Inference and Prediction in Misspecified Models

In this thesis I explore the impact of model misspecification on statistical inference and prediction, with a particular focus on applications in finance and macroeconomics. I develop novel methodologies to address challenges arising from misspecified models, presenting three self-contained studies that each address a specific inferential or predictive issue. The proposed methodologies demonstrate improved performance in both simulations and real-world applications and compare favourably to existing methods, thus offering useful tools for practitioners to produce accurate forecasts and draw reliable inferences – even when the underlying models are misspecified.