The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Vol. 29, No. 1 ECONOMIC REVIEW FEDERAL RESERVE BANK OF CLEVELAND 1993 Quarterl Vol. 29, No. 1 Generational Accounts and Lifetime Tax Rates, 1900-1991 2 by Alan J. Auerbach, Jagadeesh Gokhale, and Laurence J. Kotlikoff Unlike the federal budget, which typically measures receipts and ex penditures for one year at a time, generational accounts and lifetime tax rates focus on long-term intergenerational wealth redistribution. The accounts show that future generations can expect to pay, on average, more than twice as much to the government as current (1991) new borns if living generations continue to be treated as they are under cur rent policy. Lifetime tax rates on successive generations have increased from 22 percent for Americans born in 1900 to about 34 per cent for those born in 1991. Under the baseline economic assump tions presented here, future generations are slated to see that figure rise to more than 70 percent on average. Has the Long-Run Velocity of M2 Shifted? Evidence from the P* Model 14 Economic Review is published quarterly by the Research Depart ment of the Federal Reserve Bank of Cleveland. Copies of the Review are available through our Public Affairs and Bank Relations Depart ment, 1-800-543-3489. Coordinating Economist: William T. Gavin Advisory Board: Jagadeesh Gokhale Erica L. Groshen Joseph G. Haubrich Editors: Tess Ferg Robin Ratliff Design: Michael Galka Typography: Liz Hanna by Jeffrey J. Hallman and Richard G. Anderson The P-Star (P*) model forecasts inflation by exploiting the stability of M2 velocity and the tendency of the real economy to operate near its potential. While originally offered as a link between inflation and money growth, inverting the model provides a test of one of its primary assumptions: the constancy of M2’s long-run velocity, or V-Star (V*). If V* has increased during the last three years, predictions of inflation from the original P* model should be inferior to predictions from a model that incorporates the new, higher V* In fact, the deceleration of inflation through 1992:IIIQ was quite close to the original model’s pre diction, and simulations of the model under a variety of hypotheses regarding changes in V* provide relatively little support for a dramatic shift in that measure. Opinions stated in Economic Review are those of the authors and not necessarily those of the Federal Reserve Bank of Cleveland or of the Board of Governors of the Federal Reserve System. Material may be reprinted provided that the source is credited. Please send copies of reprinted material to the editors. ISSN 0013-0281 Examining the Microfoundations of Market Incentives for Asset-Backed Lending 27 by Charles T. Carlstrom and Katherine A. Samolyk Many view the proliferation of securitization as a response to com petitive or regulatory pressures. But to what extent would assetbacked lending occur in a less regulated environment? This paper addresses the extent to which models of credit intermediation have been able to formalize some of the market-based forces driv ing this phenomenon. The authors examine four papers that model some of the dimensions of asset-backed markets. An un derlying theme is that under certain conditions, the very informa tion costs that make financial markets important as conduits of credit can also create nonregulatory incentives for asset-backed lending as an efficient funding mode. Generational Accounts and Lifetime Tax Rates, 1900-1991 by Alan J. Auerbach, Jagadeesh Gokhale, and Laurence J. Kotlikoff Introduction Generational accounting is a new method for determining how government deficits, taxes, transfer payments, and other expenditures affect the distribution of income and wealth among different generations. The technique is still being developed, and a number of the assumptions used to estimate the accounts are controversial. Auerbach, Gokhale, and Kotlikoff (1991), Kotlikoff (1992), and Office of Management and Budget (1992) explain the basic concept and present some illustrative results. This article u p dates the baseline generational accounts report ed in the 1993 federal budget and estimates the effects of several new alternative policies. It also extends the analysis for the first time to lifetime net tax rates— the taxes that a generation pays, less the Social Security and other transfer bene fits that it receives, as a percentage of income over its entire lifetime. The new analysis reveals the following: Alan J. Auerbach is a professor of economics at the University of Pennsylvania and an associate of the National Bureau of Economic Research, Jagadeesh Gokhale is an economist at the Federal Reserve Bank of Cleveland, and Laurence J. Kotlikoff is a professor of econom ics at Boston University and an associate of the National Bureau of Economic Research. A version of this article appeared in Budget Baselines, Historical Data, and Alternatives for the Future, Office of Management and Budget, January 1993. • The net tax rates paid by future generations will be substantially higher than those paid by the baby boom and other current generations, unless policy actions are taken now to mitigate the increase. • The generational imbalance between newly bom and future Americans could be largely elimi nated either by imposing a cap on mandatory spending (excluding Social Security) from 1993 through 2004 or by instituting an appropriate sur tax. Both policies would significantly raise the net taxes paid by current Americans, but the increase for the newly bom would be considerably more under a surtax. I. The Nature of Generational Accounts The federal budget normally measures receipts and outlays for one year at a time and reports these estimates for only a few years into the fu • The lifetime net tax rates paid by Americans ture. Generational accounts, in contrast, look in the baby boom and successive generations ahead many decades, classifying taxes paid and will likely be much higher than the rates paid transfers received— such as Social Security, Medi by those born earlier. care, and food stamps— according to the generation http://fraser.stlouisfed.org/ that pays or receives the money. For an existing Federal Reserve Bank of St. Louis generation, taxes and transfers are estimated year by year over members’ remaining lifespan. These amounts are then summarized in terms of one number, the present value of the generation’s entire annual series of average future tax payments net of transfers received. For future generations, the accounts are based on the proposition that the gov ernment’s bills will have to be paid either by them or by those now living. The calculations determine how much future Americans will have to pay on average to the government, above the amount they will receive in transfers, if total government spend ing is not reduced from its projected path and if those now living pay no more than anticipated. Defined more precisely, generational accounts measure, as of a particular base year, the present value of the average future taxes that a member of each generation is estimated to pay minus the present value of the average future transfers that he or she is estimated to receive. This difference is called the “net payment” in the following dis cussion. A generation is defined as all males or females bom in a given year. Generational accounts can be used for two types of comparison. First, they allow us to com pare the lifetime net payments by future gen erations, by the generation just bom, and by different generations born in the past. Lifetime net payments by generations born in the past are based on estimates of actual taxes paid and transfer payments received through 1991, as well as on projections of taxes to be paid and transfer payments to be received in the future. Second, generational accounts can be used to compare the effects of actual or proposed policy changes on the remaining lifetime net payments of currently living and future generations. Such comparisons can be made equally well for policies that change the totals of receipts or expenditures and for those that change the composition of the budget without affecting the deficit. It should be noted that, as now constructed, generational accounts have a number of limita tions. First, they include the taxes and transfers of all levels of government— federal, state, and local— and thus do not show the separate effect of the federal budget as a whole. However, the difference in the accounts due to a federal gov ernment policy change can be analyzed alone. Second, generational accounts reflect only taxes paid and transfers received. They do not impute to particular generations the value of the govern ment’s purchases of goods and services for educa tion, highways, national defense, and so on. Thus, the full net benefit or burden that any generation receives from government fiscal policy as a whole is not totally captured. Still, the accounts can reveal the effects of a policy change that affects only taxes and transfers. In the future, it may be feasible to impute the value of certain types of government purchases to specific generations. Third, generational accounting does not, as yet, incorporate any policy feedback on the economy’s growth and interest rates. Feedback effects can be significant, but because they generally occur slowly, their impact on the dis counted values used in the accounts may be small. Moreover, there is reason to believe that they would reinforce the conclusions derived here. For example, policies that decrease cur rent generations’ net payments while increasing the burden on future generations are likely to reduce investment over time. This in turn will lower real wage growth and raise real interest rates, which on balance will harm future genera tions in absolute terms. Finally, generational accounting divides people bom in the same year into only two categories, males and females, with each designated a “gen eration.” This is an important distinction, since the sexes differ significantly in such characteristics as lifetime earnings and longevity. However, the method does not reveal differences with respect to other characteristics, such as income levels or race, nor does it show the wide diversity among individuals within any particular grouping. Thus, the results presented here should be viewed as experimental and illustrative. They are limited by the availability and quality of the data, especially for earlier years. In addition, they are necessarily based on a number of simplifying assumptions (about which reasonable people may disagree) concerning the pattern of future taxes and spending, mortality and birth rates, the interest rate used for discounting future taxes and transfers to derive present values, and so forth. The absolute amounts of the generational ac counts are sensitive to all of these assumptions. Nevertheless, like the 75-year projections is sued each year by the Social Security trustees, the accounts can be illuminating when considered in light of their assumptions. Moreover, the most fun damental result— that future generations’ average net payment will be relatively much larger than that of the generation just bom — holds for a wide range of reasonable changes in the assumptions. II. Remaining Net Payments by Existing Generations Tables 1 and 2 show the generational accounts as of calendar year 1991 for every fifth generation of T A B L E 1 Generational Accounts for Males: Present Value of Taxes and Transfers as of 1991 (thousands of dollars) Taxes Paid Transfers Received Labor Income Taxes Capital Income Taxes Payroll Taxes Excise Taxes Social Security 78.9 99.7 125.0 157.2 187.1 204.0 205.5 198.8 180.1 145.1 97.2 38.9 -23.0 -74.0 -80.7 -75.5 -61.1 -47.2 -3.5 29.2 37.5 47.8 61.1 73.5 80.4 80.4 77.6 71.0 59.8 45.8 30.2 16.2 5.7 2.4 1.1 0.6 0.2 0.0 10.1 12.9 16.5 21.2 26.5 33.1 39.9 46.8 52.3 55.4 55.3 52.2 46.4 39.0 30.9 23.6 18.0 15.0 7.1 31.8 41.0 52.3 67.1 81.3 89.5 89.8 87.0 79.9 67.6 52.0 34.5 18.6 6.6 2.7 1.3 0.7 0.3 0.0 28.2 33.3 38.7 44.6 48.3 49.1 48.5 47.8 46.9 44.5 40.7 36.2 30.8 25.6 20.4 15.5 11.0 7.6 1.7 166.5 n.a. n.a. n.a. n.a. Generation’s Age in 1991 Net Payment 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 Future generations Health Welfare 6.1 7.7 9.2 10.7 11.8 14.6 18.0 22.6 28.5 35.9 45.2 57.1 72.4 82.3 75.5 63.3 47.9 36.4 . 6.5 11.0 13.1 15.7 19.2 22.2 24.3 26.4 29.7 34.1 39.6 45.4 51.8 58.1 64.6 58.2 50.9 41.5 33.1 5.8 3.3 4.2 5.4 6.9 8.4 9.0 8.6 8.0 7.3 6.6 6.0 5.3 4.6 3.9 3.4 2.8 1.9 0.9 a n.a. n.a. n.a. n.a. n.a. n.a. Percentage Difference in Net Payment Future generations and age zero 111.1 n.a. n.a. n.a. n.a. a. $0.05 thousand or less. SOURCE: Office o f Management and Budget (1992). males and females alive in that year. The first col umn, “Net Payment,” is the difference between the present value of taxes that a member of each generation will pay, on average, over his or her remaining lifetime and the present value of transfers that he or she will receive. The other col umns show the average present values of several different taxes and transfers. All federal, state, and local taxes and transfers are included in these cal culations. Federal spending and receipts are based on the baseline calculations in the Office of Man agement and Budget’s Mid-Session Revieu>o f the 1993 Budget. The present value of future taxes to be paid by young and middle-aged generations far ex ceeds the present value of the future transfers they will receive. For males age 40 in 1991, for exam ple, the present value of future taxes is $180,100 more than the present value of future transfers. The amounts are large because these genera tions are close to their peak taxpaying years. For newborn males, on the other hand, the present value of the net payment is much smaller, $78,900, because they will pay very lit tle in taxes for a number of years. Older generations, who are largely retired, will receive more Social Security, Medicare, and other future benefits than they will pay in future taxes. That is, they have negative net payments. Females have smaller net payments than males, mainly because they earn less and thus pay less income and Social Security taxes. 5 TABLE 2 Generational Accounts for Females: Present Value of Taxes and Transfers as of 1991 (thousands of dollars) Taxes Paid Transfers Received Labor Income Taxes Capital Income Taxes Payroll Taxes Excise Taxes Social Security Health Welfare 39.5 48.7 59.4 72.4 84.0 86.4 81.1 71.9 55.3 29.5 -2.2 -39.5 -80.8 -112.5 -110.6 -100.6 -83.3 -65.6 -9.8 15.1 19.4 24.7 31.4 37.1 38.5 36.2 33.3 29.0 23.1 16.7 10.8 5.6 2.0 0.8 0.4 0.2 0.1 0.0 3.7 4.8 6.1 7.9 9.8 12.3 15.5 19.1 22.3 24.8 26.1 26.0 24.4 21.7 18.0 13.8 9-3 4.7 0.5 16.5 21.2 27.0 34.6 41.3 42.9 40.5 37.4 32.7 26.2 19.0 12.3 6.4 2.3 0.9 0.4 0.2 0.1 0.0 27.3 32.0 36.8 41.8 45.0 46.1 46.1 46.1 45.2 43.2 39.5 35.2 30.3 25.3 20.6 15.8 11.6 8.9 1.6 5.8 7.3 8.7 10.0 11.1 13.7 17.0 21.3 26.9 34.2 43.5 55.6 71.4 80.3 74.2 63.0 49.5 36.8 5.6 9.6 11.5 14.0 17.3 20.0 23.2 26.9 32.1 38.8 47.4 55.4 64.4 73.1 80.8 74.4 65.8 53.3 41.1 6.0 7.7 9.9 12.5 16.0 18.2 16.5 13.4 10.7 8.2 6.1 4.6 3.7 3.1 2.7 2.4 2.1 1.7 1.4 0.2 83.4 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Generation’s Age in 1991 Net Payment 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 Future generations Percentage Difference in Net Payment Future generations and age zero 111.1 n.a. n.a. n.a. n.a. SOURCE: Office o f Management and Budget (1992). Since the figures in these tables show the rem aining lifetime net payments of particular generations, they do not include the taxes paid or transfers received in the past. This must be kept in mind when considering the net payments of those now alive. The portion of a generation’s remaining lifetime net payment depends on whether we are talking about 10-, 40-, or 65-yearolds. The fact that 40-year-old males can expect to pay more in the future than they receive, in present-value terms, while the reverse is true for 65-year-old males, does not necessarily mean that federal, state, and local governments are treating the 40-year-olds unfairly. Because 65-year-old men paid considerable taxes when younger, and these are not reflected in their remaining lifetime http://fraser.stlouisfed.org/ net payments, direct comparisons are impossible. Federal Reserve Bank of St. Louis The lifetime net payments of different genera tions can be compared only by using lifetime net tax rates, discussed below. Estimates of future net payments by generation are affected by the amount of taxes, transfers, and other government expenditures assumed year by year in the baseline projection. These assumptions can differ widely. As explained in the appendix, the methods of projection generally seek to main tain current policy in some sense. However, cur rent policy can be interpreted in several ways, especially for expenditures such as defense. Fur thermore, long-term Medicare and Medicaid projections assume that, eventually, policy actions or other forces will hold spending growth to the overall rate of economic expansion (adjusted for shifts in the age and sex composition of the T A B L E 3 Percentage Difference in Net Payments between Future Generations and Age Zero Productivity Growth Rate Interest Rate 0.25 0.75 1.25 3.0 6.0 9.0 117 138 228 . 