oversight

Consumer Price Index: More Frequent Updating of Market Basket Expenditure Weights Is Needed

Published by the Government Accountability Office on 1997-10-09.

Below is a raw (and likely hideous) rendition of the original report. (PDF)

                   United States General Accounting Office

GAO                Report to the Ranking Minority Member,
                   Committee on Banking and Financial
                   Services, House of Representatives


October 1997
                   CONSUMER PRICE
                   INDEX
                   More Frequent
                   Updating of Market
                   Basket Expenditure
                   Weights Is Needed




GAO/GGD/OCE-98-2
      United States
GAO   General Accounting Office
      Washington, D.C. 20548

      General Government Division

      B-275450

      October 9, 1997

      The Honorable Henry B. Gonzalez
      Ranking Minority Member
      Committee on Banking and Financial Services
      House of Representatives

      Dear Mr. Gonzalez:

      The principal source of information on trends in consumer prices and
      inflation in the United States is the Consumer Price Index (CPI), according
      to the Bureau of Labor Statistics (BLS), which publishes the index.1 In fiscal
      year 1996, about $656 billion of federal tax receipts and $458 billion in
      federal spending were automatically linked to price changes measured by
      the CPI. The CPI tracks prices for a fixed “market basket” of goods and
      services that people buy for day-to-day living. Since 1940, BLS has made
      major revisions to the market basket about once a decade to reflect
      changes in what consumers buy.2

      This report responds to your request that we examine certain questions
      surrounding the issue of revising the market basket more often. As agreed
      with your office, rather than evaluating possible alternatives to the CPI’s
      basic formula or examining the whole process of making major revisions
      to the CPI,3 we focused on the market basket’s expenditure weights and
      whether they could be updated between major revisions to the CPI.4
      Historically, the expenditure weights have been changed only during
      major CPI revisions. Unlike in a major revision, the principal task in an
      update would be to change—make more current—the expenditure
      weights that had been determined during the last major revision. In a
      revision, BLS has usually changed (1) the CPI’s geographic areas and
      housing samples, which reflect where consumers live and buy goods and
      services; (2) the computer systems for processing these data; and (3) the

      1
       BLS is a part of the U.S. Department of Labor.
      2
       There has not been a uniform number of years between major revisions to the CPI’s market basket.
      Although a revision was made each decade, the number of years between revisions ranged from 9
      years to 14 years for the four revisions that occurred since 1940.
      3
       A commission chartered by the U.S. Senate has proposed an alternative formula that is intended to
      make the CPI more reflective of the cost of living. We present information later in this report about the
      alternative, which is referred to as a superlative index.
      4
       Weights allow BLS to specify the importance of the items included in the CPI market basket and
      provide appropriate emphasis to the price changes associated with those items. For example, if ground
      beef were assigned a weight representing about one-third of 1 percent of the expenditures of the
      typical urban consumer and if sirloin steak were assigned a smaller weight representing less than
      one-tenth of 1 percent, then the price changes of ground beef would have about 3 times as much
      impact on the overall CPI as similar price changes for sirloin steak.



      Page 1                                  GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                   B-275450




                   expenditure weights of the market basket items. (App. II provides
                   additional information on how the CPI is constructed and on how weights
                   are calculated.)

                   As agreed more specifically, we focused on the market basket’s
                   expenditure weights by (1) obtaining the views of individuals who were
                   knowledgeable of the CPI on updating the weights between major revisions
                   to the CPI and the practices followed by other industrialized countries in
                   updating their consumer price indexes, (2) estimating the additional cost
                   to BLS to update the weights on a 5-year cycle,5 (3) estimating the dollar
                   effect on the federal budget if the weights were updated on a 5-year cycle,
                   and (4) identifying and assessing BLS’ reasons as to why updates of the
                   weights have only occurred during major revisions to the CPI, which have
                   been about every 10 years.


                   The weight of professional opinion supported updating the market
Results in Brief   basket’s expenditure weights more frequently than major revisions to the
                   CPI have been made. We spoke with 10 individuals who were
                   knowledgeable about the CPI, and they were unanimous in believing that 10
                   years between updates was too long to reflect “current” consumer
                   spending. Two of the 10 individuals were former BLS officials, and the 8
                   others had conducted research on the CPI, including 4 who were members
                   of the Advisory Commission to Study the Consumer Price Index (hereafter
                   called the Boskin commission).6 There was less agreement among the 10
                   individuals, however, on exactly how often updates should occur. Five of
                   them, including the four Boskin commission members with whom we
                   spoke, said more frequent updating of expenditure weights was less
                   important than other ways of making the CPI more reflective of current
                   consumer spending.

                   Other major industrial countries update their consumer price indexes
                   more often than the United States, according to information provided by
                   BLS and contained in international publications. Of the six industrial
                   countries that together with the United States have made up the Group of

                   5
                    The Price Statistics Review Committee suggested, in 1961, that the more volatile categories of the
                   market basket’s expenditure weights be updated at 5-year intervals. This committee was formed under
                   a contract between the Bureau of the Budget—the predecessor of the Office of Management and
                   Budget—and the National Bureau of Economic Research. Although more than 35 years have passed,
                   the committee’s work is still recognized as an important study of the CPI.
                   6
                    The Advisory Commission to Study the Consumer Price Index was chartered by the U.S. Senate. The
                   five-member advisory commission issued a report, in December 1996, to the Senate Committee on
                   Finance titled Toward a More Accurate Measure of the Cost of Living. The advisory commission was
                   chaired by Michael J. Boskin, and it was referred to as the Boskin commission.



                   Page 2                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




Seven countries (G-7),7 two updated the weights of their consumer price
indexes annually, and the other four did so approximately every 5 years.
However, BLS officials noted that some of these countries based their
updates on national data that are not comparable to data used by the
United States; for example, some countries have not collected expenditure
data directly from consumers.

The cost of updating the expenditure weights is significantly less than the
cost of a major revision. For the purposes of estimating costs in this
report, we assumed that an update to the expenditure weights would
occur in 2003, which would be 5 years after the planned revision in 1998.
BLS estimated that the cost to update the weights in 2003 would be about
$3.1 million. In comparison, BLS estimates that it will spend about
$66 million on the upcoming 1998 revision.

Because federal tax brackets and federal payments, such as those to Social
Security beneficiaries, are adjusted for inflation, a CPI that more accurately
measures inflation could affect the federal budget. BLS estimated the range
of change in the CPI, if the expenditure weights were updated on a 5-year
cycle, from 0 (zero)—no change—to a decrease of 0.2 percentage point.
We asked the Congressional Budget Office (CBO) to use the midpoint of
BLS’ range (0.1 percentage point) to estimate the effect on the federal
budget. CBO estimated that, assuming no other changes in policy or
economic assumptions, if updating the weights in 2003 (5 years after the
planned 1998 revision) reduced CPI growth by 0.1 percentage point
annually, the projected budget surplus would be increased by a cumulative
total of $10.8 billion over the 4-year period of 2004 through 2007.

BLS cited several reasons for not updating the expenditure weights
between major CPI revisions. The foremost reasons, according to BLS, were
a lack of empirical evidence to support more frequent updates and a void
of theoretical guidance on how often to do them. BLS’ other reasons were
difficulties in obtaining funds to bring about change to the CPI and concern
with what would be the best approach to improve the CPI. In the past, data
availability was also cited as a reason, but data collection improvements
have since addressed this problem.

Although theoretical guidance is not available on all facets of updating
expenditure weights, such as exactly how often updates should occur, the
preponderance of the data we reviewed supports the need for updating


7
The United States, Japan, Italy, Germany, France, Canada, and the United Kingdom have made up the
G-7 countries that have met to coordinate economic and monetary policy.



Page 3                               GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
             B-275450




             expenditure weights more frequently than about every 10 years.
             Recognizing that the data are not perfect and do not isolate the effects of
             using outdated expenditure weights, comparisons of price indexes with
             old and new weights that go back to those made for the first revision in
             1940 indicate that price indexes computed with more current weights
             were always different from indexes computed with older weights. In
             addition, these comparisons and more recent research conducted by BLS
             tend to show lower rates of inflation with indexes using newer weights.

             BLS’ concerns about updating the expenditure weights between major
             revisions were indicated in June 1997, when BLS officials said that BLS has
             the technical ability to update the expenditure weights, but it must work
             through the challenging issues that now surround the CPI program. In
             August 1997, the BLS Commissioner said, in commenting on a draft of this
             report, that she supports updating the expenditure weights more
             frequently and that BLS was in the process of developing a new updating
             policy. As part of the process of developing this policy, the Commissioner
             said BLS was studying a number of related practical questions and would
             seek the advice of its advisory councils.


             The CPI measures the change in prices of a fixed market basket of goods
Background   and services purchased directly by urban consumers. These purchases are
             for food, clothing, shelter, fuels, transportation, medical care,
             entertainment, and other goods and services that people buy for
             day-to-day living. Only expenditures made by consumers are captured in
             the CPI.

             The CPI is used by the federal government, businesses, labor organizations,
             and private citizens. According to BLS, the CPI is used as an economic
             indicator of inflation; an escalator for wages, income payments, and tax
             brackets; and a deflator of selected economic statistical series. For
             example, through collective bargaining contract negotiations in 1996,
             1.7 million workers had their wages raised on the basis of changes in the
             CPI. As a result of changes in prices as reported in the CPI in 1996,
             43.5 million Social Security beneficiaries8 and 25.8 million food stamp
             recipients had their benefits increased for inflation in 1996.

