A1378
Title: Multidimensional panel data regression model: The case of the multidimensional homeownership vacancy rate in the USA
Authors: Talha Omer - Linnaeus University, Växjö (Sweden) [presenting]
Daniel Henderson - University of Alabama (United States)
Andros Kourtellos - University of Cyprus (Cyprus)
Abstract: The purpose is to extend the two-dimensional panel data regression model to a multidimensional setting for mixed-frequency data. Structured machine learning regression is explored in this context, incorporating techniques such as sparse group LASSO (sg-LASSO), LASSO, U-MIDAS, elastic net MIDAS, and average MIDAS. The theoretical properties and mathematical equations of these multidimensional panel data regression models are detailed. Their performance is tested via Monte Carlo simulations and two distinct data-generating processes (DGPs), each varying in sample size, number of variables, and data frequency. As an empirical application, three-dimensional home ownership vacancy rates in the U.S. are nowcasted via the extended model, and the nowcasting robustness is assessed via the Diebold and Marino test. Additionally, a random walk forecast is computed and considered a benchmark for nowcast evaluation. The performance of all the MIDAS models used against traditional two- and three-dimensional panel data regression methods is compared using the out-of-sample root mean square error (RMSE) as a nowcasting accuracy metric for both simulated and empirical data. The results indicate that the three-dimensional MIDAS panel data regression model outperforms the comparative models, demonstrating a lower out-of-sample RMSE than the other models do.
