ETFs (Exchange-Traded Funds) have about 400 stocks listed on the financial market in Korea. They are notable for their low transaction costs and high liquidity. However, studies only focus on the arbitrage strategy using the price gap between market price and net asset value of ETF, or the analysis between the ETF market and the underlying asset market. This study proposes a strategy for ETF portfolios based on underlying assets. The weight of each asset group is determined using a genetic algorithm with the Sharpe ratio of the portfolio. The performance of the portfolio is evaluated through the cumulative return and the aforementioned ratio during the test period; the experiment is conducted using the sliding window technique. In conclusion, we show that the cumulative return and Sharpe ratio of a portfolio can be improved through segmentation of the asset groups included therein.