7
Index Design and Index Purpose

As we have seen, a host of difficult questions must be answered in design ing a price index. How some of these should be resolved may depend on the particular uses to which an index is put. There is a tension between the goals of providing indexes tailored to specific purposes of public policy or for particular kinds of economic information and avoiding public confusion that might result from too many indexes.

Subject to considerations of costs, feasibility, and reliability, the publication of several overall indexes may well be warranted. For instance, we concur with the decision by the Bureau of Labor Statistics (BLS) to produce a lagged superlative index in addition to the “flagship” Consumer Price Index (CPI); we also suggest publication of an advance forecast of the superlative for compensation purposes (see below and Chapter 2). Yet there are inherent limitations in trying to produce an exact match between the design of official price indexes and the particular purposes for which they are used. Moreover, the media will inevitably emphasize one or, at most, a few indexes.1 But whether there are few or many indexes available, it is important that public policy makers and private users understand the relationships between the ways in which price indexes are designed and how they serve—or fail to serve—the various purposes for which they are employed. In this chapter we attempt to clarify some of those relationships.

1  

Research and experimental indexes, explicitly labeled as such, have been and can continue to be useful, need not be limited in number, and are less likely to create problems of public perception.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes 7 Index Design and Index Purpose As we have seen, a host of difficult questions must be answered in design ing a price index. How some of these should be resolved may depend on the particular uses to which an index is put. There is a tension between the goals of providing indexes tailored to specific purposes of public policy or for particular kinds of economic information and avoiding public confusion that might result from too many indexes. Subject to considerations of costs, feasibility, and reliability, the publication of several overall indexes may well be warranted. For instance, we concur with the decision by the Bureau of Labor Statistics (BLS) to produce a lagged superlative index in addition to the “flagship” Consumer Price Index (CPI); we also suggest publication of an advance forecast of the superlative for compensation purposes (see below and Chapter 2). Yet there are inherent limitations in trying to produce an exact match between the design of official price indexes and the particular purposes for which they are used. Moreover, the media will inevitably emphasize one or, at most, a few indexes.1 But whether there are few or many indexes available, it is important that public policy makers and private users understand the relationships between the ways in which price indexes are designed and how they serve—or fail to serve—the various purposes for which they are employed. In this chapter we attempt to clarify some of those relationships. 1   Research and experimental indexes, explicitly labeled as such, have been and can continue to be useful, need not be limited in number, and are less likely to create problems of public perception.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes There are many different uses for aggregate indexes of prices and the cost of living, most prominently: as a compensation measure to calculate how much is needed to reimburse recipients of social security and other public transfer payments against changes in the cost of living and for formal or informal use in wage setting; for inflation indexation in private contracts; as a measure of inflation for inflation-indexed Treasury bonds; as a measure with which to index the income tax system to keep it inflation neutral; as an output deflator for separating changes in gross domestic product (GDP) and its components into changes in prices and changes in real output; and as an inflation yardstick for the Federal Reserve and other macroeconomic policy makers. This chapter examines the application of indexes in each of these contexts. INDEXING PUBLIC TRANSFER PAYMENTS The CPI is widely used within government and among private parties as a means of maintaining the purchasing power of a flow of transfer payments in the face of changes in prices, sometimes specifically identified as changes in the cost of living. In a similar vein, the CPI is used to adjust eligibility limits for certain kinds of payments, usually to the poor, that were initially set in nominal dollar terms. One overarching conceptual issue that arises when cost-of-living adjustments are provided in public transfer payments is whether the adjustments should compensate recipients only for changes in the overall cost of living for the nation as a whole or should take account of any significant differences among particular groups and individuals in society. Even if it were possible to calculate a separate index for every individual, public policy would surely not seek to provide adjustments tailored to each. Indeed, taken literally, this approach would provide incentives for individuals not to substitute away from goods whose prices had risen the most (the government transfer payment would provide the means for an individual to maintain his or her consumption of expensive wines, for example, even if their prices skyrocketed).2 However, if the goal of public policy is to ensure recipients of various public transfer programs—e.g., the poor and the elderly— against changes in the cost of living, and if cost-of-living indexes for the affected group differ systematically or frequently from the aggregate CPI, then Congress 2   Of course, this does not preclude the use of democratic cost-of-living indexes, which are averages of the indexes of individual consumers or households.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes might wish to consider using a separate index for the affected group as the measure of compensation. There is no abstract criterion that can be used to determine the extent to which compensation should be tailored to particular subcategories of individuals within each broad group of targeted transfer recipients. In general, the panel believes that, if a particular category of individuals is not itself the target of a transfer program, special indexes based on detailed distinctions within the targeted groups are not a suitable basis for making cost-of-living adjustments. In the sections below, we discuss the potential use of special indexes for making cost-of-living adjustments for the elderly and the poor. Adjusting Social Security Benefits The most prominent public policy use of the CPI is for indexing benefits paid to social security retirees. Prior to 1972 Congress had periodically legislated increases in social security benefits, usually by more than enough to cover changes in inflation (as measured by the CPI) since the last increase. In 1969 the Nixon Administration announced its support for automatic indexing of benefits, declaring: “The way to prevent future unfairness is to attach the benefit schedule to the cost of living” (cited in Berkowitz, 1986:48). In 1972 this recommendation became law (along with a 20 percent one-time increase in benefits). Congress explicitly provided for annual cost-of-living-increases for people receiving benefit payments, based on changes in the CPI-W.3 The CPI is also used for the same objective in a number of other federal programs that provide transfer payments: for the military and civil service retirement systems, the railroad retirement system, veterans’ pensions, and Supplemental Security Income. When it established an indexing procedure, Congress stipulated that benefits be adjusted to offset changes in the cost of living (rather than simply to maintain the purchasing power over a fixed basket of goods) and specified that this should be done through the use of the CPI. At the time, members of Congress appeared to have accepted the widely held presumption that the CPI measures the cost of living. There is no reason to believe they explicitly considered any distinction between a fixed-weight and a cost-of-living index. The panel was not charged with recommending to Congress what specific objectives it ought to pursue in indexing social security and other benefits. We were charged to make clear the implications for public policy that flow from choosing one or another scheme of indexing and to spell out the consequences for public policy from alternative choices of index design. Moreover, even assuming 3   Various modifications have been made in the timing and details of the adjustment, but the cost-of-living terminology remains. No provision is made for decreases in benefits when the CPI declines (P.L. 92-336).

