iMpact
LOGIN
Link My iMpact  
Link Strategic Positioning Tool Kit  
To Executive Education
To Kresge Library

A Fresh Look at Forward-Looking Statements

8/19/2009 --

New research shows firms usually set the correct tone for future results in corporate filings, but the information might be buried under boilerplate.

ANN ARBOR, Mich. — Beyond the bottom line in a company's annual or quarterly report, one of the main information sources for investors and analysts is the section devoted to management's discussion and analysis. This MD&A content is one part of the report that actually may offer a glimpse into the company's future.

Or does it? The U.S. Securities and Exchange Commission (SEC), which regulates company reports, has tweaked the rules for the MD&A section to make it more informative. The concern is that these dense, detailed pages of text are too heavy on legal boilerplate and too light on relevant information.

New research by Feng Li, Ernst & Young Assistant Professor of Accounting at Ross, uses a novel computer approach to analyze millions of forward-looking statements. He found they do contain significant indicators of future performance, but that Wall Street analysts aren't able to fully reflect the information in their estimates. This suggests the SEC may have valid concerns regarding the clarity in these reports.

Li's paper, "The Information Content of the Forward-Looking Statements in Corporate Filings -- A Nave Bayesian Machine Learning Approach," also finds the information contained in these blocks of text has changed little since the SEC issued new guidelines in 2003.

"There is non-trivial information in the MD&As, and the economic magnitude is pretty big," Li says. "The analysts who cover these companies do understand most of this information, but it's interesting that they don't fully capture it. That's because the useful information might be buried under tons of boilerplate. Here, we're using a computer to read it. But a human financial analyst trying to read it might come up against 20 pages of information, and 99 percent of it is the same as last year, word-for-word. But the one percent that's different is the useful part."

The research, and subsequent follow-up initiated by Li, could help point analysts and investors in the right direction when reviewing these reports. Li's work also could serve as a useful guide for regulators.

He decided it was time to subject forward-looking statements to a serious computer analysis, given their importance to investors and concerns about usability. Until recently, the potential for serious analysis was limited by inadequate computing technology. Past studies have used dictionary-based computer programs to assess a few hundred reports.

But advances in computing technology and the wide availability of company reports online made Li's idea to use a Bayesian machine learning algorithm possible.

The Bayesian technique relies heavily on human input and sample terms, which is essential when analyzing something as specialized as the business and legal language in annual and quarterly reports. Past studies using dictionary-based programs wouldn't assign the proper values to words and phrases, Li says.

He worked with 15 research assistants -- MBA, BBA, and MAcc students at Ross -- to manually categorize about 30,000 sentences of random, forward-looking statements in MD&A copy along two dimensions: tone and content. Those coded sentences then "trained" the algorithm to categorize the tone and content of other statements in annual and quarterly reports.

The computer basically mimicked human reading, only with greater speed and accuracy. As a result, Li was able to analyze 13 million forward-looking statements from more than 140,000 corporate filings between 1994 and 2007.

He found a positive correlation with the tone of forward-looking statements and a company's future earnings and liquidity. For example, the return on assets the following year for companies with positive tone is eight percentage points higher than firms with negative tone.

The results also complement the accrual anomaly, a phenomenon researched by former Ross professor Richard Sloan. It states that when a company's earnings are inflated by non-cash benefits, or accruals, future earnings tend to fall off and the market overvalues the stock. Conversely, when earnings are based on cash flows, future earnings tend to be stronger and the market undervalues the stock.

"Our research shows that managers seem to know this phenomenon and they're anticipating the conditions of the accruals for future performance," Li says. "Combining accruals and the MD&A tone can help predict future earnings better than using accruals alone."

The research shows that Wall Street analyst forecasts were higher for companies with more positive tone in their statements, but that the information wasn't fully utilized. The statements still had predictive power for future earnings and liquidity even after controlling for the analyst forecasts.

Li's next plan for the data is to examine what types of companies have statements that are more informative. A list of specific factors such as company size, industry segment, or litigation risk could help analysts and investors.

"This could be a good supplemental information source," Li says of his future research. "You could run a program and it would give you a list of firms that you might want to give closer attention. You don't want to follow an algorithm blindly as an analyst, but it can be useful to narrow the firms you want to study. An annual report can be 150 pages. It's easy to miss information with all the boilerplate, which can be very painful to read."

Li's research also could be of interest to regulators, who have been trying to get companies to write reports in "plain English" for some time. The SEC issued new guidelines in 2003 aimed at improving the clarity of the MD&As, but Li's research shows that, so far, little has changed.

The paper continues a productive line of thought for Li. In 2006 he found firms with annual reports that are more difficult to read tend to have lower earnings.

—Terry Kosdrosky



For more information, contact:
Bernie DeGroat, (734) 936-1015 or 647-1847, bernied@umich.edu