Professional forecasts of aggregate U.S. consumption series strongly reject Robert E. Hall's (1978) random walk hypothesis. Band spectrum regressions show that low-frequency variations in growth ra tes of expenditures on nondurables and services, defined as cycles takin g more than two years to complete, primarily account for the rejection. Consumption growth and professional forecasts of GNP growth are also closely related at the low but not at the high frequencies. Liquidi ty constraints or durable characteristics of consumption goods may both explain the reported findings. Copyright 1992 by MIT Press.