A graduate’s experience of an LSE Department of Statistics research internship on financial statistics

Evelyne Ong

I am a 2025 graduate from the Department’s BSc Data Science course. In the final year of my studies, I was among three students selected for the Department’s 2-month research internship.

My internship involved working under the supervision of Assistant Professor Dr Gelly Mitrodima on original research relating to her paper ‘A Bayesian Quantile Time Series Model for Asset Returns’ (Griffin and Mitrodima, 2020).

This paper introduces a time-varying quantile-based model for asset returns. Rather than modelling returns through a single summary statistic, such as volatility, the model focuses on different quantiles of the return distribution (e.g., the median or lower-tail quantiles that capture losses). By fitting these quantiles directly, the model describes how the entire shape of the asset’s return distribution changes over time. The model uses a Bayesian framework, treating the quantiles as uncertain and updating beliefs about them as new data arrive.

Overall, the approach fits a set of quantiles to the asset returns at each time step and can construct the full conditional distribution from them. In doing so, it provides more accurate estimates of risk measures such as Value at Risk (VaR) and Expected Shortfall than traditional volatility-based models like GARCH, illustrating the model’s usefulness in estimating and forecasting stock, index and commodity returns.

The first stage of the project was a meeting with Dr Mitrodima and the other research interns, during which she introduced the paper in detail and explained that our work would focus on extending it based on our own research interests. During this first meeting, we had the opportunity to ask questions about the paper and to discuss a possible idea Dr Mitrodima had for extending the research to include multivariate (analysing relationships among two or more variables simultaneously), as well as univariate (analysing and predicting a single variable using only its own past values), modelling.

We were then provided with a set of initial materials to explore further and decide what we wanted the scope of our final research project to entail, including a number of related research papers and a range of possible project concepts ranging from simpler ideas such as converting the model code from MATLAB to Python to more complex undertakings such as the univariate to multivariate conversion mentioned above. After completing an initial literature review and code review, we would then each meet with Dr Mitrodima to discuss our ideas and any questions we had.

For my project, I chose to pursue multivariate modelling. This would involve adapting the model from a single asset to a portfolio of assets. To better understand the finer details of the paper and to investigate solutions for multivariate modelling, I read a textbook on financial statistics recommended by Dr Mitrodima. This helped me get up to speed on topics such as time series and autoregressive models, since Dr Mitrodima’s paper covered them beyond the scope of third-year statistics modules. Following this, I completed an extensive literature review on multivariate asset return modelling methods. One of these methods, copula modelling, an approach that separates a multivariate distribution into its marginal distributions (individual variable behaviours) and its dependence structure (how they link), was also being investigated by Dr Mitrodima. My project resultingly opted to extend the Bayesian Quantile time series model to a portfolio of assets using an AC skew-t copula.

I met with Dr Mitrodima bi-weekly throughout the project to discuss my progress. Since the paper covered a number of topics I had not explored in my third year, I spent a considerable amount of time undertaking the literature review. After this was completed, I moved on to implementing this approach across a portfolio of assets over a 10-year period. The results suggested improvements in VaR estimates compared to the multivariate GARCH model. Finally, I concluded the project by writing a briefing and presenting my findings to Dr Mitrodima.

Combining theoretical learning with practical implementation, this project gave me a hands-on opportunity to extend my knowledge of complex topics in asset modelling, time series models and copula methods beyond the scope of third-year modules. It provided direct insight into the process of conducting research in a leading statistics department.

Following my research internship and graduation in July 2025, I am now working as a software engineer for the commodity trading desks at Macquarie. Other students who took part in the research internship are either currently completing an MSc in Statistics at LSE or working in similar fields. I would strongly recommend the research internship to any students who are interested in postgraduate study or research.

Dr Gelly Mitrodima

I had the pleasure of running the LSE statistics student research internships in 2025 and working closely with three excellent final-year undergraduate students. While the overall quality of applications was very high, these students stood out due to their prior research experience and their strong interest in pursuing MSc or PhD studies.

I introduced them to an ongoing research project and discussed its main challenges and possible future directions, but deliberately left the choice of focus to them. I wanted the students to take ownership of their work and develop confidence in making independent research decisions by pursuing questions they genuinely cared about.

Evelyne, who was one of the students, chose to extend our model to capture interdependencies within a portfolio setting. She demonstrated outstanding coding skills and a deep understanding of concepts well beyond the undergraduate curriculum, making collaboration both productive and enjoyable.

Overall, the internship was a rewarding experience for everyone involved. The students valued the opportunity and are now either pursuing or applying for postgraduate study, and for me it was especially gratifying to see how much they were able to extend existing research across multiple dimensions, including coding, modelling, and theory-driven analysis.

By Evelyne Ong and Dr Gelly Mitrodima