The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Business Review Federal Reserve Bank o f Philadelphia Septem berOctober 1993 ISSN 0 0 0 7 -7 0 1 1 Business Review The BUSINESS REVIEW is published by the Department of Research six times a year. It is edited by Sarah Burke. Artwork is designed and produced by Dianne Hallowell under the direction of Ronald B. Williams. The views expressed here are not necessarily those of this Reserve Bank or of the Federal Reserve System. SUBSCRIPTIONS. Single-copy subscriptions for individuals are available without charge. Insti tutional subscribers may order up to 5 copies. BACK ISSUES. Back issues are available free of charge, but quantities are limited: educators may order up to 50 copies by submitting requests on institutional letterhead; other orders are limited to 1 copy per request. Microform copies are available for purchase from University Microfilms, 300 N. Zeeb Road, Ann Arbor, MI 48106. REPRODUCTION. Permission must be obtained to reprint portions of articles or whole articles. Permission to photocopy is unrestricted. Please send subscription orders, back orders, changes of address, and requests to reprint to Publications, Federal Reserve Bank of Philadelphia, Department o f Research and Statistics, Ten Independence Mall, Philadelphia, PA 19106-1574, or telephone (215) 574-6428. Please direct editorial communications to the same address, or telephone (215) 574-3805. SEPTEMBER/OCTOBER 1993 DO AMERICANS SAVE TOO LITTLE? B. Douglas Bernheim & John Karl Scholz Should policymakers encourage public saving through deficit reduction, or pri vate saving through tax incentives and pension policies? Economists debate about which method is more efficacious. Doug Bernheim and John Karl Scholz examine the private saving side of the debate by raising two questions: Is there reason to be concerned about the rate of private saving? And are there any effec tive and reliable methods of promoting private saving? HIGHWAYS AND EDUCATION: THE ROAD TO PRODUCTIVITY? Gerald A. Carlino The slowdown in productivity growth in recent decades has become a cause for concern. The decline in investment in public infrastructure and the decline in educational quality may have played a role in this slowdown. Can improved infrastructure, such as more roads, and higher educational attainment lead to in creased productivity? Jerry Carlino looks at some of the factors involved in regional productivity to determine if more high ways, increased education, and produc tivity growth are indeed linked. Do Americans Save Too Little? c B. Douglas Bernheim* & John Karl Scholz* Ita-Jince the mid-1980s, low rates of national saving in the United States have generated an enormous amount of concern among both economists and policymakers. Proposals to address these concerns fall into two broad categories: policies designed to increase public *B. Douglas Bernheim is the John L. Weinberg Professor of Economics and Business Policy, Princeton University. John Karl Scholz is an assistant professor of economics, University of Wisconsin, Madison. When this article was written, Bernheim was a visiting scholar in the Research Department of the Philadelphia Fed. The authors gratefully acknowledge the work of Robert Avery and Arthur Kennickell, who developed a clean copy of the 1983-86 Survey o f Consumer Finances and provided extensive docu mentation. saving and policies intended to promote pri vate saving. The former is synonymous with deficit reduction, while the latter includes tax incentives, pension policy, and strategies for discouraging the use of private debt. Some economists argue that deficit reduction is the most reliable and efficacious method of in creasing national saving (Summers, 1985), while others maintain that restoring adequate rates of private saving is essential (Bernheim, 1991). To evaluate the merits of strategies that target private saving, we must resolve two issues. First, aside from the obvious fact that private saving is one component of national saving, is there reason to be concerned about the rate of private saving? Second, are there any effective and reliable methods of promoting private sav ing? 3 BUSINESS REVIEW SEPTEMBER/OCTOBER1993 cal in all respects except that the elderly make THE ADEQUACY up a larger fraction of the population in A than OF HOUSEHOLD SAVING According to common wisdom, Americans in B. Since households tend to accumulate consume too much and save too little. This wealth prior to retirement and spend wealth impression is largely traceable to widely publi thereafter, we would expect to observe a higher cized statistics on aggregate personal saving. rate of aggregate personal saving in country B. International comparisons reveal that U.S. Indeed, in an economy with no growth in either households save significantly less than their population or productivity, dissaving by retir foreign counterparts. Between 1980 and 1991, ees could completely offset saving by workers: Americans saved 6.4 percent of disposable per in principle, regardless of how well individual sonal income, compared with 9.8 percent for households prepared for retirement, we might OECD Europe and 15.7 percent for Japan (Or observe virtually no aggregate personal sav ganization for Economic Cooperation and De ing. Thus, ultimately, we can judge the ad velopment, 1992). And since the mid-1980s, the equacy of personal saving only by examining rate of household saving in the U.S. has been microeconomic data on the behavior of indi vidual households. well below its historical average (Figure 1). Generally, the available evidence suggests Although these statistics raise legitimate concerns, they do not provide definitive evi that American workers have prepared poorly dence of a problem. As measured, personal for retirement. Diamond (1977) found that, saving excludes capital gains. Thus, in prin during the 1960s, 40 percent of couples and ciple, households can accumulate wealth at a more than 50 percent of unmarried individuals rapid rate even when their measured rates of reported that after retirement they received no saving are low. Rates of personal saving can money income from assets. At age 60, nearly 30 also vary across both time and countries for FIGURE 1 reasons unrelated to the adequacy of saving con Rate of Personal Saving, sidered from the per National Income Accounts spective of individual households.1*To under Percent of disposable income stand this second point, 10 consider the following hypothetical example. 8 Envision two countries, A and B, that are identi 1Indeed, Meyer, 1992, ar gues that demographic differ ences account for roughly onethird of the gap in personal saving relative to GNP be tween Germany and the U.S. during the 1980s and roughly two-thirds of the gap between Japan and the U.S. 4 FRASER Digitized for 70 72 74 76 78 80 82 84 86 88 90 Year FEDERAL RESERVE BANK OF PHILADELPHIA Do Americans Save Too Little? B. Douglas Bemheim & John Karl Scholz percent of middle-class individuals lacked suf particular assets were accumulated for retire ficient wealth to replace two years' worth of ment or for some other purpose. Consequently, income. Similarly, Hamermesh (1984) con the comparison between estimated trajectories cluded that, during the 1970s, most elderly and simulated trajectories may provide an individuals had not accumulated sufficient re overoptimistic picture of the adequacy of house sources to sustain their accustomed standards hold saving. of living. Indeed, consumption shortly after We show graphic depictions (Figures 2 and retirement exceeded the highest sustainable 3) of a simulation for a household with the level of consumption by an average of 14 per following characteristics: age 27 (as of 1991), cent. Hamermesh also found that within a few two years of college education, married, two years of retirement most retirees were forced to workers with total current earnings of $60,540, reduce their expenditures substantially.2 and the primary earner covered by a private Asset Accumulation Profiles. More recent pension plan. This household's optimal trajec evidence on the adequacy of saving appears in tory of consumption and after-tax earned in Bernheim and Scholz (1992a). Using an elabo come (including pensions and Social Security) rate model of household decision-making, we is shown in constant 1991 dollars (Figure 2, simulated asset accumulation profiles (trajec page 7).4 Note that after-tax earnings rise steeply tories) that households should follow (given early in life. Earnings growth continues at a the assumptions of the model) to prepare ad reduced level until the individual reaches age equately for retirement.3 We then compared 55, at which point it begins to fall. After retire these simulated profiles with ones estimated ment, earned income consists of Social Security from recent surveys of households' actual sav and private pension benefits. Since pensions ing behavior. (For a more detailed description are not perfectly indexed for inflation, real of the model, see Explanation o f the Model.) benefits decline gradually over time. The simulation model describes only the As a direct consequence of the household's accumulation of assets for retirement. There rapid earnings growth early in life, it saves are, of course, many reasons to save. House nothing for retirement prior to age 30. Between holds should take precautions against the pos ages 30 and 80, the consumption trajectory is sibility of illness, layoff, disability, death, and relatively flat. This flat trajectory reflects the other risks for which they are imperfectly in household's preference for a stable standard of sured. In addition, most households accumu living. However, during the 30s and 40s, con late resources to pay for large expenses such as sumption is elevated relative to the 60s and 70s. college tuition or the purchase of an automo This pattern results from changes in household bile. For some individuals, saving is motivated composition: between the ages of 30 and 50, the in part by the desire to leave a substantial typical household incurs significant child-rear bequest upon death. Unfortunately, when ex ing costs. Consumption declines rapidly after amining the data, we cannot determine whether age 80 until, at age 101, it matches after-tax retirement benefits. Falling survival probabili ties cause this end-of-life decline. Since there is a relatively low probability of reaching age 90, 2Other economists have reached somewhat more opti mistic conclusions. See Kotlikoff, Spivak, and Summers, 1982. developm ent of this model was sponsored by Merrill Lynch & Co., Inc., and is described in Bernheim, 1992b. 4We use the word "trajectory" to describe the manner in which an economic variable, such as consumption, income, or wealth, evolves as the household ages. 5 Explanation of the Model BUSINESS REVIEW SEPTEMBER/OCTOBER1993 Our simulation model reflects a "life-cycle" approach to the average household's financial decision-making process. It takes into account the fact that predictable changes in household earnings resulting from age and stage of career may not match up very well with consump tion needs. For example, the financial needs of most households are usually highest during the