When Does Regime Information Improve Volatility Forecasting? Evidence from TGARCH-X, Transformer, and Combined Models

Mwaniki, Daniel Ndirangu; Simiyu, Christine Nanjala; Ong'ala, Jacob Otieno

doi:10.22054/jmmf.2026.95229.1314

When Does Regime Information Improve Volatility Forecasting? Evidence from TGARCH-X, Transformer, and Combined Models

Articles in Press

Document Type : Research Article

Authors

Daniel Ndirangu Mwaniki ¹

Christine Nanjala Simiyu ²

Jacob Otieno Ong'ala ³

¹ School of Business, KCA University, Nairobi, Kenya.

² Dean, School of Business and Associate Professor of Economics, KCA University, Nairobi, Kenya.

³ Associate Professor of Statistics, Faculty of Biological and Physical Sciences, Tom Mboya University, Kenya.

10.22054/jmmf.2026.95229.1314

Abstract

This study examines whether latent regime information improves volatility forecasts consistently across econometric, neural, and combined models. Using daily observations for 52 firms across 11 Nairobi Securities Exchange sectors, the study evaluates forecasts from August 2016 to July 2026. The study compares matched baseline and regime-informed models within a leakage-safe chronological design, with regime-informed models using real-time filtered regime probabilities. Regimes are estimated using a zero-neutral two-state Student-t hidden Markov model that separates exact zero returns from state-dependent non-zero return variation. The forecasting set includes GARCH-family models, GRU, Transformer, PatchTST, and static and regime-dependent forecast combinations. Market-capitalisation-weighted QLIKE is the primary loss measure, supported by dependence-robust predictive-accuracy tests, Model Confidence Sets, alternative volatility targets, and specification checks. Simple econometric benchmarks remain difficult to beat on the primary squared-return target. TGARCH-X differences remain below their 80% minimum detectable effects, while equivalence testing supports a practically small overall regime-calibration effect. The probability-aware Transformer and probability-dependent combination perform worse than their respective matched baselines under the primary target. Parkinson variance strengthens the TGARCH-X result, but a Parkinson-based HMM does not outperform the primary return-based regime extraction. The economic-significance analysis finds no reliable portfolio advantage from regime calibration, despite its approximately 3.12% reduction in turnover. Overall, regime information provides selective forecasting value across models, market states, volatility targets, and specifications rather than a universal predictive benefit.

Keywords

Volatility Forecasting

Regime-Aware Forecasting

TGARCH-X

Hidden Markov Model

Transformer

Forecast Combinations

[1] Y. Amihud, Illiquidity and stock returns: Cross-section and time-series effects, Journal of Financial Markets, (2002) vol. 5 no. 1 pp. 31–56. https://doi.org/10.1016/S1386-4181(01)00024-6
[2] J. M. Bates, and C. W. J. Granger, The combination of forecasts, Operational Research Quarterly, (1969) vol. 20 no. 4 pp. 451–468. https://doi.org/10.2307/3008764
[3] T. Bollerslev, Generalized autoregressive conditional heteroskedasticity, Journal of Econometrics, (1986) vol. 31 no. 3 pp. 307–327. https://doi.org/10.1016/0304-4076(86)90063-1
[4] A. C. Cameron, J. B. Gelbach, and D. L. Miller, Robust inference with multiway clustering, Journal of Business & Economic Statistics, (2011) vol. 29 no. 2 pp. 238–249. https://doi.org/10.1198/jbes.2010.07136
[5] F. X. Diebold, and R. S. Mariano, Comparing predictive accuracy, Journal of Business & Economic Statistics, (1995) vol. 13 no. 3 pp. 253–263. https://doi.org/10.1080/07350015.1995.10524599
[6] K. Y. L. Fong, C. W. Holden, and C. A. Trzcinka, What are the best liquidity proxies for global research?, Review of Finance, (2017) vol. 21 no. 4 pp. 1355–1401. https://doi.org/10.1093/rof/rfx003
[7] R. Giacomini, and B. Rossi, Forecast comparisons in unstable environments, Journal of Applied Econometrics, (2010) vol. 25 no. 4 pp. 595–620. https://doi.org/10.1002/jae.1177
[8] R. Giacomini, and H. White, Tests of conditional predictive ability, Econometrica, (2006) vol. 74 no. 6 pp. 1545–1578. https://doi.org/10.1111/j.1468-0262.2006.00718.x