             According to BLS, to construct the CPI, the prices of more than 94,000 items
             are collected each month (e.g., margarine sold in tubs, sticks, or squeeze

             8
              Automatic adjustments of Social Security benefits, which are based on increases in the CPI, began in
             1975 (42 U.S.C. 415(i)).



             Page 4                                 GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




bottles) and aggregated into 206 “item strata” (e.g., fats and oils). In
making the monthly calculations, according to BLS, weights are used to
give proportionate emphasis for price changes of one item in relation to
other items in the CPI.

According to BLS, two sets of weights are computed from different sources
of information. BLS computes one set of weights from the Consumer
Expenditure Survey (CEX) data. These weights, which are the focus of this
report, are used to aggregate the 206 item strata into the overall index
number for the CPI. In this report, we refer to this first set of weights as
“expenditure weights.”

The second set of weights is derived primarily from information taken
from the Point-of-Purchase Survey (POPS).9 These weights, which we term
“point-of-purchase weights” in this report, are used to combine the prices
of the 94,000 items into the 206 item strata. In other words, the
point-of-purchase weights are used to aggregate the prices of the
individual items into the 206 item strata and provide the base to which the
expenditure weights are applied to calculate the CPI.

The two sets of weights are updated at different time intervals. BLS began
to publish the CPI regularly in 1921 and has changed expenditure weights
only when making major revisions to the CPI. These major CPI revisions
occurred in 1940, 1953, 1964, 1978, and 1987; another revision is scheduled
for 1998.10 BLS instituted the POPS in 1978. All of the point-of-purchase
weights are scheduled to be updated over a 5-year period, according to
BLS.


The CPI is often referred to as a cost-of-living index and is used to reflect
the cost of living to adjust, for example, federal income tax brackets and
some federal payments. Although some elements of the CPI reflect
cost-of-living concepts, the CPI was not designed to be a cost-of-living
index. As usually defined, a cost-of-living index would be broader in
coverage than an index that is based on consumer expenditures. BLS has
said through the years that the CPI is not a cost-of-living index. To date, the

9
  The CEX is used to gather data from consumers to ascertain what goods and services they are
purchasing; the POPS is used to gather data from consumers to find out where they shop for goods and
services. According to BLS, it constructs weights and selects outlet samples from sources other than
the POPS for a relatively small number of item categories.
10
  The 1987 revision was based on CEX data collected in 1982 through 1984. The planned 1998 revision
will be based on CEX data collected in 1993 through 1995. In this report, we refer to the year of the
introduction of the change of expenditure weights (e.g., 1987) rather than to the base years used to
establish the weights (e.g., 1982-84). We are using the year of introduction as the reference point when
we refer to a 5-year update (e.g., 1992 update).



Page 5                                  GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




federal government has not developed a comprehensive cost-of-living
index.11

In 1961, the Price Statistics Review Committee (hereafter called the Stigler
committee for its chairman, George Stigler) recommended that the
conceptual framework of the CPI be modified to represent a cost-of-living
index. It also supported comprehensive revision of CPI weights at least
once every decade and suggested that the more volatile categories of CPI
weights be updated at least once every 5 years. The BLS Commissioner in
1961 agreed that the CPI should be revised every 10 years. Although the
Commissioner agreed with the suggestion to update CPI weights more
often, he cited some obstacles that he thought, at the time, would preclude
BLS from doing so. We discuss these obstacles later in this report in the
section on BLS’ reasons for not updating expenditure weights more often.

In reporting to Congress in December 1996, the Boskin commission said
its overarching recommendation was that BLS establish a cost-of-living
index as its objective in measuring consumer prices. The Boskin
commission concluded that the CPI overstates inflation because of four
sources of bias: substitution bias,12 new products bias, quality change bias,
and new outlets bias.13 The commission further subdivided substitution
bias into what it termed lower-level bias and upper-level bias. The
lower-level bias concerns the aggregation of the prices of the individual
items,14 and the upper-level bias concerns the 206 item strata, which are
the subject of this report.

To address upper-level substitution bias, the Boskin commission
recommended that the fixed market basket CPI be abandoned and replaced
with two new formulas that would enable the CPI to more closely reflect


11
 For additional information about how the CPI measures the cost of living, see our report, Consumer
Price Index: Cost-of-Living Concepts and the Housing and Medical Care Components
(GAO/GGD-96-166, Aug. 26, 1996).
12
  Relative to a cost-of-living index, substitution bias gives increasing importance to items in the CPI
with higher-than-average price increases. For example, this bias occurs when consumers purchase
cheaper chicken for beef when the price of beef rises. In this illustration, the CPI overstates inflation
because it continues to track the price of beef, which consumers would have no longer purchased,
rather than the price of chicken, which consumers would have substituted for beef. A downward bias
can also occur when consumers are driven to purchase more expensive substitutes, which can occur,
for example, during wars and mandatory price controls.
13
 The Boskin commission estimated that the CPI overstated inflation by 1.1 percentage point. The
commission attributed 0.4 percentage point to substitution bias, 0.6 to new products and quality
change biases, and 0.1 to new outlets bias.
14
 In April 1997, BLS began publishing an experimental CPI that addresses the lower-level substitution
bias by using a different formula to aggregate the 94,000 individual items into the 206 item strata.



Page 6                                   GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
              B-275450




              the cost of living. One formula, according to the commission Chairman,
              would be a true superlative index; the other formula would be a modified
              superlative index.15 A superlative index, by definition, would continually
              change the market basket to reflect current consumer spending.16 BLS has
              requested funding for fiscal year 1998 to continue the fixed market basket
              CPI and to publish a CPI with a superlative index formula in 2002. Since the
              fixed market basket CPI would continue to be published, the discussion on
              how frequently to update the expenditure weights is pertinent.


              To obtain opinions on updating the CPI expenditure weights more often,
Scope and     we asked two former BLS officials and eight others who were
Methodology   knowledgeable about the CPI how often the weights should be updated.
              The eight other individuals had conducted research in connection with the
              CPI: four of the eight individuals were members of the Boskin commission,
              two were academicians, one was employed by a major economic research
              institution, and one was a member of the Stigler committee. (App. I
              describes how we selected these eight individuals.) To obtain information
              on the practices followed by other industrialized countries in updating
              their consumer price indexes, which also addresses our first objective, we
              obtained information from BLS and from publications of the Organization
              for Economic Cooperation and Development and the Canadian
              government on how often G-7 countries update their CPIs.

              To estimate the cost to BLS of updating the expenditure weights for the CPI
              on a 5-year cycle, we asked BLS to provide us with certain actual and
              estimated cost data. We asked BLS to provide us with the costs associated
              with the 1987 revision and the projected costs for the 1998 revision. In
              addition, we asked BLS for its estimate of what the costs would have been
              to update the CPI in 1992 and its estimate of what the cost might be to
              update the CPI in 2003. We did not specify to BLS what assumptions to make
              or what items to include or exclude in estimating the costs for 1992, 1998,
              and 2003. We also did not evaluate the reasonableness of BLS’ assumptions
              or estimates. BLS provided us with costs for the 1987 revision and
              estimated costs for the 1998 revision and a 2003 update of the CPI. BLS
              suggested that the cost for a 1992 update could be derived by deflating the



              15
                The Boskin commission’s report did not use the term “modified” superlative index. However, in our
              discussion with the Chairman, he agreed that one formula was not a true superlative and could be
              referred to as a modified superlative index. We use this term throughout this report.
              16
                A superlative index formula is described in appendix IV. The CPI is constructed with another
              formula, which holds the market basket of goods and services constant until a revision or update of
              the expenditure weights. That formula is described as well in appendix IV.


              Page 7                                 GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




cost of the 2003 update, which we did with the Gross Domestic Product
(GDP) price index.17

To estimate the dollar effect on the federal budget if the expenditure
weights for the CPI were updated on a 5-year cycle, we obtained assistance
from BLS and CBO, which analyzes budget-related issues and provides cost
estimates for legislative proposals to Congress. We asked BLS to provide a
range—an upper percentage point and a lower percentage point—of the
possible change that would occur to the CPI with a 5-year update.

We asked CBO to estimate the effect that a 5-year update of the CPI would
have on the federal budget, assuming no other changes in tax or spending
levels and no other changes in the economy. To do this, we asked CBO to
use BLS’ lower and upper estimates of change and the midpoint of these
estimates. To illustrate the effect of a 5-year update that would begin in
2003, we asked CBO to make projections for the years 2003 through 2007 as
it normally would and then to do a reestimation after adjusting for the
effects of a 5-year update.18

CBO’s policy is to provide projections for current and future years, but not
to provide estimates for past years. For that reason, we estimated how the
federal budget might have been affected if the expenditure weights had
been updated in 1992, which was 5 years after the 1987 major revision. In
making our estimates, we used CBO’s estimates for 1998 through 2007 to
backcast to 1992. In doing so, we assumed that the trend that was used for
the years 1998 through 2007 could be reasonably applied to the years 1993
through 1998. We discussed the methodology we used in making the
estimates with CBO officials, and they said that the methodology we used
and the results we obtained were reasonable.

In connection with impact on the federal budget, as requested by your
office, we asked the Chief Actuary of the Social Security Administration
(SSA) to estimate the effect a change in the CPI would have on the average
benefit paid to retired workers. To make the estimate, we asked SSA to use
the midpoint of BLS’ range of possible change that would occur to the CPI
with a 5-year update. Using that midpoint percentage point and economic
assumptions used in the President’s Fiscal Year 1998 Budget, SSA estimated

17
 The GDP price index, which is determined by the Bureau of Economic Analysis of the Department of
Commerce, can be used to adjust dollar amounts into inflation-adjusted dollars.
18
 CBO made projections for 1998 through 2007, which is a 10-year period. CBO does not make
projections beyond a 10-year time frame. In estimating the effects of a CPI update on the federal
budget, CBO reduced its projected rate of inflation, as measured by the CPI, by the lower and upper
estimates provided by BLS and by the midpoint of that range.



