Well-posedness theory, microstructure-faithful extensions, and empirical validation of the mean-field-game model of price formation with high-frequency participants

The 2016 model of Lachapelle, Lasry, Lehalle, and Lions describes price formation in an electronic limit order book as a mean field game (MFG): the state is the pair of sizes of the coupled bid and ask liquidity queues, and a continuum of anonymous agents — institutional investors and high-frequency traders (HFTs), differing in arrival rate and order size — repeatedly choose between joining a queue (supplying liquidity while paying a waiting cost, matched pro-rata) and paying an impact-driven premium to consume the opposite queue immediately. The paper derived the liquidity-provision/consumption decision boundaries analytically only at first order, treated the second-order PDE system numerically without any existence, uniqueness, or viscosity-solution theory, and obtained its headline conclusion — that HFTs remove inefficient liquidity imbalances and shrink the aggregate effective spread while redistributing trading costs to institutional investors — numerically in the single regime $c_1 q_1 = c_2 q_2$, under deliberately stylized assumptions (infinite order-book resiliency, near-linear market impact, constant latent fair price, exogenous Poisson arrivals, pro-rata matching). The open problem is to turn this framework into a well-founded and tested theory: (1) rigorous existence, uniqueness, and viscosity-solution analysis of the second-order equilibrium system; (2) extensions of the same framework to finite resiliency, nonlinear impact, non-constant fair prices, FIFO or size-priority matching, and endogenous arrival rates; (3) robustness of the efficiency and cost-redistribution conclusions across parameter regimes; and (4) empirical validation of the predicted invariant measures, spread dynamics, and cost redistribution against high-frequency order-book data. Later work is adjacent — market-clearing price-formation MFGs, viscosity theory for other limit-order-book equilibria, an empirical MFG price-formation model, AMM liquidity-pool MFGs, and non-mean-field HFT interaction models — but does not analyze, extend, or empirically test this specific model; the problem remains open.

Background

Electronic financial markets match buyers and sellers through a limit order book (LOB): a continuously updated public list of pending limit orders, each specifying a side (buy/bid or sell/ask), a price, and a quantity. Patient traders posting limit orders supply liquidity by joining a queue at their chosen price; impatient traders consume liquidity by sending market orders that cross the spread and eat into the opposite queue. The bid-ask spread — the gap between the best ask and best bid — together with the depths of the queues at the best quotes, is the canonical measure of liquidity and of the efficiency of the price formation process (the mechanism by which transaction prices emerge from the order flow). The effective spread — the trading cost actually paid by an executing trader, as opposed to the quoted spread posted at a given instant — is the relevant measure of realized trading costs. Market impact is the temporary price displacement caused by consuming a quantity of an available queue, and order-book resiliency is the speed at which the book's shape refills after such consumption; infinite resiliency means the shape readjusts instantly after each trade. Exchanges differ in the priority rule used to match an incoming order against the queue of resting orders at a given price: time priority or FIFO (first in, first out) fills the oldest orders first, size priority fills the largest first, and pro-rata fills each resting order proportionally to its share of the total queue size; pro-rata is used, for example, in some one-tick futures markets. Structural models of the LOB, such as Roşu (2009)'s dynamic model with FIFO priority, study queueing equilibria without a continuum of strategic players.

A long-standing debate concerns high-frequency traders (HFTs) — automated agents sending many small orders with very short horizons — trading alongside institutional investors who execute large orders slowly. Do HFTs, acting as fast liquidity suppliers, improve price formation, or do they extract rents from slower participants? Aggregate metrics such as the quoted spread can improve even while the slow traders' realized costs worsen, which motivated Lachapelle, Lasry, Lehalle, and Lions (2016) to build a strategic model in which such redistribution effects arise endogenously. Their framework is a mean field game (MFG), the class of differential games with a continuum of anonymous players introduced by Lasry and Lions (2007): each agent optimizes while taking the population's state distribution as given, and consistency between optimal behavior and the distribution yields a coupled system of nonlinear partial differential equations (PDEs) — backward Hamilton-Jacobi-Bellman (HJB) equations for the agents' value functions (their expected optimized payoffs) coupled to forward Kolmogorov equations for the distribution of states, whose stationary limit is the invariant measure (the long-run distribution of the state process).

