For the investment decision problem with uncertain exit time, this paper establishes a return maximization portfolio selection model that measures market risk with the period value at risk(PVaR). Due to the computational complexity of the PVaR, it is difficult to solve the model with conventional optimization methods. This paper proposes and proves that the optimal solution of the original model can be found by solving an equivalent mixed integer programming model. The numerical experiment analysis is carried out by using the actual stock price data. The results show that solving the mixed integer programming model can quickly find the optimal investment decision for small and mid-scale short-term investment problems.