The Order Book as a Markov Queuing System under a Fixed Reference Price

Huang, Lehalle, and Rosenbaum propose a model in which the limit order book, conditional on a fixed reference price, is represented as a continuous-time Markov chain on the state space of non-negative integer queue vectors, with order-flow intensities — submission, cancellation, and execution — allowed to depend on the full current configuration of standing liquidity across price levels. The key conceptual separation is between within-regime dynamics, in which the reference price (the mid or best bid-ask midpoint) remains fixed and the queue configurations evolve according to the Markov chain, and regime transitions, in which an event depletes a best-level queue to zero and causes a reference-price jump. Within each regime the system reaches approximate stationarity before the next price jump, which justifies fitting the state-dependent intensities from empirical data under a piecewise-stationarity hypothesis and then assembling the lower-frequency price dynamics from the renewal process of regime transitions. The state-dependent intensity structure is the empirical core of the paper: at each price level i relative to the reference price, the arrival rate λ^+_i and cancellation rate λ^-_i are estimated as functions of the current queue depths (Q_1, ..., Q_K), revealing that the system is genuinely non-homogeneous — deep queues suppress further arrivals at that level and attract arrivals at adjacent levels in ways that a constant-intensity Poisson model cannot capture. The mean-field approximation, which factors the joint queue distribution into a product of marginals evaluated at conditional mean depths, provides a computationally tractable approximation whose predictions for the stationary book shape and the first-passage distribution to boundary (price-jump time) match simulated and empirical data closely. Critically, the model reproduces not just the time-averaged book shape but also the autocorrelation structure of queue imbalance, inter-trade durations, and the conditional distribution of price moves given book configuration. For the desk, the queue-reactive model establishes a precise definition of what constitutes genuine price-formation pressure at the microstructure level: only reference-price transitions carry directional information about the underlying price process, while queue-level dynamics within a regime are essentially mechanical, mean-reverting fluctuations around the stationary book shape. This decomposition has direct consequences for how short-horizon order-book signals should be filtered: apparent imbalance at the best bid and ask is predominantly within-regime queue noise and not informative about the direction of the next reference-price move, while persistent depletion of a best-level queue approaching boundary conditions is the genuine price-formation signal.

The limit order book state is Q(t) = (Q_1(t), ..., Q_K(t)) ∈ ℤ_+^K where Q_i is the queue depth at the i-th price level relative to the reference price P*(t), and K is the tracked depth. The conditional intensity of a submission event at level i is λ^+_i(Q) and of a cancellation at level i is λ^-_i(Q), both functions of the full current state Q. The generator of the Markov chain on ℤ_+^K acts on bounded functions g as (Lg)(Q) = Σ_i [λ^+_i(Q)(g(Q + e_i) − g(Q)) + λ^-_i(Q)(g(Q − e_i) − g(Q))], where e_i is the i-th unit vector. The stationary distribution π on ℤ_+^K satisfies the balance equations L*π = 0. The reference price P* jumps when Q_i = 0 for i equal to the best level on one side; the inter-jump times constitute a renewal sequence with distribution given by the first-passage time of (Q_i(t))_{t ≥ 0} to zero starting from the stationary distribution. The mid-price autocorrelation at lag τ is determined by the renewal theory of this first-passage problem: E[ΔP*(t) ΔP*(t+τ)] = f(τ; λ^±, K), where f depends on the stationary solution of the Markov chain and decays as the renewal measure decays.

The Markov chain representation imposes a discipline that simpler tick-data models lack: every observable property of the order book — fill rates, queue depletion times, imbalance autocorrelations, mid-price autocorrelations — is in principle derivable from the single object (λ^+_i(·), λ^-_i(·)), rather than from separate ad hoc models for each observable. The empirical content lies in the functional form of the state-dependence: the estimated λ^+_i(Q) is not constant in Q_i but exhibits a saturation effect at large Q_i (the level is already full so further arrivals are discouraged by inventory constraints) and a cross-level substitution effect (queues thin at adjacent levels draw submissions from the saturated level), effects visible in the empirical intensity estimates across multiple stocks and trading days. The piecewise-stationarity assumption is the key approximation enabling identification: within each regime, the Markov chain is assumed to have reached its stationary distribution before the next reference-price jump occurs, which is testable empirically by comparing the distribution of Q at the time of price jumps to the unconditional stationary distribution π. Violations occur at high queue turnover rates or during news windows when regime durations are short relative to the Markov relaxation time, and these violations produce model misfits that appear as anomalous autocorrelations in the simulated price process relative to the observed one. The renewal-theoretic price process — constructed from the first-passage distributions to queue boundaries — generates a mid-price autocorrelation structure that is qualitatively consistent with the empirically observed near-zero autocorrelation of mid-price returns at short lags and the slow decay at medium lags, providing a microstructure grounding for the Hawkes-based descriptions of price dynamics without requiring the explicit self-excitation parameterization of the Hawkes framework.

The decomposition of order-book dynamics into within-regime Markov noise and between-regime price formation has an immediate filtering implication: queue imbalance at the best bid and ask should be filtered with a discount that increases with the ratio of within-regime turnover to between-regime jump frequency. During periods of high within-regime activity — many arrivals and cancellations per reference-price move — the queue imbalance is dominated by mechanical fluctuations, and the desk's book-imbalance conditioning signal should apply a heavier discount. Conversely, during periods of sustained queue depletion on one side with slow replenishment — which corresponds to a high estimated first-passage probability — the book state carries genuine price-formation information and the conditioning weight should be increased. Practical implementation of queue-reactive filtering requires maintaining a running estimate of the stationary queue distribution π(Q) for the current reference-price regime, which can be done efficiently by tracking the exponential-family sufficient statistics of the empirical queue depth distribution and applying a Kalman-type update as new order events arrive; the depletion probability at the best level, computed as 1 − π(Q_best ≥ 1), serves as a real-time indicator of imminent reference-price transition probability and can be incorporated directly into the desk's execution timing model.

The curse of dimensionality in the state space ℤ_+^K prevents direct nonparametric estimation of λ^±_i(Q) across the full state space: for K = 5 and queue depths truncated at M = 20, the state space has M^K = 3.2 × 10^6 elements, far exceeding the number of distinct Q values observed in any practical trading session. The basis expansion λ^±_i(Q) = Σ_α β^±_{i,α} φ_α(Q) with a chosen feature basis φ_α introduces parametric structure and must be cross-validated for the correct function class; misspecification of the basis leads to intensity estimates that violate detailed balance and produce stationary distributions that are numerical artifacts of the fitting procedure rather than reflections of the true book dynamics. Real-order-book dynamics violate the within-regime stationarity assumption most severely during news events and at the open and close of regular trading hours, when arrival and cancellation rates change discontinuously over timescales shorter than the regime duration, producing non-stationary within-regime dynamics that the model cannot capture. The regime-transition-as-first-passage framework also abstracts away the continuous information that arrives between transitions — including observable changes in order flow rates that predict the next transition — so the model's price-formation signal is necessarily less informative than would be available to an agent who observes the continuous arrival stream and uses it to update a hazard rate for the next reference-price move.