A strategy that does not guess price direction: buy spot, short perpetual, harvest funding payments. The previous walk-forward reported Sharpe 2.30 and that is precisely the reason to be suspicious. This page recomputes from scratch with strict timing, two-leg costs, stress tests, and data checks against the exchange.
Funding carry is real, but small, and shrinking. Always active generates +11.72% over three years (+3.9% per year) after costs. Its Sharpe is high because volatility is almost zero, not because returns are large. What is not proven is the timing overlay used by E17: the threshold selected from training data produces -1.02% in the next two years (Sharpe -1.22). Status: not yet conclusive as a source of alpha, only suitable as a low-risk cash product with notional 0.5x.
Before trusting the results, the data is checked first: spacing between settlements, duplicates, and match with the exchange.
What is not available: historical perpetual prices. The deriv table only stores open interest and long/short ratio since September 2026, so the spot-perp basis cannot be measured and is modeled as an assumption in the stress test. This is a real limitation, not a detail.
The old carry version determines today's position using an average funding that includes today's payment as well. That is logically wrong, so I measured its impact:
| Signal variant | 3-year return | Sharpe | Transactions |
|---|---|---|---|
| Strict signal (correct) | +4.26% | 4.09 | 48 |
| Leaky signal (old version) | +4.28% | 4.06 | 48 |
Each transaction has four cost components: spot fee 10 bps, spot slippage 5 bps, perp fee 5 bps, perp slippage 5 bps. Total 25 bps per direction on notional, so 50 bps round-trip.
| Component | Value |
|---|---|
| Gross funding (always active, BTC) | +11.86% |
| After costs | +11.72% |
| Costs paid | +0.14% |
| Average funding per 8 hours | 0.0068% |
| Entry condition | CAGR | Sharpe | Max DD | Transactions |
|---|---|---|---|---|
| always active (threshold 0) | +1.62% | 3.50 | -2.7% | 147 |
| 7-day average > 0.005% per 8 hours | +0.15% | 0.29 | -5.2% | 215 |
| 7-day average > 0.010% per 8 hours | +1.40% | 4.09 | -0.4% | 48 |
| 7-day average > 0.020% per 8 hours | +0.51% | 1.83 | -0.3% | 36 |
| 7-day average > 0.030% per 8 hours | +0.35% | 1.70 | -0.3% | 20 |
All variants produce small returns with very small drawdowns. A Sharpe above 3 here is not a sign of extraordinary superiority: the numerator is small and the denominator is nearly zero. Do not compare this figure with BTC's Sharpe of 0.86.
| Asset | CAGR | Sharpe | Max DD | Transactions |
|---|---|---|---|---|
| BTC | +1.14% | 2.82 | -0.6% | 18 |
| ETH | +1.25% | 2.94 | -0.8% | 20 |
| SOL | +1.80% | 4.60 | -0.3% | 10 |
| Year | Return | Annualized | Sharpe | Max DD |
|---|---|---|---|---|
| 2023 | +1.47% | +8.0% | 19.4 | -0.11% |
| 2024 | +6.65% | +6.6% | 18.4 | -0.01% |
| 2025 | +1.75% | +1.7% | 10.7 | -0.18% |
| 2026 | +0.58% | +0.7% | 5.3 | -0.41% |
This is the most important part: the results are shrinking. 2024 was still +6.6%, 2025 +1.7%, and 2026 only +0.58% in 298 days. This pattern is consistent with more capital competing for the same trade.
The threshold was selected from the first 12 months, then blind-tested on the next 24 months without re-selecting it. Selected threshold 0.0 with training Sharpe 12.8.
| Test period | Return | Sharpe | Max DD |
|---|---|---|---|
| 2024-10-25 to 2026-10-25 | -1.02% | -1.22 | -2.7% |
The timing overlay that underpinned the Sharpe 2.30 claim lost money once blind-tested. What survived is only the always-active version, and that needs no parameters at all.
| Scenario | 3-year return | Sharpe | Max DD |
|---|---|---|---|
| 2x costs | +2.19% | 1.47 | -1.1% |
| basis minus 5bps | +3.84% | 3.43 | -0.5% |
| basis minus 20bps | +2.60% | 1.87 | -0.9% |
With a notional of 0.5x capital, the perpetual short runs at 1x leverage: margin equal to notional, liquidation only occurs if the price rises by around 100% without additional margin. That is why its drawdown looks small. But the consequence is that the return is also small.
| Notional / capital | Short margin | Price rise to liquidation | Estimated annual return |
|---|---|---|---|
| 0.5x (used here) | 100% notional | +100% | +3.9% |
| 0.75x | 33% notional | +33% | +5.9% |
| 0.9x | 11% notional | +11% | +7.0% |
| Comparator | 3-year total | CAGR | Sharpe | Max DD |
|---|---|---|---|---|
| Carry funding (always active) | +11.72% | +3.9% | 3.5 | -2.7% |
| Buy and hold BTC | +614% | +38.6% | 0.86 | -77% |
| Cash (0%) | 0.00% | 0.00% | - | 0.00% |
E17 is not pursued as an alpha candidate. The claimed Sharpe of 2.30 is rejected because it comes from a timing overlay that failed blind testing. Carry is still recorded as a low-risk cash instrument with realistic figures: 3.8% per year in 2023-2026, falling to around 0.7% in 2026. No new paper trading for carry until historical basis data exists.
The next step is not to add more carry experiments, but to decide economically: if the target is to cover VPS costs, a carry of 0.5x notional will not be enough. If the target is to hold cash with low risk, this is worse than a deposit. Both require a decision from Fatih, not additional experiments.
Updated 25 Oct 2026 08:52 UTC+7. Funding data was re-verified against Binance. There are no real profit claims on this page.