An Online Portfolio Selection Algorithm with Dynamic Coreset Construction

Online portfolio selection in financial markets presents a significant challenge due to the vast number of tradable assets and the prevalence of noise signals. Existing methods often struggle to dynamically select asset subsets in order to optimize capital efficiency. To address these challenges, we propose the Portfolio Selection with Dynamic Coreset Construction (PSCS) algorithm. PSCS incorporates a novel clustering process that groups assets based on their cumulative returns, creating a unique feature space to facilitate effective clustering. From each cluster, a small set of assets is carefully chosen to form a coreset using a peak-to-valley feature that captures the price change ratio within recent time windows. Leveraging this coreset, a peak-based optimization problem is formulated and solved through closed-form solutions. Experimental results demonstrate the superiority of PSCS over seven existing algorithms, highlighting its promising contribution to the field of online portfolio selection in financial markets.