pith. machine review for the scientific record. sign in

arxiv: 2604.03272 · v1 · submitted 2026-03-23 · 💱 q-fin.CP · cs.AI· cs.GT· q-fin.GN

Recognition: 2 theorem links

· Lean Theorem

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

Shuchen Meng , Xupeng Chen
Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:11 UTC · model grok-4.3

classification 💱 q-fin.CP cs.AIcs.GTq-fin.GN
keywords AI adoptionsystemic riskperformative predictionalgorithmic herdingcognitive dependencyfinancial marketstail riskrisk multiplier
0
0 comments X

The pith

AI adoption in financial markets creates a convex systemic risk coupling that grows the risk multiplier superlinearly with penetration.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a unified model in which AI adoption generates systemic risk through three reinforcing channels: performative prediction where forecasts alter the outcomes they predict, algorithmic herding from correlated signals, and cognitive dependency where human decisions increasingly rely on AI outputs. Within an extended rational expectations setup with endogenous adoption, the model derives an equilibrium risk coupling r(φ) = φρβ/λ'(φ) that becomes convex because effective price impact λ'(φ) declines with adoption share φ. This convexity makes the systemic risk multiplier M = (1-r)^{-1} grow faster than adoption itself. The framework also produces a saddle-node bifurcation into algorithmic monoculture and necessity theorems showing each channel is required. Empirical checks on 99.5 million institutional holdings support tail-loss amplification of 18-54 percent.

Core claim

Within an extended rational expectations framework with endogenous adoption, the model derives 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: endogenous fragility with market depth decreasing and convex in AI adoption, embedding the convex coupling within a supermodular adoption game that produces a saddle-node

What carries the argument

The equilibrium systemic risk coupling r(φ) = φρβ/λ'(φ) with λ'(φ) decreasing in AI adoption share φ, which produces convexity and superlinear growth in the systemic risk multiplier while unifying performative prediction, algorithmic herding, and cognitive dependency.

If this is right

  • Market depth decreases and becomes more convex as AI adoption rises.
  • The supermodular adoption game produces a saddle-node bifurcation into an algorithmic monoculture.
  • Cognitive dependency as an endogenous state variable creates hysteresis that requires dynamics beyond static frameworks.
  • Each of the three channels is individually necessary for the full systemic risk effect.
  • Tail losses are amplified by 18 to 54 percent relative to Basel III buffers.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Regulators could monitor cross-AI signal correlations as an early indicator of rising risk coupling.
  • The same convex mechanism may appear in other domains where AI adoption is high, such as automated insurance pricing or crypto trading platforms.
  • Granular trade-by-trade data could be used to measure the exact functional form of λ'(φ) and test the rate of superlinear growth.
  • If adoption continues, static capital requirements may need to incorporate a dynamic term tied to measured AI penetration.

Load-bearing premise

That the endogenous effective price impact declines as the share of AI adopters in the market increases.

What would settle it

Observing that market depth or illiquidity does not decrease convexly with measured AI adoption share in post-2024 holdings data would falsify the convexity of the risk coupling.

Figures

Figures reproduced from arXiv: 2604.03272 by Shuchen Meng, Xupeng Chen.