89 111 193 65 87 162 SOURCE: Office o f Management and Budget (1992). population), even if the growth rate is quite rapid for the next few decades.1 III. Net Payments by Future Generations Future generations— those bom in 1992 and later — will be required to make a 111 percent larger net payment to the government, on average, than those bom in 1991. The average net payments of $166,500 by future males and $83,400 by future females are calculated assuming that the male-tofemale net payment ratio is the same for future generations as for those bom in 1991. The calcula tions also assume that all future Americans of a particular sex will make the same average net pay ment over their lifetimes after adjustments are made for economic growth. A growth adjustment is needed to compensate for the fact that future generations will pay more in taxes, net of transfers received, simply because their incomes will be higher. To properly assess fu ture generations’ net payment relative to that of the newly bom, it is necessary to calculate the net pay ment they will make above and beyond the amount due to economic growth. Generational accounts assume that all future generations will pay the same net amount apart from this growth adjustment. The net amount is the number shown in tables 1 and 2 for all future generations of the same sex. A generational imbalance, as defined above, is calculated in such a way that the generations now alive, including the newly bom, do not pay any more taxes (or receive any less transfers) than projected in the baseline. This assumption is an analytical device for determining the size of the nation’s fiscal imbalance; it is not meant to ■ 1 A pure extrapolation of recent trends, in contrast, implies that http://fraser.stlouisfed.org/ health care costs w ill eventually bankrupt the government. Federal Reserve Bank of St. Louis suggest that future generations will in fact close the gap all by themselves. Any actual policy change is almost certain to bear in some degree on current generations as well as on those yet to be born. If such a policy change is made, the percentage difference in net payments between the newly born and future generations would be less than shown in tables 1 and 2. Policy changes of this kind are discussed below. The size of the imbalance between future generations and the newly bom is sensitive to assumptions about both the interest rate used for discounting and the growth rate of the econ omy. Table 3 shows the percentage differential under interest rates of 3.0, 6.0, and 9-0 percent and productivity growth rates of 0.25, 0.75, and 1.25 percent. Although the difference ranges from 65 percent to 228 percent, our basic con clusion, that future generations’ net payment will be much larger than that of those just born, still holds in every case. The generational imbalance also depends on the policy assumption that all future generations of the same sex will have the same net payment (after adjusting for growth). But suppose that the future generations born between 1992 and 2001 pay only the same amount as those born in 1991- Because these future generations pay less than previously assumed, those born after 2001 will have a net payment that is 186 percent larger, rather than 111 percent larger, than that facing the 1991 generation. The greater the num ber of future generations who pay no more than current newborns, the larger will be the net payment required of generations w ho are bom still later. Change in the Imbalance between 1990 and 1991 The estimated 111 percent imbalance in 1991 between newborns and future generations can be compared with the estimated 79 percent im balance in 1990 reported in the fiscal year 1993 budget. The difference primarily reflects lower baseline receipts projected for 1993-2004. Based on last year’s projections, the estimated 1991 imbalance would be 81 percent. A second factor is that another generation, the one born in 1991, does not have to make the higher lifetime net payments required of future generations. T A B L E 4 Change in Generational Accounts Due to Alternative Policies as of 1991 (thousands of dollars) Males Generation’s Age in 1991 Females Mandatory Cap Surtax Mandatory Cap 6.4 7.7 9.1 10.5 11.1 11.8 12.6 14.0 15.9 18.2 20.7 23.0 23.2 20.0 15.6 11.0 6.6 2.5 0.0, 16.1 19.2 22.4 25.3 26.1 25.5 24.0 21.8 18.8 15.1 11.2 7.6 4.9 3.1 2.0 1.2 0.7 0.3 0.0 5.4 6.6 7.9 9.3 10.411.8 13.5 15.9 18.7 22.0 25.6 29.2 30.3 27.4 22.7 16.9 10.2 3.6 0.0 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 Future generations -71.3 -57.2 -33.2 Surtax 7.5 8.9 10.4 11.4 11.6 11.1 10.4 9.4 8.2 6.8 5.3 4.0 2.8 1.9 1.2 0.6 0.2 a 0.0 -29.3 Percentage Difference in Net Payment Future generations and age zero 11.7 15.1 11.7 15.1 a. $0.05 thousand or less. SOURCES: Office of Management and Budget (1992) and authors’ calculations. IV. Illustrative Policy Changes Table 4 compares two alternative policies aimed at rectifying the fiscal imbalance between the gen eration just bom and future generations. Both would remove the imbalance to about the same degree, but their distributive effects among differ ent generations vary tremendously. The first of these policies is a cap on all man datory spending programs except Social Security and deposit insurance. From 1993 to 2004, the sav ings from the cap would be calculated for each mandatory program with beneficiaries as the dif ference between 1) baseline spending and 2) spending limited to the growth in the number of beneficiaries plus the inflation rate (with a little ad ditional growth allowed in the first two years for transition). Medicare and Medicaid are the largest mandatory programs, and they produce most of the total savings. For these two programs, spending would be limited to the amount determined by the cap. For all other mandatory programs (except So cial Security and deposit insurance), the required savings would be spread across the board as a proportionate reduction in spending. Employing the economic assumptions used for the 1993 MidSession Review (and extended to the years after 1997), the consolidated budget is projected to be balanced under the cap in 2004.2 Thereafter, the spending growth rates for mandatory programs would be the same as in the baseline calculations. However, because the level of mandatory spending in 2004 would be lower than under the baseline, applying these same growth rates would produce permanently lower levels of subsequent spending. The cap on mandatory spending would largely eliminate the imbalance in net payments between future generations and those just bom. Future generations would pay an average of 12 percent more, instead of 111 percent more. The net pay ment by future males would be $71,300 less than under the baseline, on average, and the net pay ment by future females would be $33,200 less. All existing generations would face a larger net payment. In terms of age, the biggest increase would be for people who are now around 55 to 60. This is because the cap would mainly reduce transfer payments for health care, especially Medicare, which is received almost totally by the elderly. The increase in net payments would be higher for females than males at almost every age, because females live longer, and the cap would primarily reduce transfers to the elderly. The second policy is a surtax on the federal individual income tax. From 1993 to 2004, the amount of the surtax would equal the spending reduction required by the mandatory cap. After 2004, the surtax would increase at the same rate as other taxes generally do. The surtax would reduce the generational im balance by almost as much as the mandatory cap. Future generations would pay 15 percent more on average than those just born, com pared to 12 percent under the cap and 111 per cent under the baseline. The average future male would pay $57,200 less, and the average future female would pay $29,300 less. All exist ing generations would pay more. The distributional effect of the surtax would be quite different from that of the mandatory cap, however. The surtax would bear much ■ 2 The budget would not necessarily be balanced in all later years. Generational balance over a period taken as a whole is consistent with some years of deficit, and the illustrative policies do not entirely eliminate the imbalance. TABLE 5 Lifetime Net Tax Rates, Gross Tax Rates, and Transfer Rates (percent) Males Females Average of Males and Females Net Tax Rates Gross Tax Rates Transfer Rates Net Tax Rates Gross Tax Rates Transfer Rates Net Tax Rates Gross Tax Rates Transfer Rates 1900 1910 1920 1930 1940 1950 I960 1970 1980 1990 1991 17.8 21.8 24.2 26.4 28.2 30.6 32.3 33.6 34.1 33.9 33.9 19.6 24.6 27.7 30.5 33.0 36.8 39.6 41.7 42.4 42.7 42.7 1.8 2.8 3.5 4.1 4.8 6.2 7.2 8.1 8.3 8.7 8.8 35.3 35.7 34.0 34.4 32.7 30.6 31.5 32.5 33.1 32.9 32.8 43.9 49.6 50.4 52.8 50.6 46.9 47.9 50.3 51.6 52.0 52.0 8.7 13.9 16.5 18.5 17.9 16.3 16.4 17.8 18.5 19.1 19.2 21.5 24.7 26.3 28.1 29.3 30.6 32.1 33.2 33.8 33.6 33.5 24.8 29.8 32.5 35.3 37.3 39.9 42.3 44.5 45.5 45.7 45.8 3.3 5.2 6.2 7.2 8.0 9.3 10.2 11.3 11.7 12.2 12.2 Future generations 71.5 n.a. n.a. 69.3 n.a. n.a. 71.1 n.a. n.a. Generation’s Year of Birth SOURCE: Office o f Management and Budget (1992). more on the relatively young; the cap, on the relatively old. For example, a 65-year-old male would pay $3,100 more under the surtax than under the baseline, but $20,000 more under the cap; in contrast, a 20-year-old male would pay $26,100 more under the surtax but $11,100 more under the cap. This is because the surtax is paid disproportionately by younger people earning income, whereas the cap disproportionately re duces transfer payments to the elderly. The second distributional difference is be tween males and females. The surtax bears more on males; the cap, on females. This is primarily due to the fact that males tend to have higher in comes and pay more income taxes, whereas fe males tend to live longer and receive more health care transfers. The two policies also have different distribu tional effects between existing and future gener ations. The reduction in net payments by future generations is less under the surtax: $14,000 less for males, on average, and $4,000 less for fe males. This is partly because a larger imbalance remains between future generations and those just born, 15 percent compared to 12 percent. The improvement for future generations is less under the surtax because older generations do not pay as much more. V. Historical Lifetime Tax Rates The analysis so far has been prospective, consid ering only the present value of future taxes and transfers as of 1991 for existing generations and those yet to be bom. A prospective analysis can compare policy changes, and it can compare the lifetime fiscal burdens on the newly born and fu ture generations, since their entire lifetimes are yet to come. However, it cannot compare the lifetime fiscal burden of one existing generation with that of another existing generation born in a different year— or with future generations— because part of any living generation’s taxes and transfers occurred in the past and thus are not taken into account. A comparison of one existing generation with another must be based on their entire lifetime taxes and transfers. Table 5 shows the results in terms of lifetime net tax rates for different genera tions bom since 1900 and for future generations. The lifetime net tax rate of a generation is defined as the present value of its lifetime net taxes (taxes less transfers) divided by the present value of its lifetime income. The present values are calculated as of the generation’s year of birth, so that each cohort can be compared from the standpoint of when it was bom. The lifetime net taxes are the 9 same as the generational account for a genera tion in the year of its birth. (As shown in table 1, the lifetime net taxes of males bom in 1991 are $78,900.) Since lifetime taxes, transfers, and in come have trended upward and have fluctuated to some extent, it is more appropriate to com pare the relative fiscal burden on different gen erations in terms of lifetime net tax rates than in terms of absolute amounts. Lifetime net tax rates are calculated from his torical data on taxes, transfers, and income up to 1991 and from projections of future data as described in the previous sections. Historical data, however, are not available in the same detail as the figures for recent years underlying our projections, and in some cases they are not available at all. The appendix summarizes the methods used to construct the historical series. Lifetime calculations also introduce a number of conceptual issues. For example, how should lifetime income be measured? Lifetime income is defined as a present value, like lifetime taxes and transfers. Therefore, the present-value calcula tions should include all income that increases a generation’s resources: labor earnings, inherited wealth, and capital gains over and above the normal return to saving. The normal return to saving is not itself included in income, because that would be double counting. Saving and earning a normal rate of return do not increase the present value of a household’s resources. Data do not exist on the share of each genera tion’s income stemming from inherited wealth or supernormal capital gains, so labor earnings are used to represent income.3 The lifetime net tax rate for males in the base case exhibits a strong upward trend, rising from 17.8 percent in 1990 to about 34 percent in 1970 and succeeding years. The lifetime net tax rate for females exhibits a quite different pattern. It started much higher than for males, at 35-3 per cent, declined irregularly for half a century, and rose slightly thereafter. Since 1950, the net tax rate has been about the same for both sexes. The pattern of the female net tax rate is an artifact of w om en’s increasing labor force partic ipation and the method used to attribute labor earnings and taxes within a family. Labor earn ings are attributed to the person who receives them; some taxes, including excises, are attrib uted equally to husband and wife. The lower female earnings thus contribute to a higher fe male tax rate, especially in the early decades of ■ 3 The error due to this omission is relatively small in the aggregate, given that labor income has long accounted for three-fourths of all income and that only part of the remaining income from capital should be included. However, the errors for different generations could vary, depending on http://fraser.stlouisfed.org/ fluctuations in asset values and bequest behavior. Federal Reserve Banktrends of St.and Louis the century. At the same time, the rise in female labor force participation over time has caused their earnings to increase faster than male earn ings, without directly increasing those taxes that are attributed equally to husband and wife. This has offset the general increase in taxes that con tributed to the rising net tax rates observed in the series for males. This pattern emphasizes a conceptual question in calculating the generational accounts. How should income, taxes, and transfers be attributed within a family? Excise taxes could alternatively have been attributed in proportion to labor earn ings, or labor earnings could have been attrib uted equally between husband and wife. Table 5 displays one answer to this question by includ ing lifetime net tax rates for males and females combined, calculated as a weighted average of the net tax rate for each sex. Note that the aver age net tax rises significantly over most of this century, increasing from 21.5 percent for the gen eration born in 1900, to 32.1 percent for the gen eration born in I960, to about 33 percent for the generations born since 1970. This trend reflects the growing fiscal role of government. The aver age net tax rate for future generations is 71.1 percent, which is the same percentage differ ence relative to people newly born in 1991 as that shown in tables 1 and 2. The male and female net tax rates are virtually identical for fu ture generations. Table 5 also breaks down the net tax rates between gross tax rates and transfer rates. To calculate the latter, the present value of a gener ation’s lifetime taxes (or transfers) is divided by the present value of its lifetime income. This breakdown reveals the expanded role of gov ernment transfer payments during the past cen tury. The lifetime transfer rate for males and females taken together nearly quadrupled be tween the generations born in 1900 and those bom in 1991, starting at 3-3 percent and rising each decade to a rate of 12.2 percent. The in crease was more rapid, in both relative and ab solute terms, for the generations born before World War II than afterward. Because of the growth in the transfer rate, the gross tax rate has not leveled off in the past two decades to the same extent as the net tax rate. The gross tax rate for males and females combined nearly doubled between the generations born in 1900 and 1991, starting at 24.8 percent and in creasing each decade to a rate of 45.8 percent. A generation’s lifetime taxes pay for the govern ment’s purchases of goods and services as well as for public transfers to its own members and other generations. T A B L E 6 Lifetime Net Tax Rates (percent) Males Generation’s Year of Birth Females Average of Males and Females Baseline Mandatory Cap Surtax Baseline Mandatory Cap Surtax Baseline Mandatory Cap Surtax 1900 1910 1920 1930 1940 1950 I960 1970 1980 1990 1991 17.8 21.8 24.2 26.4 28.2 30.6 32.3 33.6 34.1 33.9 33.9 17.8 21.8 24.4 26.8 28.9 31.5 33.6 35.3 36.5 36.6 36.6 17.8 21.8 24.3 26.4 28.5 31.6 34.6 37.6 39.9 40.7 40.8 35.3 35.7 34.0 34.4 32.7 30.6 31.5 32.5 33.1 32.9 32.8 35.3 35.9 34.8 36.5 35.2 32.9 34.2 35.7 37.0 37.4 37.3 35.3 35.7 34.0 34.5 33.2 31.5 33.5 35.9 38.2 39.0 39.1 21.5 24.7 26.3 28.1 29.3 30.6 32.1 33.2 33.8 33.6 33.5 21.5 24.7 26.6 28.9 30.4 31.9 33.8 35.4 36.6 36.9 36.9 21.5 24.7 26.3 28.2 29.7 31.6 34.2 37.1 39.3 40.2 40.2 Future generations 71.5 40.9 47.0 69.3 41.7 45.0 71.1 41.3 46.5 SOURCE: Office of Management and Budget (1992). The breakdown further shows that the simi larity between males and females in lifetime net tax rates masks very different gross tax and transfer rates. Each rate is much higher for fe males, reflecting such factors as their lower life time income and greater longevity (as well as the attribution assumptions for taxes and income within the family). Table 6 shows how policy changes designed to rectify the generational imbalance would affect the lifetime net tax rates of different gener ations. For future generations, the cap on m an datory spending reduces the average lifetime net tax rate on males and females together from 71.1 percent to 41.3 percent, while the surtax reduces it to 46.5 percent. For existing generations, the