The C$23 billion Canadian fund OPTrust is using AI to reduce risk in a strategy it hopes to roll out to the wider portfolio. Wei Xie explains the benefits and challenges of machine learning including AI’s ability to identify complex dimensional relationships.

The C$23 billion Canadian fund OPTrust is embracing the power of AI to improve investment outcomes via two new strategies based around re-enforcement learning and uncertainty modelling.

The latter approach, in place since late 2020, is currently used to pare risk in a risk parity strategy and is already giving actionable signals on how to adjust risk in the portfolio. Going forward, Wei Xie, co-head of the multi-strategy investing team at the fund told FIS Digital 2021 delegates he hoped to deploy the strategy at a total fund level and combine it with the equity beta portfolio to create a risk reduced equity allocation.

Uncertainty modelling

OPTrust began exploring how to integrate AI with the support of external quant managers. Next, the fund began to build its own internal, proprietary capabilities with just Xie and an associate overseeing the project. At a high level, AI is used to better understand and inform risk in a scalable approach, he explained.

“As long as the algorithms are robust, we can adjust the approach and use it for total fund management,” he said.

Xie said that whilst very promising, machine learning is also complex. Particularly from a personnel perspective, creating challenges for investors around hiring people with the right skills. It also involves investment in a data supply chain. Indeed, even with a reasonably large investment team, OPTrust found it difficult to use AI to tap a competitive advantage because of limited resources around data and personnel. Instead, the pension fund decided to focus on risk management solutions.

“Risk management is scalable across different types of strategy and doesn’t need exotic data as it mainly involves working with price data,” he said.

Expanding on how the process works, he described the typical quant model as injecting ingredients or inputs. OPTrust sought to design a process that used AI to discover when assumptions might be ineffective due to the function of changes in the market.

The domain set out to identify times when the prevailing logic falls apart, alerting Xie to changes in external markets and the ensuing impact on risk.

“If something has changed the risk may have increased because what we are doing becomes less effective,” he explained. “The process is telling us that our strategy is less accurate, and that we should take risk out.”

Re-enforcement learning

Re-enforcement learning gives greater insight into the factors driving price action. AI allows investors to detect the non-linear impacts on prices that humans struggle to track, he said.

“Although humans are good at understanding two-dimensional linear relationships, when it becomes more abstract and relationships get to three, four or five dimensional, humans stop having any ability to interpret – accept through maths. In contrast, machine learning and AI can identify these complex relationships,” he said.

Addressing the challenge of how humans get comfortable with AI outcomes that are difficult to intuitively understand, he said the process involves testing the models and a simple checklist.

“You can do a sanity check around what is happening in your model,” he concluded.

Sarah Rundell is a staff writer for based out of London. She writes on institutional investment across all asset classes, global trade and corporate treasury.
Leave a comment