Chaos in Autobidding Auctions

Ioannis Anagnostides, Ian Gemp, Ryann Sim, Kelly Spendlove

Abstract: As autobidding systems increasingly dominate online advertising auctions, characterizing their long-term dynamical behavior is brought to the fore. In this paper, we examine the dynamics of autobidders who optimize value subject to a return-on-spend (RoS) constraint under uniform bid scaling. Our main set of results show that simple autobidding dynamics can exhibit formally chaotic behavior. This significantly strengthens recent results that went as far as quasiperiodicity. Our proof proceeds by establishing that autobidding dynamics can simulate—up to an arbitrarily small error—a broad class of continuous-time nonlinear dynamical systems. This class contains as a special case Chua's circuit, a classic chaotic system renowned for its iconic double scroll attractor. Our reduction develops several modular gadgets, which we anticipate will find other applications going forward. Moreover, in discrete time, we show that different incarnations of mirror descent can exhibit Li-Yorke chaos, topological transitivity, and sensitivity to initial conditions, connecting along the way those dynamics to classic dynamical systems such as the logistic map and the Ricker population model. Taken together, our results reveal that the long-term behavior of ostensibly simple second-price autobidding auctions can be inherently unpredictable and complex.

Introduction

In the traditional theory of auctions, bidders are assumed to be human agents manually maximizing profit. Today, however, the ecosystem is dominated by autobidders—AI agents that optimize bids across millions of auctions to achieve high-level goals, such as a specific Return-on-Spend (RoS). While recent research discovered that these systems do not always settle into a stable equilibrium, a major question remained: just how complex can this instability get?

This paper proves that automated bidding systems can exhibit formal chaos. If these systems are chaotic, long-term market forecasting becomes impossible because even the tiniest "noise"—such as a floating-point rounding error—can lead to radically different market states over time.

Methodology: Simulating Chaos via Reductions

The authors employ a reduction strategy to prove that autobidding dynamics are powerful enough to simulate well-known chaotic systems. Instead of analyzing every possible auction configuration, they developed a series of mathematical "gadgets"—specific configurations of items, values, and reserve prices—to implement dynamical building blocks.

The reduction framework uses modular gadgets to simulate complex target systems within an autobidding environment.

Key components of this reduction include:

  • Continuous Negation: A gadget that allows the system to simulate negative feedback (competition) without distorting the underlying mathematics.
  • Nonlinear Function Gadget: By using a continuum of items with carefully chosen reserve prices, the authors show an auction can recreate almost any nonlinear mathematical function.
  • Chua’s Circuit Simulation: Using these gadgets, the authors demonstrate that autobidding can $\epsilon$-simulate Chua’s circuit, a famous electronic system known for its "double scroll" chaotic attractor.

In discrete-time settings, the authors analyzed Mirror Descent, a standard optimization algorithm. They proved that these dynamics exactly recover the Ricker population model and the logistic map, both of which are textbook examples of mathematical chaos.

Key Insights and Results

The paper establishes that chaos is an intrinsic property of the continuous-time math governing RoS-constrained bidding, rather than a result of poor programming.

Numerical simulation of the autobidding-Chua system, displaying the characteristic double-scroll chaotic attractor in 3D and 2D projections.

Quantitative Evidence of Chaos

  • Positive Lyapunov Exponent: Simulations of the autobidding-Chua system yielded a largest Lyapunov exponent of $\approx 0.14214$. Any value above zero is a definitive signature of chaos.
  • The "Period Three" Smoking Gun: In discrete time, the authors proved the existence of "Period 3" orbits. In dynamical systems theory, "Period Three Implies Chaos," providing a rigorous mathematical guarantee of complexity.
  • Sensitivity to Initial Conditions: The authors formally established "Devaney chaos," meaning the system is topologically transitive and has a dense set of periodic points.

Market Implications

The research shows that even the Vickrey-Clarke-Groves (VCG) auction—often considered the "gold standard" for efficiency—can become inefficient and chaotic when bidders have RoS constraints. In chaotic regimes, both the total welfare created and the revenue earned by the auctioneer can become highly suboptimal and volatile.

Conclusion and Future Work

While the current reduction method is robust, it cannot yet simulate systems with "coupled" nonlinearities (where variables interact via cross-terms). Future research may extend these gadgets to handle such complexities. Furthermore, since long-term prediction of specific states is impossible in chaotic systems, the authors suggest that future market analysis should focus on statistical or ergodic properties—predicting the average behavior of the market rather than its exact path.

Summary updated

Abstract

As autobidding systems increasingly dominate online advertising auctions, characterizing their long-term dynamical behavior is brought to the fore. In this paper, we examine the dynamics of autobidders who optimize value subject to a return-on-spend (RoS) constraint under uniform bid scaling. Our main set of results show that simple autobidding dynamics can exhibit formally chaotic behavior. This significantly strengthens the recent results of Leme, Piliouras, Schneider, Spendlove, and Zuo (EC '24) that went as far as quasiperiodicity. Our proof proceeds by establishing that autobidding dynamics can simulate -- up to an arbitrarily small error -- a broad class of continuous-time nonlinear dynamical systems. This class contains as a special case Chua's circuit, a classic chaotic system renowned for its iconic double scroll attractor. Our reduction develops several modular gadgets, which we anticipate will find other applications going forward. Moreover, in discrete time, we show that different incarnations of mirror descent can exhibit Li-Yorke chaos, topological transitivity, and sensitivity to initial conditions, connecting along the way those dynamics to classic dynamical systems such as the logistic map and the Ricker population model. Taken together, our results reveal that the long-term behavior of ostensibly simple second-price autobidding auctions can be inherently unpredictable and complex.
BibTeX
@misc{anagnostides2026chaos,
  author = {Ioannis Anagnostides and Ian Gemp and Ryann Sim and Kelly Spendlove},
  title = {{Chaos in Autobidding Auctions}},
  year = {2026},
  eprint = {2602.09118},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2602.09118},
}
Figure