child-rearing years, while household earnings usually reach their highest point after children have left home. The household varies its rate of saving in order to achieve a better match between the ability to spend and the need to spend. It saves least in years when spending needs are high and more in years when spending needs decline.3 The model forecasts households' future income and derives the optimal consumption (and thus saving) trajectories consistent with those income forecasts. Our life-cycle calculations account for a variety of current and future household charac teristics, including age, income level, pension coverage, education, marital status, gender (if unmarried), and household composition (the numbers of children and dependent adults).b The model also projects and adjusts for future macroeconomic conditions that ought to affect savings behavior, including interest rates, inflation rates, and baseline wage growth. In addition, the model provides a realistic treatment of income taxes, payroll taxes, and social security benefits. To conduct simulations, one must also choose values for several "preference parameters." For example, the model includes a parameter commonly known as the "pure rate of time preference," which expresses the value that a household places on future consumption relative to current consumption.0The value of this particular parameter has a profound effect on the simulation results. When the pure rate of time preference is sufficiently low, it is optimal for the household to save nothing. For this reason, the absence of saving is not necessarily the result of irrationality. Rather, it may simply reflect impatience. We have calibrated our model (that is, chosen values for the preference parameters) so that the simulations produce a standard of living during retirement that is roughly comparable to the standard of living enjoyed prior to retirement.11 Consequently, it is appropriate to interpret our results as follows: if households fall significantly short of simulated asset accumulation targets, they will ordinarily be forced to accept serious reductions in their standards of living after retirement. aWhen spending needs are sufficiently high relative to income, a household may wish to liquidate or borrow against accumulated assets. Once assets are exhausted, it may be optimal for the household to borrow against future income. However, for most households, it is extremely difficult to obtain sizable unsecured loans. Our model therefore imposes a "liquidity constraint," which ensures that the household's net wealth remains positive. bOur calculations reflect the fact that larger households benefit from significant economies of scale. Research on household scale economies indicates that two adults in a household can obtain the same standard of living as one adult living alone with added expenditures of slightly more than 40 percent. Research also shows that the financial impact of adding one adult to a household is roughly equivalent to adding 2.5 children. See Cutler and Katz, 1992. cOther important preference parameters include a minimum subsistence level for consumption and a parameter known as the "intertemporal elasticity of substitution," which measures the extent to which the household's willingness to trade off current consumption for future consumption is affected by the level of current consumption relative to future consumption. dSpecifically, we use a pure rate of time preference equal to the product of 0.99 and one-year genderspecific survival probabilities (taken from standard life tables). The minimum consumption level is set equal to $10,000 (measured in 1991 dollars), and is adjusted for family size. A value of 0.25 is used for the intertemporal elasticity of substitution. Digitized for 6 FRASER FEDERAL RESERVE BANK OF PHILADELPHIA Do Americans Save Too Little? B. Douglas Bernheim &John Karl Scholz the household would prefer to accept a lower of the best available sources of data on house standard of living at age 90 and later (if it hold balance sheets.6 survived that long) in favor of a higher stan dard of living earlier in life. The associated optimal trajectory of retire 6See Avery and Elliehausen, 1988, and Avery and ment assets is also depicted (Figure 3). Assets Kennickell, 1988, for a more complete discussion of the SCF. accumulate at an increasing rate from age 30 to retire FIGURE 2 ment, peak at retirement, then decline steadily until Simulated After-Tax Income and they are exhausted at age Consumption Trajectories 100 . We then estimated actual Dollars (thousands) asset trajectories using data 70 from the Survey o f Consumer Finances (SCF) for 1983 and 1986.5 The Board of Gover nors of the Federal Reserve (in conjunction with other federal agencies) sponsored the SCF, recognized as one After-tax income 1Spending 5Our measure of accumulated net worth includes stocks and mu tual funds, bonds, checking and sav ings accounts, IRA and Keogh ac counts, money market accounts, cer tificates of deposit, profit-sharing and thrift accounts, the dollar cash value of whole life insurance, and other financial assets, as well as eq uity in property (other than primary residences) and business assets, less credit card, consumer, and other debt. This measure excludes all as sets and liabilities associated with homes and vehicles, since house holds appear to have a strong aver sion to paying living expenses dur ing retirement by drawing down the equity in their homes (see Venti and Wise, 1989). Also, it seems likely that few individuals save for retire ment by accumulating wealth in the form of vehicles. Accumulated wealth for 1983 is expressed in 1986 dollars using the Consumer Price Index. 27 33 39 45 51 57 63 69 75 81 87 93 99 30 36 42 48 54 60 66 72 78 84 90 96 102 Age FIGURE 3 Simulated Wealth Trajectory Retirement assets (thousands) 400 27 33 39 45 51 57 63 69 75 81 87 93 99 30 36 42 48 54 60 66 72 78 84 90 96 102 Age 7 BUSINESS REVIEW SEPTEMBER/OCTOBER 1993 Our analysis allows us to compare actual earner completed college are depicted in Figure and simulated optimal behavior. The results 5. The contrast between Figures 4 and 5 is for households in which the primary worker remarkable. In cases where the household has not completed college are shown in Figure head completed college, both simulated and 4. In this figure, "actual" refers to the estimated estimated changes in wealth rise steeply with change in wealth (measured as a fraction of age. Moreover, simulated asset accumulation wage income) for the representative household tracks actual asset accumulation remarkably within each age group (calculated using the well. Taken at face value, Figure 5 suggests that SCF); "Sim/no pen" indicates the simulated highly educated households saved adequately change in wealth (again as a fraction of wage for retirement between 1983 and 1986. Although it is tempting to conclude that income) for a representative household with out pension coverage for the primary earner; inadequate saving is largely confined to those and "Sim/pen" denotes the simulated change without a college education, this conclusion in wealth for a representative household with must be tempered by two considerations. First, pension coverage for the primary earner. Note as is apparent from Figure 1, personal saving that the simulated change in wealth rises steeply declined sharply after the 1983-86 period on with age. This steep increase in assets results which the estimates are based. Using a sample from two factors. First, during most of an of relatively young individuals (ages 25 through individual's working life wages rise more rap 44) surveyed in early 1992, Bernheim (1992a) idly than consumption (see Figure 2). Second, found much more pervasive evidence of inad reinvested capital income rises as the house equate saving. Second, the model probably hold accumulates assets. In contrast, the esti understates the amount of wealth that each mated change in wealth does not vary signifi household ought to accumulate. The most cantly with age. By the time the household obvious reason for this discrepancy is that the reaches middle age, simu lated asset accumulation FIGURE 4 exceeds actual accumula Rates of Asset Accumulation tion by a wide margin.7 Overall,between 1983 and No College Degree 1986, households without Annual change in wealth as a multiple of earnings a college education saved 0.25 far less than the simula tion model predicts (Fig ure 4). Results for households in w hich the prim ary 7Although estimated asset ac cumulation is actually higher at ages 27 and 32, this is of little consequence; recall that the data reflect saving for a variety of pur poses aside from retirement. Digitized for8 FRASER FEDERAL RESERVE BANK OF PHILADELPHIA Do Americans Save Too Little? B. Douglas Bernheim & John Karl Scholz simulations envision retirement planning as the sole motive for saving.8 To the extent that many households prepare poorly for retirement, there is cause to be con cerned about the rate of personal saving, per se. Historically, pension policy and tax policy have been the two most important tools for stimulat ing personal saving. We will discuss evidence on the efficacy of each of these strategies in turn. possibility that policies affecting private pen sions may have powerful effects on aggregate personal saving. Whether these effects would actually materialize depends on the way work ers would respond to an expansion of private pension coverage. Economic theory suggests that such an expansion would simply crowd out other forms of personal saving: once work ers realize that their employers are, in effect, saving for them, workers will save less them PENSION POLICY selves. The simulation results presented in the In recent years, the accumulation of assets in previous section illustrate this principle. How private pension plans has accounted for a sub ever, previous studies of personal saving have stantial fraction of personal saving (Bernheim generally failed to find evidence to support the and Shoven, 1988). This observation raises the notion that private pensions significantly re duce other forms of personal saving.9 Depend ing on whether we credit the theoretical analy 8In addition, it is quite likely that the model overstates sis or the empirical studies, we can reach dra mortality probabilities (since it does not make any allow matically different conclusions about the effect ance for the fact that these probabilities are projected to of pension policy on aggregate personal saving. decline in the future), understates the importance of health The analysis described in the preceding sec and long-term care costs for the elderly, and fails to consider tion raises an intriguing possibility: if the be the effects of mounting economic pressures that may force Congress and employers to scale back existing retirement havior of those with a college education (and benefits. higher average incomes) conforms to the pre dictions of standard eco nomic theories, while the FIGURE 5 behavior of those without Rates of Asset Accumulation a college education (who have lower average in College Degree comes) does not, perhaps private pensions do dis Annual change in wealth as a multiple of earnings p lace p erso n al saving 0.5 among the college edu cated, but not among the rest of the population. In that case, pension policy could be an effective tool for stimulating total per sonal saving, so long as it is primarily used to pro- 27 32 37 42 47 Age 52 57 62 9See, for example, the review in Shefrin and Thaler, 1988, par ticularly pages 622-24. 