The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Business Review Federal Reserve Bank of Philadelphia November •December 1995 ISSN 0007-7011 Business Review The BUSINESS REVIEW is published by the Department of Research six times a year. It is edited by Sarah Burke. Artwork is designed and produced by Dianne Hallowell under the direction of Ronald B. Williams. The views expressed here are not necessarily those of this Reserve Bank or of the Federal Reserve System. SUBSCRIPTIONS. Single-copy subscriptions for individuals are available without charge. Insti tutional subscribers may order up to 5 copies. BACK ISSUES. Back issues are available free o f charge, but quantities are limited: educators may order up to 50 copies by submitting requests on institutional letterhead; other orders are limited to 1 copy per request. Microform copies are available for purchase from University Microfilms, 300 N. Zeeb Road, Ann Arbor, MI 48106. REPRODUCTION. Perm ission must be obtained to reprint portions o f articles or whole articles. Permission to photocopy is unrestricted. Please send subscription orders, back orders, changes o f address, and requests to reprint to Publications, Federal Reserve Bank o f Philadelphia, Department o f Research and Statistics, Ten Independence Mall, Philadelphia, PA 19106-1574, or telephone (215) 574-6428. Please direct editorial communications to the same address, or telephone (215) 574-3805. Digitized for 2FRASER NOVEMBER/DECEMBER 1995 HAS DEUNIONIZATION LED TO HIGHER EARNINGS INEQUALITY? Martin A. Asher and Robert H. DeFina The past 20 years have seen a consider able rise in earnings inequality. One ap parently significant source of this in creased inequality is the large decrease in the unionized segment of the labor force. A ccord ing to som e stu d ies, deunionization may explain between 10 and 20 percent of the increase in earnings inequality of the past two decades. Bob DeFina sifts through the research on the this relationship and summarizes the re sults. MEASURING INFLATION IN A HIGH-TECH AGE Leonard Nakamura Many people use the Consumer Price Index as a guide to inflation. But how accurate is the CPI as a measure of infla tion? For example, the CPI shows that real wages for Americans are lower than they were two decades ago. But is this really the case? Probably not. In this ar ticle, Leonard Nakamura takes a close look at the CPI, discusses some of the factors involved in measuring inflation, and suggests possible solutions to the problems of collecting data and construct ing more accurate cost-of-living indexes. FEDERAL RESERVE BANK OF PHILADELPHI Has Deunionization Led to Higher Earnings Inequality? Martin A. Asher and Robert H. DeFina* o ne of the notable economic trends of the past 20 years is a dramatic rise in earnings inequality. That is, earnings are distributed much more unevenly across the population now than in the mid-1970s. The increase was especially pronounced in the 1980s, a period of strong overall economic growth. In virtually all economies, and especially in a market economy like the United States, varia tions in earnings from one person to the next ’‘’Martin Asher is an assistant professor of economics, College of Commerce and Finance, Villanova University, Villanova, PA. Robert DeFina is an associate professor of economics, College of Commerce and Finance, Villanova University. When this article was written, DeFina was a visiting scholar in the Research Department of the Phila delphia Fed. are a fact of life. Such pay differences can serve several functions. An individual's relatively high income might, for example, reflect long years of education and experience and a commensurately high level of productivity. Higher pay might also represent a reward to individu als who take greater employment and invest ment risks. For these reasons, at least some earnings disparity among individuals is not only commonplace but desirable. Still, recent inequality trends have led community leaders, policymakers, and social commentators to question whether current income disparities are in the country's best interests. Some observers, such as the National Conference of Catholic Bishops, have chal lenged the fairness of our earnings distribu tion. Academic research, meanwhile, has iden3 BUSINESS REVIEW tified a link between higher in equality and slower economic growth, although the key issue of cause and effect has yet to be determined.1 Given these concerns, interest has heightened in identifying possible reasons for the upward trend. One apparently significant source of earnings inequality is the large decrease in the union ized fraction of the labor force over the past 20 years. Depend ing on which group of earners is studied, "deunionization" ap pears to explain between 10 per cent and 20 percent of the in creases in inequality of the past two decades. The downward trend in unionization is unlikely to be reversed significantly. Thus, absent some offsetting changes— be they government interven tions or private sector develop ments—a noticeable part of the unsettling rise in inequality will remain. RECENT TRENDS In eq u ality Has Been Rising...Earnings inequality re fers to the degree of variation in earnings across a particu lar population, and it can be gauged in different ways. Two common summary measures are the Gini index and the income quintile ratio. The Gini index varies be tween zero and one, with values aRoberto Chang's 1994 article provides an accessible review of the literature on the relatio n sh ip betw een econ om ic growth and income inequality. 4 NOVEMBER/DECEMBER 1995 The Gini Index of Inequality cumulative income share 15 The Gini index of inequality is based on a construct called a Lorenz curve. The Lorenz curve plots the cumulative share of the population on the horizontal axis and the associated cumulative share of income on the vertical axis. If income were distributed perfectly equally, the Lorenz curve would be a straight line, like line A in the above graph. Thus, 10 percent of the population would have 10 percent of the income, 40 percent of the population would have 40 percent of the income, and so on. In reality, income is distributed unequally, resulting in a Lorenz curve like that in line B. The curve means that a cumulative share of the population receives less than its proportionate share of income. For example, the indicated point on B shows that 30 percent of the population receives only 15 percent of total income. The Gini index summarizes the inequality revealed by the Lorenz curve in a single number. The index is computed by dividing the area between lines A and B by the total triangular area under line A. If income were distributed equally, the Lorenz curve (line B) would lie on top of line A. In this case, the Gini index equals zero, since there would be no area between lines A and B. Alterna tively, if one person receives all income so that there is maximum inequality, the Lorenz curve would lie along the horizontal and vertical axes (the cumulative population shares would receive no income, until the last person, who receives all income). In this case, the Gini index equals one, since the area between the Lorenz curve (B) and the line of perfect equality (A) and the area under the line of perfect equality are equal. Thus, the values of the Gini index range between zero and one, with larger values indicating greater inequality. FEDERAL RESERVE BANK OF PHILADELPHIA Martin A. Asher and Robert H. DeFina Has Deunionization Led to Higher Earnings Inequality? closer to one indicating higher inequality. (See The Gini Index of Inequality.) The income quintile ratio is computed by dividing the average income of individu als in the highest income quintile by that of individuals in the low est quintile. A higher ratio sig nals growing inequality. Each of these measures has trended up during the past two decades, leaving measured in equality at its highest level in 25 years. (See An Upward Trend in Inequality.) Other measures of inequality demonstrate a similar upward trend, and the growth in inequality has not been seriously questioned.2 ...In Good Times as Well as Bad. The forces underlying period-to-period changes in earn ings inequality are complex. At times, shifts in the degree of in eq u ality sim ply m irror the economy's business-cycle fluc tuations. Other things being equal, economic expansions gen erally reduce inequality while recessions increase it.3*The in verse link arises because lower skilled, lower paid workers are 2The issues and studies discussed in this article are based on wage and salary data. The monetary value of employee benefits, such as health insurance, are typi cally not included because of a lack of data. A measure of total compensation, as opposed to wage and salary earnings, would include benefits. 