Dear Everyone,
The document of `tsdiag? says
> These tests are sometimes applied to the residuals from an ARMA(p, q) fit,
in which case the references suggest a better approximation to the
null-hypothesis distribution is obtained by setting fitdf = p+q, provided of
course that lag > fitdf.
This implies that we should pass the `fitdf' parameter when applying
`Box.test' to residuals. When checking the source code for
`getS3method("tsdiag", "Arima?)?, however, I noticed that it
didn?t pass this parameter. The relevant line is `for (i in 1L:nlag) pval[i]
<- Box.test(rs, i, type = "Ljung-Box")$p.value?. I suppose this
could be fixed by passing an additional `fitdf = sum(object$arma[1:4])?.
See also this thread on Cross Validated:
https://stats.stackexchange.com/questions/411432/significance-of-the-parameter-fitdf-in-box-test