Most of the funding posts you read on crypto Twitter are about directional bets dressed up as carry. Someone notices a perp is paying +0.04% per hour, they short it, they wait, and they call it a basis trade. It usually isn't. What they're running is a naked short with a funding tailwind — which is a different animal. The funding pays you a little when you're right about direction and ruins you when you're wrong.
A real basis trade is hedged. The perp leg is one half of a pair; the other half is a long spot position that absorbs the directional risk. What you're harvesting isn't the price move. It's the funding payment. The trade is supposed to be boring on a P&L chart and interesting only on a funding-accrual chart. If your basis trade has a chunky directional P&L, you've miscalibrated something.
Hyperliquid is the most interesting venue to run this on right now. Funding has been persistently positive on majors for several months, the venue's transparent vault layer makes the mechanics easier to inspect than on any CEX, and the spot leg settles on the same L1 the perp does, which removes a chunk of the cross-venue settlement risk that bites people running ETH-spot-on-Coinbase against ETH-PERP-on-Binance. There are also two structural quirks that change the math compared to a binance-pattern carry trade. This post is the field guide: what you actually put on, what the venue does to the carry, and the three ways a perfectly designed basis trade quietly stops working.
What the trade actually is.
The basic shape: you buy spot, you short the perp, in matched notional. The two legs cancel out on direction; the only thing left is the funding payment, which by construction you receive — because shorts receive funding when funding is positive, and most of the time funding is positive on majors.
A worked example, in numbers that match what ETH has been doing recently. ETH-PERP is trading at $3,840 and ETH spot is at $3,835. You buy $50,000 of ETH spot at $3,835. Simultaneously, you open a $50,000 short on ETH-PERP at $3,840. Your net delta is approximately zero — a small basis between the two legs, which we'll come back to.
Funding on Hyperliquid pays hourly. If the funding rate sits at +0.012% per hour over the next week, your short collects 0.012% of $50,000 every hour. That's $6 per hour, $144 per day, just over $1,000 per week. Hold the trade for a month at that rate and you're north of $4,300 on $50,000 of notional. Annualized that's roughly 100% — except you cannot hold it for a year at +0.012%, because the rate compresses dramatically as more flow piles in.
Notice what is not in the calculation: the direction of ETH. If ETH rallies from $3,840 to $4,200 over your hold, your spot leg gains $4,700 and your perp leg loses $4,700. You net zero on direction. If ETH falls to $3,500, you lose $4,400 on spot and gain $4,400 on the perp. You still net zero on direction. The carry is the only thing left. That is the whole point.
What Hyperliquid does to the math.
Three things change the trade compared to a CEX basis play.
Funding pays every hour, not every eight. Most CEXes settle funding three times a day. Hyperliquid settles every hour, on the hour. The mechanical effect is that the rate you see is smaller per period but more granular. The practical effect is that your carry compounds an order of magnitude more finely, and the rate you can lock in moves faster as flows shift. A basis trade that sees its funding rate drop from +0.012% to +0.004% over four hours has lost two-thirds of its expected forward return in that window. On an 8-hour-funding venue you'd see one print of that drop. On Hyperliquid you see four, and you have four chances to act on it.
The spot leg lives on the same L1 as the perp. Hyperliquid is its own chain, and ETH-SPOT settles on it. That removes a real chunk of execution-leg risk: no cross-venue funding to bridge, no withdrawal queue to wait through if you need to unwind, no second exchange's outage window to inherit. You can see both legs from the same wallet, which means a vault-style breakdown of the entire trade is auditable in one place. Most basis traders we have spoken to underrate this. It is what lets a single strategy file describe both legs as one position rather than two unrelated trades that have to be reconciled by hand.
The vault layer is loud. This is the awkward one. Hyperliquid's transparent vault architecture means a meaningful fraction of OI on any given day is itself basis-trade flow — vaults running the same play, in size, in public. When too much basis-trade size piles in, the funding rate compresses, and the rate you joined at is not the rate that's available a week later. The premium can also flip negative if a large vault unwinds quickly. You are not trading against a hidden crowd. You are trading alongside a visible one whose collective behavior moves the carry you're chasing. The signal upside is that you can watch the OI rotate; the cost is that the carry is more competitive than it looks at any single tick.
The three ways the trade dies.
A basis trade is supposed to be boring. The interesting paths are the ones where it quietly stops being boring and you do not notice in time.
One: the funding regime flips. The trade's whole edge is the sign of funding. You are short the perp on the assumption that funding stays positive. When it flips, you are paying instead of receiving, and the trade is hemorrhaging cost on top of whatever execution friction it carries. The flip itself is usually quick — a sharp spot selloff drags the perp below the index, funding goes negative for a few hours or a few days, and your edge has changed sign. The naive basis trader does not unwind because "it'll flip back," which it sometimes does and sometimes does not. The disciplined version has a written rule: if funding stays negative for more than four consecutive epochs, the trade unwinds regardless of where price is.