3Empirical evidence on this point is presented in articles by Rebecca Blank and Alan Blinder (1986) and Nathan Balke and Daniel Slottje (1992), among others. An Upward Trend in Inequality 67 72 77 82 87 92 The above graph shows the behavior of two income inequality measures from 1967 to 1992. The Gini index summarizes the extent of inequality with values ranging from zero to one; values closer to one indicate greater inequality. The income quintile ratio is computed by dividing the average income of individuals compos ing the top 20 percent of all earners by the average income of the lowest 20 percent of earners. As with the Gini index, higher values indicate greater inequality. Because of the different range of values possible for each measure—the Gini index has a maximum value of one while the quintile ratio has no maximum— different scales are used for each. The left axis refers to values for the Gini index, while the right axis refers to values for the quintile ratio. Each of the measures fluctu ated with no discernible trend between 1967 and the mid-1970s. Around 1975, measured inequality began an upward trend, which continued into the 1990s. The rise was especially pronounced in the 1980s, a period of generally strong economic growth. Mea sured inequality currently stands at its highest level in the past 25 years. The data used in the graph come from the federal government's Current Population Reports (P-60 series) and relate to family income. Summary data on household income are available and produce similar upward trends. Data on inequality among indi vidual incomes are not published. 5 BUSINESS REVIEW NOVEMBER/DECEMBER1995 often fired first during economic downturns. In a relative sense, recessions thus cause lower income individuals to become poorer and in equality to grow. But other things need not be equal, and temporary cycles of expansion and contrac tion thus need not be the only, or even the most important, underlying influence. More funda m ental and lon g-lastin g changes in the economy's structure can, and have, caused substantial movements in inequality, both in the United States and elsewhere.4In such cases, periods of economic growth might coincide with either rising or falling inequality. An example of a structural change is the introduction of new technology, such as the personal computer. The development and use of personal computers can help strengthen economic growth over long periods. How ever, all workers might not share equally in the additional growth. Highly educated employ ees might be better prepared to deal with the complexities of computers than their less edu cated counterparts. Indeed, low-skilled em ployees might see their jobs eliminated as the workplace becomes increasingly computer ized. Consequently, while the overall economy might improve, earnings inequality might in crease. The potential role of structural change is critical to understanding recent inequality trends, in that the rise has spanned periods of healthy overall economic growth. For example, between 1983 and 1990, the country experi enced a prolonged business expansion, with real GDP growing an average of about 3.3 percent per year and the civilian unemploy ment rate falling from 9.6 percent to 5.3 per- cent. Yet, measured inequality rose steadily throughout the period and has continued to grow in the current expansion. As mentioned before, business-cycle expansions such as those in the 1980s and 1990s tend to decrease in equality. That inequality rose suggests impor tant structural influences were also at work. The question remains as to what these are. A burgeoning literature has examined nu merous possibilities, although many have been discounted as unimportant.5 There appears to be some agreement that technological changes, such as increased computerization, have in creased the relative demand for higher paid, higher skilled workers, pulling their earnings further above those of lower paid, lower skilled individuals.6 Some studies have also focused attention on the effects of a sizable decrease in the fraction o f the unionized labor force. 4Actual examples are provided in economic historian Jeffrey G. W illiamson's book, which details and analyzes movements in American income inequality during the past two centuries. Also see the 1992 study by Claudia Goldin and Robert A. Margo. 7The effects of unions on the distribution of income are discussed in Richard Freeman (1980), Richard Freeman and James Medoff (1984), and Richard Freeman (1993). For a textbook discussion of these points, see Campbell McConnell and Stanley Brue (1995), especially chapter 11. 6 UNIONS AND THE DISTRIBUTION OF INCOME Greater Inequality...In theory, unions af fect the distribution of earnings in several conflicting ways.7*To some extent, unions pro 5John Bound and George Johnson (1992), Frank Levy and Richard Murnane (1992), and Martin Baily, Gary Burtless and Robert Li tan (1993) provide an overview of possible factors, including changes in the age, gender, industry, and occupational composition of the labor force, and increases in the returns to education and skill. In addition, the January 1995 issue of the Federal Reserve Bank of New York's Economic Policy Review also contains numerous articles describing and analyzing recent trends in inequality. 6David Howell's (1995) analysis questions the impor tance of technological change and emphasizes other changes in managerial strategies aimed at reducing short term labor costs. FEDERAL RESERVE BANK OF PHILADELPHIA Has Deunionization Led to Higher Earnings Inequality? Martin A. Asher and Robert H. DeFina unions tend to standardize wage rates across firms that have unionized work forces. Thus, unionized supermarket cashiers would receive similar wages, regardless of which store em ploys them. Finally, unions shrink the earnings differential between white-collar and blue-col lar workers because union membership, which receives the wage premium, is disproportion ately blue collar. Although unions have a theoretically am biguous effect on inequality, a long line of empirical research has found that unions have generally reduced inequality.8 These studies suggest that the declining presence of unions in the work force during the past two decades may have contributed importantly to recent trends toward inequality. RECENT TRENDS IN UNIONIZATION The decline during the past 20 years in the percentage of the work force that is unionized has been dramatic. Between 1970 and 1992, this fell by half, from about 26 percent to about 13 percent. In addition, between 1980 and 1992 the absolute number of union members fell sharply. By 1992, active union membership had fallen almost 27 percent from its 1980 level. (SeeUnion Representation Has Fallen Mark edly.) Structural Economic Change. The reasons for the decline in unionization are many.9 A major force has been numerous structural changes that have buffeted the U.S. economy over the past two decades. A much-discussed change has been a shift in employment pat terns away from traditional blue-collar, union ized manufacturing jobs toward typically lessunionized service-sector jobs. This change re flects both technological change and shifting trade patterns that have relocated manufac turing activities to lower-wage labor markets abroad. The increasing employment of women, youths, and contingent workers has likewise diminished the presence of unions, as these groups are, on average, less unionized. Greater Managerial Opposition. In addi tion to structural changes, unionism has de clined in the face of increased managerial op position. Indeed, a 1985 article by Richard Freeman argues that opposition has taken both legal and illegal forms, and that overall, mana gerial opposition tactics have emerged as the key force behind the decline of unions. In part, managerial opposition stiffened in the 1980s as the Reagan and Bush administrations took an avowedly anti-union stance through increas ingly pro-business rulings by the National La- 8The literature includes David McCabe (1912), Lloyd Reynolds and Cynthia Taft (1956), Richard Freeman (1980, 1982, 1993), Richard Freeman and James Medoff (1984), and David Card (1992). 9A detailed discussion of these factors is found in Edward Lazear, Richard Freeman, and Melvin Reder (1988) and Campbell McConnell and Stanley Brue (1995), espe cially chapter 10. mote earnings inequality, in part because unions raise the wages of members above those of similar nonunion workers. The difference is known as the union wage premium. A 1986 book by H. Gregg Lewis collected existing estimates of the wage premium. The studies cited placed the average premium at about 15 percent, although the premium varied widely across industries and has changed over time. The wage premium also contributes to in equality because union membership histori cally has been concentrated among higher skilled, blue-collar workers. As a result, work ers who already earn relatively high wages are pulled even higher above their lesser skilled counterparts. ...Or Less? Unions promote earnings equal ity in three ways. First, the imposition of union wage scales equalizes earnings across workers in a given unionized firm. For instance, cash iers in a particular supermarket would receive sim ilar w ages. Second, and in a related vein, 7 NOVEMBER/DECEMBER1995 BUSINESS REVIEW ate climates. Some au thors have argued that unions' success in gener ating wage premiums it self may have driven Percent of Millions of firm s to the sun belt where union organizing has been relatively less successful. Historically, the labor movement in the United States has been strongest in urban areas. Indeed, the 1981 estim ates of M arten Estey indicate that New York, California, Penn sylvania, Illinois, Ohio, and M ich ig an — all heavily urbanized and in dustrialized areas— ac count for about one-half of union membership. By The above graph presents data on both the number of union members contrast, unionization and on the union membership rate— the percent of the labor force that is rates in the South are a unionized. The unionization rate is measured on the left axis while small