Adaptive Multi-task Learning for Multi-sector Portfolio Optimization
This work addresses portfolio optimization for investors by enhancing multi-sector asset management, though it appears incremental as it builds on existing factor modeling frameworks.
The paper tackles the challenge of transferring information across multiple sectors for portfolio optimization by proposing a data-adaptive multi-task learning method that quantifies relatedness among factor subspaces, improving factor model estimation and portfolio performance as demonstrated on Russell 3000 index data.
Accurate transfer of information across multiple sectors to enhance model estimation is both significant and challenging in multi-sector portfolio optimization involving a large number of assets in different classes. Within the framework of factor modeling, we propose a novel data-adaptive multi-task learning methodology that quantifies and learns the relatedness among the principal temporal subspaces (spanned by factors) across multiple sectors under study. This approach not only improves the simultaneous estimation of multiple factor models but also enhances multi-sector portfolio optimization, which heavily depends on the accurate recovery of these factor models. Additionally, a novel and easy-to-implement algorithm, termed projection-penalized principal component analysis, is developed to accomplish the multi-task learning procedure. Diverse simulation designs and practical application on daily return data from Russell 3000 index demonstrate the advantages of multi-task learning methodology.