Revealing the Risk Perception of Investors using Machine Learning

Corporate disclosures convey crucial information to financial market participants. While machine learning algorithms are commonly used to extract this information, they often overlook the use of idiosyncratic terminology and industry-specific vocabulary within documents. This study uses an unsupervised machine learning algorithm, the Structural Topic Model, to overcome these issues. Our findings illustrate the link between machine-extracted risk factors discussed in corporate disclosures (10-Ks) and the corresponding pricing behavior by investors, focusing on a previously unexplored US REIT sample from 2005 to 2019. Surprisingly, when disclosed, most risk factors counterintuitively lead to a decrease in return volatility. This resolution of uncertainties surrounding known risk factors or the provision of additional facts about these factors contributes valuable insights to the financial market.

Keywords: Real Estate Investment Trust (REIT), risk, text analysis, machine learning, Latent Dirichlet Allocation, Structural Topic Model, 10-K, Item 1A, Item 7A

JEL Classification: C45, C80, G14, G18, K22, K40, M41, M48, R30

Suggested Citation: Suggested Citation

Koelbl, Marina and Laschinger, Ralf and Steininger, Bertram I. and Schäfers, Wolfgang, Revealing the Risk Perception of Investors using Machine Learning (January 5, 2024). Available at SSRN: https://ssrn.com/abstract=3686492 or http://dx.doi.org/10.2139/ssrn.3686492