The story behind the signals.

• 16 technical indicators selected and implemented from academic and practitioner literature
• Lookahead-safe backtesting framework built — position[t] = signal[t-1]
• Data pipeline for 148 tickers with 12+ years of daily OHLCV data established
• Composite score architecture designed: signals 55%, risk 30%, diversification 15%

• Pareto frontier established with 4 frontier points
• Vol-Scaling tested and rejected — individual assets do not behave like market factors
• Regime gates (VIX, Yield Curve) tested and rejected — they lag drawdowns
• Continuous sizing, per-ticker routing, and 3 other approaches eliminated

• STICKY_75 identified as optimal preset — parameters fixed since round 8
• Per-ticker routing failed twice — confirms that universal presets are more robust
• Split-OOS validation: train/test split on 148 tickers
• 30 signal-selection variants tested — no subset beats the full stack

• R29: 64 portfolios, 4 tiers, weekly rebalancing with real T-Bill yield
• Orthogonal signal enrichment (HY-OAS, VIX-Term, Connors RSI) tested
• Cash-yield sensitivity, survivorship bias, and rebalance frequency validated
• Weekly rebalancing beats daily — structurally, not by chance

• 12-window walk-forward analysis (2012–2025) — each window must pass independently
• Leave-One-Year-Out (LOYO): every year left out once — 12/12 passed
• Crash Guard validated: contrarian override during market panic, bear windows +7.6% drawdown reduction
• Permutation test (R40): p=0.000 — statistically highly significant, not random

• ADX validated as noise filter — marginal gain, but stabilizes bear windows
• RSI-14 period confirmed — alternative tuning yields no OOS advantage
• Regional signal weights: NA/EU/EM macro-heavy, Global trend-heavy
• Entry/exit thresholds, drawdown regimes, and VIX blocker systematically tested

• R59–R61: Status-quo checks — weights, ADX blocker, VIX blocker all confirmed
• R62–R64: Entry/exit grid, fast signal, Crash Guard window — engine at optimum
• R66: Dual-objective sweep — all 3 tiers independently LOYO 12/12 validated
• R67: Fast Exit (VIX spike) — Defensive +31% drawdown reduction, Balanced +10.4%

• R69/R71: Crash Guard Hysteresis — 3-day cooling-off validated across all tiers (LOYO 12/12)
• Engine version v16 in production — every improvement goes through walk-forward validation
• Research infrastructure permanently available for new hypotheses

“Simple rules beat complex models.”

Vol-scaling, regime gating, continuous sizing, and split composites — all failed. The best strategy is a fixed parameter set with clear thresholds. Complexity leads to overfitting, not better results.

“Weekly rebalancing beats daily.”

STICKY weekly: +0.37pp vs. daily: -0.33pp relative to buy-and-hold. Daily rebalancing generates noise and transaction costs. Weekly rebalancing filters that out and delivers more stable results. The signals themselves are calculated daily.

“Universal presets beat per-ticker optimization.”

Two independent attempts to find ticker-specific parameters failed due to out-of-sample instability. Features that work in training do not transfer reliably. A well-calibrated universal preset is more robust.

“Universal signals, honest limits.”

Our signals are calibrated for all common stocks and ETFs. For a small number of instruments — e.g. ultra-stable steady runners or heavily event-driven stocks — the signal stays neutral. This costs nothing and causes no harm. Crypto, options, and pure commodity positions are not supported.

“Signals are a tool, not a guarantee.”

Roughly two thirds of tested portfolios benefit measurably. For the other third, the effect is too small or the constellation is not a good fit. Edgio shows you both — and leaves the decision to you.