Non-spherical errors are related to

Solution

If the spherical errors assumption fails to hold, there are two problems with OLS. There is a better estimator available—one that is still unbiased, but is less sensitive to sampling variation (i.e., has lower standard errors). The formula we use to estimate the standard errors of OLS is invalid. Our confidence intervals and hypothesis tests will (usually) be overconfident, overstating the precision of our results. Spherical errors fails when we have either or both of: Heteroskedasticity and Autocorrelation