Tail of the stationary solution of the stochastic equation \(Y_{n+1}=a_{n} Y_{n}+b_{n}\) with Markovian coefficients (Q5921699)

The paper deals with the tail behaviour of the stationary solution \(Y_1\) of the stochastic difference equation \(Y_{n+1}=a_n Y_n + b_n,\;n \in \mathbb Z\), where the sequence \((a_n)\) is a real-valued finite state space Markov chain and \((b_n)\) is a sequence of i.i.d. real-valued r.v.'s independent of \((a_n)\). Further E \(\log| a_0| <0\) and \(E \log^+| b_0| <\infty\) are assumed ensuring the existence of a unique stationary solution \((Y_n)\). Under spectral radius properties related to the transition matrix on \((a_n)\), a technical assumption on the state space and a moment condition on \(b_0\), it is proved that \(t^\lambda P(xY_1>t)\rightarrow L(x)\) as \(t\rightarrow \infty\), for \(x\in\{-1,1\}\), where \(L(1)+L(-1)\) is positive and \(\lambda > 0\) is a parameter fulfilling an assumption. Similar results have already been proved in the case of i.i.d. matrices \(a_n\) and \(Y_n,\,b_n\) vectors by Kesten and Le Page. The proof is based on a new renewal theorem for systems of renewal equations as well as an extension of Grincevicius inequality (extension of Lévy's symmetrization inequality) to the Markovian case, given in the paper.