Policy for seasonal adjustment. A direct method is used for all-time series. This means that aggregates and their components are all adjusted independently. There are no additional adjustments made to consolidate the differences between the adjusted aggregates and the sum of their components. This is considered to provide users with a clearer view of the economy. For the same reason, there is no post-adjustment benchmarking in order to match the adjusted time series to unadjusted annual totals. The outliers for reference periods are generally determined automatically during the adjustment process. However, there are some cases where pre-determined outliers are introduced. There have been four economic events that have had a notable impact on the time series: • the Russian crisis in 1998; • the end of the dot-com bubble after the turn of the century; • the Great Recession in 2009; • the Covid-19 lockdown in 2020. These events have caused notable shifts and/or outlier periods in multiple time series. In these cases, the automatic outlier detection is used only after the introduction of the outliers for these afore mentioned events. For seasonal adjustment, Estonia uses the TRAMO/SEATS4 method with the Xxxxxxx 2.2.0 program developed and provided by Eurostat. TRAMO/SEATS utilises ARIMA models with specific filters created for the adjustment of each time series. The selection of models is partly automatic. This means that the initial selection of a model is fully automatic but a subsequent adjustment and final selection is done by experts. There are no automatic parameters or validation procedures. Instead, all automatically selected models are evaluated and manual adjustments are done based on the diagnostics of the model and an analysis of results as deemed necessary by experts. This includes graphical analysis, descriptive statistics as well as the selection of the parameters of the models. All-time series are handled separately according to the same principles. The models are selected separately for each time series. This applies to time series in current prices as well as to price- adjusted time series.
