【报告摘要】In this article, we develop a portfolio averaging strategy under the conditional mean-variance framework. It is designed for integrating all available information and achieving a desired risk-return trade-off. Specifically, a series of candidate shrinkage portfolios are constructed for controlling the portfolio positions, which differ from each other in candidate models, target portfolios, weighting matrices and/or penalty parameters. We adopt a novel criterion to determine the weights across these candidate portfolios. Theoretically, we establish the asymptotic optimality of the proposed strategy in the sense of achieving the lowest possible out-of-sample expected utility loss, and we also derive the convergence of weights arising from the criterion. Empirically, we illustrate that the proposed strategy compares favorably with 14 alternative strategies across four datasets.