New research concludes that companies’ pre-pandemic SEC 10K Risk Factors disclosures were effective predictors of future equity performance
The economic impact produced by the pandemic has been something new in modern times and continues to play out around the world. Real GDP fell 11% in the U.S., and the IMF forecasts that global output saw its most significant recorded drop in 2020. Considerable shifts in labor and technology allocation have taken place, producing many corporate winners alongside steep layoffs and bankruptcies.
Early in 2020, as the pandemic came to light, investors were presented with an array of news about the crisis’ likely impact. Analysts of all stripes gave opinions about markets and companies, producing various analyses and investment strategies to maximize pandemic returns and minimize damage. Given the unprecedented nature of the crisis, most of the analysis was speculative, of course. Most investors assumed the COVID-19 was terra incognita to be navigated as events unfolded. New research from Steven J. Davis (Chicago Booth), Stephen Hansen (Imperial College London), and Cristhian Seminario-Amez (Chicago) suggest an alternate reading of that conclusion.
Their recently published research looked at what companies had said publicly about their own risk exposures before the pandemic in their SEC 10K filings. These documents are serious matters, because companies can face significant legal and financial penalties for not disclosing material risks to regulators and investors. The researchers wondered whether information contained in the 10K’s Risk Factors (RF) section could have been used to predict how COVID-19 would affect the company’s performance in the crisis.
To find out, the researchers looked at how 2,155 equity securities moved from 24 February to 27 March, paying particular attention to what are known as “jump days” when the market rises or falls by at least 2.5%. Having established the equity dynamics during that period, the researchers analyzed the RF data for those stocks using two different but, as it turns out, complementary techniques.
Their baseline approach used what is known as the “dictionary” method, which relies “on expert-curated term sets to characterize and quantify the information content in relevant text documents.” In essence, these dictionaries are lists of terms used in financial disclosures that, to an expert reader, suggest the firm is sharing some positive or negative information. For this study, the companies’ disclosures were mapped against 36 different financial and economic dictionaries. The dictionaries, note the authors, “contain 430 terms that appear in our RF corpus, 244 after removing rare terms at the pre-processing stage.” These 244 terms appear “nearly 1.4 million times, constituting 2.4% of the RF corpus.”
The first technique provided the researchers with a map of relevant terms that the companies in the study had used to describe their future risks before the pandemic. To complement this analysis, the team also employed a supervised machine learning (SML) approach, which differs from the dictionary method in two ways. First, “it considers all terms that appear in the discussions of Risk Factors as candidates for explaining returns.” This feature is valuable because “the set of all terms is an order of magnitude larger than the term sets encompassed by the curated dictionaries.” Second, the SML approach “weights each term based on the strength of its association with the outcome of interest,” while the “dictionary approaches typically weight terms based on their frequency in the text documents of interest and perhaps in external sources as well.” In other words, the SML approach examines more text and tries to make connections that are specific to the aims of the current analysis (and not a pre-existing generic coding).
By applying this hybrid approach, the authors obtained a deeper understanding of the forces driving firm-level returns. It also helped them “uncover the role of exposures to social distancing restrictions, drug trials, e-commerce and more” as well as “the role of downstream demand linkages.”
From their dictionary-driven analysis, the team found that information contained in 10K Risk Factors disclosures was indeed a useful indicator of a firm’s likelihood to be hurt or helped by the forces the pandemic unleashed on the world. There are numerous examples in this comprehensive paper, but a few should suffice to make its point.
The stocks of companies that disclosed “high exposures to inflation, credit indicators, taxes, entitlement programs, energy and environmental regulations, and transportation, infrastructure and utilities react especially negatively to bad news about the pandemic and its economic fallout.” Firms with “high exposures to intellectual property and health-care policy perform relatively well in reaction to bad pandemic news.”
Jump days attributed to monetary policy easing “yield large positive return reactions for firms with high exposures to inflation and interest rates but not to intellectual property or transportation, infrastructure, and utilities.” Jump days attributed to “social policy news generate the largest return reaction at firms with high exposure to the tax category.”
Firms exposed to “tax-sensitive categories like real estate and business investment also outperform on fiscal policy jump days.” However, the authors note that 10K tax risk disclosures are tricky, since they capture both “exposures to both high taxes and the potential for large tax credits (e.g., for R&D or investment).”
From their SML-based analysis, different and complementary findings emerged. Disclosures about exposure to retail and card payments drove negative price changes, for example, while exposure to e-commerce risk had the opposite effect. “Video games” turned out to be a good thing to find in the 10K, unlike “aircraft” or “air travel.” Oil and gas, note the authors, “are associated with negative returns, while the opposite is true for exposures to the technology supply chain as captured by raw metals and minerals.” Interestingly, financial exposure terms were also indicative of future performance: “Exposure to financial management is also associated with negative return reactions,” though, in contrast, “exposures to banking, deposits, and investment funds yield positive return reactions.”
In summary, note the authors, “bad COVID-19 news generates a wide array of positive and negative return reactions across firms,” but the information contained in the 10K Risk Factors disclosures “enables us to uncover dozens of separate effects that play a role in driving the structure of return reactions.”
With their primary analyses complete, the authors took up one last challenge: to assess whether their new approach could also predict firm-level stock movements for other important news events. For this test, they chose the 2020 “Super Tuesday” primary won by Joe Biden, which cemented his role as the Democratic front-runner. Here again, their approach proved effective:
As a further illustration, we use the hybrid approach to characterize returns the day after the 2020 Super Tuesday elections. The market rose 4% in reaction to these elections, widely regarded as a decisive victory for Joe Biden that greatly raised his chances of securing the Democratic nomination. We again apply our hybrid approach to uncover risk factors, build associated term sets, and use them to explain firm-level returns. We find that Super Tuesday drove negative returns for firms with high exposure to hotels, gambling, fracking, and financial management; and positive returns for firms with high exposure to healthcare, health insurance, REITs, property rentals, communications and construction. Our hybrid approach lends itself to many other applications as well, including text-based analyses of what drives other outcomes.
The authors conclude by noting that “company-level stock-price reactions to COVID-19 news in early 2020 helped to predict corporate earnings surprises later in the year” and they “also foreshadowed other broad economic shifts.” Of course, those early economic changes led to subsequent shifts through economic chain reactions that have yet to run their course. This novel research helps us better understand those economic reaction chains, not just ex post facto but also with an eye toward what is to come. Indeed, as with many of the best risk-related analyses, this study reinforces the idea that the future is usually hidden in the present. It also suggests that the 10K Risk Factors disclosures, often glossed over in favor of more well-known figures and indices, may hold more insight potential than previously imagined.
Davis, Steven J. and Hansen, Stephen and Seminario-Amez, Cristhian, Firm-Level Risk Exposures and Stock Returns in the Wake of Covid-19 (September 2020). NBER Working Paper No. w27867, Available at SSRN: https://ssrn.com/abstract=3700695