Auroxyl Capital runs four systematic strategy sleeves on U.S. equities and ETFs — residual-momentum leaders, quality-value, a regime-weighted blend, and an all-weather capital shield — fully automated end-to-end, from factor computation to order execution. Currently operating as a private book; not accepting investors.
View the Four SleevesFour sleeves, one shared research discipline: pre-registered metrics, shuffle-null and multi-offset validation, and a graveyard of ideas that failed the bar. Simulated track records below run , benchmarked against QQQ.
Simulated performance at current production configuration, overlay and risk gates on, gross of fees. Log scale. Benchmark: Invesco QQQ (Nasdaq-100).
Every live component carries its own statistical evidence: shuffle-null significance, walk-forward confirmation, and parameter-flatness across perturbations.
The live value-trap guard beat 100% of label-shuffled null runs — the signal survives when the labels don't. Deployed factors clear this bar before going live.
The forward-EPS factor's information coefficient tests at t = 9.99 on leak-free, point-in-time data — positive in every year of coverage.
Perturbing the deployed guard's coefficients across 9 combinations moves Sharpe only within 1.73–1.86 — a flat plateau, not a fitted spike.
A five-agent, ten-dimension search threw ~11,500 alternative configurations at the deployed setup. None beat it out-of-sample. The production config stood.
| Live component | Statistical evidence | Status |
|---|---|---|
| Deployed factor set | Shuffle-null p = 0.000; selected by walk-forward in every fold; drop-one evidence pruning | SHIPPED |
| Momentum weighting | Robustness battery: 4 entry points × drop-top-5/10/15 winners × ex-best-2yr — advantage monotone in every column | SHIPPED |
| Forward-estimate factor | Leak-free IC t = 9.99, positive every coverage year, zero beta cost, no sector bet | SHIPPED |
| Value-trap guard | Standalone Sharpe 1.59 → 1.81; shuffle-null p = 0.000; flat 1.73–1.86 across 9 coefficient combos | SHIPPED |
| Risk overlay | Unbeaten across 7 fine-tuning axes × 3 adversarial rounds; deterministic, zero ML parameters | DEFENDED 3× |
| Regime-weighting model | Blend Sharpe 2.05 vs 1.65/1.70 standalone — the diversification is measured, not assumed (corr 0.50) | LIVE |
| Production software | 758/758 test suite green; incremental pipeline parity 0.0e+0 vs full recompute | LIVE |
This book does not accept investors. Open to collaboration and partnership enquiries only:
erlian3230@gmail.com
Important Disclosure
Past performance is not indicative of future results. All performance shown is
simulated (backtested) at the current production configuration and is gross of fees
and financing. Backtested results have inherent limitations and generally overstate live results;
Auroxyl internally discounts them by ~50%. The “AUX” series are internal strategy sleeve
designations — they are not registered funds or exchange-traded products and are
not available for investment. Concentrated portfolios (4–5 names) carry elevated single-name risk.
Investing involves risk, including possible loss of principal. This book does not accept investors
and is open to collaboration only. All trading is executed via Interactive Brokers.