On the convergence of moments in the almost sure central limit theorem for martingales with statistical applications (Q2485756)

Let \((\varepsilon_n)\) be a martingale difference sequence and \((\phi_n)\) a sequence of random variables both adapted to a filtration \(({\mathcal F}_n).\) Then it was shown earlier by \textit{F. Chaabane} [C. R. Acad. Sci., Paris, Sér. I 323, 195--198 (1996; Zbl 0858.60023)] that for the martingale transform \(M_n=\sum_{k=1}^n \phi_{k-1}\,\varepsilon_k\) an almost sure central limit theorem (ASCLT) with a bounded and continuous function \(f\) holds, i.e., \[ \lim_{n \to \infty} (\log s_n)^{-1}\sum_{k=1}^n (\phi_k^2/s_k)\; f(M_k/\sqrt{s_{k-1}})\overset{a.s.}{=}\int f(x)\, d{\mathcal N}_{0,\sigma^2}(x) \tag{1} \] where \(s_n=\sum_{k=1}^n \phi_k^2\,,\) provided some conditions hold. The main result of the present paper is the extension of (1) to functions which may increase like \(| x| ^{2p}\) with some power \(p\,.\) Some applications are given concerning estimators of moments of the residuals in a linear regression model \(X_k=\theta\phi_{k-1} +\varepsilon\), \(k=1,\dots,n,\) where the residuals form a martingale difference sequence.