From $p$-Wasserstein Bounds to Moderate Deviations

TL;DR

This paper introduces a novel approach using $p$-Wasserstein bounds to establish moderate deviation results in normal approximations, extending to dependent variables and various applications.

Contribution

It develops a new method based on $p$-Wasserstein bounds for proving moderate deviations, applicable to dependent variables and diverse probabilistic models.

Findings

Recovers optimal deviation range for i.i.d. sums.

Achieves near-optimal error rates in normal approximation.

Extends method to dependent random variables and complex structures.

Abstract

We use a new method via $p$-Wasserstein bounds to prove Cram\'er-type moderate deviations in (multivariate) normal approximations. In the classical setting that $W$ is a standardized sum of $n$ independent and identically distributed (i.i.d.) random variables with sub-exponential tails, our method recovers the optimal range of $0\leq x=o(n^{1/6})$ and the near optimal error rate $O(1)(1+x)(\log n+x^2)/\sqrt{n}$ for $P(W>x)/(1-\Phi(x))\to 1$, where $\Phi$ is the standard normal distribution function. Our method also works for dependent random variables (vectors) and we give applications to the combinatorial central limit theorem, Wiener chaos, homogeneous sums and local dependence. The key step of our method is to show that the $p$-Wasserstein distance between the distribution of the random variable (vector) of interest and a normal distribution grows like $O(p^\alpha \Delta)$, $1\leq p\leq p_0$, for some constants $\alpha, \Delta$ and $p_0$. In the above i.i.d. setting, $\alpha=1, \Delta=1/\sqrt{n}, p_0=n^{1/3}$. For this purpose, we obtain general $p$-Wasserstein bounds in (multivariate) normal approximations using Stein's method.