Strategy library

Top 30 by combined earnings-yield + ROIC rank.

Quality value: F-score ≥ 7, P/B < 1.5.

Bottom decile by ATR — empirically Sharpe 2.73 in our IC research.

Top decile of past-year-minus-1-month return.

Best out-of-sample performer in May-2026 holdout: Sharpe +2.56, CAGR +212%, MaxDD only -33%, PSR 100%. Generalises far better than in-sample.

Best triplet from May-02 sweep: Sharpe 1.93, CAGR 77.5%, MaxDD -32.8% (5pp better than best pair).

GTJA-191 + IC research top pair: Sharpe 1.86, CAGR 77.8%, PSR 100%, walk-forward 5/5 folds positive.

WARNING: walk-forward backtest with quarterly weight refits (refit using only data prior to each rebal) drops Sharpe from 1.54 → 0.69. The per-sector lift is largely in-sample fit. Use the flat composite (composite_5sig_yearly) as the honest production strategy. Vol-target overlay still useful for drawdown control on top of flat composite.

DOWNGRADED: walk-forward refit drops Sharpe to 0.69 (below flat composite's 1.28). Per-sector weights look great in-sample but don't generalise quarter-to-quarter. The earlier 'OOS' regime test was technically applying full-panel weights to historical data — a subtle leak. Use composite_5sig_yearly instead.

Constant-weight composite of 5 verified-real-alpha features. NO weight fitting needed → no overfitting risk. 8-year stress test Sharpe 1.28, CAGR 45%, MaxDD -44%, total +2033%. After walk-forward exposed per-sector weights as partially overfit, this is the honest production strategy. Pair with vol-target σ*=15% overlay for institutional-grade drawdown control.

Same composite_5sig basket but levered to a 12% annualised vol target via 60-day realised-vol scaling. Lifts Sharpe from 1.31 → 1.46 in the full window AND cuts MaxDD from -55% to -21% — ~3× lower downside for nearly identical Sharpe. Skew flips from +0.02 to +0.07. Best risk-reward profile in the library.

Sector-neutral z-score composite of 5 of the 6 features that survived EVERY bullshit gate (NW HAC, Bonferroni, sector-neutral, walk-forward): dist_52w_low + vol_252d + gap_abs + wq_alpha042 + range_true. Each weighted by its h=252d effective IR. Pre-shock 2021 Sharpe +1.23, post-shock 2022-2026 Sharpe +1.57 — works in BOTH regimes (the leader works in only one). Verified live through the production engine.

Best Sortino + best Calmar in the library. Sortino-targets 12% downside-vol (60d lookback, max 2.5x leverage) instead of total vol — preserves upside spikes that vol-target clips. 8-year backtest: Sharpe 1.83, Sortino 2.82, CAGR 46.5%, MaxDD only -16.4%, Calmar 2.84. Beats every vol-target variant (vt15 best at Sharpe 1.78, CAGR 33%, Calmar 2.24).

Same Sharpe as the aggressive variant (1.96) but with EVEN LOWER drawdown (-10% vs -19.6%). Stacks SPY>100d gate + 8% vol-target overlay. Lower CAGR (20% vs 54%) but smoothest equity curve in the library. Annualised vol 6%. For capital-preservation-first deployment with full risk-adjusted return.

Concentrated 10-stock variant of the gated leader. 8-year walk-forward 2019-2026: Sharpe +1.99, CAGR +69%, MaxDD -27%. Discovered via robustness sweep (45 configs tested) — n=10 with quarterly rebal is the sharpest. Drawback: more concentration risk vs n=20 baseline.

BEST RISK-ADJUSTED IN LIBRARY. Leader basket (low_atr × dist_52w_low), deployed only when SPY > 100-day MA. 8-year walk-forward 2019-2026: Sharpe +1.96 (vs leader's 1.66, +0.30 lift), CAGR +53.9%, MaxDD only -19.6%, in market 78% of time. The 100d MA captures medium-term trend — less choppy than 50d, more responsive than 200d. Robust to parameterization: nearby specs (50/100/150/200, single or dual) all give Sharpe 1.80-1.96. NO FITTING.

Most defensive variant. Combines 8% vol-target with 'cash if VIX > 30' regime filter. 8-year walk-forward: Sharpe +1.83, CAGR +16%, MaxDD only -12.4%, vol 8.5%. Lowest drawdown of any high-Sharpe strategy in the library. Best for capital-preservation-first deployment. Note: VIX gate only available 2020+ since VIX yfinance data starts there.

INSTITUTIONAL-GRADE STRATEGY. Same low_atr × high_dist_52w_low basket as the leader, but with daily 60-day-realised-vol targeting at σ*=8% (lev_max=2.0). 8-year walk-forward 2019-2026: Sharpe +1.76 (BETTER than unleveraged leader's 1.65), CAGR +17%, MaxDD only -15%, annualised vol 9%. Perfect for capital-preservation-first deployment. The drawdown control trick that works on the leader (because it's a fixed-weight strategy unlike the in-sample-fit composites).

VERIFIED PRODUCTION STRATEGY. 8-year walk-forward 2019-2026 (no fitting required, fixed sort weights -1×atr_14 + 1×dist_52w_low): Sharpe +1.66, Sortino +2.17, CAGR +63.5%, MaxDD -46.2%, total +3,323%. 80% of quarters positive (24/30), median quarterly +16%. Beats the flat composite (Sharpe 1.28). Worst quarter was Q1 2020 covid (-29%). Calm uptrending stocks — high dist from 52w low (in motion) but low ATR (not whippy). The signal-not-fit nature makes this immune to the in-sample issues that demoted the per-sector composites.

Combine the two: Sharpe 1.84, MaxDD only -12.6%.

Top decile by AFP composite quality score.

Top 20 by FCF/market-cap, market cap > $500M.

P/E < 15, P/B < 1.5, ROE > 10%.

Stocks at fresh 52w highs with volume confirmation.

Bottom decile of 5d return + top sortino — buys the dip on quality.

Top decile by 63d−21d return — short-horizon momentum without the 1m reversal.

Bottom quintile by β — Frazzini-Pedersen showed long-low / short-high β earns ~7% Sharpe.

Companies whose net income is well-matched by operating cash flow — high earnings quality.

Top decile of earnings yield among non-distressed (Altman Z > 2.99) names.

Smaller + cheaper names within the S&P 500 universe. Built from the 23 features that survived our IC bullshit filter. Backtest 2022-2026: Sharpe +1.42, Sortino +1.93, CAGR +52.4%, MaxDD -35%, random-baseline 100th percentile.

Counterintuitive: high-margin / consistently-profitable companies UNDERPERFORM (mean IC −10%). This strategy AVOIDS the priced-rich quality names and buys forgotten value — within S&P 500.

Built ONLY from features that survived our sector-neutral test (true cross-sectional alpha, not sector bets). Combines uptrend (101% retained), value (89% retained), and quality penalty (76% retained). Excludes the size effect which is 61% sector exposure.

Companies with low manipulation-flag M-score — combines with momentum for cleaner up-trends.