Abstract
Two families of matrix-variate hidden Markov regression models (MV-HMRMs) are here introduced. The distinction between them relies on the role of the covariates, which can be treated as fixed or random. Parsimony is achieved by using the eigen-decomposition of the components’ covariance matrices. This generates a different number of parsimonious models between the two families: 98 MV-HMRMs with fixed covariates and 9604 MV-HMRMs with random covariates. ECM algorithms are discussed for parameter estimation and, because of the high number of parsimonious models, convenient initialization, and fitting strategies are proposed. Our models are first applied to simulated data, and then to a four-way real dataset concerning the relationship between unemployment and labor force participation in the Italian provinces.
Similar content being viewed by others
Code Availability
The code, examples of simulated analyses, and the real dataset used in this manuscript are publicly available at https://github.com/danieletomarchio/Parsimonious-MV-HMRMs-with-fixed-and-random-covariates.
References
Altman, R. M., & Petkau, A. J. (2005). Application of hidden Markov models to multiple sclerosis lesion count data. Statistics in Medicine, 24(15), 2335–2344.
Altuzarra, A., Gálvez, Gálvez. C., & González, Flores A. (2019). Unemployment and labour force participation in Spain. Applied Economics Letters, 26(5), 345–350.
Apergis, N., & Arisoy, I. (2017). Unemployment and labor force participation across the US states: new evidence from panel data. SPOUDAI-Journal of Economics and Business, 67(4), 45–84.
Bacri, T., Berentsen, G. D., Bulla, J., et al. (2022). A gentle tutorial on accelerated parameter and confidence interval estimation for hidden Markov models using template model builder. Biometrical Journal, 64(7), 1260–1288.
Bartolucci, F., & Farcomeni, A. (2015). A discrete time event-history approach to informative drop-out in mixed latent Markov models with covariates. Biometrics, 71(1), 80–89.
Bartolucci, F., Lupparelli, M., & Montanari, G. E. (2009). Latent Markov model for longitudinal binary data: an application to the performance evaluation of nursing homes. The Annals of Applied Statistics, 3(2), 611–636.
Bartolucci, F., Farcomeni, A., & Pennoni, F. (2012). Latent Markov models for longitudinal data. CRC Press.
Baum, L. E., Petrie, T., Soules, G., et al. (1970). A maximization technique occurring in the statistical analysis of probabilistic functions of Markov chains. The Annals of Mathematical Statistics, 41(1), 164–171.
Biernacki, C., Celeux, G., & Govaert, G. (2003). Choosing starting values for the EM algorithm for getting the highest likelihood in multivariate Gaussian mixture models. Computational Statistics and Data Analysis, 41, 561–575.
Bulla J. (2011). Hidden Markov models with t components. increased persistence and other aspects. Quantitative Finance, 11(3), 459–475
Dang U. J., & McNicholas P. D. (2015). Families of parsimonious finite mixtures of regression models. In: Morlini I., Minerva T., Vichi M. (eds) advances in statistical models for data analysis. studies in classification, data analysis, and knowledge organization. Springer, pp. 73–84
Dang, U. J., Browne, R. P., & McNicholas, P. D. (2015). Mixtures of multivariate power exponential distributions. Biometrics, 71(4), 1081–1089.
Dang, U. J., Punzo, A., McNicholas, P. D., et al. (2017). Multivariate response and parsimony for Gaussian cluster-weighted models. Journal of Classification, 34, 4–34.
Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977). Maximum likelihood from incomplete data via the EM algorithm. Journal of the Royal Statistical Society: Series B, 39(1), 1–22.
Di Mari, R., Oberski, D. L., & Vermunt, J. K. (2016). Bias-adjusted three-step latent Markov modeling with covariates. Structural Equation Modeling: A Multidisciplinary Journal, 23(5), 649–660.
Doğru, F. Z., Bulut, Y. M., & Arslan, O. (2016). Finite mixtures of matrix variate t distributions. Gazi University Journal of Science, 29(2), 335–341.
European Parliament and Council. (2019). Regulation (EU) 2019/1700. Official Journal of the European Union (261), 1–32. http://data.europa.eu/eli/reg/2019/1700/oj
Farcomeni, A., & Punzo, A. (2020). Robust model-based clustering with mild and gross outliers. Test, 29, 989–1007.
Fraley, C., & Raftery, A. E. (2003). Enhanced model-based clustering, density estimation, and discriminant analysis software: Mclust. Journal of Classification, 20(2), 263–286.
Gallaugher, M. P., & McNicholas, P. D. (2018). Finite mixtures of skewed matrix variate distributions. Pattern Recognition, 80, 83–93.
Hubert, L., & Arabie, P. (1985). Comparing partitions. Journal of Classification, 2(1), 193–218.
Ingrassia, S., Minotti, S. C., & Vittadini, G. (2012). Local statistical modeling via a cluster-weighted approach with elliptical distributions. Journal of classification, 29, 363–401.
Kakinaka, M., & Miyamoto, H. (2012). Unemployment and labour force participation in Japan. Applied Economics Letters, 19(11), 1039–1043.
