The financial markets—the stock exchange, bonds, commodities—are a historically high-risk, high-reward avenue towards making, or losing, money. Besides working as a television weatherman, investment banking is one of the few fields where you only have to be correct sixty percent of the time to be considered a smashing success. Investment decisions are generally gauged against the S&P 500, which is a collection of the largest corporations and their stock; it’s an option that, compared to the rest of the market, promises healthy and steady returns. Of course, humans have always tried to beat this standard, and no secret that most human investors continually underperform against the S&P 500.
As a field with huge amounts of data, and an almost equal amount of human error, the field of financial services is ripe for advancements from the field of data science. By combining the history of the United States stock exchange with data regarding human investment choices, machine learning techniques are poised to comb through this huge amount of information, and maybe even come up with a stock tip you can actually rely on.
In a recent paper titled, “Should You Trust Your Money to a Robot?” CDS Faculty member Vasant Dhar investigated the possibility of using machine learning techniques to make financial decisions for humans.
So should I trust my money to a robot?
It depends on the investment at hand. Because machines only rely on previously collected data, they can’t spot a “one of a kind” investment opportunity in the same way that humans can. Granted, most humans can’t spot a “one of a kind” investment opportunity either. But because machines rely so heavily on the previously collected data, robots tend to make much smarter investment decisions. In general, humans are not very good at investing.
Can you give us a bit of background on why you chose to look into this subject?
One of my major career objectives was to design a machine that could learn to trade as well, or better, than a human counterpart, given access to the same data.
I became interested in the area of predictive models for financial decisions back in the 1990s, when I saw corporate financial data becoming more readily available. Being located in New York City, the finance industry seemed like the best place to explore all this available data, so I took a few years off from academia and went to work on Wall Street. I began mining the available data, which was mostly market and customer data.
Which part of the investment process are machines helping with?
My machine helps pick investments from a defined set of possibilities. But most other machines that are working in the financial sector are performing portfolio optimization, as a way of balancing risk; most other machines aren’t actually helping pick the investments.
In the introduction to your “Should You Trust Your Money to a Robot?” paper, you talked about several investors (George Soros and Warren Buffet) and the longevity of their careers. Are the decision making processes of these investors used to model your machine learning programs?
No, they have nothing to do with my programs. I’ve never found a human who always performed better than the market. Buffet and Soros are clearly exceptional. While some argue that their success was pure luck, I don’t believe this. They were special in spotting opportunities. But the opportunities they exploited are extremely rare.
There’s a wide range of investment opportunities within the financial markets. Which parts of the market have you been conducting your research on?
Machines have already taken over high frequency trading, because that’s an area where humans don’t stand a chance against machines. My research focuses on the mid-range investments—investments that are made and held for a period of weeks to months.
What is the biggest weakness that machines face in making financial decisions?
The noise and chatter surrounding the financial markets make markets incredibly hard to predict. There are too many random shocks and unpredictable events that impact investments; both humans and machines have a hard time being calibrated for these sorts of unpredictable events.