Communication from corporate management teams is being subjected to an increasing array of investigative methods that can detect, if not an outright lie from a CEO or CFO, a management deception that an investor can use to their advantage. Think poker and some sign that the opposition has unwittingly revealed its hand, whether that be positive or negative is less important than what you can do with the information.
Investors are already using algorithmic textual analysis, CIA lie-detection techniques, and more recently, audio analysis of management speech, to seek an edge with stock calls, sector sentiment and overall market direction.
The sheer amount of data, in this case words, that massive computing power can now store and layer will only increase. That includes text of shareholder letters, investor day presentations, management commentary prepared for earnings, and most importantly, earnings Q&As during which stock analysts are able to ask questions. Tens of thousands of earnings calls with CEOs and CFOs have already been recorded and crunched by human analysts and algorithms in the search for markers.
Cheryl Cook, CEO of Business Intelligence Advisors, which has used a Central Intelligence Agency lie detection process on behalf of investors for approximately a decade, said it has analyzed over 11,000 conference calls and one-quarter of a million earnings Q&A exchanges between management and analysts. BIA analyzes on average 200 conference calls per quarter for as many as 40 clients. "There is this huge body of content out there produced every quarter and it gets harder and harder to get an edge and investors are looking for every opportunity," Cook said.
"It's everything from discomfort and non disclosure to misleading answers," Cook said. "We think of it in terms of how much risk is in the communication. Is it reliable and complete, and if less than reliable and complete, we look for patterns in language and behavior and extrapolate about what is driving that risk, which leads to investable decisions."
BIA was founded by former CIA officers and staffers are still recruited from the agency. "That's our DNA," Cook said.
David Larcker, a James Irvin Miller professor of accounting at Stanford University's Graduate School of Business, who pioneered research in this area, said that whether a question posed by an analyst on a call is actually answered by management "is a core tell."
Research conducted by Larker showed that stocks associated with CFOs that scored high on a deception index tended to have reversals over 1,3, and 5-year periods.
Larcker published a paper five years ago that looked at the transcripts of thousands of corporate earnings calls and then compared the ones made by CEOs who later were found to lie (people like Sanjay Kumar of Computer Associates and Kenneth Lay of Enron; they were both convicted of securities fraud).
Read More Insider trading: CNBC Explains
Larcker found some three markers for a deception:
- Failure to directly answer a question. For instance, Kumar answered a question about whether the company's books could be trusted by providing an assurance about the quality of the auditor.
- Using words like "my team" and "we" more than "I."
- Using exuberant words, like "amazing" or "awesome."
Larcker is continuing to research an algorithm to detect when management has not answered the question asked by analysts, but outside the academic arena, companies like New York City-based research firm Valens Securities are moving further out on the technology frontier of deception-detection. Valens processes audio feeds from the records of conference calls and creates an analysis that maps 15 different markers that have to do with either strong negative or positive emotional effect in audio communication. The markers help Valens decide whether a CEO or another member of the management team is overconfident or underconfident about a particular business situation.
"An Electro Audio Gram (EAG) is almost like having a CEO plugged into a machine measuring hesitation and change in tone and and speed of speech," said Joel Litman, Valens CEO. Valens has focused on audio technology because management can learn to tweak their written statements, adjust body language and hide nervous habits, but it can be "very hard to adjust a tightening or loosening in one's vocal cords and what one feels in their gut."
Litman added, "You can train a CEO to say they are excited about a new product launch or they can speak in a monotone and control hand gestures, but to control their stomach from getting butterflies is impossible."
Larker said it’s not surprising hedge funds have hired linguists and others to build algorithms and conduct natural language research. "Everybody is looking for an edge," said Larcker. "Any communication that is coming out of a company now is being subjected to linguistic analysis."
“Whether you’re looking for excess value or risk mitigation, if an investor can get it by running an algorithm across transcripts … that’s huge,” Larker said. He said that’s especially true when taking into account that an equity manager may have a universe of 800 stocks that pass a stock screen and still need to toss 750 out. “It makes sense to say, ‘here are the ones that have a high score on deception. I don’t need to know which one is lying, but I can capture some of the bad ones and delete them from the screen,’” Larker explained. “That helps protects you from torpedoes,” he said.