fraction of the rates membership levels are measured on the right axis. Both the level and rate in the rest of the country. of membership peaked during the 1970s, then fell dramatically over the Why this difference ex next two decades. The 1992 membership rate of 13 percent is about half ists is debatable, but its 1972 value. strong anti-union senti The data used in the graph come from McConnell and Brue (1995) and ment, as demonstrated are based on the federal government's Current Population Survey. by right-to-work legisla tion, is often identified as a key factor. bor Relations Board. Freeman's estimates sug Alternative Provision of Services. Finally, gest that managerial opposition represents the it's likely that unionism has declined because single largest factor responsible for the decline some of the services traditionally provided by in unionization rates during the 1970s and unions are, to an extent, now dispensed by 1980s. various levels of government and by employ Business Relocations. Another factor has ers. These include workplace safety regula been the movement of production facilities tions, improved fringe benefit packages, un from the northeast rustbelt to the south and employment insurance, workers' compensa southwest sunbelt. Analysts studying the tion, pension plans, and limits on the length of movement have pointed to increases in the the workweek. A 1984 study by George relative cost of energy over time and the result Neumann and Ellen Rissman presents evi ing desire of firms to produce in more temper dence that a sizable fraction of the decline in Union Representation Has Fallen Markedly 8 FEDERAL RESERVE BANK OF PHILADELPHIA Has Deunionization Led to Higher Earnings Inequality? Martin A. Asher and Robert H. DeFina the particular set of factors held constant, these factors generally include age, gender, indus try, and education level of individuals in the work force.10 A 1992 study by David Card and a 1993 study by Richard Freeman sought to explain the rise in earnings inequality among mature private-sector male workers, roughly ages 2554 years. Card's analysis covers 1973 to 1987, HAS DECLINING UNIONISM LED while Freeman's covers 1978 to 1988. Each TO RISING INEQUALITY? study determined that decreases in unioniza The foregoing discussion suggests that the tion rates occurring over the past 20 years decline in unionism during the past two dec account for about 20 percent of the total rise in ades may have played an important causal role inequality. Thus, the measured effects of in the recent and disturbing upward drift in deunionization are large. Similar estimates inequality. But despite consistency in the broad were presented in a 1994 study by Amanda trends of unionism and inequality and a theo Gosling and Stephen Machin, who found that retical basis for believing that the link is causal, deunionization accounted for 15 percent of the the entire rise in inequality cannot reasonably rise in inequality in Great Britain during the be attributed to declining unionization rates. 1980s. As discussed earlier, other structural eco While earlier studies focused on the sources nomic changes have occurred that could also of growing inequality among mature male have contributed to greater inequality. Many workers, interest also lies in understanding the researchers, for example, have argued that upward trend in inequality among the general changing global trade patterns, reflecting population. Thus, in 1995, we conducted a worldwide relocations of production and shifts study that examined a broader sample of U.S. in product demands, have hurt low-wage, low- earnings data for the period 1974 to 1989. skilled workers regardless of their union sta Specifically, we studied earnings of all work tus. At the same time, higher wage, higher ing-age men and women. Broadening the skilled individuals have benefited from these sample to cover individuals who are typically same developments, again, independently of less unionized (young and female workers) union status. Thus, it's plausible that inequal had the expected result of lowering the esti ity grew for reasons unrelated to declining mated impact of deunionization on inequality unionization. Other possible nonunion factors that may have fed rising inequality are large influxes into the labor force of young people 10Lynn Karoly's 1992 study and our 1994 study provide and women, two groups that tend to have evidence that these factors probably had small, if any, below average wages, and employment shifts effects on recent inequality trends. Nonetheless, it is im from manufacturing industries to low-wage portant to control for these factors when trying to sepa service industries. rately identify the role of unions. A factor that does seem To identify the separate contribution of to matter in recent inequality trends is a growing financial deunionization to inequality trends, several return to education and experience. That is, for a given distribution of education levels across the population, studies have used statistical techniques to con individuals with higher levels of education are experienc trol for the impact of other possible sources of ing faster earnings growth than individuals with lower growing inequality. While the studies differ in levels. union membership is attributable to the alter native provision of services. Similarly, esti mates presented in a 1993 analysis by Henry Farber and Alan Krueger indicate that the greater availability of benefits traditionally provided only by unions has significantly un dermined the demand for union representa tion. 9 NOVEMBER/DECEMBER 1995 BUSINESS REVIEW to about 10 percent of the total rise, a smaller but still important effect.11 CONCLUSION During the past two decades, earnings in- n Our study also examined the impact of unionization trends in the public sector on inequality there. Analyzing government workers eliminates confounding factors such as changes in industry composition of the work force and shifts in trade patterns. In contrast to the private sector, unionization has been increasing in the public sector. Our analysis found that the increasing unionization has pro duced lower inequality than would have occurred other wise. This finding further supports the notion that trends in unionization are a causal factor in the evolution of inequality. equality has risen to historically high levels. The climb has sparked a debate about whether current inequality is fair and desirable and has led analysts to search for the causes of the increase. Several possible factors have been id en tified , am ong them the m arked deunionization of the labor force that has oc curred since the 1970s. Careful statistical studies have shown that about 10 percent of the rise in earnings inequal ity among all workers and about 20 percent of the rise in inequality among mature male work ers can be attributed to deunionization. As long as unionization rates fail to regain their levels of the 1970s, higher earnings inequality will remain unless some other offsetting changes occur. REFERENCES Asher, Martin A., and Robert H. DeFina. " A Decomposition Analysis of Recent Earnings Inequality Trends," mimeo, Villanova University, 1994. Asher, Martin A., and Robert H. DeFina. "The Impact of Changing Union Density on Earnings Inequality: Evidence From the Private and Public Sectors," mimeo, Villanova University, 1995. Baily, Martin N., Gary Burtless, and Robert Litan. Growth With Equity. (Washington, D.C.: The Brookings Institution), 1993. Balke, Nathan S., and Daniel J. Slottje. "A Macroeconometric Model of Income Inequality in the U.S.," mimeo, August 1992. Blank, Rebecca, and Alan Blinder. "Macroeconomics, Income Distribution, and Poverty," in Sheldon Danziger and Daniel Weinberg, e d s Fighting Poverty: What Works and What Doesn't. (Cambridge, MA: Harvard University Press), 1986. Bound, John, and George Johnson. "Changes in the Structure of Wages in the 1980s: An Evaluation of Alternative Explanations," American Economic Review, 82 (June 1992), pp. 371-92. Card, David. "The Effect of Unions on the Distribution of Wages: Redistribution or Relabelling?" National Bureau of Economic Research Working Paper 4195 (October 1992). 