{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,23]],"date-time":"2026-09-23T07:01:11Z","timestamp":1790146871960,"version":"4.0.1"},"reference-count":43,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,23]],"date-time":"2026-09-23T00:00:00Z","timestamp":1790121600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"<jats:p>\n                    Selective state-space models (SSMs) such as Mamba achieve linear-time sequence modeling through input-dependent recurrence, yet they apply the same dynamics regardless of the latent regime generating the data\u2014a structural mismatch for non-stationary signals whose statistics shift abruptly between regimes. We introduce\n                    <jats:italic>RC-Mamba<\/jats:italic>\n                    , a probabilistic forecaster whose selective-SSM backbone is dynamically conditioned on the posterior over a latent regime variable. Regime posteriors from a Student-\n                    <jats:italic>t<\/jats:italic>\n                    emission Hidden Markov Model with a sticky Dirichlet prior modulate the SSM discretisation step \u0394\n                    <jats:sub>\n                      <jats:italic>t<\/jats:italic>\n                    <\/jats:sub>\n                    together with the input and readout matrices\n                    <jats:bold>B<\/jats:bold>\n                    <jats:sub>\n                      <jats:italic>t<\/jats:italic>\n                    <\/jats:sub>\n                    and\n                    <jats:bold>C<\/jats:bold>\n                    <jats:sub>\n                      <jats:italic>t<\/jats:italic>\n                    <\/jats:sub>\n                    , so the recurrence adapts its memory decay, input sensitivity, and output emphasis to the prevailing market state. An entropy-based scale-modulation component, the Regime Uncertainty Gate, is evaluated as a route from posterior sharpness to interval width, but matched controls do not identify a performance gain for it over posterior-independent scaling. A hierarchical Feature-wise Linear Modulation (FiLM) embedding captures asset-specific and sector-level structure without an explicit asset-class prediction objective. Evaluated on 18 US-listed ETFs spanning five asset classes under five calendar-aligned expanding-window walk-forward folds (test years 2021\u20132025), RC-Mamba achieves a mean Continuous Ranked Probability Score (CRPS) of 0.00532, improving on the Vanilla Temporal Fusion Transformer by 4.7 %, DeepAR by 8.9 %, S-Mamba by 15.2 %, and an HMM-conditioned DeepAR by 18.5 %. All models share robust-MAD preprocessing, a common ceiling of 11,450 optimizer updates with validation-based early stopping, and a 30-business-day origin embargo equal to the maximum forecast horizon, so neither preprocessing nor training budget confounds the architectural comparison. Significance is assessed on 1,110 synchronized forecast origins using a moving-block bootstrap that respects serial and cross-sectional dependence rather than treating overlapping horizons as independent; every comparison remains significant after Holm correction (\n                    <jats:italic>p<\/jats:italic>\n                    <jats:sub>Holm<\/jats:sub>\n                    = 0.006). RC-Mamba also attains the best Winkler\n                    <jats:sub>80<\/jats:sub>\n                    score (0.0364) with PICP\n                    <jats:sub>80<\/jats:sub>\n                    = 0.781 and PINAW\n                    <jats:sub>80<\/jats:sub>\n                    = 0.024, and recovers three interpretable regimes (Calm, Transition, Stress) that assign every one of the 25 observed April 2025 trading days to the Stress state. Matched single-fold ablations show that removing regime conditioning degrades CRPS by 2.05 % in the most volatile test year, although several component contrasts change sign across folds and are therefore reported as design trade-offs rather than identified causal effects. A one-day-lagged long-only backtest over 1,255 out-of-sample trading days returns 8.89 % annualized against 6.56 % for an equal-weight benchmark, but its annualized alpha of 1.18 % is not statistically significant (\n                    <jats:italic>p<\/jats:italic>\n                    = 0.506) and remains insignificant under 1\u201310 basis-point costs; the economic evidence is therefore suggestive rather than conclusive. Transfer to non-financial domains is left to future work.