Paper: arXiv 2312.09353
Authors: Andrew Na, Justin Wan
Abstract
In this paper, we explore the use of a deep residual U-net with self-attention to solve the the continuous time time-consistent mean variance optimal trade execution problem for multiple agents and assets. Given a finite horizon we formulate the time-consistent mean-variance optimal trade execution problem following the Almgren-Chriss model as a Hamilton-Jacobi-Bellman (HJB) equation. The HJB formulation is known to have a viscosity solution to the unknown value function. We reformulate the HJB to a backward stochastic differential equation (BSDE) to extend the problem to multiple agents and assets. We utilize a residual U-net with self-attention to numerically approximate the value function for multiple agents and assets which can be used to determine the time-consistent optimal control. In this paper, we show that the proposed neural network approach overcomes the limitations of finite difference methods. We validate our results and study parameter sensitivity. With our framework we study how an agent with significant price impact interacts with an agent without any price impact and the optimal strategies used by both types of agents. We also study the performance of multiple sellers and buyers and how they compare to a holding strategy under different economic conditions.
Complexity vs Empirical Score
- Math Complexity: 9.0/10
- Empirical Rigor: 3.0/10
- Quadrant: Lab Rats — high math complexity, limited empirical testing
Why this score: The paper employs advanced mathematics including Hamilton-Jacobi-Bellman equations, backward stochastic differential equations, and high-dimensional viscosity solutions, but the summary focuses on theoretical reformulation and neural network architecture without mentioning backtesting, real data, or implementation metrics.
Research Flowchart
flowchart TD A["Research Goal:<br>Solve Multi-Agent Time-Consistent<br>Optimal Trade Execution"] --> B["Formulate HJB Equation<br>(Almgren-Chriss Model)"] B --> C["Reformulate HJB to BSDE<br>for Multiple Agents/Assets"] C --> D["Compute Solution via<br>Residual U-Net with Self-Attention"] D --> E["Outcomes:"] --> F["Validated Time-Consistent<br>Optimal Control Strategy"] D --> E --> G["Interaction Analysis:<br>High vs. Low Price Impact Agents"] D --> E --> H["Performance Study:<br>Multiple Sellers/Buyers vs. Holding"]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes a deep residual U‑net with self‑attention to numerically approximate the value function of a multi‑agent, multi‑asset time‑consistent mean‑variance optimal trade execution problem.
Source sentence
We utilize a residual U‑net with self‑attention to numerically approximate the value function for multiple agents and assets which can be used to determine the time‑consistent optimal control.
Assumes 5
- The market has sufficient liquidity and market frictions are ignored. market
Source sentence
For simplicity, we assume that the market has sufficient liquidity and we do not consider market frictions.
- Permanent price impact is linear in the trade rate. model
Source sentence
We use the following linear form for the permanent price impact
- Drift μi, risk‑free rate r and volatility σi are constant for each asset. model
Source sentence
In our model, we assume a constant μi, r and σi and we introduce a permanent price impact function gi(vi(t)).
- Each agent can observe a basket of d stocks and knows the price vector S_k(t). data
Source sentence
Suppose each agent k, where k ∈[1, ..., K], can observe a basket of d stocks in the market; with price process Sk(t) = [S1,k(t), ..., Sd,k(t)]⊤∈Rd.
Show 1 more
- Linear permanent price impact eliminates round‑trip arbitrage opportunities. other
Source sentence
A linear permanent price impact function is used to eliminate round‑trip arbitrage opportunities as shown in [Tse 2012].
Data
Synthetic data only.
Costs
Modeled
Permanent impact g(v)=κp v and temporary impact f(v)=[1+κs sign(v)]exp{κτ v^β} are incorporated into the price dynamics and cash equation.
Source sentence
We use the following linear form for the permanent price impact
Finds 5
- The best Sharpe ratio achieved for selling agents under poor market conditions is 0.35014. performance
Source sentence
The best Sharpe ratio we achieve is 0.35014 for selling agents under 2).
