Comparación de métodos de regresión y reducción de dimensión para el nowcasting del EMAE[A Comparison of Regression and Dimensionality Reduction Methods for EMAE Nowcasting]

This paper evaluates different alternatives for nowcasting the year-on-year growth rate of Argentina’s Monthly Economic Activity Estimator (EMAE). A reduced set of high-frequency economic indicators - electricity demand, real tax revenue, automobile sales, and cement consumption - is used for the period from January 2015 to June 2026. An autoregressive distributed lag (ARDL) model, a dynamic factor model based on principal component analysis, and three regularization methods - LASSO, RIDGE, and Elastic Net - are compared. The results show that the ARDL model provides the best in-sample fit, with an R^2=0.92, and statistical evidence of a long-run relationship between EMAE and the indicators considered. The LASSO, RIDGE, and Elastic Net models achieve similar levels of fit, while offering advantages in dealing with multicollinearity and, in the case of LASSO and Elastic Net, allowing for variable selection. The dynamic factor model exhibits a lower fit (R^2=0.64), but provides a more parsimonious representation by summarizing the information contained in the indicators into a single factor. In terms of contemporaneous effects, cement consumption and electricity demand show the strongest associations with EMAE growth. The results should be interpreted with caution, since the high R^2 values obtained by the ARDL and regularized models reflect in-sample fit and, on their own, do not establish which methodology has greater predictive ability.