Paper: arXiv 2502.07868

Authors: Moustapha Pemy, Na Zhang

Abstract

This paper studies the ubiquitous problem of liquidating large quantities of highly correlated stocks, a task frequently encountered by institutional investors and proprietary trading firms. Traditional methods in this setting suffer from the curse of dimensionality, making them impractical for high-dimensional problems. In this work, we propose a novel method based on stochastic optimal control to optimally tackle this complex multidimensional problem. The proposed method minimizes the overall execution shortfall of highly correlated stocks using a reinforcement learning approach. We rigorously establish the convergence of our optimal trading strategy and present an implementation of our algorithm using intra-day market data.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper applies advanced stochastic optimal control and reinforcement learning with multiple mathematical proofs and derivations, indicating high math complexity. It includes real intra-day market data implementation and a GitHub repository with code, demonstrating substantial empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Liquidation of Correlated Stocks"] --> B["Formulate as<br>Stochastic Optimal Control"]
  B --> C["Methodology:<br>Reinforcement Learning"]
  C --> D["Data:<br>Intra-day Market Data"]
  D --> E["Computational Process:<br>RL Algorithm Implementation"]
  E --> F["Key Findings:<br>Minimal Shortfall Strategy"]
  F --> G["Outcome:<br>Convergent & Optimal Execution"]

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 minimal shortfall algorithm that uses reinforcement learning to jointly train two neural networks for optimal liquidation of highly correlated stocks.

Source sentence
We propose a novel method based on stochastic optimal control to optimally tackle this complex multidimensional problem. The proposed method minimizes the overall execution shortfall of highly correlated stocks using a reinforcement learning approach.

Assumes 6

  • The Markov Decision Process under any policy is irreducible and aperiodic. model
    Source sentence
    (A1): We assume that under any policy π = (ak)k our Markov Decision Process is irreducible and aperiodic.
  • The policy function is differentiable with respect to its neural‑network parameters. model
    Source sentence
    (A2): We assume that our policy function π(x, a, θ) is differentiable with respect to θ.
  • Stock prices follow a discrete‑time geometric Brownian motion. model
    Source sentence
    the environment in which we train our neural networks is a discrete-time controlled dynamical system where stock prices follow a discrete version of the geometric Brownian motion
  • All stocks to be liquidated are highly correlated. market
    Source sentence
    we assume without loss of generality that those stocks are correlated and we will use their correlation to our advantage
Show 2 more
  • Trading actions incur a linear market impact. execution
    Source sentence
    we assume the stocks we want to liquidate are highly correlated and that our trading actions will incur a linear market impact.
  • Maximum trade size per stock is bounded by a fixed threshold Mi. execution
    Source sentence
    For each of these stock we have a threshold Mi that we can traded at once without been detected.

Data

  • Intra‑day market data (week of September 5‑8 2017) market-prices unknown
    2017-09-05 to 2017-09-08
    Source sentence
    For the week of September 5 to 8, 2017, the trading data of the first three days are used as training data and the trading data of the last day of the week are the testing data.

Costs

Modeled Costs are part of the reported headline results.

Linear market impact functions fi(x) = λi x with λi = −1 × 10⁻¹¹ for each stock.

Source sentence
the market impact functions are fi(x) = λix with λi = −1 × 10−11, i = 1, ..., 6.

Finds 2

  • The proposed reinforcement‑learning methodology achieves its intended objectives, minimizing execution shortfall and tracking error, and can be implemented in software. performance
    Source sentence
    We can see that our methodology clearly achieves its intended objectives and can be effectively implemented in a software package.
  • The authors rigorously establish convergence of the optimal trading strategy. theoretical
    Source sentence
    We rigorously establish the convergence of our optimal trading strategy

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 22 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

https://github.com/mpemy/optimaltradingml — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.