Machine Learning & Deep Learning Applied to Quantitative Finance

Overview

The application of machine learning to quantitative finance represents one of the most active and consequential frontiers in the field. Classical models assume linearity, normality, and stationarity assumptions that financial data routinely violates. Machine learning offers a more flexible framework, capable of capturing complex non-linear relationships and adapting to the evolving structure of markets.

The challenge, however, is that financial data is noisy, limited in quantity relative to the number of potential features, and subject to regime changes that can render historical patterns obsolete. The objective of this department is not simply to apply ML methods to finance, but to apply them correctly with a full understanding of the statistical pitfalls and a commitment to genuine out-of-sample validation.

Deep learning extends these capabilities further: sequence models that learn directly from raw price history, reinforcement learning agents that optimise trading decisions across time, and generative models capable of simulating realistic market scenarios for stress-testing and risk management. These are powerful tools and this department is dedicated to using them with rigour.

What we work on

Members of this department develop cross-sectional return prediction models using regularised regression, gradient-boosted ensembles, and neural network architectures. We study the conditions under which ML-derived factors are statistically robust and economically meaningful, with careful attention to the risk of overfitting on financial data.

Research in deep learning covers LSTM and transformer-based models for time series forecasting, reinforcement learning applied to optimal execution and dynamic portfolio hedging, and neural network approaches to option pricing that bridge classical derivatives theory with modern AI methods. We also explore generative modelling for the simulation of synthetic market data, with applications in scenario analysis and model validation.

Every project is expected to meet a high standard: reproducible code, rigorous evaluation methodology, and an honest account of both the capabilities and the limitations of the approach. We believe that the most valuable research in this field is that which is transparent about when and why a model fails.