Essays on Machine Learning approaches to Causality and Forecasting

This thesis focuses on employing machine learning estimators and related techniques in time series econometrics. Specifically, this research aims to provide a comprehensive overview of penalised linear estimators, focusing on their theoretical attributes. It also proposes methods and applications in causal inference and forecasting, introducing innovative techniques and leveraging new data sources such as sentiment and attention variables. The abundant availability of data in the modern era is a critical aspect that distinguishes contemporary econometrics from its historical counterparts. This shift has prompted researchers to adapt and develop new models and study estimators capable of handling situations with numerous variables, often exceeding the number of observations. In line with this aspect, this thesis aims to equip economists and policymakers with the necessary tools to make well-informed economic policy decisions, and to assist businesses and investors in making wise investments and strategic choices. This research will be a valuable contribution to the field of forecasting and causal inference, helping to improve the accuracy of predictions and providing useful tool in the context of penalised linear regression based inference methods applied to time series.