\(\mathcal{G}\)-inhomogeneous Markov systems of high order (Q973027)

The authors introduce and study the \({\mathcal G}\)-inhomogeneous Markov system of high order, which is a more general stochastic process, in many respects, than the known inhomogeneous Markov system. The paper is organized in 5 sections. After an Introduction, in Section 2 a rigorous definition of the \({\mathcal G}\)-inhomogeneous Markov system of high order is given. The Section 3 is entitled `The Inhomogeneous Mixture Transition Distribution Model'. Here, the superficial razor cut mixture transition distribution model is defined, extending for the inhomogeneous case the idea of mixture transition model by \textit{A. E. Raftery} [J. R. Stat. Soc., Ser. B 47, 528--539 (1985; Zbl 0593.62091)] and \textit{A. Berchtold} and \textit{A. E. Raftery} [Stat. Sci. 17, No. 3, 328--356 (2002; Zbl 1013.62088)]. As a conclusion of this section Lemma 3.1 is proved, where functional relations among the considered vector stochastic processes are established. In Section 4 (The Asymptotic behaviour of the \({\mathcal G}\)-inhomogeneous Markov system of high order) two basic theorems related to the asymptotic behaviour of the NHMS of high order are proved (Theorem 4.4 and Theorem 4.5). In Theorem 4.4, it is proved that the inhomogeneous superficilal razor cut model of mixture transition model, asymptotically has the same equilibrium distribution, as the first-order homogeneouys Markov chain to which inhomogeneous Markov chain converges as \(t\) goes to infinity. And in Theorem 4.5 is found the symptotic behaviour of the expected population structure and the expected relative population structure of the \({\mathcal G}\)-inhomogeneous Markov system of high order and provide conditions, easily met in practice, under which the convergence is geometrically fast. The last section (Illustration) is reserved to illustrate the present results with data that are typical of a hierarchical manpower system. Such data have been used from British firms and could be found among others, in earlier papers of the first author. For presentation purposes though, since, if the number of states exceeds \(k= 3\) the matrices of high order transition probabilities become too large, to be presented in a scientific paper.