Distance covariance for discretized stochastic processes (Q2203622)

For an iid sequence of continuous, bounded and measurable stochastic processes \((X_i,Y_i)\), \(i=1,2,\ldots\), on \([0,1]\), with generic element \((X,Y)\), the authors define the distance covariance \(T_\beta(X,Y)\) between \(X\) and \(Y\) by analogy with the distance covariance between vectors of data, for a parameter \(\beta\in(0,2]\). They show that \(T_\beta(X,Y)=0\) if and only if \(X\) and \(Y\) are independent. The corresponding sample distance covariance is denoted by \(T_{n,\beta}(X,Y)\). In the setting of testing for independence of stochastic processes, assume that we observe our processes \(((X_i,Y_i))_{i=1,\ldots,n}\) only at discrete time points \(t_0 < \cdots < t_p\) in \([0,1]\). This data is denoted by \(((X_i^{(p)},Y_i^{(p)}))_{i=1,\ldots,n}\). The main object of study in this paper is the sample distance covariance \(T_{n,\beta}(X^{(p)},Y^{(p)})\). The authors show that \[ n\left[T_{n,\beta}(X^{(p)},Y^{(p)})-T_{n,\beta}(X,Y)\right] \] converges in probability to zero as \(n\to\infty\) for independent \(X\) and \(Y\), provided that \(p\to\infty\) and the maximal distance between the \(t_i\) converges to zero sufficiently quickly. They also establish consistency of \(T_{n,\beta}(X^{(p)},Y^{(p)})\), as well as an asymptotic weighted \(\chi^2\) distribution for this statistic. Bootstrap procedures are also considered, and simulations are used to illustrate the main results.