In this paper, we introduce a new algorithm for vector forecasting in the singular spectrum analysis. This algorithm enables SSA users to use two different values for the window length parameter: one for reconstruction and another for forecasting. Our results on both real and simulated data support the idea of using this new algorithm.