Title: Seasonal adjustment and data inefficiency: Evidence from simulation and real-world data
Authors: Ronald Indergand - State Secretariat for Economic Affairs (Switzerland) [presenting]
Abstract: Seasonal adjustment based on moving-average filters is suboptimal from a forecast efficiency perspective. First, subsequent revisions to the data are comparatively large. Second, initial announcements are of mean-reverting character, that is, large initial growth rates tend to be revised downward. Both features increase the difficulty of assessing the dynamics of a variable and lead to suboptimal signals for policy makers. It is shown by simulation that model-based seasonal adjustment reduces the first feature of the data and eliminates the second if the data generating process is known. This results in more efficient preliminary estimates allowing for a more accurate assessment of the current state of an economic variable. Using GDP data from nine countries we demonstrate in a real-world setting how seasonal adjustment produces mean-reverting preliminary GDP releases. Overall, seasonal adjustment may account for the bulk of the results regarding mean-reverting data revisions that have been found by numerous studies.