Quarterly Futures vs. Perpetual Funding: Is There a Term Premium?

A perpetual future charges or pays funding every few hours, at a rate nobody knows in advance. A quarterly future has no funding at all; instead it trades at a premium or discount to spot, the basis, that shrinks to zero at expiry. Both instruments price the same thing: what it costs to hold leveraged exposure over time. One floats, the other is fixed at the moment you trade.

If the fixed price were systematically higher than the floating one turns out to be, you could collect the difference: lock in the basis with a short quarterly future, hedge it with a long perpetual, and pay the funding as it comes. Academic work on perpetual futures (BIS Working Paper 1087 by Ackerer, Hugonnier and Jermann) gives reasons why such a term premium might exist. We wanted to know whether it shows up in the contracts you can actually trade.

The trade

For every expired quarterly contract we opened the same position at a fixed distance before expiry (30, 60 or 90 days) and held it to the day before expiry:

• Short the USDⓈ-M quarterly future: this locks in the basis at entry.
• Long the perpetual, same notional: price moves cancel out, but the long side pays funding while it's positive.
• The result of one cycle is the basis locked in, plus a small convergence residual at exit, minus the funding paid along the way, minus fees.

For BTC with a 90-day hold, the averages across all 21 cycles look like this:

The fee line uses a taker fee of 0.055% per order on all four legs (open and close, future and perp), the same figure we use in the fee-coverage article.

What we measured

Binance USDⓈ-M quarterly contracts on BTC and ETH, 21 expired contracts per coin (expiries from June 2021 to June 2026), daily closing prices for both legs, and daily funding of the matching perpetual. Entry at the daily close 30, 60 or 90 days before expiry, exit at the close of the day before expiry.

This is a descriptive study: no signal, no threshold, nothing optimised. We did not pre-register it as a test, and with 21 cycles per coin it couldn't support a fine-grained rule anyway. The question was simply whether the average cycle earns anything.

The result: basis and funding match

Individual cycles are far apart: in some quarters the basis priced in funding of 25% a year and funding then ran at 10%; in others the basis was modest and funding spiked. But the errors go both ways. Averaged over all 21 cycles, the basis that was locked in and the funding that was later paid are almost identical: 8.5% vs. 8.4% a year for BTC and 8.2% vs. 8.4% for ETH.

All six combinations of coin and holding period tell the same story:

All figures annualized. Basis and funding are the average market rates over the cycles; "before fees" and "after fees" are the average cycle result on the perpetual notional. The t-statistic tests whether the average cycle differs from zero; none comes close.

Before fees, the spread is zero within noise in every cell, with t-statistics between −0.87 and 0.07. Between 48% and 62% of cycles end positive, and the worst single cycle lost 3.0% of the notional. After fees the result is negative everywhere, from −0.9% to −3.7% a year. The shorter the hold, the worse, because the same four fees are paid more often.

Cycle by cycle

Each cycle is a bet on whether funding over the next three months comes in above or below what the market priced at entry. On average, the market priced it right.

The measurement error that almost fooled us

Our first run showed a consistent loss at expiry: the future and the perpetual seemed to end up to 1% apart, in one case 11%. That looked like a hidden cost of the trade. It was a data artefact. Binance settles quarterly USDⓈ-M contracts at 08:00 UTC, so the future's daily candle on expiry day only covers eight hours, while the perpetual's candle covers the full day. Comparing the two closes compared prices from different times.

Exiting at the close of the day before expiry, when both candles cover the same 24 hours, makes the gap disappear: the convergence residual is between −0.24% and +0.15% in every cell. If you backtest anything around contract expiry, check that both legs' candles end at the same moment.

Why there's no free premium

The result is less surprising than it might seem. Anyone who can trade both the perpetual and the quarterly future can run this spread in either direction. If the basis regularly overpriced future funding, traders would short the future until it didn't. What remains is a fair price for the expected funding, and the realized funding scatters around it.

A conditional rule could still exist: enter only when the basis is unusually far above recent funding. We didn't test it. With an average of zero and only 21 cycles per coin, picking a threshold would mostly fit noise.

What we take away

1. Locking in a rate buys certainty, not return. The quarterly future fixes the carry at entry; on average, the fixed rate equals the floating rate that follows.
2. Zero before fees means negative after fees. Four orders per cycle cost 0.22%, enough to turn a fair trade into a losing one, especially with short holding periods.
3. A premium in the literature needs checking on your venue and period. On Binance BTC and ETH from 2021 to 2026, we found none.
4. Align your timestamps. A settlement time inside a daily candle can manufacture a loss, or a profit, that doesn't exist.

What this means for funding traders

Holding a hedged perpetual position and collecting funding means taking the floating rate as it comes, including the months when it drops or turns negative. The quarterly future offers the fixed version of the same carry. This study suggests that, on average, neither version is cheaper: you choose between a known rate and a variable one, not between a better and a worse deal. Fees, timing and the capital each leg ties up decide the rest.

Methodology