The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of Chicago Third Quarter 2002 persp 2 ives Unprepared for boom or bust: Understanding the current state fiscal crisis 26 The 2001 recession and the Chicago Fed National Activity Index: Identifying business cycle turning points 44 Why do we use so many checks? 60 Analyzing the relationship between health insurance, health costs, and health care utilization Economic . perspectives President Michael H. Moskow Senior Vice President and Director of Research William C. Hunter Research Department Financial Studies Douglas Evanoff, Vice President Macroeconomic Policy Charles Evans, Vice President Microeconomic Policy Daniel Sullivan, Vice President Regional Programs William A, Testa, Vice President Economics Editor David Marshall Editor Helen O’D. Koshy Associate Editor Kathryn Moran Production Julia Baker, Rita Molloy, Yvonne Peeples, Nancy Wellman Economic Perspectives is published by the Research Department of the Federal Reserve Bank of Chicago. The views expressed are the authors’ and do not necessarily reflect the views of the Federal Reserve Bank of Chicago or the Federal Reserve System. Single-copy subscriptions are available free of charge. Please send requests for single- and multiple-copy subscriptions, back issues, and address changes to the Public Information Center, Federal Reserve Bank of Chicago, P.O. Box 834, Chicago, Illinois 60690-0834, telephone 312-322-5111 or fax 312-322-5515. Economic Perspectives and other Bank publications are available on the World Wide Web at http:Avww.chicagofed.org. Articles may be reprinted provided the source is credited and the Public Information Center is sent a copy of the published material. Citations should include the following information: author, year, title of article, Federal Reserve Bank of Chicago, Economic Perspectives, quarter, and page numbers. chicagofed. org ISSN 0164-0682 Contents Third Quarter 2002, Volume XXVI, Issue 3 2 Unprepared for boom or bust: Understanding the current state fiscal crisis Leslie McGranahan The headlines concerning state government finances have become increasingly alarming since mid-2001. This article discusses the roots of the current state fiscal crisis by looking at the decisions made by state government leaders during the long expansion. The author suggests increased use of rainy day funds as a way to avoid future crises. 26 The 2001 recession and the Chicago Fed National Activity Index: Identifying business cycle turning points Charles L. Evans, Chin Te Liu, and Genevieve Pham-Kanter The initial release of the Chicago Fed National Activity Index (CFNAI) in early 2001 pointed to the very real possibility that the U.S. economy was teetering on the brink of recession. This article quantifies the statistical ability of the CFNAI to act as an early warning indicator of economic recessions. In simulation experiments, the CFNAI performed virtually as well as the statistical model’s ideal measure of the business cycle. 44 Why do we use so many checks? Sujit Chakravorti and Timothy McHugh The authors identify underlying disincentives for payment system participants to migrate to electronic payments. Their analysis sheds light on why check usage remains higher in the United States relative to other industrialized countries when the real resource cost of processing payments may decrease by using electronic payment networks. 60 Analyzing the relationship between health insurance, health costs, and health care utilization Eric French and Kirti Kamboj Using data from the Health and Retirement Survey and the Assets and Health Dynamics among the Oldest Old, this article provides an empirical analysis of the determinants of whether an individual purchases health insurance. The authors describe the relationship between health costs and health care utilization of individuals aged 50 and older and explore how these factors vary with access to health insurance. Unprepared for boom or bust: Understanding the current state fiscal crisis Leslie McGranahan In October 2001, the state governors sent a letter to the U.S. Senate concerning the Senate’s proposed stimu lus package. The governors sought to prevent the pas sage of a package that would be detrimental to already weak state budgets and to ask for specific assistance from the federal government for state budget items. As a result of the connection between federal and state revenues and spending, state leaders often comment on federal changes. What is remarkable about this let ter is that only seven months after the end of the great est post-war economic boom, the states were already seeking fiscal help from the federal government. In addition to its concerns about mounting defense and intelligence obligations and the flailing macroecono my, the government now faced the specter of service shutdowns by bankrupt state governments. The crisis facing the state governments emerged quickly. In August 2000, commentators at the National Conference of State Legislatures (NCSL) were boast ing that states were “in their best financial conditions in decades” (NCSL, 2000). In January 2001, the NCSL asserted that the states remained in “excellent fiscal condition” (NCSL, 2001c). But by August 2001, the NCSL was detailing how states were coping with bud getary shortfalls (NCSL, 2001a). In this article, I ask how the states found themselves in fiscal trouble so quickly. I begin by discussing the excellent revenue news from the states throughout the economic expansion. Tax revenues increased, welfare reform kept block grants at high levels, and the tobac co settlement provided a generous new form of funds. As a result, states faced the pleasant dilemma of what to do with their windfall revenues. I investigate the ways states decided to use these revenues. They faced four fundamental choices: they could spend the money on high-priority programs; they could return the money to taxpayers in the form of rebates and reductions; they could reduce indebtedness; 2 or they could save the money for a less brilliant fu ture. All states chose a combination of these four. States increased spending. Much of the spending increase was due to mounting expenditure pressures in health-care related areas. States also aggressively cut taxes, particularly personal income taxes, throughout the expansion. While the states did not move to reduce indebtedness, they did increase their savings. States save money by maintaining balances in their reserve funds. Most states have created budget stabilization or “rainy day” funds as a way to cope with unexpected shortfalls. (The only states without such funds as of October 2001 were Arkansas, Montana, and Oregon). Money transferred into these funds can be withdrawn under specific circumstances. States also maintain re serve balances in their general fund accounts. During the expansion, balance increases in these accounts were substantial but were insufficient to offset even a mild downturn. When state revenues began to deteriorate in the third quarter of 2000 as the first signs of the pending recession surfaced, state budgets soon ran into deficit. Because nearly all states are required to pass balanced budgets and are limited in their ability to issue debt, they needed to deal with the budget imbalance quick ly. State debt limits restrict states’ ability to borrow if changes in economic circumstances lead to shortfalls during the fiscal year (National Association of State Budget Officers [NASBO], 2002).1 Now, instead of the four pleasant choices outlined earlier, states faced four difficult options to deal with these revenue shortfalls. Because of the restrictions Leslie McGranahan is a consultant to the Federal Reserve Bank of Chicago and an associatefellow in the Department ofEconomics at the University of Warwick. The author would like to thank Bill Testa, David Marshall, and Helen Koshy for comments and guidance. 3Q/2002, Economic Perspectives on debt, debt issuance was not one of the options. States could increase taxes; they could cut spending; they could reduce the balances available in their reserve funds; or they could rely on the federal government to bail them out. Few states have chosen to increase taxes. Tax increases are both politically very unpopular and slow and difficult to shepherd through legislatures. Most states have used a combination of spending cuts and reserve fund withdrawals to bring their budgets into balance. While the states have asked the federal gov ernment for assistance, federal aid has not been par ticularly forthcoming. The current fiscal crisis highlights the problems inherent in the states’ balanced budget system. States cut taxes and increase expenditure during booms only to be faced with revenue shortfalls during recessions. Then, the states have to cut spending just when the need for government services becomes most pronounced and must raise taxes when taxpayers are at their poor est. To prepare for future downturns, state governments should consider some policy changes. First, states should control spending during both expansions and recessions in order to avoid the need for dramatic cuts during difficult times. Second, states should restruc ture their rainy day funds, so that they can draw on these more heavily to maintain services during diffi cult times. In order to do this, reserve fund balances would need to grow much larger than they did in the recent expansion. I highlight the experience of the five states that make up the Federal Reserve’s Seventh District— Illinois, Indiana, Iowa, Michigan, and Wisconsin. This allows me to paint a more precise picture than would come from only generalizing across 50 states. These midwestern states are interesting because they were among the first to be harmed by the economic slow down. As a result, they were forced to make difficult decisions earlier than other states. At the same time, the behavior of the midwestern states has been fairly typical of that of states nationwide. The boom: Revenues The U.S. Census Bureau segregates state funds into four separate categories—the general fund, insurance trust funds, utility funds, and funds for state-operated liquor stores. In this article, I focus on the general fund as it is the source of revenues and expenditures over which the state has the most control and it supports the largest state government expenditures. The other funds are very small, with the exception of the insur ance trust fund. This fund supports unemployment insurance, workers’ compensation, and programs for state government employees.2 Federal Reserve Bank of Chicago States generate revenues from a variety of sources, the two most important being taxes and the federal government. In 1999, the federal government provid ed just over 25 percent of general fund revenues while taxes provided 55 percent. In 1999, most state taxes came from general and selective sales taxes (48 per cent), personal income taxes (35 percent) and corpo rate income taxes (6 percent). Over time, general state revenues have been in creasing along with the rise in national income. Be tween 1980 and 1992 (the first year of positive economic growth during the recent expansion), real general rev enues increased by an average of 4 percent per year in total and 3 percent per year per capita. Throughout the recent expansion, strong national economic con ditions translated into continued strong state revenue performance. Between 1992 and 1999, state revenues grew an average of 4 percent per year and 3 percent per capita, despite significant enacted tax reductions (U.S. Department of Commerce, Bureau of the Cen sus, 1981 and 2002a).3 The reason for the continued revenue growth was that everything was going right. Robust consumer spending translated into high sales tax revenues. Sales tax revenues are procyclical both because spending is itself procyclical and because states exempt the least cyclically sensitive products from taxes—in particu lar, food and drugs. Between 1992 and 2000, real to tal state general sales tax revenues increased by 40 percent, or by an average of 4.2 percent per year. Rev enues from the even more procyclical personal income tax also increased dramatically during the expansion, by 59 percent in real terms between 1992 and 2000, or by 6 percent per year (U.S. Department of Com merce, Bureau of the Census, 2002b). One reason for the increase in income tax revenues was the high lev el of employment and earnings. But likely even more important was the dramatic increase in revenue from taxes on capital gains and dividends. The exact role of the growth of capital gains and dividends in boosting the revenue performance of the states is difficult to ascertain because data on income tax revenues from different sources are not available for most states.4 However, two sources point to a sig nificant increase in revenues derived from capital gains taxes. First, state capital gains taxes are closely linked to the federal capital gains tax and readily available data on capital gains and dividends declared on fed eral individual income tax returns show a dramatic hike over the 1990s, especially post 1994. The growth in capital gains and dividends reported on federal in come tax returns is pictured in figure 1. Second, we do have separate data on state personal income tax 3 withholding, estimated payments, and final settlements paid when taxes are filed. Trends in estimated payments give some indication of the level of capital gains and dividends received, because these are taxes paid on non-wage income. While estimated payments are high ly volatile, they did increase dramatically at times during the expansion. For example, estimated payments for 2000 taxes made between April 2000 and Febru ary 2001 were 17.1 percent higher than similar pay ments made the previous year (Jenny and Boyd, 2001). Again, in parallel with the experience of the federal government, the states’ personal income tax revenues exceeded expectations every year during the expansion, probably due to the high level of realized capital gains. Two less obvious factors also contributed to the impressive state revenue performance of the end of the millennium. First, 46 states and four major tobacco companies signed the Master Settlement Agreement in November 1998. To settle state lawsuits aimed at recovering tobacco-related Medicaid costs, the tobacco companies promised the states $206 billion over a 25-year period. The states began receiving money in late November 1999 following the approval of the agree ment by the required number of states. During 1998, 2000, and 2001, states received $2.4 billion, $6.4 bil lion, and $6.9 billion, respectively, from the tobacco settlement.5 Some states received even greater revenues than indicated by the settlement, because they used financial intermediaries to trade the 25-year stream of benefits for a single lump sum. Wisconsin, for example, arranged for a single payment. These tobacco monies are large, even in the context of multibillion-dollar 4 state budgets. The $8.3 billion due to the states in 2002 is equivalent to 1.8 percent of state general fund rev enues in 2002 recommended budgets. For 2000, with out funds from the tobacco settlement, revenue growth would have equaled 3.7 percent; including the settle ment raised the growth rate to 4.5 percent (Wilson, 1999; NCSL, 1999a). While anti-smoking groups anticipated that these funds would be spent on state smoking cessation ini tiatives and other health causes, the funds entered state coffers with no strings attached. While some of these funds have been spent to curb smoking, most have simply served to increase revenues and have not been earmarked for specific causes. The second factor, aside from taxes, contributing to state revenues during the expansion was the change in the welfare program. When Aid to Families with Dependent Children (AFDC) changed to Temporary Assistance to Needy Families (TANF) in 1996 (dis cussed in more detail below), welfare funding changed from a federal matching program to a fixed federal block grant. Under the AFDC program, the declines in caseloads that accompanied the programmatic change and economic expansion would have led to a decline in spending and, therefore, a decline in the federal match. By contrast, under the new program, block grants stayed fixed in the face of declines in the re cipient population. As a result, states could both cut their own spending down to the levels required by the legislation and use their funds to increase benefits and restructure programs to support a wide array of social services for their welfare populations. Spending Like revenues, state government spending has generally been increasing over time. Between 1980 and 1992, real general government expenditure in creased by 4.4 percent overall per year and by 3.4 percent per capita. During the expansion, between 1992 and 1999, real expenditure growth slowed to 3.5 percent per year, or 2.5 percent per capita. Cen sus data on state government spending are only avail able until 1999 (U.S. Department of Commerce, Bureau of the Census, 2002a). More recent data from NASBO show that real total state expenditure increased by 5 percent between 1999 and 2000 and 5 percent between 2000 and 2001 (NASBO, 2001c).6 Even in the presence of impressive revenue growth, by 1999 appropriations growth was expected to outpace revenue growth. In the light of the fiscal problems emerging in fiscal 2002, governors recommended that appropria tions growth slow substantially. 3Q/2002, Economic Perspectives State government spending is less cyclical than period or from just under 3 percent to just over 4 per revenue, because many of the major state services cent per year. The table also shows spending growth are not particularly cyclically sensitive. For example, in current and capital spending. Current expenditures enrollment in elementary and secondary education is grew more quickly than capital outlays. While funding increases during the expansion were a function of past fertility decisions and is not very responsive to the condition of the economy. pretty universal, two areas deserve special attention: However, spending in some programmatic areas is education and Medicaid, which is classified by the sensitive to economic conditions. The most obviously Census Bureau as part of public welfare. cyclically sensitive area is need-based services. De Trends in Medicaid spending mand for these services declines as the economy im The almost universally acknowledged source proves and employment rates increase. Then, demand of the states’ most significant spending woes is the grows in a downturn. States partially fund three cru Medicaid program. Medicaid is the health insurance cial need-based programs: unemployment insurance, program for low-income people. The program covered Medicaid (the health insurance program for the lowover 40 million recipients in 1998. The states and fed income population), and welfare (TANF, formerly eral government split Medicaid expenditures, with the AFDC). Among these, unemployment insurance is federal government picking up between 50 percent and covered by funds not considered as general expendi 76.8 percent of the program’s costs. The federal share ture in the Census Bureau definitions. decreases as state per capita income increases. Between That said, some expenditure pressures are even fiscal 1992 and fiscal 2001, real total Medicaid pro slightly procyclical. This minor procyclicality derives gram costs are estimated to have increased from $135 from the fact that states are major employers and com billion to $209 billion 1999 dollars or by 56 percent pete for their employees in the labor market. Labor in total and 5.1 percent per year (which actually rep market tightness should lead to increased wage de resents a decline from the average annual rate of growth mand among teachers, highway workers, police of ficers, and others employed by the state. Looking at teachers, for example, we see TABLE 1 that real salaries rose during the expan Changes in expenditure, 1992-99 sion. Between the 1991-92 and 19992000 academic years, average nominal 1992 1999 Percent expenditure expenditure change teacher salaries rose by 23 percent and (1999 dollars, OOOs) (OOOs) starting teacher salaries by 28 percent (American Federation of Teachers, 2002). Education 240,790,734 318,601,796 32 Public welfare 177,170,480 221,166,721 25 This second number is a better indication Highways 55,787,475 68,317,477 22 of labor market tightness because educa All other 224,402,866 281,389,231 25 tion systems compete against other em All categories, ployers for new college graduates. Between current operations 368,391,413 476,968,246 29 1992 and 2000, total inflation was about All categories, 17 percent (this is the increase in prices as capital outlays 57,117,136 68,508,917 20 measured by the gross national product Source: U.S. Department of Commerce, Bureau of the Census, 2002a. price index) (Executive Office of the Presi dent, Council of Economic Advisers, 2002). In their role as employers, states were between 1980 and 1992 of 9.1 percent per year). Over faced with increased spending pressures during the expansion, while in their role as providers of services the same period, state Medicaid program costs are esti to the needy, they faced declining pressures. On bal mated to have grown by a similar percentage (U.S. Con ance, the expansion probably cut expenditure pres gress, House Committee on Ways and Means, 2002). sures somewhat, but by no means dramatically. Figure 2 depicts the growth in total nominal To further investigate increases in expenditures, Medicaid program costs from 1992 to 2001 (costs I look at increases by spending category. Table 1 shows for 1999-2001 are estimates), compared with the over the real dollar and percentage change in expenditure all growth in the Consumer Price Index (CPI) and the in four major state spending categories between 1992 growth in the Consumer Price Index for Medical Care. and 1999. Spending increased between 22 percent and The figure shows that medical care expenses were 32 percent in all of these categories over the entire growing much more rapidly than the overall price Federal Reserve Bank of Chicago 5 level as reflected in the CPI, and Medicaid expenditures were growing dramatically more rapidly than medical care expenses. In other words, while some of the in crease in Medicaid expenditures can be attributed to an overall increase in health care costs, most of the increase needs to be explained by other factors. Medicaid ex penditures have also been increasing rapidly relative to state government expenditures more generally, as illustrated in figure 3, which depicts the growth in to tal state Medicaid program costs relative to the growth in total state expenditure from 1992 to 1999. We can break down the increase in costs into two component parts—first, the increase in the number of program recipients and, second, the increase in the cost per recipi ent. Figure 4 shows three comparisons between 1992 and 1998: the growth in number of recipients by eligibility cate gory (panel A), the growth in per capita Medicaid costs by eligibility category (panel B), and the growth in total expen ditures by eligibility category (panel C).7 Spending increases by eligibility category have been fairly similar—between 1992 and 1998, the percentage of expenditures represented by each eligibility category has been nearly constant. But the reasons underlying these similar growth rates in spending have differed somewhat. For aged recipients, an increase in costs com bined with a relatively flat recipient pop ulation led to increased total spending. For disabled recipients, both costs and 6 the recipient population grew. For children and adults, a dramatic increase in the recipient population and nearly constant costs per recipient underlie the growth in total spending. The increase in the number of re cipient children and adults derives from legislated extensions of coverage to children and parents of poor families not receiving public assistance.8 As this discussion shows, it is difficult to attribute the increase in Medicaid spending to one single force as both eligibility and costs have increased. That said, much of the debate on Medicaid program costs has naturally focused on the aged and disabled groups. Although these groups represent less than 30 percent of all recipients, they account for over 70 percent of program costs. Two particular areas of spending have received special attention: nursing facilities and pre scription drugs. Nursing facilities accounted for 22.4 percent of total Medicaid payments in 1998 and 62.9 percent of the costs for aged recipients. It is the single largest programmatic spending category. In fact, Medicaid pays 46 percent of all U.S. nursing home expenditures (U.S. Department of Health and Human Services, Health Care Financing Administration, 2000). Prescription drugs represented 9.5 percent of all Medicaid program costs in 1998 and spending for drugs has been increasing rapidly. In 1992, drugs were only 7.4 percent of program costs. These increases in drug expenditures are attributed to a nationwide in crease in drug prices and the advent of a number of new (hence, expensive) drugs. Medicaid drug expen diture increased by an additional 17.9 percent in 1999, 22.2 percent in 2000, and is estimated to increase by 3Q/2002, Economic Perspectives the size of the medically dependent population. However, this relationship should not be overstated. When I per formed simple regressions of the recipient population between 1972 and 1998 on the civilian unemployment rate, a time trend, and dummies controlling for legis lated changes in eligibility in 1990 and 1996,1 only found a statistically significant relationship between the number of adult recipients and the unemployment rate. For all other recipient categories, there is no dis cernible relationship between the unemployment rate and the size of the recipient population. Because spend ing on adults is such a small part of total Medicaid expenditure, this cyclical factor is not a huge part of the Medicaid spending story. Even though Medicaid spending is not very cy clical, worries about costs tend to be most common when the economy is weakest. This is because state budgetary problems become more acute during down turns and Medicaid is such a significant spending area. This is no exception during the current economic cli mate. I discuss potential cost saving measures for Medicaid in a later section. an additional 19.7 percent and 14.9 percent in 2001 and 2002, respectively (NASBO, 2001a). In addition to this general upward trend, Medicaid spending does have a cyclical component as well, because the loss ofjobs and health insurance increases Federal Reserve Bank of Chicago Trends in education spending Education spending, specifically spending for elementary and secondary education, represents the single greatest expenditure category for state govern ments. About $0.36 out of every $1.00 in general ex penditure is spent on education. Between 1992 and 1999, state education spending increased by 32 percent. States cover just under half of total education expen diture, with local governments funding 40 percent to 45 percent and the federal government paying the re mainder (U.S. Department of Commerce, Bureau of the Census, 2002c; U.S. Department of Education, National Center for Education Statistics, 2001). The growth in school expenditure results from a number of sources. First, there was an increase in the number of pupils in elementary and secondary schools. Between 1992 and 1999, the number of pupils increased by 9.4 percent. However, as this percentage increase is less than one-third of the percentage increase in ex penditure, other forces are needed to explain the total increase in costs. Second, over the same period, the num ber of teachers increased by 18.2 percent. The result ant increase in the teacher-pupil ratio represents the continuation of a long-standing trend in education. Costs for instruction (teachers and textbooks) represent 53 percent of total education expenditure, so the growth in teachers, combined with the increase in teacher sala ries discussed earlier, goes a long way toward explain ing the increase in total expenditures.9 Third, during the boom, states increased spending on school-related 7 capital projects. Capital expenditure jumped from 7.6 percent of school expenditures in 1990 to 9.9 percent in both 1999 and 2000 (U.S. Department of Commerce, Bureau of the Census, 2002c). This increase in capital spending was needed to help shore up deteriorating school buildings and assure compli ance with federal mandates regarding accessibility and health hazards (U.S. Congress, General Account ing Office, 1995). Despite this increase in capital spending, school buildings continue to be in poor shape, with 50 percent reporting at least one inade quate building feature as of 1999 (U.S. Department of Commerce, Bureau of the Census, 2002c). Finally, although precise national statistics on special educa tion spending are not available, there is a general consensus that increases in the number of students with diagnosed disabilities have challenged the re sources of school districts. Between the 1992-93 and 1998-99 school years, the percentage of students with a disability increased from 11.8 percent to 13.0 percent. In 1976-77, only 8.3 percent of students were diagnosed with a disability. School expenditure holds a privileged position in the debate over state expenditure. While it is the larg est single expenditure area, it is also very politically popular and somewhat sacred. Discussions of school spending on the state level often take place outside the general budget debate and it is the area most frequently exempted from across-the-board budget cuts. Overall spending growth is driven by a number of factors, but the two largest programmatic areas— Medicaid and education—go a long way toward ex plaining the overall condition of state budgets. Spending in both of these areas increased throughout the expan sion. And as budgets tighten, much of the debate in evitably focuses on these two areas. One other feature of state expenditure deserves attention. During an expansion, when state expenditure rises, increases for programs are debated and specifi cally funded. However, during budget crises, cuts tend to be across the board. (I discuss this issue in greater detail below). In other words, budget cuts are neither specific nor particularly debated. As a result, state agen cies have an added incentive to maximize their bud get by adding items that will be easy to cut in time of crisis. So, if individual agencies are concerned about the economic cycle, it is in their best interest not to save now for later, but to spend more. Tax cutting Throughout the economic boom, states reduced the tax obligations of businesses and individuals with in their borders. The federal program that sent tax 8 rebate checks to households in 2001 disbursed a total of $38 billion. Combined, the 50 states reduced taxes between 1995 and 2000 by a similar amount, $36 bil lion in 2001 dollars (NCSL, 2001c).10 Some of the state reductions are permanent, such as legislated re ductions in income tax rates. Other reductions were one-time events, such as tax rebates and refunds. States also reduced tax burdens further by providing funding to localities to reduce property tax burdens. These tax reductions served to bolster the already ro bust macroeconomy by returning funds to individu als at the same time as other forces were serving to increase personal income. The tax reductions were widespread, occurring every year between 1995 and 2001 and occurring in some manner in all 50 states. Figure 5 graphs net yearly state tax changes as a percent of the previous year’s tax collections against the year-over-year percentage change in second quar ter gross domestic product (GDP is seasonally adjust ed at annual rates). I use second-quarter GDP because most state fiscal years end at the end of the second quarter. Therefore, the two lines correspond to simi lar periods. Most tax reductions take effect in the year after the year of passage. The figure shows that as the percentage change in GDP turned positive in 1992, enacted tax increases began to fall, finally turning neg ative (into a net tax decrease) in 1995. The correlation between the two sets of numbers is a striking -0.8, showing the close connection between GDP growth and tax cuts. The figure also shows that the tax cut ting continued in earnest until 2000. The preliminary 2001 number shows a continued decline in taxes dur ing the 2001 legislative session as well. These data do not distinguish between one-time tax rebates and permanent changes in taxes. Therefore, this figure only accurately depicts the change from one year to the next and does not show aggregate changes over a number of years. Many of the enacted changes represented permanent changes and, therefore, the to tal tax reductions over time are greater than the sim ple sum of the numbers presented in the figure. As the figure shows, extensive tax cutting began in 1995—the first year since 1985 that states engaged in a net tax reduction (Mackey, 1999). In the 1995 legislative session, states reduced the taxes to be col lected in fiscal 1996 by $3.3 billion—0.9 percent of the previous year’s tax collections. Most reductions occurred in the traditionally unpopular personal in come tax. Personal income tax reductions represent ed $1.1 billion of the decline. Reflecting on the 1995 tax reduction, Scott Mackey of the National Confer ence of State Legislators wrote: “There are several reasons to think that state tax cutting activity may have 3Q/2002, Economic Perspectives other tax changes are difficult to generalize. One ex ception to this general pattern, not shown in the fig ure, is that throughout the decade, “sin” taxes on alcohol and tobacco were stable or increasing. In fact, tobacco taxes increased every year between 1995 and 2001, except 1998 when they remained unchanged. peaked during 1995. First, federal budget cuts that affect state budgets are a virtual certainty in 1995 and beyond, making states cautious about reducing revenues. Second, the strong revenue growth that states enjoyed in fiscal year (FY) 1994 and FY1995 appears to be returning to more modest levels. Finally, local property tax relief may be a higher priority than reducing state taxes” (Mackey, 1999). Mr. Mackey’s prediction proved wide of the mark. In 1996, states reduced taxes again. It was the first time that states had cut taxes in two consecutive years since FY1979-80. Further tax reductions occurred in the next five years. Figure 6 shows the reduction in taxes by year for the major tax categories—personal income tax, corporate income tax, sales and use taxes, and others. Other tax es include health care, motor fuel, cigarette, alcohol, and miscellaneous taxes. As the figure demonstrates, the tax cuts throughout the period tended to follow a general pattern. Every year, the main focus of cuts was personal income tax. Personal income tax was cut across numerous dimensions—rates were reduced in some cases, in others the base was narrowed, while other states chose to increase standard deductions or exemptions, or issue refunds. Corporate income taxes were also reduced, but not to as great an extent as per sonal income taxes. Sales and use taxes were largely stable, with some increases in exemptions for food, drugs, and other necessities. Figure 6 shows that oth er taxes were also cut throughout the period. Most of these declines in other taxes represent changes in statespecific tax programs such as Florida’s 1997 enact ment of a $411 million freeze in the special assessment for the special disability trust fund. As a result, these Federal Reserve Bank of Chicago 7ax cutting in the Midwest Tax cutting was persistent, across the board, and widespread throughout the second half of the decade. The behavior of the midwestem states was represen tative of this overall pattern. Faced with unexpectedly high revenues, state governors and legislators chose to return some monies to state residents and resident corporations. In this section, I briefly detail the major revenue actions undertaken in Indiana, Illinois, Iowa, Michigan, and Wisconsin during this period. Michigan was one of the most aggressive tax cut ters, legislating significant tax reductions on numerous occasions during the second half of the 1990s. As was the case with the overall pattern of tax cuts, the major source of cuts was the personal income tax. In 1995, Michigan increased personal exemptions and standard deductions. This was followed by a tax cut passed in 1999 that cut income tax rates and expand ed personal exemptions even further. This second cut reduced taxes by $218 million dollars. Prior to the start of the expansion, businesses in Michigan were heavily taxed, relative to corporations in other states. As of 1992, corporate income taxes represented 7.8 percent of Michigan’s general revenues, compared with 3.6 percent across all states. In order to increase state competitiveness, the Michigan legislature sig nificantly reduced business taxes in 1995 and 1999. 