How Traders Make Big Money With Algorithmic Trading

algorithmic trading

Artificial intelligence (AI) and Big Data are changing the financial markets. Even private traders are now using self-learning algorithms to make money in the markets. This article reveals how it works.

Beating the market with artificial intelligence: The big players in the financial sector, such as the world’s largest asset manager BlackRock and the British hedge fund operator MAN Group, are working on this at full speed. Artificial intelligence differs from other automated strategies in that it is no longer the human being who decides which rules the machine will follow.

Computers Making The Trading Decisions

The human no longer dictates a strategy that the computer then implements, but the machine learns (within a framework set by the human being) independently what the best strategy is to make the greatest possible profits on the markets. Even if the success so far is quite modest: In the USA there is already an ETF in which an artificial intelligence makes the investment decisions.

The general availability of automated trading is causing growing popularity also in the field of Bitcoin margin trading, which is, by the way, very restricted in the US.

But the decisions of the computers are only as good as the data available for their decisions. And here, too, big things are happening: hedge funds now routinely evaluate satellite images to determine the capacity utilization of Chinese factories, for example. Or they analyze credit card payment flows to find out where and how much people are buying. But of course the net also serves as a source of information for the machines.

BlackRock, for example, has discovered that the blog posts of employees of large corporations can be used to determine where the journey is headed: If the mood among employees is good, then as a rule the shares perform better, as BlackRock founder Larry Fink explained in an interview.

Tremendous Computing Power Needed

But huge amounts of computing power are needed to evaluate the corresponding data volumes. AI developer Sentient Technologies uses a total of two million computer processors and more than 5,000 graphics cards at 4,000 locations worldwide not only to trade on the markets but also to develop new cancer therapies (see here).

At the heart of Sentient Technologies’ strategy are so-called genetic algorithms. This involves replicating natural selection in the course of evolution in order to create algorithms that can act as successfully as possible in the markets. Algorithms that produce good results multiply. Algorithms that perform poorly become extinct.

The approach can now also be applied – albeit on a much smaller scale – by technologically fit private traders. This is made possible by the Java program Genotick, which the developer Lukasz Wojtow makes available free of charge on the Internet.

Even though the exact procedure is relatively complex, the basic idea of Genotick can be explained relatively simply: The program generates a large number of “robots”. These are mini-programs that make a decision to buy or sell. The exact calculations that the robots make to make their decisions are different for each robot and are randomly determined. After each time unit, the large number of robots vote: Will the share price rise or will it fall? The majority decides what the machine will do.

However, the next step is decisive, in which natural selection is simulated within the framework of evolution: Robots that make good predictions survive and can also produce offspring that function similarly to their creators. The robots, which often fail to make good decisions, will eventually die out. In the end, it is hoped that the program will have created a population of robots that can predict the course of the stock as well as possible. On the program’s homepage, the developer provides some impressive results.

In a practical test with the Dow Jones price data since 1988, this success could not quite be reconstructed. The generated “robot” population was able to achieve a profit of 542 percent in the simulation.

However, the program performed worse than a buy-and-hold investor who would have earned more than 1,100 percent with a long investment in the Dow Jones over the period under review (29 years). It must be noted, however: The program was provided with Dow Jones price data on a daily basis for the quick practical test. With additional price data (e.g. at tick or minute level and with price data of other underlyings) as well as fundamental data, the forecast quality could possibly be significantly improved.

Furthermore, the population of 10,000 “robots” in the test was relatively small in order to achieve a quick completion of the calculations. It is quite possible that Genotick will be able to generate considerably better forecasts if additional data is provided and larger robot populations are generated. However, this then also requires a much larger computing power.

Genotick-AI is probably not yet able to compete with the best human traders, such as the experts of our Trading Services. But it cannot be completely ruled out that in the coming years an AI will be able to catch up with the best human traders.

Comments are closed.