Holt-Winters forecast

[Manuel Pruijssers]

Hi everybody,

I have made a correllogram over the production volume for a company and thereafter I made a holt-winters forecast. The correllogram showed there was a significant trend, however the holt-winters forecast showed a beta of 0, which indicates no trend has been used to compute the forecast. How is this possible?

Sincerely,

Manuel

[Marcos Almeida]

Hello Manuel,

Welcome to the Stata Forum.

I believe the odds of an insightful reply would increase significantly, provided you share commands and results, as recommended in the FAQ.

That said, the correlogram (partial?, total? ) presents autocorrelation, and forecast may be related to other aspects of the model.

Indeed, according to the Stata Manual, there are important issues related to modeling under Holt-Winters smoothers, such as

When the Holt–Winters model fits the data well, finding the optimal smoothing parameters
generally proceeds well. When the model fits poorly, finding the α and β that minimize the in-sample
sum-of-squared forecast errors can be difficult.

In short, maybe a better description of your data would be helpful.