The C$23 billion Canadian fund, OPTrust is embracing the power of technology to improve investment outcomes. Wei Xie and Brandon Da Silva explain the fund’s focus on two subdomains of machine learning and how they can be used together: reinforcement learning and uncertainty modelling.
The exponential growth in new technology is democratising the investment process and allowing people to take investing into their own hands. The solution to staying relevant amid the disruption may be fighting fire with fire and using technology to stay ahead of the curve. But as with most things, it’s easier said than done.
For a pension fund like OPTrust, our sole purpose is to provide members with secure, predictable income in retirement. While pension plans are not profit-driven, standing still is like moving backwards in the current environment and we believe it is in the best interest of plan members to harness the power of technology to improve investment outcomes.
That said, applying artificial intelligence and machine learning in financial markets is no easy task.
The data is noisy, which leaves investors searching for insights from the lack of signal. Financial markets are also social ecosystems that experience regular regime changes.
As such, many of the approaches to AI and machine learning that work in other sectors, like health research, may not work in financial markets.
AI and machine learning efforts are resource intensive. Significant time spent on research and development may yield little to no results. The work also generally requires substantial financial resources for data, computing and storage. These are not small obstacles when our focus is on cost efficiency and ensuring investments made result in tangible benefits for our members.
To approach artificial intelligence and machine learning in the financial markets in a practical way, we’re focusing on risk management as it is scalable and relevant in many types of investment strategies.
As opposed to focusing on a niche data set applicable to only one area, the application in risk management could potentially be repurposed in many different situations and create outsized benefits for the total fund portfolio if successful.
Further, we believe success in the research and development process requires focus. The decision on what not to pursue is just as important as what to pursue.
At OPTrust we focus on two subdomains of machine learning and how they can be used together: reinforcement learning and uncertainty modelling.
Uncertainty modelling is about understanding what we don’t know. By quantifying our uncertainty, we are able to manage risk by getting out of an investment when we are not confident in its future.
On the other hand, reinforcement learning involves building an environment where artificial agents learn to behave in a way that is optimised for a reward. Think of it like giving your virtual cat a virtual treat for finding the virtual mouse faster. Humans learn in a similar way. We go into an environment, experiment and find what works and what doesn’t. Then, over time we learn behaviours that produce the desired outcomes.
If AI and machine learning in the risk management space is ultimately successful for OPTrust, we may also consider adopting the techniques in other areas of the fund.
With reinforcement learning, the general approach would be applicable to other scenarios, whether it’s in a capital markets strategy, a total fund problem or environmental, social and governance investing.
While early in our journey, we’ve seen some successes and learned along the way, including following the principle of “measure twice, cut once.”
It’s also useful to have a team-based approach that brings a mix of market experience and technical skill. Cognitive diversity is key to tackling complex problems from multiple perspectives.
We are leveraging this at OPTrust by identifying unique talent through our internship program. In fact, Brandon Da Silva started as an intern at OPTrust. Today, he is playing a leadership role in the development of our AI and machine-learning efforts.
At the end of the day, just like our artificial agents are learning, we’re learning too. Our agents are looking for rewards in their simulation environments and through them, we’re looking for rewards for our members by better managing risk with the goal of achieving better long-term returns.
With enough experimentation, we’re confident artificial intelligence and machine learning will help achieve these outcomes.
Wei Xie is co-head and director and Brandon Da Silva is an associate portfolio manager in the multi-strategy investing team at OPTrust, a defined benefit pension plan with over 98,000 members and over C$23 billion in assets under management.
