A strong law of large numbers for triangular mixingale arrays (Q1916163)

Consider a triangular array of random variables \(\{X_{nt}, 1\leq t\leq n, n\geq 1\}\) and sigma-fields \(\{{\mathcal F}_{nt}\}\). Following \textit{Andrews} [Econometric Theory 4, 468-548 (1988)], for \(q\geq 1\), \(\{X_{nt},{\mathcal F}_{nt}\}\) is a triangular \(L_q\)-mixingale array if nonnegative real sequences \(\{c_{nt}\}\) and \(\{\psi(m), m\geq 0\}\) exist such that \(\psi(m)\to 0\), and, for all \(t\geq 1\), \(m\geq 0\), \(|E(X_{nt}\mid {\mathcal F}_{n,t-m})|_q\leq c_{nt}\psi(m)\) and \(|X_{nt}-E(X_{nt}\mid {\mathcal F}_{n,t+m})|_q\leq c_{nt}\psi(m+1)\). This paper presents sufficient conditions for an \(L_q\)-mixingale array to obey \(\sum^\infty_{n=1} P[|\sum^n_{t=1} X_{nt}|>\varepsilon n]<\infty\) for every \(\varepsilon>0\) and, hence, to obey the strong law of large numbers. This theorem is used to prove a pointwise strong convergence property of kernel estimators.