- Under mixed market conditions the best Sharpe ratio for sellers is 0.1814. performance
Source sentence
The best Sharpe ratio we achieve is 0.1814 for sellers under 3).
- Experiment 1 shows the proposed method matches classical solutions where applicable. theoretical
Source sentence
We show in Experiment 1 that the proposed method matches classical solutions where applicable.
- Increasing the awareness parameter ϕ leads agents to invest more in risky assets; higher risk aversion γ reduces risky‑asset holdings. statistical
Source sentence
When an agent is more conscious other agents in the market, i.e. ϕ increases, agents tend to invest more in risky assets. When an agent is more risk adverse, i.e. γ increases, the agent invests less in risky assets.
- Higher temporary price impact makes agents liquidate more gradually, while higher permanent impact speeds up liquidation. performance
Source sentence
We saw that as the temporary price impact of the agent increases they tend to liquidate more gradually and as the permanent price impact increases the agents liquidate faster.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 34 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (36)
- Deep backward schemes for high-dimensional nonlinear PDE s (2020)
- Deep neural network framework based on backward stochastic differential equations for pricing and hedging American options in high dimensions (2021) — link
- Efficient pricing and hedging of high-dimensional American options using deep recurrent networks (2023) — link
- Deep Learning Approximation for Stochastic Control Problems (2016) — link
- Deep Learning-Based Numerical Methods for High-Dimensional Parabolic Partial Differential Equations and Backward Stochastic Differential Equations (2017)
- Adam: A method for stochastic optimization (2017) — link
- A reinforcement learning extension to the Almgren-Chriss framework for optimal trade execution (2014) — link
- Cost-Efficient Reinforcement Learning for Optimal Trade Execution on Dynamic Market Environment (2022) — link
and 28 more
- A Deep Reinforcement Learning Framework for Optimal Trade Execution (2021)
- Double Deep Q-Learning for Optimal Execution (2021) — link
- Continuous-time mean–variance portfolio selection: A reinforcement learning framework (2020) — link
- Attention is All you Need (2017) — link
- Deep Residual Learning for Image Recognition (2016) — link
- Dynamic Mean-Variance Asset Allocation (2010)
- Multiperiod Mean Conditional Value at Risk Asset Allocation: Is It Advantageous to Be Time Consistent? (2020) — link
- ON TIME CONSISTENCY FOR MEAN-VARIANCE PORTFOLIO SELECTION (2020) — link
- Optimal execution of portfolio transactions (2000)
- Optimal execution strategies in limit order books with general shape functions (2010) — link
- Incorporating order-flow into optimal execution (2016) — link
- Optimal Dynamic Portfolio Selection: Multiperiod Mean-Variance Formulation (2000) — link
- Continuous-Time Mean-Variance Portfolio Selection: A Stochastic LQ Framework (2000) — link
- A Hamilton–Jacobi–Bellman approach to optimal trade execution (2011) — link
- On time-inconsistent stochastic control in continuous time (2017) — link
- Time-Consistent Investment and Reinsurance Strategies for Mean–Variance Insurers in N -Agent and Mean-Field Games (2022) — link
- An HJB approach to a general continuous-time mean-variance stochastic control problem (2018) — link
- Investment strategies and compensation of a mean–variance optimizing fund manager (2014) — link
- Continuous time mean variance asset allocation: A time-consistent strategy (2011) — link
- Time-consistent mean–variance portfolio optimization: A numerical impulse control approach (2018) — link
- Numerical Methods for Optimal Trade Execution (2012)
- Deep Learning, Adaptive computation and machine learning (2016)
- Multi-asset Optimal Execution and Statistical Arbitrage Strategies under Ornstein-Uhlenbeck Dynamics (2022) — link
- Trade execution in illiquid markets Optimal stochastic control and multi-agent equilibria (2008)
- Feynman-Kac Representation of Fully Nonlinear PDEs and Applications (2014) — link
- U-Net: Convolutional Networks for Biomedical Image Segmentation (2015) — link
- No-dynamic-arbitrage and market impact (2010) — link
- Optimal Trading with Stochastic Liquidity and Volatility (2012) — link
0 of 36 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.