Skip to content

feat(QC): find truly complementary alpha combinations for composites #140

@jsboige

Description

@jsboige

Context

Current composite strategies combine alphas whose Sharpe ratios end up being roughly linear combinations of the underlying alphas' Sharpe ratios. The TrendWeather composite (Sharpe 1.155) works because TrendStocks dominates, not because AllWeather adds genuine diversification.

Known issues:

  • FamaFrench + AllWeather: sweep monotonic toward AllWeather (FF doesn't diversify)
  • MomentumSector + RegimeSwitching: both defensive in same periods (double-defense problem)
  • EMATrend: pending sweep, but 5 tech stocks overlap

Objective

Find alpha combinations where alphas are genuinely complementary:

  • Uncorrelated return streams (different market regimes, different asset classes)
  • One alpha profits when the other is flat/losing
  • Combined Sharpe significantly exceeds individual alphas (not just weighted average)

Approach

  1. Correlation analysis: Build a correlation matrix of all alpha return streams
  2. Regime analysis: Map which alphas work in which market regimes (bull, bear, sideways, high-vol, low-vol)
  3. Complementarity score: Rank pairs by how well they complement each other
  4. Test top pairs: Build composites from the most promising combinations
  5. Walk-forward validation: Ensure results are robust out-of-sample

Related issues

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestpriority-highBROKEN strategies to fix firstquantconnectQuantConnect strategy development

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions