Asset pricing models based on the Euler equation for consumption have not performed well empirically. One of the reasons put forward has been the lack of variability of the intertemporal marginal rate of substitution of consumption, given plausible parameter values (see, for example, Hansen and Singleton, 1982, or Kotcherlakota, 1996, for a survey).
Aggregate consumption growth, which in standard representative agent models determines the marginal rate of substitution between present and future consumption, does not exhibit enough variability to be consistent with the observed time series properties of asset prices and, in particular, with the mean and the variance of the excess return on shares over a relatively safe asset such as Treasury Bills. But the equilibrium relationship between IMRS and asset returns holds only for individuals holding complete portfolios. As more detailed micro-data on wealth and saving is made available, it is increasingly clear that the majority of individuals do not hold large stocks of financial wealth, or fully diversified portfolios. This suggests that at least part of the equity premium puzzle discussed by Mehra and Prescott (1985) could be explained by the time series properties of consumption growth for asset market participants being systematically different from those of aggregate consumption growth. this
This was a point stressed by Mankiw and Zeldes (1991). They find that a distinction between shareholders and non-shareholders is important for resolving the empirical failure of consumption-based CAPM models. But, the data they use contain information on food expenditure only. If food consumption is non-separable from the other components of consumption their estimates will be misleading. Secondly, groups of shareholders and non-shareholders are defined on the basis of share ownership in the last period of the sample. Share ownership is neither a permanent nor an exogenous state of affairs.