Methodology
How The Signal System Works — Full Parameter Disclosure
Every data source, every decision tree threshold, every position-sizing multiplier, every backtest configuration parameter, and every known limitation of the Market Signal Dashboard. This page exists so anyone — signed in or not — can audit the system before trusting it.
Data Flow
Base probabilities: 0.78 / 0.72 / 0.55 / 0.50 + live calibration bonuses
All Model Parameters
Decision Tree
| Parameter | Value | Source / Rationale |
|---|---|---|
| Earnings threshold | > -2% YoY | Walk-forward backtest (226 signals, 1998-2025) |
| CONTRARIAN_BUY base probability | 0.78 | Same dataset |
| BULLISH base probability | 0.72 | Same dataset |
| HOLD base probability | 0.50 | Same dataset |
| Credit Q1-Q4 bonuses | -0.03 / +0.02 / -0.02 / +0.05 | Bayesian-blended from quartile accuracy |
| Valuation Q1-Q4 bonuses | -0.03 / +0.03 / -0.01 / +0.05 | Bayesian-blended from quartile accuracy |
| Earnings coefficient | 0.004 per 1% growth (r²=0.01) | Very weak predictor, capped at ±0.03 |
| Probability clamp | 0.10 to 0.95 | Prevents extreme confidence |
Position Sizing
| Parameter | Value | Source / Rationale |
|---|---|---|
| Base risk per trade | 5% of stated bankroll | Conservative starting point |
| Kelly fraction | Half-Kelly (0.5x) | Standard ruin-avoidance practice |
| CONTRARIAN_BUY multiplier | 1.0x | From 81% backtest win rate, 5.1% avg return |
| BULLISH multiplier | 0.58x | From 79% win rate, 2.9% avg return |
| CAUTION multiplier | 0.06x | ~52% accuracy (near coin flip) |
| IV Rank > 80 adjustment | 0.6x size (expensive options) | Reduces exposure when IV is elevated |
| IV Rank < 20 adjustment | 1.2x size (cheap options) | Increases exposure when IV is low |
| Negative GEX adjustment | 0.8x size | Reduces exposure in volatile regime |
Pricing Model
| Parameter | Value | Source / Rationale |
|---|---|---|
| Model | Black-Scholes | Standard options pricing |
| Risk-free rate | 4.5% annual | Approximate current T-bill rate |
| IV proxy | VIX (market-wide) | NOT strike-specific — ignores smile/skew |
| Real pricing | UW bid/ask when available | Falls back to B-S estimate |
Backtest Configuration
| Parameter | Value | Source / Rationale |
|---|---|---|
| Period | 1998-2025 (226 monthly signals) | Walk-forward, no lookahead bias |
| Holdout (out-of-sample) | 2021-present | Separate from training period |
| Transaction cost | 4% per trade | Conservative estimate |
| Sharpe method | 30-day non-overlapping, √12 annualized | Avoids autocorrelation inflation |
| Earnings data lag | 60 days (FRED) / 90 days (FMP) | Simulates publication delay |
Known Limitations
| Parameter | Value | Source / Rationale |
|---|---|---|
| Earnings signal strength | r² = 0.01 (1% explanatory power) | Very weak predictor |
| CAUTION accuracy | ~52% | Near coin flip — sized at 0.06x Kelly accordingly |
| Sharpe ratio | Below 1.5 institutional threshold | Modest edge, not fund-grade |
| CSCV overfitting test | High PBO detected | In-sample may overfit out-of-sample |
| VIX as IV proxy | Market-wide, not strike-specific | Ignores volatility smile/skew |
All parameters above are derived from historical backtested data and may not reflect future market conditions. This system is an educational demonstration built by a software developer — not a registered investment advisor. See the full risk disclosure for details.
Educational Purpose Only
This methodology describes the Market Signal Dashboard — an educational research tool, not a trading product. Every parameter listed above is derived from historical backtested data and may not reflect future market conditions. The creator is a software developer, not a registered investment advisor, broker-dealer, or financial planner. Nothing here constitutes investment advice. Full Risk Disclosure
Happy to walk through the methodology, discuss the limitations, or answer questions about how this applies to your own data.