Shaoni Nandi, King’s College London
Project Description/Abstract
This paper builds a unified mixed-frequency panel data nowcasting framework for nowcasting growth rates of national real GDP. Nowcasting until now largely focuses on single country frameworks. The proposed model extends the existing Mixed-Frequency Panel Vector Auto Regression (MFPVAR) to include heterogeneous coefficients that allow diversity among countries, and multi-factor error structure that represents cross-sectional dependence between the interlinked economies. Monte-Carlo simulation results establish that the panel model outperforms time-series and pooled panel autoregressive benchmarks. The empirical application estimates the model to nowcast GDP growth for a large pool of advanced and emerging economies. Model selection and dimension reduction are simultaneously achieved through shrinkage-based estimation. We perform out-of-sample pseudo-real-time experiments, taking into account a two-level international asynchronous release calendar of macro-releases: between and within countries. The empirical results display significant gains in nowcast accuracy against standard benchmarks. Notable nowcast gains throughout the quarter are obtained during the COVID impacted quarters of 2020.