Multidimensional stochastic liquidity in Kyle's model of informed trading

Abstract: We develop a variational formulation of Kyle's model of informed trading that accommodates stochastic liquidity and multiple traded assets. The main equilibrium result is stated first: under a martingale dual condition, a matrix-valued martingale depth process generates a linear-Gaussian equilibrium with stochastic matrix-valued price impact. We derive this martingale from a primal-dual problem, inspired by causal optimal transport, that characterizes the endogenous speed at which the insider injects private information into prices; in general, this problem admits only local martingale optimizers, and the martingale dual condition is the hypothesis that the optimizer is a true martingale. We interpret informed trading as the optimal liquidation of private information and verify the construction in the scalar and common-eigenbasis cases. The fully general matrix-valued case reduces to a coupled matrix FBSDE, which we isolate as the remaining obstruction. Along the way, we establish an independently interesting Doob-Meyer decomposition for general (not necessarily symmetric) matrix-valued submartingales.

• The paper extends Kyle's model for informed trading by integrating multidimensional, stochastic liquidity via a convex variational optimization approach.
• It employs a primal–dual formulation to rigorously characterize endogenous market depth and the optimal insider trading strategy.
• The work connects classical single-asset solutions to generalized multi-asset settings and introduces novel matrix-valued stochastic analysis tools.

Variational Formulation of Multidimensional Stochastic Liquidity in Kyle's Model

Introduction and Motivation

The paper "Multidimensional stochastic liquidity in Kyle's model of informed trading" (2607.10934) addresses the equilibrium problem of informed trading under multidimensional and stochastic liquidity, extending the classical Kyle model and its continuous-time generalizations. It departs from static and deterministic price impact approaches by integrating stochasticity and asset multiplicity into the market microstructure, thus capturing heterogeneity and endogeneity in informational flows and liquidity provision. The framework utilizes a variational optimization paradigm, recasting information release and market depth determination as a coupled primal–dual problem with a transport-theoretic perspective.

Model Construction and Mathematical Formulation

The setup generalizes the Kyle–Back model to $n$ risky assets whose (Gaussian) terminal values $\tilde v$ are known to an insider from $t=0$, and to a market impacted by stochastic $\mathcal{S}^n_{++}$-valued noise volatility processes $\sigma_t$ driven by a public Brownian factor. Aggregate order flow consists of the insider's strategy and exogenous (noise) demand. Market makers only observe the sum of order flow and the public liquidity factor, filtering the insider's information via their observable filtration. Prices are determined by rational expectations conditional on aggregate order flow and the public signal.

The equilibrium search is cast as a variational problem over the market maker's posterior covariance process $(\Sigma_t)_{t \in [0,T]}$. The control variable is the rate at which the insider reveals information (i.e., reduces posterior variance) subject to martingale and monotonicity constraints arising from the dynamics of filtered estimation.

Primal–Dual Variational Construction

The primal form maximizes the expected 'value of information release', balancing speed against prevailing noise volatility and the endogenous 'inventory' of unrevealed information. The dual problem identifies the Lagrange multiplier as a matrix-valued $\mathcal{S}^n_{++}$ martingale $(M_t)_{t \in [0,T]}$, representing market depth. The core technical result is that, conditional on the solution $M^*$ being a true ($\mathbb{F}^\mathcal{W}$-)martingale (the Martingale Dual Condition, MDC), the pair consisting of the pathwise price impact $\tilde v$0 and a dynamically optimal insider trading strategy defines a linear-Gaussian equilibrium.

The duality is rigorously justified by convex analytic arguments and a matrix-valued Doob–Meyer decomposition. The equilibrium is shown to exist whenever the minimizer is interior and a true martingale; the scalar and aligned multidimensional settings are fully characterized, but for general matrix data and non-commuting volatilities, the existence question reduces to a sophisticated fully-coupled matrix FBSDE, which is isolated as open.

Main Results and Characterization

The principal theorem establishes that, under MDC, there exists an equilibrium with:

• A linear pricing rule: $\tilde v$1;
• Endogenous, stochastic matrix-valued price impact $\tilde v$2;
• The market maker's posterior covariance $\tilde v$3 satisfies a matrix Riccati equation, decreasing from prior $\tilde v$4 to zero at maturity, ensuring full revelation.
• The optimal insider strategy is absolutely continuous and inconspicuous—instantaneously conditionally mean-zero trade rates—which ensures statistical indistinguishability from noise and maximizes profit.

The relationship between information dynamics and market depth is explicitly captured by the martingale $\tilde v$5: the full path of information-releasing strategies and price responses arises as a saddlepoint in the associated primal–dual Lagrangian.

Verification in Special Cases and Link to Prior Literature

The authors demonstrate that in the scalar case ($\tilde v$6), the framework recovers classical model solutions:

• Kyle (1985): The equilibrium price impact and linear strategies are immediate, with deterministic market depth.
• Back–Pedersen (1998): Admits time-varying but deterministic (scalar) liquidity.
• Collin-Dufresne/Fos (2016): For stochastic scalar liquidity with diffusive dynamics.

For the multidimensional setting, equilibrium is characterized when all covariance and volatility matrices are jointly diagonalizable (i.e., share a common eigenbasis). In this scenario, the problem decomposes into independent scalar problems, recovering Back, Cocquemas, Ekren, and Lioui (2020).

For the general non-commuting matrix case, the analysis reveals that the associated FBSDE lacks current well-posedness theory due to its full coupling and nonlinear structure; thus, existence and regularity remain open.

Doob–Meyer Decomposition for Matrix-Valued Processes

A secondary technical contribution is a matrix-valued Doob–Meyer decomposition theorem for local submartingales. This enables rigorous construction and compactness in the matrix martingale duality and may be of independent interest in the study of stochastic processes on operator-valued filtrations.

Economic Interpretation and Implications

The variational transport framework implies that the core endogenous object is the speed at which the insider's informational advantage is liquidated, not the share inventory per se. Market depth acts as the shadow price of remaining private information, varying over time and across assets according to prevailing and anticipated noise volatility. The equilibrium endogenizes the timing and manner of information revelation, reconciling incentive compatibility and rational pricing in a multidimensional, stochastic environment.

From a theoretical perspective, the work develops a unified framework spanning key continuous-time Kyle-type equilibria. Practically, the model informs mechanism design and optimal execution in settings with state-dependent liquidity and multi-asset cross-impact, highlighting the endogenous nature of depth formation.

Future Directions

Outstanding questions involve solvability and characterization of the general matrix FBSDE governing $\tilde v$7 without shared eigenbases, the implications of richer noise structures, and connections to real-world applications such as optimal multi-asset transaction execution under information asymmetry and advanced market microstructure econometrics.

Progress on the matrix quadratic FBSDE and further understanding of multidimensional stochastic filtering under adversarial informed trading scenarios would have substantial impact on both economics and applied probability. The optimal transport variational lens introduced here may also catalyze developments in related models involving adversarial information flows and endogenous signal extraction in complex markets.

Conclusion

This paper presents a rigorous, convex-analytic reformulation of equilibrium informed trading with stochastic, multidimensional liquidity. By codifying the mutual endogenous nature of information release and market depth via a primal–dual variational approach, the work systematically unifies and extends classical Kyle models. It identifies both the structural limitations and the explicit tractable cases in higher dimensions, and introduces matrix-valued stochastic analysis tools likely to find application beyond this context. The analysis clarifies both the mechanics of price impact formation and the statistical properties of equilibrium under multidimensional information asymmetry and time-varying liquidity.