arXiv:2610.12148
Large-Scale Benchmarking of Quantum Neural Network Configurations for Financial Time Series Forecasting
Jack Waller, Xing Liang, Dimitrios Makris, Rajagopal Nilavalan
Abstract
Quantum machine learning, and quantum neural networks (QNNs) in particular, are advancing fields with growing potential. Although systematic comparisons of QNN configurations have been explored primarily for classification tasks, comparatively little attention has been given to regression problems, particularly financial time series forecasting. This study presents a large-scale systematic comparative evaluation of QNN component configurations for financial time series forecasting, using the GBP/USD spot exchange rate as a case study. A grid search across encoding methods, ansatz designs, qubit counts, layer depths, and cost functions yields 1,368 distinct model configurations, each evaluated in terms of prediction accuracy, computational cost, and convergence behaviour. The results reveal unique insights into how the choice of methods influences performance, such as that gate selection and arrangement are more critical to model success than raw parameter count, and that entanglement is a system-level property of the full circuit rather than solely at the ansatz level. The best-performing QNN configuration achieves an $R^2$ score of 0.985, outperforming a classical BiLSTM baseline. Additionally, the impact of real quantum hardware noise is assessed through execution on the IQM Emerald device, revealing that gate errors and decoherence represent a significant barrier to practical deployment, with gate selection and circuit depth identified as key determinants of hardware noise resilience. Overall, the findings provide practical architectural guidance for QNN design and establish a baseline characterisation of QNN noise sensitivity on near-term quantum devices.
AI Summary
The paper examines how quantum neural network (QNN) architecture choices affect financial time-series regression, using daily GBP/USD exchange rates as a case study. It evaluates 1,368 configurations spanning three encodings, twelve ansätze, 4–12 qubits, up to eight layers, and MSE or fidelity-based losses. Each configuration is trained in noiseless simulation with five random initialisations. The 5,000-point dataset is split chronologically into 3,000 training, 500 validation, and 1,500 test samples. The best individual run uses angle encoding, four qubits, eight layers, and an MSE loss, achieving test $R^2=0.98495$ and MSE $5.6693\times10^{-5}$. Reported best-run BiLSTM baselines achieve $R^2=0.9529$ under restricted input conditions and $0.9805$ with longer, overlapping windows. Amplitude encoding gives the strongest mean performance across configurations, although its mean $R^2$ remains negative. Comparisons indicate that gate arrangement matters beyond parameter count, additional layers generally improve accuracy, and entangling encodings can partly compensate for limited ansatz entanglement. Selected simulator-trained circuits are also evaluated during inference on IQM Emerald using 1,024 shots. The reported mean $R^2$ reduction is approximately 1.25 on hardware, versus 0.06 in an SV1 comparison attributed to readout noise; the best hardware result is $R^2=0.77$. Shallower circuits retain more accuracy, while non-native gate decompositions increase implementation costs. Stated limitations include computational restrictions on circuit size, single-run hardware evaluations, substantial hardware-noise degradation, and unresolved explanations involving barren plateaus. Generalisation to other datasets and architectures remains unestablished, and a controlled quantum-advantage study is outside scope.
Central claim may not hold: The claimed QNN superiority relies on the highest individual test score found across a large configuration-and-initialisation search, rather than demonstrated superiority of a validation-selected model on an untouched test set. The winning QNN and the constrained BiLSTM use different window lengths, and identical forecast timestamps are not established, undermining the claim of equivalent data conditions. The architecture rankings establish that parameter count alone is insufficient to predict performance, but do not establish that gate selection and arrangement are more critical than parameter count. The entanglement comparisons confound entanglement with access to input features and other architectural changes, so they do not establish entanglement as a necessary forecasting resource.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.