by Kai Cui, Head of Equity Data Science—Equity Research, Charles Murphy, Senior Research Analyst—Equity Research, Neuberger Berman
Today’s CIO Weekly Perspectives comes from guest contributors Kai Cui and Charles Murphy.
On Neuberger Berman’s Equity Research team, we are always looking for mispriced stocks or sectors, particularly underperformers that could be due a rebound.
One recent group of underperformers is U.S. travel and leisure stocks. Since mid-June and the advent of the Delta variant of COVID-19, daily U.S. infections have risen tenfold, putting a dampener on the “reopening trade” in markets.
The fundamental analysts on our team weren’t convinced this was a value opportunity, however. Despite a tentatively optimistic view on future infection rates from our health care analyst, Terri Towers, the sheer rate of new cases left many still concerned that things would get a lot worse before they got better.
The data scientists on our team had been coming to a different view, however—and when they presented their latest findings, they had quite an impact. After intense discussions, our fundamental analysts’ view on the U.S. travel sector materially changed.
It’s a story not only about the insights that data science can give us into the path of the pandemic, but also about the ways alternative data need to be marshaled to provide actionable information for our investment views.
In fact, the first set of results was not encouraging.
Working with management at online travel agents had taught us that they measured online searches as a key leading indicator for bookings. That’s why our own data scientists now gather and analyze these searches for ourselves—to the extent that many travel agents now ask us for our latest insights into these trends.
What did we find there? While there has been a slight uptick in searches in the U.K. after it dropped its COVID-19 restrictions in July, searches in the U.S. remain on the downward trend that started back in April.
But then we looked at the data science work on COVID-19 itself.
One of the datasets we maintain maps each U.S. state (as well as numerous other countries) on a grid with the Y axis showing the vaccination rate and the X axis showing an index of several personal mobility datasets, such as those generated by public transit, Google and Apple cellphones. We categorize states in the top-left (where many people are vaccinated and few are out-and-about) as “Safe.” Those in the bottom-right (where fewer are vaccinated but more are out-and-about) are “States of Concern.”
When we compared rates of new COVID-19 cases with these data, something very interesting emerged.
As we expected, States of Concern such as Mississippi, Idaho and Arkansas exhibited the earliest and biggest surge in Delta infections, as unvaccinated citizens traveled about spreading the virus. More strikingly, while personal mobility has remained steady and relatively high, infection rates appear to be plateauing. In some cases, such as Missouri, infections are even declining.
In other words, in some States of Concern, the Delta variant is not being controlled by social restrictions, but may be burning itself out—a finding that fits with Terri’s tentative thesis from a month ago. Based on these data, our data scientists believe that cases at the national level could peak within one week.
Under that scenario, travel stocks, as well as many others geared to economic reopening, appear more attractive than our analysts had thought.
This experience reinforces many of the key things we’ve learned from four years of integrating data science into our securities analysis.
First, while fundamental analysts are used to receiving material and verifiable new information about companies once a quarter, from earnings reports and calls, alternative data has the potential to generate new information week-by-week or even day-by-day.
Second, one data set provides us with information; but crosschecking it with other, related datasets, perhaps with different frequencies or longer forecasting qualities, provides us with information in context.
Third, it is important for fundamental analysts to help design data science research projects—so that they are focused on relevant, material questions and their findings and investment implications are intuitive to grasp. Equally, it is important for data scientists to challenge fundamental analysts’ assumptions when they think they are not supported by the latest, richest information.
Finally, all of this requires both fundamental analysts and data scientists to work closely together and learn one another’s languages. This is quite literally about learning computer programming languages such as Python, but at a more profound level it’s about securities and sector analysts becoming fluent in the language of alternative data and data scientists becoming fluent in business and economics.
When we get it right, we can uncover insights that are surprising, often counterintuitive—and sometimes game-changing for our investment views.