Time series assumptions


Fortran must die
Most of the evidence I have seen is that time series authors don't worry a lot about regression assumptions at least if the goal is to predict. Vinux made this point to me, but other authors such as Milhoj and Chatfield have as well. However, this is not true when you are using time series to show relationships between variables. What some refer to misspecification tests are very important. This includes the chow test and breusch pagan for AC. But I am not sure about what time series specification tests exist for equal error variance and normality. Can you use a qq plot with time series residuals. I don't know a good test for equal error variance beyond inspecting the residuals, I don't know if you can use this for time series.

After I read this I ran into the Breusch Pagan Godfrey test for hetero which apparently works for time series. I am not sure if this is the same Breusch Godfrey test used for Autocorrelation (ac) or not.