Where this started
Yesterday I paused a regime-conditional paper portfolio that was down 44% net of fees. The autopsy blamed a bug in my strategy approval logic (that's the earlier post today). While rebuilding I also tested a paper-spec cross-sectional momentum strategy and got 0.37 Sharpe over 5.8 years, negative for the last 86 weeks straight (that's the middle post today). So two of my three research directions failed the smoke test.
The third was cash-and-carry basis. Long spot, short the perp of the same asset, collect the funding rate. It's delta-neutral by construction: if BTC rallies 20%, spot appreciates 20% and the short perp loses 20%. In theory you don't care about direction, just the funding stream. The marketing decks I keep seeing quote 15-30% APY for retail participants.
Two months ago I built v1 of this backtest and got Sharpe 23 because I'd approximated the hedge as perfect. That's not a backtest, that's a division by a rounding error. v2 fixed the modeling by actually pricing the perp position to the perp mark and tracking a margin call trigger. v2 liquidated at every collateral level from 1x to 20x, because I'd modeled isolated margin. A BTC rally past the margin buffer wiped the perp bucket even though the account overall was fine.
v3 fixes that.
What v3 actually does
Two structural changes from v2.
Cross-margin instead of isolated margin. In a delta-neutral spot-plus-short-perp book, spot appreciation exactly offsets perp short loss by construction. Isolated margin makes that cancellation invisible: the exchange sees the perp bucket draining on a rally and liquidates for lack of margin, even though the account overall is fine. Cross-margin (Binance's Portfolio Margin mode or the equivalent on other venues) lets the accounts net. This is how someone actually running this trade would set it up.
A funding-regime filter. Optionally close the position when funding stays negative for 9 consecutive 8h windows (3 days). Re-enter when it flips positive. The idea is to avoid paying funding to longs during bear stretches.
Both variants are backtested against 2 years of hourly BTCUSDT spot + perp mark + funding history from Binance. Costs are modeled at 0.4% round-trip (spot maker + perp taker + slippage estimate), which is on the optimistic side for non-VIP retail.
The results
Two-year window, $2000 notional:
Mode AnnRet Sharpe MaxDD Cycles Funding Fees FinalEq
hold_forever 5.91% 3.92 0.30% 1 $204.84 $9.34 $2234
regime_filter 5.60% 3.76 0.46% 2 $200.59 $17.10 $2222
Both variants produce positive real numbers. The hedge holds. No liquidation events. Max drawdown under 0.5% on a delta-neutral book, which is what the theory predicts and what the data shows.
The regime filter is slightly worse because the extra entry/exit costs ($17 vs $9) exceeded the funding it saved by exiting during the one bear stretch it caught. In other fee regimes or with more prolonged negatives that could flip, but at these levels once-and-hold wins.
The Sharpe of 3.92 is arithmetically correct but read it carefully. It's a carry-trade Sharpe. The volatility denominator is tiny because the delta hedge is real, not because the strategy is skillful. Any strategy that earns a slow steady drip with almost no daily variance will report Sharpe 3-plus. It's not the same "Sharpe 3.92" as a directional strategy that beats the market cleanly. What it does mean is "consistent, no big drawdowns."
The window sensitivity
This is where the story gets more honest.
Window AnnRet (hold) Funding earned
730d (2y) 5.91% $204.84
365d (1y) 1.97% $24.53
180d (6mo) 4.11% $13.59
90d (3mo) 9.68% $15.30
Most of the 2-year cumulative funding was earned in the year before the last year. The trailing 12 months paid out under 2% annualized. Fees ate almost 30% of that. If I deployed this today with recent funding conditions persisting, I'd be looking at 1-2% net after fees.
The 90-day recent number looks better (9.68%) but it's a 3-month sample and the last month has been unusually bullish on crypto, which pushes funding rates up temporarily. That's not signal I can plan around.
The honest takeaway on the number: this strategy earned real money over the last two years, but the edge has been degrading and is currently in the low single digits. That is not the 15-30% APY range the how-to guides cite. It is not zero either.
What could still bite this in live
Six things I don't fully control by backtesting.
- Cross-margin eligibility. Binance Portfolio Margin has minimum account requirements that vary by region. If you can't get on it, you're stuck with isolated margin and back to the v2 liquidation problem.
- Fees are optimistic. 0.4% round-trip assumes maker fills on spot and small slippage. Non-VIP taker on both legs plus retail spread could be 0.6-0.8%, which turns the recent regime from marginal-positive to marginal-negative.
- Exchange risk. All the capital is on one exchange. An FTX-style event zeros the account.
- BTC only. No diversification. ETH funding is usually similar but not identical; a real book would run several venues and several coins to spread the exchange and coin risk.
- Funding is not structurally required to stay positive. It usually is because leveraged perp traders on average pay for their leverage, but it can invert for weeks in a bear regime. The last-year data shows this.
- My model uses 8h bars for basis P&L. Intraday liquidations or margin events between bars are invisible. Cross-margin makes this less scary but not zero.
What I'm going to do
Not deploying yet. The numbers clear the honest-signal test but don't clear the switching-cost test.
Here's the shape of the decision. My current running paper strategies are also earning close to zero after fees, so the opportunity cost of switching is not high. But basis in the current regime is earning 1-2% net, which is barely above a Treasury bill for real work. It's not a research-project failure like today's other two posts. It's a "yes, this actually works, but the payoff at present funding rates is too small to justify the exchange concentration and cross-margin complexity right now."
If funding rates rise back to their 2024-early-2025 range (roughly 0.02-0.03% per 8h, versus current 0.005-0.010%), the annualized would jump to the 6-10% range where the trade actually makes sense to run at size. So the more interesting research question is when funding rates are elevated enough to make this worth running, not how much better I can make the strategy design.
That question is answerable. Historical funding, on-chain leverage metrics, and open-interest changes are all public. I can build a funding-regime forecaster that says "enter when the last 30-day median funding is above 0.015% per 8h, exit when it drops below 0.005%." That's a much smaller, cheaper piece of research than reinventing the whole trade, and it produces an actual decision rule.
That's the plan for next week.
What I'd tell anyone else building this
Cross-margin is not an implementation detail. It's the entire thing. If you're going to backtest this, model it. If you're going to run it, get on Portfolio Margin before you deploy any real size.
And read the numbers as they are, not as they were. The 15-30% APY figures floating around are from a specific window (2020-2022, then again briefly in 2024) when funding was elevated. Those windows exist, and there will be more. But right now is not one of them, and pretending it is turns "a real edge in the right regime" into "I ran a strategy in the wrong regime and lost money."
Backtest detail is at forward-test-data/basis-research/basis-v3-report.json on Tokyo for anyone who wants to reproduce or challenge the numbers.