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes that public policy seeks to tie benefits to changes in the cost of living, a number of choices still remain about whether the present CPI (together with changes that are planned for the near future) or some alternative version is the appropriate one—for example, a separate index for the elderly. Using a Lagged Superlative Index for Escalation We conclude that, for adjusting benefits to keep pace with the cost of living, the superlative index, which the BLS will begin to publish in 2002, will be the appropriate one to use. But that index will be available only after a 2-year delay. One way to deal with this lag would be to pay an initial cost-of-living adjustment based on the change in the fixed-weight CPI and then incorporate a correction 2 years later based on the lagged superlative. Thus, the cost-of-living adjustment (COLA) for 2004 would increase the benefit payment by the change in the regular fixed-weight CPI for the past year minus a correction for the difference between the regular CPI for 2002 and the superlative index for that year, which would just have been published. If recent history is a guide, the superlative index will show an average increase of about 0.1 to 0.2 percent a year less than the real-time CPI, with a range of 0.0 to 0.5 percentage points (Aizcorbe and Jackman, 1993; Shapiro and Wilcox, 1997). A small initial “claw back” would be required; thereafter, the COLA would tend to be very close to the change shown by the current year’s real-time CPI minus an adjustment that typically fluctuated within a narrow range. There is an alternative that may offer some advantages and which the recommendation below would facilitate: Recommendation 7-1: The BLS should publish, contemporaneous with the real-time CPI, an advance estimate of the superlative index, utilizing either a constant-elasticity-of-substitution method or some other technique. There are several possible ways to construct that estimate. Individual researchers have constructed estimates of the superlative using only reference-period weights and a constant-elasticity-of-substitution formula, whose major parameter was estimated from a comparison of the real-time CPI with the superlative in prior years. Such estimates, when tested over short periods of time, have very closely tracked the superlative that later became available, although particular patterns of substantial changes in relative prices could produce larger divergences. Alternatively, the BLS could utilize other techniques for making an advance estimate, perhaps taking advantage of the latest information on relative price changes in the real-time CPI. For purposes of escalation, the panel arrived at the following: Conclusion 7-1: It would be feasible and appropriate to calculate cost-of-living allowances provided for social security and other pro-

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes grams from an advance estimate of the BLS’s soon-to-be-published superlative index. Any divergence between that estimate and the superlative that appears 2 years later could be incorporated as a correction to the cost-of-living allowance provided for that year. For example, the 2004 COLA would be based on the advance estimate of the superlative for the prior year plus or minus the difference between that estimate of the superlative for 2002 and the actual value of the superlative for that year. Independent of how the advance estimate of the superlative is arrived at, the panel supports its use. This alternative offers both an advantage and a disadvantage when compared with the 2-year delay approach. There is a high probability that the later corrections to the initial COLA adjustment would be a good bit smaller if the advance estimate is used. But it would add an additional and hard-to-explain complexity to the index used for the initial adjustment. Conclusion 7-2: On balance, the advantage of having much smaller corrections outweighs the disadvantage of the additional complexity. Compensating Beneficiaries Who Have Other Income Many social security retirees have other income, in some cases substantially exceeding their social security benefits. The broad objective of Congress in providing a cost-of-living adjustment was and is to protect the social security income of beneficiaries, and not their other income, against the consequences of price changes. In Chapter 2 we point out that it is not obvious how to design an index that holds constant the standard of living of social security recipients who have other income. Perhaps the simplest way of dealing with the problem is to define the index as one that provides the compensation needed to maintain living standards for those whose only income is their social security benefit. The Role of Taxes In its most usual formulation, a cost-of-living index provides a measure of the percentage change in expenditures a consumer would have to make to maintain a specified standard of living in the face of changes in the prices paid for goods and services. This is an expenditure COLI. An alternative approach is to measure the percentage change in the income a consumer would need to maintain that same standard of living as prices and income and payroll tax rates change. (Indirect taxes, such as sales and value-added taxes, are already included in the prices of private goods and services.) Such an index has sometimes been labeled a tax and price index (TPI; see Gillingham and Greenlees, 1987, 1990). Simply providing an additional amount of income sufficient to pay the higher prices, as

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes would be the case under an expenditure COLI, would not be enough if that income were subject to higher tax rates. A TPI would include the effect of higher income and payroll tax rates and, if used for compensation purposes, would therefore provide enough compensation to cover both the higher taxes and the higher prices. (The domain of a TPI is restricted to private goods and services; it does not impute to the consumer’s living standard any value from public goods financed by taxes.) What are the implications of tying social security payments to a tax-and-price index rather than the current expenditure-based index, assuming that a single overall CPI would continue to be used for indexing purposes? First, social security retirees pay no payroll taxes; and in the federal and many state income tax systems social security benefits are more lightly taxed than other forms of income, at least for low- and lower-middle-income taxpayers. It is likely that legislative changes in tax rates would retain the same sort of preferences for the elderly. A TPI would, therefore, be likely to overadjust social security benefits when tax rates generally are raised and have the opposite results when taxes are cut. If Congress wishes to change the after-tax benefits for social security recipients, a much fairer and more effective way to do so is through explicit changes in the tax code or benefits formula. Second, the use of an index that reflects changes in tax rates would be inequitable among social security recipients themselves. Legislated changes in income tax rates often vary among people with different incomes and in different economic circumstances, while an overall income COLI would reflect only an average of the rate changes: some beneficiaries would be overcompensated and others undercompensated. Conclusion 7-3: For purposes of indexing social security and other benefits, shifting from the current expenditure-based CPI to a tax-and-price index that reflected changes in income and payroll tax rates would pose some difficult measurement problems and create unintended distributional inequities. A Separate Index for the Elderly? To the extent that prices for goods and services paid by the elderly rise at a different rate than those paid by the population generally, Congress might consider tying retirement benefits to a special index for the elderly. At the request of Congress, BLS developed a special experimental index for the elderly (CPI-E) in which the prices of the 200-plus categories of goods and services in the regularly published CPI-U were reweighted to reflect the consumption patterns of the elderly. From 1984 to 1995 the experimental index rose by an average of 0.4 percent per year faster than the CPI-W, which is used to index social security benefits, and by 0.3 percent per year faster than the CPI-U. Several outside