Koski T. (2001). Hidden Markov models for bioinformatics, vol 2. Springer Science & Business Media
Lagona, F., Maruotti, A., & Picone, M. (2011). A non-homogeneous hidden Markov model for the analysis of multi-pollutant exceedances data (pp. 207–222). Theory and Applications: Hidden Markov Models.
Lagona, F., Jdanov, D., & Shkolnikova, M. (2014). Latent time-varying factors in longitudinal analysis: a linear mixed hidden Markov model for heart rates. Statistics in Medicine, 33(23), 4116–4134.
Langrock, R., Kneib, T., Glennie, R., et al. (2017). Markov-switching generalized additive models. Statistics and Computing, 27, 259–270.
Maruotti, A. (2011). Mixed hidden Markov models for longitudinal data: an overview. International Statistical Review, 79(3), 427–454.
Maruotti, A. (2014). Robust fitting of hidden Markov regression models under a longitudinal setting. Journal of Statistical Computation and Simulation, 84(8), 1728–1747.
Maruotti, A., & Punzo, A. (2017). Model-based time-varying clustering of multivariate longitudinal data with covariates and outliers. Computational Statistics & Data Analysis, 113, 475–496.
Maruotti, A., & Punzo, A. (2021). Initialization of hidden Markov and semi-Markov models: a critical evaluation of several strategies. International Statistical Review, 89(3), 447–480.
Maruotti, A., & Rocci, R. (2012). A mixed non-homogeneous hidden Markov model for categorical data, with application to alcohol consumption. Statistics in Medicine, 31(9), 871–886.
Maruotti, A., Punzo, A., & Bagnato, L. (2019). Hidden Markov and semi-Markov models with multivariate leptokurtic-normal components for robust modeling of daily returns series. Journal of Financial Econometrics, 17(1), 91–117.
Meng, X. L., & Rubin, D. B. (1993). Maximum likelihood estimation via the ECM algorithm: a general framework. Biometrika, 80(2), 267–278.
Merlo, L., Maruotti, A., Petrella, L., et al. (2022). Quantile hidden semi-Markov models for multivariate time series. Statistics and Computing, 32(4), 1–22.
Montanari, G. E., & Pandolfi, S. (2018). Evaluation of long-term health care services through a latent Markov model with covariates. Statistical Methods & Applications, 27, 151–173.
Ozerkek, Y. (2013). Unemployment and labor force participation: a panel cointegration analysis for European countries. Applied Econometrics and International Development, 13(1), 67–76.
Punzo, A. (2014). Flexible mixture modelling with the polynomial Gaussian cluster-weighted model. Statistical Modelling, 14(3), 257–291.
Punzo, A., & McNicholas, P. D. (2016). Parsimonious mixtures of multivariate contaminated normal distributions. Biometrical Journal, 58(6), 1506–1537.
Punzo, A., & McNicholas, P. D. (2017). Robust clustering in regression analysis via the contaminated Gaussian cluster-weighted model. Journal of Classification, 34, 249–293.
Punzo, A., Ingrassia, S., & Maruotti, A. (2018). Multivariate generalized hidden Markov regression models with random covariates: physical exercise in an elderly population. Statistics in Medicine, 37(19), 2797–2808.
Punzo, A., Ingrassia, S., & Maruotti, A. (2021). Multivariate hidden Markov regression models: random covariates and heavy-tailed distributions. Statistical Papers, 62(3), 1519–1555.
R Core Team. (2021). R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, https://www.R-project.org/
Sarkar, S., Zhu, X., Melnykov, V., et al. (2020). On parsimonious models for modeling matrix data. Computational Statistics & Data Analysis, 142(106), 822.
Schwarz, G. (1978). Estimating the dimension of a model. The Annals of Statistics, 6(2), 461–464.
Scrucca L., Fop M., Murphy T.B., et al. (2016). mclust 5: clustering, classification and density estimation using Gaussian finite mixture models. The R Journal, 8(1), 289–317. https://doi.org/10.32614/RJ-2016-021
Tomarchio S. D. (2022). Matrix-variate normal mean-variance Birnbaum–Saunders distributions and related mixture models. Computational Statistics pp. 1–28
Tomarchio, S. D., McNicholas, P. D., & Punzo, A. (2021). Matrix normal cluster-weighted models. Journal of Classification, 38(3), 556–575.
Tomarchio, S. D., Bagnato, L., & Punzo, A. (2022). Model-based clustering via new parsimonious mixtures of heavy-tailed distributions. AStA Advances in Statistical Analysis, 106(2), 315–347.
Tomarchio, S. D., Gallaugher, M. P., Punzo, A., et al. (2022). Mixtures of matrix-variate contaminated normal distributions. Journal of Computational and Graphical Statistics, 31(2), 413–421.
Tomarchio, S. D., Punzo, A., & Maruotti, A. (2022). Parsimonious hidden Markov models for matrix-variate longitudinal data. Statistics and Computing, 32(3), 1–18.
