Machine Learning and Network Analysis in the Options Market

This project focuses on the statistical analysis of the options market (financial instruments that are derivatives based on the values of underlying securities), by leveraging tools from the literature of machine learning and network analysis.

The data set used contains data for options transactions including price and volume information, broken down by market participant. The project fits well into the Information and communication technologies (ICT) research area; such a data set could not only offer itself for a myriad of potential applications but also provides motivation for the development or augmentation of existing techniques for analysis of large time-series data. While financial data are a classic instance of time series, these are widely used across various disciplines to analyse data regarding sensor measurements, personalised medicine, social networks, astronomical observations, or any other data which display an intrinsic temporal structure.

A main objective of the project is to perform dimension reduction on the options data with the primary goal of extracting risk factors for the underlying instruments, thus potentially augmenting existing risk models, either statistical or based on fundamentals, of interest to both academics and practitioners. Subsequent downstream tasks could include the detection of anomalies that signal upcoming or undergoing market shifts. Once the relevant time series of low-dimensional structural summaries have been identified, various state-of-the-art techniques for anomaly detection and change-point analysis will be explored. Community detection algorithms designed for time series data will also be performed, providing a clustering of the various instruments, which often differs significantly from the traditional sectorial classifications of the underlying instruments.

Linear dimensionality reduction methods, such as Principal Component Analysis (PCA), have been used extensively to study financial markets. In the last decade, nonlinear methods have also been used on data sets of financial indicators to capture low-dimensional representations of time series, for the purpose of prediction or detection of early signs of structural changes in the system. However, to the best of our knowledge, there has not yet been such a study based on the option market data. Robust estimation of the intrinsic dimensionality in the options and equity markets, and identification of shifts in the market regimes, would be of equal interest to both practitioners and regulators. Enhancing risk models, and improving the accuracy of algorithms predicting volume flow, are tasks of major importance for practitioners, especially market makers. On the other hand, policymakers are interested in a better understanding of systemic risk, construed as the risk of breakdown of the entire financial system and cascading behaviour that can have repercussions on the entire economy. A question of relevance for policymakers is whether, by closely tracking the various risk factors that govern the dynamics of the market, one can detect early signals of stress or upcoming crises.

The project is aligned with the following topics: (1) Artificial intelligence technologies, (2) statistics and applied probability, (3) non-linear systems, (4) operational research.