Finance professionals who are familiar with fintech already know that the future belongs to people who collaborate with robots and automated processes, not those who compete with them. David Pope, CFA, managing director at Quantamental Research at S&P Global Market Intelligence, is one collaborator who is ready to claim the future.
At the 71st CFA Institute Annual Conference in Hong Kong, Pope discussed natural language processing (NLP) and how it can yield investment insights. NLP harnesses statistics and computerized data analysis to review primary data — in Pope’s examples, transcripts of corporate earnings calls — to reveal new information, say, about a company or its position relative to its peers.
The chief benefit of NLP is its ability to analyze immense amounts of data, though it remains up to individual investment managers to determine how to apply that analysis to their investment decisions. Pope’s presentation assumed that investment managers accept the possibility that NLP could aid them, and he openly acknowledged that machine analysis has its own shortcomings and drawbacks.
A Simple Indicator
NLP can identify when executives’ answers become more complicated. One measurement that can be used with English-language material is the Gunning Fog Index, which estimates how many years of formal education are needed to understand the text.
Pope uses NLP to examine every company in the S&P 500 Index and compare their executives’ language with their willingness to take questions from analysts. “We see an inverse relationship between the complexity and the number of analysts that are called on,” he said.
These findings suggest that executives hiding bad news all share some common behavior. “The more complex my answer, the more time I’m chewing up, and then I don’t have to talk to those analysts who are going to ask me awkward questions,” Pope said.
“This is why we care about language complexity,” he said. “It translates into excess forward returns.”
“As an analyst, especially during earnings season, you can only be on so many calls physically. Just one call at a time,” Pope said. However, interesting details emerge when earnings calls are examined in bulk. Comparing nuances like sentiment across all the executives in an industry, or measuring an executive’s changing attitude over time, can yield interesting insights.
Pope listed three available methods of sentiment analysis: “Bag of Words,” N-grams, and word embedding. On the extreme end of the spectrum, word-embedding analysis compares the proximity of every word in a text to every other word in the text. For earnings calls, this sort of analysis requires exponentially more processing power since the typical earnings call transcript has about 500 billion possible combinations in it, according to Pope.
In his own work, he favors the “Bag of Words” technique, which requires much less processing power. Usually, each method provides the same result, although the more computationally intense efforts can yield greater insight. The question that analysts must answer is whether they want to perform a deeper examination on fewer companies, or take a less detailed look across a broader range of firms.
NLP’s greatest value comes from its ability to identify outliers quickly. “It could narrow your search space,” Pope said, helping identify the companies they should focus on.
Tesla CEO Elon Musk’s bizarre earnings call was brought up as an example of aberrant executive behavior, and Pope noted that NLP can also help answer questions like “How unpredictable is he being, compared to his recent past?”
Experts like Kate Darling of Massachusetts Institute of Technology (MIT) Media Lab have observed that increased automation is changing human behavior, and Pope’s work backs this up. “Management adapts quickly,” he said, noting that at least one consulting firm already uses earnings call transcripts to coach executives on how to behave.
This could lead to endless cat-and-mouse games as analysts search for new ways to probe for flaws and management find new ways to conceal them. Pope has already gained insights by comparing the polished prepared executive remarks against the improvised sentiments expressed during the less-structured Q&A session.
Whether NLP can help with lie detection has yet to be determined, Pope said. “I probably wouldn’t be standing here talking to you,” he explained, if NLP could easily spot deceit. He’d be back in his office trading on those signals.
Pope challenged the idea that algorithms could not decipher context or culture. “I think algorithms can,” he declared. “I think there are different contexts for different cultures.”
Of course, it’s up to the person performing the analysis to set the parameters that generate useful insights.
“One of our philosophies in our work is to be transparent,” Pope said. He prefers to confine his analysis to publicly available resources. He referred the audience to a Natural Language Processing Primer, available on the S&P Global Market Intelligence website, as a way to learn more about the techniques he applies. With it, he said, you “can be up and running in a matter of hours.”
Experience the 71st CFA Institute Annual Conference online through Conference Live. It’s an insider’s perspective with live broadcasts and recorded video archives of select sessions, exclusive speaker interviews, discussions of current topics, and updates on CFA Institute initiatives.
All posts are the opinion of the author. As such, they should not be construed as investment advice, nor do the opinions expressed necessarily reflect the views of CFA Institute or the author’s employer.