An Auditable Framework for Stablecoin Depeg Detection: Validation, Bias Quantification and Evidence from a Decade of Stablecoin Data

Abstract Stablecoins have become a central infrastructure of digital finance, yet the reliability of empirical conclusions about their stability depends on how peg deviations are measured. This article investigates the extent to which the frequency and severity of depegs depend on the detection rules adopted, proposing an auditable and reproducible framework for detection, validation and bias quantification. Drawing on a quantitative documentary approach with a longitudinal design, it analyses ten US dollar-pegged stablecoins with a daily price series in the Coin Metrics Community Data, between January 2016 and April 2026; the sample mixes active and discontinued coins and excludes UST and FDUSD for lack of price data.

The event count is highly sensitive to the rules: it varies from 191 to 21 when the magnitude threshold rises from 1% to 5%, and from 191 to 29 when a minimum duration of seven days is required. A threshold-selection criterion that maximises Youden's J identifies a 2.5% daily-close threshold as optimal against documented episodes. In addition, 21.4% of the raw events are post-discontinuation illiquidity artefacts, and centralised fiat-collateralised stablecoins reveal greater peg fidelity than crypto-collateralised ones, while market size does not imply greater stability.

A criterion-based validation reports a sensitivity of 1.00 at the any-event operating point and 0.375 at the severe operating point, with wide confidence intervals; a panel logit of depeg incidence is reported as an indicative, under-powered model. A second-source robustness test corroborates 13 of 21 severe episodes (61.9%) on an independent provider, leaving roughly two fifths uncorroborated. It is concluded that measured frequency and severity are not neutral properties of the market but functions of methodological and data-source choices.

JEL classification: G12; G23; G28; C55.