I've also tried GARCH(1,2) and GARCH(2,1) but both of the residuals are subject to arch after testing by Q_LB statistic. How should I do to handle it in Stata? Great thank to any help!