Reproduce “Improving Portfolio Optimization Results with Bandit Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Gustavo de Freitas Fonseca, Lucas Coelho e Silva, Paulo André Lima de Castro
This paper introduces Adaptive Discounted Thompson Sampling (ADTS) and Combinatorial ADTS (CADTS) alongside a novel Bandit Networks architecture to enhance portfolio optimization in non-stationary financial markets. The proposed methods achieve superior cumulative returns and Sharpe Ratios, outperforming classical portfolio models and existing bandit algorithms, with specific network configurations demonstrating improved robustness under market stress.
Key insights
- Selectively applying a discount factor only to the selected arm while using a sliding window for recent data allows ADTS to balance long-term experience with rapid adaptation to short-term market changes.
- A layered 'Bandit Network' architecture, inspired by neural networks, can overcome computational limitations and biases in stock selection by separating stock filtering from weight allocation.
- Applying combinatorial bandit logic with discretized weights enables effective portfolio optimization, allowing the system to learn optimal asset combinations and allocations in a large action space.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.