Skip to yearly menu bar Skip to main content

Workshop: OPT 2022: Optimization for Machine Learning

Enhanced Index Tracking via Differentiable Assets Sorting

Yuanyuan Liu · Yongxin Yang


Enhanced index tracking (EIT) aims to achieve better performance over a target equity index while maintaining a relatively low tracking error. It can be formulated as a quadratic programming problem, but remains challenging when several practical constraints exist, especially the fixed number of assets in the portfolio. In this work, we propose a new method for enhanced index tracking, subject to common practical constraints, including cardinality, which is based on a novel reparametrisation of portfolio weights integrated with a stochastic optimisation. It can simultaneously tackle asset selection and capital allocation, while being optimised by vanilla gradient descent effectively and efficiently. The proposed method is backtested with S&P 500 and Russell 1000 indices data for over a decade. Empirical results demonstrate its superiority over widely used alternatives.

Chat is not available.