Page 8                                 GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




the change in the average monthly benefit check for retired workers,
beginning with December 2003, which would be payable in January 2004,
and continued through December 2007.

To identify and assess why updates to the CPI weights have been spaced 10
years or so apart since 1940, we talked with present and past officials of
BLS and obtained their views on the reasons for this spacing. We also
reviewed the 1961 congressional testimony of a BLS Commissioner in
which he addressed the subject of BLS’ timetable for revising the CPI. In our
assessment, we (1) collected and analyzed information on past
comparisons between indexes that applied old and new expenditure
weights, (2) obtained information on how BLS collects its source data for
the CPI, (3) reviewed BLS budget information to ascertain BLS’ plans for
future changes in its indexes, and (4) compared the estimated costs and
benefits of a 5-year update to place an update in practical perspective.

In our assessment, most of the comparisons between indexes with old
weights and indexes with new weights probably reflected differences that
were not due to changes in the expenditure weights alone. Some of the
indexes we used were produced by BLS for “overlap” periods. When major
revisions to the CPI were made, BLS calculated two indexes for several
months. One index used the weights that had been in effect before the
revision, and the second index used the new weights that were created for
the revision. However, during a revision, many factors can and do change.
For example, the geographic locations where data are collected are
changed to some extent as are the items of goods and services in the
market basket. Therefore, the differences that may result from comparing
the two indexes may be due to several factors, and the effects of changes
to the expenditure weights cannot be isolated from the effects of other
changes to the index data.

With this knowledge, we treat the differences as indicators of the effects
of an update of the expenditure weights because an update of the weights
is unlikely to occur in isolation from the other factors that are associated
with revisions. For example, a 1992 update would have most likely
incorporated a market basket that was based on different geographic areas
than the areas that were used in the 1987 revision because, in 1986,
changes were made in the geographic locations where expenditure data
were collected. Such geographic changes are associated with major
revisions.




Page 9                        GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




In addition to overlap studies, we examined the effect of the age of
weights with indexes that were calculated with alternative base-year
periods. For example, comparisons were made of the official CPI’s 3-year
base of 1982 through 1984 with alternative 3-year base periods (i.e., 1987
through 1989). In these and other comparisons, we applied an economic
concept that is based upon economic literature that suggests that an index
is more accurate if the expenditure weights used to compute it represent,
as much as practical, current consumer spending.19 However, in our
review of economic literature, we did not identify any theoretical guidance
on how often (e.g., 5 years as compared with 10 years) expenditure
weights should be updated. (See app. I for more information about our
objectives, scope, and methodology.)

As previously reported in this section, this report includes, and often relies
on, estimates and comparisons prepared by BLS, CBO, or SSA.20 We did not
verify the computerized data that the agencies used in producing these
estimates and comparisons. Verification, in our opinion, would have been
impractical because it would have been costly and time consuming. In
addition, the estimates and comparisons were within the scope of
activities that BLS, CBO, and SSA normally perform. Therefore, we used their
estimates and comparisons.

The results we obtained are intended to contribute to the discussion of
how often the CPI should be updated, but they are not intended to
represent all future effects of shortening the updating cycle. Neither is our
work intended to evaluate a change in the basic formula that could
address substitution bias in the CPI. The point of shortening the updating
cycle would be to have the CPI reflect, as closely as practical, the current
spending patterns of consumers, regardless of whether the index is pushed
upward or downward.

We did our work in Washington, D.C., between November 1996 and
July 1997 in accordance with generally accepted government auditing

19
  For example, see Jack E. Triplett, “Economic Theory and BEA’s Alternative Quantity and Price
Indexes,” Survey of Current Business, Vol. 72 (Apr. 1992), pp. 49-52; Ralph Turvey, Consumer Price
Indexes: An ILO Manual (Geneva: International Labor Office, 1989), p. 38; and Price Statistics Review
Committee, The Price Statistics of the Federal Government: Review, Appraisal, and
Recommendations, A Report to the Office of Statistical Standards, Bureau of the Budget (New York:
National Bureau of Economic Research, 1961), p. 31.
20
  BLS estimated (1) the cost to update the expenditure weights and (2) the range of percentage effect
on the CPI if the expenditure weights were updated at 5-year intervals. BLS also provided us with its
past comparisons of old-weighted and new-weighted indexes. CBO estimated the dollar effect on the
federal budget if CPI growth were lowered 0.1 percentage point or 0.2 percentage point annually. SSA
estimated the impact on Social Security payments if the CPI were reduced annually by 0.1 percentage
point.



Page 10                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                           B-275450




                           standards. We requested comments on a draft of this report from the
                           Secretary of Labor, the Chair of the Council of Economic Advisers (CEA),
                           the Director of the Office of Management and Budget (OMB), and the Chair
                           of the Board of Governors of the Federal Reserve System or their
                           designees. Comments by BLS, CEA, OMB, and the Federal Reserve are
                           discussed near the end of this letter and are reproduced in appendixes V
                           through VIII.


                           We spoke with 10 individuals who were former officials of BLS or who had
More Frequent              otherwise studied the CPI, and they were unanimous in stating that 10
Updating Is Deemed         years between updates of the expenditure weights was too long. However,
Desirable by               there was less agreement among the individuals on exactly how often the
                           updating should occur. According to information obtained from BLS and
Individuals                international publications, seven major industrial countries have
Knowledgeable of the       consumer price indexes but, among them, only the United States updates
                           its CPI as infrequently as once a decade.
CPI That We
Contacted
Professional Opinions on   Two former BLS officials told us that updating the weights about every 5
Updating                   years was about right. One official told us that the POPS should be rotated
                           more frequently, which would affect point-of-purchase weights. He also
                           advocated a different method of aggregating CEX data to develop the
                           expenditure weights for the 206 item strata. The other former official, a
                           previous BLS Commissioner, noted that doing an update more frequently
                           than every 5 years would be too often.

                           We also spoke with a former member of the Stigler committee and four
                           members of the Boskin commission. The former Stigler committee
                           member said that updating the CPI only every 10 years was entirely too
                           infrequent. However, he gave low priority to updating the expenditure
                           weights more often because he believed that getting new consumer items
                           into the CPI and accounting for product improvement were more
                           important. The four members of the Boskin commission also said that the
                           expenditure weights for the market basket should be updated more
                           frequently than every 10 years. However, the members regarded more
                           frequent updating as only one step to improving the CPI. The Boskin
                           commission recommended abandoning the fixed market basket aspect of
                           the CPI and adopting a true superlative index formula and a modified
                           superlative index formula to account for changing market baskets.




                           Page 11                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




The spirit of the Boskin commission’s recommendations, according to its
Chairman, was for the CPI to be more current in order to reflect what is
occurring in the economy. He said that if there were no change in existing
products or no new products in the economy, then updating the
expenditure weights would be the only step that would need to be taken.
However, the economy is changing, with new products and product
improvements occurring constantly. Therefore, more frequent updating
was only a step toward what should be done to improve the CPI. He said
BLS should be in a permanent revision mode. Because different aspects of
the CPI interact with each other, a change in the expenditure weights
would complement other steps that could be taken, such as changing the
POPS sample more often than every 5 years, increasing the size of the CEX,
and using estimation methods to adjust for changes in the quality of items
in the CPI.

Two of the three remaining members of the Boskin commission with
whom we spoke told us that the expenditure weights should be updated
more frequently than every 5 years. The other member, citing concern
about resource constraints faced by BLS, gave preference to providing
financial support to implement the Boskin commission’s
recommendations to improve the CPI, rather than funding a more frequent
update of the market basket.

The remaining three individuals we spoke with also supported a more
frequent update than about every 10 years as a more accurate way to track
inflation. One of them said that doing so would not necessarily lead to
lower measures of inflation.

In addition, although we did not interview the Chairman of the Board of
Governors of the Federal Reserve System, we noted that he stated in a
speech, in March 1997, that there was a bias problem in the CPI given the
failure to change the expenditure weights more often than about every 10
years. However, a representative of the Federal Reserve, in commenting
on a draft of this report, said that although the Federal Reserve Chairman
has said that out-of-date weights are a source of bias in the CPI, the
Chairman does not endorse merely updating them more frequently.21 He
said that the Chairman has testified before Congress in support of changes
recommended by the Boskin commission, and that the payoff of departing
from a fixed-weight structure of the CPI is likely to be much more



21
 These comments were made by the Assistant Director and Chief, Economic Activity Section, Division
of Research and Statistics, the Federal Reserve System. His comments are reprinted in appendix VIII.



Page 12                               GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                       B-275450




                       important in improving the accuracy of the CPI than more frequent
                       updating of expenditure weights alone.


Practices of Other     As previously mentioned, the United States is one of the seven leading
Industrial Countries   industrial countries—the G-7—that have met to coordinate economic and
                       monetary policy. In addition to the United States, the other six G-7
                       countries also track consumer prices through a market basket of goods
                       and services and weight the prices of the items in the market basket.
                       According to a BLS official and published information, Japan and Italy
                       update expenditure weights every 5 years; Germany updates, on average,
                       about every 5 years; Canada updates every 4 years—except for the last
                       update (6 years), when it reengineered its index; and France and the
                       United Kingdom update every year.

