I just read Demand-Driven Forecasting: A Structured Approach to Forecasting(Wiley and SAS Business Series) and have a few doubts in Holt-Winters Model:
1) Unlike OLS Regression Modeling technique or ARIMA, no assumptions were checked before running Holt-Winters. For instance, in ARIMA, we first make the data stationary before running ARIMA or in OLS, we check normality, auto-correlation etc. However, as per the book, no test was conducted before and after running Holt-Winters. We just calculate MAPE and check if it is acceptable. So, can someone confirm if there are any tests that we should do before and after running Holt-Winters?
2) When we are running Holt-Winters multiplicative or additive model, we don't need to explicitly adjust for seasonality before?
Please note that I am calling Holt-Winters function in R directly without doing any checks or adjusting for seasonality
Best Answer
Regarding your first question, Holt Winters is simply a smoothing calculation. It's analogous to taking the average of data except that you are weighting each data point differently in this case. You would want perform some type of holdout analysis to determine whether a Holt Winters model outperforms other models available to you; MAPE would be one metric you can compare models with. So in terms of tests that should be performed prior to its use, you should probably just confirm there is no seasonality in your data.
Regarding your second question, Holt Winters will do a poor job extrapolating seasonal data. There are single season Holt Winters models and double season Holt Winters models already available in R. You can read more about those at the following links:
Holt Winters Seasonal Model
Holt Winters Double Seasonal Model