[2H6-OS-2c] OS-2

6:30 PM - 6:45 PM JST(9:30 AM - 9:45 AM UTC)

[2H6-OS-2c-05] User-Preference-Aware Portfolio Optimization via Multi-Objective Bayesian Optimization with Active Preference Learning

〇Ryota Ozaki¹, Masanori Hirano¹ (1. Preferred Networks, Inc.)

Keywords:

Finance,Portfolio optimization,Bayesian optimization,Preference

In portfolio optimization for stock investments, user preference represents one of the crucial factors. For instance, reflecting each investor's distinct risk-return trade-off preference is essential. On the other hand, investors generally find it difficult to directly specify preference parameters such as risk aversion levels as explicit numerical values. In this paper, we propose a method that efficiently converges to solutions achieving user's desired trade-off by estimating user preference using preference feedback provided by investors (through pairwise comparisons and requests for improvement directions) while simultaneously searching for the risk aversion parameter in the Mean-Variance Model. Numerical experiments using Japanese stock data demonstrate that our proposed method can identify more desirable solutions than previous methods with fewer search iterations.