Bots for Business

The Chatbot That Wants to Help Us Make Smarter Financial Decisions

Given the general hoopla about “data being the new oil,” it should perhaps come as no surprise that it’s held in extremely high regard in the financial services sector. Indeed, Thomas Egner, the Secretary General of the Euro Banking Association, recently called data the superpower of the financial services world.

By Adi Gaskell
October 19, 2021

Given the general hoopla about “data being the new oil,” it should perhaps come as no surprise that it’s held in extremely high regard in the financial services sector. Indeed, Thomas Egner, the Secretary General of the Euro Banking Association, recently called data the superpower of the financial services world.

Even as long ago as 2017, J.P.Morgan revealed that asset managers were investing around $3 billion on alternative data each year in a bid to ensure that their decisions are infused with as much information as possible. They went on to suggest that the number of analysts working with alternative data had quadrupled in the last five years.

An information advantage

The potential for data to help us make smarter investments was underlined by a study from the School of Business and Economics at Friedrich-Alexander-Universitat Erlangen-Nurnberg. The researchers trained a deep learning algorithm on around 180 million distinct data points that were collected about the various members of the S&P 500 over 22 years.

The idea was to see if the algorithm could make better stock picks than human experts. The results found that the AI-based system typically secured double-digit returns, with its performance especially strong when the economy was experiencing high levels of turbulence and uncertainty.

Similarly, robust outcomes were achieved by a second study, this time from the University of Plymouth, which highlighted how AI can be used to help commodity traders predict changes in the oil price with an extremely high degree of accuracy.

Research from the University of Toronto illustrates the potential of using rich datasets when making investment decisions. The researchers harvested around a million tweets, each of which mentioned one of the 3,600 companies analyzed in the study. The researchers used the tweets to perform textual analysis, with a machine learning algorithm then predicting whether the company concerned would meet their quarterly earnings target or not. Not only was it able to do so with a high degree of accuracy, but it even accurately predicted how the share price would respond to the event.

Data-driven decisions

All of which must mean that financial service companies are using data in pretty much everything they do, right? Yes and no. Research from PwC actually found that financial services firms were not using a great deal of data at all when they worked with customers.

While a lack of a data-driven culture played a part in this, a bigger problem was that simply having a mountain of data wasn’t enough to actually drive data-driven decision-making. Indeed, in many instances, huge quantities of data made it less likely that it would be used to drive decisions, as making sense of the data was unduly difficult.

As a result, traders and executives were striving for data that was not only plentiful but especially timely and relevant to the decisions they needed to make. Increasingly this includes the kind of alternative data sources mentioned earlier, which is typically unstructured and nonfinancial data that can be harvested from things such as social media feeds, weather forecasts, satellites, and the news.

Enabling data-driven decision-making

As research from the University of Mannheim illustrates, we often rely on our gut instinct when making decisions, even if those decisions are finance related. British fintech start-up Quant Insight is trying to make sure that such instinctive decision-making still has a robust and reliable evidence base to it.

They have developed a chatbot, which is hosted on AWS and accessed via Slack, that aims to allow users to get meaningful, timely, and actionable insights regarding particular stocks. The chatbot works closely with data from Now-Casting, alongside traditional financial data from companies such as Bloomberg, to develop bespoke models for each client's unique requirements. On average, the models cover around 30 variables and 6 million data points to help provide a financial recommendation for any particular asset.

Each recommendation is accompanied by what the team refers to as a model confidence score, which is essentially the R-squared for the recommendation, with the higher the score, the higher the confidence in the recommendation.

The service is not designed to automate the process of buying stocks but rather to provide traders with a means to locate the various insights that can guide them in their trading behaviors. For instance, they're working on a system whereby traders can "subscribe" to a particular stock, and the chatbot will monitor all of the data pertaining to that company and notify the trader if anything will elicit meaningful change in the company's share price.

"That's a perfect use of the technology, as it allows us to do things that simply wouldn't be possible through other ways," Mahmood Noorani, CEO of Quant Insight, told me.

As Olivier Sibony writes in his bestselling book Noise, wherever there is judgment, there is noise, and more of it than we think, and so the chatbot aims to dampen some of that noise and reduce the variance in financial decision-making by traders.