Investors increasingly seek to capitalize on the vast troves of data now available due to the migration of activity and commerce to the digital world. So-called “big data” is the information that we leave behind when we browse the internet, buy things online or at the store, use our smartphones and generally live our modern lives. With advances in cloud computing, machine learning and artificial intelligence, it’s now possible to extract coherent, strategic insights from these digital traces. However, whether managers can actually capitalize may depend on both the relevance and quality of the data they choose to access and their ability to use it to further investment goals.
Coronavirus: Looking to the Data
In asset management, data science and large-scale data engineering make it possible to understand and evaluate businesses at a larger scale and in more detail. This has never been more important than in the current business environment that is undergoing an unprecedented rate of change driven by the COVID-19 epidemic. Some businesses will emerge stronger and better positioned to succeed, and other businesses will fail. These changes may occur rapidly, but it will all be in the data. We are currently evaluating news and commercial activity across all sectors, globally and through several different lenses. For example, one thesis is that companies who treat their customers and employees well during this period of distress will be rewarded as the crisis passes. The data science team is working closely with the investment teams to focus our efforts.
A wide range of big data is potentially useful in the investment context. This includes “hard” data such as supplier payments, prescriptions and health care insurance claims, and laboratory results; but there is also rich information in “softer” data, such as news and social media keywords or satellite imagery. Overall, the categories that we see getting the most mileage for long-term investors include credit/debit cards and bank accounts, online transactions, internet content and search, online job postings, and investment conference call transcripts and filings.
Typically, the first major task of any data scientist is to determine which data has the potential to be useful—and which may not. A particular dataset may be of poor quality; it may be very interesting but immaterial; it may be useful for one manager type (e.g., a high volume trader) but not for another (a long-term value investor).
The next task is to take potentially useful data and make it actually useful. This requires the development of clear research hypotheses; it also requires the cleaning and interpretation of raw datasets. Having the expertise and the infrastructure to do that is, in our opinion, clearly vital. But so is being an engaged, responsive and demanding consumer of data, rather than a passive one.
Benefits to Fundamental and Quant Investors
We believe that getting the most out of big data is not just a matter of investing in personnel and computing power, but also of fully integrating data science into the traditional disciplines—whether fundamental or quantitative. Analysts with a fundamental approach often consider a vast array of information and data to understand the condition of, and prospects for, a given company—they are what most people think of when they hear the term “securities analyst.” The knowledge of such professionals is typically very deep, but comparatively narrow, encompassing a small number of companies.
Quantitative investors, in contrast, tend to be among the PhDs in the office, generally using advanced mathematical modeling, computer systems and data analysis to identify and execute on potentially profitable investment positions. Their knowledge is often broad but comparatively shallow, with an understanding of a few key things about every listed company in their investment universe. Data science can help fill the gaps in each approach by providing a more detailed view of the world, offering more texture to fundamental investment hypotheses and potentially corroborating or challenging the signals that quants may identify based on other metrics.
Importantly, portfolio managers and analysts from both camps are a vital source of the research hypotheses that define the scope of data scientists’ work and bring shape and coherence to otherwise raw data sets. Data science can help provide many answers, but it would not even know the questions without the insights of experienced investors.
Case Studies: Big Data in Action
Thus far, we’ve talked in general terms about big data and its ability, in the right hands, to develop meaningful investment insights. However, it’s probably best understood through the use of concrete examples. The rest of this article provides an array of real-life instances in which our data scientists were able to identify patterns that informed the views of our research analysts and portfolio managers.
Bank Deposits Point to Oil Production Levels
It is fairly intuitive that credit card and bank account data can tell us something about who is buying what and from whom, giving us insights into brand loyalty and sales performance for companies, stores and products. But would you have guessed that it can also provide a real-time insight into how much oil the U.S. is producing?
Some 20,000 – 50,000 Americans lease their land to petroleum fracking companies. A number of trusts manage this process, paying them royalties that are proportionate to the amount of oil being extracted at any point in time. By analyzing the aggregate trust deposits into these lessors’ bank accounts, it was possible to figure out the statistical relationship between the payments and oil extraction generally, thus creating a daily estimate of U.S. oil production.
We believe there is more to bank account data than consumer spending insights coming from payments out of accounts. Payments into accounts include things like these oil royalties and paychecks—key information for exploring what’s happening behind some of the most important macroeconomic data releases.
Job Postings Reveal Capex Commitments
When the economy was powered by traditional manufacturing, we could get a good sense of business confidence from the capital expenditures that companies reported. Investments might start to show up in revenues six or 12 months later, and in aggregate, could provide insight into where we were in the business cycle.
Those signals are much weaker in our new economy of services and technology. This is a world of operational expenditure—wages, salaries, rents and the like—rather than capital expenditure. Hiring engineers and designers is the modern world’s capex, but that information is not reported in quarterly corporate filings.
It is available in the form of job postings, however. We can now collect these from more than 8,000 U.S. public companies, representing almost three-quarters of all the job advertisements in the U.S. That can help us to see whether a company is hiring lots of engineers and designers in its early life and then filling positions in sales and marketing as it matures, providing insight into management’s confidence in its business model.
