Social Security and Household Saving in the United States: A Re-Examination

This paper examines household saving and consumption behavior in the United States in the light of recent controversies about the impact of the social security system on the consumption/ saving decision. It begins with a brief examination of the theory, then reviews the existing empirical evidence on the impact of the social security system on capital formation in the United States, and conducts some new tests. The conclusions from the empirical work are twofold: (1) When a possible specification error is corrected, the results appear to favor somewhat the existence of a significant depressing effect of social security wealth on household saving in the United States. Without this correction, there is no such evidence. Nevertheless, the results are far from being conclusive, and major problems remain. (2) There is also some evidence that social security depresses household saving by redistributing income toward those income and age groups that have higher than average marginal propensities to consume, rather than through complex calculations of social security wealth.

Abstract

This paper examines household saving and consumption behavior in the United States in the light of recent controversies about the impact of the social security system on the consumption/ saving decision. It begins with a brief examination of the theory, then reviews the existing empirical evidence on the impact of the social security system on capital formation in the United States, and conducts some new tests. The conclusions from the empirical work are twofold: (1) When a possible specification error is corrected, the results appear to favor somewhat the existence of a significant depressing effect of social security wealth on household saving in the United States. Without this correction, there is no such evidence. Nevertheless, the results are far from being conclusive, and major problems remain. (2) There is also some evidence that social security depresses household saving by redistributing income toward those income and age groups that have higher than average marginal propensities to consume, rather than through complex calculations of social security wealth.

This paper examines household saving and consumption behavior in the United States in the light of recent controversies about the impact of the social security system on the consumption/ saving decision. It begins with a brief examination of the theory, then reviews the existing empirical evidence on the impact of the social security system on capital formation in the United States, and conducts some new tests. The conclusions from the empirical work are twofold: (1) When a possible specification error is corrected, the results appear to favor somewhat the existence of a significant depressing effect of social security wealth on household saving in the United States. Without this correction, there is no such evidence. Nevertheless, the results are far from being conclusive, and major problems remain. (2) There is also some evidence that social security depresses household saving by redistributing income toward those income and age groups that have higher than average marginal propensities to consume, rather than through complex calculations of social security wealth.

I. Life Cycle Models and Social Security—Theoretical Discussion

The theoretical analysis of the impact of social security systems on household saving has been conducted in terms of the life cycle model of household saving and consumption (Feldstein (1974; 1977)). The broad conclusion has been that there are two major theoretical effects that act in different directions, so that unambiguous theoretical conclusions cannot be drawn. The first and most obvious effect is that the existence of a social security system, by reducing the need to save for retirement, is likely to reduce household saving.1 The second effect is that the existence of such a system may increase the incentive to retire; if the length of the retirement span increases relative to the length of working life, then the individual may have to save more during his shorter working life to provide for his longer retirement period. This effect would tend to increase household saving. The introduction of intergenerational transfers further complicates the picture. If the intergenerational transfer effected by the social security system is offset completely by changes in private intergenerational transfers (Barro (1974)), there is no net effect on saving and capital formation.

An aspect that has not been much discussed is the transitional behavior of the economy from a situation without a social security system to one with such a system. If the long-run, steady-state effect of social security is to reduce personal saving and household wealth, then during the transition the saving rate must first decline (beyond its ultimate long-run value) and then rise.2 This is a predictable consequence of stock/flow interactions and derives from the fact that, during the transition, some of the accumulated wealth of the economy has to be consumed. This necessity for the saving rate to overshoot has interesting implications that have not been widely recognized.3 It appears, for instance, that the transitional fluctuations in the savings rate may of necessity be two or three times as large as the long-run change.

To put the matter differently, the major impact of the introduction of an unfunded social security system occurs at its inception, when benefits are paid to those who have paid no contributions; the impact on later generations is more to change the time profile of receipt of income than to change its absolute amount. Accordingly, the major changes in saving and consumption should also occur soon after the introduction of the system, when these windfall effects are at their peak or after major changes in benefits.

