Factor-augmented sparse MIDAS regression for nowcasting

Author:

Jad Beyhum, Jonas Striaukas

Keyword:

Economics, Econometrics, Econometrics (econ.EM)

journal:

date:

2023-06-22 16:00:00

Abstract

GDP nowcasting commonly employs either sparse regression or a dense approach based on factor models, which differ in the way they extract information from high-dimensional datasets. This paper aims to investigate whether augmenting sparse regression with (estimated) factors can improve nowcasts. We propose an estimator for a factor-augmented sparse MIDAS regression model. The rates of convergence of the estimator are derived in a time series context, accounting for $\tau$-mixing processes and fat-tailed distributions. The application of this new technique to nowcast US GDP growth reveals several key findings. Firstly, our novel technique significantly improves the quality of nowcasts compared to both sparse regression and plain factor-augmented regression benchmarks over a period period from 2008 Q1 to 2022 Q2. This improvement is particularly pronounced during the COVID pandemic, indicating the model's ability to capture the specific dynamics introduced by the pandemic. Interestingly, our novel factor-augmented sparse method does not perform significantly better than sparse regression prior to the onset of the pandemic, suggesting that using only a few predictors is sufficient for nowcasting in more stable economic times.

PDF: Factor-augmented sparse MIDAS regression for nowcasting.pdf