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes studies covering earlier periods found little difference in the cost-of-living changes faced by the elderly and the general population (see Boskin and Hurd, 1985; Jorgenson and Slesnick, 1983). These results however do not necessarily mean that the elderly faced only slightly more rapid increases in living costs than the population as a whole during those years. Then, or at other times, the “true” differences could be larger and might be either positive or negative. In the BLS studies, for example, the heavier weight attached to out-of-pocket medical care expenses among the elderly accounts for the majority of the difference between the CPI-E and the CPI-U or CPI-W. It is widely believed that, quite apart from controversial issues of quality adjustment, the measurement of medical care prices in the CPI during the periods studied overstated their rise (Newhouse, 2001). BLS has recently improved the definition of the medical care services it prices, and a recent study has shown that changes of this nature significantly reduce the rise of medical care prices (Newhouse, 2001). In addition, the CPI-E did not capture any possible influence coming from changes in the magnitude and scope of senior citizen discounts. The major problem, however, lies in the fact that the elderly may, on average, buy different varieties of goods within many CPI strata, face different prices, and shop at different outlets than younger consumers. Unfortunately, as we explain in Chapter 8, there is no database that allows adequate exploration of this facet of behavior and its consequences, for the elderly as well as for the poor or other demographic groups. The special CPI-E index and the other studies cited all used the same prices for the same goods at the same outlets as were priced for the CPI-U and simply reweighted them to reflect the budget allocations of the elderly among large expenditure categories. That is one of the reasons we recommend a long-range BLS research program that would explore the use of innovative techniques (e.g., scanner technology integrated with consumer surveys) to examine this issue. Conclusion 7-4: In the absence of an index that can capture the differences in the prices or qualities of goods purchased by the elderly, we see no rationale for switching to an index along the lines of the CPI-E for purposes of indexation. However, BLS should periodically update the CPI-E to make sure that significant differences are not developing between it and the CPI-W and the CPI-U. The CPI-U Versus the CPI-W for Indexing Transfer Payments In 1978 the CPI was revised in a major way, including an expansion of its coverage from “urban wage earners and clerical workers” (one-third of the population) to “all urban consumers” (four-fifths of the population). The new index was christened the CPI-U, which is now widely accepted as the main or flagship index of consumer prices. The original CPI was renamed the CPI-W and contin-

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes ues to be used as the index for determining annual cost-of-living adjustments in social security and other programs. The weights of the two indexes differ, but only modestly. And to the extent the weighting structures do differ, the CPI-U weighting structure is closer to that of the elderly.4 Over the past 20 years, the average difference between the growth of the two indexes has been small—the CPI-U grew 0.13 percent faster, although there was substantially greater divergence over shorter time periods. Over the past 5 years, however, the difference between the growth of the two indexes averaged only 0.04 percent. The differences do not seem to be large enough to warrant switching the index used for indexing social security benefits to the CPI-U. However, if the instrument used for adjusting social security benefits is changed to the superlative index, as we recommend, that index should be based on CPI-U rather than CPI-W weights. Plutocratic Versus Democratic Indexes The current CPI is a plutocratic index. In constructing the national index each individual good is assigned a weight equal to overall consumer expenditures on that item. This procedure assigns to the spending pattern of each individual household an importance in the overall index proportional to its total consumption expenditures; the spending patterns and preferences of the rich count more than those of the poor. A democratic CPI would be one in which each household’s spending pattern received equal weight. Arguably, a plutocratic index may be the appropriate choice for an overall indicator of inflation in consumer goods. And since, in the construction of measures of national output, the individual strata indexes of the CPI are used to deflate most of the components of consumption expenditures, a plutocratic version of those individual indexes is needed. But for purposes of indexing social security benefits and other public transfer payments and for dealing with economic welfare considerations generally, a democratically constructed index seems clearly preferable since it assigns the preferences of each household equal importance. 4   When the goods in the CPI are grouped into nine broad categories, the CPI-U comes closer to the CPI-E in five cases, the CPI-W is closer in one case, and three are about equal. The median income of the population covered by the CPI-W should be closer to that of the median elderly household than to the median for the population covered by the CPI-U. If the individual items and shopping outlets whose prices enter the CPI-W were specifically chosen to reflect the purchases and shopping patterns of the wage-earning population, that index might be superior to the CPI-U for indexing social security benefits. But as we have repeatedly pointed out, the same selection of individual items and outlets goes into all the price indexes—CPI-U as well as CPI-W. The indexes differ only through the variations in their upper-level weights.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes In Chapter 8 we note that simply switching to a democratic set of weights at the upper level of CPI aggregation would probably produce only modest differences in the behavior of the index. That was the result of a research index for the elderly constructed along those lines by the BLS for the years 1989 to 1997, as well as of other studies for various earlier periods (see Technical Note 2 in Chapter 8 for a brief summary of these studies). To the extent that systematic and important differences do exist, they must arise at the very detailed level of price collection. This possibility provides yet another reason for the research project we have suggested for investigating differences among households at that level. Conclusion 7-5: To the extent that the evidence ultimately suggests significant differences between democratic and plutocratic indexes and demonstrates the feasibility of producing demographic indexes, the case for switching to such an index for compensation purposes is a persuasive one. Indexing Social Security Benefits to Wages Rather than Prices Some people have argued that social security benefits should be tied to a wage index (see, for example, Griliches, 1995). To the extent