Q-A3C2: Quantum Reinforcement Learning with Time-Series Dynamic Clustering for Adaptive ETF Stock Selection

TL;DR

This paper introduces Q-A3C2, a quantum-enhanced reinforcement learning framework with dynamic clustering for adaptive ETF stock selection, significantly improving returns by capturing evolving market regimes.

Contribution

It presents a novel quantum reinforcement learning model that integrates time-series dynamic clustering and VQCs to better adapt to changing financial markets.

Findings

Q-A3C2 achieves a 17.09% cumulative return, outperforming benchmarks.

The quantum-enhanced model improves feature representation and adaptability.

Experimental results demonstrate superior performance in dynamic market conditions.

Abstract

Traditional ETF stock selection methods and reinforcement learning models such as the Asynchronous Advantage Actor-Critic (A3C) often suffer from high-dimensional feature spaces and overfitting when applied to complex financial markets. Moreover, static clustering algorithms fail to capture evolving market regimes, as the cluster with higher returns in one period may not remain optimal in the next. To address these limitations, this paper proposes Q-A3C2, a quantum-enhanced A3C framework that integrates time-series dynamic clustering. By embedding Variational Quantum Circuits (VQCs) into the policy network, Q-A3C2 enhances nonlinear feature representation and enables adaptive decision-making at the cluster level. Experimental results on the S and P 500 constituents show that Q-A3C2 achieves a cumulative return of 17.09%, outperforming the benchmark's 7.09%, demonstrating superior adaptability and exploration in dynamic financial environments.