S&P 500 Stock Movement Prediction with CNN
Published:Dec 25, 2025 23:10
•1 min read
•ArXiv
Analysis
This paper explores stock movement prediction using a Convolutional Neural Network (CNN) on multivariate raw data, including stock split/dividend events, unlike many existing studies that use engineered financial data or single-dimension data. This approach is significant because it attempts to model real-world market data complexity directly, potentially leading to more accurate predictions. The use of CNNs, typically used for image classification, is innovative in this context, treating historical stock data as image-like matrices. The paper's potential lies in its ability to predict stock movements at different levels (single stock, sector-wise, or portfolio) and its use of raw, unengineered data.
Key Takeaways
Reference
“The model achieves promising results by mimicking the multi-dimensional stock numbers as a vector of historical data matrices (read images).”