PRACTICAL GUIDE OF USING KENDALL’S τ IN THE CONTEXT OF FORECASTING CRITICAL TRANSITIONS

The paper correctly states the modified Mann–Kendall variance with the rank-ACF correction (its Eq. 3), and argues qualitatively that overlapping windows induce positive serial correlation and inflate Kendall’s τ variability; both points align with standard practice and the candidate solution’s claims. However, the paper provides no derivation or conditions for the Normal approximation beyond citing Hamed–Rao; assumptions (stationarity, weak dependence, tie handling) needed for a valid U-statistic CLT are omitted, and the implementation details for estimating the rank-ACF are left implicit. The candidate solution supplies the missing hypotheses and a principled justification via the Hájek projection and long-run variance, and gives a concrete procedure. Hence the paper’s argument is correct in thrust but incomplete in rigor, while the model’s solution is correct and more complete. Key paper touchpoints: independence variance (Eq. 2) and its inadequacy with moving windows, and the exact modified variance formula (Eq. 3) with ρS(i) defined as the rank autocorrelation, plus evidence that larger windows create stronger overlap-induced dependence and flatter τ distributions, all present in the PDF.