F test is unreliable under presence of Heteroskedasticity… for which other violations ??
Serial Correlation (both cause smaller standard errors)
i agree on smaller std errors but the F test funda ?? do u hav d logic for it thanks !!!
like whats the logic for F test to eb too high when there is positive serial correlation ?
F-Test is Mean Square Regression (MSR) divided by the Mean Square Errors (MSE). With positive serial correlation, you will have smaller standard errors like we agreed. These smaller errors mean that we will have a smaller Sum of Squared Errors which will equate to a smaller Mean Square Error. This will lead to an inflated F-Statistic and we will assume that at least one of the independent variables is useful when in fact none of them may be useful. Anyone correct me if something is wrong.
woww how come i dun think lik tht !! is this in the cfai books … thx this sure makes it clear !!
F(stat) = MSSR/MSSE MSSE = SSE/ (n-k-1) errors small -> SSE small -> MSSE small -> F(stat) -> if (F(stat) > F(crit)) then we reject the NULL that the model is bullshi*t. as F(stat) is overstated, chances of rejecting NULL is HIGH. Chances of using a bullhi*t model is high
I am sure its in the CFAI text, couldn’t tell you a page number though.