effect of policy changes on lifetime net tax rates increases as the generation’s age declines, and for the very youngest cohort, bom in 1991, the change is quite significant. Under the mandatory cap, this generation’s lifetime net tax rate increases by 2.7 percentage points for males. For females, who will live longer, the increase is 4.5 percent age points. A surtax would raise the burden on the youngest group still more: an increase over the baseline of 6.9 percentage points for males and 6.3 percentage points for females. For older generations, the increase in the lifetime net tax rate is smaller, primarily because the absolute ef fects of the policy change are discounted over http://fraser.stlouisfed.org/ more years in order to calculate the present value Federal Reserve Bank of St. Louis as of the generation’s year of birth. In the case of the surtax, the absolute effects are also smaller for older generations, because they have fewer remaining years of labor earnings. The burden that remains on the older genera tions is greater under the mandatory cap than under the surtax, as previously explained, be cause Medicare benefits are relatively high and income taxes relatively low during their remain ing years. Since females live longer than males, the increase in their lifetime net tax rate under the mandatory cap is greater than for males at every age. O n the other hand, because males have higher labor earnings, the surtax generally hits them harder than it does females. Appendix— Construction of the Generational Accounts Present-Value Constraint Generational accounting is based on the presentvalue budget constraint of the government sector. In simple terms, this constraint says that the gov ernment must ultimately pay for its purchases of goods and services either with resources it obtains from current and future generations or with its current assets (net of debt). If current KU generations pay less in taxes (net of transfers received) to finance government purchases, future generations will have to pay more. For example, suppose that, through borrowing, pay ments for the government’s bills were repeatedly shifted to future generations by each successive current generation. Then this debt would grow, with interest. Eventually, the interest would ex ceed the lifetime income of future generations, resulting in default. More precisely, the government’s presentvalue constraint means that, at any point in time, the present value of the government’s future pur chases of goods and services cannot exceed the sum of three items: 1) the present value of future taxes to be paid (net of transfers received) by existing generations (that is, the sum of their generational accounts multiplied by the number of people in each generation), 2) the present value of taxes to be paid (net of transfers received) by future generations, and 3) the value of government assets that yield income, less the government debt. Generational accounting estimates the present value of the government’s purchases of goods and services plus amounts 1 and 3- Amount 2, the pres ent value of taxes to be paid by all future genera tions (net of transfers received), is calculated as the present value of future government purchases minus amounts 1 and 3The generational accounts for future genera tions are derived from the aggregate amount 2. For all but one of the policy experiments dis cussed here, different net payments (after adjust ing for economic growth) are not estimated for different future generations. Rather, the aggregate present-value net payment by future generations is divided on an even basis among all future gen erations so that the average net payment by the members of each keeps pace with the economy’s productivity growth. Thus, as shown in tables 1 and 2, one single (growth-adjusted) average figure stands as the generational account for all future generations of a given sex. Because the genera tional account is calculated indirectly from the above aggregates, it can be shown only as a single number and cannot be divided among specific taxes and transfers. Underlying Calculations Calculating the generational accounts is a threestep process. The first step entails projecting each currently living generation’s average taxes and transfers for each future year in which at least some of its members will be alive. The second step converts these projected values into an actuarial present value, using assumptions for the discount rate and the probability that the generation’s members will be alive in each of the future years. The sum of these present values, with transfers sub tracted from taxes, is the generational account, or net payment, for existing generations shown in the first column of tables 1 and 2. The third step estimates the other terms of the present-value constraint (ex plained in the previous section) so as to derive the average net payment by future generations. The cal culations are based on projections to the year 2200. Projection o f taxes a n d transfers. The projec tion of average future taxes and transfers begins with the national totals of all federal, state, and local taxes and transfers as reported in the Na tional Income and Product Accounts (NIPAs) for calendar year 1991. (All years in this article are calendar years unless otherwise stated.) Employee retirement and veterans’ benefits paid by the gov ernment are considered a form of employee com pensation and are classified as the purchase of a service rather than as a transfer payment. The base-year NIPA totals are distributed to all existing generations, as defined by age and sex, based on the corresponding distributions in crosssection survey data. These surveys include the Sur vey of Income and Program Participation and the Current Population Survey, both by the Bureau of the Census, and the Survey of Consumer Expendi tures by the Bureau of Labor Statistics. Those taxes that are not directly paid by individuals and so do not appear in these surveys, such as the corpo rate income tax, are allocated. Because genera tional accounting attributes taxes and transfers to individuals, household taxes and transfers are attributed to household members. No spe cial imputations are made to children, but the cross-section surveys impute some consumption to them; thus, the taxes on that consumption are attributed to children. The attribution mles affect the values of the baseline accounts, but are not likely to alter the generational implications of policy changes. The distribution of average future taxes and transfers by age and sex is assumed to equal the base-year average amounts after adjustments for growth and projected policy. In the case of federal taxes and transfers for 1993-2004, the amounts correspond to the current service estimates of taxes and transfers in the Mid-Session Revieiv o f the 1993 Budget (July 1992), extended beyond 1997 and updated for the actual fiscal year 1992 results. In the case of state and local taxes and transfers for 1993-2004, the amounts are based on the GDP assumptions in the Mid-Session Review as well as on the assumption that the ratios of ita state and local tax and transfer aggregates to GDP remain constant at 1991 levels. After 2004, the average taxes and transfers by age and sex are assumed, with two exceptions, to increase at the assumed rate of productivity growth. Pro ductivity (both labor and multifactor) is assumed to increase by 0.75 percent a year, which is close to the average annual rate of multifactor produc tivity growth since 1970. Social Security and health care transfers are the two exceptions. Projected Social Security transfers and payroll tax receipts after 2004 are based on special calculations made by the Social Security Administration assuming a productivity growth rate of 0.75 percent. Projected Medicare and Medi caid transfers from 2005 through 2030 are calculated from the growth rates in the Health Care Financing Administration’s middle-scenario estimates published in 1991 4 After 2030, health care transfers are as sumed to stabilize as a percentage of GDP apart from the effect of changes in the composition of the popu lation by age and sex. Medicare receipts are assumed to grow at 0.75 percent a year. Assum ptionsforpresent value. The appropriate discount rate for calculating the present value of future amounts depends on whether these amounts are known with certainty. Future govern ment receipts and expenditures are risky, which suggests that they should be discounted by a rate higher than the real rate of interest on government securities. On the other hand, government receipts and expenditures appear to be less volatile than the real return on capital, which suggests that they should be discounted by a rate lower than that. The baseline calculations assume a 6 per cent real discount rate, which is intermediate between the roughly 2 percent average real return available in recent years on short-term Treasury securities and the roughly 10 percent real return available on capital. The present values of future average taxes and transfers are also discounted for mortality probabilities in order to derive actuarial present values. The demographic probabilities through 2066 are those embedded in the Social Security trustees’ intermediate projection in 1992 (alter native II) of the population by age and sex. The fertility, mortality, and immigration probabilities in 2066 were used for later years. Immigration is treated as equivalent to a change in mortality. Otherprojections. Federal purchases of goods and services through 2004, like federal taxes and transfers, are from the latest Mid-Session Review extended beyond 1997 and updated for the actual fiscal year 1992 results. State and local purchases through 2004 are kept at the same ratio to GDP as http://fraser.stlouisfed.org/ 4 This scenario is discussed in Sonnefeld et al. (1991). Federal Reserve Bank of St. Louis ■ in 1991. Federal, state, and local purchases after 2004 are divided between 1) those made on behalf of specific age groups— the young, middle-aged, and elderly— such as educational expenditures, and 2) those that are more nearly pure public goods, such as defense and public safety. Purchases per person in each of the three age groups, and purchases of public goods per capita, all increase at the assumed rate of productivity growth. The economic value of the government assets that yield income, less the government debt, is es timated to be the cumulative amount of the NIPA deficit since 1900 converted to constant dollars by the GDP deflator. The average growth-adjusted net payment to be made by future generations is determined using the aggregate present value of the net pay ment (as derived through the present-value bud get constraint), the assumed productivity growth, and the projected size of future generations. The size of future generations is estimated using the Social Security alternative II projection through 2066 and the demographic assumptions for 2066 for later years. H istorical lifetim e net tax rates. Lifetime net tax rates for generations born between 1900 and 1991 are calculated by dividing the generational account of each generation at birth by its human wealth— the present value at birth of its future labor earnings. Calculating a generation’s human wealth requires knowing its average labor earnings in each future year. The average labor earnings received by particular generations in particular years are deter mined by distributing aggregate labor income by age and sex using cross-section distributions of labor income found in cross-section survey data. The lifetime generational accounts for generations bom between 1900 and 1991 are based on actual taxes and transfers between 1900 and 1991 and on projected taxes and transfers in the years thereafter. Aggregate labor earnings, taxes, and transfers were obtained from the NIPAs for 1929 and later years. Pre-1929 aggregate labor earnings are from Historical Statistics o f the United States, Colonial Times to 1970. Pre-1929 taxes and transfers are from the 1982 Census of Governments, Historical Statis tics on Government Finances an d Employment. Various cross-section surveys are used to distribute aggregate labor earnings, taxes, and transfers by age and sex. Cross-section surveys prior to the early 1960s were not available for this study, so surveys from years after I960 are used for earlier years. The Current Population Surveys are used for labor earn ings and taxes on labor earnings in 1964 and later years, and the 1964 survey is used for earlier years. ¡a References Auerbach, Alan J., Jagadeesh Gokhale, and Lau rence J. Kotlikoff. “Generational Accounts: A Meaningful Alternative to Deficit Account ing,” in David Bradford, ed., Tax Policy a n d the Economy, vol. 5. Cambridge, Mass.: Na tional Bureau of Economic Research and MIT Press, 1991. Kotlikoff, Laurence J. G enerational A ccounting: K now ing Who Pays, a n d When, fo r W hat We Spend. New York: The Free Press, 1992. Office of Management and Budget. Budget o f the U.S. Government, Fiscal Year 1993Washington, D.C.: U.S. Government Printing Office, 1992. Sonnefeld, Sally T., et al. “Projections of Nation al Health Expenditures through the Year 2000,” H ealth Care F inancing Review, vol. 13 (Fall 1991), pp. 1-27. Ifil Has the Long-Run Velocity of M2 Shifted? Evidence from the P* Model by Jeffrey J. Hallman and Richard G. Anderson Introduction Since early 1990, M2 has grown more slowly than suggested by its historical relationships with both income and opportunity cost, the lat ter measured relative to short-term market inter est rates. During the first part of this period (1990-91), although historical relationships with its opportunity cost suggested a significant decrease, M2 velocity remained quite close to its long-run average value of about 1.65. During 1992, M2 velocity increased sharply while its opportunity cost apparently decreased further. This behavior suggests that the long-run ve locity of M2, or V-Star (V*), may have risen, per haps as a result of changes in the money supply process, such as the stricter regulatory environ ment facing depository institutions. If V* has in deed increased, then the P-Star (P*) model, which assumes no change in M2’s long-run velocity, should have persistently underpredicted inflation over the last three years.We find, however, that the model has quite accurately predicted the de celeration of inflation since 1990. The paper also presents an extensive analysis, based on simulation of the P* model under a vari ety of alternative hypotheses regarding possible Jeffrey j. Hallman is an economist at the Federal Reserve Board of Governors, Washington, D.C., and Richard G. Anderson is a research officer at the Federal Reserve Bank of St. Louis. The authors wish to thank numerous colleagues at the Board and at the Federal Reserve Banks of Cleveland and St. Louis for helpful com ments and suggestions. shifts in long-run velocity, that provides little support for the view that V* has changed. Our findings reinforce other recent research conclud ing that the pickup in M2’s velocity may be largely explained by increases in an alternative opportunity cost measure based on long-term market rates.1 If correct, these results suggest that sluggish M2 growth over the last three years con tributed to both the slow pace of economic activity and the significant progress toward price stability. In addition, they suggest the potential for a rebound of M2 growth during 1993 as long-term rates fall and M2 velocity growth decelerates. I. The P* Model2 The P* model links the behavior of the price level to the growth of M2 by imposing two hypotheses on the equation of exchange, M VPQ: (i) real output Q t fluctuates around poten tial real output Q * over long periods, and (ii) ■ 1 See Feinman and Porter (1992). ■ 2 See Hallman, Porter, and Small (1991). F I G U R E 1 M2 Velocity and Opportunity Cost Ratio Percent deceleration of inflation as P t —>P]- Hallman, Porter, and Small (1991) show that the P* model can be derived as the reduced form of a special case of the expectations-augmented Phillips curve. In this case, changes in the inflation rate follow a simple autoregressive process aug mented by the lagged price gap, p t -p*t: (2 ) 4 i SOURCE: Authors’ calculations. velocity Vt has an equilibrium level V*, inde pendent of time, that it tracks in the long run.3 With these assumptions, P * is defined as the long-run equilibrium price level that could be supported by the current level of the money stock (M t) if current output (Q t ) settled down to this period’s level of potential output ( Q *): ( 1) p* — t M t V* Q) Our assumptions regarding Vt and Q t imply that if money remains fixed at M t, then Pt will fluctuate around P*. For policymakers, P* provides an index in each period t of the cumulative long-run im pact of money on the price level. The difference between the current price level and P* can pro vide a leading indicator of future acceleration or where lower-case letters denote natural logs, n , is the inflation rate, and Ai t , is the quarterly change in the inflation rate. The existence of P* depends critically on the validity of assumptions (i) and (ii). The assumption that real output fluc tuates around a growing level of potential out put is not controversial; indeed, measures of potential output are often constructed so as to ensure the validity of this assumption. The velocity assumption is more open to dispute.4 The constant velocity assumption of the P* model is motivated, in part, by the tendency of M2’s velocity since 1955 to fluctuate around 1.65, trending neither up nor down (see figure 1). Velocity at times has remained above its long-run average for several years, and recent increases do not appear particularly unusual in this respect. The assumption is likewise moti vated by the close historical correspondence be tween M2’s velocity and its opportunity cost that prevailed through 1989, also shown in fig ure 1. During this period, sustained deviations of velocity from its long-nm average tended to be