10 http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis FEDERAL RESERVE BANK OF PHILADELPHIA Has Deunionization Led to Higher Earnings Inequality? Martin A. Asher and Robert H. DeFina Chang, Roberto. "Income Inequality and Economic Growth: Evidence and Recent Theories/' Federal Reserve Bank of Atlanta Economic Review (July/August 1994). Estey, Marten. The Unions, 3rd ed. (New York: Harcourt Brace Jovanovich), 1981. Farber, Henry, and Alan Krueger. "Union Membership in the United States: The Decline Continues," National Bureau of Economic Research Working Paper 4216, June 1993. Freeman, Richard. "Unionism and the Dispersion of Wages," Industrial and Labor Relations Review, 34 (October 1980), pp. 3-23. Freeman, Richard. "Union Wage Practices and Wage Dispersion Within Establishments," Industrial and Labor Relations Review, 36 (October 1982), pp. 3-21. Freeman, Richard. "How Much Has Deunionization Contributed to the Rise in Male Earnings Inequal ity?" in Sheldon Danziger and Peter Gottschalk, eds., Uneven Tides: Rising Inequality in America. (New York: Russell Sage Foundation), 1993. Freeman, Richard. "Why Are Unions Faring So Poorly in NLRB Representation Elections?" in Thomas Kochan, ed., Challenges and Crises Facing American Labor. (Cambridge, MA: The MIT Press, 1985). Freeman, Richard, and James Medoff. What Do Unions Do? (New York: Basic Books), 1984. Goldin, Claudia and Robert A. Margo. "The Great Compression: The Wage Structure in the United States at Mid-Century," Quarterly Journal o f Economics, 107 (1992), pp. 1-34. Gosling, Amanda, and Stephen Machin. "Trade Unions and the Dispersion of Earnings in British Establishments, 1980-90," National Bureau of Economic Research Working Paper 4732 (May 1994). Howell, David R. "Collapsing Wages and Rising Inequality: Has Computerization Shifted the Demand for Skills?" Challenge (January/February 1995), pp. 27-35. Karoly, Lynn. "Changes in the Distribution of Individual Earnings in the United States: 1967-1986," Review o f Economics and Statistics, 74 (February 1992), pp. 107-15. Lazear, Edward, Richard Freeman, and Melvin Reder. "Symposium on Public and Private Unionization," Journal o f Economic Perspectives, 2 (Spring 1988), pp. 59-110. Levy, Frank, and Richard J. Murnane. "U.S. Earnings Levels and Earnings Inequality: A Review of Recent Trends and Proposed Explanations," Journal o f Economic Literature, 30 (September 1992), pp. 1333-81. Lewis, H. Gregg. Union Relative Wage Effects. (Chicago: University of Chicago Press), 1986. McCabe, David A. The Standard Rate in American Trade Unions. (Baltimore: Johns Hopkins Press), 1912. McConnell, Campbell, and Stanley Brue. Contemporary Labor Economics, 4th ed. (New York: McGrawHill), 1995. 11 BUSINESS REVIEW NOVEMBER/DECEMBER1995 National Conference of Catholic Bishops. Economic Justice For All. (Washington, D.C.: United States Catholic Conference, Inc.) 1986. Neumann, George R., and Ellen Rissman. "Where Have All the Union Members Gone?" Journal o f Labor Economics, 2 (April 1984), pp. 175-92. Reynolds, Lloyd G., and Cynthia Taft. The Evolution o f Wage Structure. (New Haven, Conn.: Yale University Press), 1956. Williamson, Jeffrey G. Inequality, Poverty, and History. (Cambridge, MA: Basil, Blackwell), 1991. 12 FEDERAL RESERVE BANK OF PHILADELPHIA Measuring Inflation In a High-Tech Age w w ¥ hen Americans ask themselves former New York Mayor Edward Koch's favorite question "How'm I doing?" many of them may answer, "Not so hot." For when they look at their "real wages" corrected for inflation * Leonard Nakamura is an economic advisor in the Research Department of the Philadelphia Fed. Len thanks Robert Parker at the BEA, Patrick Jackman and John Galvin at the BLS, and John Caskey, Robert DeFina, Mark Kuperberg, Lee Ohanian, Fred Pryor, John Roberts, Ber nard Saffran, William Stull, and Mark Wynne for com ments. Leonard 1. Nakamura* using the Consumer Price Index (CPI), it ap pears that many Americans are earning less than they were two decades ago (Figure 1). Is this really the case? Or is the CPI misleading as a standard for purchasing power? I will argue that the answer to the second question is yes, which is disturbing for several reasons. What we normally mean by inflation is the loss of purchasing power of dollars. If the CPI is giving an upwardly biased view of inflation, then our inflation-adjusted measures of consumers' purchasing power and well being will be too low. As we shall see, there is 13 NOVEMBER/DECEMBER1995 BUSINESS REVIEW income tax brackets are also indexed to the CPI. If we extrapolate esti mates from the Con Hourly Wage gressional Budget Of (Deflated by CPI) fice, between these and other corrections, it is quite possible that the budget deficit would be substantially smaller if the CPI had been cor rectly measured.2* Moreover, it is quite possible that the over statement of inflation is worsening. If mismeasurement is accelerat ing, new dangers arise. Policymakers may have greater difficulty recog 1960 1965 1970 1975 1980 1985 1990 1995 nizing progress toward price stability. And the The real wage rate of $7.43 for 1995 equals the July 1995 actual measurement. inflation-adjusted rate of interest—adjusted by the CPI— may be in some evidence that the CPI has been upwardly creasingly misleading to borrowers, lenders, biased more than 1 percent annually over the and policymakers. Why might the overstatement of inflation be past 20 years. If the CPI is revised down 1 percent annually, the post-1975 decline in real worsening? Two complementary trends are at wages disappears (Figure 2). Second, many work. First, high-tech investment goods (such economic payments—including major ones as computers and telecommunications equip such as Social Security benefits and federal ment) are playing an increasing role in the income taxes—are tied to the CPI to insulate economy and are rapidly increasing their abil them from inflation, but the CPI may be sys ity to store, process, and transmit data, text, tematically distorting them. Some Social Secu rity benefits may be too high by 20 percent, again using the 1 percent estimate.1 Personal FIGURE 1 2It has been argued that the adjustment for the cost of living in Social Security benefits is inadequate because rapidly inflating medical costs are a larger part of expenses for the retired. However, as I argue below, rapid inflation in medical expenses appears to be, at least in part, an artifact of inadequate adjustment for the increased quality of medical care. 14 2See Peterson, Congressional Budget Office, October 1994. Peterson's paper estimates that an increase of 0.1 percent in the CPI adds $ 0.5 billion to the deficit, so a CPI that is 20 percent lower would mean roughly a $100 billion lower annual deficit. To this must be added the cumulative effect of past overly large deficits on the total net interest paid, which would also lower the current deficit. How ever, the effect is actually somewhat less, since Congress has not always allowed the inflation adjustment to tax brackets to come into effect. FEDERAL RESERVE BANK OF PHILADELPHIA Measuring Inflation in a High-Tech Age Leonard I. Nakamura images, and sound. This FIGURE 2 progress has both direct and indirect impacts on high-tech consumption Hourly Wage Adjusted for Lower Inflation (1%) goods, such as home (Deflated by CPI) PCs and cars, and ser vices, such as telecom munications, cable TV, and medical care. The CPI u nd erm easures these new products and services and their con tributions to welfare. Second, improvements in consumption in the United States now in creasingly take the form of greater variety and quality, rather than in creases in qu an tity.3 1970 1975 1980 1985 1990 1995 New products and ser vices (both high tech This graph is intended to show a reasonable minimum on the bias in the CPI, based on and low tech) are pro documentation to date. It is not intended to show the most likely amount of bias or the liferating at an increas right amount of acceleration in bias. The top line represents the effects of lowering the ing rate. For example, inflation rate 1 percent. The last data entries, for 1995, reflect July 1995 data. half of all prescription based on surveys of consumer purchases taken drugs and one-third of breakfast cereals have in the years 1982 to 1984, more than a decade been introduced in the last decade. Unfortu ago. New goods and services are conceptually nately, the CPI is not designed to capture irrelevant, since the basket of goods whose changes in consumer welfare due to new goods prices are measured is held fixed. (In practice, and services. the BLS cannot stick to the pure standard of a The CPI measures the changing price of a fixed basket of goods, because many of the fixed basket of goods. Currently, that basket is goods and services available in 1982-84 have become obsolete and are no longer sold.4) If most of our economic progress is due to changes in the types of goods and services consumed, 3Th eEconomic Report of the President, February 1995, puts the CPI is not a good guide to how the cost of it this way: "The output of the economy increasingly is shifting away from standardized commodities with easily maintaining a certain standard of living is definable characteristics that change little over time, to evolving. This is not the fault of the Bureau of wards goods and services for which issues of quality and Labor Statistics (BLS), which is responsible for even definition are of primary importance." See Leonard I. Nakamura, "Is U.S. Economic Performance Really That Bad?" Federal Reserve Bank of Philadelphia Working Pa per No. 95-21 for evidence on the deceleration of physical output and consumption in the United States over the course of the 20th century. 4A s we shall see below, in practice the CPI does incor porate some new products, although it does not fully capture their benefits. 