9 BUSINESS REVIEW SEPTEMBER/OCTOBER1993 vide incentives for expanded coverage among saving displacement effect simply because they lower income, generally less educated, work did not distinguish between households on the ers. basis of education (or permanent income). To investigate this idea, we estimated equa The contrast between Figures 6 and 7 points tions that explained the median value of house to a clear and important conclusion for pension hold wealth as a function of age, total house policy: private pensions displace personal hold earnings, private pension coverage, and wealth accumulation only when the head of the educational attainment. We then used these household is college-educated. This observa equations to project asset accumulation pro tion aligns with the evidence on the adequacy files. of personal saving described in the first section Results for the median household in which of this article. Indeed, our evidence broadly the primary earner has not completed college supports a more general conclusion: collegeare presented in Figure 6. Note that pension educated households behave in the manner eligibility has little or no effect on the actual predicted by standard economic theories of path of household wealth accumulation. From saving, while less well-educated households a statistical perspective, the estimated equation do not. Past and current policies have been supports the notion that, at every age, less more successful at stimulating the expansion of educated households with private pensions pension coverage among college-educated accumulate wealth at the same rate as those without private pensions. Results for households in which the primary 10It is unlikely that the observed relationship between earner has completed college are displayed in pension coverage and saving results from spurious factors, Figure 7. Statistically, the data decisively reject since such factors would presumably also have produced the premise that the rate of asset accumulation the same patterns for less educated households. is unrelated to pension eli gibility. Note that those FIGURE 6 eligible for pensions accu Estimated Wealth Trajectories mulate resources at a sig nificantly slower rate than No College Degree those without pensions. Assets/ earnings Remarkably, at age 62, the 1.8 gap between the assets of these two groups is almost identical in magnitude to the predicted gap that emerges from our compu tations. These patterns are strongly consistent with the view that private pen sions displace other per sonal saving for collegeeducated hou seholds.10 These results suggest that other studies may have failed to find a significant Digitized for 10 FRASER FEDERAL RESERVE BANK OF PHILADELPHIA Do Americans Save Too Little? B. Douglas Bernheim & John Karl Scholz greater future consumption, even if the house hold were to save a bit less out of its current income.) Indeed, empirical estimates of the sensitivity of saving to the after-tax rate of return (called the interest elasticity of saving) vary widely (Boskin, 1978; Summers, 1981; and Hall, 1988). Individual Retirement Accounts. Most current proposals to provide tax incentives for saving are patterned after individual retire ment accounts (IRAs). IRAs were established as part of the 1974 Employee Retirement In come Security Act to give workers not covered by employer-provided pension plans added TAX POLICY The most commonly discussed strategies for incentives to accumulate resources for retire stimulating personal saving entail reductions ment. In 1981, IRA eligibility was extended to in the taxation of capital income. Economic all taxpayers. Subsequently, the Tax Reform theory suggests that households will respond Act of 1986 curtailed the tax-deductibility of to a higher after-tax rate of return on savings by IRA contributions for high income households. increasing future consumption relative to cur The existence of an income cap for IRAs raises rent consumption. However, theory does not an important question: does the sensitivity of saving to the after-tax rate of return vary sys n ecessarily p red ict that cu rren t saving w ill rise. (The reason is that a higher rate of return will tematically across income classes? The answer make wealth grow more rapidly, enabling to this question makes it possible to determine whether the current sys tem targets the most re FIGURE 7 sponsive groups. Estimated Wealth Trajectories Simulations based on the model described in this College Degree article suggest that higher Assets/ earnings income individuals will be 5.0 m uch more responsive than lower income indi Pension 4.0 No pension viduals to changes in the after-tax rate of return. Averaging across indi 3.0 viduals with pensions and individuals without pen 2.0 sions, the simulations im ply that saving by 35-year1.0 old, co llege-ed u cated h ou seh old s w ould in 0.0 crease by 10.2 percent in response to a permanent one-percentage-point in- workers than among those with less education. Analysis of the SCF data reveals that 75.2 per cent of college-educated husbands are covered by private pensions. In contrast, only 55.7 percent of husbands who lack a college educa tion are covered by private pensions. In other words, the current system is quite effective at providing pensions to those individuals who reduce other saving in response and much less effective at providing coverage to those indi viduals for whom pensions would represent incremental saving. li BUSINESS REVIEW crease in the before-tax rate of return, while the saving of 35-year-old, high-school-educated households would fall by 4.5 percent. Conse quently, policies that provide tax incentives for saving exclusively to lower income households exclude those individuals most likely to in crease saving in response to tax incentives; indeed, such policies could actually reduce aggregate personal saving. This positive relationship between income and the interest elasticity of saving results from a natural economic consideration, rather than from some peculiar feature of the simulation model. It is natural to assume that when plan ning for the future, most households are con cerned first and foremost with saving enough to assure themselves of some minimum stan dard of living. As lifetime resources increase, households have more discretion to allocate resources in a manner that increases consump tion above and beyond this minimum standard both today and in the future. For low income households, saving to achieve some minimum future consumption is prob ably far more important than saving to fund incremental consumption. Saving to provide for minimum consumption is, in effect, saving for a fixed target. An individual who saves to achieve some target will reduce saving in re sponse to an increase in the rate of return (Bernheim and Shoven, 1988). Thus, because target saving dominates the simulated behav ior of these households, they exhibit a low or negative interest elasticity of saving. For high income households, however, saving to fund incremental consumption is probably far more important than saving to achieve the minimum consumption target. Incremental saving domi nates the simulated behavior of these house holds. Thus we observe a high interest elastic ity of saving among higher income, well-edu cated households. Discretionary saving to fi nance consumption over and above the target responds positively to an increase in the rate of return. Digitized 12 for FRASER SEPTEMBER/OCTOBER 1993 Of course, in the preceding sections, we observed that the behavior of less educated (generally lower income) households may not conform to standard economic theories. Al though this finding reduces our faith in the applicability of our simulation results, it does not reverse our conclusions concerning the interest elasticity of saving. The notion that households will respond to a change in the after-tax rate of return is predicated on the assumption that households rationally antici pate and plan for future economic contingen cies. To the extent that this assumption proves incorrect, there is no particular reason to be lieve that lower income households will re spond to a change in the after-tax rate of return in the first place. Tax Policy Initiatives. Two prominent cur rent policy initiatives would reverse the direc tion of the 1986 reforms and improve tax incen tives for saving to households in higher income brackets. Family saving accounts (FSAs), pro posed by the Bush administration, would allow single individuals with adjusted gross incomes (AGI) below $60,000 and married couples with AGI below $120,000 to make contributions of up to $2500 to qualified accounts. The FSA proposal is an example of a "back-loaded" system: contributions are nondeductible, but accumulated funds are not taxed upon with drawal. An alternative proposal, the BentsenRoth "super-IRA/' would allow individuals to contribute up to $2000 to either a traditional or a back-loaded IRA.11 n On August 3, 1992, the Senate Finance Committee approved H.R. 11, the Revenue Bill of 1992. Like the BentsenRoth super-IRA, this bill would restore the deductibility of IRA contributions for all taxpayers and establish new backloaded IRAs. Contributions to back-loaded IRAs could be withdrawn without penalty after five years. The bill would also allow taxpayers to make penalty-free early withdraw als from IRAs for the purchase of a first house, for higher education expenses, for medical expenses, and for long spells of unemployment. FEDERAL RESERVE BANK OF PHILADELPHIA Do Americans Save Too Little? B. Douglas Bernheim & John Karl Scholz Unfortunately, there are sound conceptual by borrowing do not increase household sav reasons to doubt the effectiveness of extending ing. Instead, by reducing federal tax receipts, eligibility for IRA-style accounts to higher in they add to the federal budget deficit and come households. First, contributions are depress national saving. Once again, it is more capped. Under the current system, a single likely that high income households (who pos taxpayer, for example, can make no more than sess greater wealth, financial sophistication, $2000 in tax-deductible contributions. For an and access to credit markets) would engage in individual taxpayer who would have saved borrowing or asset shifting and thus defeat the more than $2000 in the absence of IRAs, the purpose of the program. availability of an IRA does not affect the costs Empirical evidence on the efficacy of IRAs is or benefits that might result from an additional mixed. Gale and Scholz (1992) find little evi dollar of saving and, therefore, provides no dence that IRAs stimulated household saving incentive on the margin for the taxpayer to between 1983 and 1986. Venti and Wise (1986, increase saving. In such cases, the IRA consti 1987, 1990, 1991) and Feenberg and Skinner tutes a "giveaway" of public funds (it reduces (1989) suggest