accompanied by comparable deviations of oppor tunity cost from its long-run average.6 The ten dency for M2 opportunity cost to return to its long-nin average provided an economic rationale for M2 velocity to do the same. Empirical models ■ 4 See, for example, Kuttner (1990) and Pecchenino and Rasche (1990). As Pecchenino and Rasche note, the inflation dynamics in Kuttner’s paper are incorrect because he confuses Q and Q* in the P* model. ■ 5 The opportunity cost shown equals the difference between the three-month Treasury bill rate (on an annualized coupon-equivalent basis) and a share-weighted average of the own rates paid on the com ponents of M2. See Moore, Porter, and Small (1991). Note that their series begins in 1959. ■ ■ 3 Equivalent alternative assumptions are (i) M2 velocity is a sta tionary stochastic process, or (ii) all shocks to the level of M2 velocity are transitory. In a nonstochastic model, P will converge to P*. For a statement of the modern quantity theory, see Dewald (1988). For antece http://fraser.stlouisfed.org/ dents to P*. see Humphrey (1989). Federal Reserve Bank of St. Louis 6 M2’s velocity and its opportunity cost have moved in opposite directions before. In 1960, velocity rose while opportunity cost fell; in 1983, velocity fell while opportunity cost rose. The duration of the most recent divergence appears unusual, however. Note that the vertical dis tance between the lines in the figure is not meaningful. m of M2’s opportunity cost developed by Federal Reserve Board staff during the 1980s seemed to confirm this long-run behavior.7 During the past three years, however, M2’s velocity and opportu nity cost have diverged sharply, with the former increasing as the latter has decreased. This diver gence raises the question of whether equilibrium velocity has indeed changed 8 II. Using the P* Model to Identify Changes in V* While the P* model was originally offered as a link between inflation and money growth, its in verse provides a test of one of its primary assump tions: the constancy of long-run M2 velocity.9 If the long-run velocity of M2 has in fact increased during the last three years, predictions of inflation from the original P* model (which assumes that long-run velocity has not changed) should be in ferior to predictions from a model that incorpo rates the “true” change in V * . This simple insight immediately suggests a testing strategy for evaluat ing alternative hypotheses regarding putative shifts in V * : Construct the various P* time series corresponding to alternative velocity assumptions; use a battery of goodness-of-fit and forecast ac curacy tests to compare the relative forecasting performance of the model under the alternative as sumptions; and accept the velocity assumption(s) most consistent with the data or, in other words, the one that yields the best model forecasting ■ 7 See, for example, Moore, Porter, and Small (1991). These models typically assumed the existence of a long-run fixed spread between the offer ing rate on a particular type of deposit and a short-term risk-free market rate (for example, the three-month Treasury bill). A similar assumption was made for money market mutual fund yields. The size of the equilibrium spread pre sumably depended on both demand and supply factors, including regulatory (capital) requirements facing the intermediary, deposit insurance premiums, and the liquidity of the deposit. ■ 8 It also raises the possibility that M 2’s opportunity cost was incor rectly measured. Recent research by other Board staff suggests that this may have been the case. A new opportunity cost measure that includes a long-term Treasury rate and a rate on consumer loans appears to track M2 velocity during 1984-92. These models are highly preliminary, how ever, and do not feature the long-run error-correction behavior of pre vious Board staff models. See Feinman and Porter (1992). ■ 9 The antecedents discussed by Humphrey (1989) also view P*- type models primarily as models of the inflation rate. A constant (or very slowly changing) velocity of money is assumed almost without mention. This is reminiscent of Irving Fisher’s quantity theory model. See Laidler http://fraser.stlouisfed.org/ (1985), chapter 5. Federal Reserve Bank of St. Louis performance.10 Suppose, for example, we learn that V * increased 6 percent in mid-1989, to 1.75 from 1.65, and has remained at that value. Using equation (1), we can construct an alternative time series of P* values that will also have shifted up by 6 percent, consistent with the higher velocity. Use of this new, more accurate measure of the equilibrium price level should improve the accu racy of inflation forecasts from the P* model. Although the divergence of velocity and op portunity cost shown in figure 1 suggests that V * may have increased, the curves tell us little about the precise form of the change. In our analysis, we consider five alternative hypotheses concerning V * during 1989-92: • It remained at its 1955-89 average value of 1.65. • It increased 6 percent in 1989:IIIQ- This quarter was chosen based on the presence of two high-visibility events that marked the end of a dec ade of regulatory forbearance for undercapitalized depository institutions: passage of the Financial In stitutions Reform, Recovery, and Enforcement Act (FIRREA) and the first resolutions of insolvent thrifts by the Resolution Trust Corporation. The depository sector, facing a stricter regulatory envi ronment and the need to improve its capital ratios, might be expected to grow more slowly or even to contract as a result. • It shifted upward by 2Va percent each year in 1990 and 1991 and by 2Vi percent in 1992. These are approximately the size of the forecast errors from the Federal Reserve Board staffs mod el of M2 demand based on income and M2’s op portunity cost relative to short-term market rates.11 • It began increasing at a IV2 percent annual rate in 1990:IQ. • It began decreasing at a V2percent annual rate in 1990:IQ. This scenario is included for two reasons. First, it directly challenges the widely held conjecture that structural changes affecting depository intermediation during the past three years must have increased M2’s long-run velocity. Second, it admits the possibility that the decrease in the inflation rate since 1989 has occurred largely as might have been expected (and perhaps even a bit more rapidly than expected), given the slow growth of M2 and the significant output gap. ■ 10 This is somewhat more complicated than stated, since the tests are non-nested. Below, we generate the empirical sampling distribution for each individual statistic. ■ 11 See Feinman and Porter (1992), figure 1. F I G U R E 2 Simulated Inflation Rates from Alternative V* Hypotheses Percent 7 Actual inflation rate Unchanged One-time shift in 1989:IIIQ Money-demand model Increasing trend Decreasing trend 1 I I I 1 II 1988 1989 I II I I I I I 1 I I 1990 1991 I I 1 1 L...1 1992 1993 NOTE: First simulated value under all five hypotheses is 1 9 8 9 :IIIQ . SOURCE: Authors’ calculations. Each of the V* hypotheses suggests a corres ponding P * series, constructed according to equation (1) using the hypothesized V*r Under the null hypothesis that V* has not changed from its 1955-89 level, the inflation-rate path for each P* series is given by equation (2). Actual data are used through 1992:IVQ.12 Under the five alternative V* assumptions, dy namic simulation of the P* model, shown in equation (2), yields the five inflation-rate paths shown in figure 2. Each simulation begins in 1989:IIIQ and is nonstochastic; that is, all of the Et error terms in equation (2) are set equal to zero over the simulation period. During the past three years, the actual inflation rate generally has been between the rates suggested by the un changed or declining V* scenarios and those suggested by a trend increase in V*. O n balance, the inflation rate appears to have most closely followed the path given by the constant V* hypothesis, at least through 1992:IIIQ. Inflation in 1992:IVQ, however, was higher than forecast by the P* model with V* unchanged. The nonstochastic simulations shown in fig ure 2, though suggestive of an unchanged longhttp://fraser.stlouisfed.org/ run M2 velocity, are not capable of answering Federal Reserve Bank of St. Louis our question about a shift in equilibrium veloc ity. In particular, the simulations assume that no stochastic factors influence the evolution of the inflation rate (ef = 0 for all t), including possible random fluctuations in M2 velocity, when M2 velocity in fact has a relatively high variance. From a statistical viewpoint, the data shown in figure 2 represent only one “draw” from the uni verse of ways velocity and inflation might have evolved under each alternative hypothesis regard ing V*. An adequate test must incorporate the in herent randomness and variability of economic variables. Furthennore, comparing the perform ance of several models (or, in our case, the same model using alternative estimates of P* ) solely on observed, actual data leaves unanswered a num ber of interesting questions, such as: • Suppose, in fact, that inflation accelerates in 1993- How long might it take before incoming data reveal a change in V? At what point, if any, will the statistical evidence compel us to reject the hypothesis that the long-mn velocity of M2 has not changed? • Which hypothesis regarding M2 velocity is believed by financial market participants? Are further decreases in long-term market interest rates waiting for clearer signals regarding future M2 velocity? We conducted a simulation study to investi gate these issues as well as the overall accepta bility of the V* hypotheses.13 Our simulation design generates, for each of the five V* hypoth eses, 1,000 simulated paths for Pt from 1989:IIIQ through 1994:IVQ. Each path is the result of a stochastic simulation of the P* model under the appropriate velocity hypothesis. The stochastic in novations e, for the simulations are drawn from a normal distribution scaled to have a mean of zero and a standard deviation of about two-thirds of 1 percent at an annual rate. This corresponds to the smaller post-1986 variance of the residuals from the P model when estimated over 1960:IIIQ-1992:IIIQ, as shown in figure 3- (A formal statistical test strongly rejects equality of the variance of the residuals be fore and after 1986.) Although the reason for this smaller variance is not apparent, it may be due to less variance in the expected inflation rate after 1986. Our simulations assume that the future ■ 12 After 1992:IVQ, M2 and 0 * are assumed to grow at annual rates of 4.5 percent and 2.5 percent, respectively. ■ 13 The simulation methodology also allows us to address some is sues of interest mainly to econometricians, such as assessing how well various statistics perform in detecting the kinds of changes in which we are interested. FI GURE Model Residuals Percent 4 3 3 and ru’12 are based on the last four, eight, and twelve forecast errors, respectively. • B ino m ial tests for an unusually high num ber of positive forecast errors, due to the assumed V* being too small. Bn4, bn8, and b n l2 are based on the last four, eight, and twelve forecast errors, respectively. The statistics are discussed further in the appen dix. For each of the 1,000 replications, we calcu lated and stored the values of the statistics for 1 each quarter from 1990:IQ through 1994:IVQ. 0 For any particular quarter within our simula tion period, the degree of support for a V* hy -1 pothesis may be inferred by comparing the -2 values of the statistics in that quarter to the simu lated distributions of possible outcomes. The -3 simulated distributions indicate the range of -4 values of the statistics that could result from ran dom, unobserved influences.14 If the value of a -5 1960 1965 1970 1975 1980 1985 1990 1995 statistic falls outside the central area of the cor responding simulated distribution, we tend to SOURCE: Authors’ calculations. reject that particular hypothesis. Our results for 1992:IVQ are shown in table 1 and figure 4. Values of the test statistics calcu lated from data for 1992:IVQ, the most recent quarter for which we have preliminary gross variance of the random innovations will resemble domestic product (GDP) data, are shown in col the smaller post-1986 period. umn 2 of the table. Columns 3-7 display a count When the precise specification of alternative of the number of model replications (out of 1,000) hypotheses in a testing situation is uncertain, as it wherein a test statistic took on a value less than is for hypotheses regarding changes in V*, the that shown in the second column. The third col choice of an appropriate test statistic is difficult. umn, for example, summarizes our simulations Some hypotheses suggest tests for omitted dummy under the hypothesis that V* has not changed variables (such as a discrete shift in the level or a from its historical average value of 1.65. Each entry nascent time trend), while others suggest the use in the column shows the number of replications of more general tests based on forecast errors. for which the value of the statistic named in the Along each simulated path P*, we computed the first column was less than or equal to the 1992:IVQ values of 12 test statistics, including tests for omit value, shown in the second column. ted variables as well as tests for general misspecifiConsider, for example, the interpretation of cation based on one-step-ahead forecast errors. the lm shift statistic for 1992:IVQ as summarized Our statistics fall into four categories: by the first row of table 1. The value of this sta • Lagrange m ultiplier (LM) tests for an tistic calculated from 1992:IVQ data is 0.026. The omitted variable in equation (1). Lm shift tests third column indicates that the lm shift statistic for a post-1989:IIQ shift dummy variable, was less than 0.026 in 266 of the 1,000 replica Im trend for a time trend beginning in 1990:IQ, tions of the unchanged V* scenario. According and Im both for both the shift and trend. to this hypothesis, then, 0.026 appears to be • Chow tests for a change in the forecast neither unusually large nor small. In contrast, the error variance, relative to the variance of the dis entry in the fourth column tells us that observing turbance Et in the simulations, perhaps due to an lm shift statistic value as small as 0.026 would a change in V*. Ch4, ch8, and ch i 2 are based be highly unusual if V* had in fact increased by a on the last four, eight, and twelve forecast er one-time 6 percent shift in 1989:IIIQ. A value that rors, respectively. • Random walk tests for autocorrelation in the forecast errors due to misspecification of the model, including a structural change. Rw4, ru>8, 2 ■ 14 In other words, the distributions shown are the empirical sam pling distributions of the statistics. RH TABLE 1 Observed Values of Test Statistics in 1992:IVQ and Cumulative Frequency of Occurrence of those Values in Simulation Number of Replications wherein Value of Statistic Is Less than in 1992:IVQ Test Statistics and 1992:IVQ Values Statistic V* Hypothesis Value HI No Change ( 2) (3) H2 One-Time Shift (4) 0.026 266 0 0.233 0.377 719 573 H3 Money-DemandModel Shift (5) H4 1Vi Percent Trend H5 -Vi Percent Trend (6) (7) 42 78 62 135 263 229 636 (1) LM tests Im shift Im trend Im botb Chow tests cb4 ch8 c h l2 4.10 12.9 15.0 626 879 753 278 351 84 227 530 388 429 752 611 576 866 727 Random walk tests rw4 rw8 rw l2 3.86 0.475 1.16 960 536 727 595 9 4 494 42 166 752 184 377 926 460 664 4 5 8 947:1,000 648: 869 816: 932 612:1,000 18: 136 16: 95 541:1,000 63: 274 254: 497 719:1,000 170: 473 431: 699 979:1,000 778: 940 895: 974 Binomial testsa bn4 bn8 b n l2 240 537 . a. The two values correspond to the value of the statistic being, respectively, either strictly less than, or less than or equal to, the value in column 2. NOTE: Each entry is the num ber o f replications out o f 1,000 trials. SOURCE: Authors’ calculations. low never occurred in 1,000 replications of the “6 percent shift” scenario. Table l ’s test statistics and simulation out comes are summarized in figure 4, with each panel corresponding to one of the 12 statistics. Each horizontal line segment in each panel rep resents the 1,000 replications of the P* model under one of the five alternative V* hypotheses, denoted H1-H5. A hypothesis regarding V* is judged more or less acceptable (in other words, consistent with the data) as the horizontal line segments for that hypothesis tend to be centered around the vertical dotted lines denoting the val ues of the statistics calculated from 1992 :IVQ data. Overall, the hypotheses that V* has not changed (H I) or has been decreasing slowly (H5) appear to be highly consistent with the data, with the 1992:IVQ value falling near the midpoint of the distribution of simulated values for a number of the statistics. The hypothesis of http://fraser.stlouisfed.org/a one-time shift in 1989:IIIQ (H2) is soundly Federal Reserve Bank of St. Louis rejected. The hypothesis that M2 velocity shifted as suggested by the Federal Reserve Board staffs money-demand model (H3) appears less consistent with the data than the hypothesis of a steady upward trend (H4), which seems fairly plausible. Neither of the trending V* hypotheses (H3 and H4) appear to be as consistent with the data as the unchanged and falling hypotheses (H I and H5), however. Market participants’ inflation expectations appear to reflect acceptance of a significant in creasing trend in M2 velocity, despite the decel eration of inflation over the past three years.15 The January Blue Chip consensus forecast, for example, calls for the GDP implicit price deflator ■ 15 Chairman Greenspan’s latest Humphrey—Hawkins report to the Congress in February of this year appears to endorse this view, as does the FOMC's reduction of its 1993 M2 target growth ranges. To avoid such bias, we use a Blue Chip forecast published before these were announced. Summary of Simulation Experiments for 1992:IVQ LM Tests C h o w Tests R a n d o m W alk Tests B in