Seasonal adjustment strategies for official statistics in an era of unpredictable events

Abstract

The Covid-19 pandemic significantly disrupted the production of official economic statistics, undermining both the reliability and predictive capacity of short-term stochastic models. With the sequence of shocks from 2020–2021 now behind us, it is essential to calibrate an ex-post strategy for revising and improving the seasonal adjustment models that were built in real time during the emergency period, particularly in relation to the treatment of anomalous observations. This study defines and tests several intervention strategies on time series of different nature that experienced clear structural breaks in their dynamics, specifically, quarterly Expenditure by Non-Residents in Italy, monthly Nights Spent by Italian Citizens in Hotels, and Industrial Production Index for the Wood and Paper Sector. Temporary level shifts, including their lagged versions and custom intervention variables, proved to be especially effective. The results show that the proposed strategies enhance the quality of seasonal adjustment: the TRAMO-SEATS procedure, combined with user-defined variables, yields better pre-processing models and reduces revisions in the adjusted series, particularly for monthly data. All analyses were conducted using JDemetra+, a comprehensive and flexible tool for seasonal adjustment and statistical diagnostics.