Consumption and saving behavior.

Zeldes, Stephen P.

Consumption and Saving Behavior

To understand business cycle fluctuations, the effects of government deficits, long-run aggregate capital accumulation, and the determination of asset prices, we need to know how households allocate their income between consumption and saving. Much of my research focuses on the effects on consumer spending and saving of imperfections in the consumer credit market that constrain the amount of household borrowing, uncertainty in household income, and changes in the value of the stock market.

The Importance of Borrowing Constraints

The permanent-income theory of consumption says that the response of consumption to temporary changes in income should be small: households will save the bulk of positive windfalls to income and will run down their assets or borrow to keep consumption smooth in the face of drops in income. This view assumes that households have access to markets in which they can borrow against their future labor income. Yet aggregate U.S. consumption fluctuates more than the permanent-income hypothesis predicts.

Credit markets are imperfect because banks do not have all possible information about loan applicants and cannot enforce loan repayments perfectly. Therefore, individuals who suffer a substantial drop in earnings cannot borrow large amounts of money with only their future labor earnings as collateral. I derive the theoretical implication of such borrowing constraints and test whether these limits on household borrowing have important effects on spending patterns.(1)

To perform the tests, I use survey data from the Panel Study of Income Dynamics (PSID) on approximately 5000 households followed over 15 years. These data include a detailed breakdown of different types of income, tax information, and information on a component of consumer spending.(2) I split the sample into two groups: those families with few or no liquid assets in a given year, and those with higher liquid assets. I derive and test two implications of the model with borrowing constraints. First, once I take variations in aftertax interest rates into account, changes in consumption should be forecastable for the low-asset group but not for the high-asset group. The second implication begins with the observation that households that are unable to borrow still have the option of saving to smooth out high current income or low anticipated future income. Because of this, households with low current assets that are unable to borrow should have a higher expected growth of consumption than the rest of the population.(3) My results generally support the view that borrowing constraints affect the spending patterns of a significant fraction of the U.S. population. Consumption growth is both more predictable and, on average, higher for households that find themselves with few or no liquid assets in a given year.

Precautionary Saving: The Effects of Income Uncertainty

I also examine the effects on saving of uncertainty about future income--including the risk of becoming unemployed or disabled, or experiencing large increases or decreases in salaries.(4) Previous researchers had shown the conditions under which uncertainty raised the level of saving, but they were unable to determine the magnitude of the effect.(5) I demonstrate that uncertainty generally will raise the sensitivity of consumption to current income. I calculate what the consumption function would be, given empirical estimates of the magnitude of income uncertainty and plausible assumptions about consumers' preferences. Therefore, I am able to calculate the exact size of the effect of income uncertainty on both the level of consumption and the sensitivity of consumption to current income, Four important results have emerged from this line of research.

1. Precautionary saving is likely to be an important component of household saving. This is especially true for households whose lifetime resources consist primarily of uncertain future labor income. Thus, one reason for the secular decline in the U.S. saving rate may have been the rise of social insurance programs.(6) In addition, uncertainty about uninsured medical expenses is likely to explain why elderly households spend so little relative to their assets.(7) Currently, I am examining the combined effects of uncertainty about income, length of life, and uninsured health expenses.(8)

2. Income uncertainty generally increases the sensitivity of consumption to transitory changes in income or wealth.(9) Previous research, which ignored the effects of income uncertainty, found it puzzling that consumers would respond so significantly to temporary changes in income.(10) I discovered that uncertainty can raise the sensitivity of consumption to income and that the degree of "excess sensitivity" in the data is of approximately the same magnitude expected from the amount of income uncertainty facing households.(11)

3. Household income uncertainty tends to lower the equilibrium interest rate on short-term government ("risk-free") bonds. Standard theories of asset pricing cannot explain the low level of the observed risk-free rate of interest. These standard theories often imply that high aggregate rates of growth will be associated with high rates of return. In the United States, however, there have been long periods in which consumer spending has grown rapidly despite low or negative real risk-free rates. Asset pricing theories also imply that uncertainty about future income increases the desired growth rate of consumption. I find that the amount of income uncertainty facing the typical household will substantially reduce the equilibrium risk-free interest rate in the economy, thus helping to resolve this "risk-free rate puzzle."

4. Deficit-financed tax cuts matter. Some argue that tax cuts financed by deficits don't stimulate consumer spending because consumers realize that their taxes will be raised in the future.(12) In work with Robert B. Barsky and N. Gregory Mankiw, I explain why tax cuts should raise consumer spending, even if individuals fully anticipate higher taxes later on.(13) Since tax revenues increase with income, individuals realize that they or their children will have to pay most of the future taxes only if they have a high future income; if they are less successful financially, the tax burden will fall on others. Thus, there is little incentive to save the money they receive today and consumer spending rises. Therefore, deficit-financed tax cuts are likely to result in higher consumer spending and lower national saving.

The Stock Market and Consumer Spending

Finance theory stresses the positive relationship between the risk of a particular asset and its expected return. A great deal of research has attempted to define and measure the "riskiness" of an asset or portfolio of assets. One measure of the riskiness of a particular stock or bond compares the covariance of its return with the growth in aggregate consumption. Under this measure, a stock is more risky if its return tends to be high when aggregate consumption is high and low when aggregate consumption is low. This "consumption beta" measure has strong theoretical underpinnings, but empirical tests have provided little support for the theory.

