Quantum Enhanced Signature Kernels for Limit Order Book mid-price predictions

Empirical evidence suggests that well-chosen quantum feature maps promise to enhance the performance of Machine Learning (ML) algorithms. Theoretical arguments suggest that the specific property that makes these quantum transformations useful for ML purposes is the amount of ‘magic’ or non-stabiliserness of the output state of a circuit defining the quantum feature map. In this work, we use an exact state vector simulation to experimentally test the use of quantum enhanced signature kernels up to 32 qubits to predict the mid-price using Limit Order Book (LOB) data. We benchmark our model against state of the art deep learning models.