Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Supervised Autoencoders with Fractionally Differentiated Features and Triple Barrier Labelling Enhance Predictions on Noisy Data

Bartosz Bieganowski, Robert Ślepaczuk

TL;DR

The paper addresses enhancing risk-adjusted cryptocurrency trading performance by combining Supervised Autoencoder-MLP (SAE-MLP) with triple barrier labeling (TBL) and fractionally differentiated features. It introduces a walk-forward framework for adaptive fractional differencing and a TBL-specific optimization metric to guide label selection, coupled with volatility-based data augmentation to improve robustness. Across Bitcoin, Ethereum, and Litecoin, SAE-MLP strategies frequently outperform buy-and-hold on information ratio metrics, particularly at 20–30 minute horizons, while controlling drawdown and volatility. The findings demonstrate the practical potential of SAE-based de-noising and TBL in algorithmic trading, with implications for asset managers and regulators, while noting limitations such as the exclusion of slippage and market impact in the evaluation.

Abstract

This paper investigates the enhancement of financial time series forecasting with the use of neural networks through supervised autoencoders (SAE), to improve investment strategy performance. Using the Sharpe and Information Ratios, it specifically examines the impact of noise augmentation and triple barrier labeling on risk-adjusted returns. The study focuses on Bitcoin, Litecoin, and Ethereum as the traded assets from January 1, 2016, to April 30, 2022. Findings indicate that supervised autoencoders, with balanced noise augmentation and bottleneck size, significantly boost strategy effectiveness. However, excessive noise and large bottleneck sizes can impair performance.

Supervised Autoencoders with Fractionally Differentiated Features and Triple Barrier Labelling Enhance Predictions on Noisy Data

TL;DR

The paper addresses enhancing risk-adjusted cryptocurrency trading performance by combining Supervised Autoencoder-MLP (SAE-MLP) with triple barrier labeling (TBL) and fractionally differentiated features. It introduces a walk-forward framework for adaptive fractional differencing and a TBL-specific optimization metric to guide label selection, coupled with volatility-based data augmentation to improve robustness. Across Bitcoin, Ethereum, and Litecoin, SAE-MLP strategies frequently outperform buy-and-hold on information ratio metrics, particularly at 20–30 minute horizons, while controlling drawdown and volatility. The findings demonstrate the practical potential of SAE-based de-noising and TBL in algorithmic trading, with implications for asset managers and regulators, while noting limitations such as the exclusion of slippage and market impact in the evaluation.

Abstract

This paper investigates the enhancement of financial time series forecasting with the use of neural networks through supervised autoencoders (SAE), to improve investment strategy performance. Using the Sharpe and Information Ratios, it specifically examines the impact of noise augmentation and triple barrier labeling on risk-adjusted returns. The study focuses on Bitcoin, Litecoin, and Ethereum as the traded assets from January 1, 2016, to April 30, 2022. Findings indicate that supervised autoencoders, with balanced noise augmentation and bottleneck size, significantly boost strategy effectiveness. However, excessive noise and large bottleneck sizes can impair performance.

Paper Structure

This paper contains 16 sections, 7 equations, 13 figures, 2 algorithms.

Table of Contents

  1. Data
  2. Feature Engineering
  3. Motivation
  4. Fractionally Differentiated Features
  5. Optimal Fractional Differentiation Order
  6. Implementation
  7. Labelling Methodology
  8. TBL-Optimized Performance Metric
  9. Model Training Considerations
  10. Data Augmentation
  11. Supervised Autoencoder MLP
  12. Walk-forward Validation
  13. Construction of equity line
  14. Performance Metrics
  15. Results
  16. ...and 1 more sections

Figures (13)

  • Figure 1: Logarithmic price index for traded currencies.
  • Figure 2: Triple-barrier-labelling visualization.
  • Figure 3: Supervised autoencoder structure.
  • Figure 4: Walk-forward validation procedure. The training set, initially expanding, is limited to a 3-period length, therefore shifting instead of expanding since split 4.
  • Figure 5: Equity value index for Bitcoin strategies.
  • ...and 8 more figures