15 NOVEMBER/DECEMBER1995 BUSINESS REVIEW collecting the basic data for and calculating the CPI, but a problem created by the increasing divergence between the conceptual basis of the CPI and a true cost-of-living index. These problems are not insuperable, but solving them may involve a change in our approach to the collection of statistics and in our use of economic theory to guide us in setting up the indexes. Ironically, the im provements in information processing and communications that are helping to make our current data-collection system outmoded could facilitate sharp improvements in price mea surement. DOES THE CPI MEASURE THE COST OF LIVING? A true cost-of-living index answers the ques tion: How much must I spend to maintain my standard of living? For example, if I must spend 5 percent more this year to give me the same enjoyment as last year, a true cost-ofliving index would rise 5 percent. Such an index ought to take into consideration chang ing prices of one good relative to another and new goods and services that become available and how people benefit from these changes. These considerations are important because to take advantage of these changes as they occur, I will likely buy a different bundle of goods this year, even if I can still afford to buy what I bought last year. Relative Changes in Price. Suppose I buy only clothing and computer supplies. In 1994, I had $200 to spend on articles of clothing priced at $10 each or on computer supplies also $10 each, and I bought 10 clothing items and 10 computer items. In 1995, I am earning 5 per cent more, or $210, and clothing increases in price to $12 while computer supplies decrease to $ 9 .1 could again buy 10 clothing items and 10 com p u ter item s, but I choose to buy seven clothing items and 14 computer items (Table 1 ). My standard of living has improved: I could have bought exactly what I did last year, but I didn't. I prefer what I am buying this year, which I couldn't buy last year, so I am better off. A true cost-of-living index should help us measure, as precisely as possible, this improve ment in the standard of living. But the CPI, conceptually, doesn't do this. It takes the base-year bundle of goods (1994, in our example) and asks: How much does it cost to buy that same bundle this year? In our example, the base bundle costs $200 in 1994 and $210 in 1995, a 5 percent increase. So if we set the CPI equal to 100 in 1994, it would equal 105 in 1995 (Table 2). Since my income has risen 5 percent and the CPI has risen 5 percent, if I think of the CPI as a measure of the cost of living, my "real income" is unchanged. But as I've shown, I'm actually better off this year because I can buy a bundle of goods I prefer to the one I bought last year. If we compare the total purchases columns for 1995 in Tables 1 TABLE 1 Actual Purchases 1994 1995 Item Price Quantity Total Price Quantity Total Clothes $10 10 $100 $12 7 $84 Computer $10 10 $100 $9 14 $126 Total 16 $200 $210 FEDERAL RESERVE BANK OF PHILADELPHIA Measuring Inflation in a High-Tech Age Leonard I. Nakamura and 2, we see the source of the difference: the CPI puts greater weight on clothes, the good whose quantity has fallen, relative to my ac tual current purchases. A price index like the CPI that uses baseyear weights (in our example, 1994) has a somewhat out-of-date comparison basket; a natural alternative would be to use an endyear weighted index (in our example, 1995).5 Using 1995 weights would result in giving greater weight to computer items whose price has fallen; buying 14 computer items and seven clothing items in 1994 would cost $210, the same as in 1995, so an end-year index would show no price increase. However, it is difficult to do this in a timely fashion, 1995 quantity weights generally not being available until 1996. An alternative is to set up a quantity index and derive an implicit price deflator. The Bureau of Economic Analysis (BEA) begins by constructing a measure of real personal con sumption expenditures (PCE) by valuing cur rent consumption at base-year prices. In our example, that means measuring 1995 consump tion (seven pieces of clothing and 14 computer items) at 1994 prices ($10 for each). By this measure, this year's consumption has risen 5 percent, to $210 (Table 3). This index of real purchases can be converted to an implicit price index by dividing the real purchases into nomi nal purchases: for 1994 this is $200/$200 = 1 and for 1995 this is $210/$210 = 1, so according to this implicit price deflator, the 1995 price level is unchanged from the 1994 level, just like the end-year index.6* In general, price indexes like the CPI (which compares today's cost of the base year consump tion bundle to what it cost in the base year) will tend to understate improvements in welfare 5A base-year quantity weighted price index is called a Laspeyres index. An end-year quantity weighted price index is called a Paasche index. 'This is not a coincidence— an end-year quantity weighted price index is the same as the implicit price deflator for a base-year price weighted quantity index for the entire period between the base and end years, but not for intermediate years. For example, if the base year is 1994 and the end year is 1996, the 1996 end-year price index and the implicit price deflator whose base year is 1994 will agree on the overall inflation rate from 1994 to 1996, but not necessarily on the rates for each intermediate year 1994 to 1995 and 1995 to 1996. The implicit price index for the intermediate years does not have a clear interpretation, being a kind of odd residual, unlike the end-year price index. TABLE 2 Consumer-Price-Index Method of Calculation 1994 Item Price Clothes $10 Computer $10 Total Quantity 1995 Total Price 1994 Quantity Total 10 $100 $12 10 $120 10 $100 $9 10 $90 $200 $210 CPI 1994=100 100 105 Real Spending deflated by CPI $200 $200 17 NOVEMBER/DECEMBER1995 BUSINESS REVIEW TABLE 3 PCE and Implicit Price Deflator Method of Calculation 1994 1995 Total 1994 Total Item Price Quantity Purchase Price Quantity Purchase Clothes $10 10 $100 $10 7 $70 Computer $10 10 $100 $10 14 $140 Real Spending in 1994 prices $200 $210 Total Spending in current year prices $200 $210 100 100 Implicit price deflator and overstate price increases, while price in dexes like the implicit price deflator (which compares today's cost of today's consumption bundle to what it would have cost in the base year) will tend to exaggerate improvements in welfare and implicitly understate price in creases.7 These effects increase as prices di verge further from those in the base year. So when there are divergent price trends, these effects accumulate over time until the base year is updated. One quick fix for the problem, proposed early in this century by economist Irving Fisher, is to update every year and average the quan tity-based index and the price-based index. Fisher's so-called Ideal Index multiplies the two indexes and then takes the square root. In our example, this would result in an inflation rate of 2.4 percent, about midway between the zero inflation of the BEA's method and the 5 percent inflation of the BLS's. The BEA has begun emphasizing Fisher Ideal price and out- 7This holds for inflation after the base year; the base year is periodically updated. 18 http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis put indexes in its reporting of the U.S. Gross Domestic Product and its components (see BEA's Chain-Weighted Price and Quantity In dexes).8 How large is the impact of the relative price problem that we have described? It depends on the rate of divergence of relative prices of goods and the average interval between up dating the market basket. Historical compari sons of a Fisher Ideal Index with the CPI provide a measure of the size of the upward bias in the CPI growth rate, and such studies broadly agree that the bias is about 0.2 percent age points annually.9 8For further discussion, see Robert Parker, "Preview of the Comprehensive Revision of the National Income and Product Accounts: BEA's New Featured Measures of Out put and Prices," Survey of Current Business, August 1995; and Jack E. Triplett, "Economic Theory and BEA's Alterna tive Quantity and Price Indexes," Survey of Current Busi ness, April 1992. 9Marilyn E. Manser and Richard J. McDonald, "An Analysis of Substitution Bias in Measuring Inflation, 19591985," Econometrica, 56, July 1988, pp. 908-30; and Ann M. Aizcorbe and Patrick C. Jackman, "The Commodity Sub- FEDERAL RESERVE BANK OF PHILADELPHIA Measuring Inflation in a High-Tech Age Leonard I. Nakamura The BEA's Chain-Weighted Price and Quantity Indexes The Bureau of Economic Analysis, beginning at the end of 1995, will emphasize chain-weighted price and quantity indexes (Fisher Ideal indexes) in its monthly reports on the national income and product accounts, rather than the more familiar output measures based on 1987 prices. The result will be lower measured real output and higher prices compared with the 1987 price-weighted output and the implicit price deflator. The major reason for the reduction in estimated output is a reduced weight for computers, cut by more than half. The BEA is entirely correct to make this change. The change makes more urgent, however, revisions to the fundamental price series to enable them to capture more of the consumer surplus lost using current measurement conventions. The BEA and the BLS (which has the primary responsibility for price measurement) are, of course, taking steps in this direction.