If saving rate behavior does not appear to have changed in these periods (when such changes should be greatest), then one might plausibly infer that the steady-state impact, through this life cycle mechanism, is very small.4 Using national income and product accounts (NIPA) definitions (U.S. Department of Commerce (1981, Table 2-1)), the personal saving rate fell from 5.3 percent the year before the introduction of social security to 4.9 percent in the first year of operation.5

In the three years before the system began, the personal saving rate averaged 3.1 percent; in the first three years of the system, this rate averaged 2.8 percent. These changes are so small as to suggest very little impact at precisely the time that a social security wealth effect should have been at its greatest. This evidence is consistent with the argument that the social security wealth effect on personal saving is minimal. Similarly, if one examines the behavior of household saving in years immediately following major changes in benefits, saving rates do not move sharply.

In his seminal paper, Feldstein (1974) asserts that the U.S. social security system has reduced household saving by 30-50 percent through a social security wealth effect. The required short-run behavior to generate a long-run, steady-state effect of 30-50 percent would be correspondingly larger, conceivably beyond the bounds of historical experience. Depending on the other assumptions, it is quite likely that a long-run reduction of 30-50 percent might require that the savings rate be negative for some years during the transition (Evans (1982)).

II. Life Cycle Models and Social Security—Empirical Evidence

The traditional means of examining the impact of social security on household saving has been to estimate an aggregate consumption function, with the addition of a social security wealth variable, representing the discounted present value of future social security benefits minus taxes. The social security effect is then inferred from the sign, size, and significance of this coefficient.

As is well known, the pre-1980 time-series studies have been almost entirely invalidated by the discovery of a computational error in the Feldstein social security wealth algorithm, by Leimer and Lesnoy (1980; 1982). Two studies that were not invalidated were those of Barro (1978) and Darby (1979), each of whom constructed his own social security variable and found statistically insignificant coefficients by conventional standards. Leimer and Lesnoy (1980; 1982) calculate several alternative social security wealth variables, each based on alternative plausible assumptions about household perceptions of future benefit and tax streams, and get ambiguous results. Depending on the particular method of construction, the effect may be significantly positive or negative or insignificant. In fact, they find a significant positive effect of social security on saving more often than a significant negative effect. They conclude that the evidence is inherently unclear, as the results depend crucially on assumptions made about perceptions of future benefits and taxes. Feldstein (1980; 1982) responds with a revised algorithm and new results that support his earlier conclusions. However, he fails to address the central point of the Leimer and Lesnoy critique—the lack of robustness of his result to plausible changes in assumptions about benefit and tax perceptions, and he makes no argument to suggest that his particular formulation is preferable. An excellent recent survey by Aaron (1982) argues that the debate using time-series data has produced no conclusive evidence; he goes on to suggest that the problems may be inherently intractable.

More fundamental difficulties were raised by Auerbach and Kotlikoff (1981 b). Using their life-cycle simulation model (Auerbach and Kotlikoff (1981 a; 1981 b; 1982)), they generated “pseudodata” for which it was known, entirely by construction, that the social security system reduced household saving substantially. However, when standard econometric tests using a social security wealth variable were run on this artificially generated data, the hypothesis that social security wealth depresses household saving was sometimes accepted and sometimes rejected, depending on the precise specification, length of sample, and so on. These results provide dramatic testimony to the bluntness of empirical methods. However, the issue at hand remains empirical; no amount of simulation of theoretical structures can answer it, even though Auerbach and Kotlikoff might like to conclude otherwise. The empirical work in this section follows largely in the Feldstein/ Leimer and Lesnoy tradition and is therefore subject to the Auerbach and Kotlikoff (1981 b) critique. The simpler approach advocated in Section III may be less vulnerable to this critique, as it does not depend on construction of a social security wealth variable and does not necessarily rely on life-cycle model mechanisms.