Figure 1
Figure 1. Figure 1: Conceptual framework: Three mutually reinforcing risk channels. The systemic [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Systemic risk multiplier M(ϕ, ρ, β) as a function of ϕ for varying ρ and β, exhibiting superlinear growth near the stability boundary. 0.0 0.2 0.4 0.6 0.8 1.0 AI adoption rate Á 0.0 0.2 0.4 0.6 0.8 1.0 Systemic risk level Á ¤ Á ¤ ¤ Diversified Equilibrium Monoculture Equilibrium Hysteresis gap Hysteresis region Forward path (Á" ) Backward path (Á# ) [PITH_FULL_IMAGE:figures/full_fig_p034_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Hysteresis in the adoption–risk relationship. The gap between the tipping point [PITH_FULL_IMAGE:figures/full_fig_p034_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Synthesis of empirical findings. (a) Portfolio convergence increases +12% post-2016 [PITH_FULL_IMAGE:figures/full_fig_p040_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Tail risk amplification. (a) Heatmap showing superlinear growth. (b) The “calm [PITH_FULL_IMAGE:figures/full_fig_p047_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Regulatory intervention comparison: baseline vs. diversity requirements, speed [PITH_FULL_IMAGE:figures/full_fig_p049_6.png] view at source ↗
read the original 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 $r(\phi) = \phi\rho\beta/\lambda'(\phi)$, where $\phi$ is the AI adoption share, $\rho$ the algorithmic signal correlation, $\beta$ the performative feedback intensity, and $\lambda'(\phi)$ the endogenous effective price impact. Because $\lambda'(\phi)$ is decreasing in $\phi$, 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.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper develops a unified model of systemic risk in financial markets arising from AI adoption via performative prediction, algorithmic herding, and cognitive dependency. In an extended rational expectations framework, it derives the equilibrium systemic risk coupling r(φ) = φρβ/λ'(φ), where λ'(φ) is decreasing in the AI adoption share φ, implying a convex coupling and superlinear growth of the systemic risk multiplier M = (1 - r)^{-1}. This leads to a saddle-node bifurcation into algorithmic monoculture and an impossibility theorem for hysteresis. Empirical validation on SEC 13F data (99.5M holdings) with a Bartik instrument (F=22.7) suggests tail-loss amplification of 18-54%.

Significance. If the central assumption on the sign of dλ'/dφ holds and is endogenously derived, the paper offers a novel theoretical unification of AI-related risk channels with quantitative implications for tail risks that exceed standard regulatory buffers, supported by large-scale institutional holdings data. The model highlights potential for hysteresis and the necessity of all three channels.

major comments (1)
  1. [Derivation of r(φ) and endogenous fragility layer] The functional form r(φ) = φρβ/λ'(φ) and the assumption that λ'(φ) is decreasing in φ are invoked to establish convexity of r(φ) and superlinear growth of M. However, no derivation is provided showing that this negative sign emerges from the performative prediction, herding, or cognitive dependency channels; the paper states that market depth is decreasing and convex in φ, which would typically imply increasing price impact. This assumption is load-bearing for the 18-54% amplification claim and the bifurcation results.
minor comments (1)
  1. [Empirical section] The Bartik instrument is reported with first-stage F=22.7, but additional details on robustness checks, exact mapping to the amplification range, and potential violations of exclusion restrictions would strengthen the empirical support.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful reading and constructive comments, which help clarify the foundations of our model. We agree that an explicit derivation of the sign of dλ'/dφ is necessary to support the convexity, bifurcation, and amplification results, and we will incorporate this in the revision.

read point-by-point responses

  1. Referee: [Derivation of r(φ) and endogenous fragility layer] The functional form r(φ) = φρβ/λ'(φ) and the assumption that λ'(φ) is decreasing in φ are invoked to establish convexity of r(φ) and superlinear growth of M. However, no derivation is provided showing that this negative sign emerges from the performative prediction, herding, or cognitive dependency channels; the paper states that market depth is decreasing and convex in φ, which would typically imply increasing price impact. This assumption is load-bearing for the 18-54% amplification claim and the bifurcation results.

    Authors: We agree that the manuscript would benefit from an explicit derivation. In the revised version we will add a new lemma in Section 2.1 deriving dλ'/dφ < 0 directly from the three channels. Performative prediction raises ρ while generating self-reinforcing price distortions that reduce effective depth; algorithmic herding increases β and produces convex reductions in depth via coordination externalities; cognitive dependency endogenizes the state variable, further lowering λ'(φ). We clarify that λ'(φ) is defined as effective market depth (the inverse of price impact), so its decrease is exactly consistent with the stated reduction in market depth and the implied rise in price impact. The revised text will include the step-by-step mapping from each channel to the functional form of λ'(φ), preserving the convexity of r(φ) and the saddle-node bifurcation. This change will be made without altering the empirical estimates or core conclusions. revision: yes

