In StatsModels there are keywords to specify how to deals with NaNs.
'none' := no checks for missing values
'drop' := drops missing values
'raise' := raises an error
But there is another keyword: 'conservative'. I don't understand what this does.
Here is the documentation, but it is not explained what is does. I'm especcially interested in what is means in the contect of Autocorrelation.