Nowcasting and forecast evaluation under large instabilities

The Covid-19 pandemic represented an unprecedented source of economic instability, causing challenges in model estimation and forecast accuracy. The magnitude of this shock is evident in the rapid growth of the recent literature, as policymakers and macroeconomic forecasters found themselves unprepared and called for urgent revisions to existing methodologies. This doctoral thesis provides a review of the recent literature on nowcasting, forecasting, and forecast evaluation, with a particular focus on the instability triggered by the Covid-19 crisis. The thesis is structured into three chapters. The first chapter, Forecasting, nowcast- ing, and forecast evaluation, introduces the most widely used econometric models used for forecasting and nowcasting key economic indicators, such as the Gross Domestic Product (GDP) growth and inflation rate. In particular, it explores dynamic factor models and mixed data sampling models are detailed, focusing on estimation methods and model specifications commonly applied in empirical work. The chapter also covers prominent approaches to forecast evaluation, highlighting the global predictive accuracy test by Diebold and Mariano (1995) and the local predictive accuracy test by Giacomini and Rossi (2010). The second and third chapters present original research contributions. In the second chapter, Nowcast combinations in presence of instabilities, a novel performance-based forecast combination scheme is introduced, which employs past realizations of the target variable as dynamic thresholds. In addition, a new class of combinations is introduced, applying smoothing to individual forecasts using a fixed-size rolling window before aggregation. The integrated method, using both the dynamic threshold scheme and the smoothing of individual nowcasts, is then evaluated in an empirical application to now- casting the Italian Gross Domestic Product (GDP) growth over three periods of interest: Pre-Covid, Covid, and Post-Covid. The results show that, compared to the individual models and competing combination schemes, the proposed method delivers more stable performance across all three periods, showcasing resilience to the Covid-19 shock. Fore- casters are thus recommended to avoid relying on complex models to capture instability and instead employs forecast combinations as a hedge against breakdowns in forecast accuracy. The third chapter, Comparing predictive ability in presence of instability over a very short time, is co-authored with Prof. Fabrizio Iacone and Prof. Luca Rossini. This chapter investigates forecast comparison in the presence of instability when this affects only a short period of time. We demonstrate that global tests do not perform well in this case, as they were not designed to capture very short-lived instabilities, and their power vanishes altogether when the magnitude of the shock is very large. Therefore, we propose and discuss alternative approaches that are more suitable to detect such situations, such as nonparametric methods (S test or MAX procedure). The results are illustrated in different Monte Carlo exercises and in evaluating the nowcast of the quarterly US nominal GDP from the Survey of Professional Forecasters (SPF) against a naive benchmark of no growth, over the period that includes the GDP instability brought by the Covid-19 crisis. Based on the findings, we recommend that forecasters should not pool the sample, but exclude the short periods of high local instability from the evaluation exercise.