A Bayesian Hidden Markov Model with Singh--Maddala Emission Distributions for Heavy-Tailed Regime-Switching Time Series
Abstract
Heavy-tailed time series with regime switching arise in environmental, financial, and engineering applications, yet conventional emission distributions such as Gaussian, Gamma, or Weibull fail to capture simultaneously extreme right-skewness, slowly decaying upper tails, and temporal dependence. This paper develops a fully Bayesian Hidden Markov Model with Singh-Maddala (Burr Type XII) emission distributions (HMMSM). We derive the complete Bayesian structure and design an adaptive Metropolis-within-Gibbs sampler combining Forward-Filtering Backward-Sampling, conjugate Dirichlet updates, and adaptive Metropolis steps. Simulations confirm parameter recovery with relative errors below 10% and 95% prediction coverage of 94.5%. On a simulated benchmark with heavy tails, Bayesian model selection via WAIC, DIC, and LOO correctly identifies K = 3, and HMMSM outperforms seven competing specifications by large margins. On a 26-year gridded rainfall record for Ireland (T = 312 monthly observations), HMMSM attains the lowest WAIC but is statistically indistinguishable from HMM-Gamma, reflecting the light tails of Irish rainfall. The selected two-regime model identifies a Moderate regime (mean approx. 87 mm) and a Wet regime (approx. 134 mm) that are persistent and seasonally coherent. We also document weak identifiability of the Singh-Maddala scale parameter. The framework is general-purpose, and reproducible code is provided.
Related
- Understanding Hidden Spatial Preferences in Bayesian Optimization
- A Bayesian Magnitude-Composition Model for Tree-Size Distributions and Stand-Tables
- Bayesian Semiparametric Hidden Markov Random Partition Fields for Factor Collapse on Graphs: A Study of Cortical Mapping of Fingertips
- Structure-Informed Bayesian Inference of Anomalous Transport and Hidden Molecular Trapping in Amorphous Media
- Posterior Regimes and Latent Deception: Variational Bayesian Inference in Hidden Markov Models for Sequential Fraud Detection in Financial Transactions
BibTeX
@misc{kadhem2026bayesian,
title = {{A Bayesian Hidden Markov Model with Singh--Maddala Emission Distributions for Heavy-Tailed Regime-Switching Time Series}},
author = {Safaa K. Kadhem},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.22656},
eprint = {2610.22656},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.22656},
}
RIS
TY - UNPB
TI - A Bayesian Hidden Markov Model with Singh--Maddala Emission Distributions for Heavy-Tailed Regime-Switching Time Series
AU - Safaa K. Kadhem
PY - 2026
DA - 2026/10/15
PB - arXiv
DO - 10.48550/arxiv.2610.22656
UR - https://arxiv.org/abs/2610.22656
AB - Heavy-tailed time series with regime switching arise in environmental, financial, and engineering applications, yet conventional emission distributions such as Gaussian, Gamma, or Weibull fail to capture simultaneously extreme right-skewness, slowly decaying upper tails, and temporal dependence. This paper develops a fully Bayesian Hidden Markov Model with Singh-Maddala (Burr Type XII) emission distributions (HMMSM). We derive the complete Bayesian structure and design an adaptive Metropolis-within-Gibbs sampler combining Forward-Filtering Backward-Sampling, conjugate Dirichlet updates, and adaptive Metropolis steps. Simulations confirm parameter recovery with relative errors below 10% and 95% prediction coverage of 94.5%. On a simulated benchmark with heavy tails, Bayesian model selection via WAIC, DIC, and LOO correctly identifies K = 3, and HMMSM outperforms seven competing specifications by large margins. On a 26-year gridded rainfall record for Ireland (T = 312 monthly observations), HMMSM attains the lowest WAIC but is statistically indistinguishable from HMM-Gamma, reflecting the light tails of Irish rainfall. The selected two-regime model identifies a Moderate regime (mean approx. 87 mm) and a Wet regime (approx. 134 mm) that are persistent and seasonally coherent. We also document weak identifiability of the Singh-Maddala scale parameter. The framework is general-purpose, and reproducible code is provided.
ER -