kvasilopoulos/exuber

Optimal number of lags

Closed this issue · 5 comments

MislavSag commented

I am comparing exuber and psymonitor packages. In psymonitor they have IC, which control for lag lengths in ADF test. Does the exuber function radf choose optimal number of lags by default or it doesnt control the number of lags by AIC or BIC criterion?

kvasilopoulos commented

Hi @MislavSag,

No, intentionally we don't want to have this option in the packages. The recommendation in PSY(2015) p. 1058 is the following:

Overall, size is reasonably well controlled when a small fixed lag length is used in the recursive
tests. This approach is therefore recommended for empirical use of the SADF and GSADF test
procedures as well as the dating algorithms that are implemented in the application later in the
article.

So, we don't want to add more complexity if there isn't a reason to do so!

MislavSag commented

So, it is recommended to use 'small fixed lag'. Should I try lags 1 to 5 and check which works best (best in sense of identifying bubbles/crises)? I suppose it depends on the frequency of the data.

kvasilopoulos commented

Sure, if you need to. I would choose lag = 1 for most of the cases though.

MislavSag commented

Thanks on info!

kvasilopoulos commented

You are welcome!