that the CPI is a reasonably good approximation to a cost-of-living index, a monthly retirement benefit adjusted for CPI inflation will “buy,” over the remaining life of the retiree, the same standard of living as it did on the date of retirement. The retiree receives protection against inflation and is insured against the economic vicissitudes that can erode the real wages and living standards of the working population—such as a large surge in energy prices or a major depreciation of the real exchange rate. With wage indexing, a worker or retiree gives up those protections but gains the advantage of sharing in the fruits of future national productivity growth and any other economic developments that improve real wages. The choice between wage and price indexing thus involves issues of broad public policy with respect to the distribution of income between social security beneficiaries and the rest of the population. The panel was not charged with providing advice or recommendations on this distributional issue, but we were explicitly asked to assess “the appropriate uses of [cost-of-living] indexes for indexing federal programs and other purposes.” While a wage index is not a COLI, the panel believes that a comparison of wage and price indexation for social security recipients, in the light of various criteria of interest to policy makers, can significantly help illuminate some of the public policy choices implicit in the selection of an indexing instrument. Alternative Wage Indexes There are many possible wage indexes that might be employed for indexing purposes. They do not all move parallel with each other, even over long periods

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes of time. Which one to adopt is far from obvious, and the choice hinges on important conceptual issues. The specific consequences of adopting wage indexing, both for retirees and the federal budget, would depend on the nature of the wage measure that was chosen and how that choice interacted with future economic developments. The initial choice of a wage index could itself be quite controversial, and the debate might recur in light of subsequent developments. However, subject to one qualification we discuss later, the wage measure could be estimated and social security benefits indexed without having to deal with vexing and often contentious issues involving the effects of substitution among goods and outlets, pervasive quality changes, and the introduction of new goods with which we wrestle throughout this report. We first describe the major conceptual alternatives for a wage index and then briefly summarize the pros and cons and the implications of using each of them. The social security system already employs an aggregate national wage index as part of the process of calculating each retiree’s initial retirement benefit. That benefit is based on the annual earnings of the retiree in each past working year, indexed up to the date of retirement using an economy-wide “average annual wage.”5 The average is the mean of the annual money wages of employees as reported by all employers on their W-2 forms. It includes wages in excess of the social security taxable maximum but excludes nonmonetary fringe benefits.6 This average is also used to index the bend points in the benefits formula and the maximum earnings subject to tax. In implementing wage indexation, Congress might well decide that the same measure used to index preretirement wages for calculating the initial benefit ought simply to be extended through the retirement years, as the index for maintaining the relationship between postretirement benefits and the real wages of the working population. However, this choice is not the only one for which a reasonable argument might be advanced. If the broad policy objective is to have the postretirement benefit rise or fall in line with the economic fortunes of the working population, a number of issues arise about the nature of the wage index that would do that appropriately. Perhaps the most important type of choice among alternative wage indexes involves an aggregation question that bears some resemblance to the issue of plutocratic versus democratic prices indexes: Should the index reflect the change in the mean wage or some other point in the wage distribution, the most likely candidate being the median wage? From the late 1970s to the mid-1990s, 5   Thus a worker’s earnings relative to other workers throughout her working life will determine the relative wage that enters into calculating the initial retirement benefit, but the absolute size of the benefit will also depend upon how fast average social security earnings have grown over time. (The formula that is applied to the resultant earnings measure to calculate benefits is itself highly redistributive.) 6   Some workers are not covered by the social security system, but an estimate of their wages is included in calculating the average social security wage.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes TABLE 7-1 Average Annual Real “Wage” Increases, 1980-2000 (percent)   Means   Period ECI Hourly Compensation ECI Hourly Wage Social Security Annual Wagea Median CPS Wage 1960-1980 n.a. n.a. 1.0 1.5 1960-1973 n.a. n.a. 2.0 2.5 1973-1980 n.a. n.a. -0.7 -0.3 1980-2000 0.9 0.6 1.2 0.3 NOTE: Nominal indexes deflated by the CPI-U (research series). aData through 1999. SOURCE: ECI data from BLS web site. Social security annual wages from U.S. Social Security Administration (2000). For CPS hourly wage, see text footnote 8. wage inequality in the United States increased substantially. As a consequence, median hourly wages rose much more slowly than their mean. A second question concerns the scope of a wage index: Should it measure changes in the overall compensation of workers, including not only wages but fringe benefits, or should it cover wages only? Over most of the past 50 years, fringe benefits—chiefly, employer-paid pensions and health insurance costs— rose more rapidly than wages, so that the growth in real compensation per hour exceeded the growth in wages by about 0.25 percent a year. Since 1980, the BLS has published a quarterly Employment Cost Index (ECI) for total hourly compensation (including fringe benefits) and one for wages alone.7 Between 1994 and 1999 the excess growth of fringe benefits relative to wages was reversed as growth in the cost of medical care and the generosity of employer health care plans were reduced. But in 2000 fringe benefits once again began to grow more rapidly. The future relationship between the two components of employee compensation is likely to depend importantly on what happens to health care costs. Table 7-1 shows for various periods the average annual growth in the real wage (or compensation) for each of the four concepts described above: the two published ECIs for hourly compensation and for wages (which are not available for the early years); the average annual social security wage; and the median 7   Here the term “wages” includes wages and salaries. The ECI defines wages as straight-time wages per hour worked. Paid leave and premium pay are included in fringe benefits. The ECI collects data from employers on a probability sample of compensation components for specific occupations. The data are combined into an overall index with fixed employment weights by occupation and industry.