Stationary Markov chains with linear regressions. (Q1888763)

In a square-integrable stationary Markov chain \(X=(X_k)\) with the linear regression function \(L\), \(E(X_k\mid X)=L(X_{k-1},X_{k+1})\), and the quadratic conditional variance function \(Q\), \(E(X_k^2\mid X)=Q(X_{k-1},X_{k+1})\), the coefficients of \(Q\) are shown to obey one of two linear constraints. Under the first constraint, the remaining coefficients satisfy one of two inequalities. Certain Markov chains are uniquely characterized through values of the coefficients.