Collecting the funding rate is not free money. The price eats what the funding pays
What this does not establish That funding carry can never be collected. The same-asset version was not tested, and this result covers one basket rule, one universe of fourteen contracts and six years.
How it is usually sold
As free money. Crypto perpetual futures pay a funding rate every eight hours: when the crowd is long with leverage, the longs pay the shorts a small fee to keep the price of the contract close to the price of the coin. The pitch says: be the one who collects. Short the coins with the highest funding, hedge the price, and pocket "fifteen percent a year, market neutral". It comes with a funding-rate dashboard that lists the juiciest rates, a "delta-neutral yield" bot that promises to do the hedging for you, and an APR calculator that multiplies eight hours by a year. What it never shows is the price of the coin you are short while you collect.
Usually sold with Funding rate dashboards"Delta-neutral yield" botsAPR calculators
Where we cast
A market-neutral basket of crypto perpetuals: each day, short the three contracts with the highest smoothed funding and go long the three with the lowest, weighted so that each side carries similar risk. Funding is paid every eight hours; in this test the payments are added up by day and the portfolio decisions are daily. The payment schedule is not the rebalancing frequency. Trade only when the market is calm, measured by the volatility of the largest coin against its own history, because funding is collected exactly when squeezes happen. Rebalance only when the ranking changes enough to matter, since carry lives on not rotating. Fourteen contracts, six years of funding and price history, real trading costs and the real funding paid and received on both sides. Four parameters, twelve combinations, all declared before the first run.
Why it looked promising
Funding is the price of impatience: leveraged longs pay to stay in the trade, and someone has to be paid to take the other side. Collecting it naked is being short the squeeze, which is how an earlier attempt of ours died. Collecting it in a basket that is long some contracts and short others should isolate the funding differential from the direction of the market, and the calm-market filter should avoid harvesting it right when the premium collects itself. The loser on the other side is the leveraged long who never looks at what the funding costs. Three separate sources of research point to the same mechanism, which is what earned it a test.
What we found
Nothing that survives. Out of sample, the Sharpe ratio, the usual measure of return per unit of risk, was zero point zero zero, against a bar of one. The test that asks whether ranking contracts by funding beats rotating the funding series in time, run on the primary configuration, found that about forty percent of the randomized shifts matched or exceeded the observed statistic, against a limit of five percent. That is not the probability that the strategy is chance; it says the benchmark gave no evidence against the null. The correction for having tried twelve combinations is a separate step: the Sharpe deflated for those attempts was zero point four six, against a bar of zero point nine. And the decisive test: remove the single best month of six years and the Sharpe goes from zero point two six to below zero. The whole edge lived in one month. Five of the seven gates fixed in advance came back red. The two that passed, a small drawdown and survival with costs doubled, do not rescue a strategy that does not earn.
The original minutes, in Spanish (secondary evidence)
| Criterio (pre-registrado) | Umbral | Resultado | |
|---|---|---|---|
| Sharpe OOS | > 1,0 | 0,00 | 🔴 |
| Degradación IS→OOS | < 40 % | 100 % | 🔴 |
| p rotación | < 0,05 | 0,405 | 🔴 |
| DSR | > 0,90 | 0,46 | 🔴 |
| Sharpe sin su mejor mes | > 0,5 | −0,03 | 🔴 |
| Sharpe costes ×2 | > 0 | 0,19 | 🟢 |
| MDD | < 20 % | 10,9 % | 🟢 |
Gate names, results and thresholds as written in the original test record. The English summary above is derived from the same rows. Units: pb = basis points per month; t_NW = t statistic with Newey-West standard errors.
What we learned
Three things. First, a basket that shorts the highest-funding contracts is, without anyone intending it, a bet against momentum: the coins that pay the most funding are the coins that are running, and the price move against the short eats the funding many times over. Second, an edge that lives in one month is not an edge; it is an event. The test that removes the best month was built for exactly this and did its job on its first outing. Third, with fourteen contracts that mostly move together, a ranking sorts clones of the same bet, and no amount of filtering turns two independent bets into ten.
Questions you may ask
Because you collect a few hundredths of a percent every eight hours while the coin you are short can move ten percent in a day. The funding is real income; the price move is real loss, and it is larger. Picture renting out a room for a hundred a month in a house whose value swings by thousands every week: the rent is not what decides whether you made money.
We did not test it here; it is a different idea with different economics. Holding the coin and shorting its own contract cancels the price move almost exactly, so what is left is the funding minus costs: a modest rent, roughly what the funding level says, with no trading edge and no upside beyond it. It belongs to the family of income, not of trading, and it would get its own pre-registered test if the funding of the cycle ever justified it.
It is the point. If a strategy needs one specific month to be profitable, its profits depend on that month happening again, and nobody can pre-register a lucky month. A strategy that survives without its best month is one you can hold through the years when the lucky month does not come.
Check your reading
Three questions to tell what this entry established from what it did not.
Is the funding paid every eight hours, or are the decisions?
Funding is paid every eight hours. In this test the payments were added up by day and the portfolio decisions were daily. The payment schedule is not the rebalancing frequency.
What does the forty percent mean?
That about forty percent of the randomized time shifts of the funding series did as well as our result or better, for the primary configuration. Not a forty percent chance that the strategy is luck.
Does this close funding carry for good?
It closes the cross-sectional basket in this form. The same-asset version, long the coin and short its own perpetual, was not tested and would be a separate income-style test.
What would bring us back
Nothing, in this form. Three attempts at collecting funding across contracts, naked, as a capitulation signal and hedged in a basket, died on three different gates, and the family is closed in its cross-sectional form.
The same-asset version, long the coin and short its own perpetual, remains a separate candidate for a separate test, judged as income and not as trading, and only when the funding level of the cycle makes the arithmetic worth the capital.