Machine Learning Time Series Regressions With an Application to Nowcasting

Real-Time Forecasting with a (Standard) Mixed-Frequency VAR During a Pandemic

“…Foroni, Marcellino, and Stevanovic (2020) generate GDP growth nowcasts and forecasts for the COVID-19 recession from a variety of mixed-frequency MIDAS models and explore adjustments of these forecasts based on the forecasting experience during the Great Recession. Babii, Ghysels, and Stiaukas (2020) use LASSO techniques to estimate MIDAS regressions for GDP nowcasts that also include text data. There is a related literature on real-activity tracking.…”

Boosting tax revenues with mixed-frequency data in the aftermath of COVID-19: The case of New York

“…Thus the large dimension of predictors comes from both the number of predictors and the fact that the predictors can be of high-frequency. Clearly it is a type of U-MIDAS model (Foroni et al, 2015), and can be implemented by machine learning (Babii et al, 2021). Note that x t+ω can include factor(s)…”

“…The literature on machine-learning forecasting with mixed-frequency data is also growing, see Lehrer et al (2019) for example. Babii et al (2021) have used sparse-group LASSO (sg-LASSO) in Mixed-Data-Sampling (MIDAS) models. But these powerful methods have not been widely used in governmental budget forecasting.…”

Forecasting Loan Default in Europe with Machine Learning

“…The range of tools available from the most recent ML literature for classification is very wide and its use in the area of credit risk analysis is still largely unexplored, with significant potentials for further study. We refer to Mullainathan and Spiess (2017) for a review of ML methods with an econometric perspective and an indication of possible applications and associated challenges if used to study economic problems, and to Carrasco and Rossi (2016), Medeiros et al (2019), and Babii, Ghysels, and Striaukas (2020) as some recent examples of ML methods applied to classical macroeconomic forecasting. This article follows the literature on ML models for default forecasting with a novel application to a multicountry European loan-level dataset.…”