This paper is about the determination and prediction of permanent income in household data. Standard static welfare indicators (e.g. per capita expenditure and income) are imperfect in this respect as they typically contain a high transitory component. The framework we employ is consistent with the permanent income hypothesis but is supplemented with a causes equation where unobservable permanent income is explicitly modelled as function of causal variables which play a key part in its determination. Simultaneous estimation of the model allows us to compare how well different standard static welfare indicators identify permanent income but more importantly enables us to predict permanent income using information contained both in the causal variables and in the standard static welfare indicators. The paper is closed by an pplication of the methodology to household data from the rual sector of two Chinese provinces.