Permanent Household Income and Consumption in Urban South America
提出一种回归模型，将家庭特征对消费的直接和间接影响分离，从而估计永久收入消费倾向，并区分估计误差与暂时性收入误差，解决了永久收入假说检验中的关键难题。
Milton Friedman's permanent income hypothesis (PIH) is considerably more complex and difficult to test than it appears at first sight, which is why certain of its postulates still cannot be accepted or rejected with certainty. One of the chief difficulties is that of separating the effects on consumption of permanent income, which is unobservable, and of various observable household characteristics which may be highly correlated with permanent income but may also be presumed to affect consumption directly. This difficulty can be overcome, provided a variable can be found which is associated with permanent income, but has no independent direct effect on the propensity to consume: if households are grouped by classes of that variable, the permanent consumption propensity or elasticity can be estimated from mean income and consumption in the groups. This technique requires the assumption of zero mean transitory income in each group, but does not depend on the postulate that transitory income is uncorrelated with consumption. For these reasons it cannot be applied to individual households. This paper presents and estimates a model which separates the direct and indirect effects on consumption of a series of observable characteristics, using regression techniques on individual household observations. This makes it possible to estimate all the parameters of the model, both the consumption propensity or elasticity and the contribution to permanent income of each observable explanatory variable. The technique still requires that at least one of the explanatory variables affect consumption only through permanent income, but for all other variables both direct and indirect influences can be estimated, and the propensity to consume a given income can vary among households with different characteristics. Another principal problem with the PIH is the assumption that transitory income and consumption are uncorrelated. Efforts to test this postulate can lead to very different results, ranging from agreement with the hypothesis to findings of high propensities to consume transitory income (see Robert Ferber, pp. 1306-09). This inconsistency, and continued controversy over the hypothesis, is due to the fact that permanent income cannot be estimated exactly, but can be known only with an which is conceptually distinct from transitory income. Empirical work on the PIH has uniformly overlooked this fact, treating estimates of permanent income as exact and regarding the whole of the unexplained residual as transitory income. The present model explicitly distinguishes the estimation from the error of transitory income or consumption. The two transitory components can then be assumed to be uncorrelated, while the part of permanent income which is not explained is consumed exactly like the explained part. Whether residual income and consumption appear to be correlated or not then depends on the relative sizes of these two components, the unexplained permanent part and the true transitory part. This technique has the further advantage that estimating the variance of the of estimation makes it possible to tell how well a set of observable variables explain permanentincome. This explanation may be better than *The Brookings Institution. Robert Solow, Richard Eckaus, and Robert Hall supervised the dissertation (1974) on which this article is based. I am particularly grateful to Ross Williams and Robert Hall for help in specifying and estimating the model, and to Denisard Alves for a crucial suggestion. William Cline, Irwin Friend, Ricardo Ffrench-Davis, and Robert Ferber provided helpful comments on earlier versions. Subsequent discussions with Surjit Bhalla and Robert Lawrence have been most useful. I am also much indebted to the managing editor and to a referee for improvements in the final version.