Risk-Sensitive Optimal Execution via a Conditional Value-at-Risk Objective

Seungki Min, Ciamac C. Moallemi, Costis MaglarasarXiv 2022 · DOI · arXiv

Abstract

We consider a liquidation problem in which a risk-averse trader seeks to liquidate a fixed quantity of an asset while balancing transaction costs from market impact against the volatility risk of holding the position. Departing from the traditional mean-variance framework, we define the objective as the Conditional Value-at-Risk (CVaR) of the implementation shortfall. By formulating the problem in continuous time and allowing for dynamic, adaptive trading strategies, we derive the first closed-form analytical solutions for this class of risk-sensitive control. Our analysis utilizes the dual representation of CVaR to transform the optimization into a zero-sum stochastic game, solved via state-space augmentation and the Emden-Fowler equation. We demonstrate that the optimal policy exhibits "aggressiveness-in-the-money," outperforming optimized static policies by 5–15% and standard VWAP policies by 15–25%.

Introduction

In quantitative finance, "optimal execution" refers to the challenge of liquidating a large block of stock without moving the market price unfavorably. Traders face a fundamental trade-off: selling rapidly to mitigate volatility risk (the risk that the asset price drops during the liquidation period) or selling slowly to minimize market impact (the transaction costs incurred by executing large orders).

The seminal model by Almgren and Chriss (2000) addressed this using a mean-variance framework, which typically results in static execution schedules determined at the start of the trading day. However, mean-variance objectives do not specifically protect against "tail risk"—the rare but catastrophic price movements that occur in the distribution's extremes. Furthermore, static strategies fail to adapt to real-time market fluctuations. This paper addresses these gaps by employing Conditional Value-at-Risk (CVaR) as the primary risk measure and seeking a dynamic strategy that adapts to price movements as they occur.

Mathematical Framework

The CVaR Challenge

CVaR is a coherent risk measure that focuses on the average of the worst-case outcomes beyond a specific quantile. While robust, CVaR is notoriously difficult to optimize dynamically because it is time-inconsistent; a strategy that is optimal at time zero may no longer be optimal at a later time.

Stochastic Game and State-Space Augmentation

To overcome this, the authors employ a dual representation of CVaR, transforming the trader's minimization problem into a continuous-time, zero-sum stochastic game. In this framework, the trader competes against a hypothetical "adversary" who attempts to choose the worst possible price scenarios.

To make the problem Markovian and solvable via dynamic programming, the authors introduce a new state variable, $Q_t$, representing the "quantile process." This state-space augmentation allows the trader to track their current risk exposure relative to the tail of the cost distribution in real-time.

The Emden–Fowler Solution

By deriving the Hamilton-Jacobi-Bellman (HJB) partial differential equation for this game, the authors find that the equation is separable. It simplifies into a specific instance of the Emden–Fowler equation, a well-known second-order ordinary differential equation. This mathematical breakthrough allows for a closed-form solution using Bessel functions, providing exact formulas for both the liquidation rate and the value function.

Strategic Insights: Aggressiveness-in-the-Money

The most significant qualitative finding of the paper is the nature of the optimal adaptive policy, termed "Aggressiveness-in-the-Money" (AIM).

  • Favorable Movements: If the asset price moves in the trader's favor, the strategy accelerates trading. By selling faster when prices are high, the trader locks in gains and eliminates future volatility risk.
  • Unfavorable Movements: Conversely, if the price moves against the trader, the strategy slows down. This avoids compounding existing losses with the high transaction costs associated with aggressive selling.

This behavior stands in stark contrast to many common heuristics and demonstrates how a CVaR-focused strategy "shapes" the tail of the cost distribution. Rather than producing a normal distribution of costs, the CVaR-optimal strategy creates a distribution with a sharp peak at the target quantile, effectively squeezing the tail of bad outcomes more efficiently than static methods.

Performance and Results

The authors validated the theoretical model through extensive Monte Carlo simulations, comparing the optimal dynamic CVaR policy against two benchmarks: an optimized static policy and a Volume Weighted Average Price (VWAP) policy.

  1. Adaptive Alpha: The dynamic policy consistently outperforms the static policy. The "adaptive alpha"—the benefit of real-time adjustment—is most pronounced for traders with moderate risk aversion, yielding improvements of 5–15%.
  2. Benchmark Comparison: When compared to an optimized VWAP strategy, the CVaR-optimal dynamic policy shows even greater gains, ranging from 15–25%.
  3. Risk Sensitivity: Interestingly, the relative benefit of the adaptive strategy increases as the trader becomes less risk-averse (closer to risk-neutral), as they have more flexibility to adjust their trading speed in response to price signals.

Conclusion and Limitations

This research provides a rigorous mathematical foundation for risk-sensitive optimal execution. By deriving closed-form solutions for a CVaR objective, the authors bridge the gap between theoretical risk measures and practical trading applications.

However, certain limitations remain for future exploration. The model assumes an infinite time horizon, which simplifies the mathematics but may not perfectly align with real-world "must-fill" orders with hard deadlines. Additionally, the model focuses on temporary market impact and assumes price follows an arithmetic Brownian motion. Future research could extend these results to include permanent price impact or more complex price dynamics, such as stochastic volatility or price jumps, although such additions may preclude the analytical closed-form solutions found here.

Summary updated

Abstract

We consider a liquidation problem in which a risk-averse trader tries to liquidate a fixed quantity of an asset in the presence of market impact and random price fluctuations. The trader encounters a trade-off between the transaction costs incurred due to market impact and the volatility risk of holding the position. Our formulation begins with a continuous-time and infinite horizon variation of the seminal model of Almgren and Chriss (2000), but we define as the objective the conditional value-at-risk (CVaR) of the implementation shortfall, and allow for dynamic (adaptive) trading strategies. In this setting, we are able to derive closed-form expressions for the optimal liquidation strategy and its value function. Our results yield a number of important practical insights. We are able to quantify the benefit of adaptive policies over optimized static policies. The relevant improvement depends only on the level of risk aversion: for moderate levels of risk aversion, the optimal dynamic policy outperforms the optimal static policy by 5-15%, and outperforms the optimal volume weighted average price (VWAP) policy by 15-25%. This improvement is achieved through dynamic policies that exhibit "aggressiveness-in-the-money": trading is accelerated when price movements are favorable, and is slowed when price movements are unfavorable. From a mathematical perspective, our analysis exploits the dual representation of CVaR to convert the problem to a continuous-time, zero-sum game. We leverage the idea of the state-space augmentation, and obtain a partial differential equation describing the optimal value function, which is separable and a special instance of the Emden-Fowler equation. This leads to a closed-form solution. As our problem is a special case of a linear-quadratic-Gaussian control problem with a CVaR objective, these results may be interesting in broader settings.
BibTeX
@misc{min2022risk,
  author = {Seungki Min and Ciamac C. Moallemi and Costis Maglaras},
  title = {{Risk-Sensitive Optimal Execution via a Conditional Value-at-Risk Objective}},
  year = {2022},
  eprint = {2201.11962},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2201.11962},
}
Figure