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes The best-known example of such adjustment is the use of the CPI-U to index the value of certain U.S. Treasury securities. In 1997 the U.S. Treasury, responding to the demand for an asset that combined the safety of Treasury securities with inflation protection, began to issue marketable 5- and 10-year notes and 30-year bonds called Treasury Inflation-Protected Securities (TIPS). By the end of 2000, $121 billion of such securities had been issued, which was 4 percent of the Treasury’s total marketable securities. Such securities are held by investors both directly and through such interme-diaries as mutual funds and pension plans.18 The TIPS are sold, and their interest rates determined, at a single-price auction of the kind used for Treasury’s other marketable securities. The principal is indexed to the seasonally unadjusted CPI-U (lagged approximately 3 months), and at maturity the securities are redeemed at the greater of their inflation-adjusted principal or their original par value. The interest rate is fixed for the term of the security, and interest is paid semiannually on the inflation-adjusted principal. Both interest payments and any adjustment to the principal are subject to federal income tax in the year they occur, but they are exempt from state and local taxes. Since individuals, organizations, and institutions save for a wide variety of purposes, the use of a broad index of general inflation for indexation seems appropriate. Arguably, the GDP price index, which covers all final goods and services, would provide the broadest protection against price changes. However, since the projected need to maintain household consumption expenditures in retirement is a major motivation for individual savings, the CPI is a quite logical alternative. In addition—and perhaps more important—the CPI’s familiarity to the public, monthly availability, and freedom from revision provide a strong rationale for its use. Obviously, a broad index like the CPI cannot fully meet the indexing needs of savers with specific and narrower savings goals, such as the college education of children or the purchase of a retirement home. In such cases, investment vehicles that more nearly track the costs of the goods in question, such as real estate investment trusts or state or college prepaid tuition programs, might be more appropriate. The large investors and institutions that play a major role in the market for Treasury securities will be aware of methodological revisions in the CPI and build them into their expectations about the future rate of measured inflation. Given no change in the rate of inflation expected under the unrevised methodology, the revisions should lead to different auction prices and effective interest yields for TIPS relative to those on other securities than would otherwise have prevailed. An investor will receive a different initial coupon yield and an offsetting difference in the inflation adjustment. However, revisions in CPI methodol- 18   Inflation-adjusted U.S. savings bonds (Series I) are also available.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes ogy that change the subsequently reported rate of inflation relative to what would have been reported under the old methodology will alter the future stream of interest payments to those who purchased TIPS prior to the revisions. The expectations about reported inflation, on the basis of which the auction prices were determined, would have been different had knowledge of the revisions then been available. For a 10- or 30-year investment, modest annual revisions occurring early in the life of the security could add up to a substantial sum and would obviously affect the market value of the TIPS in the intervening years. On the assumption that an investor in TIPS is seeking to buy a stream of returns insured against perceived changes in purchasing power or the cost of living (as contrasted with hedging liabilities themselves indexed to the CPI), the potential occurrence of future CPI revisions introduces a risk of variance uniquely related to TIPS. Potential investors must face a “revision risk” whose presence will tend to lower the average auction price and raise the cost of borrowing to the Treasury (Emmons, 2000).19 This premium, however, is likely to be a good bit smaller than the inflation premium associated with nominal yields. Conceptually, after every revision the BLS could continue to publish an unrevised index, which could be used for purposes of indexation for TIPS purchased in the year(s) prior to the revision, thereby eliminating the “revision premium.” But in some cases maintaining two sets of indexes could be very expensive, and if there are frequent revisions the number and cost of special indexes could grow rapidly. Revisions in the CPI are presumably made only after a reasoned determination that they improve the ability of the index to meet its objective, however formulated. Any unavoidable costs associated with creation of a risk premium in TIPS are likely to be much smaller than the benefits to society from an improved index in its many uses. INDEXING THE FEDERAL INCOME TAX SYSTEM In the federal income tax system, marginal tax rates rise with nominal income. A taxpaying unit whose real income is unchanged can be pushed into a higher tax bracket by inflation. In 1981 Congress provided for indexation of the individual income tax with the objective of preventing inflation from raising the burden of federal income taxes. Since 1985 (when this tax provision became fully effective) the bracket widths, the personal exemptions, and the standard deduction have been automatically indexed annually with the CPI to ensure that inflation would not result in a higher effective tax rate for any taxpayer. A 1987 paper (Gillingham and Greenlees) showed analytically and demonstrated empirically 19   Although TIPS constitute only about 5 percent of outstanding marketable Treasury notes and bonds, that percentage is increasing since about one-third to one-half of new Treasury issues are TIPS.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes that, under most circumstances, an expenditure-based index like the CPI would not accurately achieve this objective and would tend to under-index the tax system. From a cost-of-living standpoint, CPI indexation will not maintain a specified ratio of federal income taxes to the consumption spending that would yield a constant standard of living (what Gillingham and Greenlees label “direct consumption costs”).20 There are a number of reasons for this result. Some items of consumption are tax deductible: if their prices rise faster or slower than those of nondeductible consumption, a tax system indexed by the CPI will not be inflation neutral. Some items of expenditure, such as medical care, give rise to tax deductions, and some items of income, such as the imputed rent from owner-occupied housing, are tax exempt. Most importantly, during a period in which federal payroll tax rates increase, the income necessary to maintain any given standard of living in the face of price changes rises by more than the CPI. Hence indexation of bracket limits and other tax parameters by the CPI does not keep federal income taxes at a fixed ratio to direct consumption costs. Even after a much-simplified representation of the tax code and other simplifying assumptions, neutralizing the effect of inflation requires a complex indexation formula, labeled an “exact indexation measure” (Gillingham and Greenlees, 1987). It takes into account the effects on the ratio of taxes to direct consumption costs arising from the factors listed above, the most important of which have been the indirect effects of changes in federal payroll tax rates. The formula also requires the construction of individual indexes for every consumer in a sub-sample of the 1973 Consumer Expenditure Survey in order to mimic the effect of the progressive federal income tax system. Over the 1967-1985 period, the CPI (for the sample of households used in the simulation) rose by 197 percent and the exact index by 241 percent. A simulation of a simplified version of the statutory tax structure of 1973, with prices rising at the rate of growth of the CPI that actually occurred between 1967 and 1985, showed that the average ratio of federal income taxes to direct consumption costs would have risen by 192 percent over the period. CPI indexation would have reduced that to a rise of 13 percent, while exact indexation would have reduced it to zero. (In fact, marginal tax rates were adjusted periodically and other changes were made over the period, so that the actual ratio of taxes to gross consumption costs, without indexation, rose by 40 percent.) The CPI is clearly an imperfect tool for indexing the income tax system. Yet despite the high rates of inflation during the period studied, CPI indexation, while not rendering the income tax system completely neutral, would have eliminated much of the bracket creep. The Gillingham-Greenlees exact index requires a 20   Gillingham and Greenlees (1987) describe the base-weighted CPI as an approximation to a conditional COLI.