On the ergodicity of \(TAR(1)\) processes (Q1182687)

The authors consider the ergodicity of first-order autoregressive models or TAR(1)-models. Although the dynamic is linear at each step (with additive iid noise), \(X_ n= c_ i X_{n-1}+ e_ n\), the model is nonlinear, since the factor \(c_ i\) (\(i=1\) or 2) depends on the sign of \(X_{n-d}\), \(d\) the delay parameter. Given iid and integrable random variables \(e_ n\), for general delay \(d\), the authors describe the domain in the parameter space \((c_ 1,c_ 2)\) for which \((X_ n)\) is geometrically ergodic: Necessary and sufficient conditions are given in terms of \(c_ 1\) and \(c_ 2\).