In the model, the state is the pair of queue sizes $(Q_a, Q_b)$ at the ask and the bid. Agents arrive with Poisson rates, carry an order size $q$ and a waiting-cost rate $c$ per share held in a queue, and at each decision epoch choose between joining the queue on their side (supply liquidity, earn the queue price, pay the waiting cost, and be matched pro-rata) or paying an impact-driven premium to consume the opposite queue immediately. Two populations are considered: institutional investors (large $q$, low arrival rate) and HFTs (small $q$, high arrival rate). The model is deliberately stylized — infinite resiliency, near-linear market impact, a constant latent fair price (the unobserved fundamental value around which transaction prices fluctuate, treated as exogenous and fixed in time), exogenous Poisson arrivals, and pro-rata matching chosen, in the authors' words, because it keeps the dimensionality of the model tractable.

The paper's results are of three kinds. First, in a small-parameter expansion, the equilibrium decision boundaries separating liquidity-provision from liquidity-consumption regions in the $(Q_a, Q_b)$ plane are obtained analytically at first order. Second, the full second-order PDE system is written down and solved only numerically; the authors observe that the second-order terms select a particular solution among the multiple first-order solutions, state that this 'has to be linked to the notion of viscosity solutions' — the standard weak-solution concept for fully nonlinear degenerate PDEs — and explicitly decline the theory ('we do not enter in the details here'). Third, the numerics show that a market with institutional investors alone develops persistent liquidity imbalances (situations the authors call liquidity calls for liquidity, in which the transaction price moves significantly away from the fair price), while adding HFTs removes these imbalances and shrinks the global effective spread by roughly a third — yet the spread effectively paid by institutional investors increases by about twenty percent, so the gains accrue to the HFTs. The authors flag two caveats: the two-population study is performed in the regime $c_1 q_1 = c_2 q_2$ (chosen so that the two populations' value functions coincide), and the conclusions are conditioned on the accuracy of the stylized model. The only empirical contact is a qualitative comparison of the computed invariant measure with the queue-dynamics measurements of Garèche, Disdier, Kockelkoren, and Bouchaud (2013), and the paper closes by underlining the lack of liquidity measurements adapted to current market microstructure.

Problem Statement

Develop the mean-field-game analysis of the price formation process in the presence of high-frequency participants beyond the stylized first-order setting of Lachapelle, Lasry, Lehalle, and Lions (2016), with one common scientific objective: determine, within a mathematically well-founded mean-field-game model of an order-driven market with heterogeneous participants, whether the finding that HFT presence improves aggregate price-formation efficiency while redistributing trading costs to slower institutional investors is a robust property of strategic order-book equilibria or an artifact of the stylized assumptions. The problem comprises four linked open directions, each traceable to a limitation the source paper itself acknowledges; they may be pursued independently.

(1) Rigorous theory of the equilibrium system. Establish existence, uniqueness, and viscosity-solution theory for the second-order coupled Hamilton-Jacobi-Bellman and Kolmogorov system of the model as displayed in the source paper (Section 5.3 and Appendix A.3), including the selection role of the second-order terms among the multiple first-order solutions and the characterization of the stationary invariant measure of the two-queue process.

(2) Microstructure-faithful extensions of the same framework — relaxing the stylized assumptions: relax infinite order-book resiliency to finite book-reshaping timescales; allow non-linear market-impact functions; let the latent fair price vary (for example, diffusively); replace pro-rata matching with FIFO or size-priority priority rules in the full coupled two-queue, two-population model (the paper's one-queue FIFO illustration is not a solution of this); and endogenize arrival rates.

(3) Robustness of the distributional conclusions: analyze regimes with $c_1 q_1 \neq c_2 q_2$ (distinct value functions for the two populations), richer waiting-cost structures, and mixed-population configurations, to determine whether the HFT stabilization of liquidity imbalances and the redistribution of costs to institutional investors persist qualitatively.

(4) Empirical validation. The paper's only empirical contact is the qualitative comparison of its computed invariant measure with the queue-dynamics measurements of Garèche et al. (2013), and its Conclusion underlines the lack of liquidity measurements adapted to current market microstructure. Going beyond that qualitative comparison, test the model's quantitative predictions — the stationary invariant measures of the queue sizes, the dynamics of the effective spread, and the redistribution of transaction costs between participant classes — against high-frequency limit-order-book data, and, on that basis, assess or construct liquidity measures that capture adverse effects on slow traders that conventional metrics miss.

