Open mbillingr opened 8 years ago
We should also delineate the different tests, e.g. Ljung-Box and Box-Pierce test for residual autocorrelation, whereas Li-McLeod tests for residual heteroskedasticity (also according to the same source). We need to document what each test does, and we could discuss if refactoring the tests into separate functions makes sense.
A related post on Cross Validated states that the Ljung-Box test is biased and the Breusch-Godfrey test is better. Maybe this is worth implementing as well. But again, we need to check if this is true before starting to implement.
We need to verify our implementation and application of whiteness tests.
This "source" claims that Li-McLeod is similar to Ljung-Box on squared residuals. This is not evident in our implementation.
Is our implementation correct? Do we have a reference for it? Are there other sources we can check?