Optimal Execution under Liquidity Uncertainty

This paper formulates and solves a singular control problem for optimal trade execution under general price impact and stochastic liquidity regimes, proving that the value function is the unique viscosity solution to a system of Hamilton-Jacobi-Bellman inequalities and characterizing the resulting optimal strategy through its free boundary.

The Gist

Imagine you are a giant whale trying to buy a massive amount of fish (shares) in a small, crowded pond (the stock market). If you just dive in and gulp everything at once, you'll scare away all the other fish, the water will churn violently, and the price of the remaining fish will skyrocket. You end up paying way too much.

This paper is a sophisticated guide on how a whale should eat its meal without scaring the pond. It answers the question: Should I buy a little bit now, wait for the water to calm down, or make a big jump later?

Here is the breakdown of their "whale strategy" using simple analogies:

1. The Problem: The "Churned Water" (Price Impact)

When you buy shares, you push the price up. This is called Price Impact.

• The Analogy: Imagine a line of people waiting to get into a club. If one person walks in, the line moves a bit. If a whole bus of people walks in, the line stretches out, and the bouncer has to let people in faster, raising the "cost" (or effort) for everyone else.
• The Twist: The pond isn't static. After you buy, the water naturally calms down, and new fish swim in to fill the gap. This is called Resilience. The paper asks: How fast does the water calm down, and how do we use that?

2. The New Ingredient: The "Weather Report" (Liquidity Uncertainty)

Previous models assumed the pond was predictable. This paper says, "No, the weather changes!"

• The Analogy: Sometimes the pond is calm and deep (High Liquidity). Sometimes it's shallow and stormy (Low Liquidity).
• The "Regime Switch": The market can suddenly switch from a "Sunny Day" (easy to buy cheaply) to a "Stormy Night" (hard to buy without spiking prices). The authors use a Markov Chain (a fancy coin flip that remembers its last state) to model these sudden shifts. You don't know exactly when the storm will hit, but you know the odds of it happening.

3. The "Volume Effect" (The Ripple)

The paper introduces a "Volume Effect Process." Think of this as a ripple in the water.

• The Analogy: Every time you buy, you create a ripple.
  • Drift: The water naturally wants to return to flat (Resilience).
  • Diffusion: Small waves caused by other swimmers (other traders).
  • Jumps: A sudden splash caused by a shark or a whale (a massive block trade by someone else).
• The paper models this ripple as a Jump-Diffusion process. It's not just a smooth wave; it's a wave that can suddenly get hit by a splash.

4. The Strategy: The "Free Boundary" (The Decision Line)

The core of the paper is finding the Optimal Execution Strategy. They don't just say "buy slowly." They draw a map with two zones:

• The "Wait and See" Zone (Continuation Region): The water is too choppy, or the storm is coming. You hold your breath and wait. The cost of buying now is too high.
• The "Go" Zone (Exercise Region): The water is calm, or the storm is about to hit. You must buy now.
• The Free Boundary: This is the invisible line separating the two. The paper proves this line is connected and moves based on the "weather."
  • If a storm is likely: You move the line. You buy more now in the calm zone because you fear the storm will make it expensive later.
  • If the water is calm and deep: You wait longer, buying smaller amounts to let the water settle between bites.

5. The Math: The "Crystal Ball" (Viscosity Solutions)

To find this perfect line, the authors use complex math called Hamilton-Jacobi-Bellman (HJB) equations.

• The Analogy: Imagine trying to solve a maze where the walls move and the exit changes color. You can't solve it with a simple ruler. You need a "Crystal Ball" (Viscosity Solution) that looks at every possible future path (storm, calm, big splash) and finds the path with the lowest total cost.
• They proved that this "Crystal Ball" solution is unique and works even when the math gets messy (discontinuous jumps).

6. The Results: What the Computer Said

They ran simulations (computer experiments) to see how their strategy works:

• High Volatility (Choppy Water): Surprisingly, if the water is very active (lots of small waves), it actually helps you! It means liquidity is replenishing fast. You can buy more aggressively.
• High Jump Risk (Sharks): If there's a risk of a massive "shark" (big trade) hitting the market, you should buy earlier and larger to get ahead of the price spike.
• Regime Switching: If you are in a "Good Market" but know a "Bad Market" is likely coming, you should eat your fish faster while the water is still clear. If you are in a "Bad Market" but a "Good Market" is coming, you should wait and starve a little longer.

Summary

This paper tells traders: "Don't just buy slowly. Watch the weather."

If the market is calm but a storm is coming, buy now. If the market is choppy but clearing up, wait. By mathematically modeling the "ripples," the "storms," and the "sharks," the authors provide a rulebook for buying huge amounts of stock at the lowest possible price, minimizing the cost of your own hunger.

Technical Summary

1. Problem Statement

The paper addresses the optimal execution problem for a trader seeking to purchase a large block of shares ($X$) over a fixed finite time horizon $[0, T]$. The core challenge is to minimize the expected execution cost (slippage) in a market characterized by:

• Non-linear and Transient Price Impact: The price impact depends on the shape of the Limit Order Book (LOB) and dissipates over time due to market resilience.
• Stochastic Liquidity: Unlike previous models assuming deterministic resilience, this framework models the "volume effect" (the temporary removal of liquidity) as a jump-diffusion process. This captures both continuous market fluctuations and sudden shocks (e.g., block trades by other participants).
• Regime Switching: Market conditions (specifically the shape of the LOB and the resulting price impact function) shift abruptly according to a finite-state Markov chain. This allows the model to capture structural changes in market depth and liquidity regimes.

The problem is formulated as a singular stochastic control problem, where the trader controls the rate of continuous trading and the size of discrete block trades (impulses) to minimize costs.

