London, October 20 (BUS): Executives, beware! You can become your own worst enemy.
CEOs and other managers are increasingly under the microscope as some investors use artificial intelligence to learn and analyze their own language and language patterns, opening up new horizons of opportunity for slippage.
In late 2020, according to pattern language software specialist Evan Schneidman, some IT industry executives were downplaying the potential for shortages in semiconductor chips while discussing supply chain disruptions.
They said everything is fine.
However, their tone of speech showed high levels of uncertainty, according to a computational analysis designed to discover clues hidden in spoken — not perfectly written — words.
“We found that the tone of IT executives contrasted with the positive text sentiment of their feedback,” said Schneidman, who advises two of the fintech companies behind the analysis.
Within months of the comments, companies like Volkswagen and Ford warned of severe chip shortages hurting production. Stock prices in automobiles and industrial companies fell. IT executives now said there is a supply squeeze.
Schnidman finds that computer-driven quantitative boxes of scores assigned to the tone of managers’ words, versus scores assigned to written words, were in a better position before the industry turmoil.
Not a single example, though, can attest to the accuracy of speech analysis, because we don’t know whether CEOs were overly optimistic at first or honestly changed their views as circumstances changed.
However, some investors see the technology – known as natural language processing (NLP) – as a new tool to gain an advantage over competitors, according to Reuters interviews with 11 fund managers who use or run such systems.
They say that traditional financial and corporate data are so heavily mined nowadays that they offer little value.
Neuro Linguistic Programming (NLP) is a branch of artificial intelligence where machine learning loses language to understanding, then transforms it into quantifiable signals that affect quantitative money in circulation.
The field’s most ambitious program aims to analyze vocal tones, rhythm, and emphasize spoken words along with phrases, while others look to analyze speech and interview texts in increasingly complex ways.
Slavy Marinov, head of machine learning at MAN-AHL, part of the $135 billion investment management firm Man Group, told Reuters that NLP was “one of the main research areas focused on” in the computer-driven fund.
“These models turn something very chaotic into something that is easy to understand by quantum,” he said.
Indeed, advocates say NLP can unleash the untapped potential of insight from the world of “unstructured data”: calls with analysts, unrecorded questions and answers, informational interviews.
This is open to debate, though.
These AI systems can cost millions of dollars to develop and operate, which leaves out many investors and developers who save on the wealthy or specialists. Some are also relatively in beta, with no publicly available data to prove they make money. The funds interviewed declined to show evidence that NLP can increase returns, citing commercial sensitivities.
Some studies suggest that technologies can boost performance if you focus on smart places.
Analysis by quantitative strategists at Nomura in September showed a link between the complexity of CEOs’ language during earnings and equity calls. US presidents who use simple language have seen their company’s stock outperform 6% annually since 2014, compared to those using complex language.
Bank of America analysts use a model that uses statements in earnings calls to predict corporate bond default rates. This examines thousands of phrases such as “cutting costs” and “burning cash” to find phrases associated with defaults in the future. BofA said backtesting the model showed a high correlation with hypothetical probabilities.
In years past, language processing in finance has been characterized by basic and widely sold software that categorizes news or social media posts by sentiment. This loses value in the face of increasingly sophisticated NLP paradigms, spurred by technological advances and the falling costs of cloud computing.
The breakthrough came in 2018 when the developers released the source code behind NLP “Transfer Learning,” which allowed the model to pre-train on one data set of words and then run it on another, saving time and money.
The AI team at Google has since released the code behind many advanced models pre-trained on ever larger data sets.
The developers of current systems say that they process tens of thousands of words at lightning speed, extracting patterns and determining the degree to which they relate to certain important “raw” words, phrases and ideas, as identified by the user.
MAN AHL’s Marinov sees an advantage in tonal analysis but has not yet used it, focusing at the moment on clues hidden in written text.
This can be anything from comparing annual reports over time to looking for subtle changes that aren’t obvious to the reader, to identifying something intangible like company culture.
Few investors have attempted to formally measure corporate culture in the past even though it is essential for long-term performance, particularly in the area of ESG.
The Man AHL form can scan CEO feedback to search for words or phrases that demonstrate a “goal-driven” culture, as well as search through employee ratings on the job site Glassdoor.
Kai Wu, founder of hedge fund Sparkline Capital, creates “personal profiles” of companies to gauge their adherence to certain cultural values.
He chooses the primary words he thinks reflect such values. His NLP model then reduces huge amounts of words to small numbers of words with similar meanings, with the results expressed numerically.
Using his NLP model in management comments and employee reviews, he found that companies with “privacy” cultures like Apple, Southwest Airlines and Costco outperformed.
Conversely, US companies that exhibit “toxicity” — where employees use specific terms such as “good boys club” and “dogs eat dogs” — underperform significantly, Wu said.
Money without the resources to hire data scientists to build their own NLP tools can buy into analysis from outside companies, such as those recommended by Schnidman — fintech Aiera and color analytics provider Helios Life Enterprises — that sell their services to clients such as hedge funds.
However, Sparkline’s Wu sees the money as getting NLP-derived data “as close to raw as possible”, favoring in-house models.
Technology faces other challenges, and implementing it properly can take a long time.
Dutch principal NN Investment Partners uses a mixture of third-party data and its own models, some of which are still in the research stage.
One of the projects is training a model to find words that predict default rates, said Sebastian Reynders, head of investment sciences at NNIP. This initially required portfolio managers to scan long lists of statements to manually categorize them as positive or negative.
Most of the exhibitors focus on the English language, and the developers may have a difficult task adapting them to accurately read sentiments from people of different cultures who speak other languages.
Moreover, the CEOs wear cotton.
When George Mosley, chief investment officer at US-based PanAgora Asset Management, told the president of a biotech company that the AI in his fund had scanned CEO comments for passwords, the person asked for a list to help his company rank higher.
Mosli rejected the request but said documents such as transcripts of earnings calls were increasingly “well-written,” undermining their value.
However, Man Group’s Marinov believes that CEOs will ultimately not prove to be up to machines that get better with more data.
“There are no rules, it’s like a self-driving car that learns as it goes,” he added. “So in many cases it is impossible to give the executive a list of passwords.”
HF