Skip to contents

Changelog

Source: NEWS.md

rafe 0.1.0

First public release.

Metrics

  • compute_rafe(), compute_crafe() and compute_trafe() implement the risk-adjusted mean error, the operator-norm precision distortion, and their combination into the upper bound on the Sharpe-ratio gap.
  • Argument order matches the replication code of Salcher, Stöckl & Hanke (2026), so scripts from that repository run unchanged: compute_rafe() takes the realised covariance as its third positional argument, and compute_crafe() takes the realised covariance first.
  • compute_trafe() applies the theorem’s data-dependent constant by default (c = 1 when RAFE ≤ SR*, else c = 2), and returns the decomposition as attributes rafe, crafe, c and SR_star. The published replication code fixes c = 1; pass c = 1 to reproduce those numbers exactly.

Moment correction

  • mu_rafe_stein() applies positive-part James–Stein shrinkage in the risk-whitened space. Supply T_obs for the James–Stein intensity or kappa to fix it.
  • sigma_crafe_floor() raises small eigenvalues to a floor relative to the mean eigenvalue.
  • sep_tune() and joint_trafe_tune() select (kappa, tau) on an inner-validation split, sequentially or jointly. Default grids use 101 points on [0, 1] for kappa and on [0, 0.5] for tau.

Data

  • ff12: monthly excess returns on the Fama-French 12 industry portfolios, January 1964 to December 2023 — the sample of the published paper.

Vignettes

  • rafe-getting-started, rafe-evaluation (reproduces the published mean-variance experiment on 49 industry portfolios) and rafe-post-processing.