POLARIS is a 4-sleeve leveraged-ETF ensemble that uses signal mechanics fundamentally different from Phoenix and Meridian. Backtest full-sample Sharpe is 1.16 with CAGR 23.0% and MDD −24.4% over 16 years (2010-2026).
What is novel. POLARIS deliberately AVOIDS every primary signal Phoenix uses: no VIX-level macro gate, no HY-OAS gate, no 200-day SMA, no cross-sectional 12m/9m price-momentum on LETFs, no XGBoost/ML. Each sleeve uses a different information source: (1) vol-targeted RP basket with rate-velocity (DGS10 yoy) gate; (2) 40/20 Donchian channel breakout (Turtle Traders style); (3) VRP-spread harvest trading the spread between VIX and SPY 21d realised vol; (4) bond TYD on rate-direction (DGS10 vs 60d MA).
Honest disclosure. POLARIS does NOT beat Phoenix on Sharpe or CAGR standalone. Its main values are: (a) excellent IS-OOS robustness (OOS Sharpe 1.28 > IS 1.07, no overfit); (b) 0.34 correlation with Phoenix — a 20% POLARIS + 80% Phoenix blend lifts Sharpe to 2.14 with MDD −18.8%; (c) genuinely different signal sources for diversification.
| S1. VOLT_RP | Vol-targeted risk-parity on QLD (2x Nasdaq) + TYD (3x 7-10y) + UGL (2x gold). Per-asset target vols 20/10/10. Rate-velocity gate (DGS10 yoy > 1.0pp → 0.5; > 2.0pp → 0). |
| S2. DONCHIAN_BO | Long QLD when close[t-1] makes a new 40-day high; exits on 20-day low. Same rate-velocity gate. Channel-range signal, structurally distinct from moving-average filters. |
| S3. VRP_HARVEST | vrp = VIX − SPY_21d_realised. Long QLD if vrp > 5 AND realised vol < 25; long UGL if realised > 30; else cash. Trades the SPREAD, not VIX level. |
| S4. BOND_DIP | Long TYD when DGS10 < 60-day MA AND T10Y2Y > −0.5. Bond-only sleeve providing genuine asset-class diversification (correlation ~0.1 with equity sleeves). |
| Blend | Inverse-vol weights computed on IS only (2010-03 to 2018-12), held fixed for OOS. Per-sleeve self-throttle 252d HWM −20%. Portfolio overlay: vol-target 18%, DD throttle 252d HWM −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. |
Universe expansion experiment. We tested expanding the VOLT_RP equity sleeve from QLD-only to include sector LETFs (TECL, SOXL, FAS, ERX), small-cap (TNA -- not in dataset), international (EDC, YINN), and other large-cap (UPRO).
POLARIS_DIV vs baseline (full sample):
alt/polaris_div_strategy.py.Stock sleeve expansion. The dataset includes 90 individual S&P 500 large-cap stocks (no small caps or international stocks available). We added a 5th sleeve to POLARIS using 126-day Sharpe-rank on this stock universe with weekly Friday rebalance. This is genuinely novel relative to Meridian's stock sleeves, which use 126d *return*-momentum on Wednesday.
Survivorship-bias caveat. The 90-stock universe is the *current* S&P 500 with full data; bankrupt/delisted names aren't represented. Following Meridian, we report a 3% CAGR haircut on the stock sleeve as a conservative correction.
Result. POLARIS v2 (5 sleeves) substantially beats POLARIS v1: full Sharpe 1.24 vs 1.16; OOS Sharpe 1.32 vs 1.28; MDD -22.6% vs -24.4%. The haircut version still beats v1 (1.19 / 23.5% / -23.3%). The stock sleeve adds genuine diversification because individual stocks have idiosyncratic returns the LETF sleeves don't capture. See alt/polaris_v2_strategy.py.
alt/polaris_strategy.py (baseline v1), alt/polaris_div_strategy.py (multi-equity), alt/polaris_v2_strategy.py (with stock sleeve), and alt/POLARIS_DESIGN.md in the repository.