This blog post continues our series on complex investments.
There is a large and growing literature on the amount of complexity in public company filings. Over the past 14 years, research has show a correlation between statement length, high readability grade levels, and certain word patterns with a number of bad outcomes, including:
- Accounting fraud;[i]
- Restating financial reports;[ii]
- Stock price crashes;[iii]
- Higher loan rates;[iv]
- Poor and less persistent profits;[v]
- Higher stock price volatility, and;[vi]
- Negative earnings surprises.[vii]
Phrases of Fraud
Researchers have also identified a number of phrases that appear in the Management Discussion and Analysis section of 10-K filings that are correlated to fraud. These phrases, called “lexical bundles”, can be searched for using Natural Language Processing techniques, which analyze the text of company filings.
One report compared the frequency of phrases in known fraudulent 10-K filings to non-fraudulent filings. This analysis identified a number of phrases that occurred much more frequently in the fraudulent filings. Some of them included the following phrases (percent difference of appearance in fraudulent filings to non-fraudulent filings in parentheses):[viii]
- “In the year ended,” (213%);
- “Process research and development,” (160%);
- “Could have a material,” (146%);
- “Have a material adverse,” (50%);
- “The fair value of,” (50%);
- “Be no assurance that,” (17%);
The SEC's "Robocop" Regressions
The search of 10-Ks for known phrases has been taken up by the SEC using what they call their Accounting Quality Model (“AQM”). This initiative is part of the SECs Center for Risk and Quantitative Analytics. In addition to searching 10-K filings for known phrases, the AQM runs regressions on the accounting numbers reported, focusing on discretionary accrual factors and off-balance sheet items.[ix]
The AQM (which has been dubbed “Robocop” by the industry) regressions produce a score for each filing that gives a probability for it containing fraud.[x]
Taken together, the research on the complexity of 10-K filings shows that longer and more complex filings lead to bad outcomes for investors. It requires no stretch of the imagination to understand that longer and more complex explanations of company results and operations allow management teams to hide bad news, poor performance, growing debt, and other problems.
Commensurate with this, the use of certain phrases occurs more frequently in 10-K filings where the obfuscation of bad news has risen to the level of fraud.
S&P 500 Component Company 10-K Filing Length and Readability Analysis
The increasing complexity of public companies can be seen in an analysis of the S&P 500 component companies 2013 10-K filings. Some simple statistics are illustrative of the complexity in these 10-K’s:
- 79.4 percent were written at the Ph.D level;
- 19.0 percent were written at the graduate level, and;
- 1.6 percent were written at the undergraduate level.
Furthermore, 86.4 percent of all S&P 500 component companies had 2013 10-K’s of 100 pages or more.
For higher resolution versions of the charts below and additional analyses, see this.
Chart 1: 2013 S&P 500 Component 10-K Page Length and Readability Levels[xi]
A comparison of average page 2013 10-K page length by sector and industry group shows that most industry groups average well over 100 pages in their 10-Ks.
Chart 2: 2013 S&P 500 10-K Average Page Length by Sector and Industry[xii]
As we have discusseed here, 10-K page lengths have more than doubled over the past 17 years.
The increasing page lengths and Automated Readability Index levels have made understanding what individual firms do and how they do it increasingly difficult. This is important, because understanding individual firms is, in theory, easier than understanding investment vehicles that invested in individual firms.
You can find our complete roster of securities experts here.
[i] Sean, Humpherys, Kevin Moffitt, Mary Burns, Judee Burgoon, and William Felix; “Identification of fraudulent financial statements using linguistic credibility analysis”; Decision Support Systems (585-594), August 19, 2010. Available at https://tippie.uiowa.edu/accounting/mcgladrey/winterpapers/humpherys%20moffitt%20burns%20burgoon%20felix_quantitative%20methods%20for%20detection%20of%20financial%20fraud_dss%20-%20burgoon.pdf; Accessed October 2, 2014.
[ii] Rani Hoitash and Udi Hoitash, “Measuring Accounting Complexity with XBRL”; SSRN Working Paper; August 26, 2014; Available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2433677; Accessed October 2, 2014.
[iii] Mine Ertugrul, Jiaping Qiu, and Chi Wan, “Annual Report Readability, Crash Risk, and the Cost of Borrowing”, SSRN Working Paper; Available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2432797; Accessed October 2, 2014.
[iv] Id. at 17.
[v] Feng Li, “Annual report readability, current earnings, and earnings persistence”; Journal of Accounting & Economics, March 4, 2008, 221-247; Available at https://www3.nd.edu/~carecob/April2010Conference/Li%20JAE%202008.pdf; Accessed October 2, 2014.
[vi] Tim Loughran and Bill McDonald, “Measuring Readability in Financial Disclosures”, Journal of Finance, July 16, 2013; Available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1920411; Accessed October 3, 2014.
[vii] Id. at 21-22.
[viii] Kevin Moffitt, “Using Lexical Bundles to Discriminate between Fraudulent and Non-fraudulent Financial Reports”; Rutgers Accounting Seminar, February 4, 2011; Available at http://raw.rutgers.edu/docs/seminars/spring11/Moffitt.pdf; Accessed October 3, 2014.
[ix] John Carney and Francesca Harker, “How SEC’s New RoboCop Profiles Companies for Accounting Fraud, Forbes, August 9, 2013; Available at http://www.forbes.com/sites/janetnovack/2013/08/09/how-secs-new-robocop-profiles-companies-for-accounting-fraud/; Accessed October 6, 2014.
[xi] 2013 S&P 500 component company 10-K page length and Automated Readability Index level obtained from analysis of company filings.