The pitch on most perp DEX vaults sounds almost too clean. You deposit stablecoins or a basket of assets, the protocol uses your capital as the counterparty for everyone trading perpetuals on the platform, and you collect fees plus whatever the traders lose. The APY number on the dashboard is usually fat and green. What that number quietly leaves out is that you have just agreed to take the other side of every position on the venue, which means your return is not really a yield at all. It is the aggregate profit and loss of a crowd of strangers, flipped in sign, with some fee income layered on top.
I keep coming back to these vaults because the mechanism is genuinely interesting and because a lot of people size into them thinking they bought a bond when they actually bought a book of short-vol exposure to other people's trading. Those are not the same risk, and the difference only shows up on the days you would least want it to.
What you are actually earning
The GLP model, and most of the vaults that copied it, pays you from a handful of separate streams. It helps to keep them mentally separated because they behave nothing alike.
- Trading and swap fees. Every time a trader opens, closes, or swaps against the pool, they pay a fee, and a slice flows to depositors. This part is steady and roughly grows with volume. It is the closest thing to real yield in the whole stack.
- Funding and borrow fees. Traders pay to hold leveraged positions, and that cost accrues to the pool. This one leans in your favor when the book is crowded on one side, which is often but not always.
- Trader profit and loss, with the sign flipped. This is the big one and the one nobody wants to look at directly. When the platform's traders lose, you win. When they win, you eat the loss out of the pool, and the vault price drops even if fees were positive that week.
- Asset exposure. If the vault holds a basket like ETH, BTC, and some stables rather than pure USDC, part of your return is just the price of those assets moving. That can dwarf everything else and has nothing to do with how the traders did.
The first two streams are the reason the marketing APY looks appealing. The third is the reason the marketing APY is not a reliable forecast of anything. You are running a small casino, and the house edge is real over a large enough sample, but the sample is other people's directional bets, and directional bets cluster.
The toxic-flow problem
The uncomfortable part of being the house is that not all traders are the same quality of counterparty, and the good ones find you. In a healthy vault, most of the flow is retail taking leveraged punts that wash out over time, plus fee income that never stops. That mix is what makes the vault price grind up. The problem is a specific kind of trader, usually called toxic or informed flow, who is systematically right often enough that they extract more from the pool than they pay in fees.
On a perp DEX this shows up in a few concrete ways. Latency arbitrageurs pick off the pool's oracle price when it lags a fast spot move on centralized venues. Delta-neutral desks farm the funding rate and hedge elsewhere, so they collect the payment without ever taking the risk the funding was meant to compensate. And during a violent trend, a wave of momentum traders all pile in the same direction, the pool is forced to be short the whole move, and a week of accumulated fees vanishes in an afternoon. None of this is a bug in the protocol. It is the natural consequence of offering a passive counterparty to an open market. The vault is doing exactly what it promised, which is lose to the people who are right.
The mental model I use is that fee income is a slow drip and toxic-flow losses are lumpy and occasional. On most days the drip wins and the chart looks like a savings account. Then there is one day a quarter that undoes a month, and whether that day is survivable depends entirely on who trades against the pool.
How to read a vault before you become the house
Because the return is really the trader base flipped in sign, the only evaluation that matters is historical trader profit and loss on that specific platform. Crypto going up or down is almost a distraction. A vault can bleed in a raging bull market if its traders happen to be long and right, and it can print in a crash if its traders are overwhelmingly and wrongly long into the drop. So the work is not looking at BTC's chart. It is looking at the platform's trader ledger.
Here is the checklist I actually run before putting size into one of these.
- Pull the cumulative trader PnL curve, not just the vault APY. Most serious perp DEXs publish this or leave it derivable on-chain. If aggregate traders have lost steadily over a long window across different market regimes, the house edge is real. If they are net up, you would be paying to be their counterparty.
- Check the vault's worst drawdown, not its average. The average is fee income doing its slow drip. The drawdown is the toxic-flow day. Size to survive the drawdown, because that is the number that actually ends people.
- Separate asset exposure from trading edge. If the vault holds volatile assets, back out how much of the historical return was just those assets going up. In a down market that same exposure flips, and you do not want to discover that the edge you thought you were buying was really just a levered long on ETH.
- Look at flow concentration. A pool whose PnL is dominated by a handful of large winning wallets is being farmed by informed traders. A pool whose losses are spread thinly across thousands of small accounts is the retail-heavy mix you actually want to be the counterparty to.
- Watch how the oracle is priced. A vault that settles on a slow or single-source oracle is an invitation to latency arbitrage. Robust, fast, multi-source pricing is not glamorous, but it is what keeps the pickers-off from grinding you down between the big days.
Sizing it like the exposure it is
Once you accept that the position is short the trading skill of a specific crowd rather than a yield product, the sizing gets more honest. I treat a perp DEX vault the way I would treat writing options, because functionally that is close to what it is. Steady premium most of the time, a fat left tail when the market moves hard and the flow happens to be positioned for it. That framing keeps me from parking money I cannot afford to see draw down twenty or thirty percent in a bad stretch, which is very much on the table for the vaults with thin or informed flow.
The practical habit is to keep watching the trader PnL curve after you deposit, not just at entry. A vault's edge is not a constant. A platform that attracts a smarter trader base over time, or one whose incentive program pulls in delta-neutral farmers, can drift from a good house into a bad one without the APY ever flashing a warning. This is the kind of on-chain read I built a lot of Blockcircle around, watching wallet-level flow and counterparty behavior rather than just price, because on a perp DEX the price of the underlying tells you almost nothing about whether being the house is a good idea this quarter.
None of this makes the vaults a bad trade. Plenty of them have paid depositors well for long stretches, and being the house is a real, defensible edge when the flow is genuinely retail. It just means the thing you are underwriting is people, not price, and the only honest way to size it is to look hard at how those people have historically done before you agree to take the other side of all of them.