Bringing an End to Errors
Utilizing Text Analysis to Mitigate Cognitive Bias in Investing
By Alex Detmering, Director of Creative Development, Prattle, and Evan Schnidman, CEO, Prattle
The Brexit vote had just passed, and all signs pointed to a BOE rate cut on July 14, 2016. The economy was in turmoil, experts were calling for easing, and the market was pricing in a 75% chance of a cut.
One problem: the BOE held.
For many, BOE Governor Mark Carney’s July 5 speech was confirmation of the cut to come. Analysts honed in on words like, “vulnerable,” “slowed,” and “uncertainty,” pushing their cut forecast from probable to predetermined. “Governor Mark Carney sent a clear signal two weeks ago that stimulus was on the way,” reported William Schomberg of Reuters.
Not everyone interpreted Carney’s tone this way. In stark contrast to consensus, one assessment based on algorithmic analysis contended that Carney “ touted stability ” in his speech—and that the BOE would hold rates at its meeting. In the aftermath of the decision to hold, many could only scratch their heads. How could so many brilliant investors and analysts get it so wrong? And what sort of tech was behind right call? The short answers: cognitive bias and textual analysis.
Breaking Down the BOE Call
Confirmation bias is (likely) largely to blame for the incorrect forecasts of the BOE’s post-brexit policy decision. This bias can be thought of as ”the tendency for investors to seek information that supports their decision or thesis and avoid/ignore information that contradicts it.” In other words, confirmation bias encourages one to interpret evidence in favor of a conclusion one has already come to.
When one digs into news articles published in the weeks leading up to the July 14 monetary policy meeting, it seems clear that long before Carney uttered a word, many analysts had their answer. And, when he did speak, the analysts saw his words through the lens of that answer, their minds highlighting every negative word and pessimistic passage along the way.
It’s important to remember, however, that the reason analysts got this call wrong was not because they are bad at their jobs. Instead, the reason analysts got this call wrong was because they are people.
Our brains (despite what we might like to think) did not evolve to become perfect data-crunching, logic-loving machines. Because humans evolved to favor cooperation over calculation , the human mind contains a veritable storehouse of cognitive biases that it constantly calls upon during the decision-making process.
In many situations, these biases can help us. Biases focus our line of thought, often directing us towards a safe bet and always cutting down the time it takes to come to a conclusion. Biases can also help us make up for our ignorance, encouraging us to rely on the opinions others when information is scarce and action is needed.
While often beneficial or benign, cognitive biases offer shortcuts at a price that not all human endeavors can afford. Scientific research is probably the first example that comes to mind, but finance is equally (if not more) unforgiving of the mistakes our biases can cause. Cognitive biases are pervasive in human thought, and the calculations of market analysts are not immune.
It is due to this universal susceptibility that financial professionals need tools to check their biases. Although written by human analysts, the correct analysis of Mark Carney’s speech and the resulting monetary policy forecast were based on a score generated by an automated text analysis system. This system uses an algorithm to assess the historical relationship between a central bank’s language and market reaction. Using this analysis as a reference, 1 the system is then able to evaluate subsequent communications from a central bank. The resulting evaluations are unbiased, comprehensive, and quantitative scores of a particular communication’s tone.
Carney’s speech received a neutral score based on sentiment analysis, indicating the Governor’s tone was moderate, not overtly dovish.2 This, of course, contradicted the prevailing interpretation of Carney’s language. Instead of dramatically increasing the likelihood of a rate cut, this data pointed to a hold and possible rate cut if trend further deteriorated before the next BOE meeting.3 Analysts using sentiment analysis in this case were provided with an unbiased
check on their own confirmation bias, enabling them to make the right call when it was anything but popular.4
Cyborg Systems and the End of Errors
Confirmation bias is just one of the litany of cognitive biases at work in human decision-making processes.5 While not always a negative influence, these easy-to-overlook realities of human thinking ought to be recognized and compensated for whenever possible. The higher the stakes involved in a decision, the more true this principle becomes. With the trillions of dollars constantly subject to risk in the global markets, few professions are as synonymous with risk as
Now more than ever, technology is making it possible to compensate for our costly, errant tendencies. While some may fear that these advances will spell the end of human labor, the more likely scenario is cyborg systems rising to prominence. Currently the pinnacle of chess, human-machine teams employing hybrid decision-making processes hold the potential to outperform both purely discretionary and purely algorithmic investors.
The integration of text analysis technologies into investment decision-making processes represent the necessary first steps towards this future. For them, these technologies should represent one thing: an affordable means to mitigate cognitive biases, make better decisions, and, consequently, generate better returns.
1 We suggest reading the Prattle Primer to learn more about our methodology.
2 For Prattle scores, negative numbers indicate dovishness and positive numbers indicate hawkishness.
3 The data did point to neutral policy at the time. However, the subsequent dovish shift in the BOE’s sentiment suggested that a rate cut was likely at the next meeting …and that is exactly what happened.
4 Prattle has used the scores produced by their text analysis system to correctly forecast G-10 monetary policy decisions over the past year with 97% accuracy.
5 Some of the cognitive biases and limitations that have a substantial impact on investment decisions include confirmation bias, herd mentality, selection bias, and cognitive overload.
Challenges include illiquidity and fragmentation across borders.
The biannual survey enables regulators to share information and observe trends.
Fixed income managers need to show more evidence of best execution.
40% of survey respondents see similarity between digital currency and gold.
HANetf looks to lower barriers to entry in European ETF market.