that most IRA contributions federal tax receipts but does not promote more during this period represent net increases in saving). In addition, the IRA may actually household saving. Joines and Manegold (1991) induce the taxpayer to increase consumption, conclude that the effects of IRAs on household since it increases his or her total after-tax re saving are unlikely to be as large as the esti sources. For both of these reasons, the IRA mates of Venti and Wise and may be as small as would contribute to a lower rate of national the estimates of Gale and Scholz. saving. These concerns are of little significance An alternative proposal to promote house for low income households, since few of them hold saving, based on "premium saving ac would save more than $2000 in the absence of counts" (PSAs), is described in Bernheim and the program. It is far more likely that high Scholz (1992b). A PSA system would require income households would save more than the each taxpayer to save—in total— some fixed contribution limit. Thus, IRA-style proposals amount (the floor) before becoming eligible to may be a particularly ineffective vehicle for make contributions to a tax-favored account. providing tax incentives for saving to high The taxpayer would be eligible to contribute income households. each additional dollar of saving to the taxA second reason for doubting the effective favored account, up to some limit (the ceiling). ness of IRA-style accounts for high-income These floors and ceilings would rise with AGI households is that even if such a taxpayer and certain types of capital income. As with would not (in the absence of IRAs) have saved IRAs, capital income accrued on balances held more than the IRA contribution limit in a given in PSAs would be exempt from taxation.12* year, he or she could take full advantage of the The use of both floors and ceilings would IRA deduction either by financing contribu tions with previously accumulated assets or by borrowing. Indeed, the 1991 Tax Guide for 12With this essential structure, a PSA system could be College Teachers devotes a full page to the issue either front-loaded or back-loaded. Penalties could be "What If You're Short of Cash to Fund Your established to lock funds into tax-favored accounts for IRA?" (pp. 229-30). The Guide describes an IRS relatively short periods (e.g., seven years) or until some age close to retirement (perhaps age 591 /2). Accounts could be private letter ruling that allows households to established for specific purposes (e.g., retirement, purchase finance their IRAs by borrowing. Contribu of a house, college education), or the accounts could be tions funded either by shifting existing assets or unrestricted. 13 BUSINESS REVIEW create "windows" of program eligibility. Con sider, for example, a married couple with an AGI of $80,000. They might face a floor of $8000 and a ceiling of $12,000. Should they save less than $8000 in the corresponding tax year, they would not be eligible to make any contributions to a tax-favored account. If, on the other hand, they saved $9500, they would be eligible for favorable tax treatment on $1500. If they saved more than $12,000, they would be eligible to make the maximum contribution of $4000 (the difference between $8000 and $12,000). The most important distinctive feature of a PSA system is that floors and ceilings would vary with AGI. Eligibility windows could be positioned to maximize, within each income class, the number of households receiving tax breaks on the marginal dollar of saving. Doing so would maximize the incentive to save more. Higher-income taxpayers would not be de prived of tax incentives for saving; rather, they would simply be required to save much larger fractions of their incomes before becoming eli gible for PSAs. It would also be much more difficult for households to take advantage of tax-favored PSA accounts by shifting assets or by borrowing because eligibility would be based on total saving. An individual cannot increase his total saving by shifting assets from one account to another or by borrowing to invest.13 To implement a PSA system, one needs to measure a household's total saving. Bernheim and Scholz (1992b) propose the following mea sure:14156 Net purchases of assets (i.e., total purchases 13The administrative feasibility of monitoring total sav ing for each taxpayer is discussed in Bernheim and Scholz, 1992b. 14Many economists would define saving as the change in the stock of wealth between two points in time. If one adopts this definition, saving is very hard to measure: one would need to assess the market value of all assets every SEPTEMBER/OCTOBER1993 minus total sales) for assets on which investors receive capital gains and losses plus The January 1 to January 1 change in cash account balances (e.g., bank accounts), minus The January 1 to January 1 change in total debt (mortgages, consumer credit, etc.). In effect, saving is defined as the incremental resources that an individual sets aside in any year over and above reinvested capital gains.15,16 Now we'll evaluate the effects of three dis tinct strategies for promoting household sav ing: an IRA-like program with an AGI cap (hereafter referred to as the "standard IRA" system), an IRA-like program without an AGI cap (hereafter referred to as the "universal IRA" system), and a PSA system. We compare the cost-effectiveness of extending tax incen tives for saving to higher-income taxpayers through universal IRAs and PSAs. Sample schedules that define eligibility win dows for each level of AGI for a PSA system are given in Table 1. Separate schedules are given for married couples and single individuals. The schedules are chosen to maximize the ben year. The definition used in the text represents a compro mise between economic logic and administrative feasibil ity. 15Note that it is possible to compute this measure of saving without assessing the value of unrealized capital assets, since, by definition, unrealized gains are fully rein vested. 16If this definition of saving is employed, it is also impor tant to adjust each taxpayer's eligibility floors and ceilings upward by the amount of capital income other than capital gains. See Bernheim and Scholz, 1992b, for a detailed discussion of this issue. FEDERAL RESERVE BANK OF PHILADELPHIA Do Americans Save Too Little? B. Douglas Bernheim & John Karl Scholz eficial effects of the program within each popu lation subgroup.17 To facilitate comparisons with IRAs, we have adopted window widths of $2000 per year for single households, $2250 per year for married couples with one earner, and $4000 per year for married couples with two earners. For example, a dual-earner married 17Note that the floor rises with income at different rates for married couples (16.7 cents for each dollar of income over $34,000) and single individuals (34 cents for each dollar of income over $42,000). Since actual patterns of saving differ by marital status, different schedules must be used to maximize the beneficial effects of the program. couple with an AGI of $30,000 and no capital income would have a floor of $0 and a ceiling of $4000 (Table 1). In contrast, a couple with an AGI of $120,000 and dividend and interest income of $2000 would have a floor of $16,362 (.167 x $86,000 + $2000) and a ceiling of $20,362. The standard and universal IRA systems differ from the PSA proposal in that they anchor the eligibility window at $0 for all in come classes and make no adjustment for capi tal income. The standard IRA system phases out deductible contributions for married couples with incomes between $40,000 and $50,000 and for single taxpayers with incomes between TABLE l a Deductible Contribution Formula Married Couples If your income is Deductible Qualified Contribution Floor (Added to Capital Income) Deductible Qualified Contribution Ceiling (Added to Floor) Less than $34,000 0 $2250 or $4000 .167x (Income-34,000) $2250 or $4000 Greater than $34,000 Single Households If your income is Deductible Qualified Contribution Floor (Added to Capital Income) Deductible Qualified Contribution Ceiling (Added to Floor) Less than $42,000 0 $2000 .34 x (Income-42,000) $2000 Greater than $42,000 aFor the purpose of comparison with IRAs, married couples with one earner are allowed to contribute $2250 and married couples with two earners can contribute $4000. In the actual implementation of this proposal we see no compelling reason to make this distinction. 15 BUSINESS REVIEW $25,000 and $35,000.18 The universal IRA sys tem allows all households to make deductible contributions.19 We compare these plans on the basis of three criteria. The first criterion is a measure of effectiveness. Specifically, for each plan, we estimate the number of households that would receive a higher after-tax rate of return on the incremental dollar of saving. We refer to these households as the IMPACT GROUP. Our sec ond criterion is a measure of wasteful subsidi zation. Specifically, for each plan, we estimate the number of households that would make the maximum eligible contribution to a tax-favored account w hile continuing to receive the unsubsidized after-tax rate of return on the incremental dollar of investment. We refer to these households as the NO-IMPACT GROUP. Our third criterion is also a measure of wasteful subsidization: we calculate the budgetary cost of subsidizing the NO-IMPACT GROUP. We refer to this cost as the GIVEAWAY. Our calculations are once again based on data obtained from the SCF for 1983 and 1986. The interested reader is referred to Bernheim and Scholz (1992b) for details. Compare the effects of the policies on mar ried couples as shown in Table 2. The top panel shows the size of the IMPACT GROUP. Over all, the PSA system provides real incentives to 2.4 million couples, roughly 90 percent more than the IRA with AGI restrictions and 30 percent more than the universal IRA. The 18It should be noted that the current IRA system differs from the standard IRA system considered in the text in that it phases out deductible contributions only for households that are covered by private pension plans. The current system is, therefore, a blend of a standard system and a universal IRA system. 