Paper: arXiv 2412.18202

Authors: Zhuohuan Hu, Richard Yu, Zizhou Zhang, Haoran Zheng, Qianying Liu, Yining Zhou

Abstract

This paper leverages machine learning algorithms to forecast and analyze financial time series. The process begins with a denoising autoencoder to filter out random noise fluctuations from the main contract price data. Then, one-dimensional convolution reduces the dimensionality of the filtered data and extracts key information. The filtered and dimensionality-reduced price data is fed into a GANs network, and its output serve as input of a fully connected network. Through cross-validation, a model is trained to capture features that precede large price fluctuations. The model predicts the likelihood and direction of significant price changes in real-time price sequences, placing trades at moments of high prediction accuracy. Empirical results demonstrate that using autoencoders and convolution to filter and denoise financial data, combined with GANs, achieves a certain level of predictive performance, validating the capabilities of machine learning algorithms to discover underlying patterns in financial sequences. Keywords - CNN;GANs; Cryptocurrency; Prediction.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)

Why this score: The paper employs advanced deep learning architectures (GANs, CNNs, autoencoders) with mathematical complexity evident in the adversarial training objective and network design, while demonstrating strong empirical rigor through detailed data sourcing from Binance, specific train/validation/test splits, and cross-validation procedures.

Research Flowchart

  flowchart TD
  A["Research Goal: Forecast Cryptocurrency Price Fluctuations"] --> B["Input: Raw Crypto Price Time Series Data"]
  B --> C["Step 1: Denoising Autoencoder"]
  C --> D["Step 2: 1D CNN<br>Dimensionality Reduction & Feature Extraction"]
  D --> E["Step 3: GANs Network<br>Feature Generation & Pattern Detection"]
  E --> F["Step 4: Fully Connected Network<br>Classification & Prediction"]
  F --> G{"Cross-Validation & Training"}
  G --> H["Outcome: Real-time Trading Strategy<br>Validated Predictive Performance"]
  H --> I["Outcome: Discovery of Underlying Financial Patterns"]

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 novel hybrid cryptocurrency trading strategy that integrates Autoencoders, CNNs, and GANs for predicting price movements and executing trades.

Source sentence
This paper proposes a novel cryptocurrency trading strategy that leverages a hybrid approach combining Autoencoder, CNN and GANs

Assumes 2

  • The model is trained and evaluated on historical price data without accounting for market impact or slippage. execution
    Source sentence
    Positions are held for 10 time units or until a stop-loss threshold is met.
  • Data are split chronologically with training up to December 2023 and testing from January 2024 to October 2024. data
    Source sentence
    Data from October 2019 to December 2023 is used for training and validation, while data from January 2024 to October 2024 is reserved for testing.

Data

Backtest sample: 2019-10 – 2024-10 · BTC perpetual futures · crypto

  • BTC perpetual futures contract price data · Binance market-prices public
    Oct 2019 to Oct 2024 · BTC perpetual futures
    Source sentence
    This study uses data that is made up of BTC perpetual contract price data collected every 10 minutes from October 2019 to October 2024. The Binance API was used to obtain the initial and final prices, peak and bottom prices, along with trading volume, for analysis.

Costs

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

Finds 5

Reported performance (Autoencoder-CNN-GAN hybrid model): Sharpe 2.50 · total return 120.0% · max drawdown 15.0%

  • The model achieved a prediction accuracy of 61.2%, outperforming baseline methods. predictive-accuracy
    Source sentence
    The model achieved a prediction accuracy of 61.2%, outperforming three baseline models, including traditional statistical approaches and machine learning models like LSTM and standalone CNNs.
  • Starting with an initial capital of 1, the strategy yielded a net value of approximately 1,370,000, a compounded return of 120% over five years. performance
    Source sentence
    Starting with an initial capital of 1, the strategy yielded a net value of approximately 1,370,000 by the end of the testing period. This represents a compounded return of 120% over five years.
  • Maximum drawdown was kept within an acceptable range of 15%, indicating robustness. performance
    Source sentence
    Maximum Drawdown: The model maintained a maximum drawdown within an acceptable range of 15%, showcasing its robustness in adverse market conditions.
  • The strategy produced a Sharpe Ratio of 2.5, indicating favorable risk-adjusted returns. performance
    Source sentence
    Sharpe Ratio: The strategy produced a Sharpe Ratio of 2.5, indicating a favorable risk-adjusted return compared to baseline trading strategies.
  • Training time for the hybrid model was approximately 6 hours on an NVIDIA Tesla V100 GPU. performance
    Source sentence
    The training time for the proposed model was approximately 6 hours on a system using an NVIDIA Tesla V100 GPU.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 61 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive

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.