Paper: arXiv 2504.13521

Authors: Kasymkhan Khubiev, Mikhail Semenov

Abstract

Algorithmic trading relies on extracting meaningful signals from diverse financial data sources, including candlestick charts, order statistics on put and canceled orders, traded volume data, limit order books, and news flow. While deep learning has demonstrated remarkable success in processing unstructured data and has significantly advanced natural language processing, its application to structured financial data remains an ongoing challenge. This study investigates the integration of deep learning models with financial data modalities, aiming to enhance predictive performance in trading strategies and portfolio optimization. We present a novel approach to incorporating limit order book analysis into algorithmic trading by developing embedding techniques and treating sequential limit order book snapshots as distinct input channels in an image-based representation. Our methodology for processing limit order book data achieves state-of-the-art performance in high-frequency trading algorithms, underscoring the effectiveness of deep learning in financial applications.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced deep learning architectures (CNNs, LSTMs, graph filters) and novel embedding techniques for high-frequency limit order book data, indicating significant mathematical complexity. It reports backtesting on real cryptocurrency data with specific performance metrics like MAPE and Sharpe ratio, demonstrating high empirical rigor.

Research Flowchart

  flowchart TD
  A["Research Goal: Apply Deep Learning<br>to Financial Data Modalities"] --> B["Key Data Sources<br>Limit Order Books, Order Flow"]
  B --> C["Novel Methodology<br>Embedding + Image-Based Representation"]
  C --> D["Computational Process<br>Deep Learning Model Training"]
  D --> E["Key Finding: SOTA Performance<br>in High-Frequency Trading"]
  E --> F["Outcome: Enhanced Predictive<br>Signals for Portfolio Optimization"]

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 novel embedding techniques for limit order book data and a high‑frequency trading strategy based on those embeddings.

Source sentence
We present a novel approach to incorporating limit order book analysis into algorithmic trading by developing embedding techniques and treating sequential limit order book snapshots as distinct input channels in an image‑based representation.

Assumes 4

  • Orders of the minimal size are assumed to be executed with certainty. execution
    Source sentence
    The main limitation is that we expect the order to be executed for sure because the order size is the minimum available
  • Ask and bid prices are scaled using min‑max normalization relative to the mid‑price. data
    Source sentence
    We implement asks and bids prices scaling regarding to a specific LOB mid‑price accordingly
  • A sliding window of 30 consecutive LOB snapshots is used for forecasting and trading decisions. model
    Source sentence
    To perform the AMM trading strategy, we use sliding window with a size of 30 aggregation method for stacked and merged LOB snapshots
  • Forecast horizon is set to 1 second for mid‑price prediction. model
    Source sentence
    Having a sequence of N 2‑dimensional snapshots at time t we predict the mid‑price value or trend direction for a time τ = t + ∆t, having ∆t equal to 1 second

Data

Backtest sample: BTCUSDT and ETHUSDT futures · crypto · intraday

  • LOB snapshots of 11 base cryptocurrencies (BTC, ETH, etc.) order-book public
    BTC, ETH and other crypto pairs from OKeX, Binance, Huobi, Kraken
    Source sentence
    For the research, the authors used dataset of 11 unique base cryptocurrencies including BTC and ETH and trading pairs from four cryptoexchanges: OKeX, Binance, Huobi, and Kraken
  • BTC LOB dataset on Kaggle · Kaggle order-book public
    BTC/USDT
    Source sentence
    Data availability The used dataset is published and accessible on Kaggle platform: https://www.kaggle.com/datasets/kkhubiev/btc-lob/data

Recognised sources: Binance

Costs

Partial Costs appear only as a robustness check, a flat fee without spread or impact, or outside the headline numbers.

The strategy focuses on earning with negative fees and assumes AMM execution costs are close to zero.

Source sentence
strategy mainly focuses on earning with negative fees and having AMM execution costs close to zero.

Finds 4

Reported performance (CNN2LSTM trading strategy on BTCUSDT and ETHUSDT futures): Sharpe 7.89

  • The proposed LOB processing methodology achieves state‑of‑the‑art performance in high‑frequency trading algorithms. performance
    Source sentence
    Our methodology for processing limit order book data achieves state‑of‑the‑art performance in high‑frequency trading algorithms
  • Using a one‑shot setting that stacks LOB snapshots of two assets reduces MAPE by roughly a factor of four for short horizons. predictive-accuracy
    Source sentence
    The one‑shot setting helps to significantly reduce the error value approximately make it 4 times less in comparison with the single setting in case of the 0.2s and 1s forecasting horizons
  • The CNN2LSTM‑based trading strategy on BTC and ETH futures shows strong performance with a positive spread and sufficient volume to profit from negative fees. performance
    Source sentence
    The strategy demonstrates strong performance, having spread earning slightly positive, while generating sufficient trading volume to earn on negative fees
  • A SimpleCNN 2D model with merged LOBs and price‑delta regression performed poorly, yielding low trading volume and a tendency to hold long positions. negative
    Source sentence
    The SimpleCNN 2D with merged LOBs and price delta regression task had poor performance. The model tended to repeat the price trend, it had low trading volume and tendency to hold long position

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 29 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.