Paper: arXiv 2502.18177

Authors: Remi Genet

Abstract

The execution of Volume Weighted Average Price (VWAP) orders remains a critical challenge in modern financial markets, particularly as trading volumes and market complexity continue to increase. In my previous work arXiv:2502.13722, I introduced a novel deep learning approach that demonstrated significant improvements over traditional VWAP execution methods by directly optimizing the execution problem rather than relying on volume curve predictions. However, that model was static because it employed the fully linear approach described in arXiv:2410.21448, which is not designed for dynamic adjustment. This paper extends that foundation by developing a dynamic neural VWAP framework that adapts to evolving market conditions in real time. We introduce two key innovations: first, the integration of recurrent neural networks to capture complex temporal dependencies in market dynamics, and second, a sophisticated dynamic adjustment mechanism that continuously optimizes execution decisions based on market feedback. The empirical analysis, conducted across five major cryptocurrency markets, demonstrates that this dynamic approach achieves substantial improvements over both traditional methods and our previous static implementation, with execution performance gains of 10 to 15% in liquid markets and consistent outperformance across varying conditions. These results suggest that adaptive neural architectures can effectively address the challenges of modern VWAP execution while maintaining computational efficiency suitable for practical deployment.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper presents advanced mathematics with recurrent neural networks, temporal Kolmogorov-Arnold networks, and formal VWAP optimization frameworks with detailed discretization and decomposition formulas. It also demonstrates strong empirical rigor through backtesting across five major cryptocurrency markets with specific performance metrics (10-15% improvement) and discussion of computational efficiency for practical deployment.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Develop Dynamic VWAP Execution<br>with Adaptive RNNs"] --> B["Data & Inputs:<br>5 Cryptocurrency Market Datasets<br>Volume & Price Time Series"]
  B --> C["Methodology:<br>Recurrent Neural Networks RNN<br>Temporal Kolmogorov-Arnold Networks TKN"]
  C --> D["Computational Process:<br>Real-time Adaptive Adjustment<br>Dynamic Optimization Loop"]
  D --> E["Key Outcomes:<br>10-15% Execution Performance Gain<br>Superior vs Static Models<br>Adaptability to Market Conditions"]

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 dynamic neural VWAP execution framework that integrates recurrent neural networks and a dynamic adjustment mechanism to continuously optimize execution decisions in real time.

Source sentence
We introduce two key innovations: first, the integration of recurrent neural networks to capture complex temporal dependencies in market dynamics, and second, a sophisticated dynamic adjustment mechanism that continuously optimizes execution decisions based on market feedback.

Assumes 4

  • The framework assumes real‑time market feedback is available to adjust execution decisions. market
    Source sentence
    a sophisticated dynamic adjustment mechanism that continuously optimizes execution decisions based on market feedback
  • The model uses a lookback window of 120 time steps to predict the next 12 periods, implying stationarity within that window. data
    Source sentence
    The model processes market data using a lookback window of 120 time steps to generate predictions for the subsequent 12 periods.
  • The approach does not explicitly model market impact in its execution cost calculations. execution
    Source sentence
    while my framework achieves substantial improvements without explicitly modeling market impact
  • Future information does not affect current predictions due to causal padding in the architecture. model
    Source sentence
    future information does not affect current predictions. The causality of the architecture guarantees that these padded values do not influence the predictions for the current timestep

Data

Backtest sample: BTC, ETH, BNB, ADA, XRP on Binance perpetual futures · crypto · intraday

  • Hourly trading data from five major cryptocurrencies (BTC, ETH, BNB, ADA, XRP) on Binance · Binance crypto-onchain public
    inception of each contract until July 1, · BTC, ETH, BNB, ADA, XRP
    Source sentence
    The empirical analysis utilizes hourly trading data from five major cryptocurrencies (BTC, ETH, BNB, ADA, XRP) traded on Binance perpetual futures contracts. The dataset spans from the inception of each contract until July 1, 2024

Recognised sources: Binance

Costs

Unclear The classifier was not confident.