19The IRA-like proposals we simulate are superior to actual IRA schemes because, in practice, IRA schemes are susceptible to tax arbitrage strategies involving borrowing and asset shifting, which our simulations do not capture. Digitized for 16 FRASER SEPTEMBER/OCTOBER1993 difference is particularly pronounced in the top income quintile. By definition, the IRA with AGI caps ignores these households. Relative to the universal IRA, the PSA increases the num ber of couples receiving marginal incentives in the top income quintile by nearly 125 percent. Since, in this sample, over 60 percent of positive household saving is attributable to households in the top quintile of the income distribution, this improvement is particularly important. The bottom two panels of Table 2 measure the NO-IMPACT GROUP and the cost of these ineffective subsidies. The calculations show, for example, that the PSA system would reduce the number of households in the NO-IMPACT GROUP by 1.75 million (28.2 percent) and would reduce federal expenditures on ineffective sub sidies by $2.0 billion (34.0 percent), relative to the universal IRA. In terms of cost-effective ness, the PSA system increases the ratio of the IMPACT GROUP to the GIVEAWAY by 96.5 percent overall, and by 287.2 percent (that is, by a factor of almost four) in the top income quintile. The IRA with AGI caps also effec tively reduces ineffective subsidies and bud getary cost, but it achieves this reduction by excluding the very households most likely to respond to tax incentives. Note the results for single individuals (Table 3). Under a PSA system, the size of the IMPACT GROUP would increase significantly relative to other proposals. The size of the IMPACT GROUP in the highest income quintile would more than triple. Moreover, both the size of the NO-IMPACT GROUP and the GIVEAWAY would fall relative to the universal IRA. The result is a 49.7 percent increase in overall costeffectiveness (the ratio of the IMPACT GROUP to GIVEAWAY), and a 551.3 percent increase in cost-effectiveness for the top income quintile, relative to the universal IRA proposal. Other Initiatives. Pension policies and tax policies do not exhaust the full range of strate gies for stimulating personal saving. One par ticular class of policies not discussed here merFEDERAL RESERVE BANK OF PHILADELPHIA B. Douglas Bernheim & John Karl Scholz Do Americans Save Too Little? TABLE 2a A Comparison of Three Saving-Incentive Proposals, Married Couples IRA w/AGI Cap Universal IRA PSA IMPACT GROUP (in 1000s) Highest Income Quintile Full Population 0 1256 102 1840 228 2388 NO-IMPACT GROUP (in 1000s) Highest Income Quintile Full Population 0 3578 1416 6218 817 4467 ANNUAL GIVEAWAY (in $ millions) Highest Income Quintile Full Population 0 2006 1950 5861 1119 3870 COST-EFFECTIVENESS (ratio of IMPACT group to GIVEAWAY) Highest Income Quintile Full Population — .3510 .0523 .3139 .2038 .6171 Simulated Effect Simulations use data from the 1983-86 Survey of Consumer Finances. Saving and column headings are defined in the text. The PSA schedule is given in Table 1. its further attention. An accumulating body of evidence, including that contained in this ar ticle, suggests that the behavior of many house holds (particularly those with lower incomes) is not well described by traditional economic theories. To some, saving decisions appear to be governed by such factors as habit, mental accounting, and self-control. Consequently, it may be possible to design more effective poli cies by educating the population or by exploit ing the psychology of saving. The Japanese appear to have had considerable success with such a strategy during the postwar period (Horioka, 1988, and Bernheim, 1991). The de velopment of a framework for analyzing poli cies of this sort is an important research prior ity. Bernheim (1993) provides a preliminary analysis of these issues. CONCLUSION The evidence presented in this article sup ports the view that many Americans, particu larly those without a college education, save too little. Our analysis indicates that it should 17 SEPTEMBER/OCTOBER 1993 BUSINESS REVIEW TA B L E 3a A Comparison of Three Saving-Incentive Proposals r Single Taxpayers IRA w/AGI Cap Universal IRA PSA IMPACT GROUP (in 1000s) Highest Income Quintile Full Population 0 454 40 603 134 694 NO-IMPACT GROUP (in 1000s) Highest Income Quintile Full Population 0 1078 350 1405 197 1155 ANNUAL GIVEAWAY (in $ millions) Highest Income Quintile Full Population 0 460 292 845 151 650 COST-EFFECTIVENESS Highest Income Quintile Full Population .9870 .1370 .7136 .8874 1.0677 Simulated Effect Simulations use data from the 1983-86 Survey of Consumer Finances. Saving and column headings are defined in the text. The PSA schedule is given in Table 1. be possible to increase total personal saving among lower income households by encourag ing the formation and expansion of private pension coverage for such families. It is doubt ful that favorable tax treatment of capital in come would stimulate significant additional saving by this group. Conversely, the expan sion of private pensions would probably have little effect on saving by higher income house holds. However, these households are more likely to increase saving significantly in re sponse to favorable tax treatment of capital income. These findings imply that the design of Digitized for 18 FRASER the current system, which links eligibility for IRAs to an AGI cap, and which provides higher income households with more complete pen sion coverage, ensures a minimal impact on personal saving. Extending tax incentives for saving to higher income households is problematic. We have discussed two competing options: the univer sal IRA and the premium saving account (PSA). Our analysis reveals that the PSA system is a more cost-effective vehicle for providing in centives to those households most likely to respond to tax incentives. FEDERAL RESERVE BANK OF PHILADELPHIA REFERENCES Do Americans Save Too Little? B. Douglas Bernheim & John Karl Scholz Avery, Robert B., and Gregory E. Elliehausen. "1983 Survey of Consumer Finances: Technical Manual and Codebook," mimeo, Board of Governors of the Federal Reserve System, August 1988. Avery, Robert B., and Arthur B. Kennickell. "1986 Survey of Consumer Finances: Technical Manual and Codebook," mimeo, Board of Governors of the Federal Reserve System, November 1988. Bernheim, B. Douglas. The Vanishing Nest Egg: Reflections on Saving in America. Twentieth Century Fund, 1991. Bernheim, B. Douglas. "Is the Baby Boom Generation Preparing Adequately for Retirement? Summary Report," mimeo, Princeton University, August 1992a. Bernheim, B. Douglas. "Is the Baby Boom Generation Preparing Adequately for Retirement? Technical Report," mimeo, Princeton University, August 1992b. Bernheim, B. Douglas. "W hat Determines Personal Saving? The Roles of Information and Economic Literacy," in Tax Policy for Economic Growth in the 1990s. Washington, D.C.: American Council for Capital Formation (forthcoming 1993). Bernheim, B. Douglas, and John Karl Scholz. "Private Saving and Public Policy," mimeo, Princeton University and the University of Wisconsin— Madison, September 1992a. Bernheim, B. Douglas, and John Karl Scholz. "Premium Saving Accounts: A Proposal to Improve Tax Incentives for Saving," mimeo, Princeton University and the University of Wisconsin—Madison, September 1992b. Bernheim, B. Douglas, and John B. Shoven. "Pension Funding and Saving," in Zvi Bodie, John B. Shoven, and David A. Wise, eds., Pensions in the U.S. Economy. Chicago: University of Chicago Press and NBER, 1988, pp. 85-111. Boskin, Michael. "Taxation, Saving, and the Rate of Interest," Journal of Political Economy (April 1978), pp. 3-27. Cutler, David M., and Lawrence F. Katz. "Rising Inequality? Changes in the Distribution of Income and Consumption in the 1980s," mimeo, Harvard University, January 1992. Diamond, Peter A. "A Framework for Social Security Analysis," Journal of Public Economics (December 1977), pp. 275-98. Feenberg, Daniel R., and Jonathan Skinner. "Sources of IRA Saving," in Lawrence Summers, ed., Tax Policy and the Economy, Cambridge, MA: MIT Press, 1989, pp. 25-46. Gale, William G., and John Karl Scholz. "IRAs and Household Saving," mimeo, UCLA and University of Wisconsin—Madison, 1992. Hall, Robert. "Intertemporal Substitution in Consumption," Journal o f Political Economy 96, 1988, pp. 337-57. 19 REFERENCES (continued) BUSINESS REVIEW SEPTEMBER/OCTOBER 1993 Hamermesh, Daniel S. "Consumption During Retirement: The Missing Link in the LifeCycle," Review of Economics and Statistics (February 1984), pp. 1-7. Hatsopoulos, George N., Paul R. Krugman, and James M. Poterba. Overconsumption: The Challenge to U.S. Policy. Washington, D.C.: American Business Conference, 1989. Horioka, Charles Y. "W hy Is Japan's Private Saving Rate So High?" in R. Sato and T. Negishi, eds., Recent Developments in Japanese Economics. Tokyo: Harcourt Brace Jovanovich Japan/Academic Press, 1988. Joines, Douglas H., and James G. Manegold. "IRA and Saving: Evidence from a Panel of Taxpayers," mimeo, U.S.C., 1991. Kotlikoff, Laurence J., Avia Spivak, and Lawrence H. Summers. "The Adequacy of Saving," American Economic Review (December 1982), pp. 1056-69. Meyer, Stephen A. "Saving and Demographics: Some International Comparisons," this Business Review, March/April 1992, pp. 13-23. Organization for Economic Cooperation and Development, Economic Outlook, 1992. Shefrin, Hersh M., and Richard H. Thaler. "The Behavioral Life Cycle Hypothesis," Economic Inquiry 26 (October 1988), pp. 609-43. Summers, Lawrence. "Capital Taxation in a Life Cycle Growth Model," American Economic Review (September 1981). Summers, Lawrence. "Issues in National Savings Policy," National Bureau of Economic Research Working Paper 1710, Cambridge, MA (September 1985). Venti, Steven F., and David A. Wise. "Tax-Deferred Accounts, Constrained Choice and Estimation of Individual Saving," Review of Economic Studies, LIII, 1986, pp. 579-601. Venti, Steven F., and David A. Wise. "IRAs and Saving," in Martin Feldstein, ed., The Effects of Taxation on Capital Accumulation. Chicago: University of Chicago Press and NBER, 1987, pp. 7-48. Venti, Steven F., and David A. Wise. "Aging, Moving, and Housing Wealth," in David A. Wise, ed., The Economics of Aging. Chicago: University of Chicago Press and NBER, 1989. Venti, Steven F., and David A. Wise. "Have IRAs Increased U.S. Saving?: Evidence from Consumer Expenditure Surveys," Quarterly Journal of Economics 105 (August 1990), pp. 661-98. Venti, Steven F., and David A. Wise. "The Saving Effect of Tax-Deferred Retirement Accounts: Evidence from SIPP," in B. Douglas Bernheim and John B. Shoven, eds., National Saving and Economic Performance. Chicago: University of Chicago Press and NBER, 1991, pp. 103-28. Digitized for 20 FRASER FEDERAL RESERVE BANK OF PHILADELPHIA Highways and Education: The Road to Productivity? Gerald A. Carlino* rom 1948 to 1969, output per hour worked grew at an average rate of 2.5 percent per year. From 1969 to 1987, growth of labor productiv ity slowed to 1.1 percent per year. Economists and policymakers have acknowledged that the slowdown in productivity growth is one of the major economic problems facing the United States because sluggish productivity growth means slower growth in our standard of living. The decline in investment in public infrastruc ture and the decline in educational quality may have played a role in this slowdown. Growth of real government spending on nonmilitary * Gerald A. Carlino is an economic adviser in the Regional and Urban Section of the Philadelphia Fed's Research De partment. public infrastructure declined from an annual rate of 4.1 percent between 1948-69 to only 1.6 percent during 1969-87. There is also some indication that educational quality may have slipped over time as witnessed by the fact that Scholastic Aptitude Test (SAT) scores have been declining since the mid-1960s.1 The current Administration would like to increase national productivity by, among other things, increasing investment in public infra structure and by creating j ob training programs to improve the quality of the work force. Would 'The data reported in this paragraph are taken from Alicia H. Munnell, "Why Has Productivity Growth De clined? Productivity and Public Investment," New England Economic Review, January/February 1990a, pp. 3-22. 