Australia-based Credit Suisse equities analyst Richard Hitchens believes the investors poised to benefit the most right now are quantitative investment managers.
“A shortcoming of pure quant investment processes is that they typically lack direct contact with company management,” Hitchens wrote in a survey of the new approaches to management analysis published late last year. Hitchens explained, “A primary nonquantitative piece of information picked up on in their direct management dealings, that can then influence their investment decision, is the tone of the language being used, i.e. how positive or negative it is, and in particular how it has changed relative to their most recent previous dealing.”
Larker said textual analysis has a longer track record because it's been the easiest approach. The reason the older efforts have focused on transcripts is that they are easy available, but as investors go further down this path it's a combination of words, voice, and images that will all be used together to make the data "speak."
Hitchens predicts that "picking up on investment signals from the use of voice and body, e.g. voice inflection, speech volume, speech rate, speech errors, facial expressions, hand gestures, posture, and movements, will likely remain a key competitive advantage for fundamental investors for the foreseeable future."
Valens has analyzed 18,000 calls.
For instance, Facebook looked expensive late last year. "When we ran the earnings call forensics ... management showed high levels of excitement. They were very confident in what they were saying," Litman said.
The stock was trading around $75 in the last two quarters of last year; it's now up around $90.
On the flip side of the coin, Valens said, Google executives have been displaying doubt in their conference calls.
"There's hesitation and the way they are changing their tone. ... We're getting these highly questionable flags."
In particular, Valens said, Google CBO Omid Kordestani was particularly hesitant about the company's conversion to mobile and how much it would damage the companies that rely on Google; Valens flagged him for downplaying the extent to which non-mobile-friendly content providers will be penalized, and overemphasizing the boost to e-commerce expected to come from mobile.
Google shares are more or less flat over the past year, though shares have rebounded year-to-date.
Just as most investors wouldn't use the content of a conference call alone to make a buy/sell decision, you wouldn't use these kinds of technologies in isolation.
Proponents of the approach don't position the technology as its own investing discipline, but rather, an additional tool for investors with an already well-defined philosophy about stock-picking. For a hedge fund that has its own investing approach, BIA becomes a "part of the mosaic," Cook said, and importantly, uncorrelated to traditional research methods.
"We don't know whether the client is long or short … sometimes our insights increase their conviction and sometimes it tells them something new," Cook said.
Ten years ago it would have been a short only looking to find the stocks that would collapse, Litman said, but now it's becoming a more widespread approach for investors long stocks and Valens is being asked to review an entire portfolio.
""It really depends on the investor having a good process already in place," Litman said.
At the recent Exponential Finance conference in New York City, Exponential Finance, Google-funded Singularity University founding executive director Salim Ismail, a former head of innovation at Yahoo, said with nonchalant confidence that CEOs will not be able to lie in five years. The forensic accountants and analysts actually doing this work today don't believe a proclamation that bold is necessary to prove the current value of their approach to corporate analysis.
"We do not believe people will be talking about this as a 'niche' five years from now," Litman said. "We have to audit numbers but we also have to audit the words. Auditing the story will be common."
This doesn't imply an infallible lie detector. "Will it be possible for a CEO to get away with a major piece of deception on a call in 5 years? I don't think so, but the difference between lie and deception is deep and important," Litman said. "You can detect a deception. It's very difficult to know what a lie is. You only need know they are being deceptive about something. Words don't have to express what the lie is exactly."
Cook said more discussion of deception analysis is leading to more acceptance and adoption is increasing; it's now becoming a question of how firms choose to incorporate it. But added, "It is a nuanced issue emotionally and mentally. Very few people tell outright lies." The data itself falls typically falls short of investable conclusions." "Machine learning gets better and better, but I don't believe that it will ever get to point where that is enough to say, 'I want to pick this stock, make this bet.'" Cook said.
Larker said there isn't enough academic research yet for him to have confidence in any deception-detection technique that would claim 90 percent accuracy or higher. But he also said, "You don't have to get 'dead-on' good guy or bad guy. You don't have to predict tomorrow's stock price. Other things matter."
Litman put it in the most simple terms: "It's not what people say; it's how they say it."
—By Elizabeth MacBride, special to CNBC.com