A mesoscale tropical channel model is constructed to study the long-standing problem of the initiation of the Madden-Julian Oscillation (MJO). Two observed cases of MJO are chosen, one in boreal spring and one in boreal winter, without a priori knowledge of their initiation mechanism. With initial and lateral boundary conditions provided by a global reanalysis, this model is able to reproduce the initiation and gross features of two observed MJO events up to two months after the start of simulations. This leads to a conjecture that these two MJO events are generated by the influences from the lateral boundaries. This conjecture is supported by a series of sensitivity tests. These sensitivity tests demonstrate that the simulated MJO initiation does not critically depend on detailed characteristics of sea surface temperature (varying vs. constant in time, mean distribution from boreal spring vs. winter), initial conditions (within a 10 day period, perturbations in the initial conditions), the latitudinal location of the lateral boundaries (21 - 45˚N and S), and even latent heating and moist processes. The only factor found critical to the reproduction of the MJO initiation is time varying lateral boundary conditions from the reanalysis. When such lateral boundary conditions are replaced by time independent conditions, the model fails to reproduce the MJO initiation. The analysis of moist static energy has revealed that the discharge-recharge mechanism is not sufficient for the MJO initiation in the model. It is found that the latitudinal transport of westerly momentum from the extratropics is crucial in generating the lower tropospheric westerlies in the reanalysis and model. The energy source for the extratropical perturbation is through extraction of kinetic energy from the mean flow. The estimation of wave activity flux has revealed that there is a major region over the southern Indian Ocean, which produces wave activity flux towards the tropical genesis region of the MJO initiation. We have also investigated the time-scale of the boundary conditions that are responsible for the MJO initiation. Additions of small perturbations in the boundary conditions, and use of 10-day interpolated boundary conditions do not affect the MJO initiation. Thus boundary conditions responsible for the MJO initiation in the model must have time scales greater than 10 days, indicating that the shorter time scale stochastic forcing through the lateral boundaries did not play any major role. The estimation of the zonal momentum budget for the MJO initiation region has revealed the importance of the advective terms, particularly by the meridional winds before the onset of the MJO. The importance of the extratropical influences in initiating the MJO in the channel model leads to a speculation that a multi-year simulation using a tropical channel model would also produce reasonable MJO statistics if forced by time varying boundary conditions. Interestingly, the MJO statistics in the multi-year simulation using a tropical channel model is found to be not better than the global models. Increase of horizontal resolution and use of a different cumulus scheme did not improve the MJO simulation. We found that the error in the mean state was the main reason for the lack of MJO statistics in the model. The model took less than five days for the error to reach its climate bias. Thus, a good simulation of the mean state is important for the successful simulation of the MJO. Implications of these results are discussed. In short, this study has shown that the extratropical influences can be an efficient mechanism for the MJO initiation and calls for further research attention to this mechanism that has been somewhat neglected by mainstream MJO research.