Big Data in Hedge Funds: Utilizing Alternative Data Sources

Alternative Data

Introduction
The financial services industry, and hedge funds in particular, have been at the forefront of data science since well before “big data” was even a phrase. Michael Bloomberg started his eponymous company in 1981 after realizing there a gaping hole in the market for providing financial data. As Bloomberg and its competitors grew, quant hedge funds like Renaissance Technologies sprung up to take advantage of the data. These early funds built mathematical models to predict market behavior, and traded assets accordingly. Formulas like the BlackScholes equation for option-pricing assumed price movements were random. These strategies then evolved into pure data mining approaches, where computers were trained to spot patterns in terabytes of price data and build models themselves. These models were then utilized in high frequency trading schemes, where assets are bought and sold in mere seconds to take advantage of small momentum swings. 

Within the last decade, a third era in big data hedge funds began, utilizing alternative data sources like satellite imagery, web scraping, and geolocation. Whereas previously, quant hedge funds eschewed traditional, fundamentals-based investing, now they are trying to augment it with big data analysis. The surest way to beat the market is to make more accurate predictions of a company’s revenues and earnings because ultimately, these are what drives a company’s value. On the day companies report earnings every quarter, their share prices typically jump one way or the other as the market reacts to the new information. An investor who can accurately predict an earnings or revenue “beat” or “miss” can buy or short shares accordingly and have a high probability of making money. Alternative data is in conjunction 

with more traditional methods of forecasting to increase prediction accuracy and generate market-beating returns. There are, however, some drawbacks to the use of alternative data. Firstly, there are legal concerns revolving around material non-public information and data privacy. Second, there is the high cost of obtaining data, figuring out what to do with it, and finding talent to tease insights out of it. Finally, there are questions about the value of the data, especially now that its use it becoming widespread.

Alternative Data
Alternative data refers to anything that is raw or unstructured, and is distinct from things like company filings, historic market prices, or investor presentations. Web scraping, credit card transactions, satellite imagery, social media posts, app check-ins, and have all been used to try and estimate company revenues. Some data sources, like web scraping and satellite imagery, are less accurate, but can capture a larger percentage of a company’s sales. Other sources, like credit card transactions, are highly accurate, but capture only a very small sample of a company’s sales. 

Legal Concerns

 

Web scraping involves extracting data from webpages like prices, user reviews, and seat/room availability for airlines/hotels. It is particularly useful for companies that sell mostly or entirely online, like Amazon or Priceline. Satellite imagery is analyzed using image recognition techniques to estimate foot traffic at physical retail locations. It is useful for companies like Walmart or Costco that collect most of their sales in person. Social media posts can be analyzed using natural language processing and sentiments extracted. These can be used to assess how successful a new product launch has been or how many people watched an  episode of some TV show. Twitter has made its data accessible through its API, thus it is the most widely used social media data.

Credit card transaction data is great because it tells you exactly what has been purchased, where, and when, but it cannot be collected for very many customers, and thus making inferences from it carries a large margin of error. A final category of alternative data, though there are many other sources not mentioned here, comes from the most ubiquitous of all technologies today: the smartphone. Most people carry their smartphone with them all day, every day, so the location data generated is invaluable. Some of it can be obtained from wireless carriers. Some apps, like Foursquare, involve customers marking where they’ve been, so can be scraped just like a webpage When these data sources are analyzed periodically and for many different products, a reasonably accurate estimation of revenue can be made. The key is using multiple indicators in tandem to gain a more comprehensive understanding of sales. For example, satellite imagery might indicate that a Target parking lot received 100,000 vehicles in the last quarter and 150,000 this quarter. Logically, one might conclude that sales rose 50%, but actually more
people were using public transportation to get to Target last quarter and made smaller purchases on average this quarter, so sales only rose 25%. Geolocation and credit card data might have been able to identify these hidden trends, and thus create a more accurate prediction. 

However, alternative data has its limits. Thus far, it has been most effective in predicting revenues for consumer-facing companies with many customers, i.e. retailers, consumer discretionary, consumer staples, restaurant chains, some tech companies, and some industrials. It has yet to be effective for B2B companies, especially ones with few clients and who sell services or software. These companies are in sectors like financial services, health care, aerospace, consulting, materials, and utilities. 

Legal Concerns
Legal concerns surrounding alternative data involve public access to the data and potential consumer privacy violations. Insider trading laws prohibit the use of material nonpublic information when making investments in public securities. A common example of this is when an M&A analyst at a bank is working on an acquisition and buys stock in the company being acquired before the transaction is made public. Alternative data is clearly material information, since knowledge of it has the potential to change one’s valuation of a company. The question, therefore, is whether it is truly public information. Alternative data vendors sell datasets for hundreds of thousands of dollars, making it cost-prohibitive for
virtually any individual investor. Many datasets are also purchased with an exclusivity agreement that prohibits further sales of the data. 

 

Other Drawbacks
Further issues surrounding alternative data stem from its complexity. As mentioned previously, the data itself comes with a high cost. Then, there is the even greater cost of hiring data scientists to analyze the data and buying the necessary hardware and software. Alternative datasets are often unstructured and large, thus requiring the use of NoSQL databases and distributed computing systems like Hadoop. These tools are not easy to use and knowledge of them is not widespread, so hiring the right people to implement them takes time and money. 

Conclusion
Despite the drawbacks mentioned in the last couple sections, alternative data still offers great potential. First, when implemented properly, it offers a very good chance to make money, as it can provide more accurate predictions for revenue and earnings, which are at the heart of stock valuations. Second, its adoption has not been widespread enough to eliminate potential advantage. There are still many data sources, e.g. video feeds, yet to be utilized that could all offer a competitive advantage. There is also the possibility of performing more complex analyses on currently available data and further increasing the prediction accuracy.