Originally published by Traders Magazine
By David Griffiths, Director of Regulatory Affairs
Algorithmic trading has become an indispensable part of the financial world. Machines process data faster than humans and can generate greater profits. Left unchecked though, algorithms can pose significant problems when used in complex markets. Monitoring algorithms has become increasingly vital for recognizing potential risks like market manipulation, but that due diligence process has not kept pace with the proliferation of algorithmic trading across asset classes.
Understanding risks is what helps identify the types of controls your institution needs. For algorithmic trading, this starts with having robust procedures in place to monitor algorithms used for electronic trading.
Before an algorithm goes live, the model should undergo a rigorous development process that includes being tested, vetted, and signed off by multiple parties. Algorithms are scalable across sectors as the model can be trained with data to operate on different types of markets, including cryptocurrencies. Once that model is live and operating on a trading platform, monitoring procedures ensure that the algorithm works as expected. The best way to monitor an algorithm is to Know Your Algo or KYA.
KYA regarding technology should have the same emphasis as Know Your Customer (KYC) does for compliance functions and bank customers. KYC functions are a rigorous due diligence process that checks if customers are who they say they are. For financial institutions, this single check proactively eliminates significant risk regarding money laundering and other crimes. Having the same insight into technology is important. This idea is an evolving fundamental within surveillance, and KYA borrows significantly in concept from KYC.
While multiple trading platforms may leverage the same algorithms, each platform has its own goals and assesses risk differently. The first step to understanding how your algorithm works is to know the source of the algorithm. Algorithms are either supplied by clients or banks, or are colocated for high-frequency trading, and institutions need the tools to monitor all three types.
With direct market access, clients, hedge funds, and asset managers might very well all have their own algorithms that they use for trades that are executed through a prime broker. Market participants are focused on finding the right technology that helps them meet their goals. There are many client-supplied algorithms in the market, and understanding the proprietary technology that each uses is a challenge.
There’s a balance between knowing what a client is doing with an algorithm versus the business they are processing through your institution. Being able to profile algorithms is a key aspect of KYA. An algorithm that’s suddenly functioning differently, or having a behavior change, can be indicative of an update or new model that’s put in place. Knowing when this occurs is essential.
Where the algorithm comes from in some ways determines the level of risk with that algorithm. Those that are bank-supplied give you the ability to uniquely flag and tag them, and you also have more input during testing that occurs before these go live. Since these are developed in-house, there’s more transparency into how the model operates. Behavior changes are more easily identified.
Algorithms used by high-frequency traders (HFTs) require varying types of surveillance profiles depending on how that algorithm trades markets and how that algorithm is designed. These algorithms tend to be very well understood as this is an established part of the HFTbusiness.
Client-supplied algorithms on the other hand have the least amount of transparency as the model operates like a black box.
Oftentimes, the inner workings of proprietary algorithms are kept secret to protect any advantages that the algorithm might give a customer. Institutions want these monitored effectively but do not want to risk giving away the formula to the “secret sauce” of the algo. That means more detective work on the monitoring and surveillance end. Learning this algorithm’s behavior takes time, and that may require additional AI to detect irregular trading behavior that could signal a change in the model.
Markets are only getting faster, and this speed is fueled by the growing popularity of algorithms used to trade securities. As algorithms are produced and refined faster than markets move, the need for effective surveillance mechanisms only becomes greater across personas and institutions. The quickest and best way to solve the problem is to KYA.