S&P 500 Candlestick Images For Six-Class Return Prediction

501 S&P 500 constituent stocks are represented by 1,374,694 candlestick chart images from January 2010 to March 2025. Each image is labeled with a 6-class forward return category based on price movement thresholds.

Use Cases

• Train a convolutional neural network on candlestick images to classify the 6-class forward return label.
• Analyze the relationship between visual chart patterns and the ±1%/±2%/±3% return categories.
• Develop a multi-modal model combining candlestick images with technical indicator overlays for price movement prediction.
• Benchmark image classification architectures on a large-scale financial dataset of 1.37 million images.

Strengths

• Contains 1,374,694 total images for model training and validation.
• Covers 501 individual S&P 500 stocks over a 15-year period from 2010-2025.
• Provides a structured 6-class prediction task based on forward price returns.

Limitations

• The dataset is image-based, requiring computer vision expertise instead of direct tabular analysis of OHLCV data.
• Labels are derived from forward returns, which may not capture all relevant market factors for prediction.

Provenance

Source

Hugging Face dataset by author sogosonnet.

Collection Method

null

Time Range

January 2010 to March 2025.

Freshness

Data extends to March 2025, with a platform update recorded in March 2026.

Geography

United States (S&P 500 constituents).