Finds 3

  • The dynamic approach achieves execution performance gains of 10 to 15% in liquid markets. performance
    Source sentence
    execution performance gains of 10 to 15% in liquid markets
  • Dynamic‑VWAP with TKAN reduces absolute VWAP loss for ETH to 0.121, an 11.0% improvement over its static counterpart (0.136). performance
    Source sentence
    For ETH, the DynamicVWAP with TKAN achieves an absolute VWAP loss of 0.121, an 11.0% improvement over its static counterpart (0.136)
  • Training times increase by an order of magnitude (5‑8 s to >150 s) for TKAN implementations, with a 30% overhead for the dynamic framework justified by a 10‑15% execution quality improvement. performance
    Source sentence
    When applied directly, RNNs introduce substantial computational overhead compared to simple linear models like TLN - training times increase by an order of magnitude, from 5-8 seconds to over 150 seconds for TKAN implementations. However, these architectures prove essential as building blocks for our dynamic framework.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 24 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive

Code

https://github.com/remigenet/DeepDynamicVWAP — 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.

Cited References (39)

  1. The implementation shortfall: Paper versus reality (1988)
  2. High frequency trading under scrutiny (2009)
  3. The Total Cost of Transactions on the NYSE (1988)
  4. VWAP strategies, Transaction Performance: The Changing Face of Trading Investment Guides Series (2002)
  5. Optimal execution of portfolio transactions (2000)
  6. Optimal slice of a VWAP trade (2002) — link
  7. Optimal VWAP Trading Strategy and Relative Volume (2007) — link
  8. Mean Variance Optimal VWAP Trading (2012) — link
and 31 more
  1. Price, trade size, and information in securities markets (1987) — link
  2. A Theory of the Interday Variations in Volume, Variance, and Trading Costs in Securities Markets (1990) — link
  3. The Price Variability-Volume Relationship on Speculative Markets (1983) — link
  4. The Relation between Price Changes and Trading Volume: A Survey (1987)
  5. Intra-day market activity (1999)
  6. Optimal VWAP trading under noisy conditions (2011)
  7. Optimal execution of a vwap order: A stochastic control approach (2013) — link
  8. Dynamic Programming and Trade Execution (2013)
  9. VWAP Execution and Guaranteed VWAP (2014)
  10. Are the Trading Volume and the Number of Trades Distributions Universal (2011)
  11. Permanent Market Impact Can Be Nonlinear (2013)
  12. Hierarchical Deep Reinforcement Learning for VWAP Strategy Optimization — → 2212.14670
  13. An Adaptive Dual-level Reinforcement Learning Approach for Optimal Trade Execution — → 2307.10649
  14. An Optimal Control Strategy for Execution of Large Stock Orders Using LSTMs — → 2301.09705
  15. Deep Learning for VWAP Execution in Crypto Markets: Beyond the Volume Curve — → 2502.13722
  16. TKAN: Temporal Kolmogorov-Arnold Networks (2024) — link
  17. SigKAN: Signature-Weighted Kolmogorov-Arnold Networks for Time Series (2024) — link
  18. A temporal kolmogorov-arnold transformer for time series forecasting (2024)
  19. A Gated Residual Kolmogorov-Arnold Networks for Mixtures of Experts (2024)
  20. A Temporal Linear Network for Time Series Forecasting (2024)
  21. Are Transformers Effective for Time Series Forecasting? (2022) — link
  22. TSMixer: An All-MLP Architecture for Time Series Forecasting (2023) — link
  23. Long short-term memory (1997)
  24. Learning phrase representations using RNN encoder-decoder for statistical machine translation (2014)
  25. Neural machine translation by jointly learning to align and translate (2014) — link
  26. Attention is all you need (2017)
  27. Deep Smoothing of the Implied Volatility Surface (2020)
  28. Deep Learning Volatility: A deep neural network perspective on pricing and calibration in (rough) volatility models (2019) — link
  29. Financial time series forecasting with deep learning: A systematic literature review: 2005–2019 (2020)
  30. Deep learning models for price forecasting of financial time series: A review of recent advancements: 2020-2022 — → 2305.04811
  31. SigGate: Enhancing Recurrent Neural Networks with Signature-Based Gating Mechanisms (2025) — link

5 of 39 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.