ABSTRACT

To what extent do households make their consumption-savings decisions in a forward looking manner? While this question is of fundamental importance to many aspects of economics, a wide range of viewpoints prevail. The massive empirical literature studying consumer behavior with Euler equations has produced mixed results (see Martin Browning and Annamaria Lusardi (1996) for a survey), with a rough consensus emerging that consumption growth responds to some types of predictable variation in income, in violation of the orthogonality tests characteristic of basic life-cycle/permanent-income (LC/PIH) models. Indeed, it has been shown that a number of empirical facts do not accord well with the implications of basic forward-looking models. But despite these misgivings, other evidence has been gathered indicating that the forward-looking model may have more than a grain of truth to it. For example, evidence dating back to Milton Friedman (1957) indicates that consumption responds more strongly to permanent income innovations than to transitory ones.

Further potential support for the forward-looking model is provided by a number of papers by John Y. Campbell and others, which document a relation in macroeconomic data between consumption, usually scaled by income, and income changes a quarter later. More recently, Lettau and Ludvigson (2001, 2002a) have shown that a consumption measure scaled by a combination of assets and income (which they call cay) is related to investment returns from a quarter to several years later. Such regressions of income or asset returns on prior scaled consumption variables may be evidence that and alter their consumption in response, as the forward-looking model predicts they should. Under this interpretation, these regressions essentially tell the econometrician what households’ know about their future income or asset returns, which is nice since households’ information about these variables, especially their own future income, is generally vastly superior to what an econometrician can discern from observables other than consumption. So, compared to other strategies where the econometrician characterizes the household income process and tests whether consumption behaves as it should, these regressions bring more information to bear in characterizing households’ forward-looking nature. Any characterization of the household income process by the econometrician will always be limited by his or her potentially much smaller information set, and in fact the econometrician’s characterizations must coincide with households’ perceptions of the income process for these tests to be valid (see Hansen et al., 1991).