Deep Learning and the Financial Markets
- Published 28th Feb 2019
Last edited 28th Feb 2019
Deep learning is a branch of artificial intelligence and machine learning, and it is based on the idea that on exposure to specific datasets, an algorithmic model can become self-learning. The ultimate purpose of deep learning is to develop machines that can mimic the human neural network i.e. billions of neurons exchanging signals, recognizing past patterns and predicting future outcomes. The aim is that when this complex neural network is combined with the efficiency of a machine, the predictions can be more accurate and more actionable.
Over the past years, financial institutions have been working on using deep learning models to predict the direction of the markets. So far, there have been mixed results.
The relationship between deep learning algorithms and trading
All an investment trader cares about is if stock prices are going up or coming down. One of the most effective ways to do this is to study historical market data and see how certain factors moved price in the past. If the same factors occur again, the expectation is that price will behave in a similar manner. Simply put, they use past data to predict future outcomes; this sounds like a use-case that was tailor-made for deep learning.
In addition to being more efficient than humans, another advantage of a deep learning algorithm is that it can analyse large volumes of datasets within short periods and it can be programmed to filter out market noise e.g. short-term fluctuations, negative perception, and so on.
All that is required is to feed the algorithm with historical data; the more the data, the more accurate the predictions. The data is analysed and patterns are recorded. Afterward, the algorithm is used to predict outcomes in real market conditions; first as a test, then as a full-on trading system. Because of its self-learning nature, the more a deep learning model is used, the better it becomes.
Hedge funds using deep learning to trade successfully
While a number of banks are using deep learning models to augment their trading systems, few of them have ever come out to provide information on the results they find. However, some hedge funds that do the same are not so secretive about it.
A hedge fund based in Hong-Kong, Aidiya, announced a while ago that they have developed an algorithm that trades using multiple forms of artificial intelligence, machine learning, and deep learning. Every day, the algorithm analyses market prices, market volumes, corporate accounting documents, and macroeconomic data; using all these, separate engines predict how the market will go and they vote on which prediction should be used to trade. All these are done without human involvement. On its first day of trading, the algorithm made a profit of 2%.
Reports indicate that San Francisco-based Sentient Technologies, backed with funds in excess of $140M, has been using a system similar to Aidiya’s to trade even before the Hong-Kong firm.
Other hedge funds that allegedly use AI and deep learning models include Renaissance Technologies, Two Sigma, Point72 Asset Management, and Bridgewater Associates (Wired). As a matter of fact, according to a leading market research company, Preqin, over 1350 hedge funds use computer models for the majority of their trades.
There are still a number of kinks to be ironed out before the mass adoption of deep learning algorithms by banks and financial institutions. So far, though, the models have been shown to work vis-à-vis predicting how the market will behave; at least on a small scale. As the technology develops over time, the predictions will surely become more effective and more accurate.