Circularity Check

1 steps flagged

Convexity of r(φ) and superlinear M growth hinge on asserted (not derived) claim that λ'(φ) decreases with φ

specific steps
  1. self definitional [Abstract]
    "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 expression for r(φ) is written in terms of λ'(φ); the claimed convexity and superlinear M follow immediately once the sign of dλ'/dφ is asserted to be negative. The paper supplies no endogenous derivation of this sign from the three channels, so the central systemic-risk result is obtained by construction of the assumed functional form rather than from an independent step in the derivation chain.

full rationale

The paper presents r(φ) = φρβ/λ'(φ) as a derived equilibrium coupling within the rational expectations framework. However, the load-bearing property that λ'(φ) is decreasing in φ is invoked directly to establish convexity of r, superlinear growth of M = (1-r)^{-1}, and the saddle-node bifurcation. No derivation of the negative sign dλ'/dφ is supplied from the performative prediction, herding, or cognitive dependency channels; the functional behavior is asserted as part of the endogenous fragility layer. Consequently the headline implications (18-54% tail-loss amplification, impossibility of hysteresis in static frameworks) reduce to the modeling choice rather than emerging independently from the primitives.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 0 invented entities

The central claim rests on an extended rational expectations framework, the assumption that market depth decreases convexly with AI adoption, and several parameters whose values are introduced to capture the three channels rather than derived from first principles.

free parameters (3)
  • ρ
    Algorithmic signal correlation parameter appearing in the risk coupling formula.
  • β
    Performative feedback intensity parameter appearing in the risk coupling formula.
  • λ'(φ)
    Endogenous effective price impact, modeled as decreasing in adoption share φ to produce convexity.
axioms (2)
  • domain assumption Extended rational expectations framework with endogenous adoption
    Invoked to derive the equilibrium systemic risk coupling r(φ).
  • ad hoc to paper Market depth is decreasing and convex in AI adoption
    Stated as the first layer of endogenous fragility that drives the convexity of r(φ).

pith-pipeline@v0.9.0 · 5598 in / 1701 out tokens · 102558 ms · 2026-05-15T01:11:41.429976+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    Alternative Investment Management Association

    doi: 10.1093/rfs/1.1.3. Alternative Investment Management Association. AI in the hedge fund industry. Report, Alternative Investment Management Association, London, UK,

    work page doi:10.1093/rfs/1.1.3

  2. [2]

    Winfred Arthur, Winston Bennett, Pamela L

    doi: 10.1093/qje/ qjab032. Winfred Arthur, Winston Bennett, Pamela L. Stanush, and Terrence L. McNelly. Factors that influence skill decay and retention: A quantitative review and analysis.Human Performance, 11(1):57–101,

    work page doi:10.1093/qje/

  3. [3]

    doi: 10.1207/s15327043hup1101

    work page doi:10.1207/s15327043hup1101

  4. [4]

    Timothy J

    URLhttps://www.bankofengland.co.uk/report/2019/ machine-learning-in-uk-financial-services. Timothy J. Bartik.Who Benefits from State and Local Economic Development Policies? W.E. Upjohn Institute for Employment Research, Kalamazoo, MI,

    work page 2019

  5. [5]

    Philip Bond, Alex Edmans, and Itay Goldstein

    doi: 10.1016/j.jacceco.2025.101782. Philip Bond, Alex Edmans, and Itay Goldstein. The real effects of financial mar- kets.Annual Review of Financial Economics, 4:339–360,

    work page doi:10.1016/j.jacceco.2025.101782 2025

  6. [6]

    doi: 10.1016/j.techsoc.2021. 101852. Kirill Borusyak, Peter Hull, and Xavier Jaravel. Quasi-experimental shift-share research designs.Review of Economic Studies, 89(1):181–213,

    work page doi:10.1016/j.techsoc.2021 2021

  7. [7]

    doi: 10.1093/restud/rdab030. Mark T. Bradshaw, Chenyang Ma, Benjamin Yost, and Yuan Zou. Generative AI use by cap- ital market information intermediaries: Evidence from Seeking Alpha. Harvard Business School Working Paper No. 25-055,

    work page doi:10.1093/restud/rdab030

  8. [8]