o m ia l Tests Imshift ch4 rw4 bn4 926 H5 -240 H5 -576 H5 — H4 135 H4 -429 H4 — 752 H5 H3 42 H3 227 H3 494 H3 H2 0 H2 - 2 78 H2 -595 H2 HI -- 960 HI 266 HI -626 HI 979 /iy H4 UI z r947 Value in 1992:IVQ = 0.026 Value in 1992:IVQ = 4.10 Value in 1992:IVQ = 3 i Value in 19 92:IVQ = 4 Imtrend cb 8 rw8 bti8 I I H5 -- 636 H5 H4 -263 H4 -752 H4 H3 -530 H3 H3 78 H2 H2 HI -719 -460 H5 •879 •42 H2 -536 H5 — H4 -229 537 H4 62 -727 H5 H3 611 -388 rw!2 bn 12 H2 HI HI Value in 1992:IVQ = 0.377 -166 H3 H2 -573 -664 ■377 H4 HI -753 04 Value in 1992:IVQ = 5 H5 H2 18 HI Value in 1992:IVQ = 0.475 c h !2 Imboth 1 /u H3 HI Value in 1992:IVQ = 12.9 H5 H4 184 H2 -351 HI Value in 1992:IVQ = 0.233 H3 866 — Value in 1992:IVQ = 15.0 H5 H4 4j>l H3 H2 -727 Value in 1992:IVQ = 1.16 16 g HI Value in 1992:IVQ = 8 NOTE: Each horizontal line represents 1,000 replications of the P* model under either H I, H2, H3, H4, or H5. Shown after each line is the n u m ber o f replications wherein the value o f the statistic is less than in 1992:IVQ. SOURCE: Table 1. H1-H5 correspond to columns 3-7 in the table. to increase at about a 2.7 percent rate during the first half of 1993, versus its 2.1 percent pace in the second half of 1992. The inconsistency be tween the paths of the price level implied by the Blue Chip forecast and the P* model with an unchanged V* is evident in table 2. Values of our test statistics calculated from projected val ues of Pt for 1993-HQ that are based on this forecast are shown in column 2.16 The entries in column 3 show that many of our statistics will reject the constant V* hypothesis if inflation fol lows the Blue Chip forecast. The complete set of test results is displayed in figure 5. Ignoring the Chow tests and the bn4 statistic, the trending V* hypotheses H3 and H4 appear fully consistent with the Blue Chip forecast.17 ■ 16 See Blue Chip EconomicIndicators, Sedona, Arizona, January 10,1993, p. 5. Initially, it may appear somewhat surprising that the statistical support for the constancy of V* is so sharply changed by inclusion of the two additional quarters from the Blue Chip consen sus forecast. The reason for this sensitivity is that the consensus inflation forecast is very different from the forecast suggested by the P* model with an unchanged V*. P* is currently more than 8 percent below Pt, so the P* inflation model — equation (2) — forecasts that inflation will con tinue to decelerate over the next several quarters from its 2.1 percent pace in 1992:IIH. The con sensus forecast, by contrast, predicts an acceler ation during the first half of 1993. The message of table 2 is that such an acceleration is highly unlikely unless equilibrium velocity has been trending up for some time and has escaped ■ 17 Neither the Chow tests nor the bn4 test has much power against the hypothesis being tested, as is evident from examination of table 3. EU T A B L E 2 Projected Values of Test Statistics in 1993:IIQ and Cumulative Frequency of Occurrence of those Values in Simulation Test Statistics and Number of Replications wherein Value of Statistic Is Less than Projected 1993:IIQ Value 1993:IIQ Values V* Hypothesis Statistic Value HI No Change H2 One-Time Shift H3 Money-DemandModel Shift H4 H5 1Vi Percent -V i Percent Trend Trend (1) (2) (3) (4) (5) (6) (7) 0.318 1.40 1.81 791 992 982 2 198 69 124 318 387 368 750 774 693 969 962 10.8 979 970 966 885 731 514 716 659 701 888 867 866 971 964 954 1,000 955 960 961 273 47 817 184 199 951 519 544 993 898 903 943:1,000 711:1,000 224: 540 34: 125 472:1,000 149: 439 97: 294 668:1,000 340: 687 279: 550 980:1,000 959: 992 931: 985 LM tests Imshift Imtrend Imboth Chow tests ch4 ch8 chi2 17.3 22.8 Random walk tests nv4 rw8 rwl2 8.91 3.54 3.63 Binomial testsa 4 6 8 hn4 bn8 bn 12 879: 965 824: 952 a. The two values correspond to the value of the statistic being, respectively, either strictly less than, or less than or equal to, the value in column 2. NOTE: Each entry is the num ber of replications out of 1,000 trials. SOURCE: Authors’ calculations. detection by our tests for 1992:IVQ.18 Such an acceleration of inflation would provide sig nificant evidence against the constancy of V*. III. Evaluating Alternative, Less Specific Hypotheses At this point, a true believer in higher equilibrium velocity will object that, while our approach most ly rejects the specific shifted and upward-trending ■ 18 Alternatively, it may be that the variance of the innovations has increased. One way to see how inference about the constancy of V* depends on the assumed variance of the innovation process is to note that, if the Blue Chip forecast is correct, the P* model’s 1993:IQ forecast of 1.2 percent will miss by about 1.6 percent. Since we have assumed an innovation standard error of 0.6 percent, this is about a two-and-one-halfstandard-deviation miss, which is unusual. If the innovation standard deviation were instead (say) 1.6 percent, the forecast error would be only about one standard deviation, which is not so odd. http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis V* hypotheses outlined above, this does not conclusively prove that V* has not changed. Simulations with slower growth trends in V* or ones that started later than 1990:IQ, for ex ample, might not be rejected. The objection has merit. Our experiments con sider only a few specific alternative hypotheses. To evaluate rigorously, using our stochastic simulation method, the evidence for or against a less specific hypothesis— such as “V* shifted some time in the late 1980s or early 1990s”— would re quire repeating our experiments using alternative models with shifts beginning in 1989:IVQ, and again with shifts beginning in 1990:IQ, and so on. The number of required simulations in creases even further if we allow for a number of trend growth rates, rather than the 1.5 percent annual V* growth used here. We can, however, address the issue indirectly. Our test statistics should be valuable in detecting Summary of Simulation Experiments for 1993:IIQ H5 H4 H3 H2 HI H5 H4 H3 H2 HI H5 H4 H3 H2 HI LM Tests C h o w Tests R a n d o m W alk Tests B in o m ia l Tests Imshift ch4 rw4 bn4 693 368 124 •791 971 H5 H4 H3 H2 HI 716 — 885 979 H5 H4 H3 H2 HI -993 - 95 1 817 -961 - 1000 •980 Hi> H4 H3 H2 HI -668 472 -711 943 Value in 1993:IIQ = 0.318 Value in 1993:IIQ = 10.8 Value in 1993:IIQ = 8.91 Value in IS 93:IIQ = 4 Imtrend ch 8 rw8 br18 — 969 H 5 750 H4 -3 1 8 H3 H2 198 ■992 HI •898 0 h H4 H3 " 0 H2 /31 970 HI 519 184 -273 955 959 H5 H4 H3 H2 HI iAr\ 879 Value in 1993:IIQ = 1.40 Value in 1993:IIQ = 17.3 Value in 1993:IIQ = 3.54 Value in 1993:IIQ = 6 Imboth c h l2 rw l 2 bn 12 — 962 H 5 -774 H4 387 H3 H2 69 -982 HI Value in 1993:IIQ = 1.81 T" ---954 — 866 -701 -514 966 H5 H4 H3 H2 HI Value in 1993:IIQ = 22.8 903 544 -199 47 Value in 1993:IIQ = 3.63 -960 H5 H4 H3 H2 HI 931 z /y I— Q7 34 824 Value in 1993:IIQ = 8 NOTE: Each horizontal line represents 1,000 replications of the P* model under either H I, H2, H3, H4, or H5. Shown after each line is the nu m ber of replications wherein the value o f the statistic is less than in 1993:IIQ. SOURCE: Table 2. H1-H5 correspond to columns 3-7 in the table. shifts in V* that begin in other time periods or that follow time paths with somewhat different shapes than those considered above. According to table 3, when V* is subjected to a one-time upward shift of 6 percent, within six quarters the best of our test statistics (using the 5 percent critical values shown in the appendix) reject the (false) hypothesis of an unchanged V* in more than half the replications. When V* is subjected to the less dramatic change of increasing at a 1V2 percent annual rate, all of our statistics have difficulty detecting this new trend growth until at least three years have passed, as shown in table 4. In part, this slow speed of detection is due to the high underlying variance of Vt. IV. Conclusion All models used for policy analysis require peri odic revalidation of their underlying assump http://fraser.stlouisfed.org/ O f particular concern in the P* model is Federal Reserve Bank tions. of St. Louis the assumed constancy of the long-run velocity of M2. Unfortunately, the long-run velocity of M2 is no more amenable to direct observation than other “long-run” variables in economic models. Two of our findings suggest that it has not changed, however. First, the deceleration of inflation over the past three years (at least through 1992:IIIQ) closely resembles the predictions of the P* model based on an unchanged long-run M2 velocity. Second, stochastic simulation of the P* model under five alternative hypotheses regard ing putative shifts in V* provides little evidence against the constant V* hypothesis, strong evi dence against the hypothesis of a one-time shift following the FIRREA legislation, and somewhat weaker evidence against the hypothesis of an upward trend during the past three years. These results suggest little reason for policy makers to abandon the P* model when seeking to understand the future adjustment of inflation to money growth. Comparison of the P* model’s in flation forecasts to the Blue Chip consensus fore TABLE 3 Number of Rejections of Hypothesis “V* Has Not Changed” When V* in Fact Increased 6 Percent in 1989:IIIQ LM Tests Im shift Im trend Chow Tests Im both ch4 ch8 Random Walk Tests c h i2 ru>4 rw 8 rw l2 405 521 431 765 Binomial Tests bn4 bn8 bn 12 68 472 657 789 0 0 0 0 0 0 305 264 0 0 305 264 848 855 826 933 0 0 0 0 229 615 740 727 229 615 545 489 1990 IQ HQ IIIQ IVQ 459 591 245 347 491 587 252 327 420 411 252 327 419 452 252 327 419 452 207 421 641 735 796 867 908 626 720 786 812 646 733 796 856 359 358 366 386 484 580 625 590 484 580 609 596 609 640 639 595 686 806 920 886 917 953 964 862 894 907 939 888 907 932 953 380 346 281 244 575 551 481 488 677 713 709 656 598 546 481 444 893 892 875 868 897 952 978 978 0 0 0 0 712 754 675 627 583 695 765 726 991 991 994 934 960 957 966 965 969 979 980 206 181 154 139 447 376 303 281 619 561 515 475 354 314 225 194 819 759 678 616 967 967 949 935 0 0 0 0 527 460 . 404 694 875 836 353 493 995 995 998 998 977 977 979 988 989 991 993 994 125 112 101 106 231 193 155 l6 l 424 379 305 258 180 l6 l 155 134 535 439 365 320 918 834 765 692 185 0 0 0 293 240 207 171 433 370 299 421 198 341 549 673 659 1991 IQ HQ IIIQ IVQ 1992 IQ HQ IIIQ IVQ 975 1993 IQ IIQ IIIQ IVQ 981 1994 IQ IIQ IIIQ IVQ 311 529 SOURCE: Authors’ calculations. cast suggests that market participants already believe that V* has shifted. In so doing, they ap parently are discounting evidence that the steep slope of the yield curve has induced portfolio substitution away from M2 (particularly small time deposits) and toward assets such as bond mutual funds. Our results also suggest a word of caution. The high variance of Vt means that attempts to distinguish changes in V* from short-run move ments in Vt are subject to a high degree of un certainty. Our tests almost surely would have identified by now a large, discrete shift in V* that occurred other than very recently. However, they might not yet have detected an emerging slow growth trend or a more rapid trend that http://fraser.stlouisfed.org/ started later than 1990:IQ. To the extent that in Federal Reserve Bank of St. Louis flation responds with a long and variable lag to changes in money growth, this uncertainty rein forces the need for caution and vigilance in the conduct of monetary policy. If M2’s long-run equilibrium velocity has in fact shifted or is trending up, continuing slow money growth may yield less progress toward price stability than expected. The stickiness and (later) halting decline of long-term interest rates during the re covery likely reflects, in part, views by financial market participants that V* has increased and that price stability is not yet the rule of the land. H I T A B L E 4 Number of Rejections of Hypothesis “V* Has Not Changed” When V* in Fact Began Growing at a V/z Percent Rate in 1990:IQ LM Tests Chow Tests Random Walk Tests Im shift Im trend Im both ch4 ch8 c h i2 rw 4 rw 8 rw l2 50 52 54 50 50 50 50 53 53 52 51 53 56 57 53 56 57 50 52 50 53 52 64 50 50 58 60 49 64 52 61 69 84 65 80 120 128 60 75 88 104 50 61 67 87 58 70 76 80 58 70 73 84 53 78 115 128 94 135 181 216 182 238 297 383 129 159 203 263 85 102 118 147 95 110 109 141 91 108 101 118 176 188 223 266 244 317 378 440 431 576 646 745 311 386 474 557 148 182 210 230 156 180 215 274 133 168 201 244 280 343 334 380 405 512 520 823 886 925 966 683 771 820 867 223 256 266 281 302 317 344 396 282 337 358 400 437 478 454 473 Binomial Tests bn4 bn8 50 54 57 79 0 0 0 0 0 0 0 0 39 26 39 26 99 78 62 120 0 0 0 0 15 62 80 96 15 62 47 39 83 149 203 249 0 0 0 0 122 143 174 224 99 79 90 120 565 641 344 503 579 675 0 0 0 0 271 313 356 384 158 450 530 290 700 740 774 787 780 814 874 927 374 0 0 0 417 458 481 506 339 406 460 757 b n l2 1990 IQ HQ IIIQ IVQ 53 38 66 1991 IQ HQ IIIQ IVQ 46 63 94 87 1992 IQ HQ IIIQ IVQ 141 213 277 360 1993 IQ HQ IIIQ IVQ 1994 IQ HQ IIIQ IVQ 569 645 677 SOURCE: Authors’ calculations. Appendix— The Test Statistics The 12 statistics calculated during the simulations for each quarter include tests for omitted variables and for properties of forecast errors.19 The first three statistics are LM tests for omitted variables in equation (1): Im shift tests for a post-1989:IIQ shift dummy, Im trend for a time trend beginning in ■ 19 To obtain forecast errors for the tests that need them, we esti mate the P* model (using the constant V* version of P*) for each quarter of the simulation period using the simulated Pi series running up through the previous quarter. A single-step forecast error for the quarter is computed and saved, the process is repeated for the next quarter, and so on. http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis 1990:IQ, and Im both for both simultaneously. An appropriate test for a 1989:IIIQ shift in equilibrium velocity can be formulated as a test for an omitted variable, where the omitted variable itself is a dum m y variable that equals zero until 1989:IIQ and one thereafter. To see this, notice that the variable p* in equation (1) is defined as p* = m2 + v* - q*, where lower case letters indicate natural logs. A shift or trend in v* translates directly into an equivalent shift or trend in p*. If a 6 percent increase in equilibrium velocity causes us to understate p* by 0.06, this can be handled in equation (1) by adding a constant term equal to -0.06 times a , the coefficient on p - p * . The rationale for the Im trend test is identical. m TABLE A- 1 95th Percentile of Empirical Sampling Distribution of 12 Test Statistics under Null Hypothesis that V* Is Unchanged from Its Long-Run Value LM Tests 1990 IQ HQ IIIQ IVQ 1991 IQ HQ IIIQ IVQ 1992 IQ HQ IIIQ IVQ 1993 IQ HQ IIIQ IVQ 1994 IQ HQ IIIQ IVQ Chow Tests Random Walk Tests Binomial Tests Im shift Im trend Im both ch4 ch8 c h i2 rw 4 rw 8 rw l2 0.83 0.87 0.86 0.86 0.89 0.87 0.85 0.81 1.31 1.38 1.29 1.34 8.05 9.42 9.82 9.99 12.72 13.78 13.88 14.38 15.02 14.86 15.73 17.70 3.12 4.01 3.94 3.72 2.25 2.23 2.76 3.10 1.28 1.34 1.93 2.16 3 4 4 4 6 6 5 6 6 7 7 8 0.89 0.94 0.92 0.94 0.93 0.89 0.85 0.88 1.45 1.42 1.41 1.38 10.39 10.00 9.97 9.64 14.64 15.31 15.53 15.98 19.31 19.67 19.95 19.98 4.26 4.21 3.78 3.82 3.65 4.12 3.74 4.33 2.68 2.90 3.08 3.60 4 4 4 4 6 6 6 6 9 8 8 9 0.98 0.90 0.89 0.87 0.84 0.83 0.84 0.76 1.39 1.44 1.42 1.33 9.52 9.48 9.64 9.25 15.85 15.56 16.23 15.52 19.70 20.63 21.27 21.64 3.56 3.79 3.70 3.40 4.13 3.74 3.60 3.31 4.03 3.89 3.80 3.75 4 4 4 4 6 6 6 6 8 9 9 9 0.90 0.82 0.81 0.80 0.82 0.72 0.77 0.72 1.37 1.38 1.36 1.31 9.30 9.11 9.37 9.29 15.21 15.51 15.66 14.81 21.51 21.49 21.51 21.01 3.62 3.27 3.79 3.72 3.55 3.26 3.17 3.10 3.72 3.26 3.30 3.17 4 4 4 4 6 6 6 6 9 8 8 9 0.76 0.78 0.76 0.78 0.67 0.68 0.67 0.63 1.25 1.23 1.27 1.26 9.73 9.10 9.01 9-31 14.87 14.93 15.17 14.82 20.82 20.49 20.65 20.59 3.36 3.27 3.44 3.65 2.96 2.91 2.91 3.05 2.87 3.15 2.98 2.75 3 4 4 4 6 6 6 6 9 9 9 8 bn4 bn8 b n !2 SOURCE: Authors’ calculations. Chow forecast tests have long been used to determine parameter constancy and are, in fact, tests of the constancy of variances. The idea is that if the process generating the data changes at time t but the model used by the forecaster does not, the forecast error variance will in crease. The utility of the test is limited by its im plicit assumption that the variance of the true disturbances is constant. Our three Chow statis tics — ch4, ch8, and c h i2 — are calculated as the sum of the latest four, eight, or twelve squared forecast errors, respectively, divided by the variance of the simulation innovations. The rw statistics are our own invention, http://fraser.stlouisfed.org/ motivated by the idea that a persistent misspeciFederal Reserve Bank of St. Louis fication of the P* model, such as would result from a shift or trend in V*, will lead to positive autocorrelation in the forecast errors. The var iance of the sum of K consecutive forecast er rors will then be much larger than just K times the innovation variance. The rw4 statistic is the square of the sum of the four most recent fore cast errors, divided by four times the innovation variance; rw8 and rw l2 are analogous. An rw statistic can be written as the sum of a Chow sta tistic plus a term that measures autocorrelation in the forecast errors. Thus, we expect the rw test to be more powerful than the correspond ing Chow test when the alternative hypothesis involves positive forecast error autocorrelation. The binomial statistics (bn4, bn8, and bn 12) are simple counts of the number of positive forecast errors made over the corresponding inter vals. A correctly specified model should, on aver age, give about the same number of positive and negative forecast errors. The estimated coefficient in equation (2) is negative, so if V* and P* are un derstated, we would expect to see an inordinately high number of positive forecast errors. Table A-l shows the 95th percentile of the 12 statistics’ sampling distributions, based on 1,000 replications, under the null hypothesis that V* has not changed from its 1955-89 value. The number 