On square-integrability of an AR process with Markov switching (Q5937049)

Let \(X_n\) be a stationary, positive Markov chain on \(\{1,\dots,m\}\) with transition probability matrix \(P=(p_{ij})\). Consider a family \(A_1,\dots,A_m\) of \(d\times d\) autoregressive matrices and a \(d\)-dimensional stationary ergodic sequence \(\varepsilon_n\) such that \(E\|\varepsilon_1\|^2<\infty\). Define a \(d\)-dimensional AR process with Markov switching by \[ (\ast)\qquad Y_n=A_{X_n}Y_{n-1}+\varepsilon_n. \] Let \(M\) be the matrix such that its \((i,j)\) th block is \(p_{ji}(A_j\otimes A_j)\). Define \(\rho(M)\) its spectral radius. The main result of the paper says that under the condition \(\rho(M)<1\) the equation \((\ast)\) has a unique square-integrable, stationary, and ergodic solution. This theorem is applied to several special cases of the model \((\ast)\).