Applications of machine learning in finance: analysis of international portfolio flows using regime-switching models

Recent advances in machine learning are finding commercial applications across many sectors, not least the financial industry. This thesis explores applications of machine learning in quantitative finance through two approaches. The current state of the art is evaluated through an extensive review of recent quantitative finance literature. Themes and technologies are identified and classified, and the key use cases highlighted from the emerging literature. Machine learning is found to enable deeper analysis of financial data and the modelling of complex nonlinear relationships within data. The ability to incorporate alternative data in the investment process is also enabled. Innovations in backtesting and performance metrics are also made possible through the application of machine learning. Demonstrating a practical application of machine learning in quantitative finance, regime-switching models are applied to analyse and extract information from international portfolio flows. Regime-switching models capture properties of international portfolio flows previously found in the literature, such as persistence in flows compared to returns, and a relationship between flows and returns. Structural breaks and persistent regime shifts in investor behaviour are identified by the models. Regime-switching models infer regimes in the data which exhibit unique characteristic flows and returns. To determine whether the information extracted could aid in the investment process, a portfolio of global assets was constructed, with positions determined using a flowbased regime-switching model. The portfolio outperforms two benchmarks, a buy & hold strategy and the MSCI World Index in walk-forward out-of-sample tests using daily and weekly data