Micro data are used for the US over the period 1968 - 1992 to estimate time-varying specifications for the conditional variance of earnings of individual households. Specifications estimated are standard and quadratic ARCH and GARCH processes. The cross-sectional mean of the estimated time-varying uncertainty of individual households is found to have a significant impact on aggregate consumption growth implying that earnings uncertainty and precautionary saving motives matter for the aggregate economy. The estimation of a buffer stock model of consumption with time-varying earnings uncertainty provides an estimated precautionary component in aggregate consumption growth. The importance of this component is found to decrease over the sample period, a result which is in line with the existing literature.