21 BUSINESS REVIEW programs such as these improve productivity and ultimately the level of output? Differences across states in investment in public infrastructure and education provide insight into the likely effects of national spend ing in these areas. A number of recent studies have looked at the impact of public infrastruc ture and educational attainment on output at the state and local levels. Studies have found that increases in highway density and educa tional attainment improve a region's produc tivity and boost output. A recent study by Carlino and Voith found that a 10 percent increase in educational attainment of a state's residents boosts its output by 8 percent, and a 10 percent increase in highway density increases state output by 1.4 percent.2 REASONS PRODUCTIVITY DIFFERS ACROSS STATES Productivity measures the ratio of output to inputs such as land, labor, and capital. If two regions used the same quantities of inputs, output would be greater in the more produc tive region. One region might have higher productivity than another because the quality of inputs is higher. Regional productivity de pends not only on the number of machines used to produce an output but also on their age, technical quality, and degree of utilization. Regional productivity may also depend on the scale at which production takes place within a region's firms. As firms increase their size, they can sometimes increase productivity by having their workers specialize in particular tasks or by using their capital equipment more effi ciently. These internal factors may vary from one region to another and therefore may influ ence regional productivity.3 While these inter nal factors are an important source of produc- 2Gerald A. Carlino and Richard Voith, "Accounting for Differences in Aggregate State Productivity," Regional Sci ence and Urban Economics, 2, December 1992, pp. 597-617. Digitized for 22FRASER SEPTEMBER/OCTOBER 1993 tivity differentials across regions, this article focuses on public infrastructure and the quality of the region's work force, factors that are external to the firm but which influence produc tivity in a market or region. Before we look at how much public infrastructure and work force quality matter for productivity, we need to understand other external factors that affect productivity, such as a region's industry mix and the degree of urbanization, so that we can control for their effects. Industry Mix. Regional differences in pro ductivity arise partly because individual re gions often specialize in the mix of goods or services they produce. For instance, the grow ing of wheat and corn tends to be concentrated in the Plains states. Because many of the states in the Northeast and Midwest have historically specialized in the production of manufactured goods, this broad geographic area is commonly referred to as the "industrial belt" or "indus trial core." Since some industries are more productive than others, regions with a rela tively large concentration of the more produc tive industries will have greater overall pro ductivity than regions with a concentration of the less productive industries.4 Urbanization Economies. Just as a region's industry mix can influence its productivity, the 3These internal decisions by firms may be influenced by external factors. For example, the size of a region's market (external factor) may influence a regional firm's scale of operation (internal factor). 4Baumol, Blackman, and Wolff looked at national pro ductivity growth by industry during the 1947-86 period. They found that productivity growth does differ by indus try. They also reported that the traditional high-productivity-growth industries continued to perform well during the 1947-86 period, implying long-term differences in the level of productivity across industries. See William J. Baumol, Sue Anne Batey Blackman, and Edward N. Wolff, Productiv ity and American Leadership: The Long View (Cambridge, MA: The MIT Press, 1989). FEDERAL RESERVE BANK OF PHILADELPHIA Highways and Education: The Road to Productivity? percentage of a region's firms that are located in metropolitan areas also affects its productiv ity. Metropolitan areas offer their firms access to a common pool of trained labor, so that firms not only share the cost of training new workers, but any firm can vary its work force without incurring lost productivity during training pe riods or by carrying idle workers. Metropoli tan locations also help firms by providing whole saling facilities that reduce the level of invento ries any one firm needs to keep on hand and by providing access to accounting, data process ing, legal, financial, and other specialized busi ness services. Firms located in nonmetropolitan areas would need to employ people who pro vide these specialized business services on a full-time basis or else spend considerable time and money bringing them from a distance when they are needed. By locating in a metropolitan area firms can contract for these on an asneeded basis. Economists refer to the advantages offered by metropolitan areas as urbanization econo mies. These urbanization economies should increase the productivity of urban firms. Thus, other things being equal, the more urbanized regions should have greater productivity than less urbanized regions. In other words, with fewer inputs metropolitan firms can produce the same level of goods and services as nonmetropolitan firms. Urbanization economies can increase firms' productivity only up to a point. Urbanization brings not only greater productivity but also greater problems, such as congestion, that even tually balance or outweigh the efficiency gains from urbanization. At some point, increases in the number of people and firms residing in a metropolitan area clog its roads and transpor tation network and raise the average time and cost of transporting goods and commuting either to work or to leisure activities. In addi tion, as a metropolitan area grows, its bound aries may spread out, which increases both the time and distance of the average commute. Gerald A. Carlino When urban size becomes a hindrance rather than a help, firms experience urbanization diseconomies. Urbanization economies are balanced by these diseconomies, suggesting that there may be some optimal degree of urbanization. Individual firms that have incentives to ex ploit urbanization economies are guided by the "invisible hand" of the marketplace to locate in metropolitan areas. Local policymakers can lend a hand to lessen the negative consequences of congestion by providing public infrastruc ture, such as highways, airports, and mass transit facilities, that link a region's labor and product markets with one another and with those of other regions. Public Infrastructure. Some economists believe that an increase in the capital stock of the public sector leads directly to increases in private sector output because public infrastruc ture is an essential input in the production of private output.5 For example, driver produc tivity increases when a good highway system allows truck drivers to avoid circuitous back roads and to bring supplies to a firm and goods to market more quickly. Similar arguments can be made for the public provision of police and fire protection, water supply facilities, airports, and mass transit. An increase in the public capital stock, like an increase in any factor of production, directly increases private sector output.6 Of course, some public sector spending may actually substitute for private sector spending. This would be the case if close substitutes for 5For a useful survey of the recent literature, see John A. Tatom, "Should Government Spending on Capital Goods Be Raised?" Review, Federal Reserve Bank of St. Louis, March/April 1991, pp. 1-15; and Randall Eberts, "Public Infrastructure and Regional Economic Development," Eco nomic Review, The Federal Reserve Bank of Cleveland (First Quarter 1990a), pp.15-27. 6Munnell (1990a); see footnote 1 for complete citation. 23 BUSINESS REVIEW publicly provided services are available from the private sector.7 Public finance theory tells us, however, that most public sector spending should be for goods and services that would be either not provided or underprovided if left to the private sector. For example, private com panies could build roads and bridges and charge tolls for using them. But private provision may not be efficient. Although there is a large initial fixed cost associated with construction of bridges and highways, once constructed, the ad d itional cost of one m ore vehicle on uncongested roads is nearly zero. In this case, economic efficiency requires setting a zero price for use of uncongested roads. Thus, while it is possible to exclude those unwilling to pay for the use of infrastructure, such exclusion often is inefficient.8 In such cases, the public sector should provide infrastructure. Labor-Force Characteristics. Policymakers in state and local government in the U.S. have a great deal of influence on the quality of the work force because their policies affect the cost and quality of the public education system. Studies have shown that higher educational attainment of a region's labor force is an impor tant contributor to higher regional productiv ity.9 These investments in human capital may 7Studies have found that labor and public capital are complements in production, while there appears to be some degree of substitutability between public capital and pri vate capital. See Jose da Silva Costa, Richard W. Ellson, and Randolph C. Martin, "Public Capital, Regional Output and Development: Some Empirical Evidence," Journal o f Re gional Science, 27 (1987), pp. 419-37; and Alicia H. Munnell, "How Does Public Infrastructure Affect Regional Economic Performance?" New England Economic Review, September/ October 1990b, pp. 11-33. Munnell finds that highways and streets appear to be substitutes for private capital and speculates that well-maintained roads reduce wear and tear on commercial vehicles, lowering private sector main tenance and replacement of these vehicles. 8See Eberts (1990a; see footnote 5 for complete citation), for a discussion of the public goods aspects of public inputs. SEPTEMBER/OCTOBER 1993 lead to increased regional productivity because education introduces a region's workers to new techniques and skills. Since educational attainment differs across regions, these differ ences can lead to variations in regional produc tivity. THE EVIDENCE Studies on regional productivity have tended to limit their focus to specific aspects of re gional productivity. A number of studies since the mid-1970s have looked at the impact of urbanization economies on manufacturing pro ductivity at the regional level. These studies have shown that manufacturing productivity in general increases with metropolitan popula tion size (a proxy for urbanization economies), at least over the observed ranges of metropoli tan sizes.10 Another group of regional produc tivity studies has examined the role of public infrastructure in regional production, and most studies find that greater investment in public capital does raise regional productivity.11 9See, for example, Gerald A. Carlino and Edwin S. Mills, "The Determinants of County Growth," Journal o f Regional Science, 27 (1992), pp. 39-54. 