                       However, BLS officials noted that some of these countries base their
                       updates on national data that are not comparable to the U.S. continuing
                       CEX. Some of the countries do not use expenditure data collected directly
                       from consumers; others use consumer expenditure data that require
                       respondents to recall 12 months of expenditure data. BLS officials said the
                       system used in the U.S. CPI for maintaining current and representative
                       samples of items to price is more advanced than in most other countries.

                       We are not endorsing the practices in any other country over BLS’
                       practices. We provide this information for comparative purposes to show
                       the priority other countries place on keeping their market baskets current.


                       The estimated cost of updating expenditure weights is relatively small in
Cost to Update the     comparison to the cost of major revisions. For the purposes of estimating
Expenditure Weights    costs, we assumed in this report that updates of expenditure weights
Is Estimated to Be     would occur in 1992 and 2003, which is 5 years after major revisions. On
                       the basis of data supplied by BLS, the estimated cost to have updated the
Relatively Small       weights in 1992 would have been $2.4 million spread over 3 years.
                       According to BLS, the estimated cost to update the expenditure weights in
                       2003 would be $3.1 million over a 3-year budget period. BLS reported that
                       the 1987 major revision cost $47 million over 5 years. According to BLS, the
                       cost for the planned 1998 revision is expected to be about $66 million over
                       6 years.

                       BLS noted that the estimated cost of an update excluded many activities
                       that were included in the costs for revisions to the CPI. These excluded



                       Page 13                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                       B-275450




                       activities include using recent decennial census data to reselect
                       geographic areas and housing samples used in the CPI’s surveys;
                       evaluating, replacing, and updating CPI data processing systems; and
                       recategorizing the items in the CPI market basket. The activities that BLS
                       included in its estimated cost to update the expenditure weights included
                       changing the weights of the items in the market basket, redefining the item
                       strata in a limited way, and including any new items as a result of the
                       limited redefinition of the item strata. For example, the costs of an update
                       might include those associated with adding an item stratum for cellular
                       telephone services.22


                       Because the CPI is used to index federal income tax brackets and certain
More Frequent          federal spending, changes in the CPI can affect the federal budget. BLS, at
Updates Could Affect   our request, estimated the impact on the CPI if the expenditure weights
the Federal Budget     were updated on a 5-year cycle, and, with the help of CBO, we used those
                       estimates and their midpoint to illustrate the effect that updating might
                       have on the federal budget. In making these estimates, both we and CBO
                       assumed that there were no other changes in tax or spending levels and no
                       other changes in the economy during the periods under review.

                       BLS said the historical evidence suggests that shifting to a 5-year update of
                       the market basket weights could reduce the annual rate of growth of the
                       CPI by between 0 (zero) and 0.2 percentage point. Although CBO does not
                       backcast, using a method discussed with it, we estimated that an update in
                       1992 would have reduced the federal budget deficit between $0 and
                       $32.4 billion over the 6-year period until the implementation of the
                       upcoming 1998 revision. According to CBO, an update in 2003 could
                       increase the projected budget surplus between $0 and $20.2 billion over a
                       4-year period.

                       Using estimates provided by CBO for an annual 0.1 percentage point
                       reduction in CPI growth and assuming no other changes in policy or the
                       economy, we estimated that the federal deficit would have been reduced
                       by a cumulative total of $16.2 billion over the 6 years following an update
                       in 1992. According to CBO, an update in 2003 in which CPI growth would be
                       reduced annually by 0.1 percentage point and assuming that nothing else
                       changed, the projected federal budget surplus would be increased by a




                       22
                         BLS officials told us that they plan to add this stratum in the upcoming 1998 revision.



                       Page 14                                  GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




cumulative total of $10.8 billion over the 4 years following the update.23 As
shown in figure 1, most of the impact of such a reduction in the CPI would
be on federal outlays—such as reduced payments to Social Security
beneficiaries, which account for most of the outlays—and most of the
impact would occur in the later years. For example, according to estimates
by SSA’s actuaries, the average monthly benefit check for retired workers
in 2004 would be reduced by $0.91, from $939.94 to $939.03, with an annual
0.1 percentage point reduction in CPI growth; by the fourth year (2007), the
average monthly check would be reduced by $3.83, from $1,032.56 to
$1,028.73.24




23
  To compare the midpoint estimates, we converted them to 1997 constant dollars. The 1997
constant-dollar estimate of the effect of a 1992 update on the federal budget would have been a
cumulative decrease in the federal deficit of $16.4 billion over 6 years. For a 2003 update, the effect of
a 0.1 percentage point reduction in the CPI was estimated by CBO to be an increase in the projected
surplus of $8.6 billion in 1997 constant dollars over the following 4 years. A constant-dollar value is
measured in terms of prices of a base period to remove the influence of inflation. The resulting
constant-dollar value is the value that would exist if prices had remained the same as those in the base
period (e.g., 1997).
24
  According to SSA, by the fifth year (2008), the average monthly benefit check for retirees would be
reduced by $4.86, from $1,065.98 to $1,061.12.



Page 15                                  GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                                          B-275450




Figure 1: Estimated Increase in Annual
and Cumulative Federal Budget             12    Billions of dollars
Surplus If CPI Growth Were to Be
Reduced Annually by 0.1 Percentage
Point Over a 4-Year Period After a 2003   10
Update
                                           8



                                           6



                                           4



                                           2



                                           0

                                                  2004       2005      2006       2007                 2004
                                                                                                       through
                                                                                                       2007
                                                  Fiscal year


                                                            Cumulative increase in federal budget surplus

                                                            Interest on debt

                                                            Outlays

                                                            Revenues



                                          Note: The estimates assume no other changes in tax or spending levels and no other changes in
                                          the economy.

                                          Source: CBO.



                                          BLS has taken actions to respond to the 1961 Stigler committee study, and
Evidence Suggests                         the current BLS Commissioner told us one response in particular enabled
More Frequent                             BLS to markedly improve the representativeness of consumer items and

Updates Would Be                          prices in the CPI. However, BLS had not acted on the Stigler committee’s
                                          suggestion to update the more volatile categories of weights at least once
Beneficial                                every 5 years, and BLS officials cited several reasons for not doing so. The
                                          most important reason, they said, was a lack of empirical evidence to
                                          support more frequent updates and a void of theoretical guidance on how
                                          often to do them. The officials said that another reason was previous
                                          difficulties in obtaining funds for major revisions of and improvements to
                                          the CPI. They also said that, in the past, certain data necessary to update




                                          Page 16                                    GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                         B-275450




                         expenditure weights were unavailable between major revisions, but that
                         situation has changed. In addition, BLS cited its concern with what would
                         be the best approach to improve the CPI to make it more reflective of
                         current consumer spending.

                         We examined the information surrounding these reasons, and, aside from
                         the issue of funding, which is unpredictable at this point, we concluded
                         that the evidence suggests that more frequent updates would be beneficial.
                         As of August 1997, BLS was studying how often to update the expenditure
                         weights.


BLS Has Acted on Many    The current BLS Commissioner said BLS has implemented many of the
Aspects of the Stigler   Stigler committee’s recommendations. In her view, the most important
Committee Study          Stigler committee recommendation concerned the selection for price
                         tracking of individual items of goods and services. She said that in
                         response to the recommendation, BLS developed and began using the POPS
                         in 1978 to identify sales outlets, which has allowed BLS to incorporate new
                         items into the CPI that otherwise would not have been incorporated until a
                         major revision.25 She said that under the methodology used with the POPS,
                         20 percent of the outlets and items tracked are newly selected each year,
                         which changes the entire sample within 5 years. Thus, according to the
                         Commissioner, a large part of the reason for wanting to update weights
                         more frequently—maintaining the representativeness of the item and
                         outlet samples—is already accomplished on a 5-year rotation.

                         Evaluating BLS’ response to every Stigler committee recommendation was
                         not the purpose of this review. But, regarding the Commissioner’s
                         statement of making the CPI more current through the use of the new POPS
                         methodology, we believe that changing the procedures used to select retail
                         outlets and items does make the CPI somewhat more representative of
                         what consumers are purchasing. However, implementation of the POPS still
                         does not address the expenditure weights for the 206 item strata that
                         remain fixed until the entire market basket is revised. A 5-year rotation in
                         the POPS does improve the CPI in terms of keeping item samples current
                         and introducing new goods, which also updates the point-of-purchase




                         25
                           Before 1978, BLS field representatives used the same item specifications throughout the country in
                         an attempt to price an item meeting the detailed description. Since the introduction of new sampling
                         techniques in 1978, the field representatives use a store’s sales information to select a unique item
                         within the specified categories for pricing. However, once an item is selected, the field representative
                         is to continue to price it in that store. For more detailed information, see appendix II.



                         Page 17                                 GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                                 B-275450




                                 weights every 5 years.26 Even though BLS has applied a current-is-better
                                 approach to point-of-purchase weights, it has not applied that same
                                 approach to the expenditure weights. Consequently, the items in the
                                 market basket are still aggregated into the CPI with expenditure weights
                                 that reflect outdated consumer purchases.


BLS’ Reasons for Not             We spoke with the current BLS Commissioner and other current BLS
Updating Expenditure             officials about the timing of major revisions and about the obstacles to
Weights More Often               updating the weights. We discussed the timing of major revisions because
                                 expenditure weights only have been updated during these revisions. The
                                 Commissioner said she intuitively agreed that a 10-year period is long, but
                                 she was not sure what time frame was best. She provided several reasons
                                 for the length of time between major revisions to the CPI and reasons why
                                 BLS was uncertain about undertaking a weight update independent of a
                                 major revision. As previously mentioned, we also obtained the comments
                                 of two former BLS commissioners. The reasons given by these three BLS
                                 commissioners are presented in the following subsections along with, as
                                 appropriate, our related evaluation.

