AI-Driven Alpha Decay: Algorithmic Homogenization, Reflexive Signal Erosion, and the Paradox of Intelligent Markets

Shuchen Meng; Xupeng Chen

q-fin.GNMar 23, 2026

AI-Driven Alpha Decay: Algorithmic Homogenization, Reflexive Signal Erosion, and the Paradox of Intelligent Markets

Authors: Shuchen Meng, Xupeng Chen

Organizations: Department of Financial Engineering, New York University. · Department of Electrical and Computer Engineering, New York University.

Abstract

We show that AI-driven investment strategies are inherently self-defeating at scale. As AI adoption rises, three mutually reinforcing channels -- signal crowding, performative signal erosion, and Red Queen competition -- compress excess returns. We derive the alpha half-life $h(φ) = \ln 2/[θ+ δ(φ)]$ , where $θ$ is the natural mean-reversion rate and $δ(φ) = Nφρa/λ(φ)$ is the AI-accelerated decay component, which is convex-decreasing in adoption. At current adoption levels ( $φ\approx 0.7$ , $ρ\approx 0.6$ ), the model implies signal half-lives of 18 months versus 5-7 years pre-AI. We establish four theoretical results. First, the alpha half-life theorem: signal lifespans are convex-decreasing in AI adoption. Second, a signal extinction cascade: beyond a critical threshold $φ^*$ , the decay of one signal class triggers accelerated competition for remaining signals. Third, a Red Queen impossibility: in the monoculture equilibrium, net alpha is identically zero despite heavy AI investment. Fourth, a fragility-efficiency tradeoff: the adoption level maximizing price discovery strictly exceeds the level minimizing systemic fragility. Empirical validation calibrates portfolio convergence to SEC Form 13F filing patterns (99.5 million holdings, 2013-2024), documenting that simulated institutional portfolio convergence increases by 42% over the sample period. We examine simulated hedge fund return dynamics showing declining cross-sectional dispersion among AI-adopting funds, and simulate the 2010 Flash Crash to illustrate fragility consequences.

Figures & tables

Figure 1: Conceptual framework: three channels of AI-driven alpha decay. Layer 1 (Signal Crowding): convergence of AI trading signals due to shared training data and model architectures accelerates arbitrage of predictable patterns. Layer 2 (Performative Erosion): feedback from trading activity degrades signal quality through market microstructure alteration, producing a signal extinction cascade. Layer 3 (Red Queen): competitive adoption game generates overinvestment in AI and systemic externalities. Dashed arrows indicate feedback loops that reinforce the decay dynamics. Bottom boxes summarize the four main predictions.

Figure 2: The four-phase alpha lifecycle under AI adoption. Phase I (Discovery): early AI adopters extract genuine alpha; signal half-life

\approx

5–7 years. Phase II (Compression): mainstream adoption drives convergence;

h\approx

2–4 years. Phase III (Crowding): signal crowding dominates;

h<18

months. Phase IV (Red Queen): net alpha is zero in aggregate; market is efficient but fragile.

Symbol	Description	Interpretation	Range
$\phi$	AI adoption rate	Fraction of institutions using AI	$[0,1]$
$\rho_{k}$	Algorithmic homogeneity	Correlation of AI signals for factor $k$	$[0,1]$
$\theta_{k}$	Natural mean-reversion	Intrinsic decay rate of signal $k$ (month -1 )	$>0$
$\delta(\phi)$	Endogenous decay rate	AI-driven acceleration of arbitrage	$>0$
$\beta$	Performative feedback	Signal-to-price feedback intensity	$[0,1)$
$g_{k}$	Signal regeneration	Rate of new mispricing arrival	$>0$

Table 1: Summary of Model Parameters

Panel of Figure 3 — Figure 3: Theoretical predictions. (a) : Alpha half-life as a function of AI adoption $\phi$ for different homogeneity levels $\rho\in\{0.2,0.4,0.6,0.8\}$ . The half-life is convex-decreasing in $\phi$ . Horizontal dashed lines indicate pre-AI benchmark (60 months) and current estimate (18 months). (b) : Cross-sectional return dispersion for AI funds (blue) and human funds (gray). AI fund dispersion declines steeply with adoption.

Panel of Figure 4 — Figure 4: (a) : Sequential signal extinction under increasing AI adoption. Each line represents one of $K=5$ tradeable signals; signals die off sequentially as AI adoption increases through Phase III. Signals with higher regeneration rates $g_{k}$ survive longer. (b) : 3D surface of the reflexive multiplier $\mathcal{M}(\phi,\rho)$ for $\beta=0.25$ . The nonlinear explosion near the stability boundary illustrates the disproportionate fragility risk at high adoption.

