Much of the world around us is powered by machine learning. It’s an invisible force that keeps the digital world turning. The LSEG Tech Talks series, in its inaugural session on 6th August 2021, dove precisely into the domain of machine learning to bring to the forefront and showed us its impact in real-world financial markets.
Helmed by the expertise of Senior Software Architect, Dr Rasika Withanawasam and his 16 years of experience in the surveillance domain, the session explores how machine learning is currently being implemented to reduce manipulation and insider trading occurrences in the stock market.
As the LSEG Tech Talks session progressed, Dr Rasika explained that machine learning fundamentally works by learning a process from examples and experience, memorising it, and then using it to resolve a challenge faced in the real world without the need to programme by software developers. This scenario will typically occur in one of the following ways: supervised learning, unsupervised learning, semi-supervised learning, and reinforcement-based learning.
A space where machine learning can thrive with the solutions it provides needs to be a realm where there is plenty of data and where actors have a proclivity to try to gain the upper hand on the rest of the participants. Of course, the stock market is one such context where machine learning-based fraud prevention mechanisms are most needed.
How machine learning protects markets
Given the asymmetry of information inherent in the stock market, this space is prone to manipulations such as insider trading and uninformed manipulation, which refers to taking actions to mislead the market and impact stock values to favour the manipulator. Another popular tactic you may have heard of is a ‘pump and dump’, where a stock value is artificially inflated and sold.
Market surveillance departments typically use the help of rule-based methods to prevent and detect abusive, manipulative and illegal practices in securities markets. For example, using a rule-based detection algorithm, the machine learning programme can provide an alert if a broker is increasing the price of stock value with a series of small buy trades followed by a large sell trade. An expert in the domain must provide a parameter for an acceptable set of actions for the algorithm to be effective.
However, the reality is that it is inevitable that these methods will not be as accurate as desired, there will be some element of human error involved, and it is challenging to be adaptive in a space where manipulation prone actors consistently look for ways to outsmart the system.
Machine learning steps into the fray to ensure that an effective alert parameter calibration is maintained to continuously stay ahead of the curb to detect frauds in this domain. This machine learning process will take advantage of unsupervised clustering and then derive data-driven benchmarks from replacing the user-driven parameters and thereby reducing false positives and false negatives that may arise.
Another great example of machine learning was how the programme trains itself to conduct similarity detection in a price signal to detect pump and dump scenarios in the crypto markets. This self-taught programme uses computer vision techniques for attribute computation, and normalisation will then be used to train a deep learning model to pump and dump shapes in the price signals.
What makes this genuinely ground-breaking is that there is no need for any prior knowledge on manipulative steps and parameters when designing the detection system. This also reduces false positives and false negatives to boot. However, the existence of this particular solution is due to the funding by LSEG for the master’s research project for one of their employees, which is remarkable in its own right.
Getting into machine learning and learning the basics
Two queries stood out as the LSEG Tech Talks session wound down and entered into a series of Q&As with the enthusiastic audience.
For those looking at sinking their teeth into machine learning on their own time, online classes such as Coursera, Udemy and even Data Camp are great entry points into this subject. However, suppose you are interested in diving deeper into the field; In which case, it is recommended to take up a Master’s degree with any of the reputed courses in Sri Lanka.
Last but not least, the question on everyone’s mind was what is the future of machine learning in Sri Lanka. Dr Rasika made it clear that Sri Lanka is undoubtedly not lagging in machine learning markets. This is extremely clear given that the products developed at LSEG are world-leading solutions used in mission-critical contexts in major financial markets across the globe.
The biggest takeaway from the LSEG Tech Talks sessions is that there is the consistent production of brilliant cohorts who graduated from Sri Lanka’s educational institutes and given the applicability of machine learning solutions across a variety of industries, the future of this space, and the vast number of solutions it can offer makes the world looks exceptionally bright!
The next session of the LSEG Tech Talks Series will take place on 20th August from 4 PM onwards.