Fixing straight line ETS Output

[eaphymao]

I have just over 2 years of sales data (110 weeks), and the data clearly shows a downward trend and regular seasonality.

However, when I put the data through the ETS tool i'm only able to get a straight line output (auto on all ETS model settings) or a straight positively sloped line (Multiplicative error type, additive trend, no seasonal type). Every other configuration seems to give an error? 

I'd love to know why the ETS tool can't seem to create a reasonable model. Thanks! 

[danilang]

Hi @eaphymao 

I'm not sure why your ETS is returning a straight line.  I'm just starting my journey though the arcane world of forecasting myself.  According to this post, the reply from @AndrewKramer suggests that the configuration of your ETS model may need to be tweaked.  

I did, however, replace your ETS tool with the ARIMA tool as suggested in the post.  The output of this shows the downward trend with seasonal variations that you're looking for

The O output of the TS Forecast tool also gives you forecasted data values, something that the ARIMA tool does not.

Dan

[AndrewKramer · Accepted solution]

Your model has the structure ETS(ANN). This means R determined your data has Additive Error, with no Trend or Seasonality. This is why you are seeing as your confidence interval gets larger with each period into the future, but the forecast stays the same, a flat line

When trying other models, such as ETS (ANN), in RStudio, R errors out since the data is not seasonal. This is why you are getting an error with other combinations in Alteryx, just the error is an Alteryx error.

> forecast <- ets(y = data$Sales.Quantity, model = "AAM", damped = FALSE, lambda = NULL, ic = "aic")
Error in ets(y = data$Sales.Quantity, model = "AAM", damped = FALSE, lambda = NULL,  : 
  Nonseasonal data