AI is generating new financial theories faster than the actual history of the world can prove them true.
March 31, 2026
Original Paper
The Last Paper - Agentic AI and the Governance of Empirical Finance
SSRN · 6422458
The Takeaway
Credible financial research requires rare historical 'shocks' to prove cause-and-effect. While AI can run millions of statistical tests, it cannot create more history, leading to an era of 'data exhaustion' where the number of discovered 'patterns' in the market vastly exceeds the amount of data available to verify if they are real or just statistical noise.
From the abstract
Much of empirical finance works by searching already-realized data-decades of returns, filings, and transactions that sit in queryable databases. Agentic AI makes that search very fast. It does not make the data richer. The historical shocks, regulatory discontinuities, and natural experiments needed for credible causal inference are features of the past, not products of faster computing. The result is an asymmetry: the number of statistical tests a researcher can run is exploding, but the numbe