Viroli, C. (2012). On matrix-variate regression analysis. Journal of Multivariate Analysis, 111, 296–309.
Visser, I., Raijmakers, M. E., & Molenaar, P. C. (2000). Confidence intervals for hidden Markov model parameters. British Journal of Mathematical and Statistical Psychology, 53(2), 317–327.
Visser, I., Raijmakers, M. E., & Molenaar, P. (2002). Fitting hidden Markov models to psychological data. Scientific Programming, 10(3), 185–199.
Welch, L. R. (2003). Hidden Markov models and the baum-welch algorithm. IEEE Information Theory Society Newsletter, 53(4), 10–13.
Zucchini, W., MacDonald, I. L., & Langrock, R. (2017). Hidden Markov models for time series: an introduction using R. CRC Press.
Acknowledgements
This research also contributes to the PNRR GRInS Project. This research has also been partially supported by MIUR via "The SMILE project: Statistical Modelling and Inference to Live the Environment".
Funding
The authors have no financial interests to report.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflicts of interest
The authors have no conflicts of interest to disclose.
Ethical Conduct
The authors have fulfilled ethical conduct.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A
Here, we report the parameters used to generate the simulated datasets of Section 4.1 for the MV-HMRMFCs. We used the following parameters for all three models: \(\pi _1=\pi _2=\pi _3=1/3,\)
Additionally, for \(\textbf{B}_2\) and \(\textbf{B}_3\), we use the same regression coefficients of \(\textbf{B}_1\), with the exclusion of the first column containing the intercepts, which are replaced by the following vectors \(\varvec{\beta }_{2}=(-7,-8,-7)'\) and \(\varvec{\beta }_{3}=(7,8,7)'\), respectively. Then, for the VVV-VV MV-HMRMFC, we have
For the VVI-VI MV-HMRMFC, we set \(\varvec{\Sigma }_{\mathcal {Y}|1}=\text {diag}(0.64,2.12,0.25)\), \(\varvec{\Sigma }_{\mathcal {Y}|2}=\text {diag}(1.50,\) 2.16, 2.47), \(\varvec{\Sigma }_{\mathcal {Y}|3}=\text {diag}(7.16,2.51,3.56)\), \(\varvec{\Psi }_{\mathcal {Y}|1}=\text {diag}(0.58,1.92,0.90)\), \(\varvec{\Psi }_{\mathcal {Y}|2}=\text {diag}(0.46,0.73,2.98)\) and \(\varvec{\Psi }_{\mathcal {Y}|3}=\text {diag}(0.49,0.37,5.50)\), where diag\((\cdot )\) identifies a diagonal matrix.
For the EVE-EE MV-HMRMFC, we use
Appendix B
Here, we report the parameters used to generate the simulated datasets of Section 4.1 for the MV-HMRMRCs. We used the following parameters for all three models: \(\pi _1=\pi _2=0.50,\)
Similarly to before, for \(\textbf{B}_2\) we use the same regression coefficients of \(\textbf{B}_1\), with the exclusion of the first column containing the intercepts, which is replaced by the following vector \(\varvec{\beta }_{2}=(-12,-13)'\). Additionally, \(\textbf{M}_2\) is obtained by adding a constant \(c=9\) to each element in \(\textbf{M}_1\).
Then, for the VVV-VV-VVV-VV MV-HMRMRC, we have
For the VVI-VI-VVI-VI MV-HMRMRC, we set \(\varvec{\Sigma }_{\mathcal {Y}|1}=\text {diag}(1.73,0.58)\), \(\varvec{\Sigma }_{\mathcal {Y}|2}=\text {diag}(1.63,9.80)\), \(\varvec{\Psi }_{\mathcal {Y}|1}=\text {diag}(0.74,1.48,0.74,1.11,1.11)\), \(\varvec{\Psi }_{\mathcal {Y}|2}=\text {diag}(2.64,1.32,0.66,\) 0.66, 0.66), \(\varvec{\Sigma }_{\mathcal {X}|1}=\text {diag}(5.55,2.77,4.16)\), \(\varvec{\Sigma }_{\mathcal {X}|2}=\text {diag}(1.00,2.00,0.50)\), \(\varvec{\Psi }_{\mathcal {X}|1}=\text {diag}(1.06,1.77,1.06,0.71,0.71)\) and \(\varvec{\Psi }_{\mathcal {X}|2}=\text {diag}(0.57,0.57,2.30,2.30,0.57)\).
For the EVE-EE-EEV-EV MV-HMRMRC, we use
Appendix C
Here, we report the remaining parameters used to generate the simulated datasets of Section 4.3. In particular, we have \(\textbf{M}_2=\textbf{M}_1\) in Scenario A and the following parameters in common between the two scenarios
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Tomarchio, S.D., Punzo, A. & Maruotti, A. Matrix-Variate Hidden Markov Regression Models: Fixed and Random Covariates. J Classif (2023). https://doi.org/10.1007/s00357-023-09438-y
Accepted:
Published:
DOI: https://doi.org/10.1007/s00357-023-09438-y