My question is straightforward: Is there any alternative way to Box-Cox transformations to stabilize the variance of a time series?
3
-
2$\begingroup$ Surely, depending on the time series and e.g. any relationship between variance and mean. For example, a proportion in (0, 1) or [0, 1] might benefit from something quite different. $\endgroup$– Nick CoxCommented Aug 12, 2022 at 12:39
-
1$\begingroup$ stats.stackexchange.com/a/10979/919 gives an example of a different family of transformations. It also demonstrates that the choice of transformation depends on the nature of the data, as @Nick indicates. Thus, it would be difficult to offer more than a perfunctory "of course" answer to your question until you provide more details of your time series. $\endgroup$– whuber ♦Commented Aug 12, 2022 at 13:13
-
$\begingroup$ In many situations common transformations on time series (besides things like logs) include taking differences or seasonal difference. In some situations that might make sense and also do what you need. But in other circumstances you might not need to transform and instead consider models that don't assume constant variance (in some situations at least, it may be feasible to consider models where variance might relate to mean; this might preclude or modify certain kinds of analysis though). $\endgroup$– Glen_bCommented Aug 13, 2022 at 1:34
Add a comment
|