ARIMA Time Series Model

[Len_Stringer]

I am attempting to build a model with the attached dataset and workflow. 

My ARIMA configuration and results are located in the PowerPoint Presentation.

My question is that the model works fine as long as I uncheck the first or last supplier in the list of covariates.

If I check all suppliers, I get the enclosed errors.

I would appreciate any help that you can provide.

Thank you,

Len

[AkimasaKajitani]

Hi @Len_Stringer 

Are there the fields as covariates which have all same data? If then, the ARIMA model or any other models can't accept such a field.

[LiuZhang]

pic 1

To answer the original question.

1: if you want to forecast Rejected, then you should remove the variable in the covariate list itself.

2: You are giving too many covariates in the model, once those variable are considered, there is insufficient degree in the supplied data to make meaningful slice of predictor. So use only relevant variables rather than feed everything in. 

3: You selected weekly as time period frequency, which is not true in your data. You may want to consider to aggregate them (e.g. sum or avg)pic 2

Additional comment:

Rejected is a count where most of them are zero's, so a time series may not be best for you. I would suggest count regression to try, as the QTY_REJECTED should be depend on the QTY_RECIVED (and maybe other variables), rather than an event that depends on sequence of date. You can try to do time-series on Received column, which makes more sense.