← Back to Trade Fill Archive

Cross-Exchange Alpha Analysis

ETH microstructure comparison between Hyperliquid L1 and Binance Spot. One day of tick data, every fill on both venues, aligned and dissected.
Hyperliquid L1 Binance Spot ETH May 25, 2025

1. Price Spread (HL vs Binance)

HL consistently trades ETH at a discount to Binance. This is structural, not noise.
Mean Spread
-3.8 bps
HL trades below Binance on average
Max Spread
17 bps
Maximum observed divergence
Std Deviation
1.97 bps
Spread volatility (tight)
Signal
Persistent
Discount is structural, not random
A mean spread of -3.8 bps with std dev of 1.97 bps means the discount is ~1.9 standard deviations from zero. On $1M notional, this is a $380 edge per round-trip before fees. The question is whether this compensates for HL's higher execution risk and lower liquidity.

2. Lead-Lag Correlation

Cross-correlation of 10-second return buckets. Binance slightly leads Hyperliquid, but both carry signal.
Contemporaneous (0s)
0.8161
BN leads HL by 10s
0.2804
HL leads BN by 10s
0.2215
BN leads HL by 20s
0.0005
HL leads BN by 20s
0.0099
Binance leads HL at the 10s horizon (0.28 > 0.22). Price discovery happens on the deeper book first.
Information is fully priced within 20 seconds. Both lag-20 correlations are near zero. No stale signal beyond that window.

3. Smart Money Leaderboard

Top 15 wallets by realized PnL on Hyperliquid ETH, May 25 2025. Sorted by total closedPnl.
# Wallet Trades Win% Total PnL Avg PnL/Trade Notional Coins
1 0x225864ad... 1,632 99.9% +$3,133,784 $1,920 $182M 2
2 0x8af700ba... 289 100.0% +$1,607,539 $5,562 $116M 1
3 0xffffffff... 27,710 80.1% +$1,519,118 $55 $5.4M 116
4 0xd9a1ed9a... 263 95.8% +$776,363 $2,952 $51M 2
5 0x1d52fe9b... 3,029 95.8% +$562,003 $186 $22M 12
6 0x4a207d28... 600 99.0% +$518,627 $864 $1.1M 2
7 0x507e0d7c... 319 100.0% +$506,171 $1,587 $1.8M 1
8 0xdf052b53... 330 100.0% +$492,887 $1,494 $881K 1
9 0x4a96036f... 473 100.0% +$459,817 $972 $46M 1
10 0x2ba553d9... 2,467 86.3% +$457,487 $185 $41M 5
11 0x023a3d05... 25,354 74.7% +$432,459 $17 $135M 138
12 0xecb63caa... 39,788 56.7% +$392,183 $10 $119M 48
13 0x629af4d7... * 449 1.8% +$372,695 $830 $61M 1
14 0x210c6a8d... 120 100.0% +$344,697 $2,872 $454K 1
15 0xf9109ada... 26,017 69.2% +$335,984 $13 $109M 30
* See Section 4 for the 1.8% win rate anomaly

4. The 1.8% Win Rate Whale

The most interesting wallet on the leaderboard is the one that loses 98.2% of its trades.
0x629af4d7... — Textbook Breakout / Trend-Following with Tight Stops
+$372,695
Total realized PnL
1.8% win rate
8 wins / 441 losses
BTC only
Single coin, single strategy

• Winning Trades

8 trades
Average win$52,500
Biggest single win$184,474

• Losing Trades

441 trades
Average loss-$107
Loss patternTight stops, small cuts
The math:
E[trade] = 0.018 × $52,500 + 0.982 × (-$107)
E[trade] = $945 - $105
E[trade] = +$840 per trade
Massive asymmetry: winners are 490x the size of losers. Win rate is irrelevant when the payoff ratio is this extreme.

5. Maker vs Taker

HL's crossed field separates aggressive (taker) from passive (maker) fills. The difference is stark.

• Takers (Aggressive)

Profitable
Trade count1.38M
Avg fee / trade$0.007
Total fees paid$10K
Avg PnL / trade+$15.87

• Makers (Passive)

Unprofitable
Trade count1.38M
Avg fee / trade$0.967
Total fees paid$1.33M
Avg PnL / trade-$14.34
Key insight: Takers are systematically profitable. Makers get adverse-selected — they provide liquidity and get picked off by informed flow. The fee structure (takers pay ~$0.007, makers pay ~$0.967) does not compensate for the information asymmetry. Maker PnL is -$14.34/trade after rebates.

6. Volume Profile (Hourly)

HL trade count vs Binance by hour (UTC). HL has 1.7x to 4.3x Binance activity across all hours observed.
Hyperliquid Binance Bar width = trade count (max: 541K)
14:00
281K
171K
164%
15:00
314K
163K
193%
16:00
325K
150K
217%
17:00
368K
141K
261%
18:00
422K
130K
325%
19:00
460K
108K
427%
20:00
444K
135K
329%
21:00
476K
195K
244%
22:00
541K
231K
234%
23:00
390K
227K
172%
Peak ratio: Hour 19 UTC (427%) — HL has 4.3x Binance trade count. Peak volume: Hour 22 UTC — 541K HL trades, 231K Binance trades. The ratio column shows HL trade count as a percentage of Binance trade count.

7. Volatility Comparison

Same asset, same day, different microstructure. Average volatility is nearly identical, but the tails diverge.

• Hyperliquid

Avg volatility5.76 bps
Max volatility21.47 bps
Vol-of-vol4.00
Volume-vol correlation0.651

• Binance

Avg volatility5.70 bps
Max volatility48.23 bps
Vol-of-vol4.36
Volume-vol correlation0.836
Same mean
5.76 vs 5.70 bps — no meaningful difference in average realized vol. Both venues see the same asset dynamics.
Different tails
Binance max vol (48 bps) is 2.2x HL (21 bps). CEX has fatter tails — likely from larger single trades hitting the book.
Volume-vol coupling
Binance (0.836) has tighter volume-vol coupling than HL (0.651). Volume spikes on BN predict vol better.

8. Missing Math — Research Directions

Quantities we can compute from this data but haven't yet. Each represents a publishable microstructure result.

VPIN

Volume-Synchronized Probability of Informed Trading

HL's crossed field directly labels aggressor vs passive on every fill. No need to estimate trade direction (Lee-Ready, tick test) — we have ground truth. This enables exact VPIN calculation, which is typically noisy on CEX data.

Kyle's Lambda

Price impact per unit volume

Regress price change on signed order flow (net buy volume). Lambda measures how much prices move per dollar of informed trading. Compare HL lambda vs BN lambda to quantify which venue has deeper effective liquidity.

Hasbrouck Information Shares

Price discovery attribution

Decompose the common efficient price into contributions from each venue. What fraction of new information enters through HL vs Binance? The lead-lag data suggests BN leads, but Hasbrouck gives the precise split.

Amihud Illiquidity Ratio

|return| / volume, cross-exchange

Simple but powerful: how much does price move per dollar of volume? Lower = more liquid. Compute hourly for both venues to find windows where HL is more liquid than BN (and vice versa).

Optimal Execution Split

Given maker rebate + price discount, what % routes to HL?

HL offers: (a) 3.8 bps price discount, (b) maker rebates, (c) lower taker fees. Binance offers: deeper book, less slippage. Solve for the optimal split ratio as a function of order size.

jthorvaldur · Cross-exchange analysis: Hyperliquid L1 + Binance Spot · ETH · May 25, 2025