Statistical arbitrage and pairs trading on crypto perps, with live Hyperliquid pairs

What is statistical arbitrage? Statistical arbitrage bets that prices which usually move together will come back together after they drift apart. Pairs trading is its simplest form: when one asset gets expensive against the other, you short it and buy the other, sized so the market’s overall move roughly cancels, and close when the gap returns to normal. On Hyperliquid both legs can be perps, so you can short as easily as you buy. The risk is that the relationship breaks and the gap keeps widening.

How a pairs trade works

1. 01Pick a pair with a reason to move together: two layer-1 coins, two chipmakers, gold and silver.
2. 02Estimate the hedge ratio. Regress one asset’s returns on the other’s; the slope (beta) tells you how much of B hedges one unit of A.
3. 03Build the spread as the log price ratio, or A − beta × B, and measure how far it sits from its average in standard deviations (the z-score).
4. 04Enter when the spread is stretched, for example beyond ±2, by shorting the rich leg and buying the cheap one.
5. 05Exit when it returns toward zero, and stop out if it keeps widening or the relationship breaks.

Run pairs and long/short strategies on Senpi

Senpi publishes this guide, and it runs this family of strategies without code: deploy an open-source template, change its levers in plain language, and it runs on Senpi’s runtime with a stop from the first fill. Four templates cover the main styles:

• Chameleon z-scores the ETH/BTC, SOL/ETH and SOL/BTC ratios against their 48-hour average. When a ratio stretches 2 standard deviations or more and starts turning back, it takes one position in the higher-beta coin in the direction of the snap-back. It trades one leg, not both.
• Octopus ranks liquid crypto perps by 24-hour relative strength, buys the strongest and shorts the weakest, both trend-confirmed, on two equally funded wallets so the book stays close to market-neutral.
• Cougar runs the same long/short approach on tokenized US stocks such as xyz:NVDA, xyz:AMD and xyz:TSLA.
• Relativity XYZ trades the spread between the strongest and weakest stocks, commodities and indices intraday, with the S&P’s trend deciding which side it may take. It uses up to 10x leverage and is rated aggressive.

• Each book in its own wallet. Octopus and Cougar fund the long and short books in two separate strategy wallets, so each side’s margin is visible on its own.
• Maker-first entries. Orders go to the book as maker orders first (0.015% at the base tier) with a taker fallback, which matters for a strategy that pays four fills per round trip.
• A stop from the first fill. Strategies built from a rule or a template carry the Dynamic Stop-Loss (DSL), and each can have its own daily-loss or drawdown halt.
• 0% Senpi fee on paid plans (0.03% on the free trial).

A strategy you confirm runs as code on Senpi’s runtime, and by default no AI model makes live trading decisions. The agent holds no key and cannot withdraw, and copy strategies carry at most a flat stop. Senpi has no backtesting tool; some templates cite a historical replay in their notes.

Live pairs on Hyperliquid

Correlation and hedge ratio use daily log returns over the last 90 days. The z-score compares today’s log price ratio with its 90-day average. The half-life is how long a gap took to halve on average across the full history (from a first-order fit); shorter means faster reversion.

ETH/BTC and SILVER/GOLD look most like tradable pairs here: high correlation and gaps that halved in about 36 days and about 24 days. AMD/NVDA is the warning: returns were correlated at 0.66, yet the ratio moved 41% in 90 days, so shorting the ‘rich’ leg would have kept losing. HYPE/BTC barely reverted at all.

Statistical arbitrage beyond pairs

• Baskets and long/short books. Rank many assets by a signal, buy the top, short the bottom, and keep the book close to market-neutral.
• Cross-asset dispersion. Bet on the gap between the day’s winners and losers rather than on the market’s direction.
• Mean reversion on one asset. The single-asset cousin: fade a stretch from a moving average. See What is mean reversion?

Costs and risks on perps

• Fees. A pairs round trip is four fills. On crypto perps at Hyperliquid’s base tier that is 4 × 0.045% = 0.18% as taker, or 0.06% as maker. Stock and commodity pairs on the XYZ DEX cost less on most markets, which run in growth mode at 0.009% taker and 0.003% maker; xyz:GOLD, xyz:MSTR and a few others pay 0.09% taker and 0.03% maker (Hyperliquid API, Nov 12, 2026).
• Funding. Both legs pay or earn funding every hour. If both sit at Hyperliquid’s default, about 11% a year, the long pays and the short earns roughly the same; if one leg’s funding is far from the default, the carry can dominate. See Funding rates as a trading signal.
• Leg risk. Fill one leg and not the other, and you hold a naked position.
• Liquidation. Each perp leg has its own margin. A sharp move on the short leg can liquidate it while the long leg is still open.
• Regime change. Relationships break on news, listings and narratives; a stop on the spread is part of the strategy.

Test the spread with costs and funding before you trade it: How to backtest a crypto trading strategy. Hyperliquid charges 0.015% for maker orders and 0.045% for taker orders at the base tier on crypto perps. Senpi adds 0% on paid plans and 0.03% on the free trial or without a paid plan.

This is educational material, not financial advice. The market readings are data, not trade signals, and leveraged perps can lose money quickly.