There are two further problems with the empirical methodology employed by Feldstein, and replicated by Leimer and Lesnoy. The permanent income/life cycle theory, which underlies the consumption function specification, is essentially forward looking; the disposable income variable is a proxy for expected future income. Accordingly, if social security wealth (that is, a measure of the present value of future benefits minus future taxes) is added to the equation, then the corresponding component of disposable income (that is, present social security benefits minus taxes, which are a proxy for their future flows) should be subtracted.6 This has not been done by Feldstein or by Leimer and Lesnoy. This is the problem on which the empirical work in this section focuses.

The basic specification employed is as follows:

C=a+bYD+cYD1+dRE+eW1+fSSW(1)

where

C = consumer expenditure

YD = disposable income

RE = corporate returned earnings

W = household wealth

SSW = social security wealth

All variables are defined in real per capita terms. There are six alternative social security wealth series (Leimer and Lesnoy (1982)), two alternative retained earnings series, and two alternative disposable income series. (See the Appendix for details on data sources and construction.)

The two alternative ways of defining disposable income and the effect on the coefficient of social security wealth are the focus of attention. In the odd-numbered equations of Tables 2 and 3 in the Appendix, the double counting referred to earlier is present.7 In the even-numbered equations it is absent, as NIPA disposable income has been adjusted by subtracting social security benefits and adding social security contributions by households and employers. The results, although not conclusive, are striking. In each of the 12 variations, the coefficient on social security wealth is greater (more positive or less negative) when adjustment is made for the double counting than when it is not. When the Feldstein/ Leimer and Lesnoy specification is replicated, the coefficient on social security wealth is negative six times and positive six times; it is negative and significant (at the 10 percent level) five times; and it is positive and significant (at the 10 percent level) twice. The results are inconclusive, or perhaps moderately in favor of the possibility that social security wealth depresses consumption and increases saving. When the adjustment to the income variable is made, the picture is systematically different. Once again the social security wealth coefficient is negative six times and positive six times. However, none of the negative coefficients is significantly different from zero at the 10 percent level; on the other hand, the coefficient is positive and different from zero at the 10 percent level four times.

In the Appendix, the same exercise as in Tables 2 and 3 is repeated in Table 4 but with the retained earnings variable omitted, as it did not appear to add much, and the justification for its inclusion is not entirely clear. (See Auerbach and Kotlikoff (1981 b) and Evans (1982).) In the six equations (there being six distinct measures of social security wealth) in which no adjustment is made to disposable income, the coefficient on social security wealth is positive three times and negative three times. It is positive and significant (at the 10 percent level) once and is negative and significant three times. If anything, these results suggest that social security wealth may have a positive effect on household saving. However, when the adjustment to disposable income is made, as just advocated, the picture is quite different. The coefficient on social security wealth is negative and insignificantly different from zero (at the 10 percent level) three times out of six. It is positive three times out of six, and twice it is statistically significant at the 10 percent level. These further results, although far from conclusive, reinforce those found earlier, suggesting that when disposable income is adjusted appropriately, there is tentative evidence for the proposition that social security wealth increases household consumption, or depresses household saving.

A second remaining problem with the methodology of Feldstein and Leimer and Lesnoy is more difficult; it deals with the rationale underlying the construction of the social security wealth variable. Social security wealth has often been described as “fictitious wealth,” in the sense that, in an unfunded system, there are no corresponding real resources. However, as computed by Feldstein and others, social security wealth is doubly fictitious. The variable is constructed by extrapolating present benefit and tax rates into the future, allowing for demographic factors, and then discounting the resulting flows to give a present value. It is, of course, well known, and the subject of great public discussion, that the present social security benefit rates and tax rates in the United States may become increasingly incompatible in the next several decades, as the U.S. population ages;8 accordingly, either tax rates must be increased, some other source of funding must be found, or future benefits must be reduced. The social security wealth algorithms of Feldstein and his critics take account of none of this, and, correspondingly, a large component of their computed social security wealth derives from implicitly assumed future social security deficits, as current benefit and tax rates are extrapolated into the future. It seems at least plausible that public perceptions and expectations about social security are based on the premise of a roughly balanced social security budget, so that some combination of reductions in future benefits and increases in future taxes may already be built into expectations.9 Therefore, it might be preferable to construct social security wealth variables with an implicit constraint that the social security budget be balanced at each point of time. This provision would drastically reduce computed social security wealth for much of the sample period.

