Paper: arXiv 2411.10956

Authors: Hanwool Lee, Heehwan Park

Abstract

This paper presents a new approach to volume ratio prediction in financial markets, specifically targeting the execution of Volume-Weighted Average Price (VWAP) strategies. Recognizing the importance of accurate volume profile forecasting, our research leverages the Transformer architecture to predict intraday volume ratio at a one-minute scale. We diverge from prior models that use log-transformed volume or turnover rates, instead opting for a prediction model that accounts for the intraday volume ratio’s high variability, stabilized via log-normal transformation. Our input data incorporates not only the statistical properties of volume but also external volume-related features, absolute time information, and stock-specific characteristics to enhance prediction accuracy. The model structure includes an encoder-decoder Transformer architecture with a distribution head for greedy sampling, optimizing performance on high-liquidity stocks across both Korean and American markets. We extend the capabilities of our model beyond point prediction by introducing probabilistic forecasting that captures the mean and standard deviation of volume ratios, enabling the anticipation of significant intraday volume spikes. Furthermore, an agent with a simple trading logic demonstrates the practical application of our model through live trading tests in the Korean market, outperforming VWAP benchmarks over a period of two and a half months. Our findings underscore the potential of Transformer-based probabilistic models for volume ratio prediction and pave the way for future research advancements in this domain.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced Transformer architectures with probabilistic distribution heads (Student’s t-distribution) and log-normal data transformations, demonstrating high mathematical complexity. Empirical rigor is strong, featuring live trading tests over 2.5 months, specific performance metrics (outperforming VWAP benchmarks), and details on data features and model architecture suitable for implementation.

Research Flowchart

  flowchart TD
  A["Research Goal: Accurate<br>Intraday VWAP Volume<br>Ratio Prediction"] --> B["Data Preparation &<br>Feature Engineering"]
  
  subgraph B ["Data Preparation & Feature Engineering"]
      B1["Raw Volume Data"]
      B2["Log-Normal Transformation"]
      B3["External Features &<br>Stock Characteristics"]
  end

  B --> C["Model: Encoder-Decoder<br>Transformer Architecture"]
  
  subgraph C ["Model: Encoder-Decoder<br>Transformer Architecture"]
      C1["Encoder: Processes Input Features"]
      C2["Decoder: Predicts Volume Ratio"]
  end

  C --> D["Probabilistic Forecasting<br>Mean & Standard Deviation"]
  
  D --> E{"Outcomes"}
  
  E --> F["High Accuracy<br>on High-Liquidity Stocks<br>KR & US Markets"]
  E --> G["Live Trading Agent<br>Outperforms VWAP Benchmark<br>Korean Market 2.5 Months"]

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 IVE, a Transformer encoder‑decoder model with a distribution head for probabilistic minute‑level intraday volume ratio forecasting.

Source sentence
We developed a model architecture leveraging the Transformer encoder-decoder framework, optimized for predicting minute-level intraday volume ratios. This model incorporates diverse features

Assumes 4

  • Intraday volume ratios follow a log‑normal distribution after log transformation. statistical
    Source sentence
    our approach focuses on stabilizing these ratios through log‑transformation
  • The model is evaluated on high‑liquidity stocks in Korean and US markets. market
    Source sentence
    optimizing performance on high‑liquidity stocks across both Korean and American markets.
  • Predictions up to one step ahead are used for trading decisions. execution
    Source sentence
    in practical applications, we utilize predictions up to one step ahead for trading decisions.
  • Features are normalized to ensure comparability across stocks and dates. data
    Source sentence
    ... ensuring consistency and comparability across different datasets.

Data

Backtest sample: Korean market (high‑liquidity stocks) · equities · intraday

  • Refinitiv intraday volume dataset · Refinitiv market-prices commercial
    Jan 25 2023 – Aug 2023 · top 100 stocks by market cap in Korean and US markets
    Source sentence
    Our model was trained on data collected from January 25, 2023, to August 2023 from Refinitiv. This dataset includes the top 100 stocks by market capitalization in both the Korean and US markets as of September 2023.

Recognised sources: Refinitiv / LSEG (Datastream, Eikon, Tick History)

Costs

Not modeled A trading or execution result is reported without costs.

Source sentence
average execution performance that was 4.82 basis points better than the Market VWAP

Finds 3

  • IVE achieved the lowest RMSE (0.2028) and MAE (0.1229) among compared models for the Korean market. performance
    Source sentence
    IVE achieved the lowest RMSE of 0.2028 and MAE of 0.1229. This signifies a marked improvement over the BiLSTM‑HR, which was the next best model with an RMSE of 0.2055 and MAE of 0.1374.
  • IVE also outperformed other models in the US market with RMSE 0.1678 and MAE 0.0876. performance
    Source sentence
    In the US market, the IVE also outperformed its counterparts, achieving an RMSE of 0.1678 and an MAE of 0.0876.
  • Live trading tests in Korea showed IVE beating VWAP by an average of 4.82 basis points over two and a half months. performance
    Source sentence
    ... resulted in an average execution performance that was 4.82 basis points better than the Market VWAP, with a beat ratio of 59%.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 20 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 (19)

  1. Optimal Execution: A Review (2022) — link
  2. VWAP execution as an optimal strategy (2015) — link
  3. Dynamic Execution of VWAP Orders with Short-Term Predictions (2013) — link
  4. A Theory of Trading Volume (1986)
  5. Intraday Seasonalities and Nonstationarity of Trading Volume in Financial Markets: Individual and Cross-Sectional Features (2016) — link
  6. Volume Prediction With Neural Networks (2019) — link
  7. Quintet Volume Projection (2019) — link
  8. Intraday volume percentages forecasting using a dynamic SVM-based approach (2016) — link
and 11 more
  1. DeepLOB: Deep Convolutional Neural Networks for Limit Order Books (2019) — link
  2. Probabilistic Time Series Forecasting with Transformers (2022) — link
  3. Transformers for Limit Order Books (2020) — link
  4. A Transformer-based Framework for Multivariate Time Series Representation Learning (2020) — link
  5. Predicting Intraday Trading Volume and Volume Percentages (2014) — link
  6. Predicting Daily Trading Volume via Various Hidden States (2021) — link
  7. Temporal mixture ensemble models for intraday volume forecasting in cryptocurrency exchange markets (2020) — link
  8. Improving Predictive Ability of Financial Ratios: The Application of Log Transformation ( 2023 ) — link
  9. Optimal execution of portfolio transactions (2001)
  10. Attention Is All You Need (2023) — link
  11. Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles (2017) — link

0 of 19 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.