Three Imperial physics students have won an international hackathon organised by Refinitiv, with another qualifying for the final stage.
On 1 December 2022, Imperial physics students Kennedy Au, Andrew Liang and John Ow won first place and £300 in an inter-university hackathon organised by Refinitiv, a subsidiary of the London Stock Exchange Group and global provider of financial market data and infrastructure. The local organisation was supported by Dr Ovidiu Serban and Ms Gemma Ralton, and by the Data Science Society Students (ICDSS) and the Mathematics and Design School student societies.
As part of a push to empower students to learn finance and coding, Refinitiv rolled out a product called 'Codebook' to allow students from any discipline to access easy-to-use coding interfaces to learn to code and apply it to management, finance, economics, computer science or data science. In line with this initiative, they set up an international inter-university hackathon, where students were given access to Refinitiv’s data to solve a data-related problem.
The winning team, EequalsMCsquare, presented a project centred on a short-term forecasting method called Bayesian Structural Time Series which they used to forecast Bitcoin.
They said: “We chose this project for two reasons: sharpening our own understanding of Bayesian statistics, as well as it being a unique approach that may give us an edge in the competition compared with other time series techniques.”
“After phase 1.5, we decided to pivot our idea after the feedback session and focus on creating an educational Jupyter Notebook that walks through our code, rather than unnecessarily overcomplicating the model. On the final day of the hackathon, we delivered a detailed presentation on a Bayesian vector autoregression model, and compared this to two standard machine learning models that we’ve also trained on the time series; namely Long short-term memory (LSTM) and XGBoost.”
Another Imperial team, FinTech Hackers, also qualified for the final stage and included Financial Technology students: Dimitry Tertychnyy and Patrik Kovac.
Their project focused on creating an Artificial Neural Network using Heterogeneous Autoregressive Quarticity, or HARQ-ANN, which is optimised to more accurately predict prices than the existing Black Scholes model (using Historical Realized Volatility) across all moneyness levels. The motivation behind their project was to inspire the addition of a realised volatility feature to Refinitiv’s Workspace.
Article text (excluding photos or graphics) © Imperial College London.
Photos and graphics subject to third party copyright used with permission or © Imperial College London.