This paper investigates whether portfolio return autocorrelation can be explained by time-varying expected returns, nontrading, state limit orders, market maker inventory policy, or transaction costs. Evidence is consistent with the hypothesis that transaction costs cause portfolio autocorrelation by slowing price adjustment. I develop a transaction-cost model which predicts that prices adjust faster when changes in valuation are large in relation to the bid-ask spread. Cross-sectional tests support this prediction, but time-series tests do not.