Alternative-data alpha is an identification premium, not an information premium
Alternative data is observational data wearing an experimental costume: its alpha is an identification premium, not an information premium. What a fund pays for is not the satellite photo or the card-swipe panel itself but the unresolved ambiguity of the adjustment model needed to turn it into a causal claim about earnings. While that ambiguity is open, the dataset prices like alpha. The moment the adjustment standardizes — published pipelines, vendor “research-ready” panels, robust defaults — the method commoditizes and the edge decays, even though the data itself is unchanged. It is the method, not the data, that gets arbitraged away.
The decay of published predictors once they are known is documented (McLean & Pontiff, Does Academic Research Destroy Stock Return Predictability?, Journal of Finance, 2016). The fresh, falsifiable part is locating the edge in the ambiguity of the adjustment model, not the data or the crowd.
How to prove it wrongIf alternative datasets with fully standardized, vendor-published preprocessing retain abnormal returns as long as bespoke, adjustment-ambiguous datasets of equal exclusivity, the identification-premium thesis is wrong — the value would be in the data after all.