Composite Scoring System

Rather than relying on any single indicator, QuanterLab combines multiple metrics into a composite score ranging from 0 to 100. This multi-factor approach aims to identify the highest-quality mean reversion candidates.

Daily Timeframe Weights

For daily bar analysis, the composite score uses these weights:

Daily Composite Weights
  • Hurst Exponent: 25% - Core regime detection
  • RSI Win Rate: 25% - Entry signal quality
  • Variance Ratio: 15% - Mean reversion confirmation
  • Half-Life: 15% - Reversion speed
  • Bollinger Band Win Rate: 15% - Secondary entry signal
  • Autocorrelation: 5% - Additional confirmation

Hourly Timeframe Weights

Shorter timeframes emphasize different factors:

Hourly Composite Weights
  • Hurst Exponent: 30% - Even more critical for short-term
  • Variance Ratio: 20% - Increased importance
  • Half-Life: 20% - Speed matters more intraday
  • RSI Win Rate: 15% - Reduced vs daily
  • Bollinger Band Win Rate: 10% - Reduced vs daily
  • Autocorrelation: 5% - Same as daily

Score Calculation

Each component is first normalized to a 0-100 scale, then weighted:

Composite = Σ (Weight_i × Score_i) Example: (0.25 × 85) + (0.25 × 72) + (0.15 × 90) + ... = 79.5

Minimum Thresholds

Securities must meet minimum composite scores to appear in scanner results:

  • Daily timeframe: Minimum score of 45
  • Hourly timeframe: Minimum score of 35

These thresholds filter out securities that don't show sufficient mean-reverting characteristics.

Score Interpretation

Higher scores indicate stronger mean reversion characteristics and better historical performance of entry signals. A score of 80+ represents the top tier of mean reversion candidates.

Why Multi-Factor?

Single-indicator approaches have weaknesses:

  • Hurst alone doesn't tell you when to enter
  • RSI alone doesn't tell you if mean reversion is likely
  • Win rates alone can be misleading without regime context

By combining metrics, the composite score provides a more robust assessment than any individual component.

Back to QuanterLab
Report
Loading report...
Article
Loading article...