We hold inverse perpetual shorts, so funding is income rather than a cost, and we wanted to know how much of it we can count on. The answer turned out to be a fact about the instrument rather than about our position, so it belongs here.
Funding is quoted the way a price is quoted, it changes every eight hours, and it is usually described as symmetric: when longs crowd in they pay shorts, when shorts crowd in they pay longs. We measured it on the full public history of three contracts: the inverse ETH perpetual on OKX, the inverse ETH perpetual on Binance, and the linear ETH perpetual on Binance as a control. That is 2683 settlements, the longest series reaching back 400 days.
Not one settlement paid a short more than 0.0100 per cent per eight hours. Not on the inverse contracts, not on the linear one. The maximum of every series is exactly that number, to the last digit.
The first thing to check was whether this is the exchange refusing to go further. It is not. Both venues publish their own limit and it is 0.75 per cent, seventy five times higher. What we had found instead was the interest component: both venues also publish an interest rate, and it is 0.0100 per cent. Funding is the premium of the perpetual over the index plus that interest term, so whenever the premium is negligible the rate collapses onto the constant. It did so in 27 per cent of all settlements, landing on the same value to the last digit hundreds of times.
The downside carries no such shape. The same series reach minus 0.1645 per cent, sixteen times further from the constant than anything on the upside ever went.
So on this instrument, over this period, the sentence "shorts get paid when the market squeezes" did not survive contact with the data. What shorts got was the interest rate, which annualises to 10.95 per cent on the notional if it holds, against a realised average of 4.34 per cent because most settlements sat below the constant rather than on it.
The clearest illustration arrived while we were measuring. On 19 August the market ran 8.5 per cent in under an hour on what was reported as the largest short liquidation since 2021, and open interest fell 9.6 per cent in thirty minutes, which is the signature of positions being closed rather than opened. Funding through that hour sat exactly on the constant, and the premium of the perpetual over its index was negative, meaning the contract traded below the index while the market was supposedly squeezing the people short of it.
Two practical consequences for anyone reading a funding number.
The first is that the field most APIs expose is not the number that settled. On one venue the last funding rate reported by the endpoint, and shown in the terminal header, is a forecast of the next settlement rather than the result of the previous one. An hour before one settlement it read 0.00360 per cent; the rate that actually settled was 0.00757 per cent, twice as high. A comparison built on that field is a comparison of forecasts.
The second is that funding is not the reason to hold this structure. Ten per cent annualised on the notional sounds material until you notice it is available only while the premium sits at zero, and that a single quarter in our sample paid 0.41 per cent. We treat it as a pleasant regime rather than a line in the plan, and size the position on the parts that were measured.