A Reproducible Stability Audit of Multi-Criteria ETF Ranking for FinTech Decision Support

Abstract

Digital-investment decision support can reduce heterogeneous exchange-traded fund (ETF) risk–return signals to a point ranking, but that order can hide uncertainty from sampling, normalization, criterion choice, market regimes, and the alternative set. We present a retrospective, reproducibilityfocused audit of 60 ETFs using public adjusted prices from 2015 through 2025. Six risk–return criteria are evaluated by Entropy-TOPSIS and diagnostic comparators. The Primary case uses 1,000 movingblock bootstrap samples, rank intervals, top-five probabilities, and specification, temporal, weight, criterion, and universe stress tests. Entropy-TOPSIS obtains mean bootstrap Spearman agreement of 0.8902 and a mean 95% rank-interval width of 5.4167 positions. The main grouping audit uses 20 outcome-blind partitions of the same 60-ticker pool: five disjoint 12-ETF panels per partition and 500 draws per panel. Its auxiliary matched worst-gap summary averages 0.3999. In disjoint fiveyear windows, the same audit yields means of 0.3661 and 0.4319, while 100 early-versus-late fullrank correlations average 0.4026. These results characterize internal regrouping in one USD pool; they neither establish method superiority nor provide a personalized investment recommendation. The contribution is a reproducible audit and reporting protocol around standard ranking methods, not a new ranking algorithm or deployed platform.