In this study, we use the data of the Taiwan 50 Constituents to investigate the relationship between the rebalance of portfolios and their performances. We use the equally weighted portfolios as well as the portfolios, which returns, variances and/or correlations are predicted by the AR, GARCH and Exponential Smoothing models, respectively. There are altogether nine kinds of portfolio in our study. The optimal portfolios would be built by solving the maximum problESM of the Sharpe ratio. The data used in this study are the daily data of the Taiwan 50 Constituents and spans from January 2001 to December 2010. Our result shows that under consideration of transaction cost, the portfolio, which is constructed by the AR model and quarterly rebalanced, has the best performance. Furthermore, it also outperforms the same type of stock index funds and the market.index as well.