Short-term real-time forecasting during turbulent times. A model for the Spanish GDP after the pandemic

Abstract

Following the outbreak of the COVID-19 pandemic, most economic indicators experienced an increase in observed volatility, reducing the accuracy of nowcasting econometric models. In this paper, we propose a new specification for a mixed-frequency dynamic factor model used to nowcast the quarterly GDP growth rate of the Spanish economy –the Spain-STING–. With the aim of improving the predictive capacity of the model, we consider three proposals: (i) the relationship between the indicators and the estimated common factor is now contemporaneous, and not leading for some of the indicators; (ii) the variance of the common component is estimated by a stochastic process to allow it to vary over time; (iii) the set of variables is revised with the aim of including only those that add the most relevant information to the nowcast of the quarterly GDP growth rate. All these three modifications imply a notable improvement in the nowcasting performance during the period after the COVID-19 pandemic, while maintaining the accuracy obtained before it. These proposals could be also useful to revise other forecasting models.