Hello I am trying to forecast using different exponential smoothing methods(Linear and Winter's). For the optimal parameters, I am getting negative values of the forecasats. I am assuming it means that the values will be zero, since it is a sales forecast. I wanted to know if negative values denote something wrong with the model or is it possible to have negative values in forecast. Since it is just a model, I think we can get negative values for certain type of time series.
$\begingroup$
$\endgroup$
2
-
$\begingroup$ Related: stats.stackexchange.com/q/92443/27433 $\endgroup$– Waldir LeoncioCommented Aug 14, 2015 at 19:58
-
$\begingroup$ Also related: stats.stackexchange.com/q/107467/27433 $\endgroup$– Waldir LeoncioCommented Aug 14, 2015 at 20:04
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
1
Holt's or Winter-Holt's exponential smoothing methods can give negative values for purely non-negative input values because of the trend factor which acts as a kind of inertia, which can drive the time series below zero. Normal exponential smoothing doesn't have this problem, it's always smoothing inwards, it never overshoots.
-
$\begingroup$ However, the prediction intervals can get negative. How can one prevent it? $\endgroup$ Commented Aug 14, 2015 at 19:59
$\begingroup$
$\endgroup$
You can get negative values for certain kinds of models. You might want to explore more complicated models than simple exponential smoothing.
