The hypothesis that consumption evolves over time as a martingale process is tested on household panel data for three villages in south India. A novel feature of the methodology is that it gives consistent estimates of dynamic effects in short panels. The estimated coefficients of lagged consumption are generally smaller than unity and a number of the lagged income and wealth variables are statistically significant. The results are inconsistent with the proposition that consumption equals permanent income. This is also true when the data are disaggregated by household wealth.