Gaussian Boson Sampling for Asset Clustering in Statistical Arbitrage Portfolios

Battle Count: 50

Highly relevant to quantitative trading. The paper directly addresses statistical arbitrage portfolio construction, a core quant trading strategy. It introduces novel quantum clustering algorithms (GBS Roots) benchmarked against state-of-the-art classical methods (SPONGE, Spectral). Results demonstrate superior alpha generation during high-volatility periods, market neutrality maintenance during crashes, and practical considerations like photon loss compensation. The rolling-window rebalancing framework, zero-cost portfolio construction, and mean-reversion signal generation are directly applicable to real-world trading systems. The work bridges quantum computing and practical finance with concrete economic metrics (Sharpe, Sortino, total return).

High complexity due to: (1) quantum photonic computing concepts (GBS, hafnian, Takagi-Autonne decomposition, squeezed states), (2) sophisticated graph theory and correlation clustering, (3) multi-step data preprocessing pipeline (beta estimation, residual returns, RMT filtering, thresholding), (4) iterative clustering algorithms with dynamic parameter updates, (5) photon loss modeling and coherent displacement compensation, (6) rolling-window portfolio construction with rebalancing logic, (7) computational bottleneck of classical GBS simulation (O(N^3 * 2^N) for hafnian). However, the Strawberry Fields API and provided code reduce practical implementation barriers. The classical benchmarking methods (Spectral, SPONGE) are well-established.

Data and code are available at Zenodo (DOI: 10.5281/zenodo.18890167). The paper uses publicly available CRSP and Fama-French data via Wharton Research Data Services. Simulations use the Strawberry Fields API for GBS. Detailed algorithm pseudocode (Algorithms 1 and 2), parameter settings, and mathematical formulations are provided. However, physical GBS hardware is not used (simulations only), and some parameters like threshold values are referenced in appendices. The 100-run/20-run sample sizes for classical/quantum methods are specified.