# Selection of forecasting method - Winter/ARIMA/TBATS in R

#### TheMonkster

##### New Member
Please find my dataset and forecast outputs attached.

A) First sheet contains March-2011 to February 2014 data and forecast for March-2014 to February 2015 using ARIMA,Winter's,TBATS and BATS method.It also has forecast errors obtained by comparing with actual output.
B) Second sheet has forecast for June 2015 to February 2016 using above mentioned methods.
C) R code.

As it can be seen, TBATS method gave output for 2014-15 with the least error but there is no trend and seasonality (Constant values) in TBATS output for 2015-16 which is hard to believe.
BATS method gave the most erroneous output (Constant values) for 2014-15 but forecast for 2015-16 seems reasonable.

I am confused which method should I go for.Should I opt for some other technique considering my data? Or Am I missing something?

How should I select forecasting method when forecast outputs are too close to take a call?

#### vinux

##### Dark Knight
You could also try ets.
Code:
ets((valuets))
I haven't run your Rcode. So I haven't checked the diganostic part of the residuals.
You have to be careful on specifying trend and seasonality in the model. By looking at the graph, I am not able to judge about the presence of linear trend and seasonality (it may be there; for me it was not obvious). Usually the <out of sample> as well as the <in sample> deviation measures (MAPE, MAE, ...) are consider for finalizing the model.

It two-three models are giving best forecasts, you can choose the simple model (which require lesser assumptions) as the final one.

#### TheMonkster

##### New Member
You could also try ets.
Code:
ets((valuets))
I haven't run your Rcode. So I haven't checked the diganostic part of the residuals.
You have to be careful on specifying trend and seasonality in the model. By looking at the graph, I am not able to judge about the presence of linear trend and seasonality (it may be there; for me it was not obvious). Usually the <out of sample> as well as the <in sample> deviation measures (MAPE, MAE, ...) are consider for finalizing the model.

It two-three models are giving best forecasts, you can choose the simple model (which require lesser assumptions) as the final one.
TBATS gives the best forecast for year 2014-15 considering the least MAE & MPE but TBATS forecast for 2015-16 doesn't show any seasonality or trend.It just gives constant value for 8 months,which is very unlikely in reality.so my question is,should I still use this forecast?

P.S. I have forecasted using ets and got constant valuess for all months for both 2014-15 and 2015-16.The same is updated in attached file.

Last edited:

#### vinux

##### Dark Knight
TBATS gives the best forecast for year 2014-15 considering the least MAE & MPE but TBATS forecast for 2015-16 doesn't show any seasonality or trend.It just gives constant value for 8 months,which is very unlikely in reality.so my question is,should I still use this forecast?
Yes. When the relationship is not strong, the stationary series forecast may become constant after few lags (the forecast values will be or converge to the global mean).