ASRI: An Aggregated Systemic Risk Index for Cryptocurrency Markets

The Aggregated Systemic Risk Index (ASRI) is an interpretable, channel-decomposed measure of crypto-native systemic stress, aggregating four weighted sub-indices — Stablecoin Concentration Risk (30%), DeFi Liquidity Risk (25%), Contagion Risk (25%), and Regulatory Opacity Risk (20%) — into a single auditable composite. Methodologically it belongs to the financial-network tradition of connectedness and contagion measurement: the contagion channel is a cross-market connectedness construct, and a Diebold–Yilmaz (2012) connectedness series computed on the sub-indices serves as the network benchmark. The index is evaluated retrospectively against four crypto crises (Terra/Luna, Celsius/3AC, FTX, SVB); it is not a validated real-time early-warning system. Under autocorrelation-robust (HAC) inference referenced to a finite-sample fixed-b distribution, the three large-signal events are only borderline at the 5% level (p ≈ 0.04–0.06, none surviving Bonferroni); Terra/Luna is clearly non-significant. On a fair-baseline comparison on identical labels, ASRI's day-level discrimination (AUROC 0.866) is indistinguishable on marginal bootstrap intervals from its strongest single channel (Contagion, 0.851) and from the first principal component of its channels (0.858); a paired bootstrap separates them only by a small structural margin (≤0.016 AUROC, absent in AUPRC) reflecting ASRI nesting those baselines, and a standalone equity-volatility (VIX) series matches ASRI (0.875). The value of aggregation is therefore interpretive — channel attribution, lead-time, and regime structure in one transparent, reproducible composite — rather than superior discrimination. Version 2.0.0 (June 2026). Major revision: finite-sample (fixed-b) inference correction; fair-baseline demotion with a paired moving-block bootstrap analysis (small structural edge over the index's own channels, no gain over a standalone VIX series); removal of the leading/confirming Granger claim; and reproducibility hardening — the daily ASRI series is now provided as a frozen dataset of record and the generation-code bugs (hardcoded stablecoin peg, coin-ID mapping, running-maximum TVL offset) are corrected. Open code and data: github.com/studiofarzulla/asri. v3.0.0 (2026-07-11): Matches arXiv v4. Appendix T sample calculation replaced with the dataset-of-record reading (ASRI 38.1 on 2024-12-31) with explicit Eq. 6 arithmetic and provenance disclosure; ablation lead-time reconciliation note added (aggregate column vs Eq.-6 recomposition pipelines); network-contagion framing corrected to channel-decomposition adaptation; 10Y-Treasury duration-mismatch limitation added; frozen-parquet aggregate-vs-recomposition divergence disclosed (stored max 84.70 vs recomposed 81.06). The dataset of record (asri_history.parquet) and the full-recompute series (asri_open_full_20260711.parquet) are now deposited alongside the paper.