Estimation constitutes a major challenge in the implementation of mean–variance portfolios. To overcome this, we propose a partial index-tracking strategy that aims to mitigate estimation error ex-ante. Theoretically, we minimize the mean-squared error of the proposed strategy by shrinking the portfolio variance to its tracking error. Using an empirical design with over 50 years of data, our paper makes two important observations. First, we show that our proposed approach is consistent with both linear and non-linear shrinkage strategies in terms of robustness. Second, the proposed decision rule leads to a lower out-of-sample tracking error. Our findings, overall, stress the appeal of partial index tracking not only in terms of shrinkage (robustness) but also in terms of relative performance.