Essays on latent factor models in financial markets: a focus on commodity prices and volatility dynamics

This PhD dissertation investigates financial applications of factor models, or latent variable models, with a focus on model specifications that link observable processes to multiple under- lying drivers in two distinct contexts: the modelling of volatility and commodity prices. Empirical evidence indicates that the conditional volatility of financial assets is more accurately captured by models that incorporate multiple stochastic components. Numerous studies have demonstrated the superiority of two-factor models over single-factor models in effectively capturing the complex dynamics of volatility. The generalized autoregressive conditional heteroskedasticity (GARCH) model, originally proposed by Bollerslev (1986), is one of the most popular models for filtering volatility in discrete time. Despite recent advancements in the literature of two-factor GARCH models, theoretical properties, such as ergodicity and strict stationarity, and further rigorous empirical testing are still lacking. This dissertation con- tributes to the literature by proposing a novel two-factor GARCH model. Regarding the modelling of commodity prices, it is well established that the uncertainty inherent in commodity price dynamics is a complex phenomenon, making it challenging to achieve ac- curate representations of reality and reliable future predictions. Consequently, there has been significant interest in models capable of capturing the stylized facts of different commodity prices. Among these, multi-factor affine models have been particularly valued for their analytical tractability and ease of estimation using the Kalman filter. This thesis demonstrates their practical relevance through a novel empirical application, highlighting their potential to address the complexities associated with commodity price dynamics.