Using Demetra at SURS Manca Golmajer Andrejka Smukavec

Using Demetra+ at SURS Manca Golmajer, Andrejka Smukavec 20 June 2013

Organization of SA • Time series methodologists – Modelling (model selection and modification) – Training (study, preparation of literature, courses) Contact: zascas. surs@gov. si • Survey managers – Preparation of original data – Automatic seasonal adjustment – Publishing

Seasonal adjustment software and method Software: • Demetra 2. 04: until 2013 • Demetra+ 1. 0. 4: between 2013 and 2014/2015 • JDemetra+: from 2014/2015 on Method: • TRAMO/SEATS

Introduction of Demetra+ Gradual introduction of Demetra+ during 2013: • New time series: Demetra+ is used • Old time series: introduction of Demetra+ is joined with annual review FIELD LAST PERIOD PUBLICATION Industry, construction, trade, services 1/2013 February, March 2013 Labour market Q 1/2013 June, July 2013 Business tendency 10/2013, Q 4/2013 October 2013 National accounts Q 3/2013 November 2013

Associated activities • Literature: – Time series manual (for survey managers) – Time series glossary – Seasonal adjustment of time series (methodological explanations) – Commenting of seasonal adjustment of time series – Demetra+ manual (for time series methodologists) • Course for survey managers • Presentation for the Bank of Slovenia and the Institute of Macroeconomic Analysis and Development

Input and output file • Input file: – Excel • One sheet • One type (monthly or quarterly time series) • Dates: first day of the period • Output file: – Excel – By. Component • Used by most survey managers • Dates: first day of the period – Csv – VTable • Dates: last day of the period

Refreshing • Transformation, calendar effects, regression effects (pre-specified outliers) and the ARIMA model are fixed, only automatic outlier detection might be enabled for the last few periods. • The refreshing option: Partial concurrent adjustment – All outliers (+params) – Options Concurrent adjustment and Partial concurrent adjustment – Arima and outliers (+params) do the same in our case – Option Partial concurrent adjustment – Last outliers (+params) enables automatic outlier detection for the whole last year

Checking the results • Seasonal adjustment of a time series is successful if: 1. summary is Good 2. all the other diagnostics (except for the number of outliers) are Good or Uncertain 3. automatic outlier detection is disabled • If seasonal adjustment of any time series is unsuccessful, the survey manager must send the multi-processing to the time series methodologist to check and maybe change the model (check the Bad diagnostics, include an outlier, etc. ).

Some positive things about Demetra+ • It is very user-friendly: – Well organized – Colours are used – Our survey managers like the Excel output • It has a lot of possibilities, tools, tests, etc. It gives the user a lot of information and better understanding of the seasonal adjustment process. • It is much faster to select a model (transformation, calendar effects, outliers, ARIMA model) with Demetra+ than with Demetra.

Some negative things about Demetra+ • We do not like the chart scale.

• It is difficult to compare different models for the same time series. • The maximum values for the coefficients of the ARIMA model are (3, 2, 3)(1, 1, 1). In Demetra we also used (0, 1, 1)(0, 1, 2). In Demetra+ we use (0, 1, 1)(1, 1, 1) instead.

Some problems with Demetra+ • Strange values for time series components

Methodological improvements • We can send our questions to ESTATMethodology@ec. europa. eu. • We started using trading days effect (6 or 7 regressors). Before we used only working days effect (1 or 2 regressors).