1

I have a problem with identifying seasonality without using plots.


From what I have understood, the best way to detect seasonality in a time series is by understanding the data, plotting it and analyzing the autocorrelation plot.

Now I would like to apply this process to sales data about multiple products from multiple companies, which means thousands of different time series. Since I cannot take the time to understand and plot data for every product, I would like this process to become "automatic". Automatic in the sense that there would be a data pipeline where, amongst other things, there will be a seasonality test and then a seasonal difference (if seasonality is detected) etc...

Just so you can understand: the pipeline's final goal is to return forecasts. That is why I am trying to make my data stationary.


I don't think plot analysis is possible (or, at least, easy to implement for a beginner like myself) in such a pipeline. Therefore I cannot rely on it.

So my question is:
Are there any non-graphical tests that I can use in order to detect seasonality ?

I have looked into Ljung-Box but it seems that it is only used for residual analysis when fitting models, which is not what I am looking for.

Thank you for your help!

Amine
  • 11
  • There are statistical tests for seasonality but your objective seems to not require it. First, even if a series is stationary it can have seasonality. Second, taking difference at seasonal frequency does not necessarily eliminates it, even if it does it is not the best way. Third, if ultimate objective is to forecast, you'll end up fit a model to each series, so might as well check for seasonality. If you are looking for automatic forecast, do explore auto.arima in forecast package in R. It fits Seasonal ARIMA so takes account of seasonality automatically. – Dayne Oct 29 '20 at 12:33

0 Answers0