I am working with a batch of about 1000 univariate time series in R . For every time series, I have to perform following tasks , before deciding upon a model be it ARIMA, TAR or Holt Winter's Model
- Trend Detection and its type , i.e. whether trend is deterministic or stochastic
- Seasonality Detection and then deciding whether it is additive or multiplicative
- Does the series needs transformation. If yes then what kind of transformation is required, i.e whether box-cox or logarithmic.
Currently I have to visualize every series and then take a call , are there any mathematical criterion available, which can reduce this effort
Also what are the other factors that I need to consider before deciding on which model to use