Paperlane

no. 16 · Tuesday 17 November 2026

Nonsense rank correlations between independent random walks

Zhongming Xie and Liping Zhu

math.PR

Abstract

Yule's nonsense correlation persists for Spearman's rho and Kendall's tau between independent random walks. Neither coefficient concentrates at zero as the sample size grows. For centred random walks with finite increment variances, we derive a universal joint limit expressed through Brownian occupation functionals. Integral formulas give limiting variances of approximately 0.222 and 0.108, respectively, below the Pearson value of 0.241. Rank correlations thus reduce the limiting dispersion on their respective coefficient scales. We prove that both marginal limits have symmetric densities and represent them exactly as limits of polynomial densities determined by Gaussian moment integrals. Spearman's conventional t test and Kendall's normal approximation are asymptotically invalid, with rejection probabilities tending to one under process independence.

fields
Probability
pdf ↗ page arXiv:2611.20572 ↗