Essays in computational Finance

This thesis presents three independent paper-style empirical studies in computational finance, leveraging methods from computer science and complex network analysis.

In Chapter 1, we aim to contribute to the literature by introducing a novel classification model that takes into consideration unstructured Twitter data and metadata to categorise tweets based on the value-relevance of their information content. Our classifier shows promising results, achieving a 5-fold cross-validation accuracy of 82.1%. In Chapter 2, insider trading data from the U.S. Securities and Exchange Commission (SEC) are analysed to assess the profitability of their transactions. Sentiment analysis and social media data are incorporated to gauge the public sentiment surrounding these transactions. Insiders’ purchase transactions yield statistically significant abnormal returns in the short term, indicating their superior knowledge of their companies’ future performance compared to the general public. These findings can provide valuable insights for regulators, highlighting potential trading opportunities based on non-public information.

Chapter 3 examines a real central clearing counterparty network encompassing 15 central clearing counterparties (CCPs) and 350 clearing members (CMs). A simulation model is proposed to study the transmission of shocks in this network, particularly exploring the repercussions if CCPs were to face a common global stress event such as the default of multiple globally systemically important banks (GSIBs). We also investigate how variations in margin requirements as well as variations in the network structure could impact the default rate in the system. We find a positive correlation between the average default rate in the system and the number of defaulting GSIBs, and a non-linear relationship between changes in the variation margin and the default rate. Additionally, we find evidence that as the network grows, it becomes more resilient. This stress testing approach can help regulators capture and assess the shared cross-border risk factors among CCPs.