Artificial Intelligence’s New Wave on Wall Street… But Is It Fair?

Blog entry

By: Carol Ozemhoya, writer at Indexer

A good trader will do the requisite due diligence and research to make sure the investments traded on behalf of clients will maximize profits. Over the years, analytical tools have evolved and developed until today, where machine learning models can come into play and offer startling insights.

It can make for an uneven playing field when one trader or financial analyst has more information than the others. They all have access to public information provided in analyst reports, but how about when the information asymmetry is a direct result of machine learning algorithms?

According to, insider trading is illegal “because a single person with information not readily available to other investors has an unfair advantage.” In other words, because share prices should reflect all publicly available information available about a company, if a small group of investors trade on material nonpublic information, the integrity of the markets will be damaged and investors may lose confidence and stop trading. Insiders with non-public information will be able to avoid losses and benefit from gains.

The deeper question is: could machine learning (a form of artificial intelligence) produce a form of material nonpublic information? To this end, “growth in A.I.-directed investing could have radical consequences, especially in a scenario where a single investor or investment fund using proprietary A.I. is able to secure an unfair advantage over other market actors. Call it “stock market singularity.” And the groundwork for such an occurrence has already been laid,” reports

Classic Econometrics Vs. Machine Learning

Econometrics focuses on explaining variables we call Y using factors we call X. For example, Y equals the price of a house. X might represent the factors we use to explain the price of a house, such as the location (X1), the square footage (X2) and the age of the home (X3). Basically, X explains the price of the house.

Alternatively, machine learning is a scientific discipline that explores the construction and study of algorithms that can learn and decipher patterns within data.

Machine learning focuses on the accuracy of predictions. For some hedge funds, obtaining accurate predictions is highly attractive and in some cases, it’s kind of like flipping a weighted coin… it’s an easy win.

The Human Factor

No doubt, using econometrics brings in some important aspects – human instinct and human error. Consider this: How many times have you heard the phrase “you can’t teach that?” The point is, no matter what an econometric model pumps out, unlike A.I.-based systems, these predictions are still linked to the human mind’s uncanny ability to master the subject matter and to make adjustments that limit modeling error. Furthermore, these skills are not transferable.

Computers are faster, but brains can produce amazing results as well. Can the two merge and create a “super analyst?” Or do those who uses machine learning have a distinct advantage?


Countries and companies around the world use financial markets to trade, source capital and generate profits. So, this poses the question: will those global players who can afford the technology hold a distinct advantage over those still using non-A.I. systems?

At the same time, some experts say machine learning is still a concept not widely employed by “fundamental investors,” and any concerns aimed at its use are too early. Indeed, machine learning skills in market price prediction are uncommon in today’s work force, and that alone may threaten the information balance necessary for financial markets to function. Per Quora: “…machine learning techniques are still very far from being part of the menu of tools widely used and accepted in the field.”

Another point being made in artificial intelligence circles is that some machine learning models still have their kinks. For example, points out Susan Athey in Forbes recently: “The foundation of supervised ML methods is that model selection (cross-validation) is carried out to optimize goodness of fit on a test sample. A model is good if and only if it predicts well.” The point is different markets require different data, so models must be created that can easily accept different variables.

Keeping It Simple…

What would an A.I. trading specialist look like? Imagine you had a “specialist” that offered near perfect predictions of copper prices.  Alternatively, another might generate highly accurate predictions of wheat prices. Another might be great at predicting M&A activity in an industry based on commodity price regimes. The list goes on and on. But, what negative externalities may ensue?

For example, imagine we have ABC Company, which specializes in making bread. ABC has an A.I. system which specializes in predicting the price of wheat. The accuracy of the predictions produced by the system has an error rate of only 5 percent if predicting one year forward. If fact, ABC’s system predicted wheat prices will double in one year’s time.

Let’s also imagine that a competing firm, XYZ Company, is in the business of making desserts. XYZ is interested in purchasing ABC Company. XYZ is unaware ABC has an A.I. specialist, or that this system predicted that wheat prices would skyrocket in a year. The implications are clear. ABC knows that if wheat prices skyrocket, its operating margins will squeeze and that’s bad for business.

What happens if ABC agrees to be acquired by XYZ, but fails to disclose its internal estimates that ABC’s end markets will likely collapse?

What if ABC’s executive leadership sold ABC’s stock and avoided losses, based on this “specialist system” but without telling the market?

Technically, those most capable of leveraging the benefits of machine learning algorithms should have a distinct advantage over those who don’t, but in some circumstances, this may lead to a dangerous information asymmetry.