This thesis studies on the impact of appropriately modeling the fatness of left tail of the assets’ log-return distributions on the quality of scenario tree, and furthermore, on the behavior of solutions when it is applied to stochastic programming problem. To model the fat left tail, we use the Heston model. First, we introduce how to extend the single-asset Heston model to multi-asset Heston model. Moreover, we investigate the parameter calibration methods of the multi-asset Heston model. Second, we survey on the scenario tree generation methods with the focus on Moment-Matching. We also cover about the quality evaluation methods of the scenario trees. Third, we briefly introduce the formulation of Multi-Stage Stochastic Programming problem that we use in this thesis. Last, we investigate on the impact of appropriately modeling the left tail using Heston model on the solutions of stochastic programming problem, and we compare the results with the Geometric Brownian Motion model.