Lack of Empirical Evidence and   More frequent updating of the CPI weights has not been at the top of BLS’
Theoretical Guidance             priority list, according to the current Commissioner, who also said that
                                 only recently has there been any systematic evidence that inflation, as
                                 measured by the CPI, is affected by the age of expenditure weights.27 She
                                 also said that even this evidence is limited and weak. In addition,
                                 according to the Commissioner, there is neither a theoretically “best”
                                 frequency for updating the weights, nor any theoretical reason why more
                                 recent weights are “better.” Therefore, she said, the decision on how often
                                 to update must be made on commonsense and cost-benefit terms.

                                 BLS’view that there is insufficient evidence to support more frequent
                                 updates is long-standing. In 1961, the then BLS Commissioner testified
                                 before a congressional committee that there was no evidence to support
                                 more frequent weight updates, and he cited a need for additional
                                 research.28


                                 26
                                  Although the methodology does provide improvement, five of the individuals with whom we spoke
                                 said that even the POPS could be made more current with more frequent rotations than every 5 years.
                                 27
                                  John S. Greenlees, “Expenditure Weight Updates and Measured Inflation” Bureau of Labor Statistics,
                                 February 27, 1997.
                                 28
                                  The Commissioner made these comments to the Subcommittee on Economic Statistics of the Joint
                                 Economic Committee of Congress, which was holding hearings on the recommendations of the Stigler
                                 committee.



                                 Page 18                               GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




The preponderance of the data we reviewed does not support BLS’
statement, as stated currently or in 1961, that there is insufficient
empirical evidence to support the need for more frequent updating of
expenditure weights. In his 1961 testimony, the then Commissioner cited
three studies on which he based his conclusion, and each study covered a
different group of years between 1925 and 1939. For the first two periods,
the old weights produced less of a decline in prices than the new weights.
For the last period, the old weights produced more of an increase in prices
than the new weights. The differences between the indexes produced by
the old and new weights were 0.1 percentage point, 1.2 percentage point,
and 0.1 percentage point, respectively.

In 1953, BLS revised the CPI and updated the expenditure weights. It applied
the new weights to January through June 1953 consumer price data. In
response to a presidential request, BLS also applied the weights used in the
1940 revision to the January through June 1953 price data. The index with
the old weights showed an annual understatement of inflation of
0.5 percentage point in comparison with the index using the new weights.
In his 1961 testimony, the BLS Commissioner explained that the 1953
comparisons were different from those previously described for 1925
through 1939. The differences that were found in the studies for the earlier
years basically reflected changes in expenditure weights. He noted that the
comparisons for the 1953 revision reflected factors, such as changes in
cities in the CPI, in addition to expenditure weight changes.

As shown and analyzed in appendix III, additional empirical evidence has
become available since 1961 that also indicates that the measurement of
inflation is affected by the age of the expenditure weights. For example,
BLS estimates that when the upcoming 1998 revision is introduced, CPI
growth will be lowered by 0.1 or 0.2 percentage point. In addition, we
reviewed historical data from overlap studies in which BLS continued
calculating the CPI with both the old and new weights for 6 months
following a major revision. Although it is impossible to identify exactly to
what extent other factors contributed to the differences, indexes produced
by the old weights overstated inflation in comparison to those indexes
produced by the new weights in the 1964 and 1987 overlap comparisons.
The reverse was true in 1978, but that difference may have been due to a
fundamental change in BLS procedures associated with the implementation
of the POPS.

Additional BLS studies of indexes that examine the effects of the age of
expenditure weights include comparisons of indexes that were calculated



Page 19                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




with alternative 3-year base periods in which BLS compared the official
CPI’s 3-year base period of 1982 through 1984 with alternative 3-year
periods since then. For example, the analysis for an update with a 1987
through 1989 base29 averaged 0.11 percentage point lower over a 5-year
period than an index with the 1982 through 1984 base years. As a result of
our examination of these and other BLS studies, we concluded that the best
available evidence indicates that indexes with newer weights reduce the
growth of the CPI by about 0.1 percentage point per year.

Although theoretical guidance is not available on all facets of updating
expenditure weights, such as exactly when to update, economic literature
suggests that an index is more accurate if the expenditure weights used to
compute it represent, as much as practical, current consumer spending. In
addition, the Stigler committee in 1961 provided a commonsense principle
on when to revise the weights: a revision is necessary when the weight
base has changed appreciably. On the basis of the statements of
individuals with whom we talked (see discussion of these views on pp. 11
to 13) and the CPI weight comparisons we reviewed (previously presented
in this subsection and in greater detail in app. III), there is sufficient
reason to believe that the weight base had changed appreciably before
major revisions to the CPI.

The BLS Commissioner pointed to using common sense and a cost-benefit
analysis to provide guidance on how often to update the weights. At the
time of the Stigler committee’s report in 1961 and until the early 1970s, the
CPI was used in a very limited way in the federal sector to index federal
programs for the effects of inflation; the first large income program to be
adjusted with the CPI was civil service retirement payments in 1962.
Therefore, changes in the CPI had relatively little effect on federal
expenditures and had no direct impact on federal receipts. In the 1990s,
however, the federal government uses the CPI to index a much broader set
of programs, including federal income tax brackets and certain federal
payments, that directly affect a larger share of the population and a much
larger volume of federal receipts and expenditures. These uses, in our
view, provide a commonsense basis for making the index a more accurate
reflection of what consumers are buying in a rapidly changing economy.

Updating the expenditure weights more often appears to be supported
from a cost-benefit standpoint as well. BLS estimated that it would cost
$3.1 million over 3 years to update the expenditure weights in 2003. If the


29
 According to BLS, these are the base years that would have been used if the CPI had been revised in
1992.



Page 20                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                       B-275450




                       growth in the CPI decreased by 0.1 percentage point per year, CBO estimates
                       show that this would lead to a cumulative total of a $10.8 billion increase
                       in the projected budget surplus over the 4 years after the update.30
                       However, regardless of the effect on the budget, the point of doing an
                       update outside of a major revision would be the increased value of having
                       the CPI reflect, as closely as practical, the current spending patterns of
                       consumers.

Funding Difficulties   Funding for past major revisions has not always been easy to obtain. The
                       1987 revision was delayed 1 year because of funding limits. Similarly, the
                       scheduled 1998 revision’s start was delayed 1 year, until 1995, because of
                       funding limits. A former BLS Commissioner with whom we spoke also
                       identified funding as a problem. She said that funding for past major
                       revisions was held up either within the Department of Labor, by OMB, or in
                       the appropriations process.

                       Even before the tenure of this former Commissioner, obtaining funds to
                       revise the CPI was apparently a problem. In 1961, the then BLS
                       Commissioner, in testifying before a congressional committee, explained
                       that under the then present practice, revisions to the CPI were undertaken
                       only when BLS was successful in convincing the Bureau of the
                       Budget—now OMB—and Congress that there was an urgent need to bring
                       the CPI up to date.

                       BLS also has had difficulty in obtaining funds, apart from a major revision
                       for improvements to the CPI in the early 1990s. BLS officials told us that
                       they had requested $450,000 for the CPI to improve quality adjustment
                       procedures in consumer electronics, shelter, and apparel, but it did not
                       receive these funds from Congress.31

Data Availability      BLS officials said CEX data and decennial census data are essential to
                       establishing the CPI expenditure weights. CEX data are obtained from
                       consumers and identify the items of goods and services that consumers
                       have been purchasing. Among their uses in the CPI, decennial census data
                       are used in selecting the geographic locations from which samples of
                       consumers are surveyed for CEX purposes.



                       30
                         Under current budget rules, these could not offset each other. However, the estimates do give an
                       indication of the overall impact on spending and revenues.
                       31
                        This information was obtained in connection with a report we issued in 1995. See Economic
                       Statistics: Status Report on the Initiative to Improve Economic Statistics (GAO/GGD-95-98, July 7,
                       1995).



                       Page 21                                 GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                            B-275450




                            The BLS Commissioner said, in 1961, that more frequent updates required a
                            continuing CEX, which did not exist at that time.32 The current
                            Commissioner said that, until the institution of the continuing CEX that
                            allowed BLS to decrease the number of units surveyed and to make the CEX
                            an ongoing program, an update of the weights every 5 years would have
                            been costly because it would have required a special CEX. However, a
                            continuing CEX program has been established, and, from a practical
                            perspective, the first possible 5-year update of the weights using CEX data
                            would have been in 1992, which is 5 years after the first CPI revision
                            (1987) that used the continuing CEX data.

                            The need to have the CPI reflect geographic movement of the population as
                            measured by the decennial census was cited by BLS officials as a reason to
                            make a major revision every 10 years. However, BLS officials said that they
                            do not have to wait for new decennial census data before updating
                            expenditure weights.


BLS Is Deciding How Often   Updating the market basket expenditure weights is not the only way in
to Update the Expenditure   which BLS could make the CPI more representative of current consumer
Weights                     spending. The Boskin commission recommended another way, which was
                            through the concept of superlative index formulas. Although BLS plans to
                            publish a superlative index, BLS does not see it as a replacement for the
                            fixed market basket CPI. As long as BLS publishes the fixed market basket
                            CPI, updating the weights more often than once a decade would remain
                            important. BLS has the technical ability now to update the expenditure
                            weights and, in commenting on a draft of this report, the BLS
                            Commissioner said BLS was developing a new updating policy. To develop
                            this policy, according to the Commissioner, BLS was studying what
                            frequency will yield the most accurate CPI and best support the CPI’s many
                            uses.

