weight_by_blend

API documentation for tradeexecutor.strategy.weighting.weight_by_blend Python function.

weight_by_blend(alpha_signals, blend_alpha=0.5)

Linear blend of equal-weight and signal-proportional allocation.

Computes weights as:

w_i = alpha * (1/N) + (1 - alpha) * (signal_i / sum_j(signal_j))

This is equivalent to James-Stein shrinkage applied to the portfolio weight vector, shrinking signal-proportional weights toward the equal-weight prior. The literature consistently shows that shrinkage estimators outperform both pure equal-weight and pure optimised portfolios out of sample.

The blend parameter alpha controls the shrinkage intensity:

• alpha=1.0: pure equal weight

• alpha=0.0: pure signal-proportional (same as weight_passthrouh())

• alpha=0.5: 50/50 blend (recommended starting point)

Pros:

• Dead simple — one-line formula, easy to reason about

• Grounded in shrinkage estimation theory (James & Stein, 1961)

• Robust to signal noise — always partially diversified

• Weights sum to 1.0

Cons:

• Linear blending may not be optimal — equal weight is a crude prior

• Does not adapt to signal dispersion (same alpha regardless of whether signals are tightly clustered or widely spread)

References:

• James & Stein (1961), “Estimation with Quadratic Loss”, Proc. 4th Berkeley Symposium

• Jorion (1986), “Bayes-Stein Estimation for Portfolio Analysis”, Journal of Financial and Quantitative Analysis

• DeMiguel, Garlappi & Uppal (2009), “Optimal Versus Naive Diversification: How Inefficient is the 1/N Portfolio Strategy?”, Review of Financial Studies — shows 1/N is hard to beat

• Carver (2015), “Systematic Trading”, Ch. 11 — handcrafting portfolio weights as shrinkage toward equal weight

Parameters:

• alpha_signals (Dict[int, float]) – Signal objects keyed by pair ID.

• blend_alpha (float) – Shrinkage intensity. 1.0 = pure equal, 0.0 = pure signal.

Returns:

Weight map summing to 1.0.

Return type:

Dict[int, float]