Are Things Between Algo Firms Equal? — Part 1
This is part 1 of a 2 part article. In this article I explore the role of algorithms in HFT and how they are kept secret. These are also different across different firms.
Do all Algo trading firms make the same amount of money? Obviously not!!!
In India, it is said the three most successful High frequency trading firms happen to be Shastra (Tower Research), Alphagrep and Quadeye. During 2010–15 these firms contributed 80% of India’s HFT volume and profits in NSE, BSE and MCX. So, why were these firms making more money?We might be tempted to say preferential access but these firms continue to dominate India’s HFT space. Long after the alleged preferential access has ended, these firms and a few more are highly profitable. The truth is that these firms are smarter and sometimes even faster than others.
If other things were equal among all the players in an exchange, most of the players should have the same number of orders, trades and profit. But a quick look at the financials of most of the co-location players show that players operate very differently. Things are rarely equal among players. So, what are these “things” which differentiates a successful HFT firm from a regular trading firm? What makes them smarter and faster?
Typical HFT firms invest more than 50% of their revenue on trading research. These firms rely on top talent from the world’s best universities to create trading models, also known as algorithms. The algorithms are designed to predict small movements of various stocks to make risk free gains. The HFT firms are so secretive about these algorithms that they use all shenanigans to prevent them from leaking to a competitors. These companies sometimes continue to pay humongous salaries to employees years after they leave the organization, to prevent them from joining their competitors. There are numerous examples of HFT firms taking legal actions even if a simple algorithm was leaked.
The HFT algorithms can be very different in nature. At the same instant, one system might be purchasing a security while the other may be selling the same security. For example, one algorithm might be selling a security with an expectation that its price will go down in the next five minutes, while another system might be purchasing the security with an expectation that its price will go up in the next ten minutes. Both these algorithms can be right for their own timing.
These algorithms are designed to do a very specific job. This makes these algorithms very sensitive to external influences. They require constant tweaking and the parameters sometimes change even within a day.
NSE earlier followed a TBT dissemination using TCP protocol but later switched to a multicast technology called MTBT in 2014. In TBT, trade event was disseminated before a new order entry. In MTBT information of new order entry was disseminated ahead of trade event. Introduction of active tick in MTBT reversed the order of tick dissemination and would have caused many algorithms to loose efficacy. Any algorithm which uses a trade event to take decisions would get delayed. Any algorithm which uses a new order entry event to take decisions would get expedited in MTBT. Thus even though the active tick information is available to both the algorithms, each of them reacts very differently to the active tick.
Comparing trading metrics of different players without taking into account the algorithm they deploy is the same as comparing marks of two students in an examination assuming that they studied equally hard.
Next is part 2 of this article.
All articles are here. The author advises market participants in legal matters related to securities markets and has advised some noticees in this matter also.