Over the last decade, automated or algorithmic trading has taken the markets by storm. This does not come as a complete surprise, since enormous technological advances have created the possibility to trade at higher speed and frequency and recognize patterns faster than a human trader could ever dream. Unfortunately, (or luckily if you are in the risk management business) this trading system does not come risk-free.
For those unfamiliar with automated trading systems: these are electronic platforms containing pre-programmed trading commands contingent on various criteria. Depending on the actual situation at a specific point in time, it can momentarily decide what to do: to buy or to sell and how much. This makes the program highly efficient in dealing with rapid changes, which are not too uncommon in the trading business. Still, there are considerable dangers involved.
The most obvious risks involved are technical ones that each of us, whether highly experienced professional or common layman, have encountered when using technology: crashing systems, bad connections and seemingly never-ending lags. Extremely inconvenient for someone writing a long article for work, absolutely disastrous for someone in charge of an automated trading system in which many people have put their faith - and money. Obviously, these programs are extremely well protected and the risks of technical errors are minimized as much as possible. Still, they cannot be completely ruled out.
A bigger and more common risk is that of an imperfect algorithm. Since this algorithm ‘decides’ when and what to trade based on countless criteria and conditions, the slightest mistake, miscalculation or misjudgment can have huge consequences. To illustrate the consequences of an imperfect algorithm we will delve into the woes of Alex, a Dutch investment bank specialized in asset management.
In the first years since its inception in 2008, Alex Vermogensbeheer received critical acclaim for its results and was even named ‘Best Asset Manager’ by the public in 2010 and 2011. This praise was not undeserved: while the markets where going through one of the most dreadful periods in recent years, Alex managed to ensure decent, positive returns for its clients. Drawn by these promising figures, which were widely advertised, clients started to flow en masse. Everything seemed to be going perfectly, until 2014 - when the tables turned.
After being flooded with complaints from disappointed investors, the VEB (Dutch association of security holders) has now started an investigation into Alex.
The main question is whether it complied with all relevant rules and regulations and whether it created unreasonably high expectations.
Read more about High Frequency Trading and the European Regulations: Introduction to Transparency for High Frequency Trading Regulators
So, what went wrong?
According to Gerard Janssen, co-designer of the Alex asset management system, the markets are to blame. The automated system was designed so as to pull out quickly when the markets started to go down, and go back in as soon as they started recovering. This formula bore a lot of fruit during the crisis years, when markets fluctuated heavily. However, when the markets started drifting sideways, the weaknesses of the system were exposed. Mr. Janssen identifies the ‘V-spikes’ as the main cause of concern: the markets plunge down and then rise up again very quickly. The system automatically pulls out during the low, but does not go back in fast enough for the high, due to a specifically built-in delay function that is supposed to guarantee that the system does not step in too quickly in case of a false rise.
As this example illustrates, the ‘imperfect algorithm’ – risk is very high. Even systems that seem flawless for a long period of time may become worthless when circumstances and conditions change. The main point is that automated trading, however convenient and beneficial it seems, can be a very dangerous and risky enterprise, and everyone involved needs to be fully aware of the risks, even the well-hidden ones. After all, nobody is perfect – not even an extremely well designed algorithm.