InterDiff: Synthesizing Financial Time Series with Inter-Stock Correlations via Classifier-Free Guided Diffusion

Abstract

Stock prediction is hindered by data scarcity, and although existing data augmentation techniques have made significant strides, they often overlook the dynamic inter-stock interactions crucial for robust modeling. To address these challenges, we propose InterDiff , a diffusion-based framework that synthesizes realistic financial time series by dynamically modeling both intra- and inter-stock correlations. InterDiff employs hierarchical transformers to learn these correlations, encoding them into a guidance vector that steers a diffusion model via classifier-free guidance. This approach ensures that the synthetic data preserves fidelity while introducing controlled variability. Evaluations on CSI300 and CSI800 show that models trained on InterDiff-augmented data boost the information coefficient by 1.13–4.70% on CSI300 and 40.15–49.60% on CSI800, while delivering cumulative return improvements of 0.57–13.87% on CSI300 and 28.72–51.33% on CSI800 under 0.1% per-trade cost. The framework outperforms alternatives such as DiffsFormer and Quant GAN. Ablation studies reveal a fidelity-diversity tradeoff: while larger guidance strength improves synthetic data fidelity, it does not necessarily enhance prediction performance. Visualizations confirm the preservation of inter-stock correlations and a reduction in overfitting. These results demonstrate InterDiff’s ability to enhance robustness and profitability in real-world trading environments and mitigate data scarcity.