References

1. Aimé Lachapelle, Jean-Michel Lasry, Charles-Albert Lehalle, Pierre-Louis Lions. Efficiency of the price formation process in presence of high frequency participants: a mean field game analysis. Mathematics and Financial Economics 10(3):223–262 (2016). DOI: 10.1007/s11579-015-0157-1. https://doi.org/10.1007/s11579-015-0157-1 (arXiv:1305.6323, https://arxiv.org/abs/1305.6323)
2. Jean-Michel Lasry, Pierre-Louis Lions. Mean field games. Japanese Journal of Mathematics 2(1):229–260 (2007). DOI: 10.1007/s11537-007-0657-8. https://doi.org/10.1007/s11537-007-0657-8
3. Diogo A. Gomes, João Saúde. A Mean-Field Game Approach to Price Formation. Dynamic Games and Applications 11(1):29–53 (2021). DOI: 10.1007/s13235-020-00348-x. https://doi.org/10.1007/s13235-020-00348-x
4. Jin Ma, Eunjung Noh. Equilibrium Model of Limit Order Books: A Mean-Field Game View. In: Stochastic Analysis, Filtering, and Stochastic Optimization, pp. 381–410, Springer (2022). DOI: 10.1007/978-3-030-98519-6_16. https://doi.org/10.1007/978-3-030-98519-6_16 (arXiv:2002.12857, https://arxiv.org/abs/2002.12857)
5. David Evangelista, Yuri Saporito, Yuri Thamsten. Price formation in financial markets: a game-theoretic perspective. arXiv:2202.11416 (2022). https://arxiv.org/abs/2202.11416 (published version: Price Formation in Financial Markets: A Mean-Field Game Perspective, pp. 71–90, Springer (2026), DOI: 10.1007/978-3-031-97733-6_3, https://doi.org/10.1007/978-3-031-97733-6_3)
6. Agustín Muñoz González, Juan I. Sequeira, Rafael Orive Illera. A New Framework for Modelling Liquidity Pools as Mean Field Games. arXiv:2412.09180 (2024; version 4, 2026). https://arxiv.org/abs/2412.09180
7. Nicolas Baradel, Bruno Bouchard, David Evangelista, Othmane Mounjid. Optimal inventory management and order book modeling. ESAIM: Proceedings and Surveys 65:145–181 (2019). DOI: 10.1051/proc/201965145. https://doi.org/10.1051/proc/201965145
8. Taiga Saito, Akihiko Takahashi. Stochastic differential game in high frequency market. Automatica 104:111–125 (2019). DOI: 10.1016/j.automatica.2019.02.051. https://doi.org/10.1016/j.automatica.2019.02.051
9. Roman Gayduk, Sergey Nadtochiy. Liquidity effects of trading frequency. Mathematical Finance 28(3):839–876 (2018). DOI: 10.1111/mafi.12157. https://doi.org/10.1111/mafi.12157
10. Weibing Huang, Mathieu Rosenbaum. Ergodicity and Diffusivity of Markovian Order Book Models: A General Framework. SIAM Journal on Financial Mathematics 8(1):874–900 (2017). DOI: 10.1137/16M1064337. https://doi.org/10.1137/16M1064337
11. Frank Kelly, Elena Yudovina. A Markov Model of a Limit Order Book: Thresholds, Recurrence, and Trading Strategies. Mathematics of Operations Research 43(1):181–203 (2018). DOI: 10.1287/moor.2017.0857. https://doi.org/10.1287/moor.2017.0857
12. A. Garèche, G. Disdier, J. Kockelkoren, J.-P. Bouchaud. Fokker-Planck description for the queue dynamics of large tick stocks. Physical Review E 88:032809 (2013). DOI: 10.1103/PhysRevE.88.032809. https://doi.org/10.1103/PhysRevE.88.032809
13. Ioanid Roşu. A Dynamic Model of the Limit Order Book. Review of Financial Studies 22(11):4601–4641 (2009). DOI: 10.1093/rfs/hhp011. https://doi.org/10.1093/rfs/hhp011
14. Xue Cheng, Meng Wang, Ziyi Xu. Mean Field Game of High-Frequency Anticipatory Trading. arXiv:2404.18200 (2024). https://arxiv.org/abs/2404.18200
15. Yuri Ashrafyan, Tigran Bakaryan, Diogo Gomes, Julian Gutierrez. A duality approach to a price formation MFG model. arXiv:2109.01791 (2021). https://arxiv.org/abs/2109.01791