\n                  <\/jats:p>","DOI":"10.3389\/frai.2026.1903497","type":"journal-article","created":{"date-parts":[[2026,9,23]],"date-time":"2026-09-23T06:16:00Z","timestamp":1790144160000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["RC-Mamba: regime-conditioned selective state-space models for probabilistic time-series forecasting"],"prefix":"10.3389","volume":"9","author":[{"given":"Waleed","family":"Soliman","sequence":"first","affiliation":[{"name":"School of Computer Science, University of Nottingham Malaysia","place":["Semenyih, Selangor, Malaysia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhiyuan","family":"Chen","sequence":"additional","affiliation":[{"name":"School of Computer Science, University of Nottingham Malaysia","place":["Semenyih, Selangor, Malaysia"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Colin","family":"Johnson","sequence":"additional","affiliation":[{"name":"School of Computer Science, University of Nottingham","place":["Nottingham, United Kingdom"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sabrina","family":"Wong","sequence":"additional","affiliation":[{"name":"Hy-Alpha Sdn Bhd","place":["Selangor, Malaysia"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,23]]},"reference":[{"key":"B1","doi-asserted-by":"crossref","DOI":"10.3233\/FAIA240677","article-title":"\u201cTimeMachine: a time series is worth 4 mambas for long-term forecasting,\u201d","volume-title":"European conference on artificial intelligence (ECAI)","author":"Ahamed","year":"2024"},{"key":"B2","doi-asserted-by":"publisher","first-page":"1137","DOI":"10.1093\/rfs\/15.4.1137","article-title":"International asset allocation with regime shifts","volume":"15","author":"Ang","year":"2002","journal-title":"Rev. Financ. Stud"},{"key":"B3","doi-asserted-by":"publisher","first-page":"307","DOI":"10.1016\/0304-4076(86)90063-1","article-title":"Generalized autoregressive conditional heteroskedasticity","volume":"31","author":"Bollerslev","year":"1986","journal-title":"J. Econom"},{"key":"B4","doi-asserted-by":"publisher","first-page":"310","DOI":"10.1057\/jam.2010.27","article-title":"Markov-switching asset allocation: do profitable strategies exist?","volume":"12","author":"Bulla","year":"2011","journal-title":"J. Asset Manag"},{"key":"B5","article-title":"\u201cTransformers are SSMs: generalized models and efficient algorithms through structured state space duality,\u201d","volume-title":"International conference on machine learning (ICML)","author":"Dao","year":"2024"},{"key":"B6","doi-asserted-by":"publisher","first-page":"253","DOI":"10.1080\/07350015.1995.10524599","article-title":"Comparing predictive accuracy","volume":"13","author":"Diebold","year":"1995","journal-title":"J. Bus. Econ. Stat"},{"key":"B7","doi-asserted-by":"publisher","first-page":"987","DOI":"10.2307\/1912773","article-title":"Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation","volume":"50","author":"Engle","year":"1982","journal-title":"Econometrica"},{"key":"B8","doi-asserted-by":"publisher","first-page":"359","DOI":"10.1198\/016214506000001437","article-title":"Strictly proper scoring rules, prediction, and estimation","volume":"102","author":"Gneiting","year":"2007","journal-title":"J. Am. Stat. Assoc"},{"key":"B9","doi-asserted-by":"publisher","first-page":"27","DOI":"10.1016\/0304-405X(96)00875-6","article-title":"Modeling the conditional distribution of interest rates as a regime-switching process","volume":"42","author":"Gray","year":"1996","journal-title":"J. Financ. Econ"},{"key":"B10","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2312.00752","article-title":"Mamba: linear-time sequence modeling with selective state spaces","author":"Gu","year":"2023","journal-title":"arXiv"},{"key":"B11","article-title":"\u201cEfficiently modeling long sequences with structured state spaces,\u201d","volume-title":"International conference on learning representations (ICLR)","author":"Gu","year":"2022"},{"key":"B12","doi-asserted-by":"publisher","first-page":"2223","DOI":"10.1093\/rfs\/hhaa009","article-title":"Empirical asset pricing via machine learning","volume":"33","author":"Gu","year":"2020","journal-title":"Rev. Financ. Stud"},{"key":"B13","doi-asserted-by":"publisher","first-page":"3503","DOI":"10.1016\/j.jedc.2006.12.004","article-title":"Asset allocation under multivariate regime switching","volume":"31","author":"Guidolin","year":"2007","journal-title":"J. Econ. Dyn. Control"},{"key":"B14","doi-asserted-by":"publisher","first-page":"357","DOI":"10.2307\/1912559","article-title":"A new approach to the economic analysis of nonstationary time series and the business cycle","volume":"57","author":"Hamilton","year":"1989","journal-title":"Econometrica"},{"key":"B15","doi-asserted-by":"publisher","first-page":"307","DOI":"10.1016\/0304-4076(94)90067-1","article-title":"Autoregressive conditional heteroskedasticity and changes in regime","volume":"64","author":"Hamilton","year":"1994","journal-title":"J. Econom"},{"key":"B16","doi-asserted-by":"publisher","first-page":"1735","DOI":"10.1162\/neco.1997.9.8.1735","article-title":"Long short-term memory","volume":"9","author":"Hochreiter","year":"1997","journal-title":"Neural Comput"},{"key":"B17","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1016\/0304-4076(94)90036-1","article-title":"Dynamic linear models with Markov-switching","volume":"60","author":"Kim","year":"1994","journal-title":"J. Econom"},{"key":"B18","doi-asserted-by":"publisher","first-page":"1748","DOI":"10.1016\/j.ijforecast.2021.03.012","article-title":"Temporal fusion transformers for interpretable multi-horizon time series forecasting","volume":"37","author":"Lim","year":"2021","journal-title":"Int. J. Forecast"},{"key":"B19","article-title":"\u201ciTransformer: inverted transformers are effective for time series forecasting,\u201d","volume-title":"International conference on learning representations (ICLR)","author":"Liu","year":"2024"},{"key":"B20","volume-title":"Advances in Financial Machine Learning","author":"L\u00f3pez de Prado","year":"2018"},{"key":"B21","article-title":"\u201cDecoupled weight decay regularization,\u201d","volume-title":"International conference on learning representations (ICLR)","author":"Loshchilov","year":"2019"},{"key":"B22","first-page":"4765","article-title":"\u201cA unified approach to interpreting model predictions,\u201d","volume-title":"Advances in neural information processing systems (NeurIPS)","author":"Lundberg","year":"2017"},{"key":"B23","doi-asserted-by":"publisher","first-page":"1087","DOI":"10.1287\/mnsc.22.10.1087","article-title":"Scoring rules for continuous probability distributions","volume":"22","author":"Matheson","year":"1976","journal-title":"Manage. Sci"},{"key":"B24","article-title":"\u201cA time series is worth 64 words: long-term forecasting with transformers,\u201d","volume-title":"International conference on learning representations (ICLR)","author":"Nie","year":"2023"},{"key":"B25","article-title":"\u201cN-BEATS: neural basis expansion analysis for interpretable time series forecasting,\u201d","volume-title":"International Conference on Learning Representations (ICLR)","author":"Oreshkin","year":"2020"},{"key":"B26","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2403.15360","article-title":"SiMBA: simplified Mamba-based architecture for vision and multivariate time series","author":"Patro","year":"2024","journal-title":"arXiv"},{"key":"B27","first-page":"7796","article-title":"\u201cDeep state space models for time series forecasting,\u201d","volume-title":"Advances in neural information processing systems (NeurIPS)","author":"Rangapuram","year":"2018"},{"key":"B28","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2310.08278","article-title":"Lag-Llama: towards foundation models for probabilistic time series forecasting","author":"Rasul","year":"2023","journal-title":"arXiv [Preprint]"},{"key":"B29","first-page":"8857","article-title":"\u201cAutoregressive denoising diffusion models for multivariate probabilistic time series forecasting,\u201d","volume-title":"International conference on machine learning (ICML)","author":"Rasul","year":"2021"},{"key":"B30","first-page":"1135","article-title":"\u201c\u2018Why should I trust you?: explaining the predictions of any classifier,\u201d","volume-title":"Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining (KDD)","author":"Ribeiro","year":"2016"},{"key":"B31","first-page":"3543","article-title":"\u201cConformalized quantile regression,\u201d","volume-title":"Advances in neural information processing systems (NeurIPS)","author":"Romano","year":"2019"},{"key":"B32","doi-asserted-by":"publisher","first-page":"1273","DOI":"10.1080\/01621459.1993.10476408","article-title":"Alternatives to the median absolute deviation","volume":"88","author":"Rousseeuw","year":"1993","journal-title":"J. Am. Stat. Assoc"},{"key":"B33","doi-asserted-by":"publisher","first-page":"217","DOI":"10.1002\/(SICI)1099-1255(199805\/06)13:3<217::AID-JAE476>3.0.CO;2-V","article-title":"Stylized facts of daily return series and the Hidden Markov model","volume":"13","author":"Ryd\u00e9n","year":"1998","journal-title":"J. Appl. Econ"},{"key":"B34","doi-asserted-by":"publisher","first-page":"1181","DOI":"10.1016\/j.ijforecast.2019.07.001","article-title":"DeepAR: probabilistic forecasting with autoregressive recurrent networks","volume":"36","author":"Salinas","year":"2020","journal-title":"Int. J. Forecast"},{"key":"B35","doi-asserted-by":"publisher","first-page":"485","DOI":"10.55549\/epstem.1372067","article-title":"ETF markets' prediction and assets management platform using probabilistic autoregressive recurrent networks","volume":"23","author":"Soliman","year":"2023","journal-title":"Eurasia Proc. Sci. Technol. Eng. Math"},{"key":"B36","doi-asserted-by":"publisher","first-page":"e70004","DOI":"10.1002\/isaf.70004","article-title":"Improving ETF prediction through sentiment analysis: a DeepAR and FinBERT approach with controlled seed sampling","volume":"32","author":"Soliman","year":"2025","journal-title":"Intell. Syst. Account. Financ. Manag"},{"key":"B37","first-page":"5998","article-title":"\u201cAttention is all you need,\u201d","volume-title":"Advances in neural information processing systems (NeurIPS)","author":"Vaswani","year":"2017"},{"key":"B38","article-title":"\u201cTimeMixer: decomposable multiscale mixing for time series forecasting,\u201d","volume-title":"International conference on learning representations (ICLR)","author":"Wang","year":"2024"},{"key":"B39","doi-asserted-by":"publisher","first-page":"129178","DOI":"10.1016\/j.neucom.2024.129178","article-title":"Is Mamba effective for time series forecasting?","volume":"619","author":"Wang","year":"2025","journal-title":"Neurocomputing"},{"key":"B40","doi-asserted-by":"publisher","first-page":"187","DOI":"10.1080\/01621459.1972.10481224","article-title":"A decision-theoretic approach to interval estimation","volume":"67","author":"Winkler","year":"1972","journal-title":"J. Am. Stat. Assoc"},{"key":"B41","article-title":"\u201cTimesNet: temporal 2D-variation modeling for general time series analysis,\u201d","volume-title":"International conference on learning representations (ICLR)","author":"Wu","year":"2023"},{"key":"B42","first-page":"22419","article-title":"\u201cAutoformer: decomposition transformers with auto-correlation for long-term series forecasting,\u201d","volume-title":"Advances in neural information processing systems (NeurIPS)","author":"Wu","year":"2021"},{"key":"B43","doi-asserted-by":"publisher","first-page":"11106","DOI":"10.1609\/aaai.v35i12.17325","article-title":"Informer: beyond efficient transformer for long sequence time-series forecasting","volume":"35","author":"Zhou","year":"2021","journal-title":"Proc AAAI Conf Artif Intell"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1903497\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,23]],"date-time":"2026-09-23T06:16:05Z","timestamp":1790144165000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1903497\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,23]]},"references-count":43,"alternative-id":["10.3389\/frai.2026.1903497"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1903497","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,23]]},"article-number":"1903497"}}