Alternative data sets can provide a treasure trove of insight - if you know where to look. One fund came up with the idea to analyze employee feedback to predict future company performance
Traditionally, a company's financials are used as signals to determine the trajectory a company is following. Repeatedly solid quaterly results convey a company on the ascent. But that information is public, anyone can use it and everyone does use it. How can a trading desk get a leg up on its counterparts if they're all using the same information?
The next step is when Repustate comes in. Using a customized language model for sentiment and classification, Repustate classified reviews into various buckets and then applied sentiment to those buckets. For example, one bucket would be "compensation" and all aspects of a review that had to do with compensation would be assigned accordingly e.g. "granted stock options", "no raises given", "xmas bonus"
Consider the following review:
"Extensive use of outdated technology clashes with the hi-tech image the company has (or wants to have). Not enough FTE, too many temps, consultants, etc, who will be gone in a few months."
A quote like this could be categorized as follows:
|Head count||not enough FTE||negative|
|Head count||too many temps||negative|
|Head count||consultants etc. who will be gone in a few months||negative|
From this one quote alone, one could assume things don't look very promising for this company. Encounter enough quotes like this and a very negative trend can become obvious.
With some analysis in place, backtesting can be done to see if historical prices could be forecasted using employee (dis)satisfaction as a primary signal. Maybe only certain aspects of employee reviews prove to be good indicators (compensation for example) and others are just noise (parental leave policy).
Alternative data sets like these are growing in popularity and Repustate's text analytics API can help extract any alpha that might be hidden in plain sight.