Existing models of life-cycle demand typically relate household welfare to the scale of total consumption expenditures, deflated by a price index. This is only correct if preferences are homothetic and expenditure shares for particular goodswithin a period are fixed, and is sharply at odds with strong empirical evidence,including Engel’s Law. Instead of using the scale of total expenditures we show howto exploit variation in the composition of a household’s consumption portfolio toestimate demand systems that are flexible and may feature highly non-linear Engelcurves. This same procedure yields an index of household welfare closely related tothe marginal utility of expenditures within a period. We use these methods withrepeated cross-sectional expenditure surveys from Uganda to estimate an incomplete demand system and household welfare in different periods, and analyze the effects of shocks such as the 2008 food price crisis on welfare in Uganda. Our results contrast sharply with those obtained using total ‘real’ expenditures, which cannot capture the differential impact of increased food prices across the expenditure distribution.