Regression Performance of Temporally Disaggregate Variables

• Activity Number: 169
• Type: Topic Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Jens Mehrhoff*
• Companies: Deutsche Bundesbank
• Keywords: Temporal disaggregation ; Denton procedure ; Chow-Lin procedure ; Linear regression ; MIDAS regression ; Kalman filter

Abstract

A currently debated topic are temporal disaggregation techniques, e.g. for annual property prices. Although modern econometric estimation methods, such as mixed-data sampling (MIDAS) and the Kalman filter, are tailor-made for the problem at hand, much more frequently applied are the Denton and Chow-Lin procedures to the lower frequency time series in question.

We analytically derived and empirically simulated the performance of these two procedures in a linear regression framework, starting with the time series properties, e.g. seasonal heteroskedasticity and co-integration, and moving to the asymptotic distribution of the ordinary least squares estimator.

The results point to serious problems with data generated from the Denton procedure while the Chow-Lin procedure can generate reliable results even with bad, i.e. only weakly correlated, auxiliary variables (in the most extreme case independent white noise).

Nonetheless, the Denton procedure is by far the most used methodology in practice. That the Chow-Lin procedure performs reasonably well is because it pays attention to the time series's properties, albeit difficulties remain in the interpretation of infra-annual movements.