9 The reductions in personal and corporate income tax es continued to be phased in through 2002. Michigan was also typical in that any taxes that were increased were excise taxes. In 1997, motor fuel taxes were in creased, and in 1999 there was a major increase in cigarette taxation, bringing in an additional $95.2 million in revenues. Illinois’s tax reductions were similar in direction to those in Michigan, but smaller in magnitude. The main tax cut was a three-year doubling of the personal income tax exemption passed in 1998. This was viewed as a welcome change in Illinois’s historically regres sive income tax policies. Illinois also slightly reduced corporate income taxes, but the change was not as sig nificant as in Michigan. In 1999, Illinois engaged in a significant excise tax hike. In order to fund a major state public works program, state leaders increased motor vehicle and liquor taxes. The program, termed “Illinois FIRST,” was passed as a five-year, $12 billion program. Two other tax reductions were a 2000 prop erty tax rebate program and an increase in the state earned income tax program. Wisconsin’s most notable tax reductions were en acted in 1999. The state rebated $700 million in excess sales tax revenues to taxpayers who had filed income tax returns in 1998. The state also reformed the per sonal income tax by increasing standard deductions, reducing rates, and raising credits for married couples. These changes saved taxpayers $655 million. Indiana’s major tax reduction was passed during the 1999 legislative session when a major property tax decrease was coupled with an increase in the dependent child exemption to $1,000 per child. The state also changed excise taxes, reducing the unpopular automo bile excise tax in 1996, while increasing motor vehi cle license fees in 1998. Although Iowa is the least populous midwestern state, it was one of the most aggressive tax cutters. The major tax changes in Iowa were almost exclusively in the personal income tax. In 1995, personal exemptions and standard deductions were increased, while in 1997 personal income tax rates were reduced. Overall, tax changes in the midwestern states were fairly representative of those taking place across the nation. The major source of cuts was personal income tax. States both increased exemptions and deductions and lowered tax rates. Some states also decreased cor porate income taxes, but not to nearly as large an ex tent. The states only engaged in minor changes in excise taxes. Some states also acted to reduce or rebate some of the perennially unpopular property tax. We have seen how during the expansion, states used some of their windfall revenues to engage in the io very popular activity of cutting taxes. However, they also used some of these excess funds to prepare for future economic contingencies by shoring up their reserve funds. In the next section, I explore the con dition of state rainy day funds and other reserves and the extent to which states prepared for a downturn in the economic cycle. Reserve funds In order to confront unexpected shortfalls and eco nomic downturns, states maintain reserves. These re serves may be in the form of ending balances in the general fund, monies in a budget stabilization fund, or monies in a diverse array of other emergency funds. Specific rules govern when states may access the monies in budget stabilization or “rainy day” funds. By contrast, access to general fund ending balances is controlled by the same type of legislation that reg ulates other general fund appropriations. As a result, it is politically more complicated for states to access rainy day balances in the absence of an obvious need. As in the case of withdrawals, deposits for rainy day funds are controlled by specific provisions. All but three states have budget stabilization funds, which may be budget reserve funds, revenue shortfall accounts, or cash flow accounts. Those states without rainy day funds maintain all reserves as ending balances in their general fund accounts. In 2000, just under half of all reserves were maintained in rainy day funds, the other half remaining as general fund ending balances. Three-fifths of states limit the size of rainy day fund balances to between 3 percent and 10 percent of appro priations. Funds above those permitted in the budget stabilization fund remain in the state’s ending balance (NASBO, 2001a). Reserves, whether in rainy day funds or as general fund ending balances, offer states an important source of funds when unexpected contingencies threaten to disrupt fiscal functions. It is frequently cited that Wall Street views any total level of reserves in excess of 5 percent of expenditures as adequate. Figure 7 depicts total state reserves as a percent of total state expendi tures from FY1979 to FY2002. The figure also depicts the year over year percentage change in U.S. real GDP as of the second quarter (the end point for most state fiscal years). The data demonstrate a number of impor tant patterns concerning reserves. First, reserves have been quite strong. According to the data displayed in the figure, by 1998, state reserve fund balances ex ceeded the heights they had attained in 1980. Fiscal year 2002 is projected to be the ninth consecutive year with total state reserves above 5 percent of expendi tures. As of June 2001, 2002 reserves were anticipated 3Q/2002, Economic Perspectives FIGURE 7 Total reserve fund balances vs. changes in GDP Note: Total balances include both ending balances in the general fund and amounts in budget stabilization funds. Sources: GDP data from U.S. Department of Commerce, Bureau of Economic Analysis, 2002a: balances from National Association of State Budget Officers, 2001a. to be 5.9 percent of expenditures. However, while reserves remain high relative to their historical patterns, they fell between 2000 and 2001 and are expected to continue falling in fiscal 2002. Total reserve hind bal ances reached a high of 10.1 percent of expenditures in 2000 and were expected to decline to 5.9 percent by the end of FY2002. This projection for 2002, based on governors’ recommended budgets, is probably op timistic, because these estimates were published in June 2001 when the economic outlook was better. Even these optimistic forecasts predict that fiscal 2002 reserve balances as a percentage of expenditures will be low er than they have been in the past seven years. These averages mask significant variety across states. While 22 states anticipated total reserve balances below 5 per cent of expenditures for 2002, four predicted balanc es would exceed 10 percent. The decline since 2000 is widespread. In 2000,21 states had reserve hind balanc es above 10 percent and 11 had reserve fund balances below 5 percent. Reserve fund balances have declined over the past two years due to additional tax cuts, in creases in spending especially in the areas of health care and education, and the slowing of the economy. The data also show how quickly reserves can fall in responses to economic difficulties. Between 1989 and 1991, reserves fell from 4.8 percent of expendi tures to 1.1 percent of expenditures. So, while reserves were nearly adequate according to the oft-cited Wall Street rule of thumb prior to the early 1990s downturn, they nearly evaporated in just two difficult years. This begs the question whether the reserves that states built up during the booming 1990s are sufficient Federal Reserve Bank of Chicago to help them weather the current economic storm. The news reports from state governments suggest (discussed in detail below) that these reserves are not sufficient to allow states to endure the current economic situa tion without cutting spending or raising taxes. The inadequacy of state reserves to maintain ser vices in the event of a downturn was addressed in an article by the Center on Budget and Policy Priorities (CBPP) in March 1999 and updated in May 2000 (Lav and Berube, 1999, and Zahradnik and Lav, 2000). The authors calculate the amount of reserves each state would need to endure a recession without cutting programs dramatically or enacting significant tax increases. They then compare this level of needed reserves to the level available. In their calculations, the authors assume that states would face a fall in the growth rate of revenues between FY2000 and FY2003 similar to the decline experienced between FYI989 and FYI992 that cor responded to the 1990 recession. This methodology leads them to assume that the growth rate of revenues would be 43 percent of the growth rate from FYI 993 and FYI998. At the same time, they assume that state expenditures would grow at the same pace as they did between 1989 and 1998. The authors calculate the needed reserves as the gap between expenditures and revenues over the three-year period.11 These cal culations yield a conservative estimate of necessary reserves, because the 1990 recession was relative short lived and mild. In addition, as mentioned above, the demand for government services tends to increase slightly more rapidly during a recession, suggesting the growth rate in expenditures (absent government action) would be greater than experienced between 1989 and 1998. The authors conclude that only eight states (Delaware, Indiana, Iowa, Maine, Massachu setts, Michigan, Minnesota, and North Dakota) had adequate reserves on hand to combat a relatively mild recession. Other states had reserves that were lower than needed. In fact, they find that in most states re serves on hand were more than 10 percent of expen diture below what was required to maintain services. In their follow-up report, the authors noted that five of those original eight states (Delaware, Indiana, Massachusetts, Michigan, and Minnesota) had enacted tax cuts since the previous publication that left them without sufficient reserves. The CBPP report also ar gues that the statistic that 5 percent of expenditures is sufficient, while frequently cited, is “of uncertain ori gin and even more questionable validity.” They argue that reserves equal to 5 percent of expenditures are insufficient for managing recessions in all but a couple of states. The report correctly points out that needed reserves vary from state to state. States that depend 11 heavily on cyclical sources of revenue, especially in come taxes, need a greater level of reserves. The 5 per cent statistic better represents the level needed on hand for unpredictable emergencies, perhaps an event like September 11, than the level required to counteract revenue losses caused by the business cycle. This leads one to question why states did not take advantage of the strong economy and move to build adequate reserves, and why those few states with suf ficient reserves had spent them. The simple answer to both of these questions would be that it is far easier to spend money than to save it. One might say state leaders are either myopic and do not worry about fu ture economic difficulties or are overly optimistic and, thus, did not believe another recession was likely. How ever, such thinking misses the important point that states do not view reserve funds as designed to allow them to maintain services in a downturn. Rather they view these funds as allowing a window during which they can adjust their budgets and cut services or raise taxes in an orderly fashion. In other words, reserve funds are designed to allow states time to build the ark; they are not designed to carry them through the deluge. The evidence for this distinction is widespread. First, most states cap the amount of money allowed in the reserve fund. Thirty-three of the 46 states with rainy day funds cap the amount allowed in the fund. Most of the caps are at or below 5 percent of expen ditures. If states wanted these funds to counteract the fiscal effects of recession, they would not cap them at such a low level. Second, the language used by states when discussing their reserves tends to be based on concern for unexpected or short-term disruptions, not prolonged economic problems. For example, Illinois passed Rainy Day Fund legislation in April 2000. The state controller made the first deposit into the new fund on July 1, 2001. The Illinois fund is capped at $600 million (2.6 percent of 2000 state general fund expenditure). Previously, the entire ending balance had been left in the general fund. In a press release praising the legislature’s ac tion, Governor George Ryan stated that the fund was “for use at the discretion of the governor and General Assembly in the event of an unseen economic down turn that threatens state services (State of Illinois, 2000).” The language used by NASBO in explaining reserve funds is similar. They write, “[Tjotal balanc es reflect the funds states may use to respond to un foreseen circumstances after budget obligations have been met” (NASBO, 2001a). State balanced budget requirements and debt re strictions limit the ability of states to borrow to meet 12 short-term needs. In lieu of access to short-term credit markets, states maintain reserves, permitting them to dip into savings rather than borrow. These reserve funds buy states time, giving them the opportunity and flexibility to adjust their budgets in a deliberate, sen sible manner. These funds help states avoid fiscal gim mickry to affect budget balances. However, the reserve funds do not allow states to emerge unscathed from recessions. For better or worse, state governments cling firmly to their balanced budget requirements and be lieve that they ought to spend in one year what they receive in that year (or over two years in states with biennial budget cycles). That said, overall state fiscal health would improve if reserve funds were adequate to allow states to maintain, or even increase, spend ing without increasing taxes during economic down turns. Preserving balances for this purpose would require a change in thinking about state budgeting. Midwestern states ’ reserves Table 2 shows the level of reserves as a percent of general fund expenditures in the midwestern states and for the nation as a whole from FYI998 to FY2002. The data show that reserves among this group of states have been fairly typical of the U.S. averages—reach ing high levels over the past five years, though fall ing more recently. With the exception of Wisconsin in 2001-02, all midwestern states have maintained reserves above the 5 percent threshold. The final col umn of the table also displays the level of reserves needed to survive a mild recession, according to the CBPP report. These numbers tend to vary quite dra matically across states. Wisconsin’s reserves are the weakest of the group. They declined dramatically between 2000 and 2001, principally due to tax rebates passed in 1999. At the same time, Wisconsin’s required recession reserves are the highest of the Midwest states and the eighth highest among the 50 states. Illinois’s reserves have been quite stable over the past five years, hovering close to the 5 percent mark. By contrast, required reserves are quite high, suggesting that Illinois will face fiscal difficulties. Both Iowa and Indiana have reserves that have been declining over time but are slightly above the nation al average. Their required reserves are lower than the national average, but still double the level of reserves on hand. Michigan’s reserves have proven the strongest among this group of states, exceeding 10 percent in all five years presented in the table. As of 1999, Michigan exceeded the reserves required to withstand a recession by 10 percent. However, significant tax 3Q/2002, Economic Perspectives TABLE 2 Reserve balances as percent of expenditures 1998 1999 2000 2001 2002 Required reserves (as of 1999) Illinois Indiana Iowa Michigan Wisconsin 5 state total 50 state total 6.10 23.00 19.60 12.20 5.70 10.90 11.00 6.30 20.60 16.00 15.40 7.00 11.00 8.90 6.60 18.30 13.40 15.10 7.40 10.50 11.90 5.70 9.80 9.80 12.90 2.70 7.30 9.10 5.60 8.30 8.20 12.90 2.00 6.70 6.30 22.90 14.90 16.30 5.10 27.00 18.60 Note: Data for 2001 are estimated and data for 2002 are from recommended budgets. Sources: Required reserves from Lav and Berube, 1999; reserves from National Association of State Budget Officers, 1999 and 2001a. cuts enacted during the last years of the expansion have increased the state’s required reserves from the 5.1 percent reported in the table to 25.0 percent. Therefore, after aggressive tax-cutting, Michigan’s reserves were quite low relative to required levels. The experience of the midwestem states has been fairly typical. Reserves are high relative to historical levels, but low relative to cyclical requirements. The only states with low reserves as a group are a number of southern states—Alabama, Arkansas, Kentucky, Louisiana, North Carolina, and Tennessee—that did not benefit as much from the economic expansion as states in other regions. Indebtedness In addition to lowering taxes, increasing spend ing, and bolstering reserves, states also had the option of using their new-found revenues in the 1990s to re duce their indebtedness. States are not major debtors. Total outstanding state debt at the end of FYI 999 of $510 billion represented just 51 percent of total an nual state expenditure (U.S. Department of Commerce, Bureau of the Census, 2002a). By comparison, the federal government’s indebtedness is over three times its annual expenditures (U.S. Department of Commerce, Bureau of the Census, 2002c). On balance, real state indebtedness actually increased by 20 percent between 1992 and 1999, but this was less than the percentage increase in expenditure. Because most debt is long term and funds specific capital projects, it is not sur prising that debt increased at a time that capital spend ing was also increasing. While states could have used their surplus funds to pay down their debt or fund more capital projects out of current funds, they did not do so. The only kind of debt that declined during the expansion was short-term debt, which was 21 per cent lower at the end of 1999 than at the end of 1992. Federal Reserve Bank of Chicago This type of debt that matures in one year or less com prises bond and tax anticipation notes and is a barom eter of the health of state finances. The bust: Revenues The first real signs of the deterioration in state bud gets were seen in the third quarter of2000. These signs could be seen in the revenue numbers being reported by state governments. Figure 8 shows year-over-year changes in quarterly tax revenues both in total and for sales taxes. As the figure shows, between the third quarter of 1999 and the third quarter of 2000, state tax revenues, adjusting for tax changes and inflation, had grown by only 4.1 percent. These data for total tax revenues are estimates of what state revenues would have been had legislated tax changes not occurred. State revenue growth had slowed for the first time in a year and was only half the growth rate reported in the previous quarter. The changes in actual sales tax revenues were even more dramatic relative to their historical trend. Sales tax revenues grew by 4.7 per cent that quarter, the lowest growth rate reported since the first half of 1997. These sales tax revenue numbers, unadjusted for inflation or tax changes, are probably the most reliable indication of the revenue situation. Major legislated changes in the sales tax are rare and, as a result, these numbers do not rely upon predictions of the effects of legislated changes. At this juncture, there was a great deal of variation in the financial situation confronting different states and regions. Figure 9 displays changes in year-over-year sales tax revenues, by quarter, for the various regions. The figure shows that, as of the second quarter of 2000, some regions were doing far better and others far worse than the national averages. In particular, sales tax reve nues in the Far West were growing well above the na tional average, while revenues in the Mid-Atlantic, 13 Great Lakes, and Southeast were lagging the national average. Between the second and third quarter, growth rates dropped off in all areas, with the exception of the Mid-Atlantic and Rocky Mountain states, where they were flat. Revenues in these two regions fell the fol lowing quarter. Some states were still well entrenched in the impressive expansion, while others were falling quickly into revenue troubles. The decline in the growth rate of tax revenues first hinted at in the third quarter of 2000 accelerated in all the subsequent quarters for which data are avail able except one. In the third quarter of 2001, overall tax revenues and sales tax revenues fell. Between the third quarter of 2000 and the third quarter of 2001, revenues had fallen in nearly every region, exceptions being the Southeast and Southeast where revenues were close to flat. We see a similar picture when we look at the revenue growth rates for the second quar ter of 2001. Although still positive, revenue growth rates were weak in all regions. The state revenue sit uation was rapidly deteriorating in advance of the September 11, 2001, terrorist attacks. Although it is difficult to distinguish the effects of the attacks from the effects of a continuing secular decline in revenues, there is little doubt that state finances were in trouble before September 2001. Next, I look at personal income tax revenues. Because of the frequency of legislated changes in this tax, I present numbers that are unadjusted for legislated tax changes, as well as adjusted numbers. Unadjusted revenues are the revenues actually re ceived by the state government. Adjusted numbers display estimates of what receipts would have been had the legislature not changed the tax code. While income tax revenues, adjusting for legislated tax changes, held up through the end of 2000, the income tax situation was also poor by mid-2001. Figure 10 depicts the quarterly change in personal income tax revenues by quarter, both adjusted for legislated tax changes and unadjusted. Because nearly all income tax changes lowered taxes, the unadjusted line (reve nues actually received) lies almost entirely below the adjusted line. Income tax revenue changes are more difficult to interpret than changes in sales tax reve nues for two principal reasons. First, the income tax is frequently changed and even thorough estimates of the effects of legislated tax changes are bound to be imprecise. Second, because taxes must be filed prior to April 15, there is a high level of seasonality in in come tax revenues and, as a result, comparisons across quarters are quite difficult. The fall in income tax revenues is more easily understood by looking at the changes in year-over-year revenues by quarter as show in figure 11. This figure shows that in every quarter since 2000 :Q4, revenue growth rates have fallen below their level from a year earlier. The fact that income tax revenues fell slightly later than sales tax revenues suggests that the more sales-tax-dependent states were likely confronted with revenue is sues earlier than more income-tax-dependent states. When looked at from numerous angles, the state revenue situation appears poor. Revenue growth rates slowed early relative to the slowdown in GDP and have continued to decline. Revenue numbers also contin ued to fall below already reduced expectations and, by November 2001, 43 states were reporting that revenues had come in below what they had anticipat ed for FY2002. By April 2002, this number had risen to 48 states. (NCSL, 2001b; NCSL 2002) Expenditures Unsurprisingly, this decline in revenues has not coincided with a decline in the demand for state ser vices. On the contrary, among the 43 states reporting revenue shortfalls for FY2002 in November 2001 (NCSL, 2001b), 20 were also reporting that spending was exceeding levels anticipated when fiscal 2002 bud gets were passed. By April 2002, 33 states were re porting spending overruns (NCSL, 2002). Predictably enough, the main source of spending overruns and con cerns involved the Medicaid program. Nearly every state that reported spending problems, along with some that reported that spending remained on target, high lighted Medicaid spending as problematic. 14 3Q/2002, Economic Perspectives FIGURE 9 Year-over-year changes in sales tax revenues, by region and quarter Sources: Jenny, 2001, and Jenny and Boyd, 2001. Next, I look at how states might respond to this budget predicament. To do so, I examine how the states reacted to previous downturns, how the states hit ear liest by the current recession have reacted so far, and the projections and pronouncements coming from state capitols. State reactions: The 1991 recession The 1991 recession was mild and relatively short compared with previous downturns, but it hit the states very hard. States dramatically cut services and enacted large tax hikes (see figure 5). Many of these changes occurred in the middle of the fiscal year. In many cases, states were compelled to change their enacted budget mid-year to avoid running foul of their balanced bud get provisions. Reducing enacted budgets is a sign that the economy is worse than was anticipated when the original budget was passed. Thirty-five states faced a potential budget deficit at one point from 1990 to 1992, and 20 or more states acted to reduce enacted budgets during each year from fiscal 1990 to fiscal 1993. The worst year was 1991, when 30 states faced a mid-fiscal-year deficit of nearly $15 billion (2.7 percent of general expenditures). In 1991, states drew down their reserve balances. Balances at the start of the downturn were reasonably healthy, totaling 4.8 percent of ex penditures in 1989. However, by 1991 balances had fallen to 1.1 percent of expenditures. A similar pattern Federal Reserve Bank of Chicago existed during the 1980s recession. For example, bal ances declined from 9 percent to 4.4 percent in the oneyear period from fiscal 1980 to fiscal 1981 (shown in figure 7). In 1991, states were also forced to cut their budgets by $7.6 billion. Because balances had been used to deal with the 1991 fiscal situation, these excess funds were no long er available in FYI992 and FYI993. With little avail able reserves, states were forced to reduce current 15 year budgets further and raise taxes. In 1992 and 1993, 35 states and 23 states, respectively, reduced currentyear budgets, and states raised taxes by a total of $25 billion. If spending cuts and tax increases were insuf ficient, states resorted to fiscal gimmickry to affect budget balances. The most popular form of gimmick ry is for states to postpone payments to vendors, em ployees, and other recipients of state funds. Illinois was one of the main practitioners, increasing the time between the receipt and payment of bills. States could also speed up the collection of revenues by forcing vendors to remit payments to the states more quickly. Enacted budget reductions are very disruptive to service provision, because budget changes need to go into effect almost immediately (and sometimes even retroactively) leaving state agencies and their clients little time to anticipate and adapt to the changes. While declines in service provision can take effect almost im mediately, tax increases take longer. Most tax increas es go into effect in the fiscal year following the year of passage. States can only cut spending mid-year for a select range of programs. Many programs are nearly impos sible to cut mid-year. The largest item in state budgets, elementary and secondary education, is hard to reduce once teacher contracts have been signed. Also, school ing involves significant start-up costs that occur in the beginning of the school year, which is fairly early in the fiscal year in most states. By mid-fiscal year, school expenditure is fairly inflexible. In the early 1990s, there fore, states cut spending on those programs where cuts were possible mid-year, which tended to be pro grams that serve the poor (Lav and Berube, 1999). 16 The largest of these programs—then AFDC, now TANF—has undergone a series of changes that will greatly limit the states’ ability to cut funds in the future. In 1996, the program changed from an entitlement pro gram where the states and federal government split payments to a discretionary program where specific amounts were block granted to each state. States were given more control over the structure of their pro grams, with the exception that they needed to maintain spending at or above 75 percent of their 1994 spending level if they met work requirement provisions, and at or above 80 percent of their 1994 spending level if these work requirements were not met. These “main tenance of effort” provisions prohibit states from re ducing their expenditure below a certain level. Throughout the recent economic boom, caseloads have dramatically dropped and the maintenance of effort provisions has proved to be binding in a number of cases. In federal fiscal year 2000, only 11 states spent more than 80 percent of their 1994 baseline, with 15 states spending exactly 75 percent of their 1994 level and five states spending exactly 80 percent. The com bination of relative fixed funding levels and smaller case loads has meant that states have been able to sup port their dependent populations with a wide array of benefits and services in addition to cash grants. Such additional programs include work transportation, child care, and housing assistance. In the face of significant fiscal pressures, states will not be able to decrease funding levels below their maintenance of effort require ments. However, they may well keep funding at, or lower funding to, the maintenance of effort levels. A number of states will actually be able to increase total TANF funding dining an economic downturn without harm to their budget situation, because these states have not spent their entire block grants, leav ing excess amounts on account with the federal gov ernment. The TANF legislation explicitly allows states to reserve part of their block grant for future spend ing. Funds reserved with the federal government can be spent in subsequent years on “assistance.” While states will not be able to spend these monies on the wide array of “non-assistance” purposes that TANF has funded, these funds will allow greater expenditure on cash benefits. Unspent TANF funds are categorized in one of two ways for federal reporting purposes—as either unob ligated funds or as unliquidated obligations. Unobli gated hinds are monies neither committed nor expended; these funds would be available for additional cash as sistance spending during a period of economy hardship. Unliquidated obligations are payments that have been committed by state governments, but not yet spent. 