Data

Risk Exposure Measures

Published: January 01, 2021

Firm-level risk exposures are derived by first parsing all public 10-K filings available in the EDGAR database reported between 1/1/2018 and 3/1/2020. Specifically, all readable text under “Item 1” is processed by a Python program to ensure the extracted raw terms are following the standard procedure in the text-analysis literature; i.e., the text is cleaned from abbreviations, headings, plurals, stop words, and numbers, leaving only relevant lower-cased terms. Firm $i$’s level of risk exposure to category $j$ is defined each calender year as:

$Risk\, Exposure_{i,j} = \frac{\sum_{y,j}^{Y,j}term_{y,j}}{\sum_{j=1}^{J=44}\sum_{y,j}^{Y,j}term_{y,j}}$

In essence, this formula estimates each firm’s risk exposure to category $j$ as the total number of terms extracted from the document falling under risk factor $j$ divided by the total number of terms captured for all 44 risk factors for firm $i$. Risk categories, and their associated terms, are created by Davis, S. J., Hansen, S., & Seminario-Amez, C. (2021). Firm-level Risk Exposures and Stock Returns in The Wake of COVID-19. All 44 risk factors, and their terms, are obtained from the authors’ article. Categories (or factors) include Commercial Property, Display Technology, Traditional Retail, E-commerce, and Franchising. For further detail, please read the recent draft of my paper.