High-dimensional central limit theorems for homogeneous sums (Q6042056)

Given a sequence \(\mathbf{X}=X_1,X_2,\ldots\) of independent random variables with zero mean and unit variance, define the homogeneous sum \[ Q(f;\mathbf{X})=\sum_{i_1,\ldots,i_q=1}^Nf(i_1,\ldots,i_q)X_{i_1}\cdots X_{i_q}\,, \] for positive integers \(N\) and \(q\), and \(f:\{1,\ldots,N\}^q\to\mathbb{R}\) a symmetric function which vanishes unless all its arguments are distinct. The present paper establishes a quantitative multivariate central limit theorem for \(d\)-dimensional vectors of such (possibly correlated) random sums in the setting where \(d=d_n\to\infty\) as \(n\to\infty\). Letting \(\mathbf{Q}(\mathbf{X})\) be such a vector, one of the main results is a bound on \(\sup_{x\in\mathbb{R}^d}|P(\mathbf{Q}(\mathbf{X})\leq x)-P(\mathbf{Z}\leq x)|\), where \(\mathbf{Z}\) is a \(d\)-dimensional Gaussian random vector with an appropriate covariance structure. The author also establishes corresponding multivariate versions of the fourth-moment theorem, universality results and Peccati-Tudor theorems for such homogeneous sums. These results are illustrated by comparison to existing results in the literature and by a statistical application to testing for the absence of a lead-lag relationship.