Title: How the instability of ranks in non-central limit theorems affects large-sample inference under long memory

Abstract: Mathematically, under long memory, one can obtain so-called non-central limit theorems involving non-Gaussian limits. In these non-central limit theorems, a positive integer called "Hermite rank" plays a crucial role. The non-Gaussian limit appears only when the Hermite rank is greater than one. From a practical point of view, however, we argue that an Hermite rank greater than one is unstable. In contrast, an Hermite rank equal to one is stable. We provide empirical evidence supporting this argument. Assuming a higher-order Hermite rank when it is not really there will result in underestimating the order of the fluctuations of the statistic of interest. Examples are given to illustrate what happens when large-sample theories exclude higher-order Hermite ranks.