[9] L. R. Glosten, R. Jagannathan, and D. E. Runkle, On the relation between the expected value and the volatility of the nominal excess return on stocks, The Journal of Finance, (1993) vol. 48 no. 5 pp. 1779–1801. https://doi.org/10.1111/j.1540-6261.1993.tb05128.x
[10] M. Haas, S. Mittnik, and M. S. Paolella, A new approach to Markov-switching GARCH models, Journal of Financial Econometrics, (2004) vol. 2 no. 4 pp. 493–530. https://doi.org/10.1093/jjfinec/nbh020
[11] J. D. Hamilton, A new approach to the economic analysis of nonstationary time series and the business cycle, Econometrica, (1989) vol. 57 no. 2 pp. 357–384. https://doi.org/10.2307/1912559
[12] P. R. Hansen, and A. Lunde, A forecast comparison of volatility models: Does anything beat a GARCH(1,1)?, Journal of Applied Econometrics, (2005) vol. 20 no. 7 pp. 873–889. https://doi.org/10.1002/jae.800
[13] P. R. Hansen, A. Lunde, and J. M. Nason, The model confidence set, Econometrica, (2011) vol. 79 no. 2 pp. 453–497. https://doi.org/10.3982/ECTA5771
[14] D. Harvey, S. Leybourne, and P. Newbold, Testing the equality of prediction mean squared errors, International Journal of Forecasting, (1997) vol. 13 no. 2 pp. 281–291. https://doi.org/10.1016/S0169-2070(96)00719-4
[15] F. Klaassen, Improving GARCH volatility forecasts with regime-switching GARCH, In J. D. Hamilton and B. Raj (eds.), Advances in Markov-Switching Models, Physica-Verlag HD, (2002) pp. 223–254. https://doi.org/10.1007/978-3-642-51182-0_10
[16] D. A. Lesmond, J. P. Ogden, and C. A. Trzcinka, A new estimate of transaction costs, The Review of Financial Studies, (1999) vol. 12 no. 5 pp. 1113–1141. https://doi.org/10.1093/rfs/12.5.1113
[17] R. M. Leushuis, and N. Petkov, Advances in forecasting realized volatility: A review of methodologies, Financial Innovation, (2026) vol. 12 no. 1 Article 14. https://doi.org/10.1186/s40854-025-00809-5
[18] B. Lim, S. Ö. Arik, N. Loeff, and T. Pfister, Temporal Fusion Transformers for interpretable multi-horizon time series forecasting, International Journal of Forecasting, (2021) vol. 37 no. 4 pp. 1748–1764. https://doi.org/10.1016/j.ijforecast.2021.03.012
[19] R. Liu, M. Segnon, O. Cepni, and R. Gupta, Forecasting volatility of commodity, currency, and stock markets: Evidence from Markov-switching multifractal models, Journal of Forecasting, (2026) vol. 45 no. 6 pp. 2905–2941. https://doi.org/10.1002/for.70145
[20] A. K. Mishra, J. Renganathan, and A. Gupta, Volatility forecasting and assessing risk of financial markets using multi-transformer neural network based architecture, Engineering Applications of Artificial Intelligence, (2024) vol. 133 108223. https://doi.org/10.1016/j.engappai.2024.108223
[21] F. Moreno-Pino, and S. Zohren, DeepVol: Volatility forecasting from high-frequency data with dilated causal convolutions, Quantitative Finance, (2024) vol. 24 no. 8 pp. 1105-1127. https://doi.org/10.1080/14697688.2024.2387222
[22] N. Naifar, Do supply-chain stress and geopolitical risk predict strategic commodity and clean energy market returns? Evidence from explainable machine learning, Forecasting, (2026) vol. 8 no. 4 Article 59. https://doi.org/10.3390/forecast8040059
[23] Y. Nie, N. H. Nguyen, P. Sinthong, and J. Kalagnanam, A time series is worth 64 words: Long-term forecasting with Transformers, The Eleventh International Conference on Learning Representations, (2023). https://openreview.net/forum?id=Jbdc0vTOcol
[24] M. Parkinson, The extreme value method for estimating the variance of the rate of return, The Journal of Business, (1980) vol. 53 no. 1 pp. 61–65.
[25] A. J. Patton, Volatility forecast comparison using imperfect volatility proxies, Journal of Econometrics, (2011) vol. 160 no. 1 pp. 246–256. https://doi.org/10.1016/j.jeconom.2010.03.034

[26] M. Pinitjitsamut, Multi-scale forecasting of natural rubber prices using VMD-augmented BiLSTM: A hybrid architecture ablation study, Forecasting, (2026) vol. 8 no. 3 Article 43. https://doi.org/10.3390/forecast8030043
[27] Z. Qiu, C. Kownatzki, F. Scalzo, and E. S. Cha, Historical perspectives in volatility forecasting methods with machine learning, Risks, (2025) vol. 13 no. 5 Article 98. https://doi.org/10.3390/risks13050098
[28] D. Roodman, M. Ø. Nielsen, J. G. MacKinnon, and M. D. Webb, Fast and wild: Bootstrap inference in Stata using boottest, The Stata Journal: Promoting Communications on Statistics and Stata, (2019) vol. 19 no. 1 pp. 4–60. https://doi.org/10.1177/1536867X19830877
[29] M. Saidane, and E. Ben Mohamed, Forecasting regime-dependent tail risk in digital asset portfolios with a mixture of hidden Markov factor analyzers, Journal of Forecasting, (2026) vol. 45 no. 6 pp. 2735–2759. https://doi.org/10.1002/for.70159
[30] C. Tian, R. Tansuchat, and S. Sriboonchitta, Regime-aware stock index forecasting under latent market states: A hybrid statistical learning framework with cross-market validation, Forecasting, (2026) vol. 8 no. 3 Article 50. https://doi.org/10.3390/forecast8030050
[31] A. Timmermann, Forecast combinations, In G. Elliott, C. W. J. Granger and A. Timmermann (eds.), Handbook of Economic Forecasting, Elsevier, (2006) vol. 1 pp. 135–196. https://doi.org/10.1016/1574-0706(05)01004-9
[32] Y. Xiao, C. He, M. Yi, and Y. Hu, Adaptive model integration for stock price forecasting with CNN-ATTN-BiLSTM, Journal of Forecasting, (2026) vol. 45 no. 4 pp. 1331–1349. https://doi.org/10.1002/for.70098
[33] J.-M. Zakoian, Threshold heteroskedastic models, Journal of Economic Dynamics and Control, (1994) vol. 18 no. 5 pp. 931–955. https://doi.org/10.1016/0165-1889(94) 90039-6
[34] H. Zhao, M. Crane, and M. Bezbradica, MSPCIFormer: A multi-scale patching channelindependent Transformer for cryptocurrency price forecasting, Expert Systems, (2026) vol. 43 no. 9 e70343. https://doi.org/10.1111/exsy.70343