Acadian ‘Un-follows’ Microsoft Bing

Acadian hoped to use macroeconomic signals generated by Bing Predicts’ social media and internet search data—such as US retail sales and consumer sentiment—in its macro models in particular, while Bing Predicts had intended to use Acadian’s expertise to make more economic predictions. But Acadian officials say the firm found limited use for Bing Predicts’ data.

“We’ve analyzed social media data, but we haven’t found anything particularly useful,” says Mark Webster, portfolio manager at Acadian, adding that the firm looks at “many different data vendors” and frequently tests different datasets. 

“We are still experimenting. We were trying to use Bing for our macro forecasts in particular. They have a very rich set of data, and some quite encouraging results came out of that,” Webster says, noting that while other hedge funds may find the data profitable, the problem for Acadian—which holds equities for a time horizon of three to five months—is that the short-term signals generated by social media don’t immediately lend themselves to offering viable inputs to generate a medium- to long-term trading model.

“It’s challenging blending short signals with long signals. If we were reliant on three signals, and were happy to trade portfolios multiple times a day, we could produce something convincing. I have no doubt others are going to extract some juice out of this, but there is lower-hanging fruit for us that would have a bigger impact,” he says.

Typically, hedge funds have found social media to be most useful as a tool for quickly identifying and responding to news alerts—such as mergers and acquisitions, oil shocks, drug research or product releases—and market sentiment, which give short-term momentum signals and equity valuation signals after a news event. But large quant firms have so far proved resistant to buying social media data that generates such short-term signals.

Steven Schwartz, president of commercial markets at New York-based social media analytics provider Dataminr, wants to attract quant funds to use its alternative data, but says: “The jury is still out if something as unstructured as half a billion tweets a day is something that translates into a quantitative [investing] approach,” he says.

In addition, the complexities and costs of handling social media data mean some firms are automatically excluded from being able to invest in it. “Alternative data is so unique and novel it commands a much higher a price than your average news dataset, [which] is a natural barrier to entry for smaller asset management firms. They don’t have the budgets to deploy,” says Pierce Crosby, director of business development at investing social media platform StockTwits. 

Less is More

Acadian has found more value in relatively orthodox, less “sexy” data types than areas such as social media, satellite imagery or credit card data—all of which it finds are not broad-based enough to produce robust models.

One example of less sexy data is information on new cost of borrowing, derived from prime brokerage relationships with a number of banks used by the firm’s long/short equity funds. This data is not published or easily available.

Another source the firm has found useful as a signal recently is fund holdings data, which can indicate when a large, important mutual fund owns two companies. The price movements of companies held by the same mutual fund are often correlated to some extent.

A more novel signal—derived by using natural language processing—is a firm’s economic links with other sometimes apparently unrelated companies. A company with business links to a firm that goes bust might be expected to suffer similarly. Acadian’s analysis allows it to tell, for example, when an oil price cut will hurt the supplier of an important client of another company three steps removed.

“This may not sound super-sexy from a Twitter or satellite imagery perspective, but [these data sources] are really powerful. They have the breadth, and they’re really important signals,” Webster says.