Effects of heteroskedasticity and autocorrelation on Regression


Can anyone explain why are the standard errors too small when we have either Heteroskedasticity or autocorrelation ?

Both the CFA curriculum and Schweser notes state that they are too small (leading to large test statistic and thus to Type I error) but they don’t say why they are too small.



In autocorrelation, data tends to hang close to each other, since they are dependent on the prior data point, thus the small errors. That’s how i remember it.

As for heterosked… I think the residuals can be either too small or too big.