Problem with different forecast values in TS Exponential Smoothing in SAS Enterprise Miner

[Juliadrs]

Hi there! 

I have been working with SAS Enterprise Miner for several weeks now and I had to develop forecasting models using SAS Enterprise Miner for the company I work for. I used the time series exponential smoothing node. Most forecasting models work very fine with absolutely no issue, except for two of them. Indeed, I realised that the results of the node give two different set of values for the forecasting. Indeed, the forecast result including the confidence intervals and the actual vs predicted values display different forecasted values than the multiple forecast comparison plot (e.g. the first step forecasted value is 302 on the multiple forecast comparison plot and the first step forecasted value from the same model in the forecast result graph is 510).

I then decided to export the values to see which ones were really forecasted. When I export the values of the forecast using default properties, the forecasted values are the ones matching the multiple forecast comparison plot but they are not matching the one from the forecast result graph. When I export the values using the coordinated forecast data option, the forecasted values are different and then the forecasted values do not match the multiple forecast comparison but they do match the values from the forecast result graph.

This only happens when I leave the property "Best" for the model chosen in the TS Exponential Smoothing node. If I choose a model, all forecasted values are matching in both graphs. After trying several models, I did find out that when Enterprise Miner gives me two different sets of forecasted values (so one graph has different forecasted values than the other and both exported values I showed above also differs), these two different sets match the results of two different models, the additive seasonal and additive winters model.  It seems that EM is mixing the results from two models. I showed that to one of my colleagues that is more experienced than me in SAS EM and we both think that this problem seems to come from SAS EM itself. Has this problem ever been encountered before and is there a way to fix it?

I'm on the version 14.2 of SAS Enterprise Miner and my data is very limited. I have only 28 observations, I specified seasonality 12 and interval "month" in the TS exponential smoothing node.

Best,

Julia 

[alexchien]

Could you provide the data so that we can look into the issue in more details? thanks alex

[Juliadrs]

Unfortunately, the data used for this task is private so I am not allowed to share it. I can only share some of the details regarding the data.

[alexchien]

Hi Julia,

We tried a few examples we have but we can't reproduce the issue you are observing. I am wondering if u can add some noise + some transformation to skew the data but still able to reproduce the issue. We can use the skewed data for more detailed diagnoses. Or you can contact SAS tech support to help you with your issue off the forum. thanks

Alex

[mitrov]

Could you please contact tech support? They will be able to guide to provide the information that is needed to investigate the issue. If any data is shared it will be maintained confidential.