Random linear recursions with dependent coefficients (Q1957152)

The paper deals with the tail of the stationary solution \(R\) of the difference equation \(R_n=Q_n+M_nR_{n-1}\), where the coefficients \((Q_n,M_n)\in\mathbb{R}^2,\,n\in\mathbb{Z}_+\) are independent and not identically distributed (non-i.i.d.) but induced by a sequence of random variables \(X:=(X_n,\,n\in\mathbb{Z})\), where the \(X_n\) take values in a countable set (cf. Definition 1.1). If \(X\) is a stationary and irreducible Markov chain (the coefficients are a hidden Markov model) under particular assumptions it is proved that the distribution tail of \(R\) is regularly varying at infinity (Theorem 1.3 and 1.4). Then the case that the coefficients are induced by a \(C\)-chain \(X\) (which is a particular case of a chain of infinite order (claims with complete connections) is considered. Under particular additional assumptions the authors prove that the distribution tail of \(R\) is regularly varying at infinity (Theorem 1.6 and 1.7).