AI trading strategies self-destruct at scale, paper finds 18-month alpha half-life

A new paper on arXiv by Shuchen Meng and Xupeng Chen—titled "AI-Driven Alpha Decay: Algorithmic Homogenization, Reflexive Signal Erosion, and the Paradox of Intelligent Markets"—delivers a sobering verdict: the more AI is used in quantitative finance, the less effective it becomes. The authors derive an alpha half-life formula h(φ) = ln2 / [θ + δ(φ)], where δ(φ) = Nφρa/λ(φ) is the AI-accelerated decay component. At current adoption levels (φ ≈ 0.7, ρ ≈ 0.6), signal half-lives shrink to 18 months, compared to 5–7 years in pre-AI markets. The mechanism is threefold: signal crowding as too many funds chase the same patterns, performative signal erosion as trades alter the data they're based on, and Red Queen competition forcing ever-faster races that consume returns.

The paper establishes four theoretical results: an alpha half-life theorem, a signal extinction cascade beyond a critical threshold φ*, a Red Queen impossibility where net alpha becomes zero at equilibrium despite heavy investment, and a fragility-efficiency tradeoff where the adoption level maximizing price discovery exceeds that minimizing systemic risk. Empirical validation uses SEC Form 13F filings covering 99.5 million holdings from 2013–2024, showing simulated institutional portfolio convergence increased by 42%. The authors also examine simulated hedge fund return dispersion declining among AI-adopting funds and simulate the 2010 Flash Crash to illustrate fragility. The work suggests a paradox: intelligent markets become less profitable and more fragile as they become more homogeneously intelligent.