Real‐Time Data, Revisions and the Predictive Ability of DSGE Models

Abstract

We evaluate the impact of real‐time macroeconomic data and data revisions on the forecasting performance of DSGE models in the US and the euro area. We identify significant differences in data revisions between the two world regions: Negative revisions (due to overestimation in early data releases) are prevalent in the US, while the euro area data tends to be dominated by positive revisions. These biases are most significant in consumption growth, output growth, and hours worked. Parameter estimates in small‐sized DSGE models are not strongly affected by the use of real‐time data, while larger models exhibit substantial differences depending on the data used, especially during large economic downturns. Revisions significantly affect the predictive accuracy of the DSGE model for output growth in the US and for inflation and interest rates in the euro area. Our findings highlight the central role of data revisions as a determinant of predictive accuracy in macroeconomic models and thus of the quality of such specifications as an instrument to inform evidence‐based policy‐making.