Artificial intelligence beats Wall Street analysts in predicting financial data

Artificial intelligence can also beat professional financial analysts, according to a new statement on the MIT website. In fact, it was researchers at the Massachusetts Institute who developed new software that “significantly outperforms humans in predicting corporate sales.”

And it is precisely the financial sector that, according to experts, will be one of those most exploited by artificial intelligence. In this sector, in fact, usually you have to handle large amounts of data and make predictions based on these analyses, something that fits well with new artificial intelligence applications based on machine learning.

In the financial sector, then, the growing interest is increasingly linked to the so-called “alternative data”, i.e. those data, usually imprecise but nevertheless generated frequently and therefore always available and abundant, which can help to predict the earnings of a company. This alternative data is, for example, those relating to credit card purchases, those on location or those which, thanks to satellite images, show how many cars can be parked near a certain retailer.

These “abundant” data can then be combined with less frequent data, such as quarterly amounts, share prices, etc. The software is therefore able to have a clearer picture, obviously clearer than traditional human analysts can have, of the health of a company.

The new model, described in a scientific study published in Proceedings of ACM Sigmetrics Conference, uses anonymous credit card transactions along with other data and, in forecasting the quarterly earnings of more than 30 companies, has exceeded the combined estimates of Wall Street analysts. And the machine learning model created by MIT researchers used a much smaller dataset than human analysts had available.

“Alternative data are these strange proxy signals to help track a company’s underlying financial data,” says Michael Fleder, researcher at the Laboratory for Information and Decision Systems (LIDS) and lead author of the study. “We asked, ‘Can you combine these noisy signals with quarterly numbers to estimate the true financial data of a company at high frequencies? The answer is yes.”