Latent total consumption expenditure, unobservable individual preferences and panel data

This paper applies the errors-in-variables approach on Chinese panel data to estimate a system of consumer expenditure functions. In particular, total consumption expenditure is treated as a latent variable, while purchase expenditures on different goods and total household income are used as its indicators. It is shown that the common assumption of no measurement error in total expenditure can yield substantial bias in the estimators of the Engel functions, while the standard practice of assuming independence between preferences and total expenditure is acceptable on its own.