Considering the behavior of the stock market is nonstationary and thus earlier observations are less relevant to the current investment decision-making, we design online portfolio strategies only based on recent stock price data. Firstly, we design an online portfolio strategy which is the weighted average of the previous portfolio and the best constant rebalanced portfolio corresponding to recent stock price data of fixed length. Secondly, we design an adaptive online portfolio strategy by choosing the weights via online learning. We present numerical analysis by using real stock data, and the results illustrate that our strategies perform well, compared with benchmark strategies and existing online portfolio strategies.