aiqbench analysis
This paper advances sequence modeling for financial time series by substituting quadratic Transformer encoders with linear-time Mamba state space models, enabling the processing of significantly longer high-frequency contexts without saturation. The hybrid HAN-Mamba architecture achieves improved volatility forecasting accuracy and reduced parameter count while maintaining a learned attention mechanism specifically for cross-scale integration. This work demonstrates that selective state space networks can effectively capture persistent memory and abrupt regime shifts in market data, offering a more scalable alternative to standard attention mechanisms for quantitative finance applications.
HAN-Mamba: Hierarchical Selective State Space Networks for Multi-Scale Financial Volatility Forecasting
arXiv:2610.10323v1 Announce Type: new Abstract: Short-horizon realized volatility forecasting requires the integration of market information that evolves at incompatible temporal resolutions, from second-level order book dynamics to weekly regime drift. Our conference work introduced HAN-T, a hierarchical architecture in which scale-specific Transformer encoders process short, mid, and long-horizon streams and a learned attention fuser weighs their contributions. This article replaces the quadratic attention encoders with selective state space (Mamba) encoders while retaining attention only in the fuser, where the input is a three-token set rather than a long sequence. The resulting hybrid, HAN-Mamba, summarizes each stream through a recurrent state whose input-dependent gating matches two structural properties of volatility: persistent but decaying memory and abrupt regime shifts. On the Optiver Realized Volatility Prediction benchmark under time-aware five-fold cross-validation, HAN-Mamba improves mean RMSPE over HAN-T (0.1942 vs. 0.1965) with 33% fewer parameters. Its linear-time encoders further allow the high-frequency context to be extended from 60 to 240 buckets, reducing error to 0.1927 where the attention variant saturates, and support constant-time streaming updates at inference. Ablations attribute the gains to the encoder swap, confirm that the hierarchical prior transfers across sequence-model families, and show that the permutation-invariant attention fuser remains the correct mechanism for cross-scale integration.
Abstract:Short-horizon realized volatility forecasting requires the integration of market information that evolves at incompatible temporal resolutions, from second-level order book dynamics to weekly regime drift. Our conference work introduced HAN-T, a hierarchical architecture in which scale-specific Transformer encoders process short, mid, and long-horizon streams and a learned attention fuser weighs their contributions. This article replaces the quadratic attention encoders with selective state space (Mamba) encoders while retaining attention only in the fuser, where the input is a three-token set rather than a long sequence. The resulting hybrid, HAN-Mamba, summarizes each stream through a recurrent state whose input-dependent gating matches two structural properties of volatility: persistent but decaying memory and abrupt regime shifts. On the Optiver Realized Volatility Prediction benchmark under time-aware five-fold cross-validation, HAN-Mamba improves mean RMSPE over HAN-T (0.1942 vs. 0.1965) with 33% fewer parameters. Its linear-time encoders further allow the high-frequency context to be extended from 60 to 240 buckets, reducing error to 0.1927 where the attention variant saturates, and support constant-time streaming updates at inference. Ablations attribute the gains to the encoder swap, confirm that the hierarchical prior transfers across sequence-model families, and show that the permutation-invariant attention fuser remains the correct mechanism for cross-scale integration.
16 pages. Accepted for publication in Springer Lecture Notes in Artificial Intelligence (ICAART 2026 Revised Selected Papers). Extended version of the ICAART 2026 paper (DOI: https://doi.org/10.5220/0014264900004052)
Machine Learning (cs.LG); Computational Engineering, Finance, and Science (cs.CE)
arXiv:2610.10323 [cs.LG]
(or arXiv:2610.10323v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.10323 arXiv-issued DOI via DataCite (pending registration)
Full text under the publisher's open licence · 249 words
Related coverage
SeisMamba: Low-Latency Single-Station Seismic Magnitude Estimation for Spatially Distributed Earthquake Early Warning
arXiv · Aug 26, 2026
News
Quantum Computing Report · Jul 25, 2026
SFMambaNet: Spectral-Frequency Enhanced Selective State Space Model for Correspondence Pruning
arXiv · Jun 4, 2026
When Attention Beats Fourier: Multi-Scale Transformers for PDE Solving on Irregular Domains
arXiv · Jun 6, 2026