- The paper finds that EigenLayer TVL, L2 activity, and ezETH yield drive Renzo’s revenue, while mainnet TVL, tokenization events, market returns, and fee volatility are not significant predictors.
- The paper shows that bridged ezETH creates concentrated cross-chain exposure, with Linea holding 81.08% of bridged supply and Ether.Fi accounting for 65.4% of the LRT market.
- The paper estimates that a 3.33% ezETH depeg could liquidate maximum-LTV Aave positions, overwhelming local liquidity, although current restaking remains below Ethereum consensus-risk thresholds.
Overview and Motivation
This paper examines the economic drivers and systemic risk profile of liquid restaking, a DeFi service layer in which already-staked ETH is reused to secure additional applications (actively validated services, or AVSs) via protocols such as EigenLayer, while liquid restaking protocols (LRPs) issue liquid restaking tokens (LRTs) that represent these positions. The authors—Sevim and Torres—combine an econometric study of revenue generation at Renzo Protocol with an on-chain assessment of cross-protocol contagion risk arising from bridged LRTs. The restaking and liquid-restaking sectors jointly held over 28 billion USD in total value locked (TVL) as of October 2025, motivating a rigorous treatment of both the business model and its security externalities.
The paper's central thesis is that the economic and security dimensions of liquid restaking cannot be analyzed in isolation: multi-blockchain expansion of LRTs drives revenue but simultaneously propagates bridge and interconnection risks across the ecosystem.
Protocol Taxonomy and Scope
The paper catalogs the restaking landscape as of October 4, 2025, distinguishing infrastructure layers (EigenLayer at $19.00B TVL, Babylon at$7.07B, Symbiotic at $1.22B) from LRPs (Ether.Fi's eETH at$11.21B, Kernel DAO's rsETH at $1.97B, Renzo's ezETH at$1.55B). Renzo is selected for the econometric analysis because of its early and extensive multi-chain deployment (12 chains), superior data availability on The Graph, and relative under-examination compared to Ether.Fi, which dominates prior literature.
A methodological contribution worth noting is the careful de-duplication of TVL figures. Following concerns raised by Chaudhry [(2604.03274)'s cited work] and Luo et al., the authors construct uninflated TVL measures: EigenLayer TVL excludes LRP deposits, and Renzo's Ethereum TVL excludes bridged ezETH representations on L2s, since bridged tokens are claims on mainnet assets rather than additional value.
Econometric Methodology
Revenue dynamics are modeled with OLS regressions over daily data from January 22, 2024 to April 17, 2025 (452 observations), with three lag specifications (contemporaneous, t−1, t−2). The dependent variable is log total revenue; regressors include EigenLayer TVL (TVL0), Renzo mainnet TVL (TVL1), aggregated L2 TVL (TVL2), ezETH yield, ezETH price premium, market share, and controls for stETH APY, ETH returns, transaction fee volatility, the Fear and Greed Index, and tokenization event dummies. Findings are triangulated with Granger causality tests and random forest feature importance. Stationarity is verified via ADF tests, multicollinearity ruled out (all VIFs below 1.6), and robustness established through winsorization and HC3 standard errors.
Revenue Determinants
The results confirm three hypotheses and reject two:
| Hypothesis |
Variable |
Result |
Baseline elasticity |
| H2 supported |
EigenLayer TVL (TVL0) |
Significant (OLS p=0.001; Granger p<0.0001) |
+0.98% revenue per 1% TVL |
| H3 supported |
L2 TVL (TVL2) |
Significant; top random forest feature (Gini 0.2598) |
+0.08% per 1% |
| H4 supported |
ezETH yield |
Significant (Granger p<0.0001) |
+3.82% per 1% |
| H1 rejected |
Renzo mainnet TVL (TVL1) |
Insignificant across all tests |
— |
| H5 rejected |
Tokenization events |
Insignificant (though Chow test flags structural break at REZ TGE) |
— |
Two findings stand out. First, EigenLayer's TVL—not Renzo's own mainnet TVL—is the dominant parametric predictor, implying that LRP revenue is driven by ecosystem-wide growth rather than protocol-specific capital accumulation. Second, the rejection of H1 contrasts sharply with the significance of TVL2: active-user data show that usage concentrates on L2 deployments while mainnet activity remains thin, suggesting that user adoption on secondary networks, not locked value per se, sustains revenue. The yield coefficient is notably strong, reaching 24.02% revenue response per 1% yield change in the winsorized t−1 specification—the model with peak explanatory power (R² = 0.519).