3Q/2002, Economic Perspectives Additionally, in some states portions of unliquidated obligations are not truly committed and would also be available during a downturn.12 The true measure ment of available funds lies somewhere between the level of unobligated funds and the sum of both types of unspent funds. As of the end of federal FY2000, the 50 states had $2.7 billion in unobligated funds on account at the federal government and $8 billion in total unspent funds. This represents 9.7 percent and 28.6 percent, respectively, of total required state and federal TANF spending in federal FY2001, where required state and federal spending is defined as the sum of the 2001 TANF grant and the 80 percent maintenance of effort provi sion. There is a great deal of variation in the amounts available to different states. While 11 states have less than 10 percent of one year’s funding unspent, nine states have more than 50 percent of a year’s funding unspent (Lazere, 2001). These unused block grants are an additional form of rainy day reserves, providing states with an added cushion as the economy declines. Therefore, TANF spending changes will be more complicated in the com ing days. On the one hand, the saved block grants will make it easier for states with saved amounts to increase spending, and the maintenance of effort provisions will not allow states to cut spending below a certain threshold. On the other hand, states have added flexi bility to cut expenditure to the level of their mainte nance of effort requirement, because the program is no longer an entitlement program. If the dependent population increases, as occurs with a deteriorating economy, and funding levels stay relatively fixed or increase only slightly, benefits and services will inevitably be cut. The most likely targets for cutting will be those same creative new benefits in transportation and childcare that have characterized the very successful first years of TANF.13 If state behavior during the current downturn par allels that taken during the 1991 recession, we will see states begin by drawing down their balances and cutting budgets and then progress to cutting spending more dramatically and increasing taxes. The 2001 recession: Action thus far The combination of lower revenues and high or stable spending has meant that state budgets are com ing increasingly underpressure. State governments have taken various actions to confront these budget issues and bring their FY2002 budgets into balance. State governors have also begun to put forward their FY2003 budgets. As of November 2001, 36 states had cut their budgets for fiscal 2002, 24 had decided to use some Federal Reserve Bank of Chicago of their reserves, and 22 had turned to other measures, explained by the NCSL (2001b) as including “hiring freezes, capital project cancellations, and travel restric tions.” By April 2002, 40 states had reduced or were planning to reduce their budgets, 26 had turned to their rainy day funds, and 17 were eyeing tobacco settlement dollars (NCSL, 2002). Mid-year budget cuts are often across the board, with nearly all departments faced with funds a few percentage points below previously budgeted levels. In many states some sacrosanct departments, particu larly K-12 education, are spared from these cuts. Across the board spending cuts are common perhaps because they are the easiest to implement quickly. Because states need to bring their budgets back into balance quickly, they lack the ability to carefully determine areas where budget reductions would be least damaging. It then falls to the individual state agencies to choose the ex act programs where the reductions will be implement ed. More specific budgetary debate has accompanied the early discussions concerning 2003 budgets. As states have debated spending cuts, attention has inevitably turned to the Medicaid program. Through out the 1990s expansion, there was little broad discus sion of the problems with Medicaid spending. Although health care expenditure specialists did debate the is sue, debate was not widespread. As the economic sit uation has deteriorated, Medicaid spending has once again come to the fore. This is not surprising, given that it is a quickly growing program that already ac counts for nearly 20 percent of all state expenditures. To reduce costs, states have both restricted eligibility further and tried to cut costs per eligible recipient. Cost cutting can take many forms. States have limited ac cess to services and drugs by increasing the need for pre-approval and by reducing optional benefits. States have also contemplated increases in co-payments, shift ing more of the costs onto recipient families. In addi tion, states have reduced payments to service providers. This strategy is successful in reducing costs, but may lead more providers to refuse to serve Medicaid pa tients. For example, state proposals for reductions in prescription payments have led drugstores to threaten to stop filling Medicaid prescriptions (Associated Press, 2002). States have also turned to drug companies and asked for larger volume discounts on Medicaid drug purchases. Finally, states have asked the federal gov ernment for an increase in the matching rate. One jus tification for this request is that some of the increase in Medicaid costs derives from federally legislated increases in the eligible population. In addition to cutting spending, states have also drawn down their reserve fund balances. As shown in 17 figure 7 and discussed earlier, balances were expect ed to fall from 10.1 percent of expenditure in 2000 to 5.9 percent by the end of fiscal 2002. As of Novem ber 2001, seven states indicated that they would defi nitely be using reserves to balance their budgets, and 17 additional states were contemplating using reserves to balance their 2002 budgets (NCSL, 2001b). Thus far, there has been little movement, espe cially among state governors, to raise taxes. In 2001, only six states passed substantial increases in taxes. And only in North Carolina was the increase viewed as a response to revenue problems caused by the re cession (Jenny, 2002) One lesson that was reinforced during the early 1990s recession was the political un popularity of raising taxes. Governors who had en acted significant tax increases were almost universally voted out of office in favor of politicians promising to lower taxes. In many cases, these changes in lead ership coincided with the improving economy and new governors were able to keep their campaign promises. A number of the governors and legislatures that came to power in the early 1990s will soon face a similar dilemma to that confronted by their unfor tunate predecessors. So far, the limited discussion of tax increases has revolved around the cigarette and alcohol taxes. Oregon’s governor proposed increasing these two tax es in order to balance the budget. Similarly, Indiana’s governor has proposed hiking taxes on cigarettes and gambling. These taxes are politically the easiest to hike, although they are not the most lucrative revenue sources. However, for the most part, governors have chosen to speak out vociferously against tax increases. For example, New Jersey Governor McGreevey stated that he was “ruling out a tax increase” as a way to solve budget problems (Herszenhorn, 2001). This sentiment has been echoed by numerous other governors and legislative leaders across the country. That said, judg ing from the experience of the 1990s, tax hikes tend to occur late in a decline after other, easier avenues of budget balance have been exploited. The debate over 2003 budgets is quite similar to the debate over 2002 budgets. While this discussion is occur ring in a less panicked environment, the policy decisions closely parallel the decisions made concerning FY2002 budgets. In particular, states are relying on spending cuts and reserve funds rather than on tax increases. If revenue estimates for FY2003 prove too opti mistic, more states will likely turn to discussion of tax hikes. Tax increases during FY2003 may prove particularly politically challenging. If the trend in positive national economic news continues, state 18 leaders will need to justify tax increases at the same time that voters are hearing more about the overall health of the macroeconomy. Actions in the midwestern states Midwestern states were among the first hit by the downturn. Returning to figure 9, we see that the Great Lakes states had either the weakest or close to the weakest tax growth in all the quarters pictured. The midwestern (and southeastern) states had tax collec tions significantly below projections in 2001. Most other states did not begin experiencing revenue prob lems until FY2002 (see NASBO, 2001a). The midwestern states were among those hit ear liest by falling revenues. One principal reason for this was that a downturn in manufacturing production preceded the downturn in overall GDP growth. Fig ure 12 shows the trend in manufacturing relative to the trend in GDP, while table 3 details the percentage of state employment in manufacturing both overall and for the midwestern states. Illinois As of November 2001, Illinois was facing a $500 billion deficit in the FY2002 budget. This deficit was principally caused by lower-than-expected state reve nues. As of October 2001, FY2002 revenues were $262 million below the level collected over the same peri od the previous year. In order to confront the deficit, the government called on state agencies to reduce their spending by 2 percent and instituted travel restrictions, a hiring freeze, and a one-day furlough program for state workers. (In the end, the furlough program was pre vented by the state employees’ union.) The governor also cut Medicaid payments to some hospitals, although some of the original cuts were subsequently restored. The 2003 budget proposed by the governor in late February 2002 appropriated $22.7 billion from the state general fund, representing a decline in $700 million from the previous year’s appropriations. The proposed budget included no tax increases, but further across the board agency cuts of 3 percent. Additionally, the governor proposed a cut in the state work force of 3,800 workers, principally through an early retirement pro gram. Furthermore, some state penal mental health facilities were to be closed or have their opening de layed (State of Illinois, 2002). Indiana As of November, Indiana’s revenues were antici pated to be $540 million below the original forecast for FY2002. By April, the state was facing a deficit of $1.3 billion in the fiscal 2002-03 biennial budget. 3Q/2002, Economic Perspectives is $48 million that had been slated for the state rainy day fund and an additional $42 million from the ex isting rainy day fund balance. Competing budget plans from senate Republicans (the governor is a Democrat) propose reducing spending more dramati cally, including spending on education, and relying less on emergency funds (Okamoto, 2002). The governor dealt with this shortfall by freezing a series of state capital projects, instituting a hiring freeze (both in September), and calling on agencies to reduce expenditures by 7 percent. He also proposed increases in a number of different taxes, including tax es on cigarettes and casinos and further cuts in agency budgets. The legislature failed to enact tax increases before adjourning in March, and the governor put spending cuts directly into effect and recalled the leg islature for May. School funding was among the areas cut. Indiana is the only one of the midwestern states that is seriously considering tax increases. However, the discussion concerning tax increases is taking part in the context of a general tax restructuring caused by a court-ordered change in the property tax. Iowa Through the end of December 2001, Iowa’s rev enues were $200 million below original projections. In order to confront the resulting deficit, the governor implemented a 4.3 percent across-the-board spending cut. Subsequently, funding was restored for a selection of programs, including elementary and secondary ed ucation. Further bad news in February was met by an additional 1 percent cut in the 2002 budget, use of state emergency and tobacco settlement funds, and a furlough program for state workers. The governor’s proposed (revised) 2003 budget continues to avoid tax increases but proposes to bal ance the budget using a further 3 percent cut to agency budgets and funds from a number of state reserves. Education would continue to be shielded from cuts. Included in the reserves the governor proposes to use Federal Reserve Bank of Chicago Michigan As of November, Michigan’s general fund reve nues were projected to be $462 million below original estimates and overall revenues 2.5 percent below fis cal 2001 collections. The state made up for this short fall by canceling a series of capital projects, enacting spending cuts, and using money from outside the gen eral fund, including tobacco settlement money and money from the contingency fund. Spending cuts fo cused on health, welfare, and corrections, but K-12 education spending was not cut. The state considered delaying or canceling previously enacted income and business tax cuts, but chose not to do so. The governor’s proposed 2003 budget plans to make up for a $ 1 billion shortfall using additional spend ing cuts, in particular a freeze in the state-local reve nue sharing program and significant withdrawals from the rainy day fund and other state reserves. He does not recommend tax increases, aside from a small in crease in diesel taxes. The budget shields schools from cuts, in part by moving the timing of school tax payments (Cain et al., 2002). Wisconsin Wisconsin was facing a $1.1 billion deficit in its biennial budget covering FY2002 and FY2003. The deficit is primarily a result of lower than anticipated income tax collections. Because of the biennial bud get cycle, the state needed to confront 2002 and 2003 issues together. The governor’s proposed budget plan includes 3.5 percent and 5 percent reductions in TABLE 3 Percent of employment in manufacturing by state, 2000 Illinois Indiana Iowa Michigan Wisconsin U.S. Manufacturing % 50 state ranking 16.1 23.2 17.8 21.6 22.2 14.3 17 1 12 4 2 Source: U.S. Department of Commerce. Bureau of the Census, 2002c. 19 agency spending in 2002 and 2003, respectively, modest cuts in university spending, and a phase out of the provision of discretionary moneys or “shared revenues” to local governments. Education and state programs serving the needy were for the most part shielded from cuts. The governor also proposes to borrow $794 million from the state tobacco settle ment fund to fund the shared revenue program while it is being phased out (McCallum, 2002). Each of the midwestern states has chosen a dif ferent package of changes to address budget deficits for the current fiscal year. These states have also be gun debating how to ensure that the budgets for FY2003 will be balanced. While the choices made have been different, a general pattern emerges with the states en acting the least painful changes first and evolving to harder decisions as the budget situation has continued to deteriorate. Hiring freezes and travel restrictions have been followed by across the board spending cuts and a drawing down of reserve funds. States have re lied on reserves not only in their rainy day fund, but also funds from the tobacco settlement, and other more obscure places. While tax increases have largely been avoided, if state revenues continue to disappoint, fur ther agency cuts may prove too painful, reserves will be largely spent, and the states may have to resort to tax hikes to balance their FY2003 budgets. Conclusion: Lessons learned After the 1991 recession, many observers hoped that states had learned about the dangers inherent in their budget situations and would react in subsequent booms in ways that would prevent a recurrence of fis cal crisis. The current fiscal situation indicates that many of these lessons were inadequately learned. The biggest problem states face is the combination of cyclical revenues with acyclical or even counter cyclical obligations and institutions that are not per mitted to use financial markets to deal with this disjoint. States have acted in ways that exacerbate this mismatch. For example, while the reduced sales tax rates on food and prescription drugs are motivated by understandable, even admirable, policy objectives, these serve to increase the sensitivity of revenues to the business cycle. How can states deal with this problem? Rainy day funds While states’ balanced budget requirements pro hibit them from borrowing, they are permitted to save money. The principal ways this is done is through rainy day funds and cash balances in the general ac count. States should increase the levels of these 20 funds during booms to prepare for the inevitable de cline in revenues when the economy sours. As men tioned earlier, rainy day balances have been rising over recent decades. States should continue this trend. In fact, if states are successful in managing the current downturn without resorting to significant tax hikes, research may ultimately attribute this success to the health of reserves at the start of the recession. One issue regarding rainy day funds is that they are perceived as funds to cover short-term adjustment needs rather than longer-term revenue shortfalls. They are preparing states to manage for a rainy day rather than for the rainy season, or several seasons, that an economic downturn represents. State leaders would need to change their perception of these funds in order to allow them to grow to the levels needed to main tain services in the face of widespread economic dif ficulties. In keeping with the increased role of reserves, state legislatures would need to increase the permit ted size of reserve funds. As mentioned above, many states limit the level of reserves. Tax cuts The current situation, where taxes are cut during a boom and increased during a recession, both exac erbates the economic cycle and means that consider able energy is being expended in debating changes that are soon reversed. Given the political popularity of tax cuts, it would be idealistic to suggest that states should not cut taxes when the economy is booming and instead maintain all excess funds as reserves. At the same time, the political unpopularity of tax in creases means that needed tax increases occur late in the economic cycle after considerable damage has been done in terms of interruptions to state-provided services. In order to deal with this problem, states should consider enacting tax cuts that do not require offsetting legislation to be reversed in subsequent years. In particular, they might consider tax rebates and refunds rather than legislated reductions in rates. In this way, states could return money to taxpayers without jeopardizing the finances of the government during economic difficulties any more than is done by the contingencies of the economic cycle itself. Many of the tax cuts enacted during the expansion were re bates. More states should consider these during fu ture surplus years. Expenditure patterns One reason some observers argue against higher reserve balances is that they believe that governments will see these balances and find wasteful ways to spend 3Q/2002, Economic Perspectives them. By returning money to taxpayers instead, government leaders are relieved of this temptation. In other words, these commentators believe that taxpay ers are better stewards of resources than legislators. The long-term spending trends in the states justi fy this worry. State spending has been on an upward trajectory relative to personal income for quite some time. Governors and legislators should work to con front the spending demons by carefully reexamining spending priorities. Spending appears only to be care fully controlled during fiscal crisis and not during calmer times. States should look closely at how agen cies confront across-the-board spending cuts to deter mine where excess fat may be in the system. Additionally, when the economy improves, the states should continue the scrutiny of the Medicaid pro gram that is occurring during budget discussions. Medicaid spending is particularly problematic be cause its rate of growth shows no sign of abating. Also, major adjustments in the program are likely to be slow to develop because they would require the cooperation of state and federal authorities. Additionally, states should consider public rela tions programs that educate the public about the valu able services they provide. For example, do taxpayers know that states are the largest providers of school funding or do they believe that this service is princi pally funded locally? Leaning on the federal government States should not expect the federal government to bail them out when the economy sours. While the states need to act quickly to affect budget balances dur ing recessions, the federal government makes policy in a slow and considered fashion. The recent experi ence with Medicaid spending demonstrates the prob lems of relying on the federal government. While the states have been requesting additional funds for over six months, stimulus packages containing Medicaid relief for states have consistently stalled in Congress. While states may well get their additional Medicaid Federal Reserve Bank of Chicago support eventually, it will not come quickly enough to ameliorate the last-minute budget crises. The prob lems underlying the requests for added Medicaid funds are part of long-standing trends. The federal govern ment may have been more receptive to these requests during more robust economic times. Reversing balanced budget restrictions One additional option for the states would be to reverse their long-standing balanced budget restrictions and debt limits. This would allow states to borrow from financial markets when the economy deteriorates and (presumably) to pay the money back as the econ omy improves. There are two principal arguments against such a suggestion. First of all, such a recom mendation is impractical. Governors and legislators are very proud of their balanced budgets. Even the suggestion that these rules be reversed would be po litical suicide. Second, and more importantly, without these restrictions, states would be less compelled to make difficult spending decisions. As a result, state spending would likely get even more out of hand. Balanced budget restrictions mean that budgets are balanced both from year to year and (as a result) on average. While yearly balanced budgets are troubling because of the business cycle, the fact that states are not major debtors is an important strength of the state fiscal process. The current fiscal condition of the states and the difficult budget negotiations states are engaged in have come alarmingly soon after a long period of windfall revenues. While state leaders used these revenues to cut taxes and increase spending, they did not use them to plan adequately for a weak economy. After two simi lar fiscal crises only a decade apart, one might hope that states will understand the need to plan for future recessions. States could better prepare for recession by relying more completely on their reserve funds. This would allow them to escape their historical pat tern of increasing taxes when citizens are poorest and cutting services when they are most needed. 21 NOTES The majority of state budgets pertain to a fiscal year that starts on July 1 and ends on June 30. A small number of states use a different fiscal year. While most states operate annual budgets, 21 have biennial budget cycles. In some smaller states, this is in con junction with a legislature that meets every other year. In states with biennial budgets, full-blown budgets are only authorized every other year, but supplemental budget bills are often passed in off years to cope with unplanned contingencies. 2States themselves rely on different fund definitions than the Cen sus Bureau. I use Census Bureau definitions because these guar antee comparability across states. Throughout this article, I use 1992 as a dividing point for data com parisons because it was the first year of positive economic growth during the recent expansion. The last year of data used depends on data availability. I use the most recent year for which histori cally comparable data is available. All real numbers are calculated using the gross domestic product implicit price deflator from the Executive Office of the President, Council of Economic Advisers, 2002. Average yearly growth rates are based on compounding. California does collect these data. Income from options and capi tal gains in the state grew from $25 billion in 1994 to $200 billion in 2000 then fell to $70 billion in 2001 (Sterngold, 2002). 5Due to the particulars of the settlement, no monies were due for 1999. However, states received the 1998 funds in 1999. 6The levels of expenditure from the NASBO and Census data sets are quite different; however, historical comparisons find that the rates of expenditure growth tend to be very similar (Merrimam, 2000). 7Administrative costs are not included in any of these categories. The underlying numbers include only Medicaid payments on be half of recipients. 9Instruction represents 61.7 percent of current education expenditure, and current expenditure represents 85 percent of total expenditure. The 61.7 percent number is frequently cited as the percentage of spending on teachers, but this excludes non-current spending. 10This figure actually underestimates the net effects of the decline in state taxes because it treats each reduction as only reducing one year’s taxes. Many reductions were permanent and therefore reduced taxes in all subsequent years. nThis is a reasonably quick way to get an approximate calcula tion. One problem with this measure is that it leads to the predic tion that those states with the fastest growth rates of revenues from FY1993 to FYI998 would continue to face the fastest growth rates in the future. However, if revenues grew more quickly because some states rely on more cyclical forms of revenue, one would expect revenues to be slowest in those states that grew most quickly during the expansion. A more precise estimate of necessary re serves would be based on the cyclicality of the specific revenue sources relied on by each state. In a paper in 1998, Dye and McGuire provide estimates of revenue cyclicality by state, but do not esti mate required reserves. The correlation between the estimates of needed reserves and the cyclicality of revenues is -0.26. In other words, those states that CBPP calculate as needing the most re serves (as a percent of their budget) to withstand a recession are the states that Dye and McGuire find rely on least sensitive rev enue sources. 12For further information on the distinction between reported un liquidated obligations and unobligated funds, see Fazere, 2001. 13TANF will need to be reauthorized in 2002. Significant changes are not anticipated because the program has been widely perceived as being successful. 8There have been numerous changes relating to the eligibility of children since 1986. For a full discussion, see U.S. Congress, House Committee on Ways and Means, 2002. 22 3Q/2002, Economic Perspectives REFERENCES American Federation of Teachers, 2002, “Survey and analysis of teacher salary trends 2000,” February 28, available on the Internet at www.aft.org/research/ surveyOO/salarysurveyOO.pdf/. Associated Press, 2002, “Drugstores threaten to end Medicaid service,” New York, wire service, March 12. Cain, Charlie, Mark Hornbeck, and Gary Heinlein, 2002, “Critics argue Engler takes from future to fix budget,” Detroit News, available on the Inter net at http://detnews.corn/2002/politics/0202/08/ a01-411373.htm, February 8. Dye, Richard F., and Therese J. McGuire, 1998, “Block grants and the sensitivity of state revenues to recession,” in National Tax Association Proceedings of the Annual Conference on Taxation, Washington, DC, pp. 15-23. Executive Office of the President, The, Council of Economic Advisers, 2002, Economic Report of the President, Washington, DC: United States Govern ment Printing Office, available on the Internet at http ://w3. access.gpo.gov/eop/, F ebmary. Executive Office of the President, The, Office of Management and Budget, 2002, Historical Tables, Budget of the United States Government, Fiscal Year 2003, Washington, DC: U.S. Government Printing Office, available on the Internet at www.whitehouse.gov/omb/budget/fy2003/pdf/ hist.pdf/. Herszenhorn, David M., 2001, “McGreevey warns of tough fiscal future in New Jersey,” New York Times, November 8. Jenny, Nicholas W., 2002, “2001 tax and budget summary,” State Fiscal Briefs, Albany, NY: The Nelson A. Rockefeller Institute of Government, Fiscal Studies Program, No. 63, available on the Internet at www.rockinst.org/publications/ fiscal_studies/FB_63 .pdf, January. __________ , 2001, “Severe decline in state tax revenue,” State Revenue Report, Albany, NY: The Nelson A. Rockefeller Institute of Government, Fis cal Studies Program, No. 46, available on the Inter net at www.rockinst.org/publications/fiscal_studies/ RR_46.pdf, tax data from 1995-2001, December. Federal Reserve Bank of Chicago Jenny, Nicholas W., and Donald J. Boyd, 2001, “A second quarter of slow tax revenue growth,” State Revenue Report, Albany, NY: The Nelson A. Rockefeller Institute of Government, Fiscal Studies Program, No. 44, available on the Internet at www.rockinst.org/publications/fiscal_studies/ RR_44.pdf, data for 1994, June. Lav, Iris J., and Alan Berube, 1999, When it Rains it Pours: A Look at the Adequacy of State Rainy Day Funds and Budget Reserves, Washington, DC: Cen ter on Budget and Policy Priorities. Lazere, Ed, 2001, “Unspent TANF funds at the end of federal fiscal year 2000,” Washington, DC: Center on Budget and Policy Priorities, available on the In ternet at www.cbpp.org/l-22-01sfp00surplus.htm, February 23. Lex, Leo, 1998, “States’ use of surplus funds,” Con gressional Budget Office, report, available on the In ternet at www.cbo.gov/showdoc. cfm?index= 1044&sequence=0&from= 1, November. Mackey, Scott, 1999, “State tax actions 1995,” National Conference of State Legislatures, report, available on the Internet at www.ncsl.org/programs/ fiscal/STA95Pl.HTM/. McCallum, Scott, 2002, “State of Wisconsin budget reform bill summary,” Department of Administra tion, Division of Executive Budget and Finance, Madison, WI, available on the Internet at www.doa.state.wi.us/debf/sbo/state_budget/ 0103_exec_budget/pdf_files/budgetreformbill.pdf, January. McGranahan, Leslie, 1999, “State budgets and the business cycle: Implications for the federal balanced bud get amendment debate,” Economic Perspectives, Fed eral Reserve Bank of Chicago, Vol. 23, No. 3, pp. 2-17. Merrimam, David, 2000, “Sources of data about state government revenues and expenditures,” The Urban Institute, report, available on the Internet at www.urban.org/UploadedPDF/discussion00-04.PDF/. National Association of State Budget Officers, 2002, Budget Processes in the States, Washington, DC, January. 23 __________ , 2001a, Fiscal Survey of the States: June 2001, Washington, DC: National Governors Association and National Association of State Bud get Officers. Okamoto, Lynn, 2002, “Revised budget lists cuts for some agencies,” Des Moines Register, available on the Internet at www.dmregister.com/news/stories/ C4780934/17540924.html, March 7. __________ , 2001b, Fiscal Survey of the States: December 2001, Washington, DC: National Gover nors Association and National Association of State Budget Officers. State of Illinois, 2002, “Governor Ryan proposes balanced $52.8 billion 2003 budget,” Springfield, IL, press release, available on the Internet at www200.state.il.us/PressReleases/ ShowPressRelease.cfm?SubjectID=l 0&RecNum= 1684, February 20. __________ , 2001c, State Expenditure Report 2000, Washington, DC, available on the Internet at www.nasbo.org/Publications/PDFs/00exprpt.pdf/ . __________ , 1999, The Fiscal Survey of the States: December 1999, Washington, DC. National Conference of State Legislatures, 2002, “State fiscal update: April 2002”, Washington, DC, available on the Internet to members at www.ncsl.org/programs/fiscal/sfo2002.htm, April 16. __________ , 2001a, “State budget and tax actions 2001: Preliminary report,” Washington, DC, avail able on the Internet at www.ncsl.org/programs/fiscal/ presbtaO 1 .htm/. __________ , 2001b, “State fiscal outlook for FY 2002: November update,” Washington, DC, available on the Internet at www.ncsl.org/programs/fiscah sfo2002b.htm, December 3. __________ , 2001c, “State tax actions 2000: Execu tive summary,” Washington, DC, available on the In ternet at www.ncsl.org/programs/fiscal/ sta00sum.htm, January 31. __________ , 2000, “State budget and tax actions 2000: Preliminary report: Executive summary,” report, Washington, DC, available on the Internet at www.ncsl.org/programs/fiscal/presbtaOO.htm. __________ , 1999a, “State budget and tax actions 1999: Preliminary report,” Washington, DC, avail able on the Internet at www.ncsl.org/programs/fiscal/ presbta99.htm/. __________ , 1999b, “State tax actions 1996: Executive summary,” available on the Internet at www.ncsl.org/programs/fiscal/sta96sum.htm/. 24 __________ , 2000, “Ryan commends legislators for advancing education, human services, VentureTECH, tax relief, and public safety, Springfield, IL, press re lease, available on the Internet at www.state.il.us/ gov/press/OO/Apr/endssn.htm, April 15. Sterngold, James, 2002, “California officials face unprecedented budget gap,” New York Times, May 12. U.S. Congress, General Accounting Office, 1995, “School facilities: Condition of America’s schools,” letter report, No. GAO/HEHS-95-61, February 1. U.S. Congress, House Committee on Ways and Means, 2002, 2000 Green Book, Washington DC: U.S. Government Printing Office, No. 106-14, avail able on the Internet at www.access.gpo.gov/congress/ wmOO 1 .html/. __________ , 1996, 1996 Green Book, Washington, DC: U.S. Government Printing Office, No. 104-14. U.S. Department of Commerce, Bureau of the Census, 2002a, “State government finances,” Wash ington, DC, report, available on the Internet at www.census.gov/govs/www/state.html, data for 1992-99, February 27. __________ , 2002b, “State government tax collec tions,” Washington, DC, report, available on the In ternet at www.census.gov/govs/www/statetax.html, data for 1992-2000. __________ , 2002c, Statistical Abstract of the United States: 2001, 121st Edition, Washington, DC: Government Printing Office, available on the Inter net at www.census.gov/statab/www/. 3Q/2002, Economic Perspectives __________ , 1981, “State government finances in 1980,” Washington, DC, report, series GF80, No. 3. U.S. Department of Commerce, Bureau of Eco nomic Analysis, 2002a, “Current and ‘Real’ gross domestic product,” report, Washington, DC, avail able on the Internet at www.bea.doc.gov/bea/dn/ gdplev.xls. __________ , 2002b, National Income and Product Account, Washington, DC, table 6.1C. U.S. Department of Education, National Center for Education Statistics, 2001, “Public school stu dent, staff, and graduate counts by state, school year 1999-2000,” Washington, DC, available on the Internet at http://nces.ed.gov/pubsearch/ pubsinfor.asp?pubid=2001326R, May 18. U.S. Department of Health and Human Services, Health Care Financing Administration, 2000, A Profile ofMedicaid: Chart Book 2000, Washington, DC, available on the Internet at www.hcfa.gov/stats/ 2Tchartbk.pdf/. Federal Reserve Bank of Chicago U.S. Department of the Treasury, Internal Reve nue Service, 2001a, Individual Income Tax Returns: Selected Income and Tax Items for Selected Years, 1994-1998, in Current and Constant 1990 Dollars, IRS Publication 1304, Revised 4-2001, available on the Internet at www.irs.gov/pub/irs-soi/98intba.xls, federal capital gains data for 1994-98. __________ , 2001b, Individual Income Tax Returns: Selected Income and Tax Items for Specified Tax Years, 1980-1999, SOI Bulletin, Spring, available on the Internet atwww.irs.gov/pub/irs-soi/99in01si.xls, federal capital gains data for 1980-99. Wilson, Joy Johnson, 1999, “Summary of the Attor neys General master tobacco settlement agreement,” National Conference of State Legislatures, Washing ton, DC, report, available on the Internet at www.ncsl.org/statefed/tmsasumm.htm, March. Zahradnik, Bob, and Iris J. Lav, 2000, “When it rains it pours—One year later,” Center on Budget and Policy Priorities, report, available on the Internet at www.cbpp.org/5-22-00sfp.htm, May 22. 25 The 2001 recession and the Chicago Fed National Activity Index: Identifying business cycle turning points Charles L. Evans, Chin Te Liu, and Genevieve Pham-Kanter Introduction and summary On March 5, 2001, the Federal Reserve Bank of Chicago first released publicly the Chicago Fed Na tional Activity Index (CFNAI), a single, summary measure of real economic activity that is based on a weighted average of 85 economic indicators. This in augural CFNAI release explicitly mentioned the pos sibility that the U.S. economy had begun to slip into a recession. On November 26, 2001, the National Bureau of Economic Research’s (NBER) Business Cycle Dating Committee “determined that a peak in business activity had in fact occurred in the U.S. econ omy in March 2001 (NBER, 2001).’’ As the eightmonths lag of the NBER report indicates, business cycle turning points are typically only recognized many months after the event; thus, the ability of the CFNAI to identify the recession in approximately real time is important—since early recognition of business cycle turning points will enable more timely monetary pol icy responses. Although one of the first uses of the CFNAI was to gauge inflationary pressures (Fisher, 2000), there is a strong statistical relationship between this index of economic activity and business cycle movements. We can see this in figure 1, which displays the threemonth moving average index (CFNAI-MA3) from 1986 through 2001. Whenever the three-month mov ing average of this index falls into the range of -0.70 to -1.00, there is an increasing probability that the U.S. economy is in a recession. The substantial fall in the index to -1.50 in late 1990 corresponds to the 1990-91 recession. Similarly, the 2001 recession (see figures 2 and 3) is clearly evident as the index fell below -1.00. Prior to the current recession, there were five reces sions over the 1967-2000 period. In six cases, the CFNAI-MA3 fell below -0.70, after having previous ly been above zero when the economy was expanding. On five of these occasions, the U.S. economy had just entered a recession as determined later by the 26 NBER. Taken at face value, this is an 83 percent suc cess rate for the CFNAI. To further our understanding of the CFNAI and its role as a business cycle indicator, we address two main questions in this article. First, what is the general relationship between the CFNAI and economic reces sions? While economic downturns are clearly evident in the sharp reductions in the CFNAI, how much more information do we gain beyond what we would learn by simply focusing on single indicator measures of economic activity like industrial production, personal consumption expenditures, and others? We offer a graphical analysis of the data to answer this question. Second, what probabilistic statements about economic performance can we attach to specific values of the CFNAI-MA3? When the CFNAI-MA3 plunges to val ues below -0.70, what is the probability that the U.S. economy has entered a recession? We adopt a statistical approach to modeling the dynamic evolution of the 85 economic indicators in order to answer this question. To summarize our findings, our graphical analy sis indicates that individual economic indicators ap pear to predict the onset of economic recessions almost as well as the CFNAI-MA3. Indeed, many business cycle analysts prefer a relatively small number of eco nomic indicators to guide their analysis. For example, the NBER November 2001 committee report makes clear the importance of four monthly coincident eco nomic indicators of real activity; payroll employment, industrial production, real personal income less Charles L. Evans is a vice president and senior economist and Chin Te Liu is an associate economist at the Federal Reserve Bank of Chicago. Genevieve Pham-Kanter, formerly an associate economist at the Federal Reserve Bank of Chicago, is a graduate student at the University of Chicago. Throughout this project, the authors have benefited from the helpful comments of Larry Christiano, David Marshall, Ken Matheny, and Mark Watson. 3Q/2002, Economic Perspectives transfer payments, and manufacturing and trade sales in real terms.1 However, while all of the economic in dicators were signaling that the real economy was growing well below trend throughout this period, they conveyed different information about the timing of the business cycle peak. Essentially, the NBER selected the business peak based upon a peak in one very im portant indicator, total payroll employment. Visual inspection of the co-movements between industrial production, employment, and the CFNAI-MA3 sug gests that perhaps the gain in computing the index of 85 indicators is fairly small. However, the small number of economic reces sions since 1967 makes this assessment misleading. Using Monte Carlo simulations, a more careful eval uation of the statistical properties suggests substantial improvements from using the CFNAI-MA3 over in dividual indicators. For example, when the CFNAI-MA3 falls to -0.70, the probability that the economy has entered a recession is around 70 percent. When simi larly normalized three-month moving average index es of industrial production and personal consumption expenditures fall below -0.70, the probabilities are 50 percent and 35 percent, respectively. This quanti tative analysis indicates that the CFNAI-MA3 is use ful for detecting the onset of economic recessions. In the following sections, we explain the develop ment and construction of the CFNAI. Then, we exam ine how quickly and how well the index has historically identified business cycle turning points. We examine whether it is possible to accurately reflect the economy with fewer indicators or whether more than 85 indi cators may be necessary to average out the idiosyn cratic noise from the underlying economic signals. Finally, we use a statistical technique, Monte Carlo simulations, to analyze the index’s performance. This statistical approach provides us with a far greater num ber of observations than the five recessions that have actually occurred since 1967. In particular, we focus on the index’s ability to correctly identify the onset of recession in the simulated economy. Origins of the CFNAI The Chicago Fed National Activity Index is based upon an index designed by James Stock and Mark Watson in their Journal ofMonetary Economics article on “Forecasting Inflation’’ (Stock and Watson, 1999a). Stock and Watson’s Activity Index summarizes the information of 85 data series in a single index value. This is accomplished using the well-known method of principal component analysis (see box 1 for an ex planation). The current version of the CFNAI attempts to implement their monthly data selections as closely Federal Reserve Bank of Chicago as possible in real time. The CFNAI is the first prin cipal component of these data series, accounting for the largest independent variation among the econom ic indicators in the data set. Equivalently, the CFNAI is a weighted average of the 85 economic indicators. For example, a principal component index v based on three series can be expressed as yt = x xlf + w2 x x2( + w3 x x3f, where xp x2, and x, are the three original series, and w2, and w3 are the weights assigned to the data series. In practice, the weights measure the relative importance of each series in the index. An index like the CFNAI can be used in several ways. One approach is to use the index as an explan atory variable in estimating linear relationships. By computing a single index value for a large data set, the gains from data reduction allow the analyst to specify parsimonious forecasting relationships. This approach has been used by Stock and Watson (1999a, 1999b), Fisher, Liu, and Zhou (2002), and Bernanke and Boivin (2002). Another approach is to use the index to iden tify non-linear regime switches. For example, Fisher (2000) describes how movements in the activity in dex relate to broad accelerations in inflation during certain time periods. In this article, we focus on non linear regime switches from economic expansions to recessions, as in Hamilton (1989) and Diebold and Rudebusch (1996). Constructing the CFNAI One of Stock and Watson’s (1999a) findings is that the first principal component of our 85-variable dataset captures aggregate real activity in the United States. When we double the dataset by including var ious inflation rates, monetary aggregates, interest 27 BOX 1 Construction of the CFNAI Data transformations and the principal components method The CFNAI is the first principal component of a dataset consisting of 85 economic indicators. Back ground on the method of principal components may be found in most advanced statistics and economet rics books. Henri Theil’s (1971) classic text Princi ples of Econometrics provides an excellent overview of this method; we use Theil’s exposition and nota tion in the following discussion. Let x denote the 1 x 85 row vector consisting of observations at time t of the 85 data series. Let XT denote the T x 85 stacked matrix of data vectors, that is, XT = where T is the total number of observations. By this construction, each column of XTcontains T observations of an individual economic indicator. Each of the 85 raw series used to compute the CFNAI has already been inflation adjusted and, if necessary, seasonally adjusted by the original data provider. After obtaining these raw data series from HaverAnalytics, we first assess each series for its stationarity properties. If a series is determined to be non-stationary, we apply an appropriate transforma tion to render the series stationary. In most instanc es, the data are log-differenced so that the indicator series are transformed into growth rates. This is the case, for example, with employment and industrial production data. In some cases, such as the Institute for Supply Management’s Purchasing Managers In dexes, the data require no transformation. Second, each stationary series is adjusted for outlying observations. We define an outlier to be an observation whose distance away from the median is greater than six times the interquartile range of the series. That is, ,v.—the observation at time t of series i—is an outlier if |.vIf -xf°| >6(.v/5 -xf), where xf, xf, and xf are the 25th, 50th, and 75th percentiles of series x.. An outlier that is above the median has its original value replaced with xf + 6(.r’5 - xf), while an outlier that is below the median has its origi nal value replaced with xf - 6(xf - xf). Finally, we rescale each series to have a mean of zero and standard deviation of one. These standardized rates, commodity prices, and equities, the first princi pal component is essentially unchanged. That is, a measure of real activity continues to account for the largest independent, common variation in each of the 28 series are the indicator series used in XT for the prin cipal component calculation. In general, a principal component of XT is deter mined by a specific eigenvalue of the second-moment matrix XfXT. Computing the first principal component of XT requires calculating the eigenvector associated with the largest eigenvalue of XfXT. Consequently, since the CFNAI is the first principal component of X , it is simply a particular weighted average of the 85 economic indicators. In particular, CFNAIt = xfa, where a is an 85 x 1 vector of weights. Although the weights in the vector a correspond to the elements of the eigenvector associated with the largest eigen value of XfX-r , the vector a is re-scaled such the re sulting CFNAI has a mean of zero and standard deviation of one. Since we estimate a single set of weights over the entire sample period, this vector of weights remains fixed for a given set of data XT. Revisions to the CFNAI There are two main sources of revisions in the CFNAI. Firstly, because the CFNAI is designed to be released in a timely way and because indicator data are released at different times, not all of the indicators are available in time for a particular month’s CFNAI release. For example, employment data are usually available within one week of a month’s end, but in flation-adjusted retail inventory data are typically not available until another five weeks have elapsed. For any given CFNAI release in 2001, approxi mately one-third of the indicators will have had their latest monthly values forecast. In other words, the re ported CFNAI is based on the latest observed values for two-thirds of the 85 series and based on forecast values for the remaining one-third. In the following month’s CFNAI release, the data for the “lagging” series will have become available, and the previous month’s CFNAI value will be revised based on this data. In this way, forecast error is a source of revision in the CFNAI. Secondly, throughout the calendar year, the 85 monthly series are systematically revised by the origi nal reporting institutions. These revisions will alter the underlying monthly data, resulting in a change in the value of CFNAI. Although both sources of revision will also result in a change in the weighting vector a, we expect this and the re-normalization of the under lying data to have a negligible effect on the index. data series. So using 166 indicators rather than our 85 would lead to negligible changes in our threshold analysis below. Consequently, we focus on real eco nomic indicators in computing our index. 3Q/2002, Economic Perspectives The CFNAI is constructed from 85 coincident economic series that are drawn from five categories of economic activity. Table A3 in the appendix lists all 85 series. The five categories are: 1. Production and income—These data include in dustrial production growth for several industries and product classifications; component indexes from the Institute for Supply Management’s (ISM) Purchasing Managers Index (PMI); capacity utili zation measures; and real income growth mea sures (21 series). 2. Employment, unemployment, and labor hours— These data include employment growth rates for several industries from the Payroll Survey; the employment component of the PMI; changes in unemployment rates for several demographic groups from the Household Survey; initial claims for state unemployment insurance; growth rates of production hours; and changes in help-wanted measures (24 series). 3. Personal consumption and housing—These data include the growth rate of real personal consump tion expenditures for several categories; housing starts nationally and by region; building permits for new housing units; and shipments of mobile homes (13 series). 4. Manufacturing and trade sales—These data include growth rates of real sales measures for manufac turing industries; several categories of wholesale trade sales; and several categories of retail trade sales (11 series). 5. Inventories and orders—These data include com ponents of the PMI related to new orders and ven dor performance; the growth rate of inventories and inventory-sales ratios by manufacturing and trade categories; and the growth rate of new or ders for durable goods manufacturing and nonde fense capital goods (16 series). Basic properties of the CFNAI Figure 2 displays the monthly CFNAI from 1967 through 2001. By construction, the monthly index has an average value of zero and a standard deviation of one. Since many data series are deviations of growth rates from their sample average, the monthly index can be interpreted as the deviation of national activity growth from its trend rate. Consequently, an index value of zero is associated with trend rates of growth. Another reaction to figure 2 is that the index is quite volatile from month to month. Although in dex values above zero tend to stay above zero for a period of time, there are many reversals of sharp spikes from month to month. The monthly index, therefore, appears to track broad movements in the economy, but contains transitory noise. Consequently, taking a moving average of the monthly series would average out the transient noise while leaving the underlying signal in place. Figure 3 displays the trailing three-month moving average of the monthly index. We refer to this moving average as the CFNAI-MA3. Clearly, much of the transient noise in the monthly index has been filtered out. Now it is easier to see the persistent movements of the index over time. Since the index is a weighted average of 85 economic indicators, movements in each of the components contribute to movements in the CFNAI-MA3. Large positive or negative index values tend to arise when most of the individual indicators are moving together. This is especially evident during periods of economic contraction. In figure 3, sharply negative values of the CFNAI-MA3 correspond to of ficial NBER recessions. From 1967 through 2000, the moving average index fell below -0.70 after having Prior to constructing the activity index, the indi vidual data series are transformed to be stationary as denoted in table A3. In practice, this means that trend ing variables are often measured as growth rates, while variables without trends are often left untransformed. These transformed data are then each de-meaned and standardized to have a unit variance. We then com pute the CFNAI as the first principal component of the 85 data series. Box 1 presents the formal details of the methodology used to construct the index. Federal Reserve Bank of Chicago 29 previously been above zero on six occasions, and five of those were associated with recession. Perhaps it should not be surprising that a basket of economic indicators can provide a useful guide to the state of aggregate economic activity. In financial markets, individual stock prices reflect both market and company-specific risk. A portfolio of stocks, like the Standard & Poor’s 500, provides diversification of the idiosyncratic risks for individual stocks, leaving in place the undiversifiable market risk. Movements in the stock index provide indications of how the stock market is evolving. Similarly, the CFNAI-MA3 is a portfolio of economic indicators. Thus, movements in the CFNAI-MA3 are reflective of how the econo my is evolving. Over the period 1967 to 2000, the CFNAI-MA3 fell substantially whenever the U.S. economy was in a recession. Identifying business cycle turning points with the CFNAI The construction of the CFNAI highlights its prop erties as a coincident indicator of economic expansions and contractions. There are many ways to evaluate an indicator. Fisher, Liu, and Zhou (2002) examine how the CFNAI contributes to the out-of-sample explana tory power in linear models for forecasting inflation over the last 15 years. In this article, we focus on how quickly and how well the CFNAI aids in identifying business cycle turning points. Applying a simple thresh old criterion, we examine how accurately the histori cally constructed CFNAI would have identified past recessions and recoveries. 30 Identifying recessions During the period 1967-2000, there were five economic recessions, as identified by the NBER; these occurred in 1970, 1973-75, 1980, 1981-82, and 1990-91.2 Figure 3 shows the movements in the CFNAI-MA3 in the context of the NBER recession episodes, which are the shaded regions. As we men tioned earlier, figure 3 suggests that the CFNAI-MA3 may be a useful guide for identifying whether the economy has slipped into and out of a recession. Spe cifically, note that, in each of the five recessions, the smoothed CFNAI-MA3 fell below -0.70 (the dashed negative horizontal line) very near the onset of the recession. If we designate -0.70 as our recession thresh old, we see that during the 1970, 1981-82, and 199091 recessions, the index first fell below the threshold during the first month of the recession. During the 1973-75 and 1980 recessions, the index first fell below the -0.70 threshold during the third and second months of the recessions, respectively. Thus, during the period 1967-2000, the CFNAI-MA3, using the -0.70 thresh old, gave a signal of the economy being in a recession within the first three months of the recession. The -0.70 recession threshold generated one false alarm during the 1967-2000 period. Specifically, in July 1989, the CFNAI-MA3 fell to -0.94, but no re cession materialized. One explanation for the signifi cant dip in the CFNAI-MA3 is that, from mid-1988 through spring 1989, the Federal Open Market Com mittee pursued a contractionary monetary policy in an attempt to reduce inflation. This tight policy was re flected in an increase in the federal funds rate to 9.75 percent. During this time, the smoothed CFNAI exhib ited a steady decline, reaching its low level in July 1989, before returning above the -0.70 threshold. This brief analysis highlights some problems with using the CFNAI-MA3 and a simple threshold rule to identify recessions. In particular, having a thresh old low enough to prevent false signals of recessions will delay the date at which a true recession can be identified. The threshold value of -0.70 identified all five of the true recessions, but falsely signaled a sixth recession. Using a lower recession threshold of-1.50 would have eliminated the false alarm, but the true recessions would not have been identified until many months into the recession. Indeed, the 1970 recession would not have been identified until its twelfth month—the last month of this recession. Identifying historical recoveries The tension between identifying turning points early and minimizing the number of false signals also arises when we try to determine when the economy 3Q/2002, Economic Perspectives has successfully pulled out of a recession. In this case, we start with the rule that, when the CFNAI-MA3 first crosses the +0.20 threshold level from below, the reces sion has ended; this threshold is indicated in figure 3 by the dashed positive horizontal line. We see that, for four of the last five recessions, the CFNAI-MA3 crossed +0.20 from below within five months of the NBER-identified trough (official end of the recession). Following the 1970 recession, the smoothed index exceeded +0.20 two months after the trough. For the 1973-75, 1980, and 1981-82 re cessions, the threshold was crossed in the fifth, third, and fourth months, respectively, following the offi cial trough. For the 1990-91 recession, however, the smoothed CFNAI did not provide an early indication of the re cession’s end. Specifically, the CFNAI-MA3 crossed +0.20 in November 1993, even though the trough was retrospectively identified by the NBER as March 1991. In part, high levels of corporate debt and financial in stitutions’ reduced ability to extend new financing slowed the recovery from the 1990-91 recession. To mitigate the effects of these financial headwinds, the monetary policy response was to keep the federal funds rate at 3 percent until February 1994. The CFNAI signal was further delayed by the choppy nature of the recovery. The halting movements of the activity index, seen in figure 3, are consistent with contempo raneous economists’ accounts of double- and triple dips in economic activity during this period. Indeed, this recovery was so difficult to discern that the NBER only declared an end to this recession almost two years after the trough had passed. A more lax recovery threshold of +0.00, or return to trend growth, would have identified the end of the 1990-91 recession earlier. Had this threshold been in effect, the recovery would have been signaled in April 1992, or 19 months prior to the +0.20 threshold date. On the other hand, the weaker recovery threshold would also have generated false signals. In particular, a +0.00 threshold would have prematurely (by 11 months) signaled the end of the 1973-74 recession. Overall, then, the CFNAI-MA3 with a recovery threshold of +0.20 was able to identify all of the re coveries, signaling four out of the five recoveries within the first five months. Its identification of the erratic 1990-91 recovery, however, did not come until 32 months after the actual trough. The CFNAI during the 2001 recession Up until the 2001 recession, the evidence in favor of using the CFNAI as a barometer for detecting the onset of recessions was all historical. But with the Federal Reserve Bank of Chicago inaugural publication of the January 2001 CFNAI on March 5, 2001, the evaluation process moved from the sterile laboratory setting of a computer to a field test using real-time data. In the spring of 2000, the U.S. economy was con tinuing to expand at a rate that was above its potential growth rate. Second-quarter real gross domestic prod uct (GDP) growth was 5.7 percent. Monetary policy had shifted to a relatively tight stance. The federal funds rate began its initial increase in June 1999 from 4.75 percent to 6.50 percent in May 2000. By the sum mer of 2000, business analysts’ were expecting the economy to begin a transition from above-trend growth to a period of below-trend growth. Growth rates of industrial production turned negative beginning in July 2000, while other indicators began to cool notice ably. The January 2001 release of the CFNAI report ed that the CFNAI-MA3 had fallen below zero in July 2000. As zero represents the economy growing at trend, the index captured the transition that business ana lysts in the press had been discussing. Much of the initial drop in the CFNAI-MA3 comes from industrial production and the ISM Purchasing Managers Index data. These components provided strongly negative weight to the other index components that were more evenly spread around trend growth behavior. Figure 4 displays category indexes represent ing each of the five data categories of the CFNAI from 1986 through 2001. The category indexes are construct ed by summing only the weighted series in each re spective category. Each category index is then re-scaled to have a standard deviation of one. With this trans formation, if any of the category indexes captured all of the movements of the CFNAI, the two would lie on top of each other in figure 4. The production/income (panel A), employment (panel B), and inventory/orders (panel E) categories track the initial decline in economic activity pretty well for the second half of 2000. Manufacturing/trade (panel D) captured some of this decline, while the con sumer (panel C) category did not fall below zero at all in 2000. Once the recession began in March 2001, according to the NBER business cycle dating com mittee, the categories began to diverge to a greater extent. The production/income and inventory/orders categories moved with the total CFNAI-MA3 during this period. The employment category, however, fell much more sharply, particularly beginning in March 2001. This latter observation is not surprising in the context of the NBER’s announcements regarding the selection of March 2001 as the most recent business cycle peak. The dating committee mentioned that movements in payroll employment were decisive in 31 FIGURE 4 CFNAI category indexes—MA3 and business cycles (1967-2001) Notes: Shaded areas indicate NBER recessions. The black line indicates the CFNAI-MA3, and the colored line indicates the respective category index. picking the date, and very important in the overall de termination that the economy had entered recession. Within the full index, figure 4 displays some de gree of heterogeneity among the category indexes during the recession of 2001. Manufacturing/trade fell less than the overall index. And the consumer catego ry hardly registered any negative values. Simply us ing the consumer category as a proxy for the CFNAI would clearly result in different inferences. The pro duction and employment categories move much more 32 strongly with the full index, although there are peri odic differences in magnitude. Is there value in diversifying the basket of economic indicators? The previous discussion raises the issue: How many indicators are necessary to provide an accurate description of the state of economic activity? Does it take a large number of economic indicators to filter out the idiosyncratic noise, or can a single, favorite 3Q/2002, Economic Perspectives indicator do the trick? Most analysts’ first approach to answering this question would involve producing large numbers of graphs and staring. Figure 5 (on page 34) provides an abbreviated tour of the data. Figure 5 displays graphs of several baskets of economic indicators, as well as individual indicators. Each panel graphs a three-month moving average of the indicator against the CFNAI-MA3. We consider two questions here: 1) How does the individual indi cator compare with the CFNAI-MA3?; and 2) How well does the individual indicator perform at detect ing recessions and expansions using thresholds like -0.70 and +0.20? Are 85 real indicators enough? The CFNAI con tains only 85 indicators. Perhaps worse still, none of these are financial indicators that have a proven track record of periodically signaling turning points. An al ternative approach would construct an economic index using a larger set of data that include financial, mon etary, and price variables. Figure 5, panel A displays an analogous index employing 166 economic, finan cial, monetary, and price indicators. The striking fea ture of panel A is that the two measures are nearly identical. Apparently, the first principal component of this larger dataset is essentially an activity index. This observation was originally made by Stock and Watson (1999a). Do we really need 85 indicators? In our earlier analysis of the 2001 recession, we found that subcom ponents of the CFNAI performed reasonably well in tracking the economic downturn. Perhaps a smaller index performs just as well as the CFNAI. Figure 5, panels B and C display the production and employ ment category indexes, respectively. The production index follows the CFNAI-MA3 quite closely, and there appear to be few differences in inference about busi ness cycle turning points. Both suffer from the false recession signal in July 1989. The production index has an additional false positive prior to the onset of the 1980 recession. The CFNAI-MA3 was in nega tive territory in 1979 prior to the recession, but it did not cross the -0.70 threshold until 1980. Following the 1990-91 recession, the production index signals the end of the recession much sooner than the full in dex. However, the slow expansion in 1991-92 makes this virtue a bit hollow. In 1991, the production index has a close brush with calling a second recession. The CFNAI-MA3 moved lower in 1991 than the produc tion index; but since the index had not determined the end of the recession, it would not be a second re cession. In both cases, the 1990-92 period is a diffi cult one. The employment component in figure 5, Federal Reserve Bank of Chicago panel C also performs reasonably well. On the plus side, this component did not falsely signal a recession in July 1989. On the downside, it did not capture the 1973-75 recession until late 1974, almost at its end. In summary, smaller component indexes may perform about as well as the CFNAI, but more experience is required to sort this out. Many individual indicators provide false signals. As the focus narrows to individual economic indica tors, it is not surprising that many series provide false readings on the state of the aggregate economy. The discussion of the consumption and manufacturing/ trade categories of the CFNAI suggests that many of these data are poor candidates by themselves. Figure 5, panels D, E, and F display real retail sales growth, housing starts, and the PMI New Orders Index, re spectively. Each series has been transformed to be mean zero and unit standard deviation and is a threemonth moving average. Retail sales growth is quite volatile and often falls below the -0.70 threshold when the economy is not in recession. Housing starts tend to be low during most recessions. The slow recoveries following the 1973-75 and 1990-91 recessions sug gest that this indicator does not provide a quick indi cation of economic recovery. In addition, there were false recession warnings in 1967 and 1996; and the 2001 recession has been missed completely. Similarly, the PMI New Orders Index captures the five reces sions prior to 2001, as well as the current one. How ever, there are several false warnings: 1967, 1996, and 1998. Many single economic indicators contain transient fluctuations that are not related to the state of the economy. Some individual indicators do pretty well. Be cause the CFNAI gives substantial weight to data on industrial production and employment, it may be the case that single indicators in this category provide similar information to the CFNAI. Figure 5, panels G and H display growth rates of industrial produc tion and private payroll employment. Similar to the CFNAI category measures, these indicators do pretty well. For both of them, recessions are periods when industrial production and employment are low and below -1.00. Using a recession threshold of -0.70 for industrial production admits a couple of false reces sion warnings, but the performance improves if the lower threshold is decreased further to -1.00. Em ployment seems to do better than industrial produc tion. This may be because the NBER has tended to focus the recession determination on employment data more than on industrial production, at least in recent years. 33 FIGURE 5 CFNAI-MA3 versus select economic indicators, three-month moving average (1967-2001) E. Housing starts B. Production and Income Category Index F. PMI: Manufacturing New Orders Diffusion Index C. Employment, Unemployment, and Labor Hours Category Index G. Industrial production D. Real retail sales Note: Shaded areas indicate NBER recessions. The black line ii respective economic indicator. 34 the CFNAI-MA3, and the colored line indicates the 3Q/2002, Economic Perspectives Summary of analysis To sum up, our visual inspection of individual data series suggests that the CFNAI-MA3 does not perform appreciably better than the workhorse NBER coincident economic indicators like industrial produc tion and private payroll employment. But an essential question is: How much of this has been the result of data-mining from the small number of recessions un der examination? The idiosyncratic statistical noise in individual data series may have simply been small enough over this period to make a couple of data series work. In pooling 85 economic indicators for the CFNAI, the method is purchasing an insurance policy against statistical noise. And just like a home insurance poli cy, the mere fact that a house hasn’t burned down in the ten years that it has been insured does not mean that the insurance was unnecessary. To address this issue over a longer period, we now turn to simulation results of an empirically relevant statistical model. Statistical approach and Monte Carlo simulations The preceding historical analysis provides only a limited assessment of the CFNAI because the five-recession sample during the 1967-2000 period is small in statistical terms. For this reason, we have developed and analyzed Monte Carlo simulations of a business cycle index, 85 economic indicators, and the activity index to assess patterns based on a larger number of simulated observations. The challenge is to estimate an empirically rele vant set of business cycles and 85 equations for the economic indicators. For the business cycle indicator, we adopt a nonlinear Markov-switching process de veloped by Hamilton (1989). This model states that real activity transitions exogenously between expan sionary and contractionary rates of growth, while cap turing the historical average duration of business cycle expansions and recessions. Diebold and Rudebusch (1996) have also studied a system like this. For the economic indicator equations, we follow the unob served component model studied by Stock and Watson (1989). This specification states that each indicator is related to the business cycle index but also is contami nated by independent statistical noise. This captures the idea that each of the indicators has an idiosyncratic component that is not related to aggregate activity. Precise details on these specifications and the estima tion strategy are reported in the appendix. Given estimates of our statistical model from the 1967-2000 period, we conduct Monte Carlo sim ulations for the economy over a period of approximate ly 2,000 years. During this period, the nonlinear Federal Reserve Bank of Chicago Markov-switching model generates 404 recessions. We can use the simulated data for the 85 economic indicators to compute a CFNAI index over the 2,000year period. With these data, we can repeat the exer cise of using the CFNAI to decide if the economy has entered a recession. Specifically, for any given recession threshold of -r, we calculate whether the CFNAI-MA3 indicates the economy is in recession. The procedure works as follows. 1. Begin with the economy in an expansionary state. 2. If the CFNAI-MA3 falls below -r from above, then the economy is in a “CFNAI recession.’’ 3. The CFNAI recession continues until the CFNAIMA3 rises above +0.20, and then the economy is in a “CFNAI expansion.’’ Repeating steps 1 through 3 until the data sample is exhausted provides a long time-series of business cycle dates according to the CFNAI-MA3 criterion. The advantage of the laboratory environment is that the experimental design allows us to know at any date whether the true state of the simulated economy is expansion or contraction. We can tabulate what per centage of the time a CFNAI recession is in fact a true recession, and also what percentage of recessions are missed by the CFNAI using a threshold of -r. For ex ample, using a threshold of -0.70, the CFNAI-MA3 criterion determined that 394 recessions occurred during the 2,000 years of simulations. On the date that the recession call occurred, the true state of the business cycle was a recession in only 285 of the 394 recession calls. At this threshold, the frequency of success in calling a recession was 72 percent. This frequency can also be thought of as the probability that a “recession call” is correct. Notice that 119 re cessions were missed (404 true recession minus 285 correct recessions). Therefore, the frequency of fail ing to call a recession was 29 percent when the thresh old criterion insisted that the moving average index fall below -0.70.3 These are cases where the true econ omy was in recession, but the severity of the down turn was relatively modest.4 Before turning to the overall simulation results, notice that we can also compute the success of other indicators in calling recessions. We consider three additional gauges. First, given the visual success of the industrial production three-month moving index, we have tabulated the success frequencies for a mea sure we call IP3. Second, given the visual failure for consumption measures, we tabulated success frequen cies based upon a real personal consumption expen diture measure, referred to as CON3. Third, in order to assess the overall accuracy of the CFNAI-MA3 35 measure, we tabulated a success frequency for an un observed measure of the business cycle that is com mon to all 85 economic indicators. By construction, this measure has no indicator-specific idiosyncratic noise. In some sense, this is a virtually ideal indicator of the business cycle. We refer to this as ZSIM3, and it corresponds to observing z directly (as defined in the appendix). Table 1 reports the simulation frequency results for all four moving average indexes, using thresholds from -0.70 to -2.20. Recall that during the 2,000-year period of the exercise, the simulated business cycle generates 404 “true” recessions. First, consider the results for the CFNAI-MA3. As we mentioned above, with a threshold of -0.70, the probability that the economy has moved into re cession is 0.72. Apractical application of this can be seen from the March 2001 release of the CFNAI. As reported on May 31, 2001, the March 2001 CFNAIMA3 was -0.80. According to the simulation results here, that corresponds to approximately a 75 percent probability that the economy was in recession. In fact, the NBER reported in November that the economy entered a recession in March, but the Business Cycle Dating Committee also mentioned that without the ter rorist attacks of September 11, the economy might not have gone into recession. Our calculations indi cate a relatively high likelihood, three out of four chances, that the economy was in recession prior to the terrorist attacks. Next, notice that as the threshold for calling a recession becomes more stringent, the prob ability of making a false recession call becomes less likely. The CFNAI releases during 2001 pointed out that every economic recession since 1967 had breached the -1.50 level, and most had declined much more. The Monte Carlo simulations attach a 0.95 probability to a threshold of-1.50, which seems consistent with these observations. Not surprisingly, this increased reliability comes at a higher cost. As the threshold tightens, more true recessions are missed because they are not sufficiently deep. Taken to an extreme, a very large, negative threshold would likely guarantee that recessions as deep as the Great Depression would be captured, but perhaps at the cost of missing large re cessions such as 1973-75 or 1981-82. The simulation results indicate that the CFNAI MA3 filters out almost all of the idiosyncratic noise from the individual 85 economic indicators. Specifi cally, if the aggregate indicator z were directly ob servable, then the ZSIM3 measure could be constructed. At a threshold of -0.70, whenever ZSIM3 crossed this threshold, there would be a 74 percent probabili ty that the economy was in recession, compared with the CFNAI-MA3 probability of 72 percent. Across the range of thresholds considered, these differences are essentially negligible. Looking at the performance of IP3, however, the differences appear to be more substantial at first TABLE 1 Simulation frequency results for all four moving average indexes Probability that recession call is correct Probability that recession is missed ZSIM3 CFNAI-MA3 IP3 0.35 0.27 0.29 0.23 0.26 0.38 0.27 0.30 0.25 0.37 0.27 0.30 0.25 0.48 0.28 0.29 0.29 0.57 0.30 0.30 0.34 0.67 0.57 0.33 0.33 0.39 0.77 0.81 0.57 0.37 0.36 0.45 0.85 0.94 0.84 0.57 0.40 0.40 0.53 0.90 0.96 0.95 0.89 0.65 0.43 0.45 0.57 0.93 0.97 0.96 0.94 0.70 0.49 0.50 0.65 0.96 -1.70 0.98 0.97 0.96 0.67 0.56 0.54 0.71 0.98 -1.80 0.99 0.99 0.98 0.62 0.60 0.60 0.78 0.99 -1.90 0.99 0.99 0.99 N/A 0.66 0.65 0.83 N/A -2.00 0.99 0.99 0.98 N/A 0.72 0.71 0.87 N/A -2.10 0.99 0.99 0.97 N/A 0.78 0.76 0.92 N/A -2.20 0.99 0.99 0.96 N/A 0.81 0.80 0.94 N/A ZSIM3 CFNAI-MA3 -0.70 0.74 0.72 0.51 -0.80 0.78 0.76 0.58 -0.90 0.83 0.81 0.67 0.44 -1.00 0.86 0.85 0.70 0.46 -1.10 0.89 0.88 0.73 0.55 -1.20 0.92 0.92 0.77 -1.30 0.93 0.94 -1.40 0.94 -1.50 -1.60 Threshold IP3 C0N3 C0N3 Note: N/A indicates not applicable. 36 3Q/2002, Economic Perspectives glance. At a threshold of -0.70, the IP3 measure has a recession success proba bility of only 51 percent. However, in dustrial production exhibits deeper reductions during recessions, so a lower threshold may be more accurate. In fact, at -1.00, the IP3 success rate is 70 per cent, with only a missed probability of 29 percent. This is essentially the same performance as the CFNAI-MA3 at the -0.70 threshold. A cleaner comparison of these per formances is to graph the probability trade-offs for each indicator on the same graph. For each indicators’ threshold, fig ure 6 plots a pair of probabilities: the prob ability that “a recession was called” when no recession occurred versus the proba bility that “a recession occurred” but wasn’t called. The most efficient indica tors will minimize both of these proba bilities and exhibit a probability frontier, which is concentrated in the southwest portion of the figure. Figure 6 clearly displays two useful properties of the CFNAI-MA3 as a business cycle indicator. First, the performance of CFNAI-MA3 closely follows the performance of the unobservable index ZSIM3. By using the CFNAI to filter out idiosyncratic noise in the 85 economic indicators, we lose little by not ob serving the z indicator. Second, the IP3 indicator only performs better than the CFNAI-MA3 at levels of false positives that most analysts would deem unac ceptable. As long as the probability of successfully calling a recession exceeds 65 percent, the CFNAIMA3 and ZSIM3 provide lower rates of failing to call a recession than the IP3 index. Finally, focusing on consumption indicators alone is unlikely to pro vide useful indicators of business cycle turning points. The CON3 frontier is dominated by the other indica tors for all reasonable probabilities. This is not really surprising, considering that household spending indi cators do not always turn down dramatically during recessions. During the 2001 recession, which is not included in the simulated business cycle analysis here, the economy continued to experience strong growth in consumer spending during much of the downturn. Federal Reserve Bank of Chicago Conclusion The Chicago Fed National Activity Index was launched on March 5, 2001, and was promptly tested by the deceleration of U.S. economic growth that be gan in the summer of 2000. Throughout this period, the CFNAI release discussed how the low index val ues below -0.70 had previously been associated with economic recessions, and each successive month in 2001 was showing an increasing probability that the economy was already in recession. The NBER later declared that the economy officially entered recession in March 2001. The statistical analysis presented in this article indicates that the March 2001 CFNAI-MA3 of -0.80 was associated with a 75 percent probability that the economy was in recession at that time. The March CFNAI was released on May 2, 2001, com pared with the NBER report in November 2001. Con sequently, the real-time experience with the index and the statistical analysis here seem consistent with the view that the Chicago Fed National Activity In dex is a good early warning indicator of current eco nomic conditions. 37 NOTES 'See NBER (2001). 'More specifically, the NBER recession periods were December 1969-November 1970. November 1973-March 1975. January 1980-July 1980. July 1981-November 1982. and July 1990March 1991. According to the NBER definition of contraction (recession), the first date of each recession period indicates the peak of a business cycle, and the second date indicates the subse quent trough of the cycle. 4The Monte Carlo simulations of the Hamilton-Markov switching process include a restriction that recessions and expansions can be no shorter than six months. The stochastic process described in the appendix has the property that the economy could shift to recession for a single month. Although this is not likely, short re cessions do occur over a 2.000-year simulation. In the simula tions. if the economy shifted from one state to another before six months had elapsed, we assumed that the shift happened at the six-month mark. This had no noticeable effect on the average du ration of recessions in the simulations. 'Notice that given knowledge of the total number of true recessions. 404. the two percentages are sufficient to recover the number of “recessions called.” “correctly called recessions.” and “recessions not called.” APPENDIX: MONTE CARLO SIMULATION METHODS The Monte Carlo simulation consists of four parts. First, we specify two models: a model of the business cycle index and a model of the relationship between the (un observed) business cycle index and the observable parts of the economy. Second, we estimate the parameters re quired for the models using data from historical indica tors. Third, using the estimated parameters, we simulate the unobservable business cycle index and the 85 observ able series. Finally, we compute the simulated CFNAI index as a weighted average of these 85 simulated series, and we evaluate the ability of the activity index to sig nal turning points in the simulated business cycle index. S G {0,1} is the binary random variable identifying the state/regime, where 5}= 0 designates an expansion and S = 1 designates a recession, and the switching between expansion and recession states is determined by Markov transition probabil ities; and Z is a noise term that follows an AR( 1) process, namely 2) where Model of the business cycle and observable series Heuristically, a single index of the business cycle indicates whether the aggregate economy is expanding or contracting. Because of the complexities of large, dynamic, and decentralized economies, individual eco nomic indicators will be correlated with this latent in dex of the business cycle, but are measured with idiosyncratic noise. We use the Hamilton regime-switching model (Hamilton, 1989, and Diebold and Rudebusch, 1996) to formalize the behavior of the business cycle index. In particular, we express unobservable real activity as: 1)' z t= Lt• exp+Wr recessS+z, t i ? = cp?_1 + yt, v( - )V(0,cr), is independent and identically distributed (i.i.d.). £ is a deviation of current growth from trend growth, which is independent of the business cycle state .S'. In this way, latent economic activity is modeled as a non-linear process with cyclical noise. We model the observable variables ,v to be noisy measures of zt- In particular, we specify that each of the 85 observable variables is determined by: 3) A, =Y,Z, where ,v.It is the observable indicator,7 where 1 < i < 85,7 and ii.t is a noise term that follows an AR( 1) process: where Z is the (unobservable) growth in economic activity; 4) pi is the growth of economic activity during expansions; Similar specifications have been studied in many aggregate time series studies; see Stock and Watson (1989) for an extended example and additional refer ences. Combining equations 1 and 3, we have: li + Lt is the growth of economic activity during recessions; 5) u = du + e, where e - A(0, ) is i.i.d. A, =YMxP+Rra™5',) + w,,, where 38 3Q/2002, Economic Perspectives TABLE A1 Estimated business cycle parameters Aexp = 4-4% = -1-6% Pr{S, = 0 | S,_2 = 0} = 0.98 Pr{S; = 11 S(_± = 1} = 0.91 (p = 0.94 6) = 0.653 w’,7 = Y, I, + “«• It is important to recognize that S, i, and u are mu tually independent. Similar assumptions are typically employed in the empirical literature, as in Hamilton (1989), Diebold and Rudebusch (1996), and Stock and Watson (1989). months and 5 = 1 during recession months.1 According to equation 5, this regression gives us y; and wy for each of the i indicators. To estimate the remaining parameters, we make an additional assumption that allows us to use a simple method of moments estimator. The additional assump tion is that there are two “instrument” indicators that reflect the business cycle to the same degree. Specifi cally, let x, and x, be as follows: 7) Ji, = YN, +«,, =Y,(RexP +R««sA) + wi, > and 8) X2, = Y2Z, + «2, = Y2 (Rexp + ) + W2, , with the restriction that Estimation of parameters To provide empirically interesting simulation ex periments, we need estimates of the model parameters li , li , the Markov transition probabilities, cp, cr , y., d., and crV for each indicator ,v.z. *z’ z’ We estimate lx and lx from the growth rates of real GDP from 1947 to 2000. Specifically, we compute the growth rate of real GDP during each expansionary period and each recessionary period (as identified by the NBER). We then set the parameter p equal to the aver age GDP growth rate over all expansions, and lx + p equal to the average growth rate over all recessions. Table A1 lists the estimated values of these parameters. The transition probabilities are calibrated so that they correspond to the average length of historical expansions and recessions. In particular, we use the re lations that the duration of an expansion = 1/(1 - proba bility of remaining in an expansion), and the duration of a recession = 1/(1- probability of remaining in a recession). That is, the mean duration of an expansion = 1/(1 - Pr{.S'( = 01 St= 0}), and the mean duration of a recession = l/(l-Pr{S, =11 S,_, =1}). Based upon the NBER business cycle dates for 1947-2000, the mean duration of an expansion has been 50 months and the mean duration of a recession has been 12 months. The transition probabilities based on these calculations are also shown in table A1. To find estimates of y., we regress each of the 85 actual indicators on pexp + pm.„s5f for the period 1967— 2000, where .S' is the binary variable indicating expan sionary and recessionary periods. This regression is well specified, given the independence of 5, c, and u. For values of S, we note that Hamilton (1989) finds that his Kalman-smoothed inferences of the latent business cycle index correspond reasonably closely with the NBER business cycle dates. Let 5 = 0 during expansion Federal Reserve Bank of Chicago 9) Y, = Y, = Y- As noted earlier, the disturbance terms w and w, are independently distributed. This restriction is most likely to be satisfied when two indicators attempt to measure the same economic phenomenon but are from different source data. An example of this is in Prescott (1986), when he proposes a probability model of mea sured employment hours based upon the establishment survey and the household survey. Another example is where the two indicators have a similar relationship to the business cycle index, but are measured with differ ent levels of precision. For example, equation 9 might hold even when one indicator was a measure of output and another was a measure of input. In any event, the two instrument indicators used in this study are Private Payroll Employment growth from the BLS and the In stitute for Supply Management’s Purchasing Managers Index of New Orders. The source data for these two se ries are clearly independent. In addition, the restriction in equation 9 is not rejected by the parameter estimates for these two indicators. We use the restriction in equation 9 to estimate cp and cr, and (for each indicator) d. and <r. Using equation 6, we first note that: 10) w;.,- — w2f = Y2 . y2 To find an estimate of d., we use equations 4, 6, and 10 and find that: 11) £[(wfl- —w21)wfl] = £[z/2] = cl2 ; Y2 39 TABLE A2 Estimated parameters of selected indicator equations Indicator PMI, new orders Employment, private Employment, nonagriculture Employment, goods industry Unemployment rate Help-wanted ads PMI, production Industrial production IP, manufacturing Capacity utilization, manufacturing Personal income, transfers Housing permits Housing starts, Midwest Housing starts, West Manufacturing and trade sales Retail sales (real) Personal consumption expenditures New orders, construction, and materials Manufacturing and trade, inventory/sales R2 4 Y/ 0.282 0.272 0.268 0.278 -0.232 0.197 0.288 0.229 0.236 0.221 0.197 0.256 0.213 0.253 0.153 0.081 0.092 0.130 -0.117 0.715 0.297 -0.244 -0.190 -0.436 -0.431 0.724 -0.184 -0.243 -0.195 -0.128 0.915 0.765 0.839 -0.392 -0.271 -0.312 -0.250 -0.238 0.399 0.578 0.601 0.566 0.686 0.731 0.372 0.725 0.701 0.726 0.837 0.273 0.500 0.383 0.841 0.949 0.930 0.909 0.933 0.673 0.636 0.619 0.670 0.426 0.341 0.709 0.458 0.479 0.451 0.296 0.542 0.403 0.494 0.169 0.036 0.050 0.128 0.083 Note: The variables being measured in this table are derived from equation 3—x.f = YjZt+ujt—and equation 4—-uj=d.ujtl+ ejt—where e/f ~ MO, a2). 12) From equations 15 and 16, we see that: - — w2,= diE[u;t_i\ = dtf . Y2 From equations 11 and 12, it is clear that: E[(wi(-Ti-W2()w.( J 13) d,=----------- . E[(h;7 - —h2,)h',,] y2 Because we have estimates of y and w., we can use equation 13 to find d . 1 2 To find estimates of CT" , we note that equation 4 implies the following relationship between the varianc es of e and 14) o2=(l-d2)o2. Since o2 can be estimated using equation 11, we can easily obtain estimates of o2 using equation 14. Similarly, we obtain an estimate of cp by noting that equations 2 and 6 imply: 17) ■E[w„w2,-,] To obtain an estimate of cr, we note that equation 2 implies: 18) a2=(l-cp2)a2. where cr; = —£[w’„w’,,] r Finally, in estimating parameters for a particular .v , the choice of instrument indicator (,v,f) will depend on whether ,v.It is from the ISM data release or not. For example, if .v corresponds to a component of the Pur chasing Managers Index, then ,v,f will be payroll employ ment; otherwise, it will be the PMI New Orders Index. Table A2 presents a partial listing of the indicator param eter estimates. Table A3 lists all of the component data series in the CFNAI. 'NBER recessions are designated from peak to trough. For these exercises, we consider the peak and trough months as part of the recession period. 15) £[hi,h2,] = y2Oj and 16) E[w-1,u-2,_1] = Y2cpcr2. 40 3Q/2002, Economic Perspectives TABLE A3 CFNAI component data series Production and income (21 series) CUMFG DLV Capacity utilization: Manufacturing SA, percent of capacity IP DLN Industrial Production Index SA, 1992=100 DLN IP51 Industrial Production: Consumer goods SA, 1992=100 DLN IP5102 Industrial Production: Durable consumer goods SA, 1992=100 DLN Industrial Production: Nondurable consumer goods SA, 1992=100 IP51021 DLN IP52001 Industrial Production: Business equipment SA, 1992=100 DLN IP53 Industrial Production: Materials SA, 1992=100 DLN Industrial Production: Materials: Durable SA, 1992=100 IP53011 DLN IP53017 Industrial Production: Materials: Nondurable SA, 1992=100 DLN IP54 Industrial Production: Intermediate products SA, 1992=100 DLN IPDG Industrial Production: Durable manufacturing SA, 1992=100 DLN IPFP Industrial Production: Final products SA, 1992=100 DLN IPMFG Industrial Production: Manufacturing SA, 1992=100 IPMIN DLN Industrial Production: Mining SA, 1992=100 IPND DLN Industrial Production: Nondurable manufacturing SA, 1992=100 DLN IPTP Industrial Production: Products SA, 1992=100 DLN IPUTI Industrial Production: Utilities SA, 1992=100 NAPMC LV Institute for Supply Management: Manufacturing: Composite Index SA, percent NAPMOI Institute for Supply Management: Manufacturing: Diffusion Index, Production SA, percent LV YPDHM DLN Disposable personal income SAAR, billions of chained 1996$ DLN YPLTPMH Real personal income less transfer payments SAAR, billions of chained 1996$ Employment, unemployment and labor hours (24 series) A0M005 DLV Weekly initial claims for unemployment insurance SA, thousands LACONSA DLN All employees: Construction SA, thousands LADURGA DLN All employees: Durable goods manufacturing SA, thousands LAFIREA DLN All employees: Finance, insurance, and real estate SA, thousands LAGOODA DLN All employees: Goods-producing industries SA, thousands LAGOVTA DLN All employees: Government SA, thousands LAMANUA DLN All employees: Manufacturing SA, thousands LAMINGA DLN All employees: Mining SA, thousands LANAGRA DLN Employees on nonfarm payrolls SA, thousands LANDURA DLN All employees: Nondurable goods manufacturing SA, thousands LAPRIVA DLN All employees: Private nonfarm payrolls SA, thousands LARTRDA+LAWTRDA DLN All employees: Retail and wholesale trade SA, thousands LASERPA DLN All employees: Service-producing industries SA, thousands LASRVSA DLN All employees: Services SA, thousands LATPUTA DLN All employees: Transportation and public utilities SA, thousands LE DLN Civilian employment: Sixteen years & over SA, thousands LENA DLN Civilian employment: Nonagricultural industries SA, thousands LHELP DLN Index of help-wanted advertising in newspapers SA, 1987=100 LHELPR DLN Ratio: Help-wanted advertising in newspapers/number unemployed SA LOMANUA DLV Average weekly overtime hours: Manufacturing SA, hours LR DLV Civilian unemployment rate SA, percent LRM25 DLV Civilian unemployment rate: Men, 25-54 years SA, percent LRMANUA DLV Average weekly hours: Manufacturing SA, hours NAPMEI LV Institute for Supply Management: Manufacturing: Diffusion Index, Employment SA, percent Personal consumption CBHM CDBHM CDMNHM CNBHM C0ND09 CSBHM HPT HSM HST HSTMW HSTNE HSTS HSTW and housing (13 series) DLN Personal consumption expenditures SAAR, billions of chained 1996$ DLN Personal consumption expenditures: Durable goods SAAR, billions of chained 1996$ DLN Personal consumption expenditures: New autos SAAR, millions of chained 1996$ DLN Personal consumption expenditures: Nondurable goods SAAR, billions of chained 1996$ LN Construction contracts, millions of square feet DLN Personal consumption expenditures: Services SAAR, billions of chained 1996$ LN Housing units authorized by building permit SAAR, thousands of units LN Manufacturers’ shipment of mobile homes SAAR, thousands of units LN Housing starts SAAR, thousands of units LN Housing starts: Midwest SAAR, thousands of units LN Housing starts: Northeast SAAR, thousands of units LN Housing starts: South SAAR, thousands of units LN Housing starts: West SAAR, thousands of units Federal Reserve Bank of Chicago 41 TABLE A3 (continued) CFNAI component data series Manufacturing and trade sales (11 series) NAPMVDI LV Institute for Supply Management: Manufacturing: Diffusion Index, Vendor Deliveries SA, percent RSDH DLN Real retail sales: Durable goods SA, millions of chained 1996$ RSH DLN Real retail sales SA, millions of chained 1996$ RSNH DLN Real retail sales: Nondurable goods SA, millions of chained 1996$ TSMDH DLN Sales: Manufacturing: Durable Goods SA, millions of chained 1996$ TSMH DLN Sales: Manufacturing SA, millions of chained 1996$ TSMNH DLN Sales: Manufacturing: Nondurable goods SA, millions of chained 1996$ TSTH DLN Real manufacturing and trade: Sales SA, millions of chained 1996$ TSWDH DLN Sales: Wholesale: Durable goods SA, millions of chained 1996$ TSWH DLN Sales: Merchant wholesalers SA, millions of chained 1996$ TSWNH DLN Sales: Wholesale: Nondurable goods SA, millions of chained 1996$ Inventories and orders (16 series) A0M007 DLN Real manufacturers' new orders: Durable goods industries, billions of chained 1996$ A0M008 DLN Real manufacturers' new orders: Consumer goods & materials SA, millions of 1996$ A0M020 DLN Contracts and orders for plant and equipment, billions of chained 1996$ A0M027 DLN Real manufacturers' new orders: Nondefense capital goods industries SA, millions of 1996$ NAPMII LV Institute for Supply Management: Manufacturing: Diffusion Index, Inventory SA, percent NAPMNI LV Institute for Supply Management: Manufacturing: Diffusion Index, New orders SA, percent TIMDH DLN Inventories: Manufacturing: Durable goods EOPSA, millions of chained 1996$ TIMH DLN Inventories: Manufacturing EOP SA, millions of chained 1996$ TIMNH DLN Inventories: Manufacturing: Nondurable goods EOP SA, millions of chained 1996$ TIRH DLN Inventories: Retail trade EORSA, millions of chained 1996$ TITH DLN Real manufacturing & trade inventories EOPSA, millions of chained 1996$ TIWH DLN Inventories: Merchant wholesalers EOP SA, millions of chained 1996$ TRMH DLV Inventory/sales ratio: Manufacturing SA, chained 1996$ TRRH DLV Inventory/sales ratio: Retail trade SA, chained 1996$ TRTH DLV Real manufacturing and trade: Inventory/sales ratio SA, chained 1996$ TRWH DLV Inventory/sales ratio: Merchant wholesalers SA, chained 1996$ Notes: The variable mnemonics are those from HaverAnalytics. For a series yt, the stationary transformations are as follows: LV: xt = y(; DLV: xt = yt- yt_±', LN: xt = ln(y;); and DLN: xt = ln(yt) - ln(yF1). SA is seasonally adjusted. SAAR is seasonally adjusted annual rate. EOP is end of period. 42 3Q/2002, Economic Perspectives REFERENCES Bernanke, Ben, and Jean Boivin, 2002, “Monetary policy in a data-rich environment,” Journal of Mone tary Economics, forthcoming. Conference Board, The, 2001, “Leading economic indicators and related composite indexes: Methodology, revisions, and other information,” New York, report. Diebold, F., and G. Rudebusch, 1996, “Measuring business cycles: A modern perspective,” Review of Economics and Statistics, Vol. 78, February, pp. 67-77. Fisher, Jonas, 2000, “Forecasting inflation with a lot of data,” Chicago Fed Letter, Federal Reserve Bank of Chicago, No. 151, March. Fisher, Jonas D. M., Chin Te Liu, and Ruilin Zhou, 2002, “When can we forecast inflation?,” Eco nomic Perspectives, Federal Reserve Bank of Chica go, Vol. 26, No. 1, First Quarter, pp. 32M4. Prescott, Edward C., 1986, “Theory ahead of busi ness cycle measurement,” in Real Business Cycles, Real Exchange Rates and Actual Policies, CarnegieRochester Conference Series on Public Policy, K. Brunner and A. H. Meltzer (eds.), Vol. 25, Autumn, pp. 11—44. Stock, James, and Mark Watson, 1999a, “Forecast ing inflation,” Journal of Monetary Economics, Vol. 44, pp. 293-335. __________ , 1999b, “Diffusion indexes,” Harvard University, Kennedy School of Government, unpub lished manuscript. __________ , 1989, “New indexes of coincident and leading economic indicators,” NBER Macroeconom ics Annual, pp. 351-394. Theil, Henry, 1971, Principles of Econometrics, New York: John Wiley and Sons. Hamilton, James D., 1989, “A new approach to the economic analysis of nonstationary time series and the business cycle,” Econometrica, Vol. 57, pp. 357-384. National Bureau of Economic Research, 2001, “The business cycle peak of March 2001,” release, Washington, DC, available on the Internet at www.nber.org/cycles/november2001/, November 26. Federal Reserve Bank of Chicago 43 Why do we use so many checks? Sujit Chakravorti and Timothy McHugh Introduction and summary The primary question we address in this article is why consumers, merchants, and financial institutions are reluctant to embrace electronic payments even though electronic payment networks, such as the credit card and automated clearinghouse (ACH) networks, have existed for more than 25 years. While most Internetbased transactions are primarily processed via credit card networks, most noncash off-line payments by both consumers and businesses in the United States are made with checks. In the United States, there are over 15 checks written per month per person.1 This is more than three times the number of checks written per person in Canada or the United Kingdom and at least 15 times more per person than in Germany, Italy, Belgium, the Netherlands, Sweden, or Switzerland (Bank for Inter national Settlements, 2000, and Federal Reserve Sys tem, 2001).2-3 In this article, we incorporate various strands of the payment literature to provide a more integrated view as to why payment system participants are reluc tant to use electronic payments. Brito and Hartley (1995), Hirschman (1982), Mantel (2000), Murphy (1988), and Whitesell (1992) focus on consumer choice issues. Radecki (1999) and Wells (1996) discuss the revenue earned and cost to financial institutions from provid ing check services. Food Marketing Institute (1994, 1998, and 2000), Chakravorti and To (1999), and Murphy and Ott (1977) concentrate on the merchants’ perspectives. McAndrews (1997) and Weinberg (1997) investigate the network issues. Connolly and Eisenmenger (2000), Benston and Humphrey (1997), Green and Todd (2001), Guynn (1996), and Lacker and Weinberg (1998) discuss the Federal Reserve’s role in the payment system. A more integrated analy sis of the underlying incentives of various payment system participants has been developed by Baxter 44 (1983), Chakravorti and Emmons (2001), Chakravorti and Shah (2001), Rochet and Tirole (2000), and Wright (2000). We study the incentives underlying the payment network to examine why, unlike several other indus trialized countries, the United States has been slow to abandon checks. Many observers claim that electronic payments are less expensive than checks. However, these social cost comparisons usually ignore transition costs and the underlying incentives to each payment participant. Furthermore, the provision and usage of payment services exhibit network effects, more com monly referred to as the chicken-and-egg problem, which may impede the adoption of new payment tech nologies. Even if electronic payments are less expen sive and they can overcome the chicken-and-egg problem, consumers, merchants, and financial insti tutions may still be reluctant to move to electronic payments. We analyze why this is so. In addition, we explore actions by the Federal Reserve to improve the check processing system and whether this could possibly hinder the migration away from checks. Finally, we discuss potential drivers to the adoption of electronic payments. Check usage We use two different sources of check data in this article. The first source is the annual payments data published by the Bank for International Settlements Sujit Chakravorti is a senior economist and Timothy McHugh is a senior analyst in the Emerging Payments and Policy Department at the Federal Reserve Bank of Chicago. The authors would like to thank David Allardice, Ed Green, Harvey Rosenblum, and Fiona Sigalla for numerous helpful discussions. They also thank Eve Boboch, Tom Ciesielski, David Marshall, Ann Spiotto, Victor Stango, and Kristin Stanton for comments on previous drafts. 3Q/2002, Economic Perspectives (BIS). The second source of data is a comprehensive review of the retail payment systems by the Federal Reserve System (Fed) (2001). The Fed study indicated that the total volume of check payments in the United States was significantly lower than previously estimat ed. However, the data published by the BIS still pro vide valuable insights into check usage in other countries and check usage trends in the United States. We rely on the new Fed study for current check values and volumes and use the older data reported to the BIS by the Fed for trends in check values and volume. According to the new Fed benchmarking study released in November 2001, 49.6 billion checks were written in the U.S. in 2000, valued at $47.7 trillion (Federal Reserve System, 2001).4 Checks represented around 60 percent of non-cash consumer transactions. The Fed study estimates that consumers wrote around 51 percent of checks but only accounted for 19 per cent of the total value. According to the BIS (1991— 2000), per capita check volume grew at a compounded annual growth rate (CAGR) of 1.13 percent, while per capita check value grew at a 1.91 percent CAGR from 1991 to 1999.5 Unlike most other industrialized countries, the U.S. seems to have experienced growth in total check volume and value during the 1990s. For every Group of Ten (G-10) country except the United States, the volume of check usage (see figure 1) and value (see figure 2) declined during the 1990s.6 Among the rea sons that have been cited to explain U.S. check vol ume growth are differences in financial institutions per capita, cash usage, laws and regulations, and pric ing of financial services (see BIS, 1999 and 2000, and Humphrey, Pulley, and Vesala, 2000). By increasing the price of checks vis-a-vis other payment options, financial institutions in Scandinavian countries have been successful in decreasing check usage. For example, in Finland, during the mid-1980s, banks began implementing a small per-check fee of about 10 cents. Palva (2000) states that this pricing policy coincided with a drastic reduction in the use of checks (see figure 3). Adopting similar policies, Norwegian banks also successfully decreased check usage. Humphrey, Kim, and Vale (2001) found that a 1 percent increase in the price of checks resulted in a 1.07 percent decrease in check usage. They also found that online debit cards were a close substitute for checks at the point of sale.7 As a result, check usage in Norway decreased from 72 million checks in 1988 to only 6.2 million in 2000. Furthermore, the volume of payments made by pay ment cards, primarily debit cards, was 62 times that of checks in 2000 (Bank of Norway, 2000). Federal Reserve Bank of Chicago Other countries have used different approaches to reduce check usage. For example, Canadian banks give check payees immediate credit and availability of their funds (see Humphrey, Pulley, and Vesala, 2000). Furthermore, checks are backdated to remove any float benefit to paying banks. In addition, corporations are charged for the float when the distance is significant between the bank where the check is drawn and where it is first deposited. The decline in check usage across most countries indicates that, given market incentives, there is a move ment toward electronic instruments. Electronic alter natives to accessing transaction accounts for purchases are held to be less expensive than checks. Humphrey and Berger (1990) were the first to calculate the total social cost of each instrument in the United States. Social cost is the sum of the real resource cost borne by each participant to convert a given payment into good funds. They found the social cost of a cash trans action to be the lowest at 4 cents and a credit card trans action to be the highest at 88 cents. An ACH payment, an online debit transaction, and a check transaction have social costs of 29 cents, 47 cents, and 79 cents, respectively. Wells (1996) updated Humphrey and Berger’s study and found that ACH payments cost between one-third to one-half as much as a check payments but found significantly higher social cost estimates for both checks and ACH payments. A difficulty with comparing social cost among payment instruments is that a given payment instrument may not be preferred for both small and large trans actions. While cash transactions outnumber all other types of transactions, the average transaction size is relatively small compared with other payment instru ments. Consumers tend to prefer checks for larger trans actions. The average consumer check transaction is estimated at $364 (Federal Reserve System, 2001). Furthermore, consumers may not be able to use all pay ment instruments for all types of transactions. For example, cash cannot be used for bill payment via mail and checks are difficult to use for Internet trans actions. Additionally, these social cost calculations may not adequately adjust for the risk of not being able to convert the payment into good funds that may play a role in the acceptance of certain payment instru ments. Some characteristics of payment instruments are difficult to quantify, such as the convenience and comfort levels enjoyed by the participants. These estimates also ignore transition costs and network effects. Consumers, merchants, and financial institutions may be unwilling to invest in emerging payment technologies due to uncertainty about wheth er they will be widely accepted in the marketplace. 