Abstract
We develop an efficient algorithm to compute the likelihood of the phase-type ageing model. The proposed algorithm uses the uniformisation method to stabilise the numerical calculation. It also utilises a vectorised formula to only calculate the necessary elements of the probability distribution. Our algorithm, with an error’s upper bound, could be adjusted easily to tackle the likelihood calculation of the Coxian models. Furthermore, we compare the speed and the accuracy of the proposed algorithm with those of the traditional method using the matrix exponential. Our algorithm is faster and more accurate than the traditional method in calculating the likelihood. Based on our experiments, we recommend using 20 sets of randomly-generated initial values for the optimisation to get a reliable estimate for which the evaluated likelihood is close to the maximum likelihood.
Similar content being viewed by others
References
Al-Mohy A, Higham N (2009) A new scaling and squaring algorithm for the matrix exponential. SIAM J Matrix Anal Appl 31:970–989
Asmussen S, Nerman O, Olsson M (1996) Fitting phase-type distributions via the EM algorithm. Scand J Stat 23:419–441
Asmussen S, Laub J, Yang H (2019) Phase-type models in life insurance: Fitting and valuation of equity-linked benefits. Risks 7(1):17
Bickart T (1968) Matrix exponential: Approximation by truncated power series. Proc IEEE 56:872–873
Bobbio A, Cumani A (1992) ML estimation of the parameters of a PH distribution in triangular canonical form. Computer Performance Evaluation 22:33–46
Byrd R, Hribar M, Nocedal J (1999) An interior point algorithm for large-scale nonlinear programming. SIAM J Optim 9(4):877–900
Byrd R, Gilbert J, Nocedal J (2000) A trust region method based on interior point techniques for nonlinear programming. Math Program 89(1):149–185
Cheng B, Jones B, Liu X, Ren J (2021) The mathematical mechanism of biological aging. North American Actuarial Journal 25:73–93
Choudhury A, Choudhury D, Roy B (1968) On the evaluation of \(e^{A\tau }\). Proc IEEE 56:1110–1111
Cody W, Meinardus G, Varga R (1969) Chebyshev rational approximations to \(e^{-x}\) in \([0,+\infty )\) and applications to heat-conduction problems. J Approx Theory 2:50–65
Cox D (1955) A use of complex probabilities in the theory of stochastic processes. Math Proc Cambridge Philos Soc 51:313–319
Duan Q, Liu J (2016) Modelling a bathtub-shaped failure rate by a Coxian distribution. IEEE Trans Reliab 65:878–885
Faddy M (1998) On inferring the number of phases in a Coxian phase-type distribution. Stoch Model 14:407–417
Faddy M (2002) Penalised maximum likelihood estimation of the parameters in a Coxian phase-type distribution. In: Latouche G, Taylor P (eds) Matrix-Analytic Methods: Theory and Applications, 11. World Scientific, Singapore, pp 107–114
Gantmacher F (1959) Applications of the Theory of Matrices. Interscience Publishers Inc., New York
Govorun M, Jones B, Liu X, Stanford D (2018) Physiological age, health costs, and their interrelation. North American Actuarial Journal 22:323–340
Gross D, Miller D (1984) The randomization technique as a modeling tool and solution procedure for transient Markov processes. Oper Res 32:343–361
Healey M (1973) Study of methods of computing transition matrices. Proceedings of the Institution of Electrical Engineers 120:905–912
Higham N (2005) The scaling and squaring method for the matrix exponential revisited. SIAM J Matrix Anal Appl 26:1179–1193
Hyde J (1980) Testing survival with incomplete observations. In: Miller R, Efron B, Brown B, Moses L (eds) Biostatistics Casebook. John Wiley, New York, pp 31–46
Jensen A (1953) Markoff chains as an aid in the study of Markoff processes. Scand Actuar J 1953(sup1):87–91
Källström C (1973) Computing exp (A) and \(\int \) exp(As) ds, Division of Automatic Control, Lund Institute of Technology, Lund
Kirchner R (1967) An explicit formula for \(e^{At}\). Amer Math Monthly 74:1200–1204
Lin X, Liu X (2007) Markov aging process and phase-type law of mortality. North American Actuarial Journal 11:92–109
Liou M (1996) A novel method of evaluating transient response. Proc IEEE 54:20–23
Marshall A, Zenga M (2009) Recent developments in fitting coxian phase-type distributions in healthcare, ASMDA. Proceedings of the International Conference Applied Stochastic Models and Data Analysis 13:482
Marshall A, Zenga M (2012) Experimenting with the Coxian phase-type distribution to uncover suitable fits. Methodol Comput Appl Probab 14:71–86
