NEUTRINO is a high-CAGR, single-sleeve leveraged-ETF strategy. Backtest full-sample CAGR is 39.6% at Sharpe 1.26 over 16 years (2010-2026), with maximum drawdown −31.9%. NEUTRINO beats Phoenix v2 on CAGR in every window (FULL: 39.6% vs 38.0%; IS: 38.9% vs 38.4%; OOS: 40.4% vs 37.5%) at a substantially lower Sharpe and higher MDD — the growth-tilted alternative.
What is novel. NEUTRINO introduces FOUR genuinely new mechanics relative to anything in this codebase: (1) the Garman-Klass OHLC volatility estimator (statistically more efficient than realised close-to-close vol); (2) a two-horizon SMOOTH rate-velocity gate combining the trailing-yoy and trailing-90-day change in DGS10 into a continuous gate; (3) a stock-bond correlation regime gate applied only to the equity leg, that de-risks when the diversification regime breaks; (4) aggressive 45% target vol on TQQQ (3x Nasdaq), capped 2.0x.
No VIX-level macro gate, no HY-OAS gate, no 200-day SMA, no cross-sectional momentum, no ML. The risk machinery is entirely built from yield-curve dynamics and cross-asset correlation.
Honest disclosure. NEUTRINO does not beat Phoenix on Sharpe (1.26 vs 2.10) or MDD (−31.9% vs −20.5%). The single-sleeve Sharpe ceiling in this regime is ≈ 1.30; reaching Phoenix-level Sharpe would require a multi-sleeve ensemble with near-zero pairwise correlations — a structural property that's hard to engineer without copying Phoenix's specific construction. NEUTRINO's value is high CAGR and architectural simplicity.
| Universe | 3 LETFs: TQQQ (3x Nasdaq), TYD (3x 7-10y Treasury), UGL (2x gold). Plus SPY/TLT for the correlation gate. |
| Vol estimator | Garman-Klass on 21-day OHLC: σGK = √[mean(0.5·ln(H/L)² − (2ln2−1)·ln(C/O)²) × 252]. Uses the full intraday range — ~7× lower estimator variance than close-to-close. |
| Per-asset target vols | TQQQ 45% (cap 2.0×), TYD 10%, UGL 10%. Weight = clip(target / σGK, 0, cap). |
| Rate-velocity gate | gaterate = clip(1 − rvyoy/2.0, 0, 1) · clip(1 − rv90d/1.5, 0, 1) where rvT = DGS10 − DGS10.shift(T). Catches 2022 (DGS10 went 1.6% → 4.2%) without using VIX/HY. |
| Correlation gate | Applied to TQQQ only. corr60(SPY, TLT): ≤ 0 → 1.0; (0, 0.20] → 0.6; > 0.20 → 0.0. De-risks equity when stock-bond diversification breaks (e.g. 2022). |
| Self-throttle | 252d HWM drawdown, floor −25% (rarely activates). |
| Portfolio overlay | Vol-target 27% (60d realised); DD throttle 252d HWM, floor −15%. |
| Execution | All signals lagged shift(1); use info ≤ close[t−1]. Trade at open[t]; PnL uses open[t] → open[t+1]. TC 5 bps one-way on |Δw|. |
Universe expansion experiment. We tested adding small-cap (TNA, URTY -- not in dataset), international (EDC, YINN), and sector LETFs (TECL, SOXL, FAS, ERX, DRN) to the equity leg. Findings:
NEUTRINO_DIV vs baseline (full sample):
alt/neutrino_div_strategy.py.Stock sleeve experiment. The dataset includes 90 individual S&P 500 large-cap stocks (no small caps or international stocks available). We tested adding a stock-momentum sleeve in parallel to the NEUTRINO core. Used a 126-day Sharpe-rank signal (different from Meridian's 126d return-momentum) and Friday weekly rebalance (different from Meridian's Wednesday) to differentiate.
Survivorship-bias caveat. The 90-stock universe is the *current* S&P 500 with full data; bankrupt/delisted names (Lehman, WaMu) aren't represented. Following Meridian, we report a 3% CAGR haircut on the stock sleeve return as a conservative correction.
NEUTRINO_PLUS = 85% NEUTRINO core + 15% stock-Sharpe-rank. Tested weights {95/5, 90/10, 85/15, 80/20, 75/25, 70/30, 60/40, 50/50}; 85/15 gives the cleanest improvement that survives the survivorship haircut. See alt/neutrino_plus_strategy.py.
alt/neutrino_strategy.py (baseline), alt/neutrino_div_strategy.py (sector tilt), alt/neutrino_plus_strategy.py (with stocks), and alt/NEUTRINO_DESIGN.md in the repository.