One final criticism relates to the use of flow variables and stock variables. Feldstein and his critics proxy human wealth by a flow variable (disposable income), nonhuman wealth by both a stock and a flow variable (household wealth and retained earnings), and social security wealth by a stock variable (computed social security wealth). There is no discussion provided of this treatment—of the rationale for using a stock variable in one instance and a flow variable in another. Some discussion of the appropriate use of stock and flow variables would be desirable. The next section discusses this issue further and presents some empirical results.

III. An Extension and Simplification

One possible channel of influence of the social security system on saving and consumption decisions has had little attention. As well as creating a kind of contingent future wealth, social security produces substantial redistribution of current income. Social security benefits in the United States at present make up more than one half of government transfers to persons; they accounted for about 8 percent of disposable income in 1979 (Table 1), compared with only ½ of 1 percent in 1950. Government transfer payments made up 4 percent of disposable income in 1945 and rose to more than 15 percent by the late 1970s. If recipients of social security benefits (or transfers in general) have different propensities to consume from the population at large, significant effects on saving and consumption can result, even in the absence of the complex calculations of social security wealth invoked by Feldstein. Recipients of transfer payments are more likely to be liquidity constrained than the population at large and, hence, to have higher than average propensities to consume. In addition, social security recipients are predicted by the life cycle model to have higher propensities to consume, because of their greater age. In accordance with these considerations, in this section of the paper, empirical tests are conducted that examine the impact of social security payments (and of transfer payments in general) by disaggregating disposable income into transfer and nontransfer components.

Table 1.

United States: Transfer Payments and Saving Rates, 1951-79

(As percentage of household disposable income)

article image
Sources: U.S. Department of Commerce, National Income and Product Accounts, 1929-76 (Washington, 1981); Survey of Current Business, various issues.

For several reasons, this approach may be preferable to that of Feldstein, which cumulates the effect of social security into a stock variable. The first reason is that construction of a social security wealth variable requires forbidding assumptions and imputations relating to such matters as choice of discount rate, modeling of expectations, future productivity growth, future labor force participation rates, and future age-specific fertility and mortality rates. It may be easier and preferable to rely on the current flow as a proxy for the expected future flows.

The second reason for avoiding the social security wealth approach of Feldstein and his critics brings us back to the issue of the stock variable versus the flow variable that was raised in Section II. An important argument for representing human wealth (that is, future labor income) by the flow of current (and perhaps lagged) income, rather than by a discounted present value formulation, is that capital markets are such that it is not possible—or only minimally possible—to borrow against future labor income to finance current consumption. Accordingly, the stock of human wealth of an individual is notional, not effective or realizable. However, nonhuman wealth either is already liquid or is represented by an estimate of its market value; accordingly, the stock of nonhuman wealth is effective or realizable.10

Judged by the foregoing criterion of effectiveness or realizability, it is clear that the effect of social security should be represented by a flow variable. It is extremely hard to imagine capital markets permitting an individual to borrow against future social security benefits; an individual’s social security wealth is entirely a notional concept.

Consequently, the basic specification adopted for the empirical work in this section is as follows:

C=a+b(YDTF)+C(YD1TF1)+dTF+eW1(2)

where

C = consumer expenditure

YD = disposable income

TF= transfer payments (either total government transfers to persons or, simply, social security benefits)11

W = household wealth

Results are presented in Table 5 in the Appendix. When disposable income is disaggregated into social security benefits and all other disposable income, the coefficient on social security benefits is 1.49 (equation (7)) but does not appear to be stable (equations (8) and (9)). It appears that social security benefits may be a proxy for some other effect, such as, possibly, the increasing proportion of the population that is over the age of 65.

