Perp Leverage Distribution Explorer
A point-in-time snapshot of every active Hyperliquid perp trader. Tests the claim
that the majority of capital on perp exchanges runs at moderate leverage
— even when the public perception (driven by liquidation feeds and screenshot
culture) is 10x–100x. This view filters to long positions only on
perps with at least $10M aggregate long notional in our window
(plus BTC, ETH and PAXG as anchors).
Snapshot
- State captured: —
- Address universe: 24h S3 fills window —
- This is a single frozen frame — not a time series. Leverage and positions can change minute-to-minute.
Methodology
- Long-only deployer filter: an account is included if it holds any long position on the focus coins.
- Roll-up model (consistent across all views): L1 = position (per account, per coin) → L2 = per-account leverage = ntl_a / eq_a (sum a deployer's position notional, divide by their equity) → L3 = per-cohort leverage = cap-weighted mean across accounts in cohort = Σ ntl_a / Σ eq_a. This is the cohort's flow leverage. Lorenz ranks accounts by equity; Distribution histograms per-account leverage; Cohort table averages it; Coin Detail per-cohort cells use the same Σ pv / Σ eq pattern on each coin.
- Per-position leverage = pv / account_value · sign(szi) (Kelly-fraction units)
- Cohorts: equity buckets (dust<$100, retail<$1k, mid<$10k, active<$100k, pro<$1M, whale<$10M, mega ≥$10M)
- Kelly* = (μ − r) / σ² from arithmetic daily returns, annualized × 365, r = 3.5% (matches durable-leverage methodology). The long-run Kelly uses full yfinance history (BTC since 2015, ETH since 2017, etc.) and is the anchor — short-window Kelly (30/90/180/365d) is regime-sensitive and noisy.
- Capital-weighted stats weight by position notional or account equity; count-weighted stats weight every account/position equally.
Sample
- Source:
s3://hl-mainnet-node-data/node_fills_by_block for 2026-06-01 (24h window)
- ~58k perp-active addresses → per-account
clearinghouseState via Hyperliquid info API
- 34k accounts with non-zero equity at snapshot time
- Per-coin views focus on top 10 crypto perps by aggregate notional + the two RWA perps HL lists (PAXG = gold, SPX = S&P 500)
Caveats
- Snapshot, not time-series — does not capture regime dynamics
- Sample is biased toward accounts that filled in the 24h window — includes market makers and HLP-style strategies that may dominate the capital-weighted view
- Realized-window Kelly ≠ forward Kelly; in a bear regime realized Kelly is sharply negative even when long-run Kelly is positive
- Addresses anonymized as acct_NNNNN ranked by equity desc