Many time series models in economics and finance switch between different "regimes" (like calm and turbulent market periods) according to a random process called a Markov chain. In each regime, the system evolves according to a linear rule involving a matrix, producing what is called a Markov-switching affine recurrence. A key question is when the long-run statistical behavior of such a system is well-behaved. Researchers have long known that a quantity called the top Lyapunov exponent, which measures the average growth rate of random matrix products, determines whether the system settles into a stable long-run distribution at all. However, that condition alone does not tell you whether the variance and correlation structure of that distribution are actually finite and meaningful.
This paper identifies exactly when those second-order statistics, variances, covariances, and Pearson correlations, exist. The answer turns out to depend on a different mathematical object: a matrix built from the transition probabilities of the regime-switching process and the Kronecker products of the system matrices. When the largest eigenvalue of this combined operator is less than one, finite variances exist; when it is not, they blow up to infinity, even if the system is otherwise stationary. Crucially, these two conditions can come apart: a system can be stable in the sense of having a well-defined long-run distribution, yet still have infinite variances,