45 FIGURE 1 Percent change in check volume (1990-99) percent Belgium Canada Germany Italy Netherlands Sweden Switzerland Kingdom States Source: BIS (1991-2000). 46 3Q/2002, Economic Perspectives FIGURE 3 Electronic payments in Finland Source: Finnish Bankers’ Association and Palva (2000). U.S. smart card trials demonstrated that consumers and merchants may not be willing to adopt new forms of payment rapidly.8 Lack of incentives In this section of the article, we analyze the cost and incentive structure faced by each participant in the payment network. We address two fundamental questions for consumers, merchants, and financial in stitutions. First, are electronic payment alternatives less expensive than checks for each participant? Sec ond, if electronic payment forms are less expensive, are participants reluctant to abandon checks because they lack the right incentives to adopt alternative pay ment instruments? Consumers While checks might be more costly to society as a whole, several studies point out that consumers may view the marginal cost to use a check to be zero.9 Re cently, several banks have reintroduced free checking accounts to entice new customers.10 Humphrey, Pul ley, and Vesala (2000) state that most U.S. consumers prefer accounts with fixed monthly fees or no fees with minimum balance requirements to those with per-check transaction fees. Furthermore, merchants rarely impose additional fees for check payments. Federal Reserve Bank of Chicago Moreover, some consumers still view check float as a major benefit. Today, most checks are processed overnight and interest rates on transaction accounts, if they are offered, are quite low, resulting in low float benefits. Wells (1996) calculated that float is no longer significant for consumer check payments. Nonetheless, some consumers still may perceive significant float benefits. Ironically, checks do not have a built-in feature that automatically declines a transaction if the custom er’s account does not have sufficient funds. While non-sufficient-funds fees are relatively high and may lead to several other checks bouncing, most consum ers seem to ignore these costs. However, non-sufficient funds fee income is large for financial institutions, potentially reducing banks’ incentive to promote some electronic payment alternatives. Independent of the cost of check payments, we can identify three key reasons consumers have resist ed abandoning checks. First, checks are easy to use. The 1998 Survey of Consumer Finances indicated that about 87 percent of U.S. households had checking accounts, making checks the most accessible noncash payment instrument.11 Checks are also one of the most widely accepted forms of payment by merchants at the point of sale.12 For bill payments, checks are the most popular instrument because, unlike other forms of payment, they are almost always accepted by billers.13 47 Second, consumers are reluctant to switch to elec tronic alternatives unless they offer superior benefits to checks. While consumers may believe that electronic payments are less expensive overall, they are reluctant to change unless they view the shift as beneficial to them. Credit card issuers often offer additional services such as extended warranties, dispute resolution servic es, and frequent-use awards, along with interest-free short-term loans to those who pay off their balances each month. Third, some consumers feel checks give them great er control over the timing of their payments, leading to better budgeting. Hirschman (1982) argues that some consumers believe that checks may enhance their abil ity to track, budget, and control spending better than other payment instruments. Mester (2000) argues that checks give consumers more control over when to pay bills than pre-authorized ACH payments and can more easily attach remittance information. Yet, consumers can also access their checking accounts via their deb it cards and maintain budgeting, tracking, and control over their funds. When using debit cards, consumers cannot overdraw their accounts unless previous credit lines have been established. However, debit card us age in bill payment is relatively low, given the slow adoption of the necessary infrastructure. Because consumers perceive checks to be a low cost payment instrument and are comfortable with them, they are reluctant to change unless there are strong incentives to do so. From a cost standpoint, checks are relatively inexpensive if one ignores non-sufficientfiinds fees. As we noted earlier, explicit per-check charg es by financial institutions in other countries have been effective in changing consumers’ payment habits. merchants 80 cents per $ 100 in sales, though this cost varies widely depending on whether the check was verified. Given the rapid increase in the use of check veri fication systems during the last decade, it is important to analyze check costs using this technology. Nilson (2001a) estimates that 9.14 billion checks were veri fied at the point of sale. Using Federal Reserve System (2001) point-of-sale check numbers, we estimate that between 75 percent and 97 percent of checks written at the point of sale were verified in 2000.14 The typi cal cost for these services ranges from 2 cents to 20 cents per check (Nilson, 1997b). Nilson reports an average cost of 3 cents in 1998.15 Merchants have found that check verification ser vices significantly reduce the risk that they will not receive good funds. As a result, they have been able to lower their costs by over 23 percent and to reduce losses from exception items from 0.50 percent of the value of the check to 0.05 percent of the value of the check (FMI, 1998).16 According to FMI (2000), a verified check pay ment is actually the cheapest form of payment for the merchant to accept, costing a merchant 60 cents per $100 in sales versus $3.00 per $100 in sales for an unverified check.17 Since the majority of checks at the point of sale appear to be verified, it is important to concentrate on the cost of verified checks. According to FMI (2000), the cost of a verified check per $100 in sales is significantly less than cash, off-line debit cards, and credit cards. Although FMI (2000) did not report a cost for ACH-based debit card transactions, FMI (1998) reported an average cost of 82 cents in 1997. The cost difference between a verified check trans action and an online debit card transaction might also be growing. Between 1997 and 2000, FMI (1998 and 2000) found that the cost of online debit cards increased by 14 percent, or 10 cents per $100 in sales.18 Recent ly, several networks have announced plans to increase their fees significantly.19 Even if checks are more expensive than electronic alternatives, merchants may continue to accept checks Merchants A primary issue for merchants is the cost of pay ments. There are significant differences in the cost of accepting alternative payment instruments. The Food Marketing Institute (FMI) (2000) estimates the mer chant’s cost to accept each payment instrument (see table 1). Online debit cards are the second least ex pensive payment instrument for merchants to accept at 80 cents per $100 of transac TABLE 1 tions. They offer merchants immediate Merchant costs to accept a payment instrument funds, low per-transaction fees, and little, if any, settlement risk. While cash pro Cost per Check cessing costs are low at 90 cents per $100 transaction not Check Online Off-line (dollars) Cash verified verified Credit debit debit of sales, most consumers are reluctant to use cash for larger purchases. Credit card Cost per $100 sales .90 3.00 .60 1.80 .80 1.80 and off-line debit card transactions cost merchants an average of $1.80 per $100 in Source: Food Marketing Institute, 2000. sales. The average check transaction costs 48 3Q/2002, Economic Perspectives FIGURE 4 U.S. debit card use at point of sale Source: Nilson (1996-2001), Nos. 737, 726, 705. 678. 654. and 627. for three reasons. First, the potential cost savings from electronic alternatives might not be large enough to justify the transition costs to make this change and/or risk movement to more expensive payment vehicles. Second, cheaper electronic payment alternatives at the point of sale have only recently flourished. Last, checks might offer merchants some level of benefits that they are willing to “pay” for. Given the slow movement away from checks, merchants, banks, and third-party providers have start ed to convert check transactions to ACH transactions at the point of sale to reduce costs.20 However, this may involve high initial set-up costs related to imple menting a new system, purchasing equipment, and training staff. Furthermore, merchants may be reluc tant to make large investments in new payment tech nologies with uncertain futures. Furthermore, even though online debit cards are a relatively inexpensive payment option, they have be come widespread only recently. Annual per capita trans action volume in the United States increased from 0.76 transactions per year in 1990 to 11.3 transactions in 2000 (BIS, 1991-2000, and Thomson Financial, 2001). Figure 4 shows the rapid increase in debit card trans actions over the last five years. Figure 5 shows the increase in online debit card terminals installed by mer chants, indicating substantial growth in merchant ac ceptance of online debit cards over the last 13 years. Federal Reserve Bank of Chicago Lastly, some evidence suggests that merchants are willing to accept high-cost payment instruments be cause they offer benefits not offered by other instru ments. Credit cards generate sales to illiquid consumers who may not otherwise be able to purchase goods and services.21 In some instances, merchants may choose to accept certain instruments because they are tied to other instruments that they choose to accept. For ex ample, merchants accepting Visa or MasterCard credit cards are required to accept their off-line debit card prod ucts. A large group of retailers led by Wal-Mart has sued the credit card associations, alleging that this tying of their credit card and debit card products is illegal. Available U.S. data do not indicate a significant cost reduction if merchants move toward electronic payments. Furthermore, the benefits of accepting electronic payment instruments may not outweigh the investment that may be required. Merchants may be willing to accept relatively expensive payment in struments because they offer benefits such as the po tential to increase sales and profits. Financial institutions Although electronic payments are generally per ceived to be less expensive than paper-based pay ments for financial institutions, several U.S. studies indicate the costs of processing ACH and check pay ments are not very large for financial institutions. 49 FIGURE 5 EFT point-of-sale terminals The Federal Reserve Board’s 1994 Functional Cost Analysis estimated that the average cost per transaction for an ACH payment was 14 cents and the average cost for a check payment was 14.6 cents. Wells (1996) es timated that the cost of processing a check ranged from 15 cents to 43 cents, while the cost of processing an ACH payment ranged from 12 cents to 45 cents. Guynn (1996) questions whether ACH payments are really less expensive for banks to process than checks. However, evidence from Norway indicates that check transactions cost banks two to three times as much as electronic giro services and electronic funds transfer point-ofsale transactions (Robinson and Flatraaker, 1995). Furthermore, financial institutions earn significant revenue from the provision of check services. On aver age, they charge customers 21 cents and merchants 5 cents to process each check.22 In order to spur adop tion, online debit cards should offer financial institu tions similar revenue opportunities assuming similar cost structures, or similar profit opportunities result ing from offsetting cost savings. Online debit cards provide a potentially lucrative revenue stream in the form of the fee paid by the merchant’s bank to the cus tomer’s bank, commonly referred to as the interchange fee. Yet, as of the end of 2001, no online debit card network had interchange fees higher than 20 cents.23 Recently, a few online debit card networks have sig nificantly increased their processing fees. Several in stitutions have also implemented per-transaction fees 50 for using personal identification number (PIN) based debit cards at the point of sale.24 Financial institutions also earn significant revenue from fees related to overdrafts to checking accounts. According to Bank Administration Institute and PSI Global (1998), in 1995, banks earned $8.1 billion from non-sufficient-funds check fees. The Board of Gov ernors of the Federal Reserve System (1996) estimated that banks’ losses from check fraud amounted to $615 million in 1995, $215 million of which was 3Q/2002, Economic Perspectives eventually recovered. Debit cards could significantly reduce or eliminate overdrawn accounts, because transactions are only processed if funds are available. However, by promoting online debit cards, financial institutions would reduce their revenue from nonsufficient-funds fees.25 If electronic payment instruments were less ex pensive than checks, financial institutions might be able to influence consumer usage and merchant accep tance of electronic payments as evidenced in Norway and Finland. However, U.S. financial institutions may be reluctant to impose explicit per-check usage fees due to competitive pressures. Some financial observers have argued that the large number of payment providers in the United States might prevent financial institutions from implementing price increases and cost-saving measures. One initiative by a Midwest bank to charge consumers for using bank tellers initially met with consumer resistance and little support from other financial institutions. Stavins (1999) found that regional competition prevented banks from not returning canceled checks because they feared customers might switch to a rival bank that returned checks. Yet, some financial institutions were eventual ly successful in implementing both of these policies. Whether electronic payments are less expensive to process than checks for financial institutions is de batable. However, even if electronic payments are less expensive, the potential revenue from checks, especial ly in the form of non-sufficient-funds fee income, is difficult for financial institutions to forgo. The Federal Reserve Central banks differ in the roles they play in the operation and oversight of their domestic payment system (see Bank for International Settlements, 1999 and 2000). Most central banks of industrialized coun tries are involved in the settlement of retail payment transactions and some also play a clearing role. In the United States, the Fed is a provider of interbank check clearing services and is the largest ACH operator. Whether there remains a compelling need for the Fed to provide check-processing services is debatable, given technological and regulatory changes. When creating the Fed, the U.S. Congress stipulated that it should improve the efficiency and safety of the pay ment system.26 At the time the Fed was created in 1913, checks were the primary means of interbank funds transfer. Today, large-value domestic transactions are processed electronically via Fedwire, which is oper ated by the Fed. In addition, electronic retail payment alternatives, such as credit cards, debit cards, and ACH payments, are increasing their market shares.27 Federal Reserve Bank of Chicago The Federal Reserve has historically played an im portant role in the development of check processing. However, some have questioned this role. Green and Todd (2001) argue that as the United States transitions to the next generation of payment instruments, the Fed should promote “efficiency, integrity and acces sibility primarily by means other than direct service provision—such as participation in the setting of stan dards, the drafting of model legislation and the regu lation of payment services markets” (Green and Todd, 2001, p. 1). They further argue that “encroaching on activities that the private sector can perform efficiently and equitably” may erode the Fed’s reputation as a trustworthy and neutral institution. The Fed’s role in check processing has declined over the past two decades, partly due to regulatory changes such as the Monetary Control Act of 1980 (MCA), the removal of interstate branching restrictions, and changes to Regulation CC to allow banks to settle checks in same-day funds. In the five years following the passage of MCA—under the terms of which the Fed had to price its financial services and make them available to all financial institutions—the Fed’s share of interbank check clearing decreased sharply.28 Summers and Gilbert (1996) claim that the Fed’s share of check volume decreased from 61.0 percent in 1980 to 49.8 percent in 1985.29 For the same period, its share of check value decreased from 48.5 percent to 31.7 percent. The Fed also experienced a sharp decrease in the number of checks handled from 1993 to 1995. Given the lack of reliable non-Fed check volume data for this period, it is difficult to determine the cause of this de crease. Some financial observers claim that changes in Regulation CC and the removal of interstate branch ing restrictions were at least partially responsible. In 1994, Regulation CC was changed to allow banks to settle checks in same-day funds. These changes may have resulted in more institutions using private clear inghouses. We would also expect that as banks merged, the ratio of on-us transactions would increase. How ever, most estimates indicate that on-us check volume has remained constant at around 30 percent.30 Alternatively, total check volume may have de creased during this period as a result of the adoption of electronic alternatives. Therefore, the drop in Fed volume would be attributable to a decrease in overall check volume and not a change in market share. Per haps a more plausible explanation is that both a shift to private-sector alternatives and substitution of other payment instruments were responsible for the reduc tion in check volume. However, as stated before, in the absence of reliable data, we can not determine the magnitude of each effect. 51 According to Federal Reserve System (1998), the Fed processes a greater proportion of checks for smaller depository institutions than private sector providers. During the Committee on the Federal Reserve in the Payments Mechanism (also known as the Rivlin Com mittee) public forums, several community banks in dicated that they feared that private entities would not process checks at similar prices to those currently charged if the Fed left the check processing industry. Several banks and clearinghouses “freely admit that they would charge more to clear these items than the Federal Reserve now does, citing the higher costs in volved in serving these endpoints” (Federal Reserve System, 1998). These institutions suggested that the Fed subsidizes small institutions, especially those lo cated in remote locations. The Fed has denied cross-subsidizing across its priced-services product. According to Rivlin (1997, p. 5), then vice-chair of the Board of Governors of the Federal Reserve System, each product in a multi product firm “should recover at least its incremental production cost” to ensure that no individual product is being subsidized.31 Using this standard, the Fed does not subsidize any of its products. Similar argu ments could be made about the allocation of fixed costs among customers that may result in certain custom ers paying a higher proportion of fixed costs. How ever, some distributions of allocating fixed costs among customer segments may deter competitors from entering industries where economies of scale and scope are present. Chakravorti, Gunther, and Moore (1999) suggest that private-sector providers could cherry-pick the prof itable customers and leave the higher cost ones with the Fed if low-cost customers pay more than the mar ginal cost to serve them. They argue that as low-cost customers find other less-expensive check processors, the price charged should rise for those remaining cus tomers. Eventually, the rise in cost may result in an exit strategy for the Fed. Thus, these more expensiveto-serve customers may eventually choose to promote non-check payment alternatives by charging higher check fees to their customers. However, if the Fed improves check-processing technology, such incentives would be reduced. Recent ly, the Fed has made large investments in improving check-processing technology to lower its costs.32 It continues to promote electronic check presentment.33 In 2000, the Fed electronically presented about 20 percent of the checks it handled. From 1995 to 2000, the number of checks presented electronically grew from 1 billion to 3.5 billion, or a 28.5 percent com pound annual growth rate (CAGR). Though the 52 physical checks still followed, this program allowed for faster presentation of the check for payment. The Fed also participated in projects where the check was imaged or truncated at either the payee bank or at the Fed and no paper was sent to the paying bank. More than 7 percent of the checks the Federal Reserve han dled were processed in this manner in 2000. While these changes are aimed at decreasing check processing costs, they may also affect the migration to electronic alternatives. Specifically, a less-expen sive check-processing system may reduce the incen tives for financial institutions to migrate to electronic alternatives. Lacker and Weinberg (1998, p. 19) argue that “ECP (electronic check presentment) could be viewed as an attempt to stem the expected decline in check use. By reducing the cost of paper checks, ECP could slow the transition to fully electronic payment instruments that are even more beneficial.” Benston and Humphrey (1997) suggest that the Fed may not have sufficient reason to continue in the check-processing business. Furthermore, they point out that the Fed is virtually alone among central banks of developed countries in the provision of check-pro cessing services to financial institutions. Bullock and Ellis (1998) suggest that the “heavy involvement” of the Fed in check processing, along with its role in reg ulating the industry, has kept the check competitive with other payment instruments. On the other hand, improvements in check pro cessing may allow for the electronification of checks, while maintaining some features not presently avail able in competing payment instruments in the United States, ultimately facilitating the movement to electron ic alternatives. Connolly and Eisenmenger (2000) ar gue that in some instances there is a need for the Fed to take an operational role to improve the processing of checks where the private sector may not be willing to invest the necessary funds initially to adopt the necessary infrastructure. Guynn (1996) suggests that certain improvements in check processing may actu ally improve the adoption of electronic instruments. Drivers to change Given today’s underlying incentive structure, it would appear that U.S. consumers, merchants, and financial institutions are not likely to change their pay ment preferences in the near future. However, given technological enhancements and competition from nonbank payment providers, the incentives for pay ment system participants to use electronic alterna tives will increase. Financial institutions, along with merchants, have started to “electronify” checks by converting them to ACH payments. Given the right 3Q/2002, Economic Perspectives incentives, consumers may also increase their use of electronic payment instruments. In this section, we dis cuss drivers that might aid the transition away from checks in the future. U.S. consumers have only recently had electron ic alternatives for different types of payments. Credit cards have surpassed checks as the most popular in strument used for point-of-sale transactions.34 Debit cards are becoming increasingly popular at the point of sale, and ACH payments continue to gain popular ity for recurring bill payment and payroll disbursements. In addition, third-party providers are using new tech nologies to deliver payment card products and ACH payments to new market segments. Debit cards show the greatest promise to decrease check volume at the point of sale. Debit cards offer con sumers access to their transaction account like checks and allow merchants to receive their funds relatively quickly, incurring little, if any, settlement risk. Mer chant acceptance of debit cards and the number of con sumers holding debit cards are growing rapidly. From 1995 to 1998, the number of households with a debit card increased by a 27 percent CAGR. The success of promoting debit card usage is partly due to the le veraging of existing credit card and automated teller machine (ATM) networks and financial institutions’ ability to easily put the product in the wallets of their consumers by increasing the functionality of their ATM cards.35 Earlier, we noted that financial institutions may not have sufficient incentives to promote online debit cards. Recently, several electronic funds transfer net works have consolidated. In the process, several of the largest networks have publicly stated that they in tend to increase the revenue to financial institutions for participating in their networks (Breitkopf, 2001a). Two of the most prominent networks announced their plans to almost double their interchange fees (Breitkopf, 2001b). However, these plans were delayed more than six months after several large retailers indicated that they would discontinue processing transactions over the networks if the rate increases went through. Off-line debit card usage has also increased rapidly and these cards now outnumber their online counterparts in the United States. These cards provide financial in stitutions with similar levels of interchange fees to those offered by credit cards, usually a percentage of the purchase price. A few years ago, a group of merchants filed a lawsuit against Visa and MasterCard, challenging the “honor-all-cards” rules of the card associations. 36 These rules stipulate that if merchants accept one of the card association’s products such as credit cards, they must accept all of the association-branded products. 37 Federal Reserve Bank of Chicago The merchants claim that few alternatives exist to the general-purpose credit cards. Therefore, they are un willing to stop accepting credit cards but want the abili ty to decline the associations’ off-line debit cards. A significant but not often discussed payment seg ment is the person-to-person (P2P) payment segment.38 The Federal Reserve System (2001) estimates that 11.2 percent of total check volume and 6.7 percent of total check value in 2000 were consumer-to-consumer pay ments. Individuals are usually unable to accept ACH payments or credit or debit card transactions. In the last two years, banks and nonbanks have started to enable individuals to accept these payment instruments.39 While the initial impact of P2P has been mostly lim ited to the online auction community, recent initiatives by P2P providers have been geared toward capturing a larger share of the non-auction online transactions.40 Some small businesses have begun to use P2P payment services as a means to accept payments both within and outside the auction community. In the bill payment arena, depository institutions (Dis) are facing competitive pressures from both thirdparty providers and non-depository financial institu tions. Third-party providers began promoting the use of electronic payments to many of the most profitable customers for Dis. Brokerage firms and credit-card banks, which already have a connection to many of the Dis’ high-net-worth customers, have also been actively promoting electronic bill payment services. Depository institutions risk losing these integral rela tionships if they do not match or exceed the services offered by these competitors. Thus, other providers ap pear to have accelerated the incentives for Dis to pro mote some forms of account-based electronic payments. While checks continue to dominate the propor tion of non-cash payments, we have discussed several drivers that should facilitate the migration to electronic alternatives. Leveraging existing networks, debit cards have gained significant market share, and financial observers believe that debit card growth rates will con tinue to be higher than those of other established pay ment media. As more commerce is conducted remotely, such as via the Internet, electronic payments’ share of total payments will also increase, since paper instru ments may not be appropriate for these environments. Furthermore, new payment providers, especially non bank providers, have started to leverage existing net works, such as debit and ACH networks, to allow electronic person-to-person payments. As consumers and merchants become comfortable with electronic pay ments in certain payment segments, there will likely be spillover effects to other payment segments that have traditionally been dominated by checks. 53 Conclusion In this article, we have examined why U.S. con sumers, merchants, and financial institutions have been unwilling to significantly reduce their check usage. Con sumers lack incentives to change their habits. In the United States, credit card issuers have been success ful in gaining point-of-sale market share by offering incentives such as frequent-use awards and interest-free short-term loans if monthly balances are paid in full. However, in some environments, such as online pur chases, consumers may have little choice but to use electronic alternatives. Consumers may become more comfortable with electronic alternatives the more they use them, resulting in faster market adoption. For merchants, the cost to process checks, includ ing the risk of not being able to convert a payment to good funds, may not be significantly greater than for electronic alternatives such as online debit cards. How ever, merchants are also realizing the benefits of on line debit cards as evidenced by the rapid deployment of point-of-sale terminals and the merchants’ antitrust suit against the credit card associations. Merchants may be gaining sufficient bargaining power with pay ment providers to impact existing cost structures in ways that may increase usage of electronic alternatives. Financial institutions in Finland and Norway have been successful in convincing consumers to sig nificantly curtail their check usage and increase their use of electronic payment forms by imposing per-check fees. However, checks remain a substantial source of revenue for U.S. financial institutions, especially in terms of non-sufficient-funds fees. Electronic alterna tives, such as online debit cards, may not have been as 54 financially attractive, but new pricing policies by the online debit card networks may entice financial insti tutions to promote them more heavily. Some have questioned the role of the Fed in the retail payments arena. While the Federal Reserve is ac tively promoting electronic alternatives, it continues to improve check processing. Such improvements may distort the market incentives to move to electronic pay ments. Even though the United States lags behind other industrialized countries in its continued high usage of checks, no studies to date have concluded that the mi gration to electronic substitutes for checks is welfare enhancing in the U.S. context. In this article, we have identified several poten tial drivers of electronic payments, such as greater choice of payment instruments for consumers for different payment segments, greater non-face-to-face shopping opportunities, competition from non-bank payment providers, and a greater role by merchants to offer the low-cost payment alternatives. Anecdotal evidence sug gests that U.S. consumers are slowly changing their pay ment habits, and we would expect this trend to continue. Further research is warranted as to why the Unit ed States lags other industrialized countries in adopt ing electronic alternatives. We have suggested that changes need to occur in the underlying incentive structure to convince all payment participants to mi grate to electronic payments. U.S. consumers, mer chants, and financial institutions are more likely to make the transition to electronic payments, given the growth in remote purchases, developments in tech nology, and greater market-based incentives to use electronic alternatives. 3Q/2002, Economic Perspectives NOTES JThe number of checks written includes consumer, business, and government checks. We focus only on consumer checks in this ar ticle. We use U.S. check data for 2000 from the new Federal Re serve System check survey and 1999 figures from BIS for the other countries (Federal Reserve System, 2001, and BIS, 2000). 2The level of confidence in the payment statistics published by BIS is questionable. Therefore, a great amount of care should be exercised in interpreting them. 3A portion of the difference in per-capita check usage might be attributable to higher levels of cash use in some of these countries. Some countries with low check usage may have a high level of cash usage, such as Japan. See BIS (1999) for a more detailed discussion of these differences. determining the volume and value of checks is difficult. While the Federal Reserve knows the number and value of checks it pro cesses, it estimates the volume of checks processed by others. In 1999, Nilson (1996-2001), BIS (2000), and Green (1999) esti mated total U.S. check volume between 64 billion and 69 billion, accounting for $47 trillion to $83 trillion. Prior to the 2001 study, the Federal Reserve’s last benchmarking study was conducted in 1979. For more recent years, non-Fed check volume was extrapo lated from the 1979 study. 5We would expect that these numbers overestimate the growth rate in check usage, given the 2001 Fed study, but no reliable evidence indicates that growth has been negative during this period. 