Using the fuzzy AR model, of which the coefficient is symmetry fuzzy triangular number to forecast the investment returns, and the covariance of the middle value of the fuzzy return to measure the investment risk, we expand the singleperiod meanvariance model of Markowitz into multiperiod fuzzy investment programming on condition that the liquidity is introduced as the constraint. Translating the forecasting return into the clear investment return by the defuzzifier formula at the end of each phase, we establish a twophase chance programming in order to look for the Pareto solution of the binaryobjective programming, and resolve this programming with the particle swarm algorithms which includes the factor of genetic cross and mutation. To choose 32 stocks of the SSE 50 to make the empirical analysis, and use of the discrete multiperiod meanvariance model to prove the validity of the establishing model, the results show that: the investment return obtained according to the optimal portfolio of the twophase chance programming is in the permission interval of the fuzzy return value, and its forecasting error is less; the risk of the twophase chance programming is much lower than the risk of the multiperiod MV model on condition than the constraint is the same investment return.