Boosted.ai Rolls Out New Models for Navigating Covid-Specific Risks

Boosted.ai, a Toronto-based company that uses machine learning to provide asset managers and hedge funds with portfolio insights, has developed new risk models for clients struggling to accurately measure and predict the effects of the ongoing Covid-19 pandemic on portfolios.

The models differ from traditional risk factors and analyses: a necessary response to the crisis, which has baffled some existing financial models because such an event has not occurred for 100 years.

“A really interesting thing about this year is that a lot of traditional risk modeling has broken down,” says Joshua Pantony, co-founder and CEO of Boosted.ai. “Traditionally, momentum [investing] would be seen as a risky play. Traditionally, fixed income is going to be safer. Traditionally, airlines are more slow and steady. All of a sudden this year, all that stuff has broken apart.”

Like most other risk models, at their most basic level, the new models measure traditional risk factors—such as momentum exposure, style risk, or tolerance for volatility—to determine a portfolio’s risk exposure. It then goes a step further by using “topic embedding” to model the residual risks or the overall risk exposure of a portfolio that can’t be explained by traditional risk factors.

“A lot of the safer, more traditional things haven’t really worked [during the pandemic]. And the other thing is that you’ve had these weird breakdowns in traditional correlations, where suddenly hotel stocks are correlated with retirement homes, are correlated with cruise lines, are correlated with fitness,” Pantony says. “These are all things that, under a traditional risk model, you would never expect.”

As the Covid-19 outbreak has illustrated, sectors and companies, even vastly different ones, are still highly interconnected whether by geography, supply chains, or more obscure links. “Topic embedding” enables the models to seek out these less obvious correlations. It functions by linking together “clusters” of seemingly unrelated companies that are each affected similarly by a central topic or event. For example, Pantony says, we might know that nursing homes are affected by pandemics. Hotel companies are affected by social distancing, which is related to pandemics. Other types of business negatively affected by social distancing include travel companies and gyms, while companies selling home fitness equipment are affected positively.

Once these clusters are established—and users can create clusters related to any topic they choose, not only the pandemic—Boosted’s platform, Boosted Insights, will offer strategy tips to minimize or neutralize that exposure. For portfolio managers using the model inversely—perhaps they’re more confident in the likelihood of Covid-affected companies rebounding—the model can generate how best to increase their exposure.

Pantony says there is value in this type of risk modeling beyond the pandemic, as accounting for non-traditional risk factors can reduce portfolio volatility. However, he adds that it’s most useful in times of high market volatility, high correlation, and sudden, rapid changes in the economy.

Boosted, which began rolling out the new risk modeling toward the end of last year, is continuing to invest in the capability after seeing the function go from a fringe use case to its most sought-after use case in February and March. For example, the three-year-old company will use part of its May $8 million Series A funding round to build out derivatives support on the platform, which only covers equities today and to grow its client base.

Currently, the firm counts a handful of large banks as clients, but its buy-side base accounts for the bulk of platform users—mostly fundamental managers, followed by quant funds and institutional asset managers. The platform is deployed at about 15 different firms, and is used by about 10 to 15 individual users at each firm.