Research Note 04

Carry: A Real Edge I Wouldn't Trade

The research question: perpetual funding carry has genuine academic backing as a documented, economically explained inefficiency rather than a mined pattern — does it still clear a realistic capital-cost hurdle when measured on real, recent Binance data, or is an edge that is real in the literature too small to bother trading once actually measured. He, Manela, Ross & von Wachter report a Sharpe of 1.80 for a spot-short-perpetual carry trade even after realistic costs. Christin, Routledge, Soska & Zetlin-Jones report in-sample annual Sharpe ratios in the range of 7 to 10 on a major exchange, and trace a large share of that profitability to retail leverage availability, documenting a clean break when Binance cut maximum leverage from 125x to 50x in July 2021. Schmeling, Schrimpf & Todorov document crypto carry averaging roughly 7% per year on a fixed-date basis and, using a difference-in-differences design around the January 2024 launch of the first U.S. spot Bitcoin ETF, show the ETF's introduction measurably shrank the futures-spot basis available to capture.

The project's own measurement

The project's own measurement used real, unmodified Binance funding-rate data for BTCUSDT and ETHUSDT across 636 eight-hour funding periods per coin, spanning November 2025 through May 2026. Annualizing each print, mean funding came to +2.11% per year for BTC and +1.63% per year for ETH, with funding negative for roughly a third of all observed periods on both coins and the maximum single mechanically-annualized observation across the entire sample reaching only 10.95%. The regime shifted twice inside the seven-month window: a positive stretch through January 2026, three consecutive negative months, then a partial rebound in May.

Why no net-of-costs figure is reported

No net-of-costs figure is reported for this strategy, and the note explains why directly: turning a two-legged, continuously hedged position into an annualized cost figure requires assumptions about holding period, rebalancing frequency, and margin maintenance that the funding-rate dataset alone cannot responsibly support.

Verdict

The note does not need a net-of-costs figure to reach its verdict. Against the project's own pre-registered 14% annual capital-allocation hurdle — a project convention, not a market risk-free rate — both BTC and ETH's gross figures fall roughly an order of magnitude short, and even against the far more conservative comparison of roughly 4.0–4.5% available from FDIC-insured U.S. savings accounts and CDs as of August 2026, BTC's gross funding rate still falls short outright, before any cost adjustment. A second literature-backed candidate, the fixed-date futures basis trade, is closed at the literature stage rather than independently replicated: Schmeling, Schrimpf & Todorov's own reported average of roughly 7% per year already sits below the project's hurdle before any of the paper's own documented ETF-driven compression is even applied.

The verdict is explicit that carry was not falsified — the mechanism documented in the literature is still measurably present today — but was priced out of the project's opportunity set: real, measured, and simply too small relative to the operational complexity of running it and to simpler, better-documented alternatives.

Full note: GitHub ↗