Investigating the Performance of Quantum Support Vector Machines for High-Frequency Trading Strategies

Abstract

High-frequency trading (HFT) strategies operate under extreme requirements of latency, high dimensional data streams, and rapid decision-making. Classical machine learning models, including support vector machines (SVMs), are widely used in algorithmic trading, but they face limitations when confronted with ultra-high dimensionality, non-stationarity, and the need for near-real-time inference. In this manuscript we investigate the use of quantum support vector machines (QSVMs) a quantum machine-learning adaptation of the classical SVM within an HFT setting. We develop a detailed mathematical formulation of the QSVM, embed it into a prototypical HFT pipeline, and perform numerical experiments on tick-level and limit-order-book data to compare QSVM against classical SVM counterparts. We report metrics on classification accuracy, decision latency, model training and inference scalability, and robustness to noise and decoherence in near-term quantum (NISQ) settings. Our results indicate that while QSVMs currently do not universally dominate classical SVMs in all trading settings, they show promise in handling very high-dimensional feature spaces with competitive latency and accuracy under certain constraints. We conclude with an industry-oriented discussion of practical implementation issues (hardware access, latency budgets, regulatory considerations) and future research directions.