0.87 in the 1992:IVQ row and Im shift column, for example, indicates that the Im shiftstatistic for 1992:IVQ was less than or equal to 0.87 in 950 of the 1,000 replications of the constant V* model. References Dewald, William G. “Monetarism Is Dead; Long Live the Quantity Theory,” Federal Reserve Bank of St. Louis, Review, vol. 70, no. 4 (July/August 1988), pp. 3-18. Feinman, Joshua J., and Richard D. Porter. “The Continuing Weakness in M2,” Board of Governors of the Federal Reserve System, Finance and Economics Discussion Paper No. 209, September 1992. Hallman, Jeffrey J., Richard D. Porter, and David H. Small. “Is the Price Level Tied to the M2 Monetary Aggregate in the Long Run?” Am eri can Econom ic Revieiv, vol. 81, no. 4 (Sep tember 1991), pp. 841-58. Humphrey, Thomas M. “Precursors of the P-Star Model,” Federal Reserve Bank of Richmond, Econom ic Review, vol. 75, no. 4 (July/August 1989), pp. 3-9. Kuttner, Kenneth N. “Inflation and the Growth Rate of Money,” Federal Reserve Bank of Chicago, Econom ic Perspectives, vol. 14, no. 1 (January/February 1990), pp. 2-11. Laidler, David E.W. The D em and fo r Money: Theories a n d Evidence. New York: Harper and Row, 1985. Moore, George R., Richard D. Porter, and David H. Small. “Modeling the Disaggregated De mands for M2 and M l: The U.S. Experience in the 1980s,” in Peter Hooper et al., eds., F in a n c ia l Sectors in Open Economies. Wash ington, D.C.: Board of Governors of the Fed eral Reserve System, 1991, pp. 21-105. Pecchenino, R.A., and Robert H. Rasche. “P* Type Models: Evaluation and Forecasts,” Na tional Bureau of Economic Research, W ork ing Paper No. 3406, August 1990. Q Examining the Microfoundations of Market Incentives for Asset-Backed Lending by Charles T. Carlstrom and Katherine A. Samolyk Introduction The past two decades have witnessed a virtual revolution in financial intermediation. One innova tion is securitization: the packaging of loans into pools that are funded by marketable securities. At the same time, the selling of individual loans has itself grown tremendously over this period. While individual loans are primarily sold to other depos itory institutions, securitization involves the sales of securities to nonbank investors as well. Both loan sales and securitized loan pools are broadly identified as asset-backed lending. A financial asset is a claim to future cash flows as stipulated by the issuer. What distinguishes asset-backed lending is that the securities involved are backed by specific financial assets and then sold. Alternatively, these financial assets might have been pooled and funded by issuing general claims on the firm. Instead, when a loan is either securitized or sold individually, it is funded sepa rately rather than with the other assets on the bal ance sheet of the loan originator.1 Hence, loan sales and securitization, from the perspective of the seller of the asset-backed securities, are a means of off-balance-sheetfinance. http://fraser.stlouisfed.org/ The proliferation of asset-backed lending has been commonly viewed as a response to com Federal Reserve Bank of St. Louis Charles T. Carlstrom and Katherine A. Samolyk are econ omists at the Federal Reserve Bank of Cleveland. The authors wish to thank Joseph G. Haubrich, Anjan Thakor, and James B. Thomson for helpful comments. petitive and regulatory pressures, which have prompted institutions to participate in credit markets in ways that are not directly reflected on their balance sheets. In particular, capital re quirements are cited as reducing the profitability of funding certain investments on-balance-sheet with deposit liabilities. However, nonbank firms that are not subject to the regulations associated with the federal safety net are also engaging in asset-backed financing. This indicates that there are important nonregulatory incentives for loan sales and securitization. Asset-backed lending has become an impor tant mode of funding for particular types of credit. Though depository institutions are the primary originators of home mortgages, more than 40 per cent of these claims are ultimately financed through the government-sponsored secondary mortgage market. In the past several years, how ever, asset securitization has spread beyond government-sponsored sales of mortgage-backed ■ 1 Although securitized loan pools are funded separately, they are frequently sold with some type of recourse, which means that they are partially backed by the general claims of the firm that originated the loan. securities to include private pools that are backed by increasingly diverse types of loans, from credit-card receivables to Third World debt. Cur rently, more than 15 percent of consumer install ment credit is funded through securitization.2 The evolution of financial market innovations in tandem with changing banking regulations makes it difficult to assess what is driving the trends in asset-backed markets. Because we wish to evaluate why asset-backed lending occurs in the absence of regulations, we examine how successful economists have been in applying formal models to this phenomenon. Although off-balance-sheet funding can arise for either market-based or regulatory-based reasons, we focus on four papers that attempt to model assetbacked lending in the absence of governmentsponsored insurance and regulations. We first outline the general nature of inter mediation and describe asset-backed markets in this context. Information costs have long been viewed as a rationale for financial intermedia tion. The literature on asset-backed lending has picked up on this theme to argue that loan sales and securitization are also best understood as a means of minimizing information costs. There fore, in order to understand some of the models that have attempted to formalize asset-backed lending, we first discuss several models of finan cial contracting under imperfect information, which have been useful in characterizing the roles that financial intermediaries play in chan neling credit.3 Finally, we analyze how existing government policies may affect the incentives for firms, primarily banks and thrifts, to engage in these activities. I. An Overview of Interm ediation In a decentralized economy with significant in formation and transaction costs, the financial sector affects how resources are channeled from lenders to borrowers. As financial conduits, in termediaries pool lenders’ resources to fund a portfolio of claims on many, often diverse, bor rowers. In doing so, intermediaries are said to conduct indirectfin an ce , allowing them to issue indirect claims with cash flows that differ in vary ing degrees from those of the borrowers. Thus, 2 See Federal Reserve Bulletin, Domestic Fi nancial Statistics, Table 1.55, Consumer Installment Credit, March 1993. intermediaries perform asset transform ation in making their investment and funding choices. To the extent that information is costly to ob tain, financial contracts and institutions also can reduce the information costs associated with channeling resources to the most productive in vestment opportunities. Thus, intermediation yields more attractive portfolio choices for inves tors while facilitating a more efficient flow of credit to borrowers. Interm ediation and Asset Transformation Three of the types of asset transformation pro duced by intermediaries are 1) denomination transformation, 2) credit risk transformation, and 3) maturity transformation. How effectively these methods can mitigate information costs is an important part of our subsequent analysis. Denomination transformation allows inter mediaries to lend to borrowers with large credit needs by issuing smaller-denomination claims to many savers. For example, mutual funds that invest in government bonds and Treasury bills pool the funds of a group of small investors to fund a portfolio of relatively similar claims. Denomination transformation also allows small savers to diversify by enabling them to hold a wider variety of investments. Credit risk transformation pools the resources of many lenders to fund several projects. This al lows intermediaries to diversify the risks of the assets in their portfolios, and thus to issue in direct claims to investors with a more predict able return than the individual assets being funded. This is the main role of stock or bond mutual funds, although most intermediaries engage in credit risk diversification. Finally, intermediaries also perform maturity transformation by issuing indirect claims that offer a pattern of promised cash flows different from those promised by borrowers. Banks and thrifts are noted for the degree of maturity trans formation in their portfolios. They fund mediumand long-term projects by issuing short-term liquid deposits that serve as close substitutes for legal tender.4 Contractual savings institutions, such as insurance companies and pension funds, ■ ■ 3 Two important papers surveying this literature are Gertler (1988) http://fraser.stlouisfed.org/ and Bhattacharya and Thakor (1991). Federal Reserve Bank of St. Louis ■ 4 McCulloch (1981) emphasizes that this degree of maturity trans formation is actually “misintermediation” that reflects the regulatory in centives for banks to assume credit risks as well as the risk associated with mismatching the durations of their assets and liabilities. produce a very different sort of cash flow trans formation. They fund portfolios of assets by sell ing contracts promising cash flows that are contingent on specific events, such as property loss, death, or retirement. Much of the intermediation associated with these types of asset transformation channels funds to borrowers who place debt or equity directly in credit markets. A distinguishing characteristic of some intermediaries is that they specialize in lend ing to borrowers who would find it prohibitively costly to obtain funds through direct market place ments because of the relative costs associated with screening, monitoring, and servicing their claims. Depository institutions and finance compa nies, for example, profit by developing a compara tive advantage in lending to small or information intensive borrowers. Thus, some intermediaries are special in the sense that they provide lenders with new investment opportunities— that is, they are asset originators. nonbank investors. These pools are generally originated by large firms. From the perspective of the pool originator, however, nonmortgage securitization is basically a means of separating the financing of certain assets from that of its general portfolio. Finally, securitization of mortgages takes place in the secondary market in order to fund pools of insured mortgages. These pools in clude claims from many, often geographically diverse, mortgage originators. This form of se curitization simultaneously creates a pool of sim ilar loans (mortgages) purchased from loan originators in different localities. Hence, a unique characteristic of mortgage-backed securities is that they are collateralized by loans from various financial firms. Loan Sales versus Securitization A major difference between loan sales and An Overview of Asset-Backed Markets In contrast to funding a portfolio of assets by the issue of unsecured claims, asset-backed lending is an alternative funding mode by which an asset or set of assets is sold by its originator. We use the term asset-backed lend ing to refer to both securitization and individual loan sales. A loan sale is usually made by a bank to another bank, and involves no asset pooling in and of itself.’ However, the process of making loans marketable, by increasing the access of other lenders to investment opportunities, can improve the allocation of credit. Loan sales in volve transactions between two (or more) finan cial institutions, whereas securitization generally involves the sale of claims (against the securi tized asset portfolio) to individual investors who hold these in their portfolios for investment pur poses. Consequently, securitized claims are priced like other capital-market instruments, but loan sales are priced based on bilateral (multi lateral) negotiations. Alternatively, nonmortgage securitization usually takes the form of a bank or nonbank firm funding a pool of similar assets by forming a subsidiary that markets claims to the pool to ■ 5 For a comprehensive overview of the loan sales market, see Gor http://fraser.stlouisfed.org/ ton and Haubrich (1990). Federal Reserve Bank of St. Louis securitization is that loan sales usually provide no recourse for the party buying the loan. Most view this as the result of regulators’ treatment of loan sales in their assessment of capital ade quacy requirements for depository institutions. Banks and thrifts are not required to hold capi tal against loans sold, except for those sold with recourse, which are treated as if they are onbalance-sheet items in determining capital ade quacy. Thus, given the incentives to maximize leverage, these institutions tend to sell loans without recourse to truly “get them off the regulatory books.” Securitization, on the other hand, is generally associated with the provision of some form of credit enhancement that increases the market ability of the asset-backed securities. One common form of enhancement for securitized assets is backing by a bank-issued standby letter of credit (SLC). For a stipulated fee, banks issue SLCs, which are promises to insure the purchasing party up to a prespecified amount for losses incurred on the securitized loans. Before a loan pool is funded, both the loans and the bank issuing the SLC are rated. Because the rating of the pool is affected by the rating of the bank issuing the guarantee, the extent to which this method of credit enhance ment is used is limited. Moreover, to avoid regu lated capital assessments, a bank securitizing a pool of loans usually does not issue the creditenhancing SLC. Thus, the originator of the pool is generally not also its guarantor. An increasingly popular enhancement, the cash-collateral-account method, has the pool originator covering potential losses with cash placed in an escrow account. Another method to enhance loan quality is to overcollateralize the loan pool. That is, extra loans are included in the pool so that the value of the loans exceeds the value of the securities issued to fund it. Why Fund OffBaiance-Sheet? Given the attributes of asset pooling, it is natural to question the benefit of funding a loan or pool of loans off-balance-sheet. The answer, of course, is that this method is more efficient— less expensive— than on-balance-sheet fund ing. As we have asserted, asset-backed lending is commonly viewed as a response to both regu latory costs and market incentives. In its early years, regulations were clearly an important factor motivating securitization via the secondary mortgage market.6 Regulated branch ing restrictions in tandem with information costs caused banks and thrifts to operate in relatively localized markets. The government-sponsored secondary mortgage markets allowed these institu tions to hold portfolios from many different parts of the country. These regulatory restrictions are less important today. This suggests that informa tion costs are becoming the more relevant deter minant of interregional lending. A fundamental role of intermediation is to produce the information involved in channeling credit in the most cost-effective way. In particular, lenders do not always have good information about the risk and return of borrowers’ investment opportunities. Intermediaries specialize in produc ing this information, as well as in structuring and servicing contracts. Therefore, in order to under stand why off-balance-sheet funding may be more efficient, it is useful to examine the roles of both financial contracts and intermediation in mitigating information costs. Here, the primary focus is on market incen tives— specifically due to information costs— as a motive for asset-backed lending. In the follow ing section, we discuss several models of finan cial contracting and intermediation. We then proceed to examine why asset originators might choose asset-backed lending as an alternative to on-balance-sheet funding. 