Just after lunch on a Friday afternoon, a professional trader sends instant messages to several colleagues about a big tech-stock bet he is about to make. The messages suggest that he is very excited about the trade, but the transaction ultimately results in major losses for his firm.

Could those losses have been avoided if the trader had a better understanding of his own emotional exuberance and how it may be affecting his judgment? And could his instant messages have provided that understanding?

Possibly, according to new research by Brian Uzzi, a professor of management and organizations at the Kellogg School.

Uzzi, along with Bin Liu of Google and Ramesh Govindan of the University of Southern California, wanted to know if people’s electronic communications could offer clues about their emotional state. If so, this could be a powerful tool, as emotions are understood to affect the quality of our decision making.

The researchers analyzed millions of instant messages (IMs) and trades from 30 traders over a two-year period. They observed that the best trading decisions, and highest profits, happened when traders showed a moderate level of emotion in their IMs—that is, not too much, and not too little. Beyond simply improving trades, this finding could help anyone tasked with making risk-related decisions, from air traffic controllers to humanitarian aid groups.
The Rise of Unstructured Data
The study addresses an entirely new way to make informed decisions.

“The architecture, theory, and practice of finance revolve around analysis of quantitative data— expressed in things like balance sheets, income statements, prices, and analyst reports—to help understand where to place investments,” Uzzi says. Financial firms and researchers “have figured out how to squeeze every little bit of insight out of structured numerical data.”

But these days, structured data are not the only game in town. Over the last decade, emails, text messages, and IMs have provided a deluge of available unstructured data.

“The explosion of unstructured data represents the next frontier of information,” Uzzi says. “Because traders are among the many professionals who routinely communicate electronically, we thought it would be of value to analyze the data they generate for insights.”
We Need to Talk—About Risky Decisions
The researchers suspected that IMs could provide a snapshot of traders’ emotional states as they made trading decisions.

“When people set out to do something risky, they actually like to talk to other people before they do it,” Uzzi says. “They chat with others to get a sense of what other people think is risky and incorporate that into their own decision making.”

But most of us take an oblique approach to soliciting such feedback. “People want to appear as good decision makers, and in trading, specifically, there is a need to keep your actual trading secret,” Uzzi says. So, traders fish for information within their social network without being overt about their intentions. For example, when asking what others think, they can reveal their emotional state in their choice of words. “Do they call the market for a stock ‘changeable,’ ‘shifting,’ or ‘volatile?’ Each word reveals a different level of emotional activation, and when many words over many instant messages are combined, you can develop a revealing picture of the trader’s emotional state of mind.”

Thus Uzzi and his coauthors propose that when people talk about risky things, they use language that evokes their emotions, even if only subconsciously. “Most of us tend to be unaware of our own emotional states, unless we’re raging mad or euphoric,” Uzzi says. “So traders may experience a range of emotions that are either beneficial or disadvantageous to their trades without realizing it.”

How might emotions be a boon or hindrance for traders?

Previous research by others has shed some light on how emotions affect decision making. One study compared two groups of people—those who had experienced brain trauma that prevented them from experiencing emotion and those who had not experienced such trauma—on a range of mental tasks.

The study found that while both groups could perform quantitative and other rational calculations, only the non-brain-damaged participants could make risk-related decisions easily. This suggests that people need an emotional response to push them toward one choice or another when risk is involved.
The Just-Right Level of Emotion
Observations of traders at work, as well as lab-based studies, confirm that emotions matter. If traders are “too emotional,” says Uzzi, “they may make poor decisions; if they’re not emotional enough, they may be too slow to make decisions.”

But no one had previously tested whether digital communications could be used to measure people’s emotional state.

To do so, the researchers analyzed all 886,000 trade-related decisions and 1,234,822 IMs from 30 professional day traders over a two-year period. They also tracked each trader’s median daily profit—the best measure of trading performance. IMs were coded for level of emotion, based on the specific words used. For example, “nice,” “gold,” and “hit” were associated with moderate levels of emotion.

The study showed that traders were more likely to use IMs when making trades—meaning they were eager to talk about their risk-related decision—and that the emotion expressed in IMs correlated with profitability. As predicted, traders made the highest-quality decisions at a moderate level of emotional activation.

Excessive emotion is problematic. “If you’re over-activated, your emotional state is drawing cognitive resources away from the analytical brain,” Uzzi says. “You then fail to attend to the right information or your perception of the information becomes distorted.”

But perhaps surprisingly, zero emotion is also not ideal. “We show that unless you reach a certain level of emotionality, you can’t pull the trigger on the trade, and you don’t buy the stock at just the right time,” Uzzi says. The finding goes against conventional wisdom that an emotionless state—the proverbial poker face—is best for decision making in business. As the authors note, even Warren Buffett emphasizes controlling emotion in investing, rather than channeling the right amount of it.
Training a Better Trader
The findings point to valuable practical applications within the red-hot field of financial technology (FinTech).

One possibility is to provide traders with feedback about their current emotional state, to help guide decision making in the face of risk. “Based on their emotional state, they can decide whether to lean into a trade or lean out of it,” Uzzi says.

Traders could also benefit from understanding factors that influence their emotions. “Companies could develop a very sophisticated system that analyzes electronic communication to tell a trader, ‘Every time you talk to Joe, you get overly emotionally activated, and that might drag you toward a bad trade,’” Uzzi says. “Or just the opposite: ‘Every time you talk to Nancy, it seems to sap you of all emotion and you might be too slow on your trades.’ Or that after lunch and on Friday afternoons your emotional activation is not optimal.”

Additionally, traders could use this feedback to train themselves to modulate their emotions, learning how to nudge them toward the optimal state for high-stakes decisions. At the organizational level, a smart system could help identify the employees best suited to the emotional elements of their work.
And the benefit of this sort of analysis extends well beyond traders.

Air-traffic controllers make hundreds of high-stakes decisions daily, and understanding how their emotions influence their judgment could potentially save lives. So do emergency response personnel and those in the military.

“Military and security groups deal with highly emotional decision making,” Uzzi says. “Do I put 100 boots on the ground? A thousand? When do I pull them out? I need to be in the right emotional state to make those decisions.”

Many organizations are already working on smart monitoring systems, Uzzi says. Some are using widely available, unstructured digital data such as Twitter posts to “capture the mood of the masses,” Uzzi says. His research could help optimize such systems. As the authors write, “[T]he tracking of IMs or Tweets during an evacuation or natural disaster could indicate which targets of aid are least likely to make good decisions.”

Finally, these findings could prove useful to entrepreneurs who aim to serve organizations that care about risk-related decision making. Entrepreneurs might tackle whether wearable devices can capture emotional indicators, for instance, or how emotion-related feedback can most effectively be displayed on computer or phone screens.

“There’s lots of blue sky here for creating ventures that take advantage of this science,” Uzzi says.