Why Algorithmic Systems May Not Always Work in Trading

Though many companies sell algorithmic trading systems as a holy grail of trading, this is simply natural. The reason is simple: an algorithmic system is designed by humans and may suffer from the same flaws that humans do. And when the overall framework of an algorithmic trading system is not robust, the chances of success in trading are likely to diminish.  

Algorithms in Trading

Let’s understand this with trend-following systems which are, in all likelihood, the most popular example of algorithmic trading systems. They often fail in bringing the desired result for traders. The reason is when a large number of traders start using the same methodology to make money, the profit is diminished.  

Another reason is the unpredictability of the markets. There are always some factors, which cannot be predicted such as what Donald Trump is going to do next. Or will the Feds adjust rates? Or will there be a trade deal between the United States and China? Such factors are impossible to be considered while designing systems. In short, the unknown is always there. Whenever you feel like designing the best algorithmic trading system, a black swan event could wipe out all your money.  

Algorithmic trading systems do not suit investors with low personal risk tolerance. The problem becomes acute when the drawdown is high. 

Now let’s come to the main point. Why does this happen? Algorithmic trading has a wider spectrum; it may involve anything from a simple moving-average based system to highly sophisticated and complex mathematical models for trading thousands of securities based on their statistical relationship.  

Delving Into Factors Causing Failure 

First, an algorithmic system may not be robust because of the lack of sufficient risk management systems. Such flaws are in-built within algorithmic trading systems as they are based on certain rules and conditions. A system with not-so-robust input parameters is likely to flounder in real world trading.  

Second, these systems often don’t take slippage into account, resulting in less-than-optimal results when deployed in real world trading. Slippage is the difference between the price at which a trader expects the price to be executed and actual price at which the trade is executed.  

Third, algorithmic trading systems may have limitless variations. For example, a system or framework for initiating trade may be devised on a list of valuation and growth metric conditions based on both technical and fundamental parameters. However, in order to succeed, the rules and conditions should be made precise to the maximum extent possible, or else the algorithmic system may give wrong Buy and Sell signals. The problem can be even more acute if the system is based on identification of patterns such as a double bottom or a head and shoulders pattern.  

Fourth, a system may stop working as the fundamental underlying factors change. However, it is almost impossible to figure out whether this happens because of a drawdown or because the system no longer remains effective.  

Fifth, smart humans may outsmart systems. Some traders take advantage of the chinks in such algorithmic trading systems. They know how to make a tidy profit right under the nose of these systems, which are prone to make costly mistakes if they are not set up correctly or if some traders get an insight into these trading patterns. 

When Humans Outsmarted Algorithms 

In 2007, Svend Egil Larsen, a Norwegian day trader, figured out the functioning of the trading algorithm of Timber Hill, a unit of US-based Interactive Brokers, vis-a-vis the trades in certain illiquid stocks. Whatever the bid was, the algorithm would change the stock price in a uniform way. Taking full advantage of the flaw, Larsen was able to earn $50,000 in just a few months. 

Larsen was charged for ‘market manipulation’. However, the court pronounced him not guilty, accepting the defense that what they were doing was a market practice and that he had actually helped out the system by exposing an existing flaw. 

Later on, the same Mr. Larsen capitalized on UBS’s mistake of not setting a bottom limit on one of its trading algorithms. Picking up some stocks at a discount, he minted $14,000 in a matter of a few minutes.  

Larsen is just one of the wannabe examples. Eivind Stolen, a Swedish day trader, rode on the malfunctioning of a Morgan Stanley client algorithm to earn thousands of dollars in just seven minutes. 

This time, the algorithm had begun buying at the offer price and selling at the bid. Eivind and many others made a small spread on every trade they bought and sold back right away. 

In 2010, Deutsche Bank’s trading algorithms in Japan erroneously exited a $182-billion stock position. 

In 2011, ten Focus Morningstar exchange traded funds launched recently on Nasdaq OMX and NYSE Euronext were wiped out. 

Every time an algorithm goes wrong, there are many across the world who benefit from it. 

The moral of the story is that algorithmic trading based on rigid ‘if-else-then’ rules may not work in trading every time. Perhaps, it is time we began thinking about some alternatives.  Artificial Intelligence could be that next step – taking advantage of automation as has been done with traditional algorithmic trading (the upside), but in this new era applying automation on systems that think like humans.  

Why Algorithmic Systems May Not Always Work in Trading Click To Tweet