	Model Layer
Equilibrium Property	Layer 1	$+$ Layer 2	$+$ Layer 3
	(Signal Crowding)	(Performative Erosion)	(Red Queen)
Alpha decay $h(\phi)$ decreasing	✓	✓	✓
Convex acceleration of decay	✓	✓	✓
Return dispersion compression	✓	✓	✓
Signal extinction cascade	—	✓	✓

Table 2: Model Hierarchy: What Each Layer Enables

Panel of Figure 5 — Figure 5: (a) : Adoption incentive $\Delta\Pi(\phi)$ showing multiple equilibria: stable diversified ( $\phi_{L}$ ), unstable tipping point ( $\phi_{U}$ ), and stable Red Queen ( $\phi_{H}$ ). Career concern shifts push the system toward the monoculture. (b) : Price discovery welfare $W_{eff}(\phi)$ and stability welfare $W_{stab}(\phi)$ . The market equilibrium $\phi^{*}$ overshoots the efficiency-optimal and socially optimal adoption levels.

Strategy	Variation	Key Assumption	Expected Sign	Main Threat
IV: Cloud costs	Time-series $\times$ cross-section	Compute costs $\perp$ equity returns (conditional on $\mathbf{X}$ )	$\hat{\beta}_{1}^{(1)}<0$ , $\hat{\beta}_{1}^{(3)}>0$	Correlated tech shocks
DiD: LLM releases	Event $\times$ NLP exposure	Parallel pre-trends; LLM timing exogenous to markets	$\hat{\beta}_{\tau}<0$ for $\tau\geq 0$	Contemporaneous confounds
Observable $\hat{\rho}$	Cross-section of homogeneity	$\hat{\rho}$ proxies true algorithmic correlation	$\hat{\beta}_{3}<0$ (interaction)	Measurement error in $\hat{\rho}$

Table 3: Identification Strategy Summary

Figure 6: Average pairwise cosine similarity of institutional portfolios from SEC Form 13F filings (2013Q1–2024Q4). AI–AI pairs (blue) show a 58% increase, compared to 19% for non-AI pairs (gray). Vertical dashed lines indicate structural break dates (Bai-Perron test): 2018Q2 (deep learning mainstreaming), 2020Q3 (COVID regime shift), 2023Q1 (GPT-4 adoption).

Panel of Figure 7 — Figure 7: Pairwise cosine similarity heatmaps for 30 institutions (15 AI-adopting, 15 non-AI). (a) : In 2014, similarity is diffuse. (b) : By 2024, the AI block (upper-left) shows markedly higher convergence, consistent with signal crowding.

Panel of Figure 8 — Figure 8: (a) : Rolling 12-month cross-sectional standard deviation of monthly returns. AI/quantitative funds (blue) show a 29% decline versus 10% for fundamental/human funds (gray). (b) : Rolling 36-month risk-adjusted alpha for median AI fund (blue) and median human fund (gray), with 95% confidence bands.

Figure 9: Estimated half-life of factor strategies by vintage. All factors show dramatic shortening: Momentum from 84 to 12 months, Value from 72 to 20 months. The model prediction (dashed line) closely matches the empirical estimates.

Panel of Figure 10 — Figure 10: (a) : Simulated Flash Crash price path (black) versus fundamental value (green dashed). Shaded area represents algorithmic amplification ( $\mathcal{M}\approx 1.4$ ). (b) : Scatter plot of fundamental shock versus observed price decline across 20 simulated stress events. Regression slope $>1$ confirms systematic amplification.

Panel of Figure 11 — Figure 11: (a) : EDGAR full-text analysis showing $>$ 50 $\times$ growth in AI-related keywords (bars, right axis) and percentage of institutions mentioning AI (line, left axis). (b) : AI disclosure intensity versus portfolio convergence. Positive correlation ( $R^{2}$ shown) confirms the signal-crowding mechanism.

Figure 12: Sensitivity of the alpha half-life

h(\phi)

to model parameters. Each panel varies one parameter while holding others at baseline values (

\phi=0.7

\rho=0.6

\beta=0.25

N=100

K=5

). The half-life is most sensitive to

\rho

(homogeneity) and

N

(number of AI competitors).