Two: the basis widens against you on entry or exit. The perp can trade meaningfully off spot — a few basis points in normal conditions, dozens during stress. If you put the trade on when the perp is rich (perp above spot) and the basis converges to flat over your hold, you give up the convergence on entry. Worse: if you have to unwind during a stress moment when the basis is wide in the wrong direction, the unwind itself costs you weeks of carry. We have measured this on real prints. The worst exit cost we have seen on a $100,000 basis position during a sharp vol spike was 14 basis points round-trip, which is roughly two weeks of average +0.012% funding evaporated in one trade.
Three: opportunity cost catches up. This bites people running the trade in low-funding regimes. The spot leg requires capital, and capital tied up in spot is capital not earning anything else. If the implied yield on the spot side — lending it out, staking it, parking it in a stable-spot vault — exceeds the funding you're collecting, the trade is negative-carry once you account for opportunity cost. ETH-PERP basis at +0.012% per hour is roughly +100% annualized, which beats almost any spot yield you can find. ETH-PERP basis at +0.004% per hour is +35% annualized, which is roughly what you can get on a stable-spot strategy with a tenth the operational complexity. The trade still works on paper. It just does not beat the next-best use of capital, and you need to know your hurdle rate before you commit to the lockup.
The common thread is the same in all three: a basis trade is not a hold-forever position. It is a yield-harvest trade that you actively manage, with explicit rules for when the regime has changed enough to step out. The naive version — "I'll just leave it on" — is how a working trade goes from boring to a slow drain.
How to run one without losing the boring.
In an Engine strategy file, a basis-trade playbook is a four-block markdown file like any other, with sizing logic that explicitly pairs the two legs and a regime gate keyed off funding sign and venue stress. Here is a stripped-down version of what we ship:
## Universe
- ETH-PERP / ETH-SPOT, BTC-PERP / BTC-SPOT
- Min 24h perp volume: $100M
## Signals
- funding_rate averaged > +0.008% over last 6 epochs
- basis (perp − spot) between -10 bps and +25 bps at entry
- Skip if BTC realized vol > 70% (annualized, 1d window)
## Risk
- Per-pair notional cap: 25% NAV; total: 50% NAV
- Per-leg size: equal notional, matched at the same tick
- Unwind if funding < 0 for 4 consecutive epochs
- Unwind if basis widens past +60 bps or below -25 bps
- Daily PnL stop: -0.8% NAV (basis trades should not lose much in a day)
A few things to notice. The signal block is not a "fire when funding > X" trigger. The strategy is allowed to be on whenever funding has averaged positive over a window — it is a continuous position, not a discrete entry. The risk block is where the basis-specific logic lives: the leg-pairing, the basis-widening unwind, the funding-sign unwind. The daily PnL stop is unusually tight because a working basis trade should not lose much in a day. If it is losing 1% in a day, the assumptions baked in have already broken, and the right move is to step out and re-evaluate, not to sit through the loss waiting for the carry to make it back.
The agent reads this file the same way you do. Every funding-payment hour is logged. Every basis check is logged with the perp price, the spot price, and the spread. Every regime check fires a line into the decision log with the numbers it saw. Six weeks into a basis-trade run, you can scroll back and see exactly how much carry the trade earned versus what it was supposed to, and where any gap came from. Most basis trades we have audited that "stopped working" turned out not to have stopped at all — they were earning roughly what they were supposed to, and losing the difference in execution cost on unnecessary rebalances. Without the log, you would never see that. With it, you can fix it.
A short conclusion.
A basis trade on Hyperliquid is the most boring profitable thing you can run on the venue, and the boredom is the feature. You are not betting on direction; you are harvesting funding while a price-neutral position absorbs the move. The trade pays you to be patient about it.
The three failure modes are not exotic. Funding flips, basis widens, opportunity cost catches up. All three can be encoded as rules in a strategy file the agent reads, and all three are visible in the decision log when they happen. Run them, and the basis trade stays boring. Skip them, and the trade quietly turns into a directional bet you did not sign up for.
The funding rate on majors will compress further as more flow piles in. The trade gets less interesting at +0.004% than at +0.012%, and at some point a stable-spot strategy beats it on a per-NAV basis. But the structure of it — paired legs, regime-aware unwinds, audited carry — is durable across the rate. If you want a starting playbook, the file above is defensible. Fork it, run it for a few weeks, and let the decision log tell you which line to edit next. That is the whole loop.