This paper tests two versions of the random walk model for portfolios of Brazilian stocks. It found evidence of persistency in daily and weekly returns, rejecting the random walk models. Those evidences were weaker in recent periods. The paper also found a Monday effect, and other seasonality effects for monthly returns. Additionally there were asymmetric first-order cross-correlations on portfolios ranked by size, with large firm returns predicting small firm returns. Nonlinearities in returns were also detected at several time horizons.