sm.stats.acorr_ljungbox(res.wresid) needs helpful return

[josef-pkt]

sm.stats.acorr_ljungbox returns the values for all lags, this can be difficult to interpret.

A recommended answer would be easier to interpret. What's the recommended lag to look at?

Also acorr_ljungbox or similar should be added to summary() as more powerful alternative to Durbin-Watson, (DW doesn't have p-value)

[jseabold]

Box and Jenkins recommends that the lag be long enough so that the 'weights' go to zero. I think where

w_t = \phi^{-1}(B)\theta(B)a_t=\psi(B)a_t

In any event, in the examples I've seen the Q-stat is either never significant blows up at higher lag-lengths so it's obvious there are problems. I think stata's default is 40 as with the other corrgram type functions.

[jseabold]

Not a release blocker.

[josef-pkt]

http://stats.stackexchange.com/questions/200267/interpreting-ljung-box-test-results-from-statsmodels-stats-diagnostic-acorr-lju/200283#200283

some links from the answer

"
Hyndman http://robjhyndman.com/hyndsight/ljung-box-test/ recommends min(10, T/5) for non-seasonal time series. His simulations show that ljung-box overrejects if the lag length is too large.

Stata uses min(n/2 − 2, 40) http://www.stata.com/manuals13/tswntestq.pdf

It looks like in R the default number of lags depends on which package is used.

This http://www.r-bloggers.com/story-of-the-ljung-box-blues-progress-not-perfection/ shows a similar pattern of p-values decreasing with lag length.
"

[josef-pkt]

docstring is wrong: maxlag is hardcoded at 41, but docstring has copy paste error
"If lags is None, then the default maxlag is 12*(nobs/100)^{1/4}"