* For example, the BLS is correcting the rotation bias in foods, where it appears to be most important, and it is improving its methodology for measuring prices of prescription drugs as well. But other steps will take longer to implement, particularly under the tightening budget constraints these agencies face. The net result is that, in the short run, measures of price and output may be further from, rather than closer to, reality. *BEA, "Mid-Decade Strategic Review of BEA's Economic Accounts: An Update," Survey of Current Business, April 1995, pp. 48-56. New Products in Old Price Indexes. The solution outlined above gets us only part of the way toward a true cost-of-living index because the price indexes typically ignore new goods, and when they do consider these goods, they do so too late. Suppose, continuing our earlier example, a new good, the CD ROM, appears in 1995 and costs $10.50.1 decide to buy two CD ROMs and therefore reduce my purchases of both clothing and computer supplies by one item each (Table 4). Now I am even better off because I prefer the CD ROMs to the articles they replaced. Note that the new good would not affect the CPI. Since the CPI uses base year (1994) quan tities of each item, and for CD ROMs that quantity is zero, the price of the new good is irrelevant, conceptually. But the BLS in prac tice must confront the problem, because goods stitution Effect in CPI Data, 1982-1991," Monthly Labor Review. These studies cover a longer period and extend over periods where the base year has been changed. They average out the effects of the different base years that have been applied to achieve the long-run index. in the base market basket sometimes disap pear from the marketplace as they succumb to competition from newer products. Moreover, in constructing the PCE deflator or a true costof-living index, the BEA needs to construct a 1994 price for CD ROMs. There are four basic ways to deal with this problem. Ignore it. One could assume no change in price for the good. For many years, the BEA had no statistics on actual prices of mainframe computers and kept the computer price index at l . 10 Since the general price level was rising, this implied that mainframe computer prices were falling relative to the prices of other goods. However, there is no reason to believe that this method provides very accurate an swers. Alternatively, one could use changes in the prices of similar products to measure the price change for new goods, which may make some 10This period was before personal computers were an important share of computer sales. 19 NOVEMBER/DECEMBER1995 BUSINESS REVIEW TABLE 4 Actual Purchases—With New Good Added 1994 1995 Total Total Item Price Quantity Purchase Price Quantity Purchase Clothes $10 10 $100 $12.00 6 $72 Computer $10 10 $100 $9.00 13 $117 CD ROM ? 0 $0 $10.50 2 $21 Total Spending $200 sense if the products are moving similarly in price. However, this is probably not the case as the "similar" products are typically older prod ucts and may well be outmoded. Moreover, the very existence of the new products in creases the standard of living by making pos sible a greater variety of purchases. So ignor ing the problem is highly unsatisfactory. Use sample rotation. A partial solution is to add new items to the price survey as they appear, for example, to begin collecting data on CD ROMs in 1995 and to lump them in with some pre-existing good. For example, CD ROMs could be considered computer sup plies. To keep abreast of new products and stores, the BLS rotates the sample of goods and outlets whose prices it collects. It rotates 20 percent of the sample each year, but it takes two years after new items are identified before they are introduced into the surveys. As a result, the actual goods priced are, on average, about five years out of date. Unfortunately, the BLS's sample rotation procedure itself biases the CPI upward. If an item is selling at a temporarily reduced price when the BLS rotates it into the sample, the quantity sold will be temporarily high, so the item will appear in the CPI basket of goods with a higher quantity weight than justified by its average sales. And items that are at a sale price when rotated into the sample show a 20 $210 high rate of price increase as they return to a more normal price. So sample rotation gives a bigger weight to goods whose prices are likely to rise. This effect alone is estimated to have erroneously raised inflation between 0.2 and 0.3 percent annually over the 15 years since the BLS introduced sample rotation.11 The BLS has recently taken steps to remove this bias from the foods category, where the effect has been quite pronounced. Use product characteristics. When a variant replaces an existing product, if the value of the changes in the product can be given a price, then the new and old products can be com pared. In the auto industry, for example, the resource costs to the manufacturer of optional equipment that has been made standard are used to compare new models with old ones. For example, if a new model has a driver's-side air bag as standard equipment and the old model did not, BLS deducts the wholesale cost of the bag, electronics, and installation, plus a standard markup, before it calculates how much the price of the new model differs from the old. The hedonic method is a somewhat more n Brent R. Moulton, "Basic Components of the CPI: Estimation of Price Changes," Monthly Labor Review, De cember 1993, pp. 13-24. FEDERAL RESERVE BANK OF PHILADELPHIA Measuring Inflation in a High-Tech Age Leonard I. Nakamura Hedonic methods help to capture the re source cost, including the cost of the sales effort, of new characteristics of products. But they do not fully capture the benefits that accrue to consumers from the introduction of new goods. Estimate consumer surplus. When a new product is introduced, consumers receive a welfare gain (called "consum er surplus"), which is the difference between what they would be willing to pay for the product and what they actually have to pay. In our ex ample, suppose that I would have been willing to pay $35 to buy the two CD ROMs, but I only had to pay $21.00. To be explicit, suppose I would be willing to pay $20 for one CD ROM, $15 for a second one, and $10 for a third. At a price of $10.50 each, I will buy two, and the welfare gain I will receive is the same one I would have got if the price had fallen from $17.50 to $10.50. Note that a product need not be wholly new 12See Jack E. Triplett, "The Economic Interpretation of to provide surplus. Consider the opening of a Hedonic Methods," Survey of Current Business, January discount clothing store near my home. Again 1986, pp. 36-40. using the example, suppose the price of cloth 13Product characteristic adjustments are used for hous ing has gone up from $10 to $12 at the store I ing and new autos, two areas in which quality improve patronized in 1994, but the discounter sells the ments are sometimes thought to be overstated rather than same items for $10, so I decide to purchase the understated. But this doesn't mean that the CPI is overstat clothes from the discounter. If I am just as ing quality as a whole, since the upward biases produced happy to shop at the discounter as to go to the by sample rotation and relative price changes are much larger than the downward biases in the measurement of old store, my price of clothes has not risen. The housing and auto prices. Between 1983 and 1988, the BLS arrival of the discounter gives me "surplus" failed to adequately correct for the aging of housing units. compared with shopping at the old store—I This resulted in a downward bias of roughly 0.1 percent would be willing to pay up to $12 for clothes annually and was eliminated in 1988. See William C. and would still switch to the discounter, but I Randolph, "Housing Depreciation and Aging Bias in the Consumer Price Index," Journal of Business and Economic pay only $10. Statistics, July 1988. The other widely cited indication of The assumption implicitly made in the CPI downward bias is the correction for environmental con and in the PCE deflator in this example and trols in new cars. However, this effect has been estimated others is that there is no consumer surplus to be less than 0.05 percent per year. See John F. Peterson, from the introduction of the new product. "Is the Growth of the CPI a Biased Measure of Changes in Cost of Living?" Congressional Budget Office paper, 1994. Under this assumption, there is no measure Thus, over the past 10 or 15 years, the average effect of ment problem. The arrival of a new product, these overstatements of quality must be less than 0.1 per such as a CD ROM, makes no difference to my cent annually. The well-documented errors of underesti w elfare. In the discounter exam ple, the mation of quality due to sample rotation bias and relative discounter's clothes are considered new clothes price bias are roughly five times as large. theoretically satisfying statistical technique that values changes in goods by the implicit retail price of the changed characteristic. This method uses statistical regressions to estimate how much consumers pay for each of a product's specific characteristics or parts (like an air bag), controlling for other characteristics.12 In the computer industry, at least in the 1980s when the BEA introduced this technique, the most important characteristics of computers were processing speed and memory size. So a computer with twice as much processing speed and twice as much memory was considered to be twice as much computer. Since the process ing speed of different computers and their memory sizes could be calculated, at least for some given benchmark set of tasks, then even though whole new generations of computers appeared, their prices could be compared.13 21 BUSINESS REVIEW items, different from the old store's. There is no consumer surplus from the arrival of the discounter— it is assumed that any consumer surplus I would have gained is eaten up in the switch (the old retailer's friendliness and ser vice that have been lost are equal to the full difference in price). An alternative assumption is that shifts in consumer demand in response to changes in price are informative about the amount of consumer surplus consumers receive. If, when the CD ROM is introduced, consumers