6 See Pavel (1986) for a comprehensive description of the histori http://fraser.stlouisfed.org/ cal evolution of this market. Federal Reserve Bank of St. Louis ■ II. Financial Structure in Response to Information Costs Even in a world where there is complete informa tion about available investment opportunities, credit intermediation can occur if individuals without wealth have more profitable projects than do those with greater financial resources. How ever, while intermediation can help in diversifying the portfolios of the individuals supplying finan cial resources, the nature of the claim on these in vestment projects is uncertain. In particular, as Modigliani and Miller (1958) state, it is not clear why a project should be funded via a debt con tract, which stipulates a predetermined promised cash flow and default (should that cash flow not be met), versus an equity contract, which prom ises only to pay a cash flow that is contingent on the project’s return— precluding the event of de fault. Modigliani and Miller show that in a world without taxes, transaction costs, and information costs, entrepreneurs would be indifferent between funding projects with debt or equity. Debt versus Equity Contracts Information costs thus play an important role in explaining the stnicture of the contracts between borrowers and lenders that we observe in reality. One model of financial contracting under imper fect infonnation is presented in Townsend (1979). He demonstrates that when it is costly for lenders to monitor the performance of a borrower’s project, debt contracts allow lenders to minimize monitoring costs.7 In his model, borrowers can observe the proceeds of their in vestment opportunities, while lenders can do so only by paying a fee. In this setting, an equitytype contract stipulating a payoff that always depends on the project’s realization implies that investors will always have to expend resources to monitor the project’s outcome. Alternatively, debt contracts minimize these monitoring costs by specifying a contractual inter est payment to lenders. Borrowers pay this pre specified amount except when default is declared. In that situation, lenders receive the realized value of the project (or firm), which they must ascertain ■ 7 This suggests that debt would be preferred to equity. One reason equity might be preferred is if bondholders cannot observe the riskiness of the investments undertaken by the firm ’s management. In that situa tion, the investments undertaken w ill be too risky, which transfers wealth from bondholders to equityholders. by incurring monitoring costs. Here, debt con tracts minimize monitoring costs because lenders must monitor investment outcomes only in the event of borrowers’ default.8 Information Costs and Credit Risk Transformation One function of financial intermediation, as mentioned earlier, is to pool assets in order to reduce portfolio risks, thus enabling investors with limited wealth to hold a diversified portfolio. Another, indirect advantage of diver sification is that it helps to minimize information costs by decreasing the need for investors to monitor privately observed portfolio risks. Diamond (1984) examines how asset diversi fication by banks mitigates the need for depositors to monitor the performance of bank investments. He describes a world in which information about realized project returns is costly. If many lenders are needed to fund one borrower, an intermediary could group these lenders to fund the project. However, because the project’s return is costly to observe, each lender would in general have to monitor the intermediary’s investment. Diamond demonstrates that by diversifying across many projects, an intermediary can decrease the variability of the return on its port folio, and thus the need for lenders to monitor the performance of the portfolio. Depositors in essence loan funds to the bank in exchange for debt contracts. A reduction in portfolio risks lowers expected monitoring costs by reducing the probability that the firm will default on its liabilities by not paying depositors their stipu lated return. In the extreme case, complete diversification of asset returns eliminates portfo lio risk and thus the need for depositors to m on itor the bank. Hence, Diamond describes how asset pooling allows the monitoring function to be delegated to intermediaries.9 ■ 8 This result is predicated on the assumption of deterministic auditing. That is, auditing occurs with a probability of either one or zero. Mookherjee and Png (1989) show that, in general, random auditing will be optimal. That is, even when bankruptcy occurs, the probability of being audited is less than one. ■ 9 Ramakrishnan and Thakor (1984) show that financial intermedi aries will also arise with ex ante monitoring costs. Diamond’s paper as http://fraser.stlouisfed.org/ sumes ex post monitoring costs. Federal Reserve Bank of St. Louis III. Asset-Backed Lending as a Funding Mode Diamond’s analysis illustrates an interesting point, but in more realistic settings, firms may be limited in how much they can benefit from asset pooling. This restriction is useful to consider in examining why loan sales and securitization may be efficient ways of funding certain invest ments. Asset-backed lending in its most general sense is the sale of an asset by its originator, which separates the financing of the asset from that of the originator’s portfolio. Imperfect information about the portfolio choices of intermediaries can help to explain market-based incentives for asset-backed lend ing. The first two papers we discuss below cite the inability of localized or specialized banks to diversify portfolio returns as a rationale for fi nancial firms to engage in both loan sales and securitization. The models developed in these papers formalize this rationale, motivating assetbacked lending as a means for local borrowers to tap into nonlocal sources of funds. The second two models of asset-backed lending emphasize the differences in the information available to in termediaries versus the individuals who hold their debt prior to investment choices. These models formalize asset-backed lending as a means of collateralizing, thus enabling investors to obtain financing terms that better reflect the underlying quality of the projects being funded. Portfolio Risks and Capital Constraints While perfect diversification removes the need to monitor imperfectly observed portfolio risks, imperfect diversification creates the need for a more complicated financial structure. For exam ple, when banks cannot perfectly diversify risks, the amount of their equity capital assumes greater importance. Without sufficient equity capital, banks may be unable to attract funding in order to finance risky investments. By buffering potential portfolio losses, equity capital serves as an alterna tive means of mitigating the need for lenders to monitor an intermediary: It cushions portfolio losses and thus protects depositors. Bemanke and Gertler (1987) and Samolyk (1989a,b) show that when depositors’ costs of monitoring an institution are prohibitive, inter mediaries may face market-imposed capital con straints on the risks associated with their portfolio choices. Capital inadequacy arises when a bank is unable to attract funds to finance profitable in vestments because it has inadequate capital to absorb possible portfolio losses. The key to this result is that it is assumed to be extremely costly for depositors to monitor the out come of a bank’s portfolio. Depositors recognize that banks have the incentive to report large losses on their risky assets, in effect claiming that they are unable to meet depositors’ claims. Hence, banks will not be able to attract depositors un less they have sufficient capital to cover poten tial portfolio losses on risky investments.10 Limits to the Benefits of On-Balance-Sheet Interm ediation Capital constraints can arise because banks are both unable and unwilling to diversify their port folios adequately. Government policies have affected the incentives for intermediaries— espe cially banks and thrifts— to manage portfolio risks prudently. Portfolio and branching restrictions have limited the ability of banks and thrifts to di versify credit risks as well as the risks associated with maturity transformation. Regulatory limits on the types of depository lending, such as the “Qual ified Thrift Lender Test,” also constrain portfolio diversification.11 Finally, the provision of federal ly sponsored deposit insurance creates moral haz ard problems in both the management of credit risks and the interest-rate risks associated with maturity transformation. These policies reduce the potential for depositors (and regulators) to dele gate the monitoring function. Given the partial deregulation of the banking industry, these restrictions are probably not as important an impediment to diversification as they once were. Ironically, a major factor limit ing intermediaries from diversifying and hence minimizing information costs is the very costs of identifying, monitoring, and funding borrowers that make financial contracts and intermediation important. These costs may cause intermedi aries to specialize in lending to certain types of borrowers (industry versus consumers) or to borrowers in certain regions. Asset-Backed Lending as a Response to Localized Capital Constraints Carlstrom and Samolyk (1993) present a model in which capital constraints motivate one rationale for off-balance-sheet lending. Their model predicts that loan sales occur as a response to differences in project returns across regions that arise when some regions are capital constrained and others are not. Similar to the model used by Samolyk (1989b), banks operate in distinct, informationally segmented regions or markets. Bankers within a particular region have a comparative advantage in supplying loans there because they have better in formation about credit conditions or would-be bor rowers. However, the inability of banks to diversify localized portfolios perfectly can cause some regions to be capital constrained.12 The authors demonstrate that in the absence of asset-backed lending, a region with a relatively large set of profitable— albeit risky— investment opportunities and limited bank capital can be con strained. That is, the region will be unable to at tract sufficient deposits to fund all of its profitable investment opportunities. A constrained bank must channel resources instead into safer but less profitable investments. Binding capital constraints cause interregional differences in returns on projects. These, in turn, create the incentive for banks in constrained markets to originate and sell unfunded profitable investments to banks in unconstrained regions. Unconstrained banks, though adequately capi talized, would not lend to constrained banks via deposit liabilities because these liabilities are claims on the constrained banks’ entire portfolios, which nonlocal firms have no comparative advan tage in monitoring. Alternatively, unconstrained bankers will purchase individual projects from these banks. They recognize that banks are con strained because of excess profitable investment opportunities in their region. Thus, binding capital constraints give rise to asset-backed lending by al lowing a bank to separate the funding of certain projects from the performance of its portfolio. ■ 10 In this discussion, depositors should be understood as either uninsured depositors or banking regulators. ■ 11 The Qualified Thrift Lender Test refers to the regulation that re quires thrifts to hold a certain fraction of their portfolio In the form of http://fraser.stlouisfed.org/ home mortgages. Federal Reserve Bank of St. Louis ■ 12 Capital constraints arise because of short-term variations in lending opportunities that do not create the incentive for a structural re allocation of bank equity capital. 33 Asset-Backed Lending as a Means of Delegating Nonlocal Monitoring Carlstrom and Samolyk’s model shows how cap ital constraints in informationally segmented banking markets can cause banks to sell loans, facilitating a more efficient allocation of resources. These capital constraints are one example in which capital markets may not be as efficient as suggested by textbooks. Loan sales may arise to help correct the associated regional imbalances. Another potential problem with intermedia tion is that information costs may cause credit to be rationed for some borrowers. Credit rationing exists when someone is unable to obtain credit even though he or she is (ex ante) identical to a borrower who does obtain financing. W hen in formation is costless, economic theory predicts that credit rationing will not arise because loan rates will increase until the quantity of loans sup plied equals the quantity of loans demanded. Williamson (1986) demonstrates that it may be efficient for intermediaries that face monitor ing costs to ration credit. As in Diamond, he characterizes banks as issuing claims to a large number of lenders and lending to a large num ber of borrowers. Because of ex post project monitoring costs, banks issue debt contracts to many ex ante identical borrowers, monitor projects only in the event of default, and pay a noncontingent return to depositors. Unlike Diamond, who assumes that banks can fund any number of investments at a given cost of funds, Williamson analyzes an economy in which banks face an increasing marginal cost of funds: They must charge higher loan rates to offer returns that will attract the funds of inves tors with better alternatives. Higher loan rates, however, lead to greater monitoring costs be cause higher interest charges raise the probabil ity that borrowers will default on their loans. Although lenders get all of a project’s proceeds in the event of default, the increase in expected monitoring costs may actually decrease the ex pected return of a loan. In this setting, interme diaries may be unwilling to charge higher loan rates in order to fund more projects and instead choose to ration credit. In a related paper, Boyd and Smith (1989) ex tend this analysis to show another way in which asset-backed lending may improve the perform ance of informationally segmented credit mar kets. As in Carlstrom and Samolyk, differences in interregional returns on projects lead to a type of asset-backed lending. Boyd and Smith consider a variation of the con tracting model described by Williamson (1987).13 In their model, identical borrowers, whose proj ects require costly ex post state verification, con tract individually with lenders to supply funds. To observe the ex post returns on borrowers’ invest ments, lenders must incur monitoring costs, but such costs are assumed to be larger for lenders in other markets. Thus, like Carlstrom and Samolyk’s model, there is a comparative advantage to fund ing projects within one’s own region. Boyd and Smith consider two banking regions that differ in the local ratios of potential lenders to borrowers, creating a scenario in which a Williamson-type credit rationing occurs in only one of the regions. Securitization allows lenders in unrationed markets to fund projects in rationed markets: An intermediary pools and monitors the loans of local borrowers, funding them by issuing claims to other markets. Like D iam ond’s model of intermediation, diversification by this inter mediary allows the ultimate investors, lenders in the unrationed market, to delegate the monitoring to the intermediary in the market where the loans are being originated. Lenders do not find it profitable to fund proj ects in other markets directly because of the large intermarket monitoring costs. However, asset pooling, which completely diversifies away the risk of the pool, eliminates the need for in vestors to incur the large intermarket costs of monitoring the underlying assets. All monitoring takes place locally by the coalition at the lower intramarket monitoring cost. Similar to Carlstrom and Samolyk’s model, loan sales occur in order to equalize expected project returns across mar kets. Credit rationing, however, may still occur in markets where assets are being securitized. How Well Do These Models Describe Off-Balance-Sheet Financing? In Boyd and Smith’s model, securitized loan pools are originated by a coalition of individual borrow ers within one locality, but are funded by lenders in another. Most mortgage securitization takes place via an interregional intermediary, which pools loans from loan originators in many ■ 13 Williamson (1987) shows that credit rationing can occur in a model with debt contracts, where individual borrowers contract with in dividual lenders. This paper is sim ilar to his earlier one (Williamson [1986]), except that there are no financial intermediaries. localities. To the extent that interregional diver sification is conventionally viewed as an impor tant rationale for mortgage securitization, the Boyd-Smith model is limited in the extent to which it can be interpreted as a model of the secondary mortgage market. Instead of being a model of regional mortgage securitization, their analysis is a better descrip tion of most nonmortgage securitization. They do not, however, depict an intermediary that funds a share of its projects off-balance-sheet through a subsidiary. Rather, each individual borrower (not a “bank”) funds his entire project along with other borrowers. Carlstrom and Samolyk depict loan sales and not securitization. However, they model one im portant aspect of nonmortgage asset-backed lending in the sense that banks fund parts of their portfolio on- and off-balance-sheet. These models help explain some of the bene fits of both loan sales and securitization. For two reasons, however, the models are limited in de scribing some dimensions of asset-backed mar kets. First, both the Carlstrom-Samolyk and BoydSmith models rely on regionally segmented bank ing markets to drive their results— an increasingly less likely scenario given the consolidation of the depository industry and the increase in nonbank intermediation. Second, as discussed earlier, secu ritized assets are usually backed by some type of credit enhancements or provide some sort of recourse for the purchasing party that helps make them marketable. Neither of these papers explains why credit enhancements might be an important part of the securitization process. The next two papers discuss the importance of credit enhance ments in making risky bank assets attractive to nonbank investors. Asset-Backed Lending as a Means of Signaling Credit Quality Greenbaum and Thakor (1987) present a model in which the choice of on- versus off-balancesheet funding (which they refer to as the deposit funding mode [DFM] and securitized funding mode [SFM], respectively) is a sorting mecha nism whereby borrowers choose one or the other based on the quality of their project. If a borrower selects the SFM, he must also choose the degree to which the bank will provide re course in the event of default. The degree to which a loan is collateralized signals the quality of the asset to nonbank investors. This elimi nates the need for them to screen the borrower. The model consists of borrowers with projects that differ in quality. Borrowers must choose be tween one of two funding modes. If a project is funded on-balance-sheet, a bank’s entire stock of equity capital effectively collateralizes the project. The bank screens the borrower to ascertain the quality of his project, while depositors screen the bank. This redundancy is necessary because banks are unable to convey the outcome of their screening directly to depositors. Under the DFM, the value of the bank’s collateralization and both of these screening costs are priced into the borrower’s risk-adjusted loan rate. Alternatively, under the SFM, a bank offers to fund the project off-balance-sheet by providing a credit enhancement in the form of bank col lateralization. A borrower