10For a survey of this literature, see Ronald Moomaw, "Spatial Productivity Variations in Manufacturing: A Criti cal Survey of Cross-Sectional Analysis," International Re gional Science Review, 8 (1983), pp. 1-22. n See Randall Eberts, "Estimating the Contribution of Urban Public Infrastructure to Regional Economic Growth," Working Paper 9004, Federal Reserve Bank of Cleveland (May 1990b). While Eberts concentrates on the influence of public capital on manufacturing output, an article by Alicia Munnell and one by Teresa Garcia-Mila and Therese J. McGuire extend the analysis of public infrastructure to aggregate output at the state level. See Munnell (1990b; footnote 7 has complete citation); and Teresa Garcia-Mila and Therese J. McGuire, "The Contribution of Publicly Provided Inputs to States' Economies," Regional Science and Urban Economics, 22 (1992), pp.229-41. Both studies find that public infrastructure has positive effects on aggregate pro ductivity at the state level. FEDERAL RESERVE BANK OF PHILADELPHIA Highways and Education: The Road to Productivity? The examination of each of these factors in isolation can result in misleading conclusions. For example, the contribution of public infra structure to regional productivity may be over stated if the other factors thought to influence regional productivity are not taken into consid eration. The clustering of firms in metropolitan areas creates urbanization economies, which, in turn, increases a region's overall productiv ity and output. More output leads to increased tax revenue for state and local governments. Some of the increased tax revenue may be used to supply public infrastructure. Perhaps it is urbanization economies that largely contribute to regional productivity, and public infrastruc ture contributes to a much lesser extent or not at all. Since increased urbanization economies lead to more output, which, in turn, leads to more public infrastructure, studies that look at the role of public infrastructure on regional productivity but fail to control for urbanization economies run the risk of overstating the rela tive importance of public capital.12 The Carlino and Voith study provides a more comprehensive view of the factors affect ing state productivity by considering the rela tive importance of industry mix, urbanization economies, public infrastructure, and labor quality on aggregate production at the state level during the 1967-86 period (see Appendix, page 30).13 12An unresolved issue is whether public capital pre cedes private capital formation or vice-versa. There is evidence that the formation of public capital and private capital is a simultaneous process. See Eberts (1990a; foot note 5 has complete citation). 13Carlino and Voith (1992; see footnote 2) used multiple regression analysis to examine the relative importance of industry mix, labor-force quality, urbanization economies, and infrastructure on state aggregate productivity. One problem with analyzing the results from a multiple regres sion analysis is that the variables are generally measured in different units. For example, educational attainment is measured in years, and public infrastructure is measured in Gerald A. Carlino Industry Mix. Carlino and Voith measured industry mix by the share of state output attrib utable to each of the nine major industry group ings.14 By including these industry-mix vari ables, their study controlled for industrial struc ture differences across states, which helped to isolate the effects of the other variables thought to have independent effects on state productiv ity. Carlino and Voith found that state produc tivity varies a lot, running from about 50 per cent above the national average in Delaware to about 35 percent below average in Wyoming. They also found that controlling for industry mix alone explains about 26 percent of the variation (see Industry Mix Is an Important Com ponent o f a Region's Aggregate Productivity). Urbanization Economies. The Carlino and Voith study used the percent of a state's popu lation that is metropolitan to capture the effects of urbanization economies. The percent of the population living in metropolitan areas varied widely across states in 1984; for example, it is as low as 14.7 percent in Wyoming and as high as 100 percent in New Jersey.15The positive effects terms of highway density. To facilitate the comparison of the effects of different variables, we must standardize our find ings. A common approach couches relationships in per centage terms—the percent change in one variable associ ated with the percent change in another. This unitless measure is known as an elasticity. The elasticity for state output tells us the percent change in state output given a percentage change in any of the explanatory variables, while holding all other explanatory variables constant. 14These groupings are agriculture; mining; construction; manufacturing; transportation, communication, and public utilities; trade (wholesale and retail combined); finance, insurance, and real estate (FIRE); services; and government. Since the industry shares of state output sum to one, it is necessary to drop the percentage share of one of the indus tries. Although agriculture is the excluded industry in the Carlino and Voith study, the study could just as easily have excluded any one of the other industries. 15Every county in New Jersey is part of a metropolitan area even though large parts of some counties are rural. 25 Industry Mix Is an Important Component of a Region's Aggregate Productivity BUSINESS REVIEW SEPTEMBER/OCTOBER 1993 The estimates of total factor productivity from the Carlino/Voith study can be used to compare aggregate productivity across states by looking at the ratio of productivity in a state relative to productivity averaged across all states. If productivity in a state is equal to the national average, the ratio would equal one. If the state is more productive than the average state, the ratio would be greater than one. And the ratio is less than one if the state is less productive than the average state. State productivity varies from about 50 percent above the national average (48-state average) in Delaware to about 35 percent below the national average in Wyoming (see Table). Even with the exclusion of Delaware, there is a 58 percent differential between Rhode Island, the second most productive state, and Wyoming, the least productive state. But controlling for industry mix alters the picture substantially. Industry Mix. Total productivity was recalculated for each state, controlling for industry mix differences across states by assigning the national industry mix to each state. Controlling for industrial structure reduces the differential in total productivity across states by 26 percent. The differential in state productivity runs from about 43 percent above the national average (compared with 50 percent above average before standardization) to 19 percent below the national average (compared with 35 percent below before standardization). Of the 16 states in the top one-third of the productivity distribution before standardization, 13 states remain in the top one-third after standardization. Indiana, Maine, and Massachusetts, which were in the top one-third before standardization, moved to the middle third after standardization. Three states, Louisiana, Oklahoma, and New Mexico, were in the bottom one-third before standard ization but moved to the top one-third after standardization. Wyoming is an interesting example of how industry mix can affect a state's productivity in that it moves from being 35 percent below the U.S. before standardization to just about at the national average after controlling for industrial structure. A relatively large portion of total employment in Wyoming is in the extractive industries, especially oil and gas. Mining employment in Wyoming accounted for 22 percent of total employment in 1980, compared with only one percent nationally. Wyoming also tends to be much less manufacturing oriented. In 1980, only 6 percent of total employment in Wyoming was accounted for by manufacturing, compared with 28 percent nationally. One recent study shows that while productivity in the mining industry fell dramatically during the period 1947-86, it improved slightly in manufacturing.3 aWilliam J. Baumol, Sue Anne Batey Blackman, and Edward N. Wolff, Productivity and American Leadership: The Long View (Cambridge, MA: MIT Press, 1989). Digitized for 26FRASER FEDERAL RESERVE BANK OF PHILADELPHIA Highways and Education: The Road to Productivity? Gerald A. Carlino Aggregate Productivity Differences Across States3 Total Controlling for Industry Mix Total Controlling for Industry Mix 1 Delaware 1.5002 1 1.4338 25 Nevada 1.0210 42 0.9058 2 Rhode Island 1.2282 2 1.1886 26 Arkansas 1.0178 19 1.0149 3 South Carolina 1.2081 3 1.883 27 Maryland 1.0177 31 0.9660 4 Georgia 1.1833 12 1.0603 28 Arizona 0.9953 36 0.9350 5 Connecticut 1.1685 6 1.1210 29 Oregon 0.9867 39 0.9144 6 New Hampshire 1.1456 10 1.0723 30 Virginia 0.9751 26 0.9908 7 Tennessee 1.1230 13 1.0595 31 Iowa 0.9687 25 0.9922 8 Vermont 1.1156 14 1.0591 32 Mississippi 0.9653 20 1.0127 9 Indiana 1.1061 17 1.0481 33 Colorado 0.9364 40 0.9102 10 West Virginia 1.1007 5 1.1609 34 Florida 0.9343 46 0.8332 11 Missouri 1.0949 29 0.9780 35 Washington 0.9289 44 0.8700 12 North Carolina 1.0913 11 1.0603 36 Kansas 0.9274 32 0.9642 13 Alabama 1.0868 9 1.0730 37 Nebraska 0.9250 38 0.9184 14 Massachusetts 1.0864 24 0.9997 38 Idaho 0.9138 41 0.9089 15 Michigan 1.0832 16 1.0506 39 Oklahoma 0.9061 7 1.1194 16 Maine 1.0830 21 1.0075 40 Utah 0.9022 43 0.8881 17 Illinois 1.0788 27 0.9870 41 Texas 0.8433 28 0.9816 18 Ohio 1.0662 22 1.0005 42 North Dakota 0.8409 35 0.9426 19 Wisconsin 1.0515 18 1.0324 43 California 0.8285 48 0.8090 20 New York 1.0450 34 0.9528 44 Louisiana 0.8236 4 1.1647 21 Pennsylvania 1.0386 33 0.9533 45 South Dakota 0.8143 45 0.8626 22 New Jersey 1.0381 37 0.9350 46 New Mexico 0.7989 15 1.0531 23 Minnesota 1.0337 30 0.9765 47 Montana 0.7744 47 0.8296 24 Kentucky 1.027 8 1.0880 48 Wyoming 0.6457 23 0.9998 aIndex represents ratio of aggregate productivity in each state to the national average. 