$p$-Wasserstein distance and Cramér-type moderate deviations for the multi-dimensional elephant random walk

Abstract

In this paper, we establish $p$-Wasserstein bounds and Cramér-type moderate deviations for the multi-dimensional elephant random walk in the diffusive and critical regimes. Using a Gaussian convolution estimate with random covariance and the recursions for the conditional covariance, we obtain $W_{p,2}(\mathcal L(\widehat T_n),N(0,I_d))\le C_{a,d}pL_n$, where $\widehat T_n$ is normalized by its exact covariance and \[ L_n\asymp \begin{cases} n^{-1/2}\log n,&a\le1/3,\\ n^{-3(1/2-a)},&1/3<a<1/2,\\ (\log n)^{-3/2},&a=1/2. \end{cases} \] Consequently, we obtain radial Cramér-type moderate deviations for $x=o(L_n^{-1/3})$. We also determine the covariance bias under the uncorrected normalization and establish the corresponding Wasserstein bounds, lower bounds and moderate deviation estimates. The dimension and the model parameters are fixed.