Don't you guys ever test the obvious features :)?

[startz]

If you estimate with
ls nrnd c
and then choose serial correlation LM test you get an error alert "Matrix size mismatch."

[EViews Glenn]

Note that any use of nrnd as the dependent variable has problems when you perform tests since the data aren't stable.

That said, what's happened is that your clever little escapade has triggered two errors -- one related to the use of the random number generator as the dependent variable and the other related to the fact that when the first error fails, a matrix doesn't get sized to a specific dimension. Long story short, the second error message is masking the first. Will be fixed in an update.