Job postings can also tell us about the performance of certain suppliers to the companies that are hiring. For example, when Microsoft’s Azure started to surprise with its success in the cloud computing market, our technology analysts looked for real-time confirmation of the sustainability of this trend in IT job postings, the nature of which could shed light on future cloud expenditures. Based on our results, we believed that Azure was likely to sustain its success—but also that it was a preferred solution in the financial sector.
The Downside of ‘Very Optimistic’ Earnings Calls
Our firm has developed a natural language processing model that has identified more than 5,000 phrases of two or more words, which, when they appear in the transcripts of quarterly earnings calls, it may consider to carry positive or negative implications about the company.
Sometimes it is intuitive that these phrases should be construed as positive or negative. For example, “raise guidance” is seen as positive and “reduce guidance” is seen as negative. Similarly, “ahead of schedule” is better than “slower than.” We would rather see “repurchase shares” than “equity offering.”
But is “very optimistic” positive or negative? Correlation between this bigram (two-word combination) and subsequent stock performance, picked up from thousands of transcripts, indicates that it is a bad sign, on average. Management tends to say it when results have been poor and they want to persuade investors that a turnaround is on the way. Bigrams such as “closer look,” “briefly review” or “quick update” may seem completely neutral, but analysis shows that the first tends to be positive and the other two negative—because management likes to get into details that reflect well on the business and gloss over the problems.
Extract From an Earnings Call
Source: Thinknum. For illustrative purposes only.
Are Rewarded Customers Good Customers?
One of our portfolio management teams wanted to know what benefits Starbucks was getting from its customer loyalty program. Our data science team turned to credit and debit card transactions for answers.
After identifying loyalty-program members, it found that they spent around $88 on average each quarter versus the non-member spend of around $30. We were also able to show that average revenue per user jumped substantially at the point of conversion to loyalty-program membership and that, for all but the very highest-spending customers, it remained high and even rose further over subsequent years.
Furthermore, the revenue growth associated with our sample of reward members closely tracked revenue growth subsequently reported by Starbucks. In our view, this correlation could be a tool when analyzing future performance.
Loyalty Pays at Starbucks
Quarterly Average Revenue per User
Source: Second Measure, Neuberger Berman.
Would Investors Pick Up Lyft and Uber?
The market leaders in ride-hailing apps both took their stock public last year, and both similarly suffered inauspicious first-day trading. Our analysts were not surprised by this because our data science team had confirmed their existing concerns about potential rising costs and declining sales growth.
On the costs side, from direct deposit information we were able to identify Lyft and Uber drivers from our sample of bank account data. Through 2017 and 2018 that data showed a rising percentage of drivers working for both companies, which points to declining driver loyalty and increasing driver churn even as the absolute number of drivers was rising.
On the sales side, from credit and debit card transactions we found evidence of rapidly declining new customer growth for all three of the leading ride-hailing firms. That suggests that future revenue growth could be ever more reliant on increasing use by existing customers.
Dwindling Customer Growth
New Customer Rate: Rideshare Rivals
Source: Second Measure, Neuberger Berman.
This example shows how data can help provide a unique window into the true performance of pre-IPO companies before they begin reporting publicly. It is also a great reminder that there may be more to bank account data than consumer spending insights. Payments into accounts include paychecks—key information for exploring what’s happening to the labor mix and labor costs for individual companies, sectors and the economy as a whole.
Is Comcast a Good Place to Work?
Big data has a key role to play in analyzing environmental, social and governance (ESG) factors because many of them are either not reported or not standardized to afford useful comparison. The problem is most acute when it comes to “softer” social factors like human capital management.
U.S. telecom giant Comcast appears to have good reason to promote itself as a rewarding place to work, ranking high in multiple corporate and workplace comparisons, including for disabled employees. However, ESG rating agencies saw a different picture. One agency pointed out a planned acquisition and “multiple labor controversies” in rating the company far lower on labor management than the sector average.
Whom to believe? Big data provided some perspective through the opinions, reviews and ratings left by employees on the recruitment website Glassdoor. When we “scraped” those for the telecom sector, we found that Comcast ranked well above average. A look at job postings also helped. We like to compare the proportion of a company’s workforce that is represented by currently live job postings with subsequent expense growth. We believe that gives us an insight into how many of those job postings relate to genuine expansion of employment at the company and how many are due to churning of the same role. On that metric, Comcast’s ratio of 0.38 compared well with its sector peers, ranking seventh out of 22 companies.
Conclusion: Capitalizing on Big Data
Big data is enjoying a surge of popularity, but whether this trend lasts may depend on expertise and execution. Indeed, many of the players that have tiptoed into data science could eventually become disillusioned when they find that data can be hard to read and potentially misleading without the right people and infrastructure to make sense of things. A real pitfall also would be to assume that big data can be an end in itself. Yes, it has great potential, but from our perspective it will likely prove more useful if integrated with more traditional research and quantitative techniques in seeking to produce genuine insights with meaningful investment implications. Managers who understand this dynamic, and are willing to commit sufficient time and resources, are more likely to enjoy success in capitalizing on big data over the long haul.