Matthew Lawrence
Thesis Title: Modelling the Dynamics of Financial Risk: Non-Parametric Latent Volatility Measures and Dimension Reduced Modelling in Portfolio Construction
Although traditional parametric methods of multivariate volatility modelling are prominent in the financial econometrics literature (for example the VEC(p,q)) they are inseparable from two key weaknesses: computational complexity and, crucially, over-fitting - the statistical phenomenon whereby models with many parameters are so dependent on their training data that they perform poorly in out-of-sample forecasting tests. Both of these failings can be attributed to the high number of parameters (dimensions) required by traditional parametric models, but dimension reduction techniques such as latent factor modelling, neural networks (particularly autoencoders), and LASSO regression analysis help to reduce this.
The essence of my research is to apply and manipulate popular techniques for dimension reduction such as latent factor modelling and machine learning methodologies to non-parametric estimators of portfolio risk in order to separate observed co-volatilities from the underlying elements driving them. In addition to providing more accurate risk measures and less computational complexity for investors, dimension reduced non-parametric risk modelling would also enable us to tackle problems such as asynchronous portfolios. By incorporating a regime switching mechanism for periods when each asset is traded/untraded, we can circumnavigate the mathematical imposition of there being a zero probability of a zero return (a necessary by product of assuming asset log-prices follow a continuous-time stochastic volatility process). As risk modelling in itself is only useful to construct better performing portfolios, I will seek to further manipulate and apply forecasted Realised Volatility matrices in order to judge the portfolio by common metrics such as the Sharpe Ratio which will give my research further applicability.
Biography:
Matthew Lawrence is currently an ESRC Midlands Graduate School PhD Candidate in Economics, having previously completed his MSc Financial Economics (Distinction) and BA Modern History and Economics (1st Class Joint Honours) at the University of Manchester. His research field is financial econometrics, specifically using machine learning, latent factor modelling, and time-series forecasting in the context of financial statistics and portfolio optimisation.
Matthew has previously worked with the Quantitative Research group at the University of Manchester Ethical Investment Society, and is passionate about applying his research to the wider context of financial markets.
Management and Business Studies,
University of Birmingham
2025 Cohort
Email:
Supervisory Team:
Dr Marco Barassi
Dr Ekaterina Kazak