Uncertainty in saving: Is the life cycle model correctly specified?

The theoretical backdrop for this dissertation is Modigliani's life cycle hypothesis of saving. A central proposition of the life cycle model is that utility-maximizing individuals form consumption and saving plans around expected lifetime incomes. Another heretofore separate body of theory holds that such expectations are inherently uncertain, and owing to risk aversion on the part of the utility maximizers, current consumption will decrease--saving will increase--with an increase in the uncertainty of anticipated income. The research reported here is essentially an empirical investigation of the relationship between consumption, saving, and uncertainty. An uncertainty-explicit aggregate consumption function is developed through a series of steps. First, the relevant theory is cast in a life cycle framework. Then a micro-level consumption function, which includes an uncertainty variable, is derived through maximization of the individual's intertemporal utility function. Finally, the micro-level function is aggregated to obtain the model used for empirical estimation and testing. In general, each of two uncertainty variables--the variance and semivariance of aggregate income, calculated over the period of the distributed lag used in representing expected lifetime income--behaves as expected and proves statistically significant when the lags are set at fewer than five years. In the preferred specification, aggregate income is defined to include all contributions for social insurance. This definition is motivated by evidence reported in the literature that the accumulation of social security wealth displaces--possibly on a dollar-for-dollar basis--private saving and wealth accumulation. This research has two implications for economic policy. First, policy-based attempts to simultaneously increase the rate of private saving and stabilize income growth may be at cross purposes. Reducing income fluctuations may lower perceptions of the uncertainty of future income, thereby leading to higher consumption and lower saving. Second, if an uncertainty variable should be but is not included in models on which policy analyses are based, parameter estimates in those models will be biased. The magnitude and direction of such specification bias is examined here.