Combining LSTM Time-Series and CNN Chart Features for Stock Prediction

Summary

The document summarizes a 2019 study that combines two representations of the same stock data to predict prices: numerical time series and images of stock charts. An LSTM extracts temporal patterns from the sequence data, while a CNN extracts visual patterns from chart images; their features are combined in an LSTM-CNN model. The study uses SPDR S&P 500 ETF data and compares the combined approach with standalone LSTM and CNN models.

According to the summary, the combined model predicts prices better than either individual model, and candlestick charts perform best among the chart image types considered. These findings suggest that combining distinct representations may reduce prediction error in the studied setting. The document provides only a summary, not the paper's experimental details, evaluation metrics, or implementation procedure. Its reported result is specific to the study data and does not establish profitability, robustness across assets, or live trading performance.

Key ideas

• The approach combines numerical price sequences with images derived from the same market data.
• An LSTM extracts time-series features, while a CNN extracts chart-image features.
• The study reports lower prediction error for the combined model than for standalone LSTM or CNN models.
• Candlestick charts were reported as the most suitable image representation in the study.
• The summary does not establish trading profitability or generalization beyond the evaluated data.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.