Quantum Weighted Moving Average for Predicting Limit Order Book Trends

This paper investigates whether quantum machine learning can be effectively applied to forecasting multivariate financial time series, specifically focusing on predicting price trends from limit order book (LOB) data. The authors propose the Quantum Weighted Moving Average (QWMA) model, a hybrid classical-quantum architecture. The approach involves two primary stages: classical preprocessing via bilinear normalization to handle the LOB's feature and temporal dimensions, and a quantum routine based on a Linear Combination of Unitaries (LCU) to process the embedded time series. The authors also explore fixed-weight variants, specifically the Quantum Simple Moving Average (QSMA) and the Quantum Exponential Moving Average (QEMA), to reduce the complexity of state preparation.

The QWMA model achieves predictive performance comparable to leading classical benchmarks, such as the Temporal Attention Augmented Bilinear Network (TABL), on the FI-2010 benchmark dataset and a dataset of China A-share stocks. The authors find that the quantum model is expressive enough to identify and focus on the most predictive parts of the time series. While the QWMA model with trainable weights performs well, the QEMA variant—which uses exponentially decaying weights to emphasize recent data—often matches this performance while being easier to implement. The study highlights that proper input normalization is critical; without it, the quantum model's performance degrades significantly.

Financial forecasting is a high-stakes domain where even marginal improvements in predictive accuracy can be valuable. This paper provides a concrete, modular framework for integrating quantum routines into existing financial machine learning pipelines. By demonstrating that quantum models can compete with classical deep learning architectures on standard LOB benchmarks, the work offers a potential path for future quantum applications in market trading, even if a definitive quantum advantage remains to be proven.