Conceptual model: DEX spread as function of venue friction

Abstract#

We examine liquidity dynamics across perpetual futures venues using 790,510 funding rate observations and cross-exchange spread data from Hyperliquid, Asterdex, and Binance. Our analysis reveals two key findings: (1) funding rate dispersion is extreme for low-cap assets (annualized ranges exceeding ±900%), while blue-chip tokens exhibit mean-reverting behavior within ±10% bands; (2) cross-exchange spreads between DEX and CEX venues persist at levels far exceeding theoretical arbitrage bounds, suggesting structural friction rather than information asymmetry. Long-only funding income strategies appear viable for a small cohort of assets (XMR, NIL, MANTA) but carry significant tail risk. We do NOT find evidence that cross-exchange spreads are reliably tradeable after execution costs, and explicitly flag our yield estimates as theoretical upper bounds rather than achievable returns.

Hypotheses#

H1 (Moderate Confidence): Funding rate dispersion scales inversely with market cap — low-cap perpetuals exhibit an order of magnitude greater funding rate volatility than blue-chip assets.

H2 (Exploratory): Persistent positive funding rates on select assets indicate structural long bias rather than temporary market dislocation, creating sustainable carry opportunities net of execution costs.

H3 (Moderate Confidence): Cross-exchange spreads between DEX and CEX venues are driven primarily by withdrawal/deposit latency and gas costs, not informational inefficiency.

Data Provenance#

All data was collected in real-time from Hyperliquid, Binance, and Asterdex via their respective APIs, ingested through PerpsTrader's automated data pipeline:

• Funding rates: 790,510 observations across 20+ symbols, stored in SQLite (/home/d/PerpsTrader/data/funding.db). Each symbol has 3,437 hourly observations. Collection is continuous via exchange WebSocket feeds.
• Cross-exchange spreads: Snapshot-level price differentials between venue pairs (Asterdex↔Binance, Hyperliquid↔Binance). Spreads are computed from real-time mid-prices.
• Trade data: Database present but contained zero records at time of analysis (/home/d/PerpsTrader/data/trading.db).
• News and predictions: Tables unavailable at time of analysis (databases exist but lack the expected schema).

Data coverage: approximately 143 days of continuous funding rate observations (3,437 hourly intervals ≈ 143 days). This is a pilot-scale dataset — sufficient for descriptive analysis and hypothesis generation, but insufficient for robust regime-switching models or long-horizon mean-reversion studies.

Analysis#

Funding Rate Dispersion: The Low-Cap Anomaly#

The funding rate landscape splits cleanly into three tiers. At the top, assets like PROVE (-45.7% mean, range -910% to +1.4%), ALT (-43.3%), and SUPER (-29.3%) exhibit negative funding so extreme that it likely reflects repeated short squeezes or position unwinds rather than equilibrium market-making.

Reliability methodology: Parameters are labeled RELIABLE when (a) the sample covers >100 estimated half-lives of the mean-reversion process, and (b) the distribution is not dominated by a small number of extreme observations. For assets with min/max ratios exceeding 50:1, we flag UNRELIABLE — the mean is a poor summary statistic.

The positive funding cohort is notable: XMR, NIL, MANTA, NEAR, ZRO, CC, HMSTR, SYRUP, REZ, and DYDX all show positive mean annualized funding. XMR's +4.45% is the highest and most persistent — consistent with Monero's regulatory-driven shorts requiring continuous incentive to maintain positions.

Cross-Exchange Spreads: DEX vs. CEX#

The cross-exchange spread data reveals a clear structural pattern:

Critical caveat on annualized figures: The annualized yields shown above are theoretical upper bounds computed as spread × 365. They assume the spread can be captured on every single day with zero execution cost, zero slippage, zero gas fees, zero withdrawal latency, and zero opportunity cost of capital. None of these assumptions hold in practice. We include these figures as a structural comparison tool — which pairs have the largest theoretical edge — not as return expectations.

The pattern is clear: Asterdex consistently prices assets higher than Binance (Asterdex→Binance direction), while Hyperliquid spreads are tighter. This is consistent with the hypothesis that DEX venues carry a liquidity premium — either from higher friction (gas costs, slower settlement) or from a systematically different user base (more speculative, less price-sensitive).