Broader market controls (ETH price returns, transaction fee volatility) are insignificant, indicating that Renzo's revenue is insulated from general crypto-market fluctuations and tethered instead to restaking-specific dynamics. The premium variable is likewise insignificant throughout.
Interconnected Risk Assessment
The security analysis maps asset flows across staking, restaking, lending, AMM, and yield protocols. As of October 2025, total restaked TVL (~$19.23B, of which 72.79% is natively staked ETH rather than stETH) represents roughly 13% of all staked ETH (35.7M ETH), well below the ~33% threshold associated with threats to finality. The paper therefore concludes that cascading slashing through restaking does not currently pose a consensus-level security risk to Ethereum—a deliberately conservative claim grounded in current scale.
Bridged ezETH constitutes 90,862 tokens (~30.18% of total ezETH supply, ~2.1% of the LRT market cap) across Linea, Arbitrum, Base, Blast, and BSC, with Linea alone holding 81.08% of bridged ezETH (~24% of total supply). This concentration exposes Renzo to Linea-specific rollup risks—sequencer centralization, proof-validation failure, compromised multisigs. At current scale, however, the paper finds no systemic threat to the broader staking ecosystem. Concentration risk at the protocol level is also flagged: Ether.Fi's eETH holds 65.4% of the LRT market, a single point of failure should a dominant issuer be compromised.
Stress Test Results
The paper's most striking quantitative result comes from a liquidation stress test using Aave v3 Linea parameters (LTV = 72.5%, LT = 75%). Under worst-case assumptions (borrowing enabled at maximum LTV), the critical depeg threshold is:
$1.22B) from LRPs (Ether.Fi's eETH at$0
A mere 3.33% decline in bridged ezETH's price would render all maximum-LTV Aave positions liquidatable. With 64,890 ezETH posted as collateral on Aave v3 Linea but only 149 ezETH in native DEX liquidity (Etherex)—a liquidity coverage ratio of 0.23%—liquidation sell pressure could not be absorbed locally. Even aggregating all mainnet ezETH/stETH liquidity (~5,251 ETH equivalent across Uniswap, Balancer, Curve) covers only 8.09% of potential liquidation volume, implying substantial price impact and forced propagation back to mainnet. Redemption pressure transmitted to Lido would correspond to roughly 0.76% of liquid-staked ETH—not consensus-threatening, but sufficient to generate short-term validator exit pressure and volatility amplification. The November Balancer exploit (~$120M loss) is cited as evidence that such cross-contamination events are realistic rather than hypothetical.
An important caveat applies: at observation time, ezETH borrowing on Aave v3 Linea was disabled with 0% reserve utilization, so the stress test reflects a hypothetical re-enablement scenario rather than realized exposure.
Limitations
The authors are candid about several constraints. Model explanatory power is moderate (baseline R² ≈ 0.31), leaving substantial unexplained variance, and the relationships identified are exploratory rather than causal—co-integration testing is suggested as a needed extension. The analysis covers a single LRP over roughly fifteen months of a rapidly evolving market, limiting generalizability across protocols and cycles. Subgraph data may contain inconsistencies, and the youth of the restaking sector restricts long-horizon inference. The stress test itself rests on strong assumptions (maximum-LTV positioning, borrowing enabled, oracle behavior under stress) that bound its realism.
Conclusion
This paper delivers two complementary contributions. Econometrically, it establishes that LRP revenue is driven by underlying infrastructure growth (EigenLayer TVL), LRT yield, and multi-chain deployment—with the counterintuitive result that the protocol's own mainnet TVL is not a significant predictor. From a risk perspective, it quantifies how the same multi-chain strategy that generates revenue introduces bridge-dependent contagion channels, demonstrating through stress testing that even modest depegs of bridged LRTs can trigger liquidation cascades far exceeding available liquidity, while confirming that current restaking scale remains below thresholds threatening Ethereum's consensus security. The open questions left by the paper concern whether these findings generalize to other LRPs, how contagion scales as restaked TVL grows toward security-relevant proportions, and which complementary methodologies—network contagion models, agent-based simulations—can best capture the propagation dynamics the authors only begin to quantify.