Real estate investment trusts (REITs) are incomebased financial assets whose yields are largely related to financial and operating conditions. Their revenue comes mainly from the rent, and more than 90% distribute dividends. Therefore, the return on investment is stable and predictable. This paper attempts to study the relationship between yield and key indicators, and then constructs a good portfolio of REITs. Empirical data are 165 listed REITs with 50 key statistics (including three kinds of indicators: enterprise valuation, financial highlights, and trading information). Stepwise regression and support vector regression are used for modeling. A strategy, the Top 10% Strategy, is proposed as an application of this study, and this approach is validated by simulation. With STEP and SVR methods, the strategy has gained approximately 20% yield, which is much higher than the average of dataset, 8%. The results show that using key statistics can significantly increase profitability of REITs investment.