Hedge Funds Use Shopping Center Cameras in Hunt for Alpha

Prior to the spread of the coronavirus, the US stock market was nearing an historic milestone: the S&P 500 equity benchmark was 5% away from posting its largest ever rise without falling more than 20%—a bull run not seen since the tech boom of the 1990s, according to the Financial Times. The market has started to stabilize, and the bull run is expected to continue, at least in the near term.

If investors can see their money grow by simply following an index, what’s the point of paying a fee to a hedge fund manager? This is leading buy-side firms to increasingly look to alternative datasets in order to find alpha and justify their services. Tim Harrington, co-founder and CEO of alt data platform provider BattleFin, says that, specifically, sentiment data is seeing a growing interest of late, with insights taken from camera footage in and around shopping centers becoming one particular area of demand.

“We’ve had a massive bull market for so long, it’s very hard to just rely on earnings estimates and private equity ratios,” he says. “[Portfolio managers] are really looking for the shift to sentiment.”

Harrington says that the demand for sentiment data is driven by buy-side firms looking to better predict the long-term valuations of company shares. This is where shopper sentiment comes into play. The idea is that valuable trading insights can be derived from camera footage in and around shopping centers to evaluate how customers might feel or react to a product or brand.

The data can also be used to assess factors such as age, gender, and shopper experience, which can also offer some indication of how well that company is performing amongst key demographics. This is done using biometric techniques and a training model to detect and analyze the facial features and expressions of individuals, says Geoff Horrell, head of the London Labs unit at Refinitiv. (Last year, Refinitiv made a strategic investment in BattleFin.)

“The industry uses mobile data to work out how many people have visited [the Westfield, Stratford shopping center in London], but if you can also detect if they were happy or sad, if they were a man or a woman, if they were old or young—that information can potentially give you a few early alerts to revenue or potential sales generated for retail,” he says.

To some, this demand for shopper sentiment is seen as a natural progression. While retailers stockpile this type of data for security reasons, to inform their ad spending, and for marketing, the next logical use-case is for monetizing these insights.

However, sentiment data is only one small piece of the puzzle.

Harrington says that sentiment analysis should be used alongside traditional market and reference data, as well as other forms of alternative datasets—like geolocation, web data, credit card transactions, and receipt data—to create company profiles and performance measurements.

“On the other side of this, you can come up with all this alternative data, build your models, but unless you really have fundamental and reference data … you only have half the equation. So, you really need alternative data, plus fundamental data to really execute or use this effectively,” he adds.

Ethics and Bias

This kind of information is likely to become more prevalent in the future with the advent of 5G networks and the Internet of Things (IoT). But, there are major security and privacy concerns that regulators around the globe are currently grappling with.

Knowing that traders are using your facial expressions and physical data to inform their strategies is enough to make anyone feel uneasy. This is where data providers and buy-side firms using this type of data will have to scrutinize how they source and manage this sensitive information to avoid breaching data privacy regulations, such as the General Data Protection Regulation (GDPR) in Europe.

Harrington says that data collectors like retailers or phone app providers will have to be fully transparent in how they use their customer’s data.

“We want clear rules that when you are signing up for that free service, it says: ‘You’re signing up for this free app; in exchange for the free app, we’re going to anonymize your information and we have the ability to sell it,’” Harrington says.

Other pressing challenges with using personalized or biometric data include bias. Machine-learning models require vast amounts of training data to accurately identify or distinguish factors—for example, whether a person is laughing or crying, or whether someone is male or female.

“To train a model and verify that the information is accurate, you need a lot of information about those individuals,” Horrell says. “One of the things that we talk about is the potential for bias. To train your facial recognition system, you need to have a good cross-section of people from a certain age, gender, race, people with different heights, complexions, and hairstyles—[this is] to do effective training and make sure that you’re not introducing bias.”