Is there a way to compute robust standard errors for potential autocorrelation problems in probit models ?