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes complex formula even for a stripped-down representation of the tax code. Constructing a more precise index would substantially add to that complexity and still require the use of some judgmental simplifications. The result would be an opaque indexing measure that would be very difficult to explain and justify to taxpayers. Therefore, we conclude the following: Conclusion 7-6: Despite the imperfections of the CPI, it should continue to be the basis for tax indexation and the tax law should not be changed to require the construction of an “exact” index. MEASURING OUTPUT CHANGES In the national income accounts of all countries, changes in national output are estimated by using price indexes to “deflate” the changes in the components of current dollar expenditures and then combining them into aggregate indexes of output (quantity) or, correspondingly, constant dollar output measures. In this context an aggregate index of inflation ought to be evaluated in terms of its ability to partition expenditures into two symmetric components—an index of inflation and an index of output change, which when multiplied together produce the observed change in current dollar expenditures.21 In the United States, as in most other industrial countries, the overall CPI or its equivalent is not used as a deflator for aggregate consumption expenditures, but its individual components are the deflators for most of the individual categories of consumption expenditures. Up-to-date estimates of consumer expenditures in current dollars are made quarterly by BEA in the U.S. Department of Commerce to produce the national income and product accounts. The expenditure estimates are based on sales and other data, collected mainly from sellers rather than households. For most categories of consumption goods, the NIPA estimates of expenditures are substantially higher than those derived from the CEX survey.22 While the NIPA estimates undoubtedly pick up a large volume of expenditures that are missed in the CEX interviews and diaries, for many expenditure categories (e.g., automobiles and computers) one must cull from NIPA the sales made to business firms or to individuals for business use, which introduces a potential source of error. Consumer expenditures constitute about two-thirds of GDP. If the NIPA estimates of aggregate consumer spending are seriously over- 21   See Diewert (2000a:sections 2, 3) for a treatment of this topic. 22   There are some differences in scope between the NIPA and CPI universes of consumption goods, dictated by the structure of the national income accounts. For example, the NIPA classifies the value of in-kind transfers to consumers as consumption, rather than government expenditures. And in the NIPA price index for personal consumption expenditures, the weight assigned to medical services includes Medicare- and Medicaid-financed outlays; in the CPI the medical service weights reflect only out-of-pocket consumer expenditures.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes stating their true magnitude, the statistical discrepancy (which measures the excess of the independently collected gross domestic income relative to gross domestic expenditures) would tend to be a substantially larger (positive) number than it typically is. Most of the NIPA consumer expenditure data are grouped into detailed categories that can be associated with one or a combination of several CPI strata price indexes. These expenditure categories are then deflated, principally with the use of the CPI price indexes, to produce quantity indexes of real consumption outlays for each category. The detailed price and quantity indexes are aggregated into annual and quarterly Fisher indexes for total consumption and its major subcategories.23 Fisher quantity and price indexes have the desirable property that an index of current dollar expenditures divided by a Fisher price index produces the corresponding Fisher quantity index. The quarterly data are available shortly after the quarter ends. Inflation and output changes for each year or quarter reflect the weights of the beginning and ending periods in the comparison. The Fisher price and quantity indexes are superlative indexes and take approximate account of the effect of consumer substitution among the individual strata categories of goods in response to changes in relative prices. For this as well as other reasons, the measured inflation rate is a little lower and the output increase a little higher than would be the case with Laspeyres weighting.24 Even though the overall CPI is not itself used as an output deflator for consumption expenditures, its individual components are the critical elements (along with the estimates of current dollar expenditures) in measuring what is happening to the level and structure of national output. Decisions about how to deal with such problems as quality change and new goods thus feed into the aggregate measures through the individual price indexes. The estimation of national output data through the deflation of current dollar expenditures for the individual components of consumption requires either that the individual strata indexes used for deflators be themselves aggregated from individual prices with expenditure weights or that the individual prices be selected with the sampling procedure based on expenditure weights. That is, they must be plutocratic indexes. If the research into alternative data collection tech- 23   The annual Fisher indexes are based on weights in adjacent years and (in a very recent change in methodology) the quarterly indexes on weights in adjacent quarters. The quarterly indexes through the most recent completed year are adjusted so that their annual average corresponds to the annual indexes. 24   BLS recently switched to geometric averaging of individual prices for calculating the individual indexes for strata that account for about 61 percent of the weight in the CPI. Analysis led the agency to conclude that in these strata substitution among goods in response to price changes was large enough to warrant that geometric weights would be superior to arithmetic weights for approximating the effects of substitution behavior on the cost of living.