6The G-10 countries are Belgium, Canada, France, Germany, Italy, Japan, the Netherlands, Sweden, Switzerland, the United Kingdom, and the United States. Japan is not included in the figure because check usage in Japan is extremely limited, and France is not in cluded because it did not report check data for 1999. 7There are two types of debit cards—online and off-line. Online debit cards use ATM networks to authorize and process transac tions and require customers to use a PIN (personal identification number) code. Off-line debit cards use credit card networks and are signature-based. 8For a discussion of why stored-value did not succeed in the United States, see Chakravorti (2000). 9However, there are fixed costs such as the opportunity cost of holding funds in a zero- or low-interest bearing account and po tential monthly fees. 10Bank of America, Washington Mutual, Bank One, Harris Bank, and Fifth Third have recently started to promote free checking accounts (Thomson Media, 2001). HIn comparison, 72.5 percent of households had a credit card and 34.5 percent of households had a debit card in 1998 (a substantial increase from 17.6 percent of households in 1995). 12However, there are several types of merchants, such as gas sta tions and restaurants, that do not usually accept checks. 13For more on bill payment, see Andreeff et al. (2001). 14Federal Reserve System (2001) estimates that of the 49.6 billion checks written in 2000, 19 percent were written at the point of sale and 12 percent were written at either the point of sale or for remittance. Therefore, between 9.42 billion and 12.4 billion checks were written at the point of sale. Federal Reserve Bank of Chicago 15In addition, merchants used third parties to guarantee another 2 percent of the total check volume. According to Nilson (1999b), the cost for check guarantee services averaged 1.56 percent of the value of the check in 1999. Note that guaranteed check costs are significantly higher than those of online debit cards and may be more than off-line debit cards and credit cards. 16FMI (2000) does not break out the cost of fraud for verified and unverified checks, so we are using FMI (1998) for this portion of the analysis. 17We should note that based on the average cost per transaction, FMI (2000) shows that online debit cards are still the cheapest means of payment. On a per transaction basis, online debit cards cost $0.34, verified checks cost $0.36, and unverified checks cost $0.38. ^Comparatively, on a per transaction basis, the cost of debit cards rose from an average of $0.30 in 1994 and $0.29 in 1997 to $0.34 in 2000. 19Both the Star and NYCE networks have increased their maximum fees for supermarkets to 19 cents, while Interlink has raised this fee to 20 cents. 20In 2000, 32 million checks were converted to ACH payments at retail locations (National Automated Clearing House Association, 2001). 21For a theoretical exposition of credit extensions and their ben efits to merchants, see Chakravorti and To (1999). 22The Board of Governors of the Federal Reserve System (2001) states that the average per check fee was about 21 cents for inter est-bearing accounts. Check revenue would be similar for accounts that did not charge per-check fees if the average account holder wrote 9.5 checks. See FMI (1998) for merchants’ fees from checks. 23As of the end of 2000, Interlink charged the highest online debitcard-processing fee of 20 cents for point-of-sale transactions. Af ter recent price increases, Interlink still charges the highest fees with a maximum of 45 cents for point-of-sale transactions. 24According to a recent survey by Dove Consulting and Pulse EFT, 6 percent of financial institutions charge extra fees for using PINbased debit cards at the point of sale. The study found that an av erage fee of $1 is being charged to consumers because the lowinterchange offered on PIN-based debit transactions does not ad equately cover processing costs. See Breitkopf (2002). 25Off-line debit cards offer issuers significant revenues in the form of interchange fees that could offset the decrease in insufficientfunds fees. This lucrative interchange revenue may be partly re sponsible for the increase in popularity of debit cards since their introduction in the early 1990s (see figure 4). 26For a historical perspective on the Fed’s role, see Gilbert (1998) and Summers and Gilbert (1996). 27See Federal Reserve System (2001). 28For a discussion of the private sector response to Fed pricing polices resulting from the MCA, see Frodin (1984). For a histori cal perspective on retail payment services and the MCA, see Kuprianov (1985). 55 29The Fed’s share is taken as a percentage of interbank check vol ume ignoring any changes in on-us volume. 30The last comprehensive study of the payment system by the Federal Reserve prior to the 2001 study placed the share of on-us check volume at 30 percent in 1979. The 2001 Fed study also put the on-us check share at 30 percent. 31For more discussion of cross-subsidization, see Faulhaber (1975). 32By mandate, the Fed must recover costs from the financial services it provides. Investments in equipment are amortized over years. Thus, it appears that the Fed expects the demand for its check clearing services will not decline significantly in the near future. 33A number of other private-sector initiatives have also been un dertaken to truncate checks at the point of sale and through lock boxes. In most cases, these initiatives take the magnetic ink char acter recognition (MICR) information and turn the payment into an ACH transaction. 35There are differences in consumer liability across payment instru ments. For a discussion, see Spiotto (2001). For a discussion of how evolving payment instruments and applications have leveraged the existing payment infrastructure, see Mantel and McHugh (2002). 36This case is currently awaiting trial. 37See Evans and Schmalensee (1999) for a discussion of network rules and history. 38For more details on online P2P systems, see Kuttner and McAndrews (2001) and McHugh (2002). 39In some cases, P2P providers have created their own medium of exchange, but most also allow consumers to easily convert the value into good funds. 40PayPal, the leading provider of electronic person-to-person pay ments, indicated that 66.9 percent of its payment volume in 2001 originated from online auctions. See McHugh (2002) for a discus sion of PayPal and its services. 34Some observers do not categorize credit cards as electronic pay ments, because in most instances the statement is provided on paper and most payments are made by check. However, for most mer chants, credit cards are processed, cleared, and settled electronically. REFERENCES Andreeff, Alexandria, Lisa C. Binmoeller, Eve Boboch, Oscar Cerda, Sujit Chakravorti, Thomas Ciesielski, and Edward Green, 2001, “Is EBPPjust a click away?,” Economic Perspectives, Federal Re serve Bank of Chicago, Vol. 25, No. 4, pp. 2-16. Bank Administration Institute and PSI Global, 1998, Profilingfrom Change in the U.S. Payments System, Chicago, IL: Bank Administration Institute. Bank for International Settlements, 2000, Clear ing and Settlement Arrangementsfor Retail Payments in Selected Countries, Basel, Switzerland, September. Benston, George J., and David B. Humphrey, 1997, “The case for downsizing the Fed,” Banking Strategies, January/February, pp. 30-37. Board of Governors of the Federal Reserve Sys tem, 2001, Annual Report to the Congress on Retail Fees and Services ofDepository’ Institutions, Wash ington, DC: Federal Reserve System. __________ , 1998, Survey of Consumer Finances for 1998, Washington, DC: Federal Reserve System. __________ , 1996, Check Fraud Survey, Washing ton, DC: Federal Reserve System. __________ , 1999, Retail Payments in Selected Coun tries: A Comparative Study, Basel, Switzerland, Sep tember. __________ , 1994, Functional Cost and Prof1 Anal ysis, Washington, DC: Federal Reserve System. __________ , 1991-2000, Statistics on Payment Sys tems in the Group of Ten Countries, Basel, Switzerland. Breitkopf, David, 2002, “PIN-signature debit tug-ofwar escalates,” American Banker, February 25, p. 6. Bank of Norway, 2000, Annual Report on Payment Systems for 2000, Oslo, Norway. __________ , 2001a, “B of Amove to Visa sign of EFT shakeout,” American Banker, August 31, p. 1. Baxter, William F., 1983, “Bank interchange of transactional paper: Legal and economic perspec tives,” Journal ofLaw & Economics, Vol. 24, Octo ber, pp. 541-588. __________ , 2001b, “Visa retreats on Interlink fee increase,” American Banker, October 9, p. 13. 56 Brito, Dagobert L., and Peter R. Hartley, 1995, “Consumer rationality and credit cards,” Journal of Political Economy, Vol. 103, pp. 400-433. 3Q/2002, Economic Perspectives Bullock, Michelle, and Luci Ellis, 1998, “Some fea tures of the Australian payments system,” Reserve Bank ofAustralia Bulletin, pp. 1-9. Food Marketing Institute, 2000, It All Adds Up: An Activity-Based Cost Study ofRetail Payments, Wash ington, DC. Chakravorti, Sujit, 2000, “Why has stored value not caught on?,” Federal Reserve Bank of Chicago, Emerg ing Issues Series, working paper, No. S&R-2000-6, May. __________ , 1998, EPS Costs: A Retailer’s Guide to Electronic Payment Systems Costs, Washington, DC. Chakravorti, Sujit, and William R. Emmons, 2001, “Who pays for credit cards?,” Federal Reserve Bank of Chicago, Public Policy Series, report, No. EPS2001-1, February. Chakravorti, Sujit, Jeffrey W. Gunther, and Rob ert R. Moore, 1999, “Cream skimming and the nonviability of universal access: A theory of the Federal Reserve’s evolving role in retail payments under MCA,” Federal Reserve Bank of Chicago, mimeo. Chakravorti, Sujit, and Alpa Shah, 2001, “A study of the interrelated bilateral transactions in credit card networks,” Federal Reserve Bank of Chicago, Public Policy Series, report, No. EPS-2001-2, July. Chakravorti, Sujit, and Ted To, 1999, “A theory of merchant credit card acceptance,” Federal Reserve Bank of Chicago, working paper, No. WP-99-16, November. Connolly, Paul, and Robert Eisenmenger, 2000, “The role of the Federal Reserve in the payments system,” presented at the conference “The Evolution of Monetary Policy and the Federal Reserve System Over the Past Thirty Years,” sponsored by the Feder al Reserve Bank of Boston, October. Evans, David S., and Richard L. Schmalensee, 1999, Paying with Plastic: The Digital Revolution in Buying and Borrowing, Cambridge, MA: MIT Press. Faulhaber, Gerald R., 1975, “Cross-subsidization: Pricing in public enterprises,” American Economic Review, Vol. 65, No. 5, pp. 966-977. Federal Reserve System, 2001, “Fed announces re sults of study of the payments system: First authori tative study in 20 years,” Washington, DC, press release, November 14. __________ , 1998, The Federal Reserve in the Pay ments Mechanism, prepared by the Committee on the Federal Reserve in the Payments Mechanism, Wash ington, DC. Federal Reserve Bank of Chicago __________ , 1994, Benchmarking Comparative pay ment Methods: Costs and Case Studies, Washington, DC. Frodin, Joanna H., 1984, “Fed pricing and the check collection business: The private sector response,” Busi ness Review, Federal Reserve Bank of Philadelphia, January/February, pp. 13-22. Gilbert, R. Alton, 1998, “Did the Fed’s founding im prove the efficiency of the United States payments system?,” Review, Federal Reserve Bank of St. Lou is, May/June, pp. 121-142. Green, Edward J., and Richard M. Todd, 2001, “Thoughts on the Fed’s role in the payments system,” Region: 2000 Annual Report, Federal Reserve Bank of Minneapolis, Vol. 15, No. 1, April, pp. 5-27. Green, Paul, 1999, Checks at the End of the 20th Century’ ... and Beyond!, Petaluma, CA: The Green Sheet, Inc. Guynn, Jack, 1996, “Improving the check system while planning its demise,” Financial Update, Federal Reserve Bank of Atlanta, October, pp. 1-2 and 5. Hirschman, Elizabeth, 1982, “Consumer payment systems: The relationship of attribute structure to preference and usage,” Journal ofBusiness, Vol. 55, No. 4, October, pp. 531-545. Humphrey, David B., and Allan Berger, 1990, “Market failure and resource use: Economic incen tives to use different payment instruments,” in The U.S. Payment System: Efficiency, Risk and the Role of the Federal Reserve, David B. Humphrey (ed.), Boston: Kluwer Academic Publishers, pp. 45-56. Humphrey, David B., Moshe Kim, and Bent Vale, 2001, “Realizing the gains from electronic payments: Costs, pricing, and payment choice,” Journal ofMoney, Credit, and Banking, Vol. 33, No. 2, May, pp. 216-234. Humphrey, David B., Lawrence Pulley, and Jukka Vesala, 2000, “The check’s in the mail: Why the Unit ed States lags in the adoption of cost-saving electronic payments,” Journal ofFinancial Sendees Research, Vol. 17, No. 1, February, pp. 17-39. 57 Kuprianov, Anatoli, 1985, “The Monetary Control Act and the role of the Federal Reserve in the inter bank clearing market,” Economic Review, Federal Reserve Bank of Richmond, July/August, pp. 23-35. Kuttner, Kenneth N., and James J. McAndrews, 2001, “Personal on-line payments,” Economic Policy Review, Federal Reserve Bank of New York, Decem ber, pp. 35-50. Lacker, Jeffrey, and John Weinberg, 1998, “Can the Fed be a payment system innovator?,” Economic Quarterly, Federal Reserve Bank of Richmond, Vol. 84, No. 2, pp. 1-25. Mantel, Brian M., 2000, “Why don’t consumers use electronic banking products? Towards a theory of ob stacles, incentives, and opportunities,” Federal Re serve Bank of Chicago, Public Policy Series, No. EPS-2000-1, September. Mantel, Brian, and Timothy McHugh, 2002, “Evolving e-payment networks: The strategic, com petitive, and innovative implications,” Payment Sys tems Worldwide, Spring. McAndrews, James J., 1997, “Network issues and payment systems,” Business Review, Federal Reserve Bank of Philadelphia, November/December, pp. 15-25. McHugh, Timothy, 2002, “The growth of person-toperson electronic payments,” Chicago Fed Letter, Federal Reserve Bank of Chicago, August, No. 180. Mester, Loretta, 2000, “The changing nature of the payments system: Should new players mean new rules?,” Business Review, Federal Reserve Bank of Philadelphia, March/April, pp. 3-26. Murphy, Michael M., and Mack Ott, 1977, “Retail credit, credit cards and price discrimination,” South ern Economic Journal, Vol. 43, No. 3, January, pp. 1303-1312. Murphy, Neil, 1988, “Determinants of household check writing: The impacts of the use of electronic banking services and alternative pricing of checking services,” Board of Governors of the Federal Reserve System, Finance and Economic Discussion Series, No. 3 8, August. 58 National Automated Clearing House Association, 2001, “News release: Commercial ACH payments increase by 14 percent in 2000,” Herndon, VA, April 13, available on the Internet at www.nacha.org/news/ news/pressreleases/2001/PR042301 b/ pr042301b.htm. Nilson, H. Spencer, pub., 2001a, The Nilson Report, Oxnard, CA, No. 742. __________ , 2001b, The Nilson Report, Oxnard, CA, No. 737. __________ , 2000a, The Nilson Report, Oxnard, CA, No. 726. __________ , 2000b, The Nilson Report, Oxnard, CA, No. 715. __________ , 1999a, The Nilson Report, Oxnard, CA, No. 705. __________ , 1999b, The Nilson Report, Oxnard, CA, No. 691. __________ , 1998, The Nilson Report, Oxnard, CA, No. 678. __________ , 1997a, The Nilson Report, Oxnard, CA, No. 654 __________ , 1997b, The Nilson Report, Oxnard, CA, No. 644. __________ , 1996a, The Nilson Report, Oxnard, CA, No. 627. __________ , 1996b, The Nilson Report, Oxnard, CA, No. 618. Palva, Marianne, 2000, “Developments in the Finn ish retail payment system,” Proceedings from the Workshop on Promoting the Use ofElectronic Pay ments: Assessing the Business, Technological, and Legal Infrastructures, Chicago, IL: Federal Reserve Bank of Chicago, pp. 31-40. Radecki, Lawrence, 1999, “Banks’payment-driven revenues,” Economic Policy Review, Federal Reserve Bank of New York, pp. 53-70. 3Q/2002, Economic Perspectives Rivlin, Alice M., 1997, “Role of the Federal Reserve in the payment system, appendix 2,” testimony be fore the Subcommittee on Domestic and Internation al Monetary Policy of the Committee on Banking and Financial Services, U.S. House of Representa tives, September 16. Robinson, Pal Erik, and Dag-Inge Flatraaker, 1995, “Costs in the payment system,” Economic Bul letin, Bank of Norway, Vol. 66, No. 2, pp. 201-216. Rochet, Jean-Charles, and Jean Tirole, 2000, “Co operation among competitors: The economics of pay ment card associations,” Institut D’Economie Industrielle, mimeo. Spiotto, Ann, 2001, “Credit, debit, or ACH: Conse quences & liabilities—A comparison of the differences in consumer liabilities,” American Bankers Associa tion Bank Compliance, September/October, pp. 4-11. Stavins, Joanna, 1999, “Checking accounts: Fees and features, consumer preference, impact on bank revenues,” New England Banking Trends, Federal Reserve Bank of Boston, Fall, No. 22. Federal Reserve Bank of Chicago Summers, Bruce J., and R. Alton Gilbert, 1996, “Clearing and settlement of U.S. dollar payments: Back to the future?,” Review, Federal Reserve Bank of St. Louis, September/October, pp. 3-27. Thomson Financial, 2001, Card Industry Directory 2002 Edition, Chicago. Thomson Media, 2001, “Competing in a checking ‘free for &W f American Banker, August 1, p. 1. Weinberg, John A., 1997, “The organization of private payment networks,” Economic Quarterly, Federal Re serve Bank of Richmond, Vol. 83, Spring, pp. 25-43. Wells, Kristen, 1996, “Are checks overused?,” Quar terly Review, Federal Reserve Bank of Minneapolis, Fall, pp. 2-12. Whitesell, William C., 1992, “Deposit banks and the market for payment media,” Journal ofMoney, Credit, and Banking, Vol. 24, No. 4, November, pp. 431^198. Wright, Julian, 2000, “An economic analysis of a card payment network,” Network Economics Con sulting Group and University of Aukland, mimeo. 59 Analyzing the relationship between health insurance, health costs, and health care utilization Eric French and Kirti Kamboj Introduction and summary In this article, we provide an empirical analysis of the determinants of whether an individual purchases health insurance coverage. We describe the relationship be tween health insurance, health costs, and health care utilization of the elderly, using data from the Health and Retirement Survey and the Assets and Health Dynamics among the Oldest Old. We show how health costs and health care utilization depend upon access to health insurance for individuals aged 50 and older. Given the public interest in extending health in surance coverage to those who are currently unin sured, it seems worthwhile to better understand why some people do not purchase health insurance. For example, 2000 Democratic presidential candidate A1 Gore advocated that individuals aged 55-64 be allowed to “buy in” to Medicare. The idea was that eligible individuals would have to pay for Medicare coverage, but would potentially pay less than the price of pri vately available insurance. Medicare would potentially be cheaper because of the cost advantages associated with the group coverage that Medicare provides. By understanding the determinants of the health insurance purchase decision, we can better understand how pro posed reforms may affect health insurance coverage. First, we investigate the factors influencing a per son’s decision to purchase health insurance. A General Accounting Office study found that in 1998, private health insurance premiums for a family of four ranged from $3,000 to $ 14,000 per year. Although health care coverage can be expensive, very few households are unable to buy private health insurance. Nevertheless, many households choose to be uninsured rather than purchase private health insurance.1 Therefore, we as sume that even low-income households are able to buy basic health insurance. Given that almost all individuals in our data are able to purchase health insurance, the most likely 60 reason that they remain uninsured is that they expect their health costs without insurance to be significant ly lower than their health costs with insurance. We test four potential reasons why this might be the case: 1) adverse selection in the insurance market—because insurers cannot distinguish between high-cost and low cost individuals in a group—leading to potentially prohibitive costs of health insurance for healthy indi viduals; 2) moral hazard—the idea that if the price of something is low, people use more of it—leading to potentially prohibitive costs of general insurance; 3) potentially prohibitive administrative costs of pro viding health insurance for private individuals; and 4) many of the uninsured already receive explicit insur ance through Medicaid and implicit insurance through hospitals that will treat indigent patients, which may obviate the need for them to purchase additional health insurance. Most studies of the health insurance purchase decision focus on the importance of adverse selection and moral hazard as potential reasons why individu als may not purchase insurance. Our results provide evidence that neither adverse selection nor moral hazard is the key determinant of the health insurance purchase decision. We find no evidence that adverse selection makes private insurance too expensive and only moderate evidence that moral hazard may make private health insurance prohibitively expensive. How ever, we find significant evidence that high adminis trative costs drive up the price of private insurance. Moreover, we find a large amount of evidence that the existence of Medicaid and implicit insurance ob viates the need for individuals to purchase additional health insurance. This last result suggests that changes Eric French is an economist at the Federal Reserve Bank of Chicago. Kirti Kamboj is a graduate student at the University of Chicago. 3Q/2002, Economic Perspectives in government-provided health insurance, such as al lowing younger individuals to “buy in” to the Medicare program, would likely have a small effect on the health insurance coverage of older Americans. The data show that many of those currently “uninsured” already have access to low- or no-cost health care coverage from the government and hospitals. Data: Health and Retirement Survey and Assets and Health Dynamics among the Oldest Old We use data from the Health and Retirement Sur vey (HRS) and Assets and Health Dynamics among the Oldest Old (AHEAD). These two datasets are collected by the same organization and have a similar sample design for much of the sample period. Both contain detailed information on health costs, health insurance, and demographics. The HRS is a sample of non-institutionalized2 individuals aged 51-61 in 1992. Spouses of these in dividuals were also interviewed, regardless of the spouse’s age. The HRS includes both a nationally rep resentative core sample and an additional sample of blacks, Hispanics, and Florida residents. A total of 12,652 individuals in 7,608 households were inter viewed in 1992 and re-interviewed in 1994, 1996,1998, and 2000, creating up to five separate responses for each individual. The AHEAD is a nationally representative sample of non-institutionalized individuals aged 70 and older in 1993. Like the HRS, spouses ofAHEAD respondents were also interviewed, regardless of age. Also like the HRS, the AHEAD dataset includes both a nationally representative core sample and additional samples of blacks, Hispanics, and Florida residents. A total of 8,222 individuals in 6,047 households were interviewed in 1993. These individuals were interviewed again in 1995, 1998, and 2000, creating up to four separate responses for each individual.3 In order to assess the quality of the HRS/AHEAD data, we present means of several key variables of in dividuals aged 50 and older and compare them with aggregated statistics from other sources.4 Consider sources of insurance first. Table 1 shows that most individuals receive employer-provided in surance, including insurance from current employers, past employers, and unions, as well as from the spouse’s current employer, past employers, and unions.5 Almost all individuals over age 65, as well as those who draw disability insurance, are eligible for Medicare. Individ uals with low incomes and asset levels are also eligible for Medicaid. Those not eligible for any of the above forms of insurance are faced with either purchasing Federal Reserve Bank of Chicago private health insurance or having no insurance at all. Table 1 shows that many individuals who do not have access to government- or employer-provided health insurance choose not to purchase insurance on the pri vate market. Of our sample, 17 percent have private insurance, while 7 percent have no insurance. Much of the remainder of this article is devoted to under standing the health insurance purchase decision for people who are neither covered by employers nor by the government. The central variable of interest in our study is the level of health costs paid by the household. For sin gle households, we compute this as the individual’s health costs. For married households, it is the sum of the husband’s and wife’s health costs. Health costs are the sum of insurance premiums, drug costs, and costs for hospital, nursing home care, doctor visits, dental visits, and outpatient care. See the appendix for a more detailed description of these variables. For our sam ple, mean household out-of-pocket health costs are $2,527 per year and mean health costs for those aged 65 and older are $2,716. The U.S. per capita average is $2,831 for non-institutionalized households headed by an individual aged 65 or older (Federal Interagen cy Forum, 2000). This means that health costs in the HRS/AHEAD are likely significantly below the na tional average when accounting for the institutional ized population. One important reason why average health costs in the HRS/AHEAD data are below the national av erage is that individuals in the HRS/AHEAD spend far fewer nights in a nursing home. Households head ed by someone aged 65 or older spend 7.2 nights in a nursing home per year in our sample versus 15.8 nights in the aggregate statistics (National Center for Health Statistics, 1999).6 Selden et al. (2001) find that 9 per cent of total aggregate health costs and 13 percent of costs paid out of pocket arise from nursing home visits. Why is there a market for health insurance? In the next two sections of this article, we describe some of the important determinants of the health in surance purchase decision. Then, we provide empiri cal evidence on these issues. The most obvious reason people purchase health insurance is to limit uncertainty associated with cata strophic health costs.7 The idea behind health insurance is that uncertain health expenditures are diversifiable risks. That is, health insurers provide health insurance to many individuals. While there is a great amount of uncertainty about how much insurers must pay out for any individual, there is very little uncertainty about aver age medical expenses for a large pool of individuals. 61 TABLE 1 Descriptive statistics Mean Standard deviation Observations Fraction with insurance plan Employer-provided Private None Medicaid Medicare 0.50 0.17 0.07 0.10 0.17 0.50 0.37 0.25 0.29 0.38 46,991 46,991 46,991 46,991 46,991 Medical costs (1998 dollars) Out of pocket costs Drug costs Insurance premiums Total household expenses 744 753 1,085 2,527 Variable actuarially fair insurance. As we noted in table 1, however, many people do not pur chase insurance. The most common ex planation why people do not buy insurance is that it is impossible to buy actuarially fair insurance. Next, we examine why this is so. Why doesn’t everyone purchase health insurance? Above, we argued that people should purchase health insurance to reduce un certainty if their expectation is that they will pay the same amount for health care Health care utilization whether or not they are insured. However, Nights in nursing home 4.09 34.65 42,638 insured individuals are, on average, likely Nights in hospital 1.69 6.13 42,418 Doctor visits 6.95 9.99 41,757 to pay more than the uninsured. Below, Had outpatient surgery 0.13 0.24 42,663 we highlight four reasons for this and cite Saw a dentist 0.65 0.48 36,315 existing evidence for each of the reasons. Did not take prescribed drugs 0.09 0.28 36,316 First, prices of health insurance may Demographics Fraction married 0.54 0.50 46,953 be potentially high because of adverse Good health 0.69 0.46 41,606 selection. Adverse selection occurs when Economic resources there are high health cost individuals and Assets <$50,000 0.32 0.47 45,627 low health cost individuals in a group, Assets >$50,000<$200,000 0.34 0.47 45,627 Assets >$200,000 0.34 0.48 45,627 but health insurers cannot distinguish be Income <$5,000 0.05 0.22 45,874 tween the two. Income >$5,000<$30,000 0.49 0.50 45,874 Income >$30,000 0.46 0.50 45,874 Recall that if markets are competi Working 0.38 0.48 46,442 tive and there are no administrative costs, insurers will set the price of health insur Sources: HRS/AHEAD data and authors’ calculations. ance equal to the average medical expen diture of individuals who purchase health Therefore, even if the health insurer is risk averse, by insurance. If individuals with low health costs are able pooling health costs of many individuals together, the to reveal that, on average, they will have low health insurer faces very little risk. As a result, the insurer costs, health insurers will charge those individuals low cares only about expected medical expenditures of insurance premiums. However, in practice it is very difficult for insur the individual when setting the insurance premium. ers to distinguish between the two groups. Individuals Suppose that the firm’s only cost of providing health insurance is medical expenditures. In other may know whether they are “high cost.” However, this words, we ignore administrative costs to the insurer. information is not available to the insurer of a group Also, assume that there are a large number of indi plan. For example, Blue Cross/Blue Shield health in viduals in the market, and that all of these individuals surance merely requests home address, date of birth, face the same distribution of health costs. If markets sex, whether the individual smokes, and whether the are perfectly competitive, the firm’s expected profit individual wants maternity coverage. is zero. If the insurer makes profits, new health insur As a result of not being able to distinguish between ance providers will enter the market and bid down high-cost and low-cost individuals, insurers charge insurance premiums to the expected health costs of everyone (conditional on the information listed imme the individual. Therefore, insurers will offer insurance diately above) the same price for health insurance.8 If only high health cost individuals purchase health in to individuals at “actuarially fair” prices, that is, prices equal to the expected health costs that individuals face. surance, and health insurers charge premiums equal Assuming that individuals are risk averse, they to average health costs of people who buy health in would rather pay their expected health costs than surance, then the cost will be relatively high. Although face the possibility of extremely high health costs. low health cost individuals may value health insurance Therefore, individuals will be better off purchasing at more than the cost to insurers of providing it to 62 2,516 2,523 3,197 5,057 41,876 41,807 34,251 33,005 3Q/2002, Economic Perspectives them, since they are risk averse, they may value it at less than what insurers charge to provide health in surance to high health cost people. In this scenario, the low health cost individuals will not purchase health insurance.9 If insurers could distinguish between high-cost and low-cost individuals, they would provide insur ance to low health cost individuals at a price equal to their expected health costs. This would make low cost individuals better off. Insurers would still charge high-cost individuals their expected health costs, making them no worse off. Cutler and Zeckhauser (2000) survey the evidence on adverse selection. They argue that empirical work has repeatedly documented its importance when com paring insurers that offer multiple plans. For example, individuals who opt for Medicare health maintenance organizations (HMOs) (that offer less generous ser vice than most Medicare plans but cover some servic es, like drug costs, that most Medicare plans do not cover) are more likely to have consumed few medical services in the past than those who do not opt for Medicare HMOs. Many researchers also cite the high price of pri vately provided health insurance as evidence that ad verse selection does drive up the price of health insurance. For example, Gruber and Madrian (1995) document that Blue Cross/Blue Shield health insur ance for a family of four in New England costs $10,310 in 1998 dollars.10 However, it is not clear in the above example that individuals who buy Blue Cross/Blue Shield are any less healthy than those who decide not to pur chase insurance. Moreover, many studies that consid er the health insurance purchase decision have found relatively little evidence that adverse selection exists in the market for health insurance (see Cardon and Hendel, 2001, for example). The second reason insurance may be so expen sive is the cost of administering plans for large em ployers. Administrative costs account for 10 percent to 15 percent of the costs of the health plans (Cutler and Zeckhauser, 2000). However, these costs are po tentially higher for insurance plans administered to small groups of people. For example a Congressional Budget Office (U.S. Congress, 1988) study found that large firms (10,000+ workers) pay 35 percent less than small firms (one to four workers). Gruber and Madrian (1995) argue that this price difference reflects some combination of adverse selection and administrative costs. Given that it is not obvious that adverse selec tion is more serious for small employers than large em ployers, it is likely that the cost difference is largely Federal Reserve Bank of Chicago from the lower administrative costs at large firms. Pauly (1986) finds that administrative costs may ac count for 50 percent of the cost of “Medigap” health insurance plans.11 Moral hazard is the third reason health insurance costs are high. Moral hazard is a consequence of downward sloping demand curves: If the price of a good becomes cheaper, people buy more of that good. People purchase health insurance to reduce the costs of medical procedures. For example, many “in demnity” plans like Blue Cross/Blue Shield allow people to obtain whatever health care they wish, but the insurer pays most of the price. If individuals have a 20 percent co-payment, then the price of medical services is only 20 percent of what it would be with out insurance. This potentially leads people to use medical services that are of very little value to them. Recall that if markets are competitive, then the price of health insurance is equal to expected medical ex penses of purchasers of health insurance plus admin istrative costs. The high level of medical services consumed by insured individuals will be reflected in the price of health insurance. Evidence from the RAND Health Insurance Ex periment (Manning et al., 1987) suggests that a 1 per cent rise in the price of health care services results in a .2 percent reduction in the quantity of health care services consumed, or a price elasticity of .2. Given that the price of health care services