Graveyard / Basis and spreads / Binance perps
Coinbase premium flips
Does a flip in the Coinbase premium predict Bitcoin?
The idea
A positive Coinbase premium is read as US spot demand. Commentators claim a sign flip, not the level, carries information and that price continues for hours. Go with the flip.
How we tested it
| Data | Coinbase and Binance 1-minute prices for BTC, ETH and SOL, pooled equally. 4-hour mean premium crossing +5 bp after being below −5 bp, and the mirror. Stop 1.5%, target 2.0%, 72-hour safety stop, flat when the 7-day BTC move exceeds 10%. |
|---|---|
| Period | October 2023 to October 2026 (34 months, 10 bull, 5 bear, 19 choppy). |
| Costs | 0.13% per round trip. |
| Holdout | In-sample and out-of-sample. |
| Control | No random-entry control was run for this test. |
Result
| Premium after USDT/USD adjustment, 5th-95th percentile | −2.0 to +2.6 bp (BTC) |
|---|---|
| Same, 4-hour mean, 1st-99th percentile | −4 to +5.5 bp |
| Unadjusted premium, 5th-95th percentile | −12.8 to +7.2 bp |
| USDT/USD rate, same percentiles | −12.3 to +6.0 bp |
| Adjusted flips, full sample | 16 trades, Sharpe −0.13, profit factor 0.89 |
| Raw flip, out-of-sample | 10 trades, −128 bp each |
Why it failed
The Coinbase premium on charts is about 90% the USDT/USD exchange rate. After dividing that out, arbitrage holds the real gap to a few basis points, far below the 13 bp cost. No peer-reviewed study tests this premium for return predictability.
What would make it worth retesting
Only as a USDT/USD premium idea, with a freshly frozen rule.
Test id: premium-cbflip-perps-event. Past results of a backtest say nothing certain about the future.