CEOs’ language under ML microscope

Executives, beware, you could become your own worst enemy.

CEOs and other managers are increasingly in control as some investors use AI to learn and analyze their language patterns and tone, opening up a new frontier of opportunities to make mistakes.

At the end of 2020, according to language modeling software specialist Evan Schnidman, some IT executives were downplaying the possibility of semiconductor chip shortages while discussing supply chain disruptions.

Everything was fine, they said.

However, according to an algorithmic analysis aimed at identifying hidden clues in ideally spoken, unwritten words, the tone of his speech showed a high degree of uncertainty.

“We found that the tone of IT industry executives did not match the positive textual sentiment of their comments,” said Schnidman, who advises two fintech companies behind the analysis.

Months after the comments, companies like Volkswagen and Ford warned of a serious chip shortage affecting production. Auto and industrial companies’ share prices fell. IT executives have now said there is a reduction in supply.

Schnidman argues that before the turmoil in the industry, computerized quantitative funds that access values ​​assigned to the tone of the words of managers would have been better positioned compared to the values ​​assigned to written words.

However, an example cannot confirm the accuracy of the speech analysis because we do not know whether the executives were initially too optimistic or whether they changed their minds sincerely as circumstances changed.

However, some investors see the technology known as Natural Language Processing (NLP) as a new tool to gain an edge over competitors, according to Reuters interviews with 11 fund managers who use or test such systems

They say traditional financial data and business statements are so exploited these days that they offer little value.

very messy”

NLP is a branch of artificial intelligence in which machine learning is used on language to make sense of it and then turn it into quantifiable signals that quantitative backgrounds represent for trading.

The most ambitious software in this field aims to analyze audible tones, cadence and accent of spoken words as well as phraseology, while others seek to analyze transcriptions of speeches and interviews in a more in addition to sophisticated.

Slavi Marinov, head of machine learning at Man AHL, which is part of the $ 135 billion investment management firm Man Group, told Reuters that NLP is “a major research area” of the computer-driven fund. .

“These models turn something very messy into something that is easily understood by a quant,” he said.

Proponents of Effect claim that NLP can unleash the untapped information potential of the “unstructured data” world: calls with analysts, unwritten questions and answers, interviews with the media.

These AI systems can cost millions of dollars to develop and operate, preventing many investors and developers deep-pocketed or niche. Some are even at a relatively experimental stage, with no publicly available data to prove that they are making any money. Funds interviewed declined to prove that NLP can increase returns, citing commercial sensitivities.

However, some studies suggest that the techniques could increase performance if concentrated in smart places.

September’s analysis by Nomura’s quantum strategists showed a link between the complexity of the language of leaders in revenue calls and actions. bosses who used plain language have seen their company’s stocks outperform by 6% per year since 2014, compared to those who use complex words.

BofA analysts use a model that uses profit call phrases to predict corporate bond default rates. This one examines thousands of phrases such as ” costs cutting” and “cash burn” to find phrases associated with future defaults. Backtesting the model showed a strong correlation with the probability of default, BofA said.

Both systems analyze the transcripts.


In recent years, finance language processing has characterized the basic and widely sold software that ranks news or social media posts based on sentiment. This is losing value in the face of increasingly sophisticated NLP models, driven by technological advances and falling costs of cloud computing.

The breakthrough came in 2018 when the developers released the source code behind NLP’s “transfer learning”, which made it possible to pre-train a model on a dataset of words and then put it in implemented on another, saving time and money.

Since then, Google’s artificial intelligence team has released the code behind several previously trained top models on ever larger data sets.

Today’s systems designers say they can process tens of thousands of words at lightning speed, extract patterns, and quantify how closely they are related to certain meaningful “starting” words, phrases, and ideas set by the user.

Marinov from MAN AHL sees merit in sound analysis, but has not yet used it and is initially concentrating on hidden clues in written texts.

This could be anything from comparing annual reports over time to look for subtle changes that are not obvious to the reader, to quantifying something as intangible as corporate culture.

Few investors have attempted to formally measure corporate culture in the past, although it is critical to long-term performance, especially in the hot ESG investment sphere of environmental, social and governance considerations.

Few investors have attempted to formally measure corporate culture in the past, although it is critical to long-term performance, especially in the hot ESG investment sphere of environmental, social and governance considerations.

Man The AHL’s model can analyze executive comments for words or phrases that demonstrate a “goal-oriented” culture, as well as search for employee reviews on the Glassdoor Jobs website.

Kai Wu, founder of the hedge fund Sparkline Capital, has created “personality profiles” for companies to measure their adherence to certain cultural values.

Select keywords that reflect these values. His NLP model then reduces large volumes of words to a small number of words with similar meanings, with results expressed numerically.

Using his NLP model of management feedback and employee reviews, he found that companies with “idiosyncratic” cultures like Apple, Southwest Airlines, and Costco were outperforming.

In contrast, American companies exhibiting “toxicity,” where employees use phrases such as “good ol boy’s club” and “dog eat dog,” underperformed, Wu said.


Funds without the resources to hire data scientists to build their own NLP tools can buy in analysis from third-party firms, like those Schnidman advises – fintech Aiera and tonal analytics provider Helios Life Enterprises – which sell their services to clients such as hedge funds.

Wu believes that funds should receive NLP-derived data “as close to raw data as possible,” with internal models being preferred.

Technology faces different challenges and can take a long time to get right.
Dutch manager NN Investment Partners uses a combination of third-party data and its own models, some of which are still under investigation.

A project trains a model to find words that predict bond default rates, said Sebastiaan Reinders, director of investment science at the NNIP. However, portfolio managers initially had to go through long lists of phrases in order to manually mark them as positive or negative.

Most models are focused on English and developers could face a difficult task adjusting them to read the moods of people of different cultures who speak other languages.

Plus, the executives are thrilled.

When George Mussalli, chief investment officer at U.S.-based PanAgora Asset Management, told the director of a biotech firm that his fund’s AI analyzed executive comments for watchwords, the person asked for a list to help their business rank higher.

Mussalli denied the request, but claimed that documents such as income call transcripts were increasingly “well scripted”, undermining their value.

Still, Man Group’s Marinov believes executives will eventually fall short of machines that improve with more data.

“There are no rules, it’s like an autonomous car that learns as you go,” he added. “So in many cases it is impossible to give the manager a list ofwatchwords.”

Most Popular