to trade secret theft is –1.74% over the [–4,+1] event window, corresponding to an aggregate loss across all events of approximately $181 billion (2020 dollars). These estimates are robust to controlling for weak corporate governance and reputational capital loss. Victim firms respond by acquiring a large number of smaller firms to replenish intangible capital.

• The FOMC Announcement Premium Asymmetry [PDF] [SSRN]
  

  Excess equity returns around Federal Open Market Committee (FOMC) meetings are concentrated in recessions. On FOMC announcement days, the difference between stock returns in recessions and expansions is 73-119 basis points (bps). For reference, the unconditional difference between the announcements and all other trading days is 21 bps. The asymmetry remains statistically significant after accounting for the elevated volatility in economic downturns. The pre-announcement drift and the compensation for bearing risk on announcement days are also much more pronounced in recessions. Overall, the state-dependent equity market behavior around FOMC news releases reflects the asymmetric risk accumulation over the business cycle.

• Pairwise Dissimilarity and Risk-Seeking Portfolio Construction [PDF] [SSRN]
  

  I propose a heuristic portfolio construction rule that weights assets according to their dissimilarity from the remaining holdings. The methodology is insensitive to estimation error and does not assume a return-generating process. Relative to the 1/N rule, in a sector rotation setting, the dissimilarity approach delivers annual excess returns of 6 to 121 bps while maintaining comparable levels of diversification, volatility, tail risk, and turnover. This outperformance reflects compensation for bearing tail risk and harnessing momentum by dynamically increasing the weights of rapidly appreciating assets.

Work-in-Progress

• Financial Texts and Strategic Information Sequencing

  I propose a new methodology for quantifying topic importance based on the order and context of words within a text. It captures strategic information sequencing representing managerial intent to emphasize or obfuscate risk factors. The procedure does not require dimensionality reduction and has low measurement error.

• Cybersecurity Risk Disclosure and Pricing

  Cybersecurity risk exposure and managerial decision to disclose it strongly depend on the firm’s industry and size. After accounting for the industry-level effects, only small firms have statistically significant cyber risk alpha. For small firms, a potential incident is a tail event; large firms are either too big to fail or self-insure.