Whitepaper: Future of TCA and Machine Learning

Medan Gabbay

Whitepaper: Future of TCA and Machine Learning
  • 1 Version
  • 735 Download
  • 4.72 MB File Size
  • 1 File Count
  • 19th May 2019 Create Date
  • 17th February 2020 Last Updated
Please login to download LinkedIn

In an era where the risk-free interest rate never hovers above the low single-digits, low investment returns have become the norm. Lowering the cost of investment with efficient execution and streamlining costs is the most certain way to preserve overall investment performance. Managing the cost of execution is far more deterministic than an investment strategy outcome.

The buy-side tech stack was most recently reviewed to improve TCA capabilities following MIFID II regulations. ‘You can only improve what you can measure” and this change represented a fundamental shift in the capabilities of institutions to measure and therefore improve performance using an arsenal of analytical tools and data.
The next step in this evolution is to develop TCA from the tactical ‘tick-in-the-box” compliance product to a pre-trade predictive tool for strategy selection and process automation. Machine Learning allows us to extract information from raw data and represent it in a model. The Short-Term applications for AI/ML with the greatest impact for buy-side are:

Trending

Using data to find trends within time-series. Examples are the evolution of Spread, Hit-Ratio or Fill-ratios over time. Trends are well suited to ML and provide rapid improvement over current practices for monitoring large number of trends in multiple instruments / asset classes as well as detected them much earlier than the current techniques.

Predictive

In this case historical or modelled data can predict the outcome. Statistical / probabilistic methods in use by the typical buy-side currently are labour intensive and therefore expensive. For example price reversion predictions.

The Future Architecture

The current systems in use by the buy-side were built during an era when connectivity was the issue to be solved. These systems were on operated on a ‘fire and forget’ paradigm which is mostly true to this day. Despite the large amount of work to incrementally improve these systems such as adding Algos, Smart order routing and automation to move to an AI/ML driven model requires a more fundamental improvement. This new architecture is much more data driven mimicing the neural structure of the brain. Actions are split into predictive agents that feed execution agents. Real-time Data collection forms a loop with a real-time oversight of Human Operators and automated risk controls. New ML techniques can implicitly update models that are trained with big data sets. These models are tightly integrated into the execution software to provide new predictions but also be re-trained with each execution.

Conclusion

Machine learning is in its infancy with a lot of innovation yet to be delivered (or even imagined). As with all new technologies there is a cost to travel the learning curve and bring about organisational change. TCA is an efficient starting point to test ideas and trigger initiatives for building core system capabilities. With little insight being provided to traders, AI/ML in TCA represents a giant leap forward for all market participants and provides the fuel for the slow transformational journey to improving the trading architecture and trading decisions.
File
Quod+Machine+Learning+TCA+-++MAR19+-+FINAL.pdf