Testing for independence by the empirical characteristic function (Q1070713): Difference between revisions

Let \(C(t_ 1,...,t_ p)\) be the characteristic function of a p-variate distribution F(x) and let \(C_{[k]}(t_ k)=C(\underline 0,t_ k,\underline 0)\) be its k th marginal characteristic function. Then the hypothesis of total independence of p components can be expressed as \(H_ 0: C(t_ 1,...,t_ p)=\prod^{p}_{k=1}C_{[k]}(t_ k).\) Let \(C_ n(t)\) for \(t\in R^ p\) be the empirical characteristic function based on a random sample \(X_ 1,...,X_ n\) from F(x) and let \(C_{n[k]}(t_ k)\) be its k th marginal empirical characteristic function. The author gives a necessary and sufficient condition for the weak convergence of the process, under \(H_ 0\), \[ S_ n(t)=n^{1/2}\{C_ n(t)-\prod^{p}_{k=1}C_{n[k]}(t_ k)\}\equiv S_ n^{(\quad 1)}(t)+iS_ n^{(2)}(t) \] to a Gaussian process in a space of continuous functions on a compact set in \(R^ p\). Then he gives a sufficient condition for the weak convergence of a quadratic form based on \(S_ n^{(1)}(t^{(n)})\) and \(S_ n^{(2)}(t^{(n)})\) to the \(\chi^ 2_ 2\)-distribution under \(H_ 0\) where \(t^{(n)}\) is an estimate of \(t^ 0\) that maximizes the total variance function of the limiting Gaussian process on a given compact set.