When disposable income is separated into total government transfer payments to persons and nontransfer income, the coefficient on transfer payments is 0.98 (equation (4)) and the other coefficients are reasonable. This coefficient seems to be reasonably stable across subperiods (equations (5) and (6)). The fit of equation (4) is marginally better than that of the comparable equation without disaggregation of disposable income (equation (1)), although the difference is small. The results presented are consistent with the contention that the propensity to consume out of transfer payments in the United States is unity, compared with a propensity to consume out of nontransfer disposable income of about 0.7. Given the increasing share of transfer payments in disposable income (Table 1), this result provides an explanation for lower household saving rates. Such changes in the importance of transfer payments imply a decline in the household saving rate of about 2 percent, ceteris paribus, from 1960 to 1979.

Consequently, it appears possible that transfer payments may exert their influence less by changing lifetime calculations than by redistributing income across heterogeneous groups, with different propensities to consume. In many respects, this mechanism may be simpler and more plausible than that proposed by Feldstein.

IV. Summary and Conclusion

In Section I, it was argued that the short-run response of the saving rate to social security should be larger than the long-run response. Accordingly, the claim by Feldstein (1974) that the social security system has reduced household saving by 30-50 percent through a social security wealth effect seems dubious, as the necessary transitional fluctuations in the saving rate would be huge, perhaps beyond that which has been experienced historically.

In Section II, it was argued that the empirical work of Feldstein and his critics has been flawed by a specification error. When this is corrected, the results—although still far from conclusive—appear to favor the existence of a depressing effect of social security wealth on household saving. Before correction of the possible specification error, the results were inconclusive. A further difficulty is that much of measured social security wealth may derive from implicitly assumed social security deficits in the future, which is an extremely debatable assumption.

In Section III, a simpler approach is advocated. It is argued that government transfer programs in the United States may have lowered household saving by redistributing income to those with higher than average propensities to consume—the poor, the liquidity constrained, and the old. In support of this approach, the tests that were conducted found a propensity to consume out of transfer payments that was close to unity.

APPENDIX

In all cases, the dependent variable is real per capita consumer expenditure. The estimation technique is ordinary least squares, with a Cochrane-Orcutt correction for first-order serial correlation. Data are annual and are from the U.S. Department of Commerce (a) National Income and Product Accounts, 1929-76 (Washington, 1981), and (b) Survey of Current Business, various issues; (c) the Flow of Funds Section, Board of Governors of the Federal Reserve System (Washington, August 1982); (d) Evans (1969); and (e) Leimer and Lesnoy (1982).

Notation and data construction

article image
Table 2.

United States: Consumer Expenditure Equations with Social Security Variables, Excluding Inventory and Capital Consumption from Retained Earnings, 1947-76

article image
Table 3.

United States: Consumer Expenditure Equations with Social Security Variables, Including Inventory and Capital Consumption in Retained Earnings, 1947-76

article image
Table 4.

United States: Consumer Expenditure Equations with Social Security Variables, Omitting the Retained Earnings Variable, 1947-76

article image
Table 5.

United States: Consumer Expenditure Equations with Disaggregated Disposable Income, 1947-79

article image

BIBLIOGRAPHY
  • Aaron, Henry J., Economic Effects of Social Security, The Brookings Institution (Washington, 1982).

  • Ando, Albert, and Franco Modigliani, The ‘Life Cycle’ Hypothesis of Saving: Aggregate Implications and Tests,American Economic Review, Vol. 53 (March 1963), pp. 5584.

    • Search Google Scholar
    • Export Citation
  • Auerbach, Alan J., and Laurence J. Kotlikoff (1981 a), “National Savings, Economic Welfare, and the Structure of Taxation,” National Bureau of Economic Research, Working Paper No. 729 (Cambridge, Massachusetts, August 1981). (This article is scheduled for publication in Behavioral Simulation Methods in Tax Policy Formation, ed. by Martin S. Feldstein, University of Chicago Press.)

