VAdam : A volatility-aware adaptive gradient optimization framework for financial time-series learning
*Aditya NagdiyaCorresponding authoraditya22.ad@gmail.comDepartment of Information TechnologyInstitute of Engineering & TechnologyDevi Ahilya VishwavidyalayaIndore, Madhya Pradesh, 452017, IndiaView full profile → , Vivek Kapoorvkapoor@ietdavv.edu.inDepartment of Information TechnologyInstitute of Engineering & TechnologyDevi Ahilya VishwavidyalayaIndore, Madhya Pradesh, 452017, IndiaView full profile → , Vrinda Tokekarvrindatokekar@yahoo.comDepartment of Information TechnologyInstitute of Engineering & TechnologyDevi Ahilya VishwavidyalayaIndore, Madhya Pradesh, 452017, IndiaView full profile →
* Corresponding author · click or hover a name for details
- Received:
- 01 Mar 2026
- Published Online:
- 31 Oct 2026
- Article type:
- Research Article
- Language:
- EN
- Article no.:
- JIOS-2486
- Pages:
- 3585–3595
Abstract
Financial time-series forecasting is difficult because of constantly changing market conditions, non-stationary behavior, and sudden changes in volatility which can negatively affect optimization stability during deep neural network training. Existing adaptive optimization methods such as Adam and its variants employ fixed optimization dynamics that do not directly respond to changing market conditions, which may lead to inefficient learning under varying volatility regimes. This study proposes VAdam, a Volatility-Aware Adaptive Gradient Optimization Framework for financial time-series learning. The proposed method builds on the conventional Adam optimizer by incorporating a volatility-conditioned update mechanism that dynamically adjusts learning rate and momentum behavior according to normalized market volatility. The method was evaluated on real cryptocurrency datasets obtained from Yahoo Finance, including Bitcoin, Ethereum, Solana, XRP, and Cardano, using a controlled optimizer-isolation benchmark with a shared MLP forecasting model. A comparison with SGD, RMSProp, Adam, AdamW, and Nadam demonstrates that VAdam achieves modest but consistent improvements over standard Adam, including 0.65% lower mean MSE, 0.67% lower price sMAPE, and 1.28% fewer epochs to convergence. The largest improvement was observed for XRP, where VAdam reduced MSE by 4.65% relative to Adam. The findings suggest that volatility-aware optimization can improve Adam-family optimization behavior while maintaining a simple drop-in optimization framework for financial forecasting tasks.
Keywords
Subject Classifications
References
[1] R. F. Engle, “Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation,” Econometrica, vol. 50, no. 4, pp. 987–1007 (1982), doi: 10.2307/1912773.
[2] T. Bollerslev, “Generalized autoregressive conditional heteroskedasticity,” J. Econometrics, vol. 31, no. 3, pp. 307–327 (1986), doi: 10.1016/0304-4076(86)90063-1.
[3] F. Corsi, “A simple approximate long-memory model of realized volatility,” J. Financ. Econometrics, vol. 7, no. 2, pp. 174–196 (2009), doi: 10.1093/jjfinec/nbp001.
[4] B. Lim and S. Zohren, “Time-series forecasting with deep learning: A survey,” Philos. Trans. R. Soc. A Math. Phys. Eng. Sci., vol. 379, no. 2194, Art. no. 20200209 (2021), doi: 10.1098/rsta.2020.0209.
[5] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 30, pp. 5998–6008 (2017).
[6] B. Lim, S. Ö. Arik, N. Loeff, and T. Pfister, “Temporal fusion transformers for interpretable multi-horizon time series forecasting,” Int. J. Forecast., vol. 37, no. 4, pp. 1748–1764 (2021), doi: 10.1016/j.ijforecast.2021.03.012.
[7] M. M. Patel, S. Tanwar, R. Gupta, and N. Kumar, “A deep learning-based cryptocurrency price prediction scheme for financial institutions,” J. Inf. Secur. Appl., vol. 55, Art. no. 102583 (2020), doi: 10.1016/j.jisa.2020.102583.
[8] M. Saad, J. Choi, D. Nyang, J. Kim, and A. Mohaisen, “Toward characterizing blockchain-based cryptocurrencies for highly accurate predictions,” IEEE Syst. J., vol. 14, no. 1, pp. 321–332 (2020).
[9] R. Kaur, M. Uppal, D. Gupta, S. Juneja, S. Y. Arafat, J. Rashid, J. Kim, and R. Alroobaea, “Development of a cryptocurrency price prediction model: Leveraging GRU and LSTM for Bitcoin, Litecoin and Ethereum,” PeerJ Comput. Sci., vol. 11, Art. no. e714 (2025).
[10] W. Badar, S. Ramzan, A. Raza, N. L. Fitriyani, M. Syafrudin, and S. W. Lee, “Enhanced interpretable forecasting of cryptocurrency prices using autoencoder features and a hybrid CNN-LSTM model,” Mathematics, vol. 13, no. 12, Art. no. 1908 (2025), doi: 10.3390/math13121908.
[11] D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in Proc. Int. Conf. Learn. Represent. (ICLR), (2015).
[12] T. Dozat, “Incorporating Nesterov momentum into Adam,” in Proc. Int. Conf. Learn. Represent. (ICLR) Workshop, (2016).
[13] I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” in Proc. Int. Conf. Learn. Represent. (ICLR), (2019).
[14] L. Luo, Y. Xiong, Y. Liu, and X. Sun, “Adaptive gradient methods with dynamic bound of learning rate,” in Proc. Int. Conf. Learn. Represent. (ICLR), (2019).
[15] J. Zhuang, T. Tang, Y. Ding, S. Tatikonda, N. Dvornek, X. Papademetris, and J. S. Duncan, “AdaBelief optimizer: Adapting stepsizes by the belief in observed gradients,” in Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 33, pp. 18795–18806 (2020).
[16] L. Liu, H. Jiang, P. He, W. Chen, X. Liu, J. Gao, and J. Han, “On the variance of the adaptive learning rate and beyond,” in Proc. Int. Conf. Learn. Represent. (ICLR), (2020).
[17] S. J. Reddi, S. Kale, and S. Kumar, “On the convergence of Adam and beyond,” in Proc. Int. Conf. Learn. Represent. (ICLR), (2018).
[18] A. Samanta and S. Williamson, “Adaptive learning guided by bias-noise-alignment diagnostics,” arXiv:2512.24445 (2025).
[19] D. Dhabliya, A. Lavhale, S. Thakur, R. M. Gomathi, A. Kumar, and S. De, “Enhancing cryptographic security through zero-knowledge proofs in theoretical mathematics,” J. Discrete Math. Sci. Cryptogr., vol. 29, no. 2-B, pp. 913–921 (2026), doi: 10.47974/JDMSC-2542.