Planned Publication of      According to the BLS Commissioner, the use of superlative indexes, such as
Superlative Indexes         those BLS is producing on an experimental basis, is the appropriate way to
                            address what the Stigler committee sought in its recommendation for
                            more frequent updating of weights. That is, a superlative index reflects
                            changes in consumer spending in response to changes in relative prices,
                            and, under certain assumptions, a superlative index is free of upper-level

                            32
                              Data for the CEX were collected only when funding was provided for a CPI revision, and the
                            Commissioner was correct in saying that a continuing CEX was not available in 1961. Since 1979, CEX
                            data have been collected continuously, whereby respondents in the interview portion of the CEX are
                            interviewed every 3 months over 5 calendar quarters. In each quarter, one-fifth of the respondents are
                            deleted and replaced by new respondents.



                            Page 22                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




substitution bias. In commenting on a draft of this report, the BLS
Commissioner said that since superlative indexes can be published only
with a representational lag because of the lack of current-period
expenditure as well as price data, their use in the CPI is precluded. The CPI,
she said, is produced monthly and revised only in unusual circumstances.

Although BLS plans to begin publication of a superlative formula index in
2002, BLS officials stated that they also plan to continue to publish the
fixed market basket CPIs.33 Therefore, those who use the CPI for escalation
purposes would have to choose among the published CPIs, including the
fixed market basket CPI.

In addition, the federal government’s use of the CPI is legislatively tied to
the fixed market basket concept in some instances. For example, the U.S.
tax code specifically identifies the use of the CPI-U for automatic inflation
adjustments of federal income tax brackets and deductions for personal
exemptions.34 Therefore, unless otherwise changed by legislation or unless
BLS named its superlative index the CPI-U, the fixed market basket CPI
would be used in these programs.

The Boskin commission recommended that BLS replace the fixed market
basket CPI with two new index formulas as follows: (1) an index that
would be updated annually and revised historically to incorporate
measurement improvements and (2) a monthly index that would be based
on a “trailing” 2- or 3-year average of CEX data. According to the Chairman
of the Boskin commission, the annual index would use a true superlative
index formula, whereas the monthly index would use a modified
superlative index formula. He said the spirit behind the commission’s
recommendation was for BLS to move as close as practical to creating a CPI
that would be reflective of the cost of living.

Although BLS plans to publish a superlative index in 2002, it has not
decided how this index will be constructed. The Commissioner said that
true superlative indexes cannot be produced in “real time,” or monthly,
because they require current expenditure data, which are impossible to
collect and process on a monthly basis. In July 1997, another BLS official
told us that the superlative index that BLS plans to publish in 2002 may be

33
  Publishing variations of a consumer price index would not be unusual. Originally, when BLS
announced that it would begin publishing the CPI-U, which represents all urban consumers, it planned
to discontinue the older CPI-W, which represents urban wage and clerical workers. Since users, such as
labor unions, supported publication of the CPI-W, BLS changed its plans to drop the CPI-W and now
publishes both indexes.
34
  26 U.S.C. 1(f).



Page 23                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                               B-275450




                               (1) an annual number with a 2-year lag that, for example, would reflect
                               inflation for the year 2000 or (2) a current measure that would be subject
                               to revision as more current expenditure data become available. According
                               to the BLS official, this second measure would not be considered a true
                               superlative index until the more current expenditure information was
                               incorporated.

                               Although BLS has not decided on the construction of a superlative index,
                               one approach or index formula it is considering is referred to as a Fisher
                               Ideal superlative index. As explained in appendix IV, two different index
                               values are combined to produce a Fisher Ideal superlative index. One of
                               the two index values is based on the Laspeyres index formula, which is the
                               formula used to produce the official CPI. In other words, values from the
                               CPI’s fixed market basket would be inputs to the Fisher Ideal index. The
                               other index value used in this superlative computation is based on a
                               Paasche formula, which, unlike the Laspeyres formula, incorporates
                               current expenditure weights. In general terms, the Fisher Ideal superlative
                               index reaches the middle ground between the Laspeyres and Paasche
                               formulas.

BLS Has Technical Ability to   According to BLS officials, BLS has the technical ability to update the
Update and Plans to Consult    expenditure weights more frequently. However, as of June 1997, BLS was
With Advisory Groups           undecided as to whether it would update the weights outside of major
                               revisions to the CPI. The Deputy BLS Commissioner said BLS was still
                               considering the matter and, as a first step, needed to make a decision
                               within BLS about updating the expenditure weights at times other than
                               major revisions. In August 1997, in commenting on a draft of this report,
                               the BLS Commissioner said BLS was developing a new updating policy. To
                               develop this policy and before making a final determination, BLS plans to
                               study a number of practical questions and the Commissioner said that BLS
                               will seek the advice of its advisory councils, all with the intent of
                               determining the best frequency for updating the CPI weights.


                               As the principal source of information on consumer prices and inflation in
Conclusions                    the United States, the CPI should reflect current consumer expenditures as
                               much as practical. That clearly was the view of the Stigler committee in
                               1961 and of the Boskin commission in 1996.

                               One step that BLS could take to advance that concept is to update the
                               expenditure weights of the CPI more often than only during major revisions
                               to the CPI. The current practice of updating weights only as part of a



                               Page 24                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




revision means that it is 10 years or more between updates of the
expenditure weights; this appears too long to achieve a reasonable
representation of current consumer spending. The BLS Commissioner said
that she intuitively believed that 10 years between updates was too long.
The two former BLS managers and eight CPI researchers with whom we
spoke all believed that, conceptually, the weights should be updated more
often than every 10 years. According to a BLS official and published
information, other G-7 countries update expenditure weights more often
than every 10 years.

However, BLS has held for some time that, although 10 years between
updates may seem inappropriate, there is no strong empirical evidence
that suggests a connection between the age of the weights and the
measurement of inflation. Our examination of BLS data, however, showed
that the age of expenditure weights affects the measure of inflation.
Although the data are not perfect and do not isolate the effects of using
outdated expenditure weights, comparisons of price indexes employing
BLS data with old and new weights indicate that price indexes computed
with more current weights were always different than indexes computed
with older weights. This result has been the case going back to
comparisons made for the first revision in 1940. In addition, comparisons
generally tend to show lower rates of inflation with indexes using newer
weights.

There are also reasons for making certain that the expenditure weights
are, as much as practical, reflective of current consumer spending. Since
1962, the CPI has been legislatively connected to adjusting some benefit
payments for inflation and more recently to adjusting federal tax brackets.
As a result, any overstatement or understatement of inflation by the CPI
can have a major impact on the federal budget. For example, if, beginning
in 2003, CPI growth were annually reduced by 0.1 percentage point and all
policies and the economy remained unchanged, CBO estimates that the
federal budget surplus over the 4 years following 2003 would be
cumulatively $10.8 billion higher.

We recognize that gaining financial support for revising or improving the
CPI has been a problem at times. We cannot predict the ease or difficulty
BLS might have in getting funds to update the weights more often (e.g.,
$3.1 million over 3 years that BLS estimated it would need for an update in
2003).




Page 25                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                     B-275450




                     We also recognize that adjusting the weights more often does not have the
                     highest priority among all commentators on the CPI. For example, the
                     Boskin commission would rather see BLS replace the fixed market basket
                     CPI with various types of superlative indexes. However, even if BLS
                     published a superlative index, which it plans to do, the updating of the
                     weights more often would remain significant because BLS does not view
                     superlative indexes as replacements for the fixed market basket CPI. It
                     plans to continue to publish the long-standing fixed market basket CPI. In
                     addition, since BLS is still trying to address basic conceptual issues in
                     designing the superlative-type index that it plans to publish in 2002, the
                     uncertainty surrounding this planned index suggests to us that making the
                     fixed market basket index as current and accurate as possible should be
                     done.


                     We recommend that, as long as a fixed market basket CPI is published, the
Recommendation       Commissioner of BLS should update the expenditure weights of the CPI’s
                     market basket of goods and services more frequently than every 10 years
                     to make it more timely in its representation of consumer expenditures.


                     We sent a draft of this report to the Secretary of Labor, the Chair of CEA,
Agency Comments      the Director of OMB, and the Chairman of the Board of Governors of the
and Our Evaluation   Federal Reserve System and requested comments from them or their
                     designees. The Commissioner of BLS provided comments for the
                     Department of Labor, and said she supports more frequent updates of the
                     expenditure weights. However, the Commissioner said neither economic
                     theory nor empirical evidence demonstrates the superiority of any
                     particular update interval. She said that BLS needs to consider carefully
                     what frequency will yield the most accurate CPI and best support the many
                     uses of the index. There are, she said, a number of practical questions
                     related to developing a new updating policy that BLS must address. BLS is
                     currently studying these questions but, she said, the ultimate decision rests
                     largely on commonsense judgment. Finally, the Commissioner emphasized
                     that BLS will not evaluate potential changes to calculating the CPI on
                     whether they raise or lower the measured rate of price change. Rather, BLS
                     will evaluate potential changes on whether they produce a more accurate
                     index. We agree with the Commissioner’s statement that potential
                     changes to the CPI should be predicated on whether they produce a more
                     accurate index. The Commissioner was silent as to whether the new policy
                     would direct an expenditure weight update between major revisions,
                     which have been every 10 years or so. Although we cannot say exactly



                     Page 26                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




how often the expenditure weights should be updated, the evidence we
reviewed suggested that updating only once every 10 years or so was
insufficient. The Commissioner’s August 8, 1997, letter is reprinted at
appendix V. BLS provided technical comments on the draft report by
separate communication, and we incorporated them as appropriate.