27 BUSINESS REVIEW SEPTEMBER/OCTOBER 1993 of increased urbanization make up one side of the urban size ledger. The negative effects of congestion brought on by increased urbaniza tion make up the other. Thus, Carlino and Voith allowed for the fact that increasing the degree of urbanization would increase pro ductivity up to a point, after which productiv ity would decrease.16 Both forces influence productivity: increased urbanization encour ages growth, and increased congestion dis courages it. Carlino and Voith found that the positive effects of urbanization economies are greatest when roughly half of a state's popula tion is metropolitan.17 Infrastructure. A state can mitigate the effects of congestion by building and main taining streets and highways. The Carlino and Voith study employed highway density (high way miles per square mile of land area in a state) as a proxy for state infrastructure, partly because of the relative importance of high ways and partly because data for the other categories of public capital are generally not available.18 The study found that state produc tivity responds to the availability of a highway network. A 10 percent increase in a state's highway density leads, on average, to a 1.4 percent increase in total output. The Carlino and Voith study corroborates the findings re ported in several recent studies in terms of the importance of infrastructure spending on state output. One study, by Garcia-M ila and McGuire, employed annual expenditures on highways by state and local governments dur ing 1969-83 as a measure of public sector capi tal. The study found that a 10 percent increase in highway spending results in a 0.7 to 1.7 percent increase in aggregate state output.19 A study by Munnell, using a broader measure of infrastructure than the one employed by Carlino and Voith, found that a 10 percent increase in infrastructure led to a 1.5 percent increase in aggregate state output during the 1970-86 period.20 The similarity of the findings among the three studies supports the concept of public infrastructure spending as a public policy instrument for fostering productivity growth at the state level.21 16T o capture the effects of congestion Carlino and Voith took the percent of a state's population that is metropolitan and squared it. This follows William BaumoTs reasoning that if each resident of a metropolitan area imposes exter nal costs on every other, and if the magnitude of the cost borne by each resident is roughly proportional to a metro politan area's population size, then since these costs are borne by each of R residents involved, the total external cost will increase not with R but with R2. See William J. Baumol, "Macroeconomics of Unbalanced Growth: The Anatomy of Urban Crisis," American Economic Review, Vol.57 (1967), pp. 415-26. Most of this infrastructure consists of assets owned by state and local governments. The largest single item is highways and streets, which account for 39 percent of total state and local wealth. See Munnell (1990b; footnote 7 has complete citation). 17Of course, factors other than percent of a state's popu lation that is metropolitan can influence the urbanization economies states offer. For example, urbanization econo mies may spill over state boundaries so that states that are not highly urbanized may benefit from urbanization econo mies if they are near highly urbanized states. 18In 1988 nonmilitary infrastructure amounted to $2 trillion, compared with $4.4 trillion in private capital. 19Garcia-Mila and McGuire (1992; footnote 11 has com plete citation). 20Munnell (1990b; footnote 7 has complete citation). Munnell found that an additional dollar of public infra structure spending yielded the same increase in aggregate state output as an additional dollar spent on private capital. Munnell used the stock of state and local public capital, which includes highways and streets, water and sewer systems, buildings (schools, hospitals, etc.), and equipment. The results of this study are somewhat controversial. See John A. Tatom, "Public Capital and Private Sector Perfor mance," Review, Federal Reserve Bank of St. Louis, M ay/ June 1991, pp. 3-15; and Alicia H. Munnell, "Infrastructure Investment and Economic Growth," Journal o f Economic Perspectives, 16, Fall 1992, pp. 189-198. FEDERAL RESERVE BANK OF PHILADELPHIA Highways and Education: The Road to Productivity? Gerald A. Carlino Labor-Force Characteristics. Differences in labor-force composition—education, experi ence, degree of unionization—across states can result in differences in aggregate productivity. The Carlino and Voith study uses educational attainment, defined as the percent of a state's population that is 25 years old and over with 12 or more years of schooling, as its measure of labor-force quality. The percent of a state's 25and-over population with at least a high school diploma varies widely across the United States; for example, in 1980 it was as low as 53 percent in Kentucky and as high as 80 percent in Ari zona. Carlino and Voith's results indicate that a 10 percent increase in educational attainment leads, on average, to an 8 percent increase in aggregate output.*22 This finding suggests that education is an important public policy instru ment for promoting productivity growth at the state level.23 21The magnitude of the effect of public infrastructure on state level output is about half as large as that found for the national economy. For example, Aschauer found that a 10 percent increase in the stock of public capital led to a 3.9 percent increase in national output. See David A. Aschauer, "Is Public Expenditure Productive?" Journal o f Monetary Economics, 23, March 1989, pp. 177-200. When one state adds to its stock of public infrastructure, this increased investment most likely has a beneficial effect on the output of neighboring states. For example, the opening of Interstate 476 in Pennsylvania in 1992 not only made Pennsylvania's workers more productive, but it may have improved the productivity of workers in Delaware and New Jersey as well. For a general critique of Aschauer's findings, see Laura Rubin, "Productivity and the Public Capital Stock: Another Look," Working Paper No. 118, Board of Gover nors of the Federal Reserve System, May 1991. CONCLUSION The research summarized in this article sup ports the view that increased infrastructure spending and greater educational attainment do improve productivity and ultimately the level of output. Further research should help determine the relative effects of additional spending on infrastructure and education. But the findings so far suggest that state govern ments should pay close attention to investment in public capital and to the level of educational attainment of their workers. 22Of course, more productive workers may place a higher value on educational attainment. To some extent, therefore, productivity and educational attainment may be a simulta neous process. 23Factors other than those discussed here could affect state productivity, including state policies and regulations, the degree of unionization, research and development spend ing, and technical progress. While these factors may deter mine differences in state productivity, few, if any, data are available to determine the relative importance of these omitted variables. 29 APPENDIX BUSINESS REVIEW SEPTEMBER/OCTOBER 1993 A state's output of goods and services depends on the quantities of inputs, such as capital and labor, and on the productivity of those inputs. The relationship among output, inputs, and productivity is given in the following production function: Q = AF(K, L) Accordingly, the amount of real output, Q, that a state can produce during some period, such as a year, depends on the size of its capital stock, K, and the number of hours worked, L. The symbol F is a function, or equation, relating output to capital and labor inputs. The symbol A measures the overall effectiveness with which a state uses its capital and labor resources. The symbol A is therefore referred to as a measure of total factor productivity. If two states used the same levels of capital and labor, the more productive state would have a larger A term and would therefore produce more output than the state with a lower A term. While some studies have treated the various productivity factors as inputs in the produc tion function, the Carlino/Voith study treated them as affecting the efficiency parameter, A. Specifically, the value of A depends on industry mix, urbanization economies, public capital, and the quality of labor. This means that the various productivity factors augment private sector use of labor and capital. In this case, an increase in the level of public capital increases the efficiency of both private capital and labor. The Empirical Model. Empirical analysis of state productivity has had to deal with an important data problem, namely, data on the stock of capital at the state level are not available. Fortunately, a production function technique has been developed that permits the estimation of productivity without the need for data on the capital stock.3 The technique involves estimating a wage equation. It is assumed that workers are paid according to their productivity (that is, there is perfect competition in and across local labor markets), and therefore wages and the demand for labor reflect the differentials in productivity across states. Under these conditions, the following wage equation is derived from the aggregate produc tion function: where W.( = Annual aggregate real wage bill divided by number of employees in state i for time t. S..t = The real output share of the j-th one-digit industry (mining; construction; manufactur ing; transportation, communication, and public utilities; wholesale and retail trade; finance, insurance, & real estate; services; and government) in state i for time t. P. = For each year, the percent of state i's population living in metropolitan areas in 1970 or 1980 (whichever is closest) based on 1983 metropolitan area definitions. I. = Total primary Federal-Aid Highway System miles per square mile of land area in state i for 1980. E. = Educational attainment (percent of the population 25 years old and over with 12 or more years) in i in 1980. aSee Gerald A.Carlino, "Increasing Returns to Scale in Metropolitan Manufacturing," Journal of Regional Science, 19,1979, pp. 363-73. FEDERAL RESERVE BANK OF PHILADELPHIA APPENDIX (continued) Highways and Education: The Road to Productivity? Gerald A. Carlino T Zt = Technical progress, represented by a time index. = Dummy variable to capture the effects of the energy shock years; Z( = 1 if t = 1973 to 1978; and 0 otherwise. U = Union membership as a percent of employees in nonagricultural establishments in i for 1970. Q = Real gross state product in state i at time t. Lu = Aggregate employment in state i at time t. The findings reported in the text of this article are based on a random-effects estimation of a pooled cross-section time series model for the 48 contiguous states for the period 196786 (providing 960 observations).13 While a wage equation was estimated, we obtained the effects of industry mix, urbanization economies, public infrastructure, and labor force quality on output indirectly by transforming the appropriate estimated coefficients of the wage equation.0 bThe estimated coefficients for industry mix, urbanization economies, public capital, and labor quality capture the direct effect of these variables on labor productivity. There may also be important indirect effects that are not captured by the estimates. For example, states with high educational attainment may also attract the more productive industries. °Let a krepresent the output effect of the k-th productivity variable. Then the output effect is calculated indirectly as a k = (3 / p, where p = 0 -1. For details see Gerald A. Carlino and Richard Voith, "Accounting for Differences in Aggregate State Productivity," Regional Science and Urban Economics, 22,1992, pp. 597617. 31 FEDERAL RESERVE BANKOF PHILADELPHIA BUSINESS REVIEW Ten Independence Mall, Philadelphia, PA 19106-1574
@FedFRASER