pur chase them in large amounts and their pur chases thereafter are relatively insensitive to price increases, that is an indication that had the CD ROMs been available earlier, consum ers would have been willing to pay more than required. In other words, consumers are re ceiving a substantial consumer surplus. If con sumers switch to discount outlets in large numbers and are relatively insensitive to price swings at the new outlets, they are receiving a substantial consumer surplus. To measure the price sensitivity of consumers requires statis tical analysis and extensive data. This is the textbook problem of estimating the consumer's demand curve. One such study has been conducted for pharmaceuticals. After Eli Lilly's drug Keflex lost patent protection in 1987, generic versions became available at roughly half the price of the brand-name product. Eli Lilly did not cut the price of Keflex but, indeed, raised it 24 percent over the next three years. Meanwhile, the price for the generic drug fell 30 percent, and generics garnered 83 percent of the market measured in doses. Until recently, in such instances, the BLS treated the generic drug as a separate product that was worth only half as much as the brand-name product (the relative price of the generic at introduction) and would have likely recorded a rising price index for this drug. But many purchasers are probably indifferent to the brand name and would have experienced a pure price decline when the 22 NOVEMBER/DECEMBER1995 generic became available. A study by Franklin Fisher and Zvi Griliches argues that the most reasonable measure of the consumer surplus realized by purchasers who switched would lead to a price index that declines 48 percent, compared with the BLS approach, which would show a price increase of 14 percent. This is a 19 percent annual rate difference over the 45 months covered in the study.14Infanuary 1995, the BLS changed its procedure to reflect a price decline when a therapeutically equivalent ge neric version of a drug replaces a brand-name patent drug. This change should substantially reduce this problem of pricing drugs. Although there is some arbitrariness to esti mates of the value of consumer surplus from the introduction of new goods and services, they are surely an improvement over assum ing that new product introductions result in no consumer surplus.15* Moreover, private data companies, as well as many retailers, regularly 14Franklin M. Fisher and Zvi Griliches, "Aggregate Price Indices, New Goods, and Generics," Quarterly Jour nal o f Economics 110, February 1995, pp. 229-44. They make the assumption that consumers vary in their relative valu ations of the generic product, but that those who switch are evenly spread over the spectrum of possible types— from those for whom the entire difference in price is surplus, to those who receive no surplus. This and other studies on prescription drugs have been done at the wholesale level where the data are relatively easy to come by. In summa rizing them, F.M. Scherer has estimated conservatively that the BLS's treatment of the introduction of generics implies an upward bias of 1.2 percent a year in the Pro ducer Price Index for prescription drugs. F.M. Scherer, Journal of Economic Perspectives 7, Summer 1993, pp. 97-115. 15When goods and services and outlets disappear from the marketplace, consumer surplus is lost. But, in general, this consumer surplus is likely to be small compared with the consumer surplus of new product introductions be cause consumers will not choose to switch from high consumer-surplus goods to low ones, and because suppli ers of high consumer-surplus goods can raise their prices and still retain demand, and so are unlikely to go out of business. FEDERAL RESERVE BANK OF PHILADELPHIA Measuring Inflation in a High-Tech Age collect the data necessary for these estimates. A study by Jerry Hausman examines ready-toeat breakfast cereals, specifically Apple-Cin namon Cheerios, using supermarket data.16 Hausman found that accounting for new ce real introductions, as measured in a true costof-living index, might have reduced the CPI for ready-to-eat breakfast cereals by 20 percent over the 10-year period 1980 to 1990, or about 2 percent a year.17 IMPROVEMENTS IN QUALITY PLAY IMPORTANT ROLE IN ECONOMY What is driving these rapid product and service innovations? Dramatic improvements in high-tech investment goods are an impor tant force in the availability of new goods and services. Investment in high-tech goods ac counts for more than 4 percent of the economic output of the United States. For many of these goods, substantial price deflation is normal. For example, between 1951, when the first com puter was designed, and 1984 the cost of com puters— adjusting for changes in quality—fell an estimated 1000-fold.18 Since 1984 computer prices have continued to fall dramatically. Large price declines appear in other recent high-tech investment goods, such as telecom munications equipment and medical equip ment. While there have always been some goods whose prices are declining, the declines haven't been as large. In the early part of this century and over roughly the same period as that for computers, the price of automobiles 16Jerry A. Hausman, "Valuation of New Goods Under Perfect and Imperfect Competition," NBER Working Pa per 4970, December 1994. 17This measure depends on an assumption of a "repre sentative consumer," i.e., that all consumers are alike. 18Robert J. Gordon, "The Postwar Evolution of Com puter Prices," in Dale G. Jorgenson and Ralph Laundau, eds., Technology and Capital Formation (MIT, 1990). Leonard I. Nakamura declined substantially, but only about tenfold.19 High-tech investment goods are primarily purchased by businesses rather than consum ers; their effect on consumer prices is indirect and often is not captured statistically. For example, buying automatic teller machines may improve banking services, but the impact is difficult to measure and, in practice, is not captured. Similarly, as new medical equip ment is introduced, its effect on the quality of diagnosis and treatment is typically not in cluded in the pricing of medical services. High-Tech Entertainment Goods and Per sonal Computers. As measured in the CPI, PC prices fell about 10 percent in 1994, about the same rate of decline for the whole period 1987 to 1994. This appears to underestimate the rate of decline: computer magazine advertisements show a much sharper rate of decline: approxi mately 24 percent over the course of 1994.20 A recent study by Ernst Berndt and Zvi Griliches shows a decline of 30 percent annually for 1982-88.21 The likely reason for this discrep ancy is that the CPI tends to underestimate the price decline for personal computers because the sample being surveyed is too old.22* 19Daniel M. G. Raff and Manuel Trajtenberg, "QualityAdjusted Prices for the American Automobile Industry: 1906-1940," NBER Working Paper 5035, February 1995. 20I collected data from advertisements of Dell and Gateway, the two largest mail-order PC firms. Price level data indicate a decline of $570 on an average price of $2380, or almost 24 percent. Complete data for the study can be found on the Federal Reserve Bank of Philadelphia In te rn e t site at h ttp :/ / lib e rty n e t.o rg / ~ fe d re srv / fedpage.html. 21Ernst Berndt and Zvi Griliches, "Price Indexes for Microcomputers: An Exploratory Study," in Murray F. Foss, Marilyn E. Manser, and Allan H. Young, eds., Price Measurements and Their Uses, NBER Studies in Income and Wealth No. 57 (University of Chicago, 1993). 22The introduction of the P6 Intel processor in 1995 marks the sixth generation of PC microprocessors since the 23 BUSINESS REVIEW Similar measurement problems exist for a broad spectrum of products that have elec tronic components embedded in them. A readily identifiable segment in personal con sumption expenditures includes video and audio products, computing equipment, and musical instruments; this segment accounted for 1.6 percent of consumer purchases in 1994. Electronic parts play a large role in a wide variety of other consumer products. For ex ample, the value of the electronic components of new cars is approximately the same as that of the steel in them— and improvements in the quality of these components are not well cap tured in the CPI for new cars. Medical Goods and Services. The develop ment of new drugs and diagnostic and treat ment methods make medical care an area where quality change has been important. Studies of prescription drugs suggest that annual infla tion in this product group may be overstated 6 or 7 percent because the BLS does not account for quality improvements. Although such drugs account for only about 1 percent of consumer expenditures, the net effect may be close to 0.1 percent on the overall index.23 And these drugs are less than one-fifth of the por tion of consumer expenditures accounted for by medical costs. One interesting study of CT scanners shows that the average price rose about 160 percent between 1973 and 1982. However, correcting for quality, using characteristics like resolu tion and speed, produces a price decline of 72 percent. Finally, a third method that includes consumer surplus shows a 1000-fold price de crease!24 first IBM PC in 1979, or a new generation every three years. It is obviously extremely difficult to keep up with this rate of product introduction using conventional methods of price data collection. 