Related topics: stocks, computers, Cisco, Best Buy, News Corp.
Last month's smashing victory by IBM's advanced supercomputer Watson over two human champs on the TV game show "Jeopardy!" might have some portfolio managers worried about a looming "Terminator"-like risk to their jobs.
Putting that aside, however, what if a computer attempted to rate the quality and reliability of corporate managers' responses on conference calls? Could a Watson or a Terminatrix T-X help a portfolio manager by combing through thousands of company earnings calls in search of encouraging -- and alarming -- signs?
It sounds implausible, but something like that has been developed by Business Intelligence Advisors, a Boston-based outfit that draws on staffers with Central Intelligence Agency experience.
Reading between the lines
Regular readers may remember an article that Barron's published five years ago on BIA's training of fund managers to do behavioral analysis of responses from corporate executives ("Is Your CEO Lying?").
More recently, BIA has been scaling up and mechanizing what essentially had been a bespoke practice, with clients paying for training and analysis of specific conference calls or corporate presentations.
Under that system, says BIA CEO Richard Leggett, its analysts were producing some 250 conference-call reports each quarter. Given the intense labor required and the universe of 5,000 publicly traded U.S. companies, there was no realistic way to increase the company's capabilities without adding some software muscle or recruiting a stock-savvy Watson.
Early in its history -- Business Intelligence Advisors was founded in 2001 -- as its analysts produced research reports, they instituted a system to identify and tag words and sentences (including about 270,000 of the latter) found in conference-call transcripts. From 2002 to 2009, these were put into a computerized database.
Then, Leggett says, the company developed "an algorithm that models the methodology, using linguistic pattern-recognition software." Unlike a word-search list, such programs look for response patterns that BIA says are related to certain behavioral indicators. Using methodology developed by the CIA as a noninvasive, nonconfrontational alternative to a polygraph test, the company decided to focus on 32 of them, covering a wide range of verbal and nonverbal behaviors.
Among them: behavior that indicates "a lack of confidence and uncertainty all the way to the other end of the spectrum, [such as] someone's actually trying to avoid disclosing or disclosing something that might not be correct." Despite its origins, Leggett is at pains to point out that the program -- called the BIA Earnings Call Analytics Platform -- is not equivalent to a lie detector.
Beginning in late 2010, BIA began selling its computer-generated research to analyze conference calls. Typically, these reports are produced for a client within five minutes after a transcript is available. The report initially provides what's called a decile score, ranging from 1 to 10, of the conference call, based on the 32 behavioral indicators. It also notes any key topics that raise the most behavioral concerns and provides the scores for the previous four quarterly calls.
In the ranking, each company's call is graded in relation to about 3,000 other calls for that quarter. Companies that score a 1 exhibit the least potentially problematic behavior; those with a 10, the most. The outliers, like 9's and 10's, are further scrutinized by human analysts in an additional report.
Says Leggett: "Management knows more about their business than anybody else. You can glean insights by the way executives talk about the topics of interest to the market." BIA does not evaluate the facts in the call, but rather the "quality of the delivery . . . of the commentary." It's a "filter to know where the information is highly reliable and high-quality or where it is not as reliable and you need to do more work," he observes.
The service isn't cheap. "Our relationships tend to average in the low six-figure range on an annual basis . . . depending on which services the client wants," Leggett says. "It's consumption-based and can cost considerably more."
Is it worth it?
Reading the results
A back test on the bespoke rankings conducted by Harvard Business School professor Christopher Malloy, a BIA consultant, suggests that there is a correlation between the rankings and future stock-price performance.
According to Malloy's work, the stocks of companies exhibiting the highest-concern rankings underperformed the market by about 16%, while those of companies that raised the fewest red flags tended to outperform the market by about 15%. Some 60% of the high-risk companies were laggards, and 61% of the low-risk outfits were outperformers.
A BIA-sponsored study says its conference-call rankings -- from low levels of concern to high levels -- are correlated with subsequent stock performance. Companies with low concern ratings outperformed by 15.9%, BIA says. Companies with high concern ratings underperformed by 16.1%.
