Nowcasting Macroeconomic Variables Using High-Frequency Fiscal Data

Róbert Ambriško

Macroeconomic data are published with a time lag, making room for nowcasting macroeconomic variables using fiscal data. This is because a) monthly and daily fiscal data are available from the state budget in a very timely manner and b) many fiscal data are the function of macroeconomic variables. I employ two nowcasting models, bridge equations and MIDAS regressions, which link quarterly macroeconomic variables to monthly fiscal data for the Czech Republic. Bridge equations are found to be particularly suitable for nowcasting the wage bill using social contributions, achieving a 2% improvement in the root mean square error (RMSE) of one-quarter recursive forecasts compared to historical CNB forecasts. Further, I propose a tractable method for incorporating daily data into the nowcasting models, relying on STL decomposition by Cleveland et al. (1990). Depending on the timing, the RMSE for the wage bill can be up to 4% lower when the available daily data on social contributions are taken into account in the nowcasting models too.

JEL codes: C53, C82, E37

Keywords: Bridge equations, daily data, fiscal, midas, nowcasting, real-time data, short-term forecasting, STL

Issued: June 2022

Download: CNB WP No. 5/2022 (pdf, 817 kB)