pays for the amount collateralized with an up-front fee. Banks screen borrowers and then announce a fee schedule as sociated with a borrower’s choice of collaterali zation. As with insurance, lower-risk projects are charged less for any given level of coverage (collateralization). A borrower’s choice of cov erage is public information and thus can signal a project’s quality, eliminating the need for the purchasing party also to screen the asset. For higher-quality projects, the fee associated with the borrower’s choice of bank collateraliza tion is offset by the reduction in depositors’ screen ing costs. For poorer-quality projects, however, the fee necessary to purchase collateralization is greater, outweighing the benefits from the elim ination of screening by nonbank investors. Thus, poorer-quality borrowers forgo the fee and choose the DFM with full collateralization, al though depositors’ screening costs wrill be priced into their loan rates. An important implication of this framework is that higher-quality assets will tend to be securi tized, while lower-quality assets will tend to be held on-balance-sheet. The intuition is as follows: Higher-quality borrowers receive a lower interest cost than lower-quality borrowers under either funding mode. However, because the choice of collateralization under the SFM produces informa tion about project quality and eliminates the need for asset-backed investors to screen the underly ing assets, higher-quality borrowers can take ad vantage of low credit enhancement rates to obtain a better term of finance. Moreover, their cost of funding is lower despite the increased risk asso ciated with less-than-full bank collateralization from the investors’ perspective. The Greenbaum-Thakor framework repre sents an important step in characterizing the 35 trends in securitization, especially to the extent that asset-backed lending separates the col lateralization and monitoring of the underlying claims from their funding. Similar to the BoydSmith model, this model depicts asset-backed lending as a means of eliminating the need for investors to monitor the performance of the un derlying asset(s). Here the reduction in monitor ing costs occurs, however, because a borrower’s choice to fund via a collateralized loan sale signals project quality and eliminates investors’ need to screen. Alternatively, in Boyd and Smith, the diver sification associated with borrowers’ pooling of claims facilitates delegated monitoring. Asset-Backed Lending as a Means of Securing Credit Quality James (1988) presents a model that characterizes a different rationale for asset-backed lending. Spe cifically, he emphasizes that loan sales with re course are a means of obtaining lower funding costs by separating the cash flows on a particular claim from those to the unsecured claimants fund ing a bank’s balance sheet. He argues that loan sales with recourse are equivalent to a firm issuing secured debt. Because banks are prohibited from issuing secured claims, loan sales with recourse are likely to occur for the same reasons that firms issue secured debt. Firms issue secured debt in part to mitigate an underinvestment problem that may occur with fixed-rate bond contracts. If firms with outstand ing debt are constrained to raise funds by issu ing additional unsecured claims, they may forgo financing certain new profitable projects— in particular, projects that would reduce the over all risk of the firm’s portfolio. This occurs be cause banks cannot reprice existing unsecured claims to reflect accurately changes in the risk of their portfolio due to new asset acquisitions. Thus, if a firm chooses to issue unsecured claims to finance a project that reduces portfolio risk, existing bondholders receive a wealth transfer from stockholders as the risk-adjusted value of their claims increases. James refers to the underinvestment problem that motivates the use of secured debt as the col lateralization hypothesis. The key to this problem is that banks are locked into a fixed cost of funds on their liabilities. With secured debt, the existing bondholders do not have access to the newly ac quired assets should the firm declare bankruptcy. Since regulations restrict banks and thrifts from is suing secured debt, loan sales with recourse— by separating the funding of new projects from that of a firm’s existing investments— can mitigate a potential underinvestment problem. Banks cannot issue secured debt, so the ex tent to which they fund their portfolios by issu ing term liabilities such as certificates of deposit (CDs) may motivate them to finance certain as sets off-balance-sheet with some form of recourse. Still, James’ model may be limited as an explanation for asset-backed lending by banks and thrifts, because the bulk of their liabilities are short-term deposits. Such liabilities have a return that can be readjusted to reflect the risk of a bank’s portfolio after new assets are acquired. Thus, any wealth transfers from bank equityholders to depositors (in an unregulated environment) could be mitigated by readjusting short-term deposit rates. Regulatory Factors and Asset-Backed Lending In reality, the fact that banks are insured, and that the FDIC (not insured depositors) must consider the risk of a bank’s portfolio, complicates this analysis. As the residual claimant of a bank’s as sets, the FDIC, not insured depositors, bears the credit risk of these assets. If capital requirements and deposit insurance premiums were correctly priced (and effectively repriced) to reflect a bank’s risk, the incentives for banks to engage in assetbacked lending would be reduced. To the extent, however, that the FDIC does not price the provi sion of insurance to reflect a bank’s risk accurately, James’ model motivates asset-backed lending. The interpretation here is that safer assets will be funded off-balance-sheet to maximize the value of FDIC insurance to bank equityholders. The models in both James and Greenbaum and Thakor explain why firms would provide credit enhancements for their off-balance-sheet funding. In reality, these enhancements are gen erally issued by a third party— to some degree because of regulations. This is especially true for bank loan sales, as loans sold with recourse are viewed as on-balance-sheet assets in the as sessments of capital requirements. In spite of these limitations, however, these frameworks are useful in characterizing a widely accepted rationale for the proliferation of nonmortgage securitization: to separate the securitized assets from the general portfolios of financial intermediaries. 36 The proliferation of asset-backed lending is merely one way that the financial scene is chang ing. As evidenced by nonbank activities in this market, securitization is both the result of techno logical innovations in information production and an artifact of banking regulations. In this paper, we have focused primarily on models that formal ize market-based reasons for asset-backed lend ing. However, the existence of government regulations, in tandem with the provision of the federal safety net, is widely viewed as a significant factor impacting both the volume of securitization and the types of loans securitized. IV. Regulatory Incentives for Securitization Regulatory models of asset-backed lending gen erally focus on how regulations impact a bank’s choice of funding. For example, Benveniste and Berger (1987) argue that credit enhancements for asset-backed securities allow banks to maxi mize the value of deposit insurance by issuing claims that are senior to those of the FDIC. Al though their argument is similar to that posited by James, he argues that this adverse tendency is offset by the likelihood that loan sales backed by SLCs mitigate the underinvestment problem. The incentive to shift risk to the FDIC is also limited by the marketplace. The creditworthiness of both the loans being securitized and the issuer of credit enhancements affects the rating of a pool. Thus, banks that issue SLCs are generally lowerrisk institutions. Other regulatory incentives for banks to en gage in asset-backed lending are the regulatory taxes associated with on-balance-sheet funding. For example, capital requirements— the mini mum legal fraction of an investment that must be held as equity capital— are popularly viewed as the primary regulatory incentive for banks and thrifts to sell assets. These requirements are designed to protect the FDIC and uninsured depositors in the case of bank failure. Regulation-based models, however, empha size that if capital requirements on a particular class of loans are greater than merited by the inherent risk of the claims, banks will have an incentive to either sell or securitize the loan.14 That is, there will be an incentive to move a loan from on-balance-sheet, where it is subject to capital requirements, to off-balance-sheet, where it is not. http://fraser.stlouisfed.org/ ■ 14 See Pennacchi (1988). Federal Reserve Bank of St. Louis This will be the case when the cost of the reg ulated equity buffer exceeds the cost of market ing the claims. Two other regulatory taxes that have been cited as potential inducements for asset-backed lending are fractional reserve requirements and flat-rate FDIC insurance premiums on deposit liabilities. These assessments are viewed as raising the cost of deposit funding, thus encouraging depository institutions to fund loans off-balance-sheet. Yet, securitization has continued to expand in spite of decreases in the reserve requirements set by the Board of Governors of the Federal Reserve Sys tem. In addition, to the extent that deposit in surance is subsidized, flat-rate deposit insurance premiums are unlikely to be a major factor in the growth of securitization. For example, if the premiums charged to insure the deposits funding relatively risky loans allow an institution to obtain funds more cheaply than from other sources, then even though there are other costs associated with deposit funding, this may be a relatively cheap source of finance. Because deposit insurance premiums are currently not risk based, they may still have the undesirable effect of causing banks to securitize their safest and most liquid loans. V. Conclusion Although market-based reasons are an impor tant factor driving off-balance-sheet lending, this type of lending may still impact the risk of lending that is funded on banks’ balance sheets. For example, Greenbaum and Thakor’s model predicts that the safest assets will be securitized while the risky assets will be held on-balancesheet. Regulations provide similar incentives for securitizing the safest assets. Because these fac tors can clearly impact the exposure of the FDIC, policymakers are understandably con cerned about the rapid growth of this practice. In its role as an insurer, the government aims to maintain the solvency of the insurance fund by regulating deposit insurance premiums and capital requirements. But it is precisely these as sessments that can affect the risks undertaken by depository institutions, as regulatory costs create an incentive for banks to shrink their balance sheets by securitizing loans. However, the trend toward asset-backed lending should not be viewed as either a boon for nonbank competitors or the bane of the FDIC. Depository institutions can earn fee income for participating in various dimensions of the secu ritization process. Moreover, with prudent regulatory supervision of banks’ off-balance- sheet activities, asset-backed lending can miti gate the rising costs of the federal safety net as it reduces the share of credit funded on the books of depository institutions. Thus, securitization is better viewed as an important innovation in the financial sector— one that allows new suppliers of credit to enter the market and existing ones to inter mediate credit more efficiently. References Benveniste, Lawrence M., and Allen N. Berger. “Securitization with Recourse: An Instrument that Offers Uninsured Bank Depositors Se quential Claims,”Jo u rn a l o f B anking a n d Finance, vol. 11, no. 3 (September 1987), pp. 403-24. Bemanke, Ben, and Mark Gertler. “Banking and Macroeconomic Equilibrium,” in William A. Barnett and Kenneth Singleton, eds., New Ap proaches to M onetary Economics. New York: Cambridge University Press, 1987, pp. 89-111. Bhattacharya, Sudipto, and Anjan V. Thakor. “Con temporary Banking Theory,” Indiana Univer sity Discussion Paper 504, November 1991. Boyd, John H., and Bruce D. Smith. “Securitiza tion and the Efficient Allocation of Invest ment Capital,” Federal Reserve Bank of Minneapolis, Working Paper 408, May 1989Carlstrom, Charles T., and Katherine A. Samolyk. “Loan Sales as a Response to Market-Based Capital Constraints,” Federal Reserve Bank of Cleveland, Working Paper, 1993 (forthcoming). Diamond, Douglas W. “Financial Intermediation and Delegated Monitoring,” Review o f Eco nom ic Studies, vol. 51, no. 3 (July 1984), pp. 393-414. Gertler, Mark. “Financial Structure and Aggre gate Economic Activity: An Overview,” Jo u r n a l o f Money, Credit, a n d B anking, vol. 20, no. 3 (August 1988), pp. 559-88. Gorton, Gary B., and Joseph G. Haubrich. “The Loan Sales Market,” in George G. Kaufman, ed., Research in F in a n c ia l Services: Private a n d P ublic Policy, vol. 2 (1990), pp. 85-135. Greenbaum, Stuart I., and Anjan V. Thakor. “Bank Funding Modes: Securitization versus Deposits,”Jo u rn a l o f B anking a n d Finance, vol. 11, no. 3 (September 1987), pp. 379-401. James, Christopher. “The Use of Loan Sales and Standby Letters of Credit by Commercial Banks,”Jo u rn a l o f M onetary Economics, vol. 22, no. 3 (November 1988), pp. 395-422. McCulloch, J. Huston. “Misintermediation and Macroeconomic Fluctuations,”Jo u rn a l o f M onetary Economics, vol. 8, no. 1 (July 1981), pp. 103-15. Modigliani, Franco, and Merton H. Miller. “The Cost of Capital, Corporation Finance, and the Theory of Investment,” A m erican Econom ic Review, vol. 48, no. 3 (June 1958), pp. 261-97. Mookherjee, Dilip, and Ivan Png. “Optimal Auditing, Insurance, and Redistribution,” Q ua rterlyJou m a l o f Econ om ics, vol. 104, no. 2 (May 1989), pp. 399-415. Pavel, Christine. “Securitization,” Federal Reserve Bank of Chicago, Economic Perspectives, vol. 10, no. 4 (July/August 1986), pp. 16-31. Pennacchi, George G. “Loan Sales and the Cost of Bank Capital,”Jo u rn a l o f Finance, vol. 43, no. 2 (June 1988), pp. 375-96. Ramakrishnan, Ram T.S., and Anjan V. Thakor. “In formation Reliability and a Theory of Financial Intermediation,” Review o f Economic Studies, vol. 51, no. 3 (July 1984), pp. 415-32. Samolyk, Katherine A. “Portfolio Risks and Bank Asset Choice,” Federal Reserve Bank of Cleve land, Working Paper 8913, October 1989a. ______ . “The Role of Banks in Influencing Regional Flows of Funds,” Federal Reserve Bank of Cleveland, Working Paper 8914, November 1989b. Townsend, Robert M. “Optimal Contracts and Competitive Markets with Costly State Verifi cation,”Jo u rn a l o f Econom ic Theory, vol. 21, no. 2 (October 1979), pp. 265-93. Williamson, Stephen D. “Costly Monitoring, Financial Intermediation, and Equilibrium Credit Rationing,”Jo u rn a l o f M onetary Economics, vol. 18, no. 2 (September 1986), pp. 159-79. ______ . “Costly Monitoring, Loan Contracts, and Equilibrium Credit Rationing,” Q uarterly Jo u rn a l o f Economics, vol. 102, no. 1 (February 1987), pp. 135-45. First Quarter Working Papers Current Working Papers of the Cleveland Federal Reserve Bank are listed in each quarterly issue of the Economic Review. Copies of specific papers may be re quested by completing and mail ing the attached form below. ■ 9217 Commitment as Irreversible Investment by Joseph G. Haubrich and Joseph A. Ritter Single copies of individual papers will be sent free of charge to those who request them. A mailing list service for personal subscribers, however, is not available. Institutional subscribers, such as libraries and other organiza tions, will be placed on a mail ing list upon request and will automatically receive Working Papers as they are published. ■ 9219 ■ 9302 Cross-Lender Variation in HRM Policy and Home Mortgage Lending Increasing Inequality by Robert B. Avery, in a Salary Survey Patricia E. Beeson, and MarkS. Sniderman by Erica L. Groshen ■ 9218 The Determinants of ■ 9301 Airport Hub Locations, Sharing with a RiskService, and Competition Neutral Agent by Neil Bania, Paul W. Bauer, and Thomas J. Zlatoper by Joseph G. Haubrich Please complete and detach the form below and mail to: Research Department Federal Reserve Bank of Cleveland P.O. Box 6387 Cleveland, Ohio 44101 Check item(s) requested Please send the following Working Paper(s): □ 9217 □ 9219 □ 9218 □ 9301 □ 9302 Send to: Please print Name Address City State Zip Economic Review H 1992 Quarter 1 ■ Recent Behavior of Velocity: Alternative Measures of Money by John B. Carlson and Susan M. Byrne H 1992 Quarter 3 Comparing Central Banks’ Rulebooks by EJ. Stevens Commodity Prices and P-Star by Jeffrey J. Hallman and Edward J. Bryden Forbearance, Subordinated Debt, and the Cost of Capital for Insured Depository Institutions by William P. Osterberg and James B. Thomson The Causes and Consequences of Structural Changes in U.S. Labor Markets: A Review by Randall W. Eberts and Erica L. Groshen An Introduction to the International Implications of U.S. Fiscal Policy by Owen F. Humpage 1992 Quarter 2 I 1992 Quarter 4 Intervention and the Bid-Ask Spread in G-3 Foreign Exchange Rates by William P. Osterberg White- and Blue-Collar Jobs in the Recent Recession and Recovery: W ho’s Singing the Blues? by Erica L. Groshen and Donald R. Williams An Ebbing Tide Lowers All Boats: Monetary Policy, Inflation, and Social Justice by David Altig Assessing the Impact of Income Tax, Social Security Tax, and Health Care Spending Cuts on U.S. Saving Rates by Alan J. Auerbach, Jagadeesh Gokhale, and Laurence J. Kotlikoff Sluggish Deposit Rates: Endogenous Institutions and Aggregate Fluctuations by Joseph G. Haubrich History of and Rationales for the Reconstruction Finance Corporation by Walker F. Todd