Real-Time Filtered Estimates of Final Superlative CPI Based on an Estimated Monthly VAR Model of Initial and Final Superlative CPI

• Activity Number: 320
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Business and Economics Statistics Section
• Author(s): Peter Zadrozny*+
• Companies: Bureau of Labor Statistics
• Keywords: Kalman filtering ; missing data ; revised data ; reduced-form modelling

Abstract

Initial SCPI estimates are released every month with 1 month delay. Final SCPI estimates are released every February for all months 2 years ago, with 15-25 months delay. Estimates are in historical form (bivariate, compact, with no missing values) when indexed by months of occurrence and are in real-time form (50-variate, sparse, with many missing values) when indexed by months of release. A bivariate VAR model is estimated using historical-form data. A Kalman filter, which can handle missing data, is applied to the estimated model in state-space form and the released estimates in real-time form in order to compute filtered estimates of final SCPI. Applied to recent U.S. data in log form, the method produces RMSE < 1% of filtered estimates of final SCPI, compared with RMSE < 20% of initial releases of SCPI, as estimates of final releases of SCPI.