By a letter dated August 8, 1997 (see app. VI), CEA’s Director of
Macroeconomic Forecasting said more frequent updating would be a small
improvement and ought to be considered. However, the Director hoped
that readers of this report do not confuse more frequent updating with the
adoption of a true cost-of-living index.

In an August 12, 1997, letter (see app. VII), OMB’s Associate Director for
Economic Policy said that frequent updating of expenditure weights is one
important option. However, OMB believed that BLS should consider more
frequent updating in context with other potential improvements. The
Associate Director pointed out that substitution bias would remain in a
Laspeyres-type index even with more frequent updating of expenditure
weights. The CPI is a Laspeyres-type index.

The Federal Reserve’s designee, the Assistant Director and Chief of the
Economic Activity Section, Division of Research and Statistics, said in a
July 31, 1997, letter (see app. VIII) that the draft report addressed a very
important public policy issue. He said that more frequent updating of the
expenditure weights would be desirable absent other actions to improve
the CPI’s accuracy. However, other changes to the CPI, such as those
recommended by the Boskin commission, may do more to improve the
CPI’s accuracy. He said our recommendation for more frequent updating
seemed to be only a “second-best” solution, which the Federal Reserve did
not endorse. The “first-best” solution, he said, is for BLS to depart from the
fixed-weight structure of the CPI.

The comments from CEA, OMB, and the Federal Reserve all convey a similar
message that the CPI should be a true or ideal cost-of-living index, which
has been discussed in terms of changing the CPI’s construction from a
Laspeyres index formula to a superlative index formula. As we stated in
the sections of this report on our objectives and methodology, it was not
our intent to evaluate a change in the basic formula used to construct the
CPI. However, the BLS Commissioner in her comments on the draft report
said, in part, the following:




Page 27                       GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




“Economic theory provides an elegant rationale for the use of superlative index
formulas . . . to provide approximations to a cost-of-living index. . . . The unfortunate
limitation of superlative formulas is that their calculation requires current-period
expenditure as well as price data, so superlative indexes can be published only with a lag.
This precludes their use in the CPI . . . .”


As we previously discussed in this report and as the Commissioner
mentioned in her comments, the administration has asked Congress for
funds to produce a BLS superlative index beginning in 2002. BLS also plans
to continue to publish the Laspeyres fixed market basket CPI. While we
agree with those who commented that updating the expenditure weights is
not a fix for turning the CPI into a true cost-of-living index, we believe that
such updating makes sense for the fixed market basket CPI as long as BLS
continues to publish it.

The Federal Reserve and CEA designees also expressed concern as to
whether we overstated the effect of more frequent updating on the CPI.
Both cited one estimate (0.04 percentage point reduction) from a
February 1997 research paper written by a BLS official to support their
concern.35 We believe we have not overstated the potential effect. We
report that a 5-year update of the expenditure weights could reduce the
CPI’s rate of growth by between 0 (zero) and 0.2 percentage point per year.
This range was estimated by BLS on the basis of historical evidence, which
was provided in this February 1997 research paper. BLS has raised no
second thoughts to us about the reasonableness of this range. For
example, BLS did not question the range in commenting on our draft report.
As we report in appendix III, the BLS research paper provided a number of
different point estimates that are based on regression analyses, overlap
comparisons, and other studies. The regression analysis from which the
0.04 percentage point estimate was derived found evidence of a small
effect—rather than no effect—on measured inflation. In addition, the
0.04 percentage point estimate was within the lower end of the range of
estimates that BLS provided to us.

CEA, OMB,and the Federal Reserve each made additional comments, which
are addressed as appropriate in appendixes VI, VII, and VIII.


As arranged with your office, unless you publicly announce its contents
earlier, we plan no further distribution of this report until 30 days after its
issue date. At that time, we will send copies of this report to the Chairman

35
  Greenlees, op. cit.



Page 28                             GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
B-275450




of this Committee; the Chairmen and Ranking Minority Members of other
interested congressional committees; the Secretary of Labor and the
Commissioner of BLS; the Director and the Chief Statistician of OMB; the
Chair of the Council of Economic Advisers; and the Chairman of the Board
of Governors of the Federal Reserve System. We will also make copies
available to others on request.

Major contributors to this report are listed in appendix IX. If you have any
questions about this report, please call either of us. Bernard Ungar can be
reached on (202) 512-8676, and James Bothwell can be reached on
(202) 512-6209.

Sincerely yours,




Bernard L. Ungar
Associate Director
Federal Management and
  Workforce Issues




James L. Bothwell
Chief Economist




Page 29                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Contents



Letter                                                                                              1


Appendix I                                                                                         34

Objectives, Scope,
and Methodology
Appendix II                                                                                        40
                       Construction of the CPI                                                     41
Background
Information on the
Consumer Price Index
Appendix III                                                                                       45
                       BLS Overlap Studies Consistently Show Differences Between                   45
Analysis of Price        Indexes Calculated With Old and New Weights
Indexes With           Alternative CPI Series Computed With Old and New Weights Also               48
                         Show Differences
Alternative
Expenditure Weights
Appendix IV                                                                                        51
                       Laspeyres Index Formula                                                     53
Price Index Formulas   Paasche Index Formula                                                       53
                       Fisher Ideal Index Formula                                                  54
                       Observations                                                                55

Appendix V                                                                                         56

Comments From the
Bureau of Labor
Statistics
Appendix VI                                                                                        59
                       GAO Comments                                                                61
Comments From the
Council of Economic
Advisers



                       Page 30                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                        Contents




Appendix VII                                                                                       62
                        GAO Comments                                                               64
Comments From the
Office of Management
and Budget
Appendix VIII                                                                                      66
                        GAO Comments                                                               68
Comments From the
Board of Governors of
the Federal Reserve
System
Appendix IX                                                                                        70

Major Contributors to
This Report
Related GAO Products                                                                               72


Tables                  Table III.1: Results From 1940 CPI Revision Overlap Tests                  46
                        Table III.2: Estimates of Difference in Annual Rates of Growth             47
                          Between Old and New Weights, by Major Revision
                        Table III.3: Estimates of Difference in Annual Rates of Growth             50
                          Between 1982 Through 1984 Weights and 1987 Through 1989
                          Weights
                        Table IV.1: Hypothetical Shopper’s Grocery Bills Over a 3-week             52
                          Period
                        Table IV.2: Laspeyres, Paasche, and Fisher Ideal Formulas Index            52
                          Values for Hypothetical Illustration

Figures                 Figure 1: Estimated Increase in Annual and Cumulative Federal              16
                          Budget Surplus if CPI Growth Were to Be Reduced Annually by
                          0.1 Percentage Point Over a 4-year Period After a 2003 Update
                        Figure II.1: Expenditure Weights Used for the 1987 CPI Revision            43




                        Page 31                     GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Contents




Abbreviations

BEA        Bureau of Economic Analysis
BLS        Bureau of Labor Statistics
CBO        Congressional Budget Office
CEA        Council of Economic Advisers
CEX        Consumer Expenditure Survey
CPI        consumer price index
CPI-U      consumer price index representing all urban consumers
CPI-W      consumer price index representing all urban wage and
                clerical workers
GDP        gross domestic product
OMB        Office of Management and Budget
PCE        Personal Consumption Expenditures
POPS       Point-of-Purchase Survey
SSA        Social Security Administration


Page 32                    GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Page 33   GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix I

Objectives, Scope, and Methodology


              To obtain views on updating the Consumer Price Index’s (CPI) expenditure
              weights more often than every 10 years, which addressed our first
              objective, we asked two former Bureau of Labor Statistics (BLS) officials
              and eight individuals who have studied the CPI for their views on how often
              the CPI should be updated. Of these eight individuals, one was a member of
              the Stigler committee; four were members of the Boskin commission; one
              had developed a superlative index theory that BLS was considering in
              connection with the CPI; and two had studied the Boskin commission’s
              report. These latter three researchers were recommended to us by the
              Boskin commission members we interviewed or others because their
              views were neutral or differed from the Boskin commission’s position on
              the amount of bias in the CPI. We also reviewed public statements made by
              the Chairman of the Board of Governors of the Federal Reserve System
              concerning the frequency of updating the CPI. To obtain information on the
              practices followed by other industrialized countries in updating their
              consumer price indexes, which also addresses our first objective, we
              obtained information from BLS and from publications of the Organization
              for Economic Cooperation and Development and the Canadian
              government on how often the G-7 countries update their CPIs.

              To estimate the cost to BLS of updating the CPI on a 5-year cycle, our
              second objective, we asked BLS to provide us with certain actual and
              estimated cost data. We asked BLS to provide us with the costs associated
              with the last major revision of the CPI, which took place in 1987, and the
              projected costs for the 1998 revision. In addition, we asked BLS for its
              estimate of what the costs would have been to update the CPI in 1992 and
              its estimate of what the cost might be to update the CPI in 2003. In other
              words, we asked BLS to provide cost data for a prior revision (1987), a
              planned revision (1998), and two updates (1992 and 2003). The interval
              between 1992 and 1998 is 6 years rather than 5 years, but that difference
              was unavoidable given that a major revision is scheduled for 1998. We did
              not specify to BLS what assumptions to make or what items to include or
              exclude in estimating costs for 1992, 1998, and 2003. We did not evaluate
              the reasonableness of BLS’ assumptions or estimates.

