Forecasting with the maximum of correlated components having bivariate t-distributed errors