Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beyond Trend Following: Deep Learning for Market Trend Prediction

Fernando Berzal, Alberto Garcia

TL;DR

Trend-following strategies are often backward-looking and struggle to predict regime shifts. This paper argues that context-rich supervised ML and deep learning, trained on hundreds of leading indicators, can produce nonlinear risk indicators and forward trend predictions, with AutoML-driven hyperparameter optimization and walk-forward validation to ensure out-of-sample robustness. It provides a comprehensive treatment of data inputs, feature engineering, network architectures (including RNNs, CNNs, and transformers), evaluation, and XAI considerations, plus practical risk-on/risk-off and systematic allocation strategies. The work highlights the potential for improved responsiveness to market changes and better risk-adjusted performance for asset managers who retain human oversight while leveraging advanced ML risk indicators.

Abstract

Trend following and momentum investing are common strategies employed by asset managers. Even though they can be helpful in the proper situations, they are limited in the sense that they work just by looking at past, as if we were driving with our focus on the rearview mirror. In this paper, we advocate for the use of Artificial Intelligence and Machine Learning techniques to predict future market trends. These predictions, when done properly, can improve the performance of asset managers by increasing returns and reducing drawdowns.

Beyond Trend Following: Deep Learning for Market Trend Prediction

TL;DR

Trend-following strategies are often backward-looking and struggle to predict regime shifts. This paper argues that context-rich supervised ML and deep learning, trained on hundreds of leading indicators, can produce nonlinear risk indicators and forward trend predictions, with AutoML-driven hyperparameter optimization and walk-forward validation to ensure out-of-sample robustness. It provides a comprehensive treatment of data inputs, feature engineering, network architectures (including RNNs, CNNs, and transformers), evaluation, and XAI considerations, plus practical risk-on/risk-off and systematic allocation strategies. The work highlights the potential for improved responsiveness to market changes and better risk-adjusted performance for asset managers who retain human oversight while leveraging advanced ML risk indicators.

Abstract

Trend following and momentum investing are common strategies employed by asset managers. Even though they can be helpful in the proper situations, they are limited in the sense that they work just by looking at past, as if we were driving with our focus on the rearview mirror. In this paper, we advocate for the use of Artificial Intelligence and Machine Learning techniques to predict future market trends. These predictions, when done properly, can improve the performance of asset managers by increasing returns and reducing drawdowns.
Paper Structure (16 sections, 5 figures)

This paper contains 16 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Investment Philosophy
  3. Risk Indicators
  4. The Limits of Linear Models
  5. On the Use of Risk Indicators by Fund Managers
  6. Model Training for Market Trend Prediction
  7. Machine Learning
  8. Machine Learning Techniques
  9. Model Output
  10. Input Variables
  11. Feature Engineering
  12. Model Hyperparameters
  13. A Configuration Nightmare: AutoML to the Rescue
  14. Model Evaluation
  15. Inside the Black Box: The Limits of XAI Techniques
  16. ...and 1 more sections

Figures (5)

  • Figure 1: The limits of a linear indicator: The linear risk indicator (in blue) is unable to react quickly to a market shock, as happened at the start of the 2020 pandemic. In contrast, a non-linear risk indicator (in red) is much more reactive.
  • Figure 2: ACCI S&P 500 risk indicator. Source: ACCI Wealth Technologies, https://www.acciwealth.com/.
  • Figure 6: Machine Learning: Learning from data using Artificial Intelligence.
  • Figure 7: Correlation matrix for 101 alphas, from fernandez2022: Indicators employed by algorithmic trading strategies often exhibit some correlation between them.
  • Figure 8: Walk-forward cross validation, from fernandez2022: Models are trained on historical data (in blue) and tested using test sets (in red) that always posterior to the data used to train them.