Optimal investment decision is an investor's rational allocation of risky assets in a complex and volatile environment from a long-term perspective in order to obtain the maximum desired utility. In this paper, the investment decision-making problem with uncertain rate of return in multi-stage investment is studied. Firstly, the ARMA-GARCH model is established to forecast the future rate of return of assets based on historical data of risky assets, Monte Carlo simulation method is used to simulate the possible future situation of yield, and random sampling is applied to build scenario trees. Secondly, based on the scenario tree, the mean-variance model proposed by Markowitz is extended to multiple periods according to stochastic programming theory. Finally, the data of six stocks in the Chinese securities market is selected to empirically investigate the model. The results of the study find that scenario tree is effective in describing the uncertainty problem. The model is suitable for multi-period investment and can provide radical, stable and conservative investors with intuitional and clear investment decision guidance.