    • Search Google Scholar
    • Export Citation
  • Auerbach, Alan J., and Laurence J. Kotlikoff (1981 b), “An Examination of Empirical Tests of Social Security and Savings,” National Bureau of Economic Research, Working Paper No. 730 (Cambridge, Massachusetts, August 1981). (This article is scheduled for publication in Social Policy Evaluation: An Economic Perspective, ed. by Elhanan Helpman.)

    • Search Google Scholar
    • Export Citation
  • Auerbach, Alan J., and Laurence J. Kotlikoff (1982), “Investment versus Savings Incentives: The Size of the Bang for the Buck and the Potential for Self-Financing Business Tax Cuts,National Bureau of Economic Research, Working Paper No. 1027 (Cambridge, Massachusetts, November 1982).

    • Search Google Scholar
    • Export Citation
  • Auerbach, Alan J., and Laurence J. Kotlikoff, and Jonathan Skinner, The Efficiency Gains from Dynamic Tax Reform,International Economic Review, Vol. 24 (February 1983), pp. 81100.

    • Search Google Scholar
    • Export Citation
  • Barro, Robert J. (1974), “Are Government Bonds Net Wealth?Journal of Political Economy, Vol. 82 (November/December 1974), pp. 1095117.

    • Search Google Scholar
    • Export Citation
  • Barro, Robert J. (1978), The Impact of Social Security on Private Saving: Evidence from the U.S. Time Series, American Enterprise Institute for Public Policy Research, AEI Studies, No. 199 (Washington, 1978).

    • Search Google Scholar
    • Export Citation
  • Boskin, Michael J., John B. Shoven, and Laurence J. Kotlikoff, The Reforms Leave Many Problems,New York Times (January 23, 1983), Section C, p. 2.

    • Search Google Scholar
    • Export Citation
  • Darby, Michael R., The Effects of Social Security on Income and the Capital Stock, American Enterprise Institute for Public Policy Research (Washington, 1979).

    • Search Google Scholar
    • Export Citation
  • Evans, Michael K., Macroeconomic Activity—Theory, Forecasting, and Control: An Econometric Approach (New York, 1969).

  • Evans, Owen J. (1982), “The Life Cycle Inheritance: Theoretical and Empirical Essays on the Life Cycle Hypothesis of Saving” (unpublished doctoral dissertation, University of Pennsylvania, May 1982).

    • Search Google Scholar
    • Export Citation
  • Evans, Owen J. (1983), “Tax Policy, the Interest Elasticity of Saving, and Capital Accumulation: Numerical Analysis of Theoretical Models,American Economic Review, Vol. 73 (June 1983), pp. 398410.

    • Search Google Scholar
    • Export Citation
  • Feldstein, Martin S. (1974), “Social Security, Induced Retirement, and Aggregate Capital Accumulation,Journal of Political Economy, Vol. 82 (September/October 1974), pp. 90526.

    • Search Google Scholar
    • Export Citation
  • Feldstein, Martin S. (1977), “Social Security and Private Savings: International Evidence in an Extended Life-Cycle Model,” Chap. 8 in The Economics of Public Services: Proceedings of a Conference Held by the International Economic Association in Turin, Italy, ed. by Martin S. Feldstein and Robert P. Inman (London, 1977), pp. 174205.

    • Search Google Scholar
    • Export Citation
  • Feldstein, Martin S. (1980), “Social Security, Induced Retirement, and Aggregate Capital Accumulation: A Correction and Update,National Bureau of Economic Research, Working Paper No. 579 (Cambridge, Massachusetts, November 1980).

    • Search Google Scholar
    • Export Citation
  • Feldstein, Martin S. (1982), “Social Security and Private Saving: Reply,Journal of Political Economy, Vol. 90 (June 1982), pp. 63042.

