A0823
Title: Factor-based nowcasting with missing data
Authors: Mark Hallam - University of York (United Kingdom) [presenting]
Abstract: Motivated primarily by the problem of producing nowcasts for large unbalanced panels of data, methods are developed for estimating the latent factor structure and values of missing observations in panels with general patterns of missing data, including those with mixed sampling frequencies. The approaches combine recent developments in factor-based imputation for partially observed panels proposed by prior studies with the Kalman filtering and QML-based approaches for estimating approximate dynamic factor models in prior studies. This results in two-step estimators of the latent factors and missing observations that allow for general patterns of missing data, dynamics in the latent factors, and cross-correlation in the idiosyncratic components. Simulation exercises are conducted, using DGPs commonly employed in the factor model literature, modified to allow for various patterns of missing data. It is found that the proposed methods generally perform well, providing accuracy gains relative to the simpler one-step factor-based imputation methods in both the estimation of the latent factors and imputation of the missing values. Finally, the proposed methods are applied to a macroeconomic nowcasting exercise using a real-time US macro-financial dataset over the period 1990 to 2025.
