Analyzing the Effect of Data Revisions on Predictive Densities in a Small-Scale DSGE Model

• Activity Number: 20
• Type: Topic Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Dean Croushore*+ and Keith Sill
• Companies: University of Richmond and Federal Reserve Bank of Philadelphia
• Keywords: real-time data ; DSGE model ; forecasting

Abstract

We evaluate the impact of data revisions on predictive densities from a small-scale, New Keynesian, DSGE model. We begin by estimating the model in a quasi-real time framework using final vintage data and rolling over an expanding sample size. Predictive densities at multiple horizons are then generated (in output, inflation and nominal interest rates). These densities are then compared to those obtained when the model is estimated and forecasted using, instead of final vintage data, initial release, intermediate release, and annual revision vintage data. We document significant differences in predictive densities depending on data vintage.

The main revision that affects the predictive densities is the annual revision to the data that occurs once a year when the past three years of data are revised. We illustrate how the contours of the predictive densities change at each annual revision and explore how the predictive densities are affected by: (1) changes in jumping-off point for forecasts (revisions to the last three years of data); and (2) changes in structural parameter estimates.