Firms Hone Use of New Alt Datasets to Pick Covid-19’s Winners

The Covid-19 outbreak has accelerated the use of alternative data in investing. The first stage of its use saw firms apply alternative data mostly in macroeconomic forecasting or to assess the impact of the pandemic at sector level. Increasingly, though, as investors refine their use of the data, they are searching out indicators that tell them about the prospects for individual companies.

PanAgora Asset Management is using natural language processing to gather and process data on Covid-19 scientific research. The firm has repurposed an existing machine learning model designed to determine—based on medical trial results—which biotech stocks would do well and which would perform poorly.

“We took our model that looks at individual companies and instead applied it to all the drug trials currently underway for Covid-19,” says George Mussalli, chief investment officer and head of equity research at PanAgora. News reports imply a vaccine is expected in the fourth quarter of this year or the first quarter of next year. “But we asked ourselves: What’s the probability that one of these therapeutic trials works, or one of the vaccines gets approved by the the Food and Drug Administration (FDA)?” he says.

The model found there is a better than 80% chance of successful treatments for the virus by the fourth quarter. But the outlook for a vaccine is less promising, with the cumulative probability of success a little above 50%. There are more than 130 drug trials for a Covid-19 vaccine and more than 1,100 therapeutic drugs in testing.

To train the model, the firm fed it data on every vaccine trialed by the FDA for the past 30 years, gathered from clinicaltrials.gov. The results come with caveats, though, Mussalli says. There are many more trials for Covid-19 than other viruses, perhaps raising the chances of success. Equally, the FDA may approve a drug that turns out not to be up to standard.

Once drugs are available, the firm will probably start tracking data on the section of the FDA website where doctors and patients report drug side effects. “The big risk is for the first year. You might have unknown side effects that cause the drug to get pulled off the market. We’ve seen that before. Then there is a higher probability for the drug company’s stock to tank,” he says.

Mussalli admits it’s difficult to understand the impact of Covid-19 on stocks and to gauge what effects are already priced into the market.

“We saw early on, before unemployment went up, people searching on websites and talking on social media about how to apply for unemployment benefit. So now we’re looking for things like bankruptcy filings and foreclosures to get an earlier sense of what’s going on,” he adds.

Jonathan Berkow, senior quantitative research analyst and the alternative data lead for equities at Alliance Bernstein, says the group has ramped up its web scraping and natural language processing in the wake of the pandemic, to track activity levels in different regions of the world.

The firm is looking at number of cases of Covid-19 per population cross-referenced with economic activity, what kinds of activities are taking place, whether people are working or staying home, even how many are searching on the internet to buy new cars. Some of this data is publicly available. Other data, the firm has purchased. All of the data is then combined with company-specific locations to determine how exposed individual stocks are to the pandemic. “The geographic footprints are critical for determining how firms are going to respond to this pandemic,” Berkow says

Neuberger Berman, meanwhile, has also been looking to alternative data in recent weeks, including US credit card transactions broken down into households’ spending on gasoline, which is increasing with personal car usage, and use of public transportation, which has remained low. The firm is also employing natural language processing to analyze the text of job postings in order to determine the types of projects companies are hiring for. “The amount of data that’s available is exploding. Obviously, it was growing already,” says Michael Recce, chief data scientist at PanAgora. But the pandemic has forced investors to “go figure things out” as well. “This is accelerating the use of alternative data in the investing world.”

Recce says much of the use of alternative data started with restaurants and retail, but has gradually spread into technology, media, telecoms, subscription businesses and healthcare. The business-to-business sector and industrials have lagged behind, but now Neuberger is gathering that information also.

Recce sees additional value in data that can help detect bias in existing signals. Combining geolocation and credit card data, for example, can show how many customers are visiting a store as well as how much they spend.

“The alpha for fundamental investors like us is in understanding what [new data] tells us about the business we didn’t know,” Recce adds.