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes niques that we recommend (see Chapter 8) ultimately leads to the production of democratically weighted CPI indexes at the lower (strata) level, it will be necessary to retain a parallel system of expenditure-weighted individual strata indexes for use in the NIPA. Since the detailed components of the CPI are the building blocks of the NIPA as output deflators and inflation measures for consumer goods, which make up two-thirds of GDP, decisions about conceptual and measurement issues in its construction will not only determine what is known about the rate of inflation but will also strongly influence information about the rate at which the U.S. economy grows and the nature of its structural changes over time. The most important issues in this regard are decisions about how to deal with quality change and the introduction of new goods. One would surely, for example, not consider it inflationary (and thus calling for restrictive monetary policies) if households consumed a large fraction of their growing real incomes upgrading the quality of autos, housing services, and other goods and, in the process, paid higher nominal prices. But the quality issue is not only a matter of doing a better job of capturing product improvements. Some analysts have argued, and provided indirect evidence, that sellers often use the occasion of changes in models and styles to raise prices. As a consequence, the current BLS procedures for pricing the new models may attribute too much of that markup to quality change and too little to “pure” price change (see Chapter 5). For purposes of well-informed macroeconomic policy making, measures of national inflation, and corresponding measures of national output, ought to incorporate the prices and output of new goods as soon as practicable and also reflect changes in the quality of goods and services to the extent that they can be reliably measured, subject to the conceptual limitations discussed in Chapters 2 and 5. Pursuit of this objective must fully recognize the difficult problems that surround the use of statistical estimation techniques to produce measures of quality change and observe the cautions we express in Chapter 4 about introducing quality adjustments into the index before sufficient preparatory research is done. These difficulties underline the potential value of a continuing research and operational effort directed toward further improving and monitoring the ability of the BLS to deal with changes in quality, the timely introduction of new goods, and related issues. INFLATION INDICATORS FOR MACROECONOMIC POLICY The central reason that inflation of any significant magnitude is considered undesirable and economically destructive is that it creates obstacles and uncertainties for business firms and individuals in planning and making commitments for the future. In theory, if the rate of future inflation were known with certainty, then with minor inconvenience one could operate with equal efficiency under a high or a low rate of inflation. But the future rate of inflation is never known with

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes certainty. Moreover, inflation tends to be more variable and relative prices subject to larger changes when overall inflation is high than when it is low. To the extent that inefficiencies in business planning are a major concern, an index that somehow combined both input and output prices would be desirable conceptually, perhaps supplemented by some measure of the variation in relative prices. But since the cross-sectional and time-series variations in inflation tend to be positively correlated with the overall inflation rate, some broad measure of inflation—the CPI or the price index for GDP—is a suitable indicator for purposes of monetary policy. The Technical Note at the end of this chapter compares the GDP price index, the CPI, and the NIPA price index for consumer expenditures as measures of inflation. Over the last 10 to 15 years, central banks of the economically advanced countries have increasingly come to define their principal, if not always sole, objective as the pursuit of low and stable inflation rates. In some countries these are embedded in numerical targets set by law. The U.S. Federal Reserve has no statutory numerical target, but it does seek a low inflation rate as a top-priority objective.25 Even if modest changes in inflation may not be harmful to macroeconomic performance or consumer welfare, charging the central bank with achieving sustained low inflation is likely to be desirable. It may, for instance, reduce the temptation to adjust macroeconomic policy to avoid politically difficult decisions by allowing inflation to inch persistently upward. When inflation strongly overshoots a central bank’s targets, “errors” of 0.5 or 1 percent a year in measuring inflation clearly do not matter very much—the steps the central bank ought to take are obvious. But when inflation is in the general neighborhood of the target, small measurement differences can loom large in policy making (which would necessarily be the case if the inflation target were set in law). In this view, the monetary authorities would mistakenly restrict the economy if inflation were above the target due to measurement error, whereas such restrictive policy would be unnecessary if the inflation measure were “corrected” to remove the source of error. Because the transition costs of pushing the inflation rate down by, say, a further 1 percent are very substantial, the magnitude of any upward bias in the inflation index can be a critical question of monetary policy. It is not at all clear, however, that this is a correct view of how monetary authorities should react to a change in index measurement techniques. 25   The Federal Reserve Act directs the Federal Reserve to seek “to promote effectively the goals of maximum employment, stable prices, and moderate long term interest rates.” But as the Federal Reserve Board has stated, “A stable level of prices appears to be the condition most conducive to maximum sustained output and employment and to moderate long-term interest rates” (Federal Reserve Board, 2000). History makes it clear that the Federal Reserve considers an inflation rate persistently in the very low single digits to be consistent with the statutory objective of “stable prices.”

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes Assume that the actual rate of inflation is a little above the target and that new statistical procedures are introduced into the index which lower the reported inflation rate by x percent, where x is a small number. Two alternative policy responses could occur. First, the central bank finds that reported inflation now meets its target and relaxes its efforts to drive inflation lower. Alternatively, it could reduce the target inflation rate by x percent. Since nothing in the real economy has changed by virtue of the statistical correction, the former response represents an implicit judgment that (unbeknownst to policy makers) the old target was in fact too high and was unduly restrictive for the economy. But mere changes in measurement can hardly be taken as evidence regarding the appropriateness of the original inflation targets. In the absence of any reason to believe that the old targets were inappropriate relative to the way inflation was reported at the time, the logical reaction to a change in inflation stemming from revised measurement is to alter the target correspondingly. Because the CPI is a focal point for many transactions, a statistically induced change in the index may, during some transition period, have important real effects—for example, on wage bargains and therefore on actual inflation. The central bank would have to take those changes into account in formulating its operating policy; but it would seem logical to adjust the target itself to reflect the statistical change. Since the CPI is used for indexing public transfer payments, the income tax code, and private contracts, even those methodological changes that produce relatively small differences in the annual rate of inflation can have significant consequences for government budgets and the welfare of individuals as their effects cumulate over time. When used as an output deflator, even small changes in the GDP price index can cumulate into larger differences in reported output gains and have at least modest effects on the assessment of real economic performance. But as an inflation indicator for the central bank and other policy makers, small differences in reported inflation rates resulting from statistical changes in the CPI or design differences between different price indexes are probably not very significant. Thus, the difference between the use of a Laspeyres or a superlative index (at the upper level) represents a choice of at least some importance for most uses of the CPI, but not for its use by the central bank as an inflation indicator. Even if such small differences should matter, they will likely have been incorporated into the decision making of these sophisticated users of price indexes. In addition to the use of an inflation index as a measure of their policy target, central banks and other policy makers use various derivative measures of current inflation as devices to help them to filter out the substantial “noise” in the month-to-month changes in inflation so as to detect significant changes in its underlying level. Many different kinds of measures can be used, such as the “core” rate of inflation (excluding such volatile items as food and energy); a “stripped” rate of inflation, removing (for example) the 10 percent of items with