Mitchell H (2016) Latent phase-type models for Italy’s ageing population. PhD Thesis, Queen’s University, Belfast
Moler C, Loan C (2003) Nineteen dubious ways to compute the exponential of a matrix twenty-five years later. SIAM Rev 45(2003):3–49
Nelder J, Mead R (1965) A simplex method for function minimization. Comput J 7:308–313
Okamura H, Dohi T, Trivedi K (2009) Markovian arrival process parameter estimation with group data. IEEE/ACM Trans Networking 17(4):1326–1339
Okamura H, Dohi T, Trivedi K (2011) A refined EM algorithm for PH distributions. Perform Eval 68(10):938–954
Okamura H, Dohi T, Trivedi S (2013) Improvement of Expectation-Maximization algorithm for phase-type distributions with grouped and truncated data. Appl Stoch Model Bus Ind 29(2):141–156
Okamura H, Dohi T (2009) Faster maximum likelihood estimation algorithms for Markovian arrival processes. 2009 Sixth International Conference on the Quantitative Evaluation of Systems, IEEE, 73-82
Putzer E (1966) Avoiding the Jordan canonical form in the discussion of linear systems with constant coefficients. Amer Math Monthly 73:2–7
Rice J (2006) Mathematical Statistics and Data Analysis. Thomson Brooks/Cole, Belmont
Rizk J, Walsh C, Burke K (2021) An alternative formulation of Coxian phase-type distributions with covariates: Application to emergency department length of stay. Stat Med 40(6):1574–1592
Ross S (2014) Introduction to Probability Models. Academic Press, Oxford
Saff E (1971) The convergence of rational functions of best approximation to the exponential function. Trans Am Math Soc 153:483–493
Sidje R (1998) Expokit: A software package for computing matrix exponentials. ACM Transactions on Mathematical Software 24:130–156
Siu C, Yam P, Yang H (2015) Valuing equity-linked death benefits in a regime-switching framework. ASTIN Bulletin 45(2):355–395
Stewart W (1994) Introduction to the Numerical Solution of Markov Chains. Princeton University Press, Princeton
Su S, Sherris M (2012) Heterogeneity of Australian population mortality and implications for a viable life annuity market. Insurance Math Econom 51(2):22–332
Waltz RA, Morales JL, Nocedal J, Orban D (2006) An interior algorithm for nonlinear optimization that combines line search and trust region steps. Math Program 107(3):391–408
Ward R (1997) Numerical computation of the matrix exponential with accuracy estimate. SIAM J Numer Anal 14:600–610
Acknowledgements
This work is supported by the Natural Sciences and Engineering Research Council of Canada through R. Mamon’s Discovery Grant (RGPIN-2017-04235). The authors thank Bruce Jones, Jiandong Ren, the Associate Editor, and two anonymous reviewers for their valuable comments.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Declarations
Both authors certify that they have no affiliations with or involvement in any organisation or entity with any financial interest or non-financial interest in the subject matter or materials discussed in this paper.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
Proof of Theorem 2.1
Suppose \(h_1=h_m=h\). It follows that \(h_i=h\) for \(i=1,\ldots ,m\).
Suppose \(h_1<h_m\). When \(s>0\), \(h_1^s< h_m^s\). For any \(1 \le i<j \le m\),
where \(\displaystyle p_1=\frac{m-i}{m-1}>\frac{m-j}{m-1}=p_2\) and the inequality holds using the lemma below. Therefore, \(h_i<h_j\) using the fact that \(x^s\) is an increasing function of x when \(x,s>0\).
Similarly, when \(s<0\), \(h_1^s>h_m^s\), we have, for any \(1 \le i<j \le m\),
Hence, \(h_i<h_j\) since \(x^s\) is a decreasing function of x when \(x>0\) and \(s<0\).
When \(s=0\),
which is an increasing function of i if and only if \(h_1<h_m\).
In summary, for any value of s, \(h_i\) is an increasing function of i if and only if \(h_1<h_m\). Likewise, it may be shown that \(h_i\) is a decreasing function with respect to i if and only if \(h_1>h_m\). \(\square \)
Lemma
For any \(0<a<b\) and \(0\le p_2<p_1 \le 1\), the inequality \(ap_1+b(1-p_1)<ap_2+b(1-p_2)\) holds.
Proof
\(\square \)
Rights and permissions
Springer Nature or its licensor 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
Cheng, B., Mamon, R. A uniformisation-driven algorithm for inference-related estimation of a phase-type ageing model. Lifetime Data Anal 29, 142–187 (2023). https://doi.org/10.1007/s10985-022-09577-1
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10985-022-09577-1