One possible problem with the tests performed thus far is that they assume that all individuals have easy and inexpensive access to the equity markets, and therefore that all individuals hold stock. The measure of consumption used in tests of the theory is therefore aggregate spending of the U.S. population. Yet nonstockholders comprise an important part of all consumers. If consumption patterns differ between stockholders and nonstockholders, estimating a model with aggregate time-series data is likely to give misleading results.

In recent work with Mankiw, I use data from the PSID to examine such differences.(14) We use 17 years of extensive data on a representative sample of approximately 5000 families, including data on the size and allocation of each family's wealth. We find that:

1. Only a small fraction of the population holds stock, either directly or through defined-contribution pension plans. In 1984, only 28 percent of U.S. households held any of their wealth directly in the stock market and only a small number held defined-contribution pension wealth in the stock market. The consumption of nonstockholders comprises a significant fraction of aggregate consumption.

2. The prevalence of stock ownership is strongly (positively) related to both the labor earnings and the education level of the household.

3. The consumption of stockholders differs from that of nonstockholders. This suggests that disaggregating the data could improve the performance of a variety of consumption-based asset pricing models.

4. The growth of consumption of stockholders is slightly more volatile and has a significantly higher correlation with the stock market than that of nonstockholders. The riskiness of the stock market cannot explain the large difference between the average return on U.S. stocks and Treasury bills (the "equity premium"). Intuitively, if the random movements in stock returns are not associated with large changes in consumption, then the randomness does not represent true riskiness to the consumer and therefore should not require a very large risk premium. Our results show that the covariance of consumption growth with the excess return on the market is four to seven times higher for stockholders than for nonstockholders. Thus, separately examining the consumption of stockholders and nonstockholders helps explain this equity premium puzzle.

Future Research

My research has raised a number of questions that merit additional investigation. First, I am exploring how the interaction of three important sources of uncertainty (earnings, health expenses, length of life) influences aggregate saving. Second, now that theoretical models have been developed showing the potential importance of precautionary saving, econometric techniques need to be used to estimate the size of precautionary saving in household and aggregate data. Third, further research is needed to explain why certain wealthy households do not own stock in their portfolios. Finally, I am trying to figure out ways of combining the microdata with aggregate data to create a proxy for the consumption of stockholders, so that the relationship between stockholder consumption and the return on the market can be examined using a longer historical time series. (1)S. P. Zeldes, "Consumption and Liquidity Constraints: An Empirical Investigation," Journal of Political Economy 97, 2 (April 1989), and "The Effects of Borrowing Constraints on Consumption in a Multiperiod Model with Income Uncertainty," manuscript, 1987. (2)The survey asks for the current value of consumption expenditures on food at home and food away from home. (3)Technically, this observation corresponds to a one-sided inequality in the consumer's intertemporal first-order condition. (4)S. P. Zeldes, "Optimal Consumption with Stochastic Income: Deviations from Certainty Equivalence," MIT Ph.D. dissertation, 1984, and Quarterly Journal of Economics (May 1989). (5)These authors (for example, H. E. Leland, "Saving and Uncertainty: The Precautionary Demand for Saving," Quarterly Journal of Economics 86 [1968], and A. Sandmo, "The Effect of Uncertainty on Saving Decisions," Review of Economic Studies 37 [1970]) showed that if the third derivative of the utility function is positive, increased income uncertainty will raise saving. However, they were unable to derive a closed-form analytic solution, and thus the magnitude of this effect was not known. Rather than use analytic techniques, I use numerical methods to closely approximate the solution. (6)This by no means implies that the programs were not welfare improving. One reasonable interpretation is that the drop in saving, while optimal from each family's standpoint, was a negative side effect of these programs from the point of view of the aggregate economy. (7)See also L. J. Kotlikoff, "Health Expenditures and Precautionary Savings," NBER Working Paper No. 2008, August 1986, and in What Determines Savings? Cambridge, MA: MIT Press, 1989. (8)I am exploring this in current research with R. Glenn Hubbard and Jonathan S. Skinner. (9)This presents an alternative to the borrowing-constraints explanation of "excess sensitivity" discussed in the previous section. (10)R. E. Hall and F. S. Mishkin, "The Sensitivity of Consumption to Transitory Income: Estimates from Panel Data on Households," Econometrica 50 (1982). (11)I found this "excess sensitivity" result for constant relative risk aversion utility functions. In work subsequent to my original finding, Roell ("Capital Market Imperfections and the Excess Sensitivity of Consumption to Transitory Income," manuscript, 1986) and Kimball ("Precautionary Saving and the Marginal Propensity to Consume," manuscript, 1988) have verified this excess sensitivity result analytically and have shown which properties of the utility function lead to this effect. However, they are unable to calculate the magnitude of the effect. The numerical results discussed in the text calculate both the direction and the magnitude of this effect. (12)R. J. Barro, "Are Government Bonds Net Wealth?" Journal of Political Economy (1974), and "The Ricardian Approach to Budget Deficits," Journal of Economic Perspectives (1989). (13)R. Barsky, N. G. Mankiw, and S. P. Zeldes, "Ricardian Consumers with Keynesian Propensities," NBER Reprint No. 821, February 1987, and American Economic Review 76, 4 (September 1986). (14)N. G. Mankiw and S. P. Zeldes, "The Consumption of Stockholders and Nonstockholders," NBER Working Paper, forthcoming 1989.