Approximating the null distribution of a class of statistics for testing independence (Q2423537)

In the paper, a class of tests is considered for the independence of two random variables $X$ and $Y$. The tests proposed are related with values of the quantity \[ T_n(w)=n\|C_n(t,s)-C_n(t,0)C_n(0,s)\|_w^2, \] where $\|\cdot\|$ stands for the norm in the separable Hilbert space \[ L_2(w)=\left\{f:\mathbb{R}^2\rightarrow\mathbb{C}:\|f\|_w^2=\int\int |f(t,s)|^2w(t,s)\text{d}t\text{d}s<\infty\right\}, \] $w$ is a non-negative weight satisfying some regularity conditions and $C_n$ is the empirical characteristic function related with sample $(X_1,Y_1),\ldots, (X_n,Y_n)$ consisting of independent copies of the vector $(X,Y)$. More exactly, \[ C_n(t,s)=\frac{1}{n}\sum\limits_{j=1}^n\exp\{\mathbf{i}tX_j^*+\mathbf{i}sY_j^*\}, \] where $(X_1^*,Y_1^*)=(X_{\pi(1)},Y_{\sigma(1)}), \ldots, (X_n^*,Y_n^*)=(X_{\pi(n)},Y_{\sigma(n)})$, and $(\pi(1),\ldots,\pi(n))$, $(\sigma(1),\ldots,\sigma(n))$ are two arbitrary permutations of the indices $(1,\ldots, n)$. \par The consistency of the proposed estimators is analysed, a simulation study is presented, and an application to real data set for testing independence is included.