• forecasting
• nowcasting
• time-series
• VAR

This thesis discusses the relatively novel set of techniques aimed at \textit{nowcasting}, namely the practice of using recently published data to update, in real-time, key economic indicators that are published with a significant lag, such as real GDP. We review the two main approaches in achieving real-time predictions: joint models that can be cast in state-space representation (like VARs and Dynamic Factor Models), and "partial" models such as Mixed Data Sampling (MIDAS). We then explore some applications of nowcasted data in forecasting, by testing their short-term forecasting performance up to one quarter in different time frame specifications, using the theoretical framework of the HVAR model of Corsi, Longo and Cordoni (2023), which achieves correct identification thanks to high-frequency data. We find that, At high frequencies, nowcasted data is able to produce very good forecasts for up to one quarter. At the quarterly frequency, the forecasting performance is poor as the predictors carry much noise and, more importantly, the aggregation of high-frequency variables causes loss of information and the flattening of lead-lag relations across variables.