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes the largest positive and negative changes; and the weighted median rate of change of the individual items in the index (see Cecchetti, 1997). Policy makers also typically want to examine current trends in producer prices for finished, intermediate, and crude goods, as well as import prices and the employment cost index. However, the central banks and governments of advanced countries have sizable economic and statistical staffs who can manipulate individual components of the CPI, the PPI, and other indexes to produce these sorts of analytical tools. Except perhaps by generating a demand for additional or different subaggregations of goods, these are not matters that involve the basic design of the indexes. TECHNICAL NOTE: THE CPI VERSUS THE NIPA PRICE INDEX AS AN INFLATION MEASURE As measures of aggregate inflation, the NIPA indexes have some advantages over the CPI. First, an overall measure of inflation ought to include the prices of all privately produced goods and not only consumer goods. Some have argued that if an index of inflation for the goods purchased by consumers remains stable, changes in the prices of investment goods can be ignored, since satisfying consumer needs is the objective of the economy. But that is only true in a steady state and is not a very helpful assumption when analyzing economic conditions, which often change rapidly and unexpectedly. Aggregate NIPA inflation indexes are available quarterly for total GDP and for gross domestic purchases (including imports, by all domestic users) and for a fairly detailed set of components, including, of course, many categories of consumer goods.26 The fact that an index that includes capital goods is a useful inflation indicator does not imply that indexes of consumer prices, CPI or NIPA, should abandon the current practice of pricing the service flows from the housing stock and return (in the case of the CPI) to pricing house purchases. The broader price index, congruent with the definition of GDP, should (and does) price both the flow of goods and services for consumption and the stream of outlays on all capital goods, including housing. One characteristic of the NIPA inflation indexes is that they provide a Fisher-type index on a real-time basis, whereas such an index for the CPI will only be made available after a 2-year lag. To estimate a real-time Fisher index for the CPI, it would be necessary to rely on a forecast approximation. In a comparison made for recent years, the real-time CPI, which is Laspeyres at the upper 26   Publishing a measure of inflation for the private-sector GDP might be helpful for some individual users. It would exclude the output of government employees (in which, by assumption, productivity gains are zero and price changes equal wage changes) and would save users the effort of calculating it from the published data.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes level, has averaged 0.15 percent a year higher growth than the Fisher, although the differences varied from year to year, reaching 0.5 percent in one year. The domain of the NIPA index includes all medical consumption expenditures, including those paid for by Medicare, Medicaid, and employer-paid health insurance, which currently account for 17.5 percent of the weight in the index. The CPI only includes consumer out-of-pocket costs, which represent about 6.0 percent of the weight in that index. In most (but not all) years of the past several decades, the price of medical care has been rising faster than the average for other consumer goods and services, raising the NIPA inflation rate above the CPI rate. However, the NIPA (atypically) uses not the CPI but the PPI index for hospital service inflation, which is differently constructed and shows a lower rate of increase. On balance, the treatment of medical care costs in the NIPA index raises its measure of consumer goods inflation slightly relative to the CPI rate in most years. As a measure of inflation (as opposed to an index used for compensation purposes) the NIPA estimate, which comes closer to covering all medical service prices, is the more relevant one. Table 7-2 compares the annual inflation rates of three alternative indexes since 1991: the NIPA GDP and personal consumption expenditure (PCE) price indexes and the BLS research index (CPI-U-RS) that extends backward in time the major revisions in CPI measurement techniques that have been introduced over the last several decades. (Many of these changes had been incorporated earlier in the NIPA indexes, and several others were extended back to 1978 in the latest NIPA revisions.) With the exception of 1996, the differences between the two measures of consumer goods inflation are small. There are a number of conceptual differences TABLE 7-2 Comparison of NIPA and CPI Indexes, 1991-2000 (percent change) Year NIPA, GDP NIPA, PCE CPI-U-RS 1991 3.6 3.8 3.7 1992 2.4 3.1 2.7 1993 2.4 2.4 2.6 1994 2.1 2.0 2.2 1995 2.2 2.3 2.5 1996 1.9 2.1 2.7 1997 1.9 1.9 2.2 1998 1.2 1.1 1.4 1999 1.4 1.6 2.0 2000 2.3 2.7 3.4 NOTES: NIPA, National Income Product Accounts; GDP, gross domestic product; PCE, personal consumption expenditures; CPI-U-RS, research series using the urban consumer price index.

OCR for page 191
At What Price?: Conceptualizing and Measuring Cost-of-Living and Price Indexes between the two indexes, including the treatment of medical services; the fact that the PCE has updated weights every 4 or 5 years whereas the CPI had 1982-1984 weights until 1998; and the use of a Fisher index for aggregation in the PCE versus a fixed-weight Laspeyres for the CPI. On balance, these differences tend to lower the PCE slightly relative to the CPI: over the 10 years shown in the table, the average annual increase in the PCE was 0.2 percent less than in the CPI. Overall, however, the two indexes move very closely together, exhibiting much the same downward trend in inflation over the period and, with the exception of 1996, roughly the same pattern of small fluctuations around the trend. Except for a 1-year deviation in 1992, the price index for GDP moves closely with the consumption price indexes—not too surprisingly since consumption is two-thirds of GDP. There is not much to choose among the indexes as an indicator of general inflation. The NIPA indexes allow an analyst to keep track of indexes for investment goods, exports, and consumer goods estimated on a basis consistent with the index for total GDP. It has up-to-date weights and a small advantage in being aggregated as a Fisher index. The CPI, in contrast, is produced monthly—although this advantage is attenuated by the degree of noise that accompanies monthly data. Luckily the Federal Reserve, the executive branch, and the Congressional Budget Office have enough analysts to use and compare both sets of indexes and their various components.