23Zvi Griliches and Iain Cockburn, "Generic and New Goods in Pharmaceutical Price Indexes," Harvard Discus sion Paper 1664, Cambridge, MA 1993. 24 NOVEMBER/DECEMBER 1995 Technological advance in medical care has been profound. Yet the benefits to the con sumer of such advances are not included in our price statistics. Medical care is a large propor tion of the consumer budget, accounting for roughly 7 percent of the CPI.25 If we extrapo late the 6 percent quality improvement in drugs to all of medicine, this one source can reduce the inflation rate, as measured by the CPI, 0.4 percentage points annually. HOW BAD ARE THESE PRICING PROBLEMS? The best estimates for the impact of the first problem we discussed, relative price move ments, show the CPI about 0.2 percent too high annually. The upward bias from sample rota tion has been reliably estimated at 0.2 to 0.3 percentage points annually. For quality gains whose costs go unmeasured there is consider able uncertainty, but they are likely to account for more than half a percentage point.26 Thus the CPI may well be overstated 1 percent annu ally even if we take no account of the consumer surplus associated with the introduction of new goods and services. Studies of CT scan ners and of generic pharmaceuticals show that consumer surplus in high-tech goods may be very large. And Hausman's study of breakfast cereals suggests that non-high-tech goods may also have very substantial consumer surpluses. 24Manuel Trajtenberg, Economic Analysis of Product In novation: The Case ofC T Scanners (Harvard, 1990). 25It has been argued that its proportion of the consumer budget is actually twice as large because the 7 percent figure does not take into consideration government and corporate subsidies of these payments. 26The extrapolations discussed in the text for medicine and high-tech consumer goods alone would amount to 0.6 percentage points annually on the CPI, and these goods account for less than one-twelfth of consumer expendi tures. FEDERAL RESERVE BANK OF PHILADELPHIA Leonard I. Nakamura Measuring Inflation in a High-Tech Age HOW CAN WE DO BETTER? Right now, most price data are obtained by sending BLS price collectors on their rounds of retail establishments to fill out pricing forms. This seems unnecessary. Improvements in com puter and telecommunication facilities mean that many retailers, wholesalers, and manu facturers routinely capture detailed item-by item transactions data as part of their billing systems. If the BLS could tap this information, collecting price data might be less expensive and more accurate than current methods. A pilot project to use scanner codes (universal product codes) as a basis for price-data collec tion is under way at the BLS.27 Moreover, broad collection of data of this type would facilitate further studies on demand curves that could help measure the impact of in creases in variety and quality on welfare. Another question arises in regard to data collection: Should a private firm collect the data for and publish the C onsum er Price In dex? Having private firms collect price data might enhance efficiency, but there are a num ber of problems. One is that a private data collector might not be able to protect the pri vacy of voluntary data providers, and this might discourage such providers. Another problem is that the private data collector might be tempted to use the data for private gain. If the private data collector were to pick up indications that the price index was going to be 27Marshall Reinsdorf, "Constructing Basic Component Indexes for the U.S. CPI from Scanner Data: A Test Using Data on Coffee," U.S. Bureau of Labor Statistics, mimeo. surprisingly high, it might be tempted to use this information to make speculative gains in interest rate markets. Thus, it would appear unlikely that purely private data collection would enhance price measurement. Public-private cooperation would probably be essential to an improved effort to collect statistics. In the past, private corporations have been very useful to the BLS and the BEA in assisting them in improving statistics. For ex ample, special arrangements with automobile manufacturers help provide more accurate sta tistics on new car and truck sales. And a team of economists from IBM was very helpful in setting up the hedonic price regressions for computers. Large firms may have the resources to assist in such an effort, and the incentive to do so, since sound government policy-making and improved public perceptions of economic reality are likely to be in their best interest. CONCLUSION U.S. economic performance is most prob ably better than has been reported in our offi cial statistics. Documented biases and extrapo lations from other studies suggest that the CPI may distort our picture of recent U.S. eco nomic history. A true cost-of-living index would show prices rising more slowly than reported in the CPI; most likely, inflation would be at least one percentage point per year lower. Improving data collection and providing firmer economic foundations for the price measures themselves are important elements in improv ing economic decision-making and contract ing in the United States. 25 Philadelphia /RESEARCH Working Papers The Philadelphia Fed's Research Department occasionally publishes working papers based on the current research of staff economists. These papers, dealing with virtually all areas within economics and finance, are intended for the professional researcher. The papers added to the Working Papers series thus far this year are listed below. To order copies, please send the num ber of the item desired, along w ith your address, to WORKING PAPERS, D epartm ent of Research, Federal Reserve Bank of Philadelphia, 10 Independence Mall, Philadelphia, PA 19106. For overseas airmail requests only, a $3.00 per copy prepayment is required; please make checks or money orders payable (in U.S. funds) to the Federal Reserve Bank of Philadelphia. A list of all available papers may be ordered from the same address. 95-1 Satyajit Chatterjee and Dean Corbae, "Valuation Equilibria with Transactions Costs" 95-2/R Sherrill Shaffer, "Structural Screens in Stochastic Markets" (supersedes Working Paper No. 9223) 95-3 Franklin Allen and Douglas Gale, "A Welfare Comparison of Intermediaries and Financial Markets in Germany and the U.S." 95-4 Franklin Allen and Douglas Gale, "Financial Markets, Intermediaries, and Intertemporal Smoothing" 95-5 Gregory P. Hopper, "The Dynamics of the Exchange Rate Under a Crawling Peg Regime: A Game Theory Approach" 95-6 Franklin Allen and Douglas Gale, "Universal Banking, Intertemporal Risk Smoothing, and European Financial Integration" 95-7 Paul Calem and Michael Stutzer, "The Simple Analytics of Observed Discrimination in Credit Markets" 95-8 Joseph Hughes, William Lang, Loretta Mester, and Choon-Geol Moon, "Recovering Technolo gies That Account for Generalized Managerial Preferences: An Application to Non-RiskNeutral Banks" 95-9 Ana Castaneda, Javier Diaz-Gimenez, and Jose-Victor Rios-Rull, "Unemployment Spells and Income Distribution Dynamics" 95-10 Paul Calem and Loretta J. Mester, "Consumer Behavior and the Stickiness of Credit Card Interest Rates" (Supersedes No. 92-24/R) 26 http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis FEDERAL RESERVE BANK OF PHILADELPHIA 95-11 Richard Voith, "Parking, Transit, and Employment in a CBD" 95-12 Gary Gorton and Richard Rosen, "Banks and Derivatives" 95-13 Sherrill Shaffer, "Translog Bias Under Declining Average Costs" 95-14 Alberto Trejos and Randall Wright, "Toward a Theory of International Currency: A Step Further" 95-15 Gerald Carlino and Robert DeFina, "The Differential Effects of Monetary Policy Shocks on Regional Economic Activity" 95-16 Paul Calem, "Mortgage Credit Availability in Low- and Moderate-Income Minority Neighbor hoods: Are Information Externalities Critical?" 95-17 Leonard I. Nakamura, "New Directions in Information and Screening in Real Estate Finance" 95-18 William J. Stull, "Is High School Economically Relevant for Noncollege Youth?" 95-19 James McAndrews and George Wasilyew, "Simulations of Failure in a Payment System" 95-20 Keith Sill, "An Empirical Investigation of Money Demand: Evidence from a Cash-In-Advance Model" 95-21 Leonard Nakamura, "Is U.S. Economic Performance Really That Bad?" 27 1995 INDEX January/February Dean Croushore and Tom Stark, "Evaluating McCallum's Rule for Monetary Policy" Gerald A. Carlino, "Do Education and Training Lead to Faster Growth in Cities?" March/April Rafael Rob, "Entry and Exit of Firms and the Turnover of Jobs in U.S. Manufacturing" Theodore M. Crone, "Making Money in the Housing Market: Is There a Sure-Fire System?" May/June Gregory P. Hopper, "A Primer on Currency Derivatives" Sherrill Shaffer, "Rethinking Disclosure Requirements" July/August Loretta J. Mester, "There's More Than One Way to Sell a Security: The Treasury's Auction Experiment" Robert P. Inman, "Do You Know How Much Money Is in Your Public Purse?" September/October James McAndrews, "Antitrust Issues in Payment Systems: Bottlenecks, Access, and Essential Facilities" Satyajit Chatterjee, "Productivity Growth and the American Business Cycle" November/December Martin A. Asher and Robert H. DeFina, "Has Deunionization Led to Higher Earnings Inequality?" Leonard I. Nakamura, "Measuring Inflation in a High-Tech Age" FEDERAL RESERVE BANK OF PHILADELPHIA Business Review Ten Independence Mall, Philadelphia, PA 19106-1574 Address Correction Requested