2. Methodology

Model Setup

• State Variables:
  • $x$: Remaining inventory to be purchased.
  • $y$: Volume effect process (representing the "thinning" of the LOB).
  • $i$: Current liquidity regime (state of the Markov chain).
• Dynamics:
  • Reference Price ($A_t$): Modeled as a martingale.
  • Volume Effect ($Y_t$): Evolves according to a Stochastic Differential Equation (SDE) with jumps:
    $$dY_u = -h(Y_{u-})du + \sigma(Y_{u-})dW_u + \int q(Y_{u-}, z)M(du, dz) + dX_u$$
    Here, $-h(Y)$ represents the deterministic resilience (replenishment), $\sigma$ represents continuous volatility, and the integral term represents jumps from external market activity. $dX_u$ is the trader's control (purchase).
  • Price Impact: The ask price is $P_t = A_t + D_t$, where the deviation $D_t = \psi_{I_t}(Y_t)$. The function $\psi_i$ is derived from the LOB shape function $F_i$, allowing for general, potentially discontinuous, order book shapes.
• Objective: Minimize the expected total cost, which includes the integral of the price impact over continuous trades and the discrete cost of block trades.

Mathematical Analysis

• Value Function: The value function $v_i(t, x, y)$ represents the minimal expected cost starting from state $(t, x, y)$ in regime $i$.
• Viscosity Solutions: Due to the lack of smoothness in the value function (caused by the singular control and jump terms), the authors do not seek classical solutions. Instead, they characterize $v$ as the unique viscosity solution to a system of Hamilton-Jacobi-Bellman Quasi-Variational Inequalities (HJBQVI).
  • The system involves a partial integro-differential operator $L$ (accounting for diffusion and jumps) and a coupling term from the Markov chain.
  • The HJBQVI takes the form:
    $$\max \left( -\frac{\partial v_i}{\partial t} - \mathcal{L}v_i - \sum_{j \neq i} (v_j - v_i)Q_{ij}, \quad -\frac{\partial v_i}{\partial x} - \frac{\partial v_i}{\partial y} - \psi_i \right) = 0$$
• Free Boundary: The solution defines a free boundary separating the continuation region (where it is optimal to wait) from the exercise region (where it is optimal to execute a trade).

Numerical Scheme

• The authors propose a finite difference scheme to approximate the viscosity solution.
• They use an Implicit-Explicit (IMEX) method to handle the stiffness of the PIDE (Partial Integro-Differential Equation).
• The integral term (representing jumps) is discretized using a trapezoidal quadrature rule with linear interpolation for off-grid points.
• The scheme ensures stability via a Courant-Friedrichs-Lewy (CFL) condition.

3. Key Contributions

1. Dynamic Stochastic Resilience: The paper introduces a jump-diffusion process for the volume effect, moving beyond the deterministic resilience assumptions of earlier models (e.g., Predoiu et al., Alfonsi et al.). This captures both continuous market noise and discrete liquidity shocks.
2. Regime-Switching Framework: It incorporates a Markov chain to model structural shifts in market liquidity. Crucially, the regime switch affects the shape of the LOB (the function $\psi_i$), allowing the model to reflect how price impact changes under different market conditions (e.g., high vs. low liquidity).
3. Rigorous Mathematical Characterization:
   • Proves the continuity of the value function.
   • Establishes that the value function is the unique viscosity solution to the HJBQVI system using a strong comparison principle.
   • Analyzes the geometry of the free boundary, proving it is path-connected and establishing a "partial smooth-fit" principle.
4. General LOB Shapes: The model accommodates arbitrary LOB shapes (block, power-law, etc.) and allows for discontinuous price impact functions, making it highly flexible for empirical calibration.

4. Results and Numerical Findings

The authors present numerical simulations under various configurations (single regime and regime-switching) using a square-root price impact model.

• Impact of Volatility and Resilience:

  • Higher Volatility ($\sigma$): Leads to lower execution costs and a smaller exercise boundary. High volatility implies more active market replenishment, allowing the trader to wait for better liquidity.
  • Higher Resilience ($h$): Similarly reduces costs and shrinks the exercise region, as liquidity recovers faster.
  • Higher Jump Intensity ($\lambda$): Increases execution costs and expands the exercise region. Traders anticipate future liquidity shocks (which will make trading more expensive) and choose to execute larger trades earlier to avoid these future costs.

• Regime-Switching Dynamics:

  • Anticipatory Behavior: Traders adjust strategies based on the probability of switching regimes.
    • If the current regime is low-impact but there is a risk of switching to a high-impact regime, the trader expands the exercise region (trades more aggressively now) to lock in lower costs before the switch.
    • Conversely, if in a high-impact regime with a chance of switching to low-impact, the trader contracts the exercise region (waits) to trade later when conditions improve.
  • Asymmetry: The frequency and direction of regime switches significantly alter the optimal strategy. Higher switching intensity generally amplifies these anticipatory behaviors.

5. Significance

This paper significantly advances the theory of optimal execution by bridging the gap between theoretical models and complex, real-world market microstructure.

• Practical Relevance: By modeling liquidity as a stochastic process with regime shifts, the framework provides a more realistic tool for institutional traders to manage execution risk in volatile or changing market environments.
• Theoretical Depth: The rigorous treatment of singular control with jumps and regime switching, including the proof of uniqueness for viscosity solutions and the analysis of the free boundary, provides a robust mathematical foundation for future research in stochastic control under uncertainty.
• Strategic Insight: The results highlight that optimal execution is not just about minimizing immediate impact but involves strategic timing based on the stochastic evolution of liquidity and the anticipation of future market regime shifts.