18th IIAI International Congress on Advanced Applied Informatics, pp. 780-787, July 17, 2025
3rd International Conference on Computational and Data Sciences in Economics and Finance (CDEF 2025) in 18th IIAI International Congress on Advanced Applied Informatics (IIAI AAI 2025)
Financial time-series are essential for evaluating trading strategies in financial markets; however, historical data alone are often insufficient for comprehensive assessments. Recent studies increasingly emphasize generating financial time-series using generative models to capture potential future scenarios. Although volatility is a crucial factor in strategy evaluation, previous studies have largely overlooked estimating volatility in generated data. To address this gap, we propose a conditional diffusion model that simultaneously generates financial price sequences and estimates volatility. Similar to previous methods, our method includes forward (diffusion) and reverse (denoising) processes. However, in the reverse process of our method, each step not only denoises the data but also estimates the volatility of the final output. Validation experiments using simulated data evaluate the performance of our method in data generation and volatility estimation. The results demonstrate that our method matches previous methods in generation performance and produces volatility estimates that significantly correlate with realized volatility computed from returns.
Diffusion Model; Conditional Generation; Volatility Estimation;
10.1109/IIAI-AAI67470.2025.00143
@inproceedings{Yoshida2025-model-merging,
title={{Conditional Diffusion Model with Volatility Estimation for Financial Time-Series Generation}},
author={Ryoya Yoshida and Masanori Hirano and Kentaro Imajo},
booktitle={18th IIAI International Congress on Advanced Applied Informatics},
isbn={979-8-3315-9937-9},
pages={780-787},
publisher={IEEE},
doi={10.1109/IIAI-AAI67470.2025.00143},
year={2025}
}