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

Shuchen Meng; Xupeng Chen

q-fin.CPMar 23, 2026

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

Figures & tables

Figure 1: Conceptual framework: Three mutually reinforcing risk channels. The systemic risk multiplier

\mathcal{M}=(1-\phi\rho\beta/\lambda^{\prime})^{-1}

is superlinear in the product of channel intensities. The bottom panel illustrates the diversified equilibrium (left) and monoculture equilibrium (right), with hysteresis barriers creating path-dependent transitions.

Symbol	Description	Interpretation	Range
$\phi$	AI adoption rate	Fraction of institutions using AI	$[0,1]$
$\rho$	Algorithmic homogeneity	Correlation of AI signals	$[0,1]$
$\beta$	Performative feedback	Price-to-fundamental feedback intensity	$[0,1)$
$d_{i}(t)$	Cognitive dependency	Weight on AI vs. human judgment	$[0,1]$
$\kappa$	Skill atrophy rate	Speed of human skill degradation	$>0$
$\gamma$	Peer pressure	Conformity incentive in adoption	$>0$

Table 1: Summary of Model Parameters

	Model Variant
Equilibrium Property	Kyle	$+\;\rho>0$	$+\;\beta>0$	$+\;\kappa>0$
	(baseline)	(herding)	(performative)	(dependency)
Endogenous $\lambda(\phi)$	—	✓	✓	✓
Excess volatility	—	additive	amplified	amplified
Coordination failure	—	possible a	✓	✓
Multiplier $\mathcal{M}>1$	—	—	✓	✓

Table 2: Model Hierarchy: What Each Channel Enables

Figure 2: Systemic risk multiplier

\mathcal{M}(\phi,\rho,\beta)

as a function of

\phi

for varying

\rho

and

\beta

, exhibiting superlinear growth near the stability boundary.

Figure 3: Hysteresis in the adoption–risk relationship. The gap between the tipping point

\phi^{*}

and recovery point

\phi^{**}

widens with the atrophy rate

\kappa

and duration of the monoculture phase.

	Pre-AI (2013–2016)		Post-AI (2017–2022)		Diff.	$t$ -stat
Measure	Mean	SD	Mean	SD
Cosine Similarity	0.089	0.007	0.100	0.013	$+0.011^{***}$	3.43
Portfolio HHI	0.066	0.004	0.076	0.005	$+0.011^{***}$	7.13
Top-10 Overlap	0.377	0.124	0.519	0.058	$+0.142^{***}$	4.30

Table 3: Portfolio Convergence: Pre-AI vs. Post-AI Periods (Real 13F Data)

Panel of Figure 4 — Figure 4: Synthesis of empirical findings. (a) Portfolio convergence increases $+12\%$ post-2016 (real 13F data, $p=0.002$ ). (b) Cumulative abnormal returns during algorithmic amplification episodes. (c) AI-related language in SEC filings shows $>$ 50 $\times$ growth (EDGAR EFTS, 2013–2024). (d) Cross-sectional relationship between AI adoption intensity and portfolio convergence.

	LOO Bartik	Supply-Side	Overidentified	Top-10 Overlap
	(Cos. Sim.)	(NVIDIA Rev.)	(Both IVs)
$Z_{i,t}^{LOO}$	$3.74^{***}$	—	$3.21^{***}$	$24.16^{***}$
	$(0.78)$	—	$(0.81)$	$(9.02)$
$Z_{i,t}^{supply}$	—	$2.89^{***}$	$1.47^{**}$	—
	—	$(0.77)$	$(0.72)$	—
First-stage $F$	$22.7^{***}$	$14.2^{***}$	$16.8^{***}$	$7.2^{***}$

Table 4: Bartik Instrumental Variable: First-Stage Results (Real 13F Data)

Panel of Figure 5 — Figure 5: Tail risk amplification. (a) Heatmap showing superlinear growth. (b) The “calm before the storm” paradox: lower unconditional volatility coexists with higher conditional tail losses.

Figure 6: Regulatory intervention comparison: baseline vs. diversity requirements, speed bumps, human-in-the-loop mandates, and the combined regime.

Result	Key Prediction	Empirical Test	ABM Consistency
P1: Monoculture Trap	AI–AI pair convergence $>$ baseline	13F cosine sim. $+12\%$ ( $t=3.43$ , § 4.1 )	$\phi_{T}\to 0.405$ from $\phi_{0}=0.1$ (Exp. 1)
P2: Tail Risk	$\mathcal{M}$ superlinear in $\phi\rho\beta$	Event study CAR–concentration gradient (§ 4.2 )	2.9 $\times$ drawdown amplification across $(\phi,\rho)$ grid (Exp. 2)
P3: Irreversibility	Hysteresis gap $\phi^{}-\phi^{*}>0$	NLP dependency ratio ratchet (§ 4.3 )	$\sigma_{H}(T)/\sigma_{H}(0)\approx 6\times$ (Exp. 1)
P4: Calm Before Storm	Low $\sigma_{unc}$ , high $\sigma_{tail}$	VPIN escalation across event eras (§ 4.2 )	Conditional tail CTE 95 divergence (Exp. 2)
Bifurcation (Prop. 3.11 )	Discontinuous $\phi$ jump at $c^{*}$	Dose-response: strict $>$ broad adopters (§ 4.4 )	S-curve adoption trajectory (Exp. 1)
Impossibility (Thm. 3.15 )	No recovery without skill restoration	Asymmetric ADV language shifts pre/post-crisis (§ 4.3 )	$c^{*}<c^{}$ in reversal experiments

Table 5: Structural Mapping: Theoretical Results to Empirical and Simulation Tests

Performative Prediction

Performative prediction studies how deployed predictions change the data and behavior they are meant to predict.

See all 18 papers

AI-Driven Alpha Decay: Algorithmic Homogenization, Reflexive Signal Erosion, and the Paradox of Intelligent Markets
Mar 23, 2026Shuchen Meng, Xupeng ChenFinance Algorithmic Trading
Department of Financial Engineering, New York University. · Department of Electrical and Computer Engineering, New York University.
Cognitive Debt: AI as Intellectual Leverage and the Dynamics of Systemic Fragility
Jun 13, 2026Shuchen MengAI Risk Management Cognitive Offloading
New York University
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

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

Shuchen Meng, Xupeng Chen

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

Finance Algorithmic Trading

Open paper AskPDF arXiv

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.

Abstract

Performative Prediction

Explore similar work

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

Cognitive Debt: AI as Intellectual Leverage and the Dynamics of Systemic Fragility

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