Skip to content

markowitz.backtest.strategies

markowitz.backtest.strategies

Portfolio strategies for use inside :class:WalkForward.

Each strategy implements the :class:Strategy protocol: given an in-sample window of asset returns it produces a weight vector (numpy array) summing to one. Strategies are kept dependency-light so the backtest layer remains usable even when the convex-optimization extras are not installed.

GMVSample

Global Minimum Variance with the sample covariance, closed form.

MaxSharpeNaive

Tangency portfolio with sample mean and sample covariance.

Falls back to :func:markowitz.optimizer.MeanVariance when available, otherwise uses the closed-form Sigma^{-1} (mu - rf) solution.

OneOverN

Equal-weight benchmark of DeMiguel, Garlappi, Uppal (2009).

RiskParity(*, max_iter: int = 500, tol: float = 1e-08)

Equal-risk-contribution portfolio via cyclical coordinate descent.

Solves the Spinu (2013) / Griveau-Billion-Richard-Roncalli (2013) convex programme min_x 0.5 x'Sigma x - sum_i b_i log(x_i) with b_i = 1/n (equal risk targets), then renormalizes to sum to one.

Source code in src/markowitz/backtest/strategies.py 
133
134
135
def __init__(self, *, max_iter: int = 500, tol: float = 1e-8) -> None:
    self.max_iter = int(max_iter)
    self.tol = float(tol)

Strategy

Bases: Protocol

Static portfolio-construction protocol used by :class:WalkForward.