Please rotate your device to landscape mode to view the charts.

(Re-)imag(in)ing Price Trends

Journal: Journal of Finance

Date: 20231201

Author: Jiang, Jingwen; Kelly, Bryan; Xiu, Dacheng

Abstract:
We reconsider trend-based predictability by employing flexible learning methods to identify price patterns that are highly predictive of returns, as opposed to testing predefined patterns like momentum or reversal. Our predictor data are stock-level price charts, allowing us to extract the most predictive price patterns using machine learning image analysis techniques. These patterns differ significantly from commonly analyzed trend signals, yield more accurate return predictions, enable more profitable investment strategies, and demonstrate robustness across specifications. Remarkably, they exhibit context independence, as short-term patterns perform well on longer time scales, and patterns learned from U.S. stocks prove effective in international markets.

Link: Google Scholar

Key Findings

Superior Performance of CNN-Based Image Analysis

Image-based CNN predictions outperform traditional trend signals with annualized Sharpe ratios up to 7.2 for equal-weight portfolios

Successful International Transfer Learning

CNN models trained on U.S. data transfer effectively to international markets, outperforming locally trained models in 21 out of 26 countries

Robust Time-Scale Transfer

Models trained on high-frequency data successfully predict returns at lower frequencies, demonstrating the fractal nature of price patterns

Weekly Strategy Performance Comparison

  • CNN strategies achieve Sharpe ratios of 7.2, 6.8, and 4.9 for 5-, 20-, and 60-day images respectively
  • Traditional strategies like WSTR and TREND achieve lower Sharpe ratios of 2.8 and 2.9
  • CNN strategies maintain superior performance even in value-weighted portfolios

International Transfer Learning Success

  • Transfer learning improves performance in 80% of international markets
  • Average Sharpe ratio increases from 2.3 (local training) to 3.6 (transfer) for equal-weight portfolios
  • Benefits are particularly strong for markets with fewer stocks

Time-Scale Transfer Performance

  • Time-scale transfer achieves 2.1 Sharpe ratio for 20-day forecasts
  • Performance comparable to direct training (2.2 Sharpe ratio)
  • Demonstrates pattern self-similarity across time scales

Contribution and Implications

  • Introduces novel approach combining CNN with price chart images for return prediction
  • Demonstrates the universal nature of price patterns across markets and time scales
  • Provides practical framework for applying deep learning to technical analysis

Data Sources

  • Weekly strategy performance chart based on Table I performance metrics
  • International transfer results derived from Table X comparing local vs transfer performance
  • Time-scale transfer visualization constructed using data from Table XI equal-weight portfolio results