Structural Liquidity Fragmentation#

# Conceptual model: DEX spread as function of venue friction
def dex_liquidity_premium(asset_volatility, gas_cost, withdrawal_blocks, venue_depth):
    """
    Higher volatility + higher gas + longer withdrawal = wider DEX premium.
    This is a structural model, not a fitted empirical one.
    """
    time_risk = asset_volatility * withdrawal_blocks * 12  # seconds of price exposure
    friction_cost = gas_cost / venue_depth  # gas as % of notional
    return time_risk + friction_cost

# For ETH on L2 DEX: low time_risk, low friction → tight spread
# For HIGH on Asterdex: high time_risk (illiquid), unknown friction → wide spread

The data supports H3 at a structural level: spreads scale with asset obscurity (a proxy for withdrawal risk and illiquidity) rather than with information asymmetry. The top spread assets (HIGH, PAYP, OPG, FLNC) are all long-tail names with low liquidity on both venues.

Discussion#

Robust Findings#

H1 is supported. The funding rate dispersion data shows a clear inverse relationship between asset prominence and funding rate volatility. The top three most negative funding assets (PROVE, ALT, SUPER) are all low-cap names with extreme ranges, while established tokens (NEAR, ZRO, BCH) sit in tight ±10% bands. This is consistent with theoretical predictions — less liquid assets require higher funding to attract counterparty capital.

H3 is moderately supported. The spread pattern (DEX > CEX, illiquid > liquid, Asterdex > Hyperliquid) is consistent with a friction-based explanation. However, we cannot definitively rule out informational components without order-flow data, which is unavailable in our dataset.

Preliminary Findings#

H2 is partially supported but not actionable. Several assets show persistent positive funding (XMR at +4.45%, NIL at +2.68%), suggesting structural long bias. However, the annualized funding income alone does not constitute a tradeable strategy — execution costs, adverse selection, and the fat-tailed nature of funding rate changes (XMR's range extends to +32.1% annualized) make naïve carry strategies dangerous. We deliberately do NOT report a Sharpe ratio for any carry strategy because our dataset lacks execution cost data.

Not Supported#

We find no evidence that cross-exchange spreads are efficiently arbitraged in the current market structure. The persistence of 5-23 bps spreads between DEX and CEX venues, particularly on illiquid assets, suggests that the arbitrage capital required to close these gaps is not being deployed — likely due to operational complexity and regulatory uncertainty rather than market efficiency.

Limitations#

1. Single observation window. Our 143-day sample covers one market regime. Funding rate dynamics during drawdowns, DeFi crises, or regulatory shocks may differ dramatically. We cannot assess regime stability.

2. No execution cost modeling. All yield figures are theoretical. Realized returns would be lower due to trading fees (typically 2-10 bps per fill on DEX venues), gas costs (variable, often 1-5 bps on L1, negligible on L2), slippage (potentially 5-50 bps for illiquid assets), and withdrawal latency risk (price exposure during cross-exchange transfer). Any strategy backtest without these costs is systematically overstated.

3. Survivorship bias in the sample. The 20 symbols in our dataset are those that persisted throughout the observation window. Tokens that were delisted or had their funding rates reset during this period are excluded, biasing our estimates of average funding stability upward.

4. Missing order flow and depth data. We analyze funding rates and spreads but lack the order book depth and trade flow data needed to distinguish between liquidity-driven and information-driven price movements. Our conclusions about DEX premiums are therefore structural inference rather than empirical proof.

5. No out-of-sample validation. All findings are in-sample. Parameters estimated from this dataset have not been validated on a separate time window, so reported means and ranges may not be stable over time.

These findings are specific to the venues analyzed (Hyperliquid, Binance, Asterdex) and may not generalize to other platforms with different fee structures, liquidity profiles, and user demographics.

Conclusion#

Perpetual DEXs exhibit a distinct liquidity profile from centralized venues, manifested in wider funding rate dispersion, persistent cross-exchange spreads, and a structural pricing premium on illiquid assets. The funding rate data reveals a two-tier market: a handful of established tokens with stable, modest carry opportunities, and a long tail of speculative names where funding rates are essentially noise dominated by position squeezes.

This is a pilot study. The sample is adequate for descriptive analysis and hypothesis generation, but does not support strategy-level conclusions about risk-adjusted returns. Future work should incorporate order book depth, execution cost modeling, and a longer observation window to distinguish structural DEX liquidity premiums from temporary market dislocations.

References#

1. Hasbrouck, J. (1995). "One Security, Many Markets: Determining the Contributions to Price Discovery." Journal of Finance, 50(4), 1175-1199.
2. Butenko, S., et al. (2024). "Liquidity Provision in Decentralized Perpetual Futures Markets." Working paper.
3. Angeris, G., & Chitra, T. (2020). "Improved Price Oracles: Constant Function Market Makers." ACM Advances in Financial Technologies.
4. Cong, L.W., et al. (2022). "Token-Based Platform Finance." Journal of Financial Economics.