    Markus K

    doi: 10.1093/rfs/ hhu032. Markus K. Brunnermeier. Deciphering the liquidity and credit crunch 2007–2008.Journal of Economic Perspectives, 23(1):77–100,

    work page doi:10.1093/rfs/ 2007

  9. [9]

    Stephen M

    doi: 10.1257/jep.23.1.77. Stephen M. Casner, Richard W. Geven, Matthias P. Recker, and Jonathan W. Schooler. The retention of manual flying skills in the automated cockpit.Human Factors, 56(8): 1506–1516,

    work page doi:10.1257/jep.23.1.77

  10. [10]

    doi: 10.1177/0018720814535628. Alain P. Chaboud, Benjamin Chiquoine, Erik Hjalmarsson, and Clara Vega. Rise of the 54 machines: Algorithmic trading in the foreign exchange market.Journal of Finance, 69(5): 2045–2084,

    work page doi:10.1177/0018720814535628 2045

  11. [11]

    Rama Cont

    doi: 10.1111/jofi.12186. Rama Cont. Volatility clustering in financial markets. InRisk Management in Volatile Financial Markets. Springer,

    work page doi:10.1111/jofi.12186

  12. [12]

    J´ on Danielsson, Robert Macrae, and Andreas Uthemann

    doi: 10.1016/j.jfineco.2006.09.007. J´ on Danielsson, Robert Macrae, and Andreas Uthemann. Artificial intelligence and systemic risk.Journal of Banking and Finance, 140:106290,

    work page doi:10.1016/j.jfineco.2006.09.007 2006

  13. [13]

    doi: 10.1016/j.jbankfin.2021. 106290. Hermann Ebbinghaus.Memory: A Contribution to Experimental Psychology. Teachers College, Columbia University, New York,

    work page doi:10.1016/j.jbankfin.2021 2021

  14. [14]

    Financial Stability Board

    doi: 10.1038/460685a. Financial Stability Board. Artificial intelligence in finance: Financial stability implications. Report, Financial Stability Board, Basel, Switzerland,

    work page doi:10.1038/460685a

  15. [15]

    org/2024/11/artificial-intelligence-in-finance/

    URLhttps://www.fsb. org/2024/11/artificial-intelligence-in-finance/. F. Douglas Foster and S. Viswanathan. Strategic trading when agents forecast the forecasts of others.Journal of Finance, 51(4):1437–1478,

    work page 2024

  16. [16]

    tb04075.x

    doi: 10.1111/j.1540-6261.1996. tb04075.x. Paul Goldsmith-Pinkham, Isaac Sorkin, and Henry Swift. Bartik instruments: What, when, 55 why, and how.American Economic Review, 110(8):2586–2624,

    work page doi:10.1111/j.1540-6261.1996 1996

  17. [17]

    20181047

    doi: 10.1257/aer. 20181047. Itay Goldstein and Alexander Guembel. Manipulation and the allocational role of prices. Review of Economic Studies, 75(1):133–164,

    work page doi:10.1257/aer

  18. [18]

    Jillian Grennan and Roni Michaely

    doi: 10.1111/j.1467-937X.2007.00467.x. Jillian Grennan and Roni Michaely. FinTechs and the market for financial analysis. Journal of Financial and Quantitative Analysis, 56(6):1877–1907,

    work page doi:10.1111/j.1467-937x.2007.00467.x 2007

  19. [19]

    Moritz Hardt and Celestine Mendler-D¨ unner

    doi: 10.1038/nature09659. Moritz Hardt and Celestine Mendler-D¨ unner. Performative prediction: Past and future. Statistical Science, 40(3):417–436,

    work page doi:10.1038/nature09659

  20. [20]