differs greatly between those with and without insurance, moral hazard potentially leads insured individuals to con sume far more medical services than is ideal, leading to expensive medical insurance. A final reason many people may find private medi cal insurance expensive is that they already receive insurance from the government or through hospitals. Medicaid provides insurance to individuals with low income and assets. Moreover, hospitals that receive federal funding cannot turn away indigent patients. Therefore, individuals with low income and assets do not need to purchase insurance. They already have it provided explicitly by Medicaid or implicitly by hos pitals. This explanation has received less attention than the other explanations (see Cutler and Gruber, 1996, for an exception). However, as shown in the empiri cal work below, this may be an important oversight. With these explanations in mind, table 2 describes the problems associated with universal government insurance relative to employer-provided health insur ance and private insurance. The main advantage of nationalizing health insurance, such as expanding Medicare to all individuals aged 55 and older, is to overcome adverse selection problems.12 Because the 63 plans. The Employee Benefit Research Institute (1999) reports that employers contribute an average of $3,288 to their employees’ health insurance. Therefore, the total cost of employer-provided insurance premiums is the sum of the employee contribution plus the em ployer contribution, or $4,442. Compare this amount with insurance premiums for households headed by individuals aged 50-64 with private insurance. These households spend $4,067, on average. This would imply that the total cost of a private plan is slightly less than the cost of an employer-pro vided plan.13 However, not only do households with private insurance spend more on insurance premiums than households with employer-provided insurance, they also have higher out-of-pocket expenses. This may reflect the higher deductibles and co-pays of private health insurance policies. When we sum up the insur ance premiums paid by the individual and the firm plus what the individual pays out of pocket, the total Health insurance coverage, health costs, health cost for households with employer-provided and health care utilization insurance is $5,489 (a $3,288 employer contribution In this section we provide some new empirical plus total household expenses of $2,201) and the total evidence on the four potential reasons individuals do cost for private insurance is $5,871, a difference of 7 not purchase health insurance We find no evidence of percent. Given that administrative costs constitute adverse selection and limited evidence in favor of moral about 13 percent of insurance costs for employer-pro hazard and high administrative costs. Instead, we be vided plans, these costs make up 20 percent of private lieve the main reason some individuals do not purchase insurance costs. insurance is that they are already receiving insurance, Moreover, table 3 shows that households with either through the government or implicitly through private insurance receive fewer medical procedures hospitals. than households with employer-provided health insur In order to assess the importance of administra ance. This may reflect the fact that private insurance tive costs, we compare individuals with private insur does not usually cover pre-existing conditions. Given ance with individuals with employer-provided insurance. that those with private health insurance consume few Recall that Cutler and Zeckhauser (2000) find that er medical services than those with employer-provid administration accounts for 10 percent to 15 percent ed insurance, the cost (net of administrative cost) of of the total cost of health insurance at large firms. Our private health insurance is likely lower than the cost goal is to find out whether individuals who purchase (net of administrative cost) of employer-provided in private insurance face significantly higher administra surance. Therefore, the calculation of administrative tive costs than those who receive health insurance costs above likely understates the administrative cost through their employer. of private health insurance. Table 3 shows household health costs by age group The second potential reason people do not purchase and health insurance type. For households headed by health insurance is adverse selection, which implies someone aged 50-64, health insurance premiums are that only the most unhealthy purchase private insurance, $ 1,154 per year for those with employer-provided which makes premiums prohibitively expensive for healthy people. However, the evidence presented in table 3 refutes this explana TABLE 2 tion. Fully 81 percent of people aged Problems with health insurance, by payment system 50-64 with private insurance report that Employerthey are in good health. However, only Nationalized provided Private 65 percent of the uninsured do likewise. Therefore, the uninsured are more likely Administrative costs yes yes yes Moral hazard yes yes yes to be unhealthy than those who purchase Adverse selection no some yes private insurance. Comparing those older government would expand coverage to everyone, both the healthy and unhealthy would be covered. Indeed, Akerlof (1970) points out that most individuals aged 65+ were uninsured before Medicare was passed into law and argues that adverse selection was one reason for the low insurance rates of these people. However, nationalizing health care would do little if anything to overcome high administrative costs or moral hazard. Administrative costs would potentially be the same for insurance plans administered by large private in dustries and the government. And moral hazard is in herent in the very nature of insurance contracts and is not specific to the insurance provider. Therefore, ar guments in favor of nationalizing health insurance must rest on the assumption that adverse selection exists in the marketplace for health insurance coverage and partly on the assumption that, because of risk aver sion, health insurance makes people better off. 64 3Q/2002, Economic Perspectives TABLE 3 Descriptive statistics by age group Employerprovided A. Ages 50-64 Fraction with insurance plan Medical costs (1998 dollars) Non-drug out-of-pocket costs Drug costs Insurance premiums Total household expenses Health care utilization Nights in nursing home Nights in hospital Doctor visits Had outpatient surgery Saw a dentist Did not take prescribed drugs Demographics Fraction married Good health Economic resources (1998 dollars) Assets <$50,000 Assets >$50,000<$200,000 Assets >$200,000 Income <$5,000 Income >$5,000<$30,000 Income >$30,000 Working B. Ages 65-79 Fraction with insurance plan Medical costs (1998 dollars) Non-drug out-of-pocket costs Drug costs Insurance premiums Total household expenses Health care utilization Nights in nursing home Nights in hospital Doctor visits Had outpatient surgery Saw a dentist Did not take prescribed drugs Demographics Fraction married Good health Economic resources (1998 dollars) Assets <$50,000 Assets >$50,000<$200,000 Assets >$200,000 Income <$5,000 Income >$5,000<$30,000 Income >$30,000 Working None 0.67 0.09 0.12 0.07 0.04 659 513 1,154 2,201 1,049 709 4,067 5,871 599 585 110 1,277 224 465 53 712 994 1,354 366 2,792 0.241 1.067 6.415 0.133 0.806 0.061 0.008 0.809 5.702 0.115 0.769 0.084 0.627 0.932 4.532 0.055 0.487 0.209 5.350 3.146 10.492 0.094 0.385 0.205 1.527 2.791 9.608 0.118 0.469 0.311 0.683 0.822 0.583 0.809 0.451 0.646 0.201 0.235 0.454 0.332 0.229 0.392 0.379 0.020 0.266 0.714 0.729 0.197 0.252 0.550 0.058 0.320 0.622 0.653 0.572 0.285 0.143 0.189 0.552 0.257 0.554 0.857 0.122 0.019 0.275 0.692 0.031 0.071 0.631 0.265 0.102 0.148 0.706 0.146 0.072 0.37 0.22 0.01 0.10 0.30 732 669 1,255 2,601 651 1,394 2,408 4,329 605 880 211 1,657 333 585 149 1,032 647 1,055 433 2,088 1.135 1.845 7.776 0.174 0.762 0.040 0.798 1.835 7.260 0.159 0.701 0.095 0.693 2.133 6.336 0.077 0.463 0.136 10.005 3.394 9.305 0.102 0.384 0.139 1.921 1.690 6.650 0.120 0.594 0.110 0.662 0.735 0.581 0.723 0.492 0.583 0.273 0.396 0.509 0.650 0.153 0.357 0.489 0.010 0.429 0.561 0.201 0.156 0.343 0.501 0.012 0.528 0.460 0.221 0.468 0.321 0.205 0.174 0.626 0.200 0.176 0.760 0.186 0.049 0.122 0.831 0.047 0.059 0.323 0.369 0.306 0.031 0.661 0.308 0.170 than 65 who purchase private Medigap health insur ance with those who only have Medicare or who have no health insurance at all, again we see that those who purchase private insurance are healthier than those with no insurance other than Medicare.14 The third potential reason individuals may not purchase health insurance is the moral hazard prob lem. Those who are insured face a low price of health care services, so they tend to consume more, which drives up the price of premiums. Therefore, controlling Federal Reserve Bank of Chicago Medicare Private Medicaid for health status, those who have private health insur ance should consume more health care services than those who have no insurance. Table 3 shows that for households headed by someone aged 50-64, those with private health insurance are the least likely to spend a night in a nursing home or a hospital. Those 50-64 with private insurance spend .01 nights in a nursing home and .8 nights in a hospital per year, on average. Those without insurance spend .6 nights in a nursing home and .9 nights in a hospital per year, on average. 65 TABLE 3 (continued) Descriptive statistics by age group Employerprovided C. Ages 80 and older Fraction with insurance plan Medical costs (1998 dollars) Non-drug out-of-pocket costs Drug costs Insurance premiums Total household expenses Health care utilization Nights in nursing home Nights in hospital Doctor visits Had outpatient surgery Saw a dentist Did not take prescribed drugs Demographics Fraction married Good health Economic resources (1998 dollars) Assets <$50,000 Assets >$50,000<$200,000 Assets >$200,000 Income <$5,000 Income >$5,000<$30,000 Income >$30,000 Working Medicare Private None 0.23 0.28 0.01 0.16 0.32 1,773 765 832 3,198 1,241 1,230 2,033 4,431 477 1,086 73 1,141 646 391 165 1,123 1,018 944 345 2,281 16.168 2.638 7.446 0.151 0.631 0.019 12.622 2.366 7.094 0.132 0.558 0.055 25.198 1.984 6.055 0.102 0.422 0.094 44.476 3.275 8.442 0.080 0.274 0.060 12.279 1.975 6.146 0.099 0.454 0.066 0.369 0.560 0.309 0.592 0.294 0.564 0.140 0.359 0.266 0.586 0.214 0.375 0.410 0.010 0.640 0.350 0.018 0.241 0.376 0.382 0.024 0.719 0.257 0.044 0.433 0.328 0.239 0.209 0.657 0.134 0.029 0.804 0.154 0.039 0.123 0.867 0.010 0.008 0.391 0.341 0.265 0.056 0.789 0.155 0.040 Medicaid Sources: HRS/AHEAD data and authors’ calculations. These findings are not consistent with the moral haz ard explanation. However, those with private insurance on average have more doctor visits, are more likely to have outpatient surgery, are more likely to see a dentist, and are less likely to not take prescribed drugs than those without insurance. These findings are con sistent with moral hazard. These patterns hold when com paring those older than 65 with private insurance with those older than 65 who have either Medicare insurance or no insurance. The privately insured older than 65 are less likely to spend time in a nursing home or in a hospital, but have more doctor visits, are more likely to have outpatient surgery, are more likely to see a dentist, and are less likely to not take prescribed drugs than those without insurance or those who only have Medicare. One possible reason those with private insurance are less likely to spend the night in a nursing home or a hospital than those who are uninsured is that the pri vately insured are healthier. For two people with equal health, the person with private insurance is potential ly more likely to spend time in a nursing home or a hospital than the person without insurance. We return to this issue when conducting our multivariate analy sis in the next section. Another possible reason that complements the previous explanation is that hospitals may have a difficult time turning away those without 66 insurance who are very ill. However, it is easy for dentists and doctors offering elective surgery to turn away the uninsured. The final explanation why people do not purchase health insurance is that they receive implicit insurance through the government and hospitals. One testable implication of this hypothesis is that those without insurance pay less for a unit of health care services than those with insurance. As discussed previously, table 3 provides evidence that those without health insurance consume only slightly fewer health care services than those with private insurance. Note, how ever, that those aged 50-64 with no insurance spend only $ 1,277 per year on health care, versus $5,871 per year for those with private insurance. This difference in costs is not completely an artifact of differences in in surance premiums either. Those who have no insur ance spend less on out-of-pocket expenses such as drugs and co-pays than those with private insurance. We also note that households with no insurance are more likely to have low assets and low income than those with private insurance. This is important for two reasons. First, if an individual is indigent, public and non-profit hospitals must treat them. There fore, low-asset individuals have implicit insurance through hospitals. Second, individuals with low assets are potentially eligible for Medicaid. 3Q/2002, Economic Perspectives Individuals receiving Medicaid insurance con sume more medical services and spend less on health care than any other group. That expenditures by Medicaid beneficiaries, who have low income and assets, are low is not surprising. After all, the govern ment spent $10,243 per Medicaid beneficiary aged 65 or older and $9,097 per blind or disabled individ ual in 1998 (U.S. House of Representatives, 2000). One way to test whether Medicaid is a significant source of insurance for the uninsured is to estimate the probability that a household that is uninsured be comes covered by Medicaid health insurance two years later. For uninsured households headed by someone aged 50-64, there is a 9.2 percent probability that they will be covered by Medicaid two years later. For house holds that purchase private insurance, there is only a 2.6 percent probability that they will be covered by Medicaid two years later. This shows that individuals with “no” insurance are more likely to be covered in the near future. This may mean that individuals who believe that they will be eligible for Medicaid in the event of a household emergency feel less compelled to purchase private health insurance than those who do not believe that they will be eligible for Medicaid. Multivariate analysis of determinants of health costs As noted above, we find that the total cost of employer-provided insurance plans (that is, the sum of costs paid by both employees and employers) is similar to that of private plans for households headed by individuals younger than 65. We also find that those with no insurance pay much less for medical care than those with private insurance. However, these compari sons are difficult to interpret because there may be important differences in the quality of care provided across health care plans. Even though having private insurance leads to higher health costs than having no insurance, having private insurance may also lead to a significantly higher quality of care. In this section, we use multivariate regressions to control for the quality of health care received. Although we cannot control for all aspects of health care quality, we can control for many of the determinants of health costs, such as nights in a hospital or nursing home. In this analysis, we aim to explain differences in costs of different types of health insurance, controlling for health care utilization. Table 4 presents estimates of some of the deter minants of health costs for the three age groups. Each age group has two columns, the first one with the health care utilization and health status measures and the second one without. By controlling for health Federal Reserve Bank of Chicago care utilization in the regressions, we can assess whether differences in health care utilization explain differ ences in health costs among households with differ ent types of insurance. First, we need to infer administrative costs. Re call that although the total cost of employer-provided insurance is similar to that of private plans for house holds headed by an individual younger than 65, house holds with employer-provided insurance consume more health care services than households with private in surance. Here, we assess whether controlling for the quantity of health care services consumed affects the estimated cost differences for the two groups. We are interested in whether the total cost of private insurance is greater than the total cost of employer-provided in surance, holding utilization constant. Column 2 in each category of table 4 provides evidence on this. Controlling for the health utilization variables those with employer-provided insurance pay $324 more than those with no insurance.15 Those with private health insurance plans pay $4,132 more than those who are uninsured and ($4,132 - $324 =) $3,808 more than those with employer-provided health insurance. Recall that firms contribute about $3,288 toward em ployees’ health insurance plans. Therefore, the total cost of obtaining private health insurance is only $520 greater than the cost of obtaining private health insur ance. This estimate is not much different from the difference in mean health costs described in the pre vious section. Therefore, our estimate of administra tive costs of private health insurance above is little changed by the multivariate analysis. Another finding in the previous section is that those who are uninsured pay much less for health care than those who purchase private insurance. However, those who purchase private insurance are also more likely to consume certain medical services, such as dentist visits. Controlling for assets, income, education, race, marital status, age, and health care utilization does not affect the difference in health costs between those who are insured and those who are uninsured. The gap between health costs for households that are privately insured and uninsured is $4,132, almost the same as the difference in mean health costs shown in table 3. There fore, our central finding, that the uninsured are implic itly insured by hospitals and the government, is not overturned by the multivariate regression analysis. As further evidence on the hypothesis that “im plicit” insurance is important, it appears that greater household resources lead to greater health costs, even after controlling for health care utilization and health insurance. This is true when the proxy for household resources is assets, income, or education. In other words, 67 TABLE 4 Determinants of medical costs Ages 65-79 Ages 50-64 1 2 Insurance type Employer-provided 430 (99) 324 (122) Private 4,057 (135) 4,132 (160) Medicaid -172 (140) -550 (165) Medicare 1,237 (168) 927 (197) Demographics High school graduate 225 (80) 161 (98) College graduate 528 (105) 361 (130) Age 58 (9) 48 (11) Black -34 (83) 7 (100) Married 1,751 (72) 1,623 (90) Good health 159 (83) 316 (101) Economic resources (1998 dollars) Assets >$50,000<$200,000 531 (97) 736 (120) Assets >$200,000 32 (126) 26 (152) Income >$5,000<$30,000 55 (144) 11 (175) Income >$30,000 -260 (77) 12 (95) Working -2,937 (555) -2 (2) Health care utilization Nights in nursing home 66 (8) Nights in hospital 28 (4) Doctor visits 860 (176) Had outpatient surgery 185 (92) Saw a dentist 830 (126) Did not take prescribed drugs -2,784 (684) Constant R2 0.148 0.180 1 Ages 80+ 1 2 2 240 2,037 -241 110 (418) (421) (426) (416) 161 1,892 -653 -26 (435) (437) (442) (431) 1,423 2,681 47 929 (911) 1,121 (905) 2,524 (912) -681 (901) 787 (937) (932) (940) (927) 241 521 46 -130 1,848 -32 (107) (153) (10) (132) (100) (120) 228 490 31 -89 1,499 70 (111) (159) (11) (137) (108) (125) 580 461 114 -794 1,987 -353 (196) 400 (332) 380 (20) 78 (251) -783 (231) 1,358 (216) -37 (196) (329) (20) (250) (243) (215) 344 30 376 -38 -2,455 (142) (247) (272) (122) (874) 484 -67 247 45 19 (149) (255) (281) (126) (2) -516 (268) -231 580 (387) 550 2,038 (477) 1,854 -1,065 (499) -680 -9,237 (1,935) 17 (267) (387) (474) (485) (1) 23 41 662 421 1,216 -1,931 (7) (4) (196) (107) (162) (904) 0.091 0.121 44 (14) 77 (11) 259 (389) 211 (185) 1,531 (377) -6,989 (1,961) 0.065 0.117 Notes: For each age category, column 1 shows results with no health utilization controls, while column 2 shows results with controls. Numbers in parentheses are standard errors. Sources: HRS/AHEAD data and authors’ calculations. poor people seem to pay less than rich people for the same medical services. One caveat to this last finding is that more affluent households may be paying more for medical care because they are purchasing better care. For example, they are potentially spending nights in better hospitals and seeing better doctors. Conclusion This article provides an empirical analysis of the determinants of whether an individual purchases health insurance coverage. Using data from the Health and Retirement Survey and the Assets and Health Dynam ics among the Oldest Old, we document the distribu tion of health costs and health care utilization of individuals aged 50 and older. We show how health costs and health care utilization depend upon access to health insurance. 68 Of our sample, 6.5 percent are uninsured. We con sider four potential reasons for this. We show that ad verse selection is unlikely to be an important factor in driving health insurance costs. There is some evidence that administrative costs and overconsumption due to moral hazard raise the cost of health insurance. How ever, we find a large amount of evidence that the ex istence of Medicaid and implicit insurance obviates the need for individuals to purchase additional health insurance. Given that many of the “uninsured” already have access to low- or no-cost health care coverage from the government and hospitals, we argue that Medicare buy-in proposals (that is, proposals to allow younger individuals to pay a premium to join Medicare) would most likely have a small effect on the health insurance coverage of older Americans. 3Q/2002, Economic Perspectives APPENDIX: CODING TWO HEALTH COST VARIABLES This appendix describes the coding of the two main health cost variables used in our analysis: medical costs paid by the individual and total costs of medical treat ment. Medical costs paid by the individual are equal to the sum of drug costs, out-of-pocket expenses on items other than drugs, and insurance costs. The total cost of medical treatment is the total cost of medical services (whether or not the individual pays for those services) plus drug costs. During waves one and two, members of the HRS and AHEAD samples were asked different sets of ques tions. Wave three includes only HRS respondents. Dur ing waves four and five, HRS and AHEAD respondents were asked the same questions. Health cost information for wave one of the HRS is limited to insurance premiums. The insurance pre mium question only refers to insurance purchased directly from an insurance company or through a mem bership organization, such as the American Associa tion of Retired Persons (AARP). It does not include employee contributions to employer-provided insur ance plans. Given that the information is incomplete, we do not include wave one information. From wave two onward of the HRS, however, we are able to com pute the total cost of medical treatment, the out of pocket expenses, the drug costs, the cost of the insur ance, and the total medical costs the individual pays. In wave two, the insurance premium question in cludes employee contributions to employer-provided insurance plans and insurance directly purchased or through a membership organization. In wave two, respondents were asked whether they had any hospital stays, nursing home stays, or visits to a doctor. If they answered yes to any of these ques tions, they were then asked both the total cost and out of pocket cost for the visit or stay. Also in wave two, respondents were asked wheth er they purchased medicines prescribed by a doctor. If they did, they were asked how much these medicines Federal Reserve Bank of Chicago cost per year. It is not clear whether the cost measure refers to the cost paid by the individual or what the pharmacy charges the individual and the insurer. Nevertheless, we use this variable to determine the drug costs variable, which is also added into the total costs variable. For wave three onward, total health costs and out of pocket costs are clarified to include the amount paid for doctors, hospitals, nursing homes, outpatient surgery, dental expenses, in-home medical care, and special facilities and services. Note that the costs (both out-of-pocket and total) of outpatient surgery, dental expenses, in-home care, and special facilities and ser vices are not included in wave two. The procedure for determining the insurance costs also changes for HRS waves three through five. The insurance premium variable is now the sum of premi ums of all employer-provided insurance, Medicare through HMO plans, supplemental plans, private/AARP/ professional coverage, and long-term care plans. Note that in wave two, Medicare insurance costs are missing. For wave one of the AHEAD dataset, we can determine the out-of-pocket and total costs. The outof-pocket costs include the costs of nursing home stays as well as “any part of hospital and doctor bills and any other medical or dental expenses in the last 12 months.” We infer that drug costs are included in this measure, although respondents were not asked directly about drug costs. AHEAD wave one also asked what policies besides Medicare respondents have, includ ing long-term care policies, and how much they paid yearly for such policies. From this, we determine the insurance costs of the respondent. We do not use wave one in our analysis in this article, however, be cause the source of health insurance is incomplete. For wave two of the AHEAD dataset, imputation procedures for total costs, and out of pocket costs, in surance costs, drug costs, and medical costs are the same as in waves three through five of the HRS. 69 NOTES ’Blue Cross/Blue Shield is willing to cover most people, although its plans often do not cover pre-existing conditions. institutionalized individuals include individuals in nursing homes. 3In 1998 and 2000, individuals in the HRS and AHEAD (as well as an additional sample of older individuals) were asked the same questions. In the HRS and AHEAD waves before 1998, many of the questions asked were the same across the two datasets, allow ing us to merge the datasets. Because the health insurance and health cost data are incomplete in wave one of both datasets, we use waves two through five in our analysis here. "Health costs and health care utilization in the survey instrument are for the past two years or since the individual was last inter viewed. We divide health costs and health care utilization mea sures by the number of years since the individual was last interviewed, or by two if the individual was never previously interviewed. We dropped individuals with missing information on their health in surance coverage. This reduced the original sample from 49,843 person-year observations to 46,991 person-year observations. How ever, we kept individuals with missing information on other vari ables. There are 33,005 observations with health cost information. 5Employer-provided insurance includes individuals with federally provided health insurance plans through the Veterans Administra tion and the Post Office. It does not include Medicare or Medicaid. Individuals with federally provided health insurance also have rather similar characteristics to people with employer-provided insurance. The main difference is that households with federal insurance tend to spend less on insurance premiums. 6B ecause the HRS/AHEAD sample was drawn from the non-institutionalized population—which excludes individuals in nursing homes—it is not surprising that the number of nights in a nursing home is lower in the HRS/AHEAD sample than the national aver age. Nevertheless, many HRS/AHEAD household members do enter a nursing home after they are initially interviewed. 7There are other explanations for why people purchase health in surance. One is that people who purchase health insurance through their employer pay for it with pre-tax dollars. If individuals face a 30 percent tax rate, they will be indifferent between spending $1,000 on medical expenses pre-tax and $700 post-tax. Unless they have very large medical expenses, making them eligible to deduct medi cal expenses on taxes, they will pay for any medical procedures not covered by employer-provided health insurance on a post-tax 70 basis. Therefore, there are important tax advantages to employerprovided insurance. Another reason is that health maintenance organizations potentially have market power in the market for medical services and thus can bargain with hospitals for lower prices. ^ote, however, that pre-existing conditions, such as cancer, are usually not covered by private health insurance plans. 9Note that if everyone purchased health insurance, average medi cal expenses would fall. Therefore, if low-cost individuals value health insurance at more than the cost of providing insurance to them, two equilibriums potentially exist: one in which health in surance is purchased only by high-cost individuals (resulting in high health insurance premiums), the other in which health insur ance is purchased by everyone and the price of health insurance premiums lies between the costs of high- and low-cost individuals. 10Gruber and Madrian (1995) document that Blue Cross/Blue Shield was $8,640 in 1993. We adjusted this number to 1998 dollars us ing the medical care component of the Consumer Price Index. 1’Medigap plans are private health insurance plans that cover health costs that Medicare does not cover, such as co-pays. Note that Medigap plans likely have high administrative costs because Medigap only pays for the smaller expenses that Medicare does not cover. 12However, Gore’s “buy in” proposal still might have had problems, given that the buy in would be voluntary. Healthy individuals would potentially not buy into Medicare. ^Nevertheless, $4,067 is surprisingly small given that Gruber and Madrian (1995) found that Blue Cross/Blue Shield charged over $10,000 in 1998 dollars for private health insurance for a family of four in New England. 14Note that differences in health costs amongst the different health insurance groups fall greatly after age 65. Private health insurance premiums fall from $4,067 for those aged 50-64 to $2,408 for those aged 65-79. This is not surprising given that after age 65, Medicare becomes the primary source of health insurance. Private insurance pays many of the costs that Medicare does not pay. This is why these health insurance plans are also often referred to as “Medigap” plans. 15The omitted category in these regressions is an uninsured, lowasset, low-income, white individual, who dropped out of high school. 3Q/2002, Economic Perspectives REFERENCES Akerlof, G., 1970, “The market for ‘lemons’: Quality uncertainty and the market mechanism,” Quarterly Journal ofEconomics, Vol. 84, No. 3, pp. 488-500. Arrow, K., 1963, “Uncertainty and the welfare economics of medical care,” American Economic Review, Vol. 53, No. 5, pp. 941-973. Cardon, J., and I. Hendel, 2001, “Asymmetric information in health insurance: Evidence from the National Medical Expenditure Survey,” RAND Journal ofEconomics, Vol. 32, No. 3, pp. 408-427. Cutler, D., and J. Gruber, 1996, “Does public insurance crowd out private insurance?,” Quarterly Journal ofEconomics, Vol. Ill, No. 2, pp. 391-430. __________ , 1995, “Health Insurance availability and the retirement decision,” American Economic Review, pp. 938-948. Gustman, A., and T. Steinmeier, 1999, “Effects of pensions on savings: Analysis with data from the health and retirement study,” Carnegie-Rochester Conference Series on Public Policy, pp. 271-324. Hubbard, R., J. Skinner, and S. Zeldes, 1995, “Precautionary saving and social insurance,” Journal ofPolitical Economy, Vol. 103, pp. 360-399. __________ , 1994, “The importance of pre-cautionary motives in explaining individual and aggregate saving,” Carnegie-Rochester Series on Public Policy, pp. 59-125. Cutler, D., and R. Zeckhauser, 2000, “The anatomy of health insurance,” in Handbook ofHealth Economics, A. Culyer and J. Newhouse (eds.), New York: John Wiley and Sons, pp. 563-643. Manning, Willard G., Joseph P. Newhouse, Naihua Duan, Emmett B. Keeler, Arleen Leibowitz, 1987, “Health insurance and the demand for medical care: Evidence from a randomized experiment,” American Economic Review, Vol. 77, No. 3, June, pp. 251-277. Employee Benefit Research Institute, 1999, EBRI Health Benefits Databook, Washington, DC. National Center for Health Statistics, 1999, “Health, United States, 1999,” Washington, DC, report. Federal Interagency Forum on Aging-Related Statistics, 2000, “Older Americans 2000: Key indicators of well-being,” Hyattsville, MD, report. Palumbo, M., 1999, “Uncertain medical expenses and precautionary saving near the end of the life cycle,” Review ofEconomic Studies, Vol. 65, No. 2. Feenberg, D., and J. Skinner, 1994, “The risk and duration of catastrophic health care expenditures,” Review ofEconomics and Statistics, Vol. 76, No. 4, pp. 633-647. Pauly, Mark, 1986, “Taxation, health insurance, and market failure,” Journal ofEconomic Literature, Vol. 24, No. 2, pp. 629-675. French, E., and J. Jones, 2002a, “The effects of health insurance and self-insurance on retirement behavior,” Federal Reserve Bank of Chicago, mimeo. _________ , 2002b, “On the distribution and dynamics of health costs,” Federal Reserve Bank of Chicago, mimeo. Gruber, J., and B. Madrian, 1996, “Health insurance and early retirement: Evidence from the availability of continuation coverage,” in Advances in the Economics ofAging, D. A. Wise (ed.), Chicago, University of Chicago Press, pp. 115-143. Federal Reserve Bank of Chicago Rust, J., and C. Phelan, 1997, “How Social Security and Medicare affect retirement behavior in a world of incomplete markets,” Econometrica, Vol. 65, pp. 781-831. Selden, T., L. Levit, J. Cohen, S. Zuvekas, J. Moeller, D., McKusik, and R. Arnett, 2001, “Reconciling medical expenditure estimates from the MEPS and NHA, 1996,” Health Care Financing Review, Not. 23, No. 1. Smith, J., 1999, “Healthy bodies and thick wallets: The dual relationship between health and economic status,” Journal ofEconomic Perspectives,Not. 13, No. 2, pp. 145-166. 71 U.S. Congress, Congressional Budget Office, 1988, Costs and Effects ofExtending Health Insurance Coverage, Washington, DC. U.S. Congress, General Accounting Office, 2000, Private Health Insurance: Potential Tax Benefit of a Health Insurance Deduction Proposed in HR. 2990, No. GAO/HEHS-00-104R, April 21. 72 U.S. House of Representatives, Committee on Ways and Means, 2000, The 2000 Green Book: Background Material and Data on Programs within the Jurisdiction of the Committee on Ways and Means, Washington, DC: Government Printing Office, No. WMCP 106-14. 3Q/2002, Economic Perspectives