The paper takes the maximum income of each asset portfolio of banks as the objective function and the risk loss unit income bears and the VaR as restrains and applies Backward Induction Theory and non-linear programming to setting up Multi-stage Dynamic Optimal Model of Loan Portfolio for Commercial Banks based on Default Loss Control. It studies the Downside-risk per profits of bank can bear in this period with applying Backward Induction Method to optimize the next period asset portfolios. This reflects the influence on the whole loan allocation of Downside-risk per profits of bank can bear. It optimizes the whole period loan portfolios by considering the single period loan portfolio's optimal and use Downside-risk per profits to control default loss of portfolio. This changes the phenomenon of popular research taking the variances as the risk and better depicts portfolio risk.