Paper: arXiv 2509.16137

Authors: Ruslan Tepelyan

Abstract

OHLC bar data is a widely used format for representing financial asset prices over time due to its balance of simplicity and informativeness. Bloomberg has recently introduced a new bar data product that includes additional timing information-specifically, the timestamps of the open, high, low, and close prices within each bar. In this paper, we investigate the impact of incorporating this timing data into machine learning models for predicting volume-weighted average price (VWAP). Our experiments show that including these features consistently improves predictive performance across multiple ML architectures. We observe gains across several key metrics, including log-likelihood, mean squared error (MSE), \(R^2\), conditional variance estimation, and directional accuracy.

Complexity vs Empirical Score

  • Math Complexity: 4.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper employs standard machine learning models with conventional statistical metrics (MSE, R², log-likelihood), showing moderate math complexity, while the rigorous dataset (Russell 3000, 2021), temporal splits, and detailed filtering criteria demonstrate strong empirical backtest-ready implementation.

Research Flowchart

  flowchart TD
  A["Research Goal: Does OHLC<br>timing data improve<br>VWAP prediction?"] --> B{"Methodology"}
  B --> C["Data: Bloomberg OHLC Bar Data<br>with Timing Features"]
  B --> D["Preprocessing & Feature Engineering"]
  C --> D
  D --> E{"ML Training & Evaluation"}
  E --> F["Multiple Architectures<br>e.g., Regression, Trees, NN"]
  F --> G["Key Findings & Outcomes"]
  G --> H["Improved Predictive Performance<br>across all metrics"]
  G --> I["Metrics: Log-Likelihood, MSE,<br>R², Directional Accuracy"]

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 adding intra‑bar timestamp features to OHLC data for machine‑learning models that predict VWAP, showing consistent performance gains across architectures.

Source sentence
Our experiments show that including these features consistently improves predictive performance across multiple ML architectures.

Assumes 2

  • Only trades executed during regular U.S. market hours and using the default set of trade condition codes are included. data
    Source sentence
    included only trades executed during regular U.S. market hours and subject to the default set of trade condition codes.
  • Bars with insufficient trading activity, low price or low tick counts are filtered out to ensure data quality. data
    Source sentence
    We removed bars in which the price or number of ticks fell below minimum thresholds, bars for which the target variable was unavailable, and bars with insufficient data or variation in their recent historical values.

Data

  • Bloomberg OHLC bar data with timing features · Bloomberg market-prices commercial
    2021 calendar year · Russell 3000® Index constituents
    Source sentence
    For our analysis, we used Bloomberg’s intraday service to generate OHLC bar data for all members of the Russell 3000® Index during the 2021 calendar year.

Recognised sources: Bloomberg

Costs

Not applicable The paper reports no trading or execution result.

Finds 1

  • Adding timestamp features improves all reported metrics (log‑likelihood, MSE, R², conditional variance estimation, directional accuracy). performance
    Source sentence
    We observe gains across several key metrics, including log-likelihood, mean squared error (MSE), R2, conditional variance estimation, and directional accuracy.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 5 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.