    Terrence Hendershott, Charles M

    doi: 10.1111/j.1540-6261.1991.tb03749.x. Terrence Hendershott, Charles M. Jones, and Albert J. Menkveld. Does algorithmic trading improve liquidity?Journal of Finance, 66(1):1–33,

    work page doi:10.1111/j.1540-6261.1991.tb03749.x 1991

  21. [21]

    doi: 10.1111/j.1540-6261.2010. 01624.x. International Monetary Fund. Global financial stability report: AI and financial stability. Technical report, International Monetary Fund, Washington, DC,

    work page doi:10.1111/j.1540-6261.2010 2010

  22. [22]

    URLhttps: //www.imf.org/en/Publications/GFSR. Amir E. Khandani and Andrew W. Lo. What happened to the quants in August 2007? Evidence from factors and transactions data.Journal of Financial Economics, 98(2): 171–222,

    work page 2007

  23. [23]

    Andrei Kirilenko, Albert S

    doi: 10.1016/j.jfineco.2010.08.003. Andrei Kirilenko, Albert S. Kyle, Mehrdad Samadi, and Tugkan Tuzun. The flash crash: 56 High-frequency trading in an electronic market.Journal of Finance, 72(3):967–998,

    work page doi:10.1016/j.jfineco.2010.08.003 2010

  24. [24]

    Albert S

    doi: 10.1111/jofi.12498. Albert S. Kyle. Continuous auctions and insider trading.Econometrica, 53(6):1315–1335,

    work page doi:10.1111/jofi.12498

  25. [25]

    Donald MacKenzie.An Engine, Not a Camera: How Financial Models Shape Markets

    doi: 10.2307/1913210. Donald MacKenzie.An Engine, Not a Camera: How Financial Models Shape Markets. MIT Press, Cambridge, MA,

    work page doi:10.2307/1913210

  26. [26]

    doi: 10.7551/mitpress/1967.001

    ISBN 978-0-262-13460-2. doi: 10.7551/mitpress/1967.001

    work page doi:10.7551/mitpress/1967.001 1967

  27. [27]

    Jos´ e Luis Montiel Olea and Carolin Pflueger

    doi: 10.1038/238413a0. Jos´ e Luis Montiel Olea and Carolin Pflueger. A robust test for weak instruments.Journal of Business and Economic Statistics, 31(3):358–369,

    work page doi:10.1038/238413a0

  28. [28]

    doi: 10.1080/07350015.2013. 803869. Alan Moreira and Tyler Muir. Volatility-managed portfolios.Journal of Finance, 72(4): 1611–1644,

    work page doi:10.1080/07350015.2013 2013

  29. [29]

    Raja Parasuraman and Dietrich H

    doi: 10.1111/jofi.12513. Raja Parasuraman and Dietrich H. Manzey. Complacency and bias in human use of automation: An attentional integration.Human Factors, 52(3):381–410,

    work page doi:10.1111/jofi.12513

  30. [30]

    doi: 10.1177/0018720810376055. Juan C. Perdomo, Tijana Zrnic, Celestine Mendler-D¨ unner, and Moritz Hardt. Performa- tive prediction. InProceedings of the 37th International Conference on Machine Learn- ing (ICML), volume 119 ofProceedings of Machine Learning Research, pages 7599–7609. PMLR,

    work page doi:10.1177/0018720810376055

  31. [31]

    Stefan Thurner, J

    doi: 10.1006/ijhc.1999.0252. Stefan Thurner, J. Doyne Farmer, and John Geanakoplos. Leverage causes fat tails and clustered volatility.Quantitative Finance, 12(5):695–707,

    work page doi:10.1006/ijhc.1999.0252 1999

  32. [32]

    2012.674301

    doi: 10.1080/14697688. 2012.674301. Xavier Vives. Complementarities and games: New developments and future directions. Journal of Economic Literature, 43:437–479,

    work page doi:10.1080/14697688 2012

  33. [33]

    doi: 10.1257/0022051054661558. Proofs, simulation implementation details, additional empirical results, detailed experiment designs, robustness checks, and supplementary figures are provided in the Online Appendix. 58

    work page doi:10.1257/0022051054661558