The small sample bias of the 2SLS coefficient estimator for general dynamic simultaneous equation models with innovation errors, lagged-dependent and strongly-exogenous explanatory variables is approximated through large sample asymptotic. Results for the reduced form and structural form of general DSEM are obtained. Structural form large-T approximations bias are then used to construct corrected estimators for the parameters of interest in high order DSEM. Alternatively, the numerical bootstrap method results suggest that the non-parametric bootstrap could be used in 2SLS for improving estimation. Some simulation results are present for the Quenouille jackknife and the bootstrap.