Binance 19c — 5 Years
Hyperliquid 55c — 5 Months
Executive Summary
Continuous Forecast-Weighted Trend-Following on Crypto Perpetual Futures
1.33
Sharpe
Binance 19c, 5yr
2.239
Sortino
36.6%
CAGR
357.3%
Total Return
-18.3%
Max Drawdown
2.0
Calmar
87% (13/15)
WFO Pass
15-fold OOS
83.6%
MC P(>1)
10k bootstrap
-0.04
Corr BTC
near zero
43.9%
Win Rate
daily
2.213
Sharpe
Hyperliquid 55c, 5mo
4.271
Sortino
92.0%
CAGR
22.4%
Total Return
-10.5%
Max Drawdown
8.76
Calmar
93.5%
MC P(>0)
10k bootstrap
85.8%
MC P(>1)
10k bootstrap
-0.21
Corr BTC
near zero
47.8%
Win Rate
daily
Performance
Equity curves, benchmarks, returns, and risk metrics
Consistent growth from $10,000 to $45,734. Worst DD -18.3% recovered in 276d.
Low correlation with BTC (-0.04) — returns are alpha, not leveraged beta. Click legend to toggle.
| Metric | Value |
|---|---|
| Sharpe | 1.33 |
| Sortino | 2.239 |
| CAGR | 36.6% |
| Total Return | 357.3% |
| Max Drawdown | -18.3% |
| Max DD Duration | 276d |
| Calmar | 2.0 |
| Win Rate | 43.9% |
| Profit Factor | 1.31 |
| Max Consec. Losses | 11d |
| Corr BTC | -0.04 |
| Beta BTC | -0.01 |
Cross-validation on native Hyperliquid data. Sharpe 2.213, CAGR 92.0%.
| Metric | Value |
|---|---|
| Sharpe | 2.213 |
| Sortino | 4.271 |
| CAGR | 92.0% |
| Total Return | 22.4% |
| Max Drawdown | -10.5% |
| Max DD Duration | 41d |
| Calmar | 8.76 |
| Win Rate | 47.8% |
| Profit Factor | 1.58 |
| Max Consec. Losses | 8d |
| Corr BTC | -0.21 |
| Beta BTC | -0.14 |
Signal Speed Comparison
2-speed (32/128 + 64/256) vs 3-speed (16/64 + 32/128 + 64/256)
| Config | BN Sharpe | BN CAGR | BN MaxDD | HL Sharpe | HL CAGR | HL MaxDD |
|---|---|---|---|---|---|---|
| 3speed50 | 1.33 | 36.6% | -18.3% | 2.213 | 92.0% | -10.5% |
| 2speed | 1.226 | 23.6% | -9.7% | 1.792 | 51.6% | -8.0% |
Vol-Target Analysis
Each config uses correctly scaled emergency DD — labels show (vol_target, dd)
Sharpe is constant at ~1.55 across vol_target levels when emergency DD is correctly scaled. Vol-target is a pure leverage dial — same risk-adjusted return, different absolute return/drawdown.
| Profile | vol_target | emergency_dd | Sharpe | MaxDD | CAGR |
|---|---|---|---|---|---|
| Conservative | 0.20 | 20% | ~1.55 | ~10% | ~32% |
| Moderate (LIVE) | 0.35 | 30% | ~1.55 | ~18% | ~71% |
| Aggressive | 0.50 | 40% | ~1.55 | ~25% | ~126% |
Emergency Stop Impact
What happens when emergency DD is too tight for the vol_target
vt=0.35 — Emergency Stop Impact
vt=0.50 — Emergency Stop Impact
vt=0.35: All three curves are identical — the natural MaxDD (~18%) never reaches the emergency threshold (20%, 35%, or OFF). The emergency stop is irrelevant at this vol_target.
vt=0.50: The dd=20% curve (red) clearly diverges — the emergency stop fires prematurely, flattening all positions and locking in losses. dd=40% and noES are identical (DD never reaches 40%).
Rule: emergency_dd_pct >= max(0.20, vol_target). Too-tight ES is the #1 performance killer.
vt=0.50: The dd=20% curve (red) clearly diverges — the emergency stop fires prematurely, flattening all positions and locking in losses. dd=40% and noES are identical (DD never reaches 40%).
Rule: emergency_dd_pct >= max(0.20, vol_target). Too-tight ES is the #1 performance killer.
Correlation Filter
Impact of blocking correlated instruments (lb=168h, threshold=0.7)
| Config | Sharpe | Return | MaxDD | Calmar |
|---|---|---|---|---|
| corr OFF | 1.549 | 211.5% | -17.5% | 4.16 |
| corr ON | 1.655 | 154.4% | -12.8% | 4.42 |
| Config | Sharpe | Return | MaxDD | Calmar |
|---|---|---|---|---|
| corr OFF | 1.531 | 18.8% | -14.7% | 5.06 |
| corr ON | 2.213 | 22.4% | -10.5% | 8.76 |
The correlation filter blocks instruments correlated >0.7 with stronger forecasts, improving diversification. It consistently improves Sharpe and reduces MaxDD on both datasets, at the cost of lower absolute return (fewer positions).
Fee Sensitivity
Sharpe at different fee levels — Nautilus event-driven backtests
| Fee | Sharpe | Return | MaxDD |
|---|---|---|---|
| 0 bps | 1.33 | +357.3% | 0.0% |
| 1.5 bps (maker) | 1.226 | +339.3% | 0.0% |
| 3.0 bps | 1.122 | +321.3% | 0.0% |
| 4.5 bps (taker) | 1.018 | +303.3% | 0.0% |
| 7.0 bps | 0.845 | +273.3% | 0.0% |
| 9.0 bps | 0.707 | +249.3% | 0.0% |
| Fee | Sharpe | Return | MaxDD |
|---|---|---|---|
| 0 bps | 2.213 | +22.4% | 0.0% |
| 1.5 bps (maker) | 1.713 | +18.8% | 0.0% |
| 3.0 bps | 1.214 | +15.2% | 0.0% |
| 4.5 bps (taker) | 0.714 | +11.6% | 0.0% |
| 7.0 bps | -0.119 | +5.6% | 0.0% |
| 9.0 bps | -0.785 | +0.8% | 0.0% |
Continuous sizing strategies are highly fee-sensitive due to frequent rebalancing. Hyperliquid fees: 1.5 bps maker / 4.5 bps taker. POST_ONLY (maker) execution is critical — at taker fees, Sharpe degrades by ~25% on BN and the strategy becomes marginal on HL. All live orders use POST_ONLY with IOC fallback.
Walk-Forward Validation
15-fold, BN 19c 5yr — not applicable to HL (too short)
V4_2speed: 13/15 positive (87%), mean=1.317, median=1.338
V4_3speed50: 13/15 positive (87%), mean=1.493, median=1.776
12-month in-sample / 3-month out-of-sample, rolling quarterly, fixed config. No re-optimization = true OOS. Only available for BN 5yr (HL too short for WFO).
Monte Carlo Analysis
Block bootstrap (10k samples, block=20d)
P(Sharpe>0)=100.0%, P(Sharpe>1)=83.6%. 95% CI: [0.655, 1.964]. Statistically significant edge.
P(>1)=85.8%. Wider CI due to shorter history. Median Sharpe 2.828 confirms signal on independent data.
Configuration Sensitivity
Anti-churn sweep — BN 35c 3yr dataset
Optimal: top_k=7, rebal=8h. Rebalancing every 24h loses 30-50% Sharpe. CS deadband=0.15 adds +5% Sharpe by preventing ranking churn.