    • Search Google Scholar
    • Export Citation
  • Kopits, George, and Padma Gotur, The Influence of Social Security on Household Savings: A Cross-Country Investigation,Staff Papers, Vol. 27 (March 1980), pp. 16190.

    • Search Google Scholar
    • Export Citation
  • Leimer, Dean R., and Selig D. Lesnoy (1980), “Social Security and Private Saving: A Reexamination of the Time Series Evidence Using Alternative Social Security Wealth Variables,U.S. Department of Health, Education, and Welfare, Office of Research and Statistics, Working Paper No. 19 (Washington, November 1980).

    • Search Google Scholar
    • Export Citation
  • Leimer, Dean R., and Selig D. Lesnoy (1982), “Social Security and Private Saving: New Time-Series Evidence,Journal of Political Economy, Vol. 90 (June 1982), pp. 60629.

    • Search Google Scholar
    • Export Citation
  • Lesnoy, Selig D., and Dean R. Leimer, Social Security and Private Saving: New Time Series Evidence with Alternative Specifications,U.S. Department of Health and Human Services, Office of Research and Statistics, Working Paper No. 22 (Washington, January 1981).

    • Search Google Scholar
    • Export Citation
  • Modigliani, Franco, and Richard Brumberg, “Utility Analysis and Aggregate Consumption Functions: An Attempt at Integration,” in The Collected Papers of Franco Modigliani, Vol. 2: The Life Cycle Hypothesis of Saving, ed. by Andrew Abel (The MIT Press, 1980), pp. 12897.

    • Search Google Scholar
    • Export Citation
  • U.S. Department of Commerce, Survey of Current Business (Washington), various issues.

  • U.S. Department of Commerce (1981), The National Income and Product Accounts of the United States, 1929-76 (Washington, 1981).

*

Mr. Evans, economist in the European Department, was in the Fiscal Affairs Department when this paper was prepared. He was a Research Fellow at the Brookings Institution and is a graduate of the University of Sydney and the University of Pennsylvania.

This paper expands on work from his doctoral dissertation and has benefited from discussions with Albert Ando, Barry Bosworth, Jared Enzler, John B. Shoven, and colleagues in the Fund. The views expressed are the author’s alone.

1

If the system is financed on a pay-as-you-go basis, there is no offsetting government saving, so that national saving is reduced.

2

For a detailed exposition, see Evans (1982, Chap. 2).

3

For a discussion of savings rate overshooting in the context of interest rate changes, see Evans (1983). A useful way to check the plausibility of a proposition couched in terms of steady states is to examine the required transitional behavior; to be fulfilled, an otherwise plausible long-run result may require startling transitional behavior.

4

Such qualitative inferential evidence should, of course, be regarded only as supplementary to that provided by econometric testing.

5

Using 1937 as the first year of the system.

6

If one believes that the disposable income term is essentially backward looking, representing current and previous income, then this problem may not exist at the theoretical level. Nevertheless, as the variation in measured social security wealth will be generated largely by changes in current social security benefits and taxes, an insignificant coefficient on social security wealth, if no adjustment is made to disposable income, would not be surprising, even if Feldstein’s hypothesis were correct. The existence of this double-counting issue was brought to the author’s attention by Jared Enzler, of the Board of Governors of the Federal Reserve System.

7

The sample used is for the period 1947-76. Prewar data are avoided, following the arguments by Leimer and Lesnoy (1982). Social security wealth data for the years after 1976 had not been published when this paper was written.

8

This was certainly true before the reform package that was enacted early in 1983 and, according to many well-placed critics, is still true today. See, for example, Boskin, Shoven, and Kothkoff (1983).

9

The argument bears a resemblance to that which claims that government debt is not net wealth, because the public anticipates future tax increases to repay the debt. However, the argument here may be stronger; it is well known that social security benefits and taxes roughly balance each year. It, therefore, may be unwise to construct an expected future social security wealth variable without building in this constraint.

10

These arguments seem to be influential in much of the consumption function literature, although they are often not explicit.

11

With disposable income adjusted consistently.