Research by Refinitiv found that almost three-quarters of the quant community said COVID-19 had harmed their models, with more than one in ten saying that the pandemic had made their models obsolete. What caused these problems and how can quants adapt to the new environment?
- Quants struggled in 2020, but the underperformance gap has narrowed in 2021.
- Size, value and momentum are important in quant research, but quants can neglect to create models responsive to real-world shocks.
- As the quant community continues to recover in 2021, adaptive data inputs and algorithms need to be top of mind.
For more data-driven insights in your Inbox, subscribe to the Refinitiv Perspectives weekly newsletter.
Quant funds’ computer models rely on finding patterns in historical data, and very few could successfully trade through 2020’s once-in-a-century pandemic. There has been some narrowing of the underperformance gap in 2021, but the crux of the issue remains – most quants need to update.
The quants that thrived in 2020 have adaptive data and models. As concerns about COVID-19 spread through the investment media, models based on media sentiment were able to quickly re-balance.
Why did COVID-19 cause problems for the quant community?
Academic and industry research has long identified that share prices move in patterns in response to information flow, called overreaction (mean-reversion) and underreaction (trending).
Data based on such events helps quants identify breaking events – even anomalous events such as COVID-19 – by extracting the most predictive interpretations from the market’s reactions. Despite this awareness, many quants have been slow to adopt such insights.
Figure 1 below compares the performance of a quant factor built from media sentiments and themes (MMS) with traditional factors Size (SMB), Value (HML), and price momentum (MOM) from January 2020 through to April 2021.
Size, value, and momentum, the three of the major investment “factors” that economists have discovered, tend to lead to above-average returns in the long run. They involve grouping stocks according to a defining characteristic, such as their size, their cheapness, or their price change.
Systematically mining such factors is the heart of the hype around the computer-powered, algorithm-driven quantitative investment industry. But in the rush to find factors that worked historically, quants often forgot the importance of creating models responsive to real-world shocks and regime changes, and this neglect led to significant quant underperformance.
Investors and media sentiment
News about a stock’s earnings, management, or share price has a different impact on investors, and their reactions to such news produces different price patterns. Additionally, when such events break, the media’s sentiment about the event, audience, vividness, visibility and anticipation modulate the news impact on prices.
The MMS factor we show in Figure 1 is a machine learning-based based model of media reactions predictive of share prices.
This media factor was launched in production at the beginning of January 2020, and – as designed – it performed well over the tumultuous 16 months since its launch, uncorrelated with and besting the three traditional factors noted above.
StarMine MarketPsych Media Sentiment
The StarMine MarketPsych Media Sentiment (MMS) model is a stock ranking system that provides a 1 to 100 daily percentile ranking for over 16,000 global stocks. MMS complements the StarMine suite of equity models and follows a similar methodology in research and implementation. The model is derived from Refinitiv MarketPsych Analytics.
The MMS scores are designed to forecast the next month’s relative share price returns, with higher-ranked stocks outperforming lower.
Historical evaluation demonstrates significant outperformance of higher deciles versus lower ones, with the top-bottom decile spread for global stocks averaging 10.4 percent annually from 2006 to October 2020, including 12.3 percent in the out-of-sample period.
The MMS scores are uncorrelated with traditional market factors and complement fundamental models.
There has been a partial quant recovery in 2021, and the hope of surviving quants lies in adaptive data inputs and algorithms. These models are often based on AI, which has become a critical tool for understanding subtleties and unusual relationships in markets.
It is increasingly clear post-COVID-19 that quant models must be designed to handle surprising events. Doing so requires not only cutting-edge methods but also next-generation datasets to match. Quants who ignore either requirement may risk performance crowding and ultimately obsolescence.