I have a time series sales data and I am trying to figure a model for decomposition the components, but I am not sure on what model to choose Additive or Multiplicative as the graph looks confusing.
Is there way other than looking at the graph to decide the model?
I know that I must choose Multiplicative if the seasonal variation is increasing or decreasing as the trend increasing does that mean the difference between Jan and Dec of each sales year?
What do you mean by multiplicative/additive? I will give you some keywords, if you are interested you can read them up, e.g. on Wikipedia, and specify your model then accordingly.
1) Test whether your data is stationary or not. A quick and dirty check would be by plotting the actual time series. If it has a trend, the series is not stationary.
2) If the series is stationary: You could specify your model by plotting the correlogram. From here you can decide whether a (seasonal) MA(q)- or AR(p)-model (or ARMA(p,q)) might be appropriate (p,q denote the number of lags).
3) If the series is not stationary, you have to determine what kind of trend you have in the data (stochastic or deterministic), e.g. via a Dickey-Fuller test. Based on the outcome of the test, specify the model according to step 2.