              BLS provided cost data for the 1987 revision, the upcoming 1998 revision,
              and a 2003 update. BLS suggested that the cost for a 1992 update could be
              derived by deflating the cost of the 2003 update. For this conversion, we
              compared the Object Class 11 index published by the Office of
              Management and Budget (OMB), which is used to adjust pay categories, and
              the Gross Domestic Product (GDP) price index as determined by the
              Department of Commerce’s Bureau of Economic Analysis, which is used



              Page 34                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix I
Objectives, Scope, and Methodology




to adjust all other budget categories. We found minor differences between
using the two deflators and chose to use the GDP price index, which, in
comparison to the Object Class 11 deflator, led to a slight overstatement of
the cost and, in reference to our cost-benefit comparison, provided a
conservative estimate.

To address our third objective—estimate the dollar effect on the federal
budget if the CPI weights were updated on a 5-year cycle—we obtained
assistance from BLS and the Congressional Budget Office (CBO). We asked
BLS to estimate whether the CPI would go up or down as the result of a
5-year update. More specifically, we asked BLS to provide a range—an
upper percentage point and a lower percentage point—of the possible
change that could occur to the CPI with a 5-year update.

To gauge the reasonableness of estimates that BLS provided, we compared
them with the results of BLS’ overlap studies of old and new weights during
the first 6 months of major revisions and other BLS studies that examined
the impact of more frequent updates.1 The results of these comparisons
are reported in appendix III.

We then asked CBO to estimate the effects that a 5-year update of the CPI in
2003 would have on federal outlays, revenues, debt service, and the overall
budget, assuming no other changes in tax or spending levels and no other
changes in the economy. To do this, we asked CBO to use BLS’ lower and
upper estimates and the midpoint of these estimates of change in the CPI
that would result from a 5-year update. To illustrate the effect of a 5-year
update that would begin in 2003, we asked CBO first to apply its standard
projections for the years 2004 through 2007;2 the results represented CBO’s
baseline. CBO then made additional projections for the years 2004 to 2007
to account for changes in the CPI as estimated by BLS for a 5-year update,
and we compared these projections against CBO’s baseline. To adjust the
CPI for the effects of a 5-year update of the expenditure weights, CBO
reduced its estimated CPI by 0.2 percentage point, which BLS had estimated
could be the upper estimate of change in the CPI from a 5-year update. The
difference between the baseline and the adjusted CPI estimates was

1
 Mary Lynn Schmidt, “Comparison of the Revised and the Old CPI,” Monthly Labor Review, Vol. 110
(Nov. 1987), pp. 3-6. Mary Lynn Schmidt, “Effects of Updating the CPI Market Basket,” Monthly Labor
Review, Vol. 116 (Dec. 1993), pp. 59-62. Government Price Statistics: Hearings Before the
Subcommittee on Economic Statistics of the Joint Economic Committee of the Congress of the United
States. 79th Cong., 1st Sess., pp. 582-585, (1961), (statement by Ewan Clague, Commissioner of Labor
Statistics). John S. Greenlees, “Expenditure Weight Updates and Measured Inflation,” Bureau of Labor
Statistics, February 27, 1997.
2
Because CBO’s projections do not go beyond 10 years, CBO did not make a projection for 2008, which
would be the fifth year of our illustrative 5-year span. The first year of CBO’s projections was 1998.



Page 35                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix I
Objectives, Scope, and Methodology




reported by us as the upper estimate of the dollar effect of a 5-year update
on the federal budget. Similar estimates and calculations were made with
the midpoint—0.1 percentage point—of BLS’ estimate. No further estimates
and calculations from the baseline were necessary to account for BLS’
lower estimate of change, which was 0 (zero).

We also reported these estimates in footnote 23 of this report in 1997
constant dollars by applying the GDP price index. We calculated the 1997
constant-dollar amounts using the GDP price index. Dollar amounts for
years other than the base year (1997) were adjusted for the effect of
inflation with the GDP price index. These adjustments had the effect of
increasing the amounts for the years before 1997 and of decreasing the
amounts for the years after 1997.

CBO would provide projections for current and future years but not provide
estimates for past years. Therefore, after discussion with CBO, we
estimated how the federal budget deficit might have been affected if the
expenditure weights were updated in 1992. In making our estimates, we
assumed that there were no other changes in tax or spending levels and no
other changes in the economy for 1992 through 2007. We also assumed
that the economic trend that was found for the years 1998 through 2007
could be reasonably applied to the years 1993 through 1998.

We first replicated CBO’s estimates for outlays and revenues that are
affected by the CPI for 1998 through 2007 from projections for the relevant
categories published in CBO’s Economic and Budget Outlook in
January 1997.3 We then asked CBO to follow the previously described
methodology and to make estimates of a 0.1 percentage point reduction in
the CPI beginning in 1997 and 2002. We also replicated CBO’s estimates of a
1.0 percentage point change in the CPI on the federal deficit that was
published in January 1997 and adjusted the estimates to represent a
0.1 percentage point reduction.4 We then compared the effects of these
two estimates on revenues and outlays and found significant differences in
the revenue estimates that were due to rounding rules for tax revenues.
We chose to use CBO’s published 1.0 percentage point estimates that were
adjusted to represent a 0.1 percentage point reduction rather than those
that were calculated for us by CBO because the published 1.0 percentage


3
  See The Economic and Budget Outlook: Fiscal Years 1998-2007, CBO (Washington, D.C.: 1997), pp. 24
and 36. The relevant category for revenues is individual income taxes. Spending programs affected by
changes in the CPI include Supplemental Security Income, veterans’ compensation and pensions,
Social Security, and federal civilian and military retirement.
4
 See The Economic and Budget Outlook: Fiscal Years 1998-2007, p. 41.



Page 36                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix I
Objectives, Scope, and Methodology




point estimates that were adjusted to represent a 0.1 reduction provided a
more conservative estimate, as well as a smooth trend. The effects from
the adjusted 1.0 percentage point reduction were then applied to the
outlay and revenue totals that are affected by changes in the CPI in each
year from 1993 to 1998.5 Debt service costs were calculated from the
year-to-year change in CBO’s baseline debt, less the saving from changes in
outlays and revenues. The current-dollar estimates derived from these
calculations were adjusted with the GDP price index to 1997 constant
dollars. The estimates for the 0.1 percentage point reduction were doubled
to obtain estimates for a 0.2 percentage point reduction in the CPI. We met
with CBO to discuss our approach, and CBO staff stated that the method and
results appeared reasonable.

In connection with the effect on the federal budget, we asked the Chief
Actuary of the Social Security Administration (SSA) to estimate the effect a
change in the CPI would have on the average benefit paid to retired
workers. Your office had requested that we obtain this information to
illustrate how a federal revenue or payment program that is adjusted
periodically because of changes in the CPI might be indirectly affected by
more frequent updating of the expenditure weights. To make the estimate,
we asked SSA to use the midpoint of BLS’ range of possible change that
would occur to the CPI with a 5-year update. Using that midpoint
percentage point and the President’s Fiscal Year 1998 Budget assumptions,
SSA estimated the change in the average monthly benefit, beginning with
December 2003, which would be payable in January 2004, and continued
through December 2007.

Our fourth objective had the following two elements: (1) identify the
reasons for the 10 years or so between revisions and (2) assess those
reasons. For the first element, we interviewed present and past officials of
BLS and obtained their views on why major updates to the CPI have been
spaced about 10 years apart. Among the officials we interviewed were the
present Commissioner of BLS and a former Commissioner of BLS. We also
reviewed the 1961 congressional testimony of another BLS Commissioner
in which he addressed the subject of BLS’ timetable for revising the CPI. In
our assessment of BLS’ reasons, we (1) collected and analyzed information
on past comparisons between indexes that applied old and new
expenditure weights, which were also used to address the reasonableness
of BLS’ estimates under our third objective; (2) obtained information on
how BLS collects its source data for the CPI; (3) obtained BLS fiscal year
1998 budget information to ascertain BLS’ plans for future changes in its

5
 Actual amounts were used for 1992 through 1995; CBO estimates were used for 1997 and 1998.



Page 37                               GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix I
Objectives, Scope, and Methodology




indexes; and (4) compared the estimated costs and benefits of a 5-year CPI
update cycle obtained under our second and third objectives.

In our assessment of past comparisons between indexes that applied old
and new weights, we noted that the indexes reflect differences in addition
to those directly related to changes in expenditure weights, such as
conceptual changes in the structure of the market basket. These
differences are a result of data limitations in that the overlap periods
incorporate many factors that can be changed in a revision. With this
knowledge, we treated the differences as indicators of the effect of an
update of the expenditure weights because an update of the weights is
unlikely to occur in isolation from the other factors that are associated
with revisions. For example, a 1992 update would have incorporated a
market basket that would have been based on different geographic areas
because changes were made in 1986 in the geographic locations where
expenditure data were collected. Such geographic changes are associated
with major revisions.

In addition to reviewing overlap studies, we examined the effect of the age
of weights with indexes that were calculated with alternative base periods.
For example, comparisons were made of the official CPI’s 3-year base of
1982 through 1984 with alternative 3-year base periods (i.e., 1987 through
1989). In these and other comparisons, we applied an economic concept
that is based upon economic literature that suggests that an index is more
accurate if the expenditure weights used to compute it represent, as much
as practical, current consumer spending.

As previously reported, this report includes, and often relies on, estimates
and comparisons prepared by BLS, CBO, or SSA.6 We did not verify the
computerized data that the agencies used in producing these estimates and
comparisons. Verification, in our opinion, would have been impractical
because it would have been costly and time consuming. In addition, the
estimates and comparisons were within the scope of activities that BLS,
CBO, and SSA normally perform. Therefore, we used their estimates and
comