Nowcasting Argentina's GDP: A Comparison of Multiple Methodologies

Congreso; International Conference on New Trends in Econometrics and Finance; 2020

Nowcasting, the task of measuring GDP in real time and forecasting its short term forward trajectory, has been the goal of macroeconomists and econometricians for several decades. Starting in the early 90's with simple linear regressions as in the seminal work of Braun (1990) [1] and Fitzgerald and Miller (1989) [2] econometric models have evolved with the adoption of a variety of techniques. In 2008, the famous paper by Giannone et al [3] inaugurated a brand new framework that was further developed by Stock and Watson [4]: dynamic factor models that exploit common unobserved factors in macroeconomic series by means of a state-space representation and the Kalman filter. In the last decade, econometric models that make use of the novel machine learning algorithms have proliferated, and nowcasting hasn't been an exception to this trend. Elastic net regularization as a strategy to mitigate overfitting has been applied to the forecasting of US GDP with considerably good performance Smalter Hall (2018) [5]. However this approach requires that all predictor variables share the same frequency. Finally, in an alternative approach, St Louis Fed researchers Grover, Kliesen and McCracken (2016) [6] propose a nowcasting method based on an "Economic News Index" (ENI), which consists on the surprise component-defined as the difference between the median consensus estimate and the realized value of the economic variable-of key monthly economic data releases and is then used to update the forecast of real GDP growth. The ENI is built using a weighted linear combination of 68 predictor macroeconomic variables, and the weight is computed by applying a 3-Pass Regression Filter to these variables. The current paper builds on these 3 approaches to conduct a nowcasting exercise on Argentina's GDP. The objective of this work is twofold: on one hand, a comparison of the performance of the three different models to estimate Argentina's GDP in real time, and on the other hand an evaluation of the reliability of the predictive capacity of these models when they are applied to a highly unstable country as is Argentina, with marked oscillations and high volatility in its macroeconomic variables.