Moments of Markov switching models (Q1973430)

Let \(\{\varepsilon_t\}\) be i.i.d. \(N(0,1)\) random variables and \(S_t\) an unobserved stationary ergodic \(k\)-state Markov homogeneous process. The author deals with three types of Markov switching models, namely (MS I) \(y_t=\mu_{S_t} +\sigma_{S_t}\varepsilon_t\), (MS II) \(y_t=\mu_{S_t} +\varphi_1(y_{t-1}-\mu_{S_{t-1}})+\sigma_{S_t}\varepsilon_t\) with \(|\varphi_1|<1\), and (MS III) \(y_t =\mu_{S_t} +\varphi_{1S_{t-1}}(y_{t-1} -\mu_{S_{t-1}}) +\sigma_{S_t}\varepsilon_t\) with \(|\varphi_{1S_{t-1}}|<1\). In some cases it is assumed that \(k=2\). Central moments \(E(y_t-Ey_t)^n\), covariance function of \(y_t\), and covariance function of \(y_t^2\) are calculated for all three models. The results are used for finding relationships between volatility clustering, regime switches, and structural breaks in time series models.