Data Cost vs Alpha ROI

Alternative datasets — satellite imagery counting parking-lot cars, credit card panels, web-scraped pricing — routinely cost anywhere from tens of thousands to several million dollars a year in licensing fees. Before buying, a fund needs to estimate whether the dataset's expected contribution to strategy returns actually clears that cost, not just whether the data sounds interesting.

The basic calculation compares the dataset's annual cost against the extra dollar profit its signal is expected to add, scaled by how much capital that signal can actually be deployed against — a highly predictive signal that only works on $20 million of AUM may still lose money against a $2 million data bill, while a modestly predictive signal usable across a $2 billion book can be worth it even at a high price. This is why the same dataset can be a clear buy for a large multi-strategy fund and a clear pass for a small one running the identical model.

A fund evaluating a $500,000/year web-scraped pricing feed might estimate it adds 15 basis points of annual alpha, deployable across a $400 million allocation — roughly $600,000 in expected annual profit before costs, a modest but positive net after the $500,000 fee, though thin enough that a modeling error in the alpha estimate easily flips the sign.