Parameter	Estimate	Source	Method
$\phi$	0.70	13F convergence trends	Portfolio similarity analysis
$\rho$	0.60	Within-AI-group dispersion	Return decomposition
$\beta$	0.25	Price–fundamental gap	Performative attenuation (Eq. 14 )
$g_{k}$	0.02	Factor regeneration	New-factor discovery rate post-extinction
$\kappa$	1.0	Baseline (linear)	Robustness check at $\kappa=2$
$h_{k}(0)$	58 months (pre-AI)	McLean and Pontiff, (2016)	Half-life $\ln 2/\theta_{k}$ ; $\theta_{k}=0.012$ month -1

Table 4: Calibration Summary

Result	Mechanism	Testable Prediction	Ref.
Alpha Half-Life Theorem	Signal crowding by correlated AI traders accelerates arbitrage	Factor half-lives shorten convexly with AI adoption; currently $\approx$ 18 months vs. 60 months pre-AI	Prop. 3.3
Signal Extinction Cascade	Performative feedback erodes signal quality; death of one signal increases pressure on survivors	Sequential factor death ordered by crowding intensity; momentum dies first	Prop. 3.7
Red Queen Equilibrium	Career concerns lock in overinvestment; net alpha is zero	Aggregate AI fund alpha $\to 0$ despite rising AI expenditure; arms race in compute/data	Prop. 3.10
Efficiency–Fragility Tradeoff	Price discovery improves but tail risk grows convexly	Flash crash amplification $\mathcal{M}>1$ increasing with $\phi$ ; “calm before the storm” pattern	Prop. 3.11

Table 5: Summary of Main Results and Testable Predictions

Finance

Finance studies markets, institutions, and financial decisions using data and models to understand risk, value, and resource allocation.

See all 281 papers

Artificial Intelligence and Systemic Risk: A Unified Model of Performative Prediction, Algorithmic Herding, and Cognitive Dependency in Financial Markets
Mar 23, 2026Shuchen Meng, Xupeng ChenPerformative Prediction Algorithmic Trading
Department of Financial Engineering, New York University. · Department of Electrical and Computer Engineering, New York University.
AI Trading's Alpha Singularity: Emergent Market Reasoning through Agent-to-Agent Self-Evolution
Jun 28, 2026Yuqi Li, Siyuan Liu, Bingjun LiuFinance Emergent Behavior in Multi-Agent Systems
Panda AI
Representation Homogeneity and Systemic Instability in AI-Dominated Financial Markets: A Structural Approach
Apr 14, 2026Yimeng Qiu, Qiwei HanAlgorithmic Trading Emergent Behavior in Multi-Agent Systems

Mar 23, 2026J/K move · Enter open · S save

Artificial Intelligence and Systemic Risk: A Unified Model of Performative Prediction, Algorithmic Herding, and Cognitive Dependency in Financial Markets

Shuchen Meng, Xupeng Chen

Department of Financial Engineering, New York University. · Department of Electrical and Computer Engineering, New York University.

Performative Prediction Algorithmic Trading Cognitive Offloading

Open paper AskPDF arXiv

We develop a unified model in which AI adoption in financial markets generates systemic risk through three mutually reinforcing channels: performative prediction, algorithmic herding, and cognitive dependency. Within an extended rational expectations framework with endogenous adoption, we derive an equilibrium systemic risk coupling

r(φ) = φρβ/λ'(φ)

, where

φ

is the AI adoption share,

ρ

the algorithmic signal correlation,

β

the performative feedback intensity, and

λ'(φ)

the endogenous effective price impact. Because

λ'(φ)

is decreasing in

φ

, the coupling is convex in adoption, implying that the systemic risk multiplier

M = (1 - r)^{-1}

grows superlinearly as AI penetration increases. The model is developed in three layers. First, endogenous fragility: market depth is decreasing and convex in AI adoption. Second, embedding the convex coupling within a supermodular adoption game produces a saddle-node bifurcation into an algorithmic monoculture. Third, cognitive dependency as an endogenous state variable yields an impossibility theorem (hysteresis requires dynamics beyond static frameworks) and a channel necessity theorem (each channel is individually necessary). Empirical validation uses the complete universe of SEC Form 13F filings (99.5 million holdings, 10,957 institutional managers, 2013--2024) with a Bartik shift-share instrument (first-stage

F = 22.7

). The model implies tail-loss amplification of 18--54%, economically significant relative to Basel III countercyclical buffers.

Abstract

Finance

Explore similar work

Artificial Intelligence and Systemic Risk: A Unified Model of Performative Prediction, Algorithmic Herding, and Cognitive Dependency in Financial Markets

AI Trading's Alpha Singularity: Emergent Market Reasoning through Agent-to-Agent Self-Evolution

Representation Homogeneity and Systemic Instability in AI-Dominated Financial Markets: A Structural Approach