We derive a closed-form solution to a continuous-time optimal portfolio selection problem with return predictability and transaction costs. Specifically, we assume that asset returns are predicted by stochastic signals, and that transaction costs are of quadratic form. The agent chooses a trading strategy to maximize the expected exponential utility of his terminal wealth. Our feedback trading strategy indicates that the agent should trade gradually toward a dynamic aim portfolio, which is a weighted sum of the expected future Merton portfolios. The agent's aim portfolio converges to the Merton portfolio as time approaches the terminal date. Our analysis offers new insights to the existing literature. First, our optimal trading strategy is affected by the volatility of return-predicting factors, while such an effect is absent in Gârleanu and Pedersen (2016). Secondly, the agent invests more into the assets with more persistent signals and with less transaction costs.