Stablecoins as Intra-Crypto Safe Havens? Dynamic Correlations and Flight-to-Quality in Digital Asset Markets

Technical Summary: Stablecoins as Intra-Crypto Safe Havens? Dynamic Correlations and Flight-to-Quality in Digital Asset Markets

Problem Statement
Traditional financial markets possess canonical safe havens (e.g., gold, sovereign bonds) that decouple from equities during stress. Digital asset markets lack an internal equivalent to government bonds, and Bitcoin has historically failed safe-haven tests against equities. Consequently, USD-pegged stablecoins are widely treated as the internal refuge of the crypto ecosystem. However, existing literature offers conflicting evidence: early studies (pre-2020) classified stablecoins as havens, while subsequent research documents their fragility and peg deviations. A critical gap exists in understanding whether the "haven" property is intrinsic to stablecoins or contingent on the specific type of crisis and the coin's collateral design. This paper addresses whether stablecoins deliver on their design promise as a refuge across the full sequence of recent market crises: the COVID-19 liquidity shock (March 2020), the Terra collapse (May 2022), the FTX exchange failure (November 2022), and the Silicon Valley Bank (SVB) episode (March 2023).

Methodology
The study utilizes daily data from January 2018 to April 2026, covering nine USD-pegged stablecoins (USDT, USDC, DAI, TUSD, PAX, GUSD, HUSD, BUSD, USDK) against two risk anchors: Bitcoin (BTC) and Ether (ETH). The analysis employs a unified framework to test four hypotheses regarding hedge status, crisis dependence, activity vs. price responses, and design mechanisms.

1. Dynamic Correlations: Rolling correlations (60-day baseline, with 30/90-day sensitivity checks) serve as the primary transparent instrument to visualize comovement shifts during crisis windows.
2. Safe-Haven Regression: The paper applies the Baur and McDermott (2010) regression taxonomy, distinguishing between a hedge (uncorrelated/negative on average), a diversifier (positive but imperfect), and a haven (uncorrelated/negative during stress). Strong havens exhibit significant negative tail comovement; weak havens show comovement indistinguishable from zero.
3. Crisis Interaction Models: To test episode-specific behavior, the authors interact risky asset returns with indicators for the four distinct crisis windows.
4. Robustness and Inference: Given the short duration of crisis windows (24–67 days) and heavy-tailed returns, the study employs wild bootstrap inference (Rademacher weights, 4,999 replications) alongside standard HAC (Newey-West) errors to ensure statistical validity. Dynamic Conditional Correlation (DCC) models are used as a parametric check.
5. Flight-to-Quality Tests: The study distinguishes between "flight to activity" (trading volume) and "flight to price" (peg premiums). Abnormal activity is measured via both reported exchange volume and manipulation-proof on-chain transfer counts.
6. Fidelity vs. Capacity: A Spearman rank correlation tests the link between average peg fidelity (mean absolute deviation) and tail comovement capacity.

Key Results
The analysis yields four primary findings that challenge static classifications of stablecoins:

• The Refuge Rotates: No single stablecoin acts as a safe haven across all four crises. The identity of the "haven" coin depends entirely on the origin of the stress.
  • COVID (2020): DAI (crypto-collateralized) exhibited the strongest negative tail comovement (-0.092 slope vs. BTC), driven by forced deleveraging.
  • Terra (2022): No major coin showed significant negative comovement; the episode was characterized by peg deviations (USDT traded at a discount) rather than haven behavior.
  • FTX (2022): Second-tier coins (GUSD, PAX) showed directional evidence of haven properties, though significance varied by inference method.
  • SVB (2023): A complete inversion occurred. USDC and DAI (reserves exposed to banking) fell to discounts exceeding 300 basis points. Conversely, USDT (reserves held outside the banking system) rose to a premium and became the only coin with a significant negative crisis slope, effectively acting as the haven.
• Flight to Quality is Universal in Activity, Selective in Price: Stress days trigger a universal increase in trading activity across all nine stablecoins, confirmed by on-chain transfer data (ruling out wash trading). However, price premiums (flight to price) are selective. Only specific designs, notably DAI during the 2020 crash, converted this demand into significant price premiums (reaching 663 basis points). USDT showed no significant price change under stress in most episodes.
• The Collateral Mechanism: The strongest intra-crypto haven (DAI) operates via a mechanical deleveraging channel. When its collateral (ETH) crashes, holders of collateralized debt positions are forced to buy DAI to unwind leverage, creating demand exactly at the point of maximum stress. This results in negative tail comovement despite DAI having one of the weakest average peg fidelities.
• Fidelity Does Not Predict Haven Capacity: There is no monotonic relationship between average peg fidelity and crisis protection. USDC, which maintained the tightest average peg (2.4 bps deviation), suffered the largest crisis discount in the SVB episode. Conversely, DAI, with a higher average deviation, provided the strongest tail protection.

Key Contributions

1. Unified Crisis Testing: This is the first study to test the haven classification across the complete 2020–2023 crisis sequence within a single framework, reconciling early positive findings with later fragility evidence by demonstrating that the classification is episode-dependent.
2. Design-Specific Mechanisms: The paper identifies the "collateral channel" as a distinct mechanism for safe-haven behavior, showing how crypto-collateralized designs can generate demand during the very stress events that crash their underlying assets.
3. Methodological Rigor: The study highlights that standard asymptotic inference is insufficient for short crisis windows with heavy tails, advocating for wild bootstrap methods to avoid false positives in haven classification.
4. Distinction of Properties: It empirically decouples "peg fidelity" (an average property) from "refuge capacity" (a tail property), arguing that they are distinct design outcomes.

Significance and Claims
The paper claims that the safe-haven property of stablecoins is not intrinsic but contingent on two factors: the state of the market (specifically the origin of the crisis) and the specific design of the coin (collateral type and reserve custody).

• For Investors: The results argue against concentrating refuge allocations in the coin with the tightest historical peg. Instead, a diversified refuge strategy across different reserve exposures is necessary, as the coin that anchors the system in one crisis may be the epicenter of the next.
• For Regulators: The study challenges regulatory frameworks that prioritize peg fidelity metrics as proxies for crisis stability. It posits that rules concentrating reserves in bank deposits (as seen with USDC) may inadvertently create the very systemic risk that leads to crisis discounts. A supervisory framework must monitor the concentration of reserve custody and stress-test for specific crisis origins, rather than relying solely on average peg maintenance.
• For Literature: The paper resolves the tension between "haven" and "fragility" literature by demonstrating that both are accurate descriptions of the same market, provided the crisis context is acknowledged.

The authors conclude modestly that while the haven role exists, it is dynamic. The "flight to quality" in digital assets is a flight to activity that is universal, but a flight to price that is highly selective and design-dependent.