Learning patterns/Evaluation of data affected by seasonal or calendar effects

A learning pattern forevaluation
Evaluation of data affected by seasonal or calendar effects
problemSeasonal or calendar variation can make it hard to understand the trends and processes behind your data.
solutionSeparate the seasonal or calendar effects from the base data, and create trends from the cleaned data.
creatorSamat
endorse
created on15:57, 14 September 2019 (UTC)
status:DRAFT

What problem does this solve?

edit

We often need to work with time series. These data are typically affected by seasonal or calendar effects (most of the time by both of them). These effects can be smaller or larger compared to the base data itself. In many cases, they make it hard to see and understand what is going on, what is the trend and what processes are behind the data.

What is the solution?

edit

Estimating and removing the seasonal and calendar effects from our time series help us see better and clearer what our data means. After trying out and testing several software, I used JDemetra+ for my study.

JDemetra+ is a tool developed by the National Bank of Belgium (NBB) in cooperation with the Deutsche Bundesbank and Eurostat, and it is the recommended[1] statistical tool for members of European Statistical System (ESS) and for members of the European System of Central Banks (ESCB). The software is open source and is under the EUPL free license[2]. JDemetra+ implements the concepts and algorithms used in the two leading seasonal adjustment methods: X-13 ARIMA and TRAMO/SEATS. X-13 ARIMA is a filter based method, while TRAMO/SEATS is a model based method.

Personal experience
  • The software is quite user friendly and has an easy to use graphical interface; it does not require too much knowledge about the methods themselves, if you do not want to go into details. You select your dataset and the software gives you back the results of the analysis graphically, in tables, and in a report style (with quality measurements and details of the analysis).
  • The tool separates the seasonal and calendar effects, the trend of the dataset, and the irregular part of the series. You can visually analyze the results inside the software, or copy the table of results into a spreadsheet program for further analysis and adjustments.
  • It provides forecasts for your data (trend, seasonal and irregular series, and the time series cleaned from the seasonal and calendar component) based on the input dataset.
  • It is very useful not only for analyzing the trend of the dataset, but also for analyzing the seasonal and calendar component of the original time series, which provides valuable extra information about the series.
  • It is not clear which one of the two methods is better. The best way is to try both of them, and use the one which works better with the specific dataset.
  • For the data cleaned from the seasonal and calendar effects, the TRAMO/SEATS method looked a little bit better, but for the trend lines, X-13 ARIMA handled sudden drops or jumps more nicely.

When to use

edit

If you have a time series with a high enough number of data points affected by seasonal or calendar effects. (You can see in the software if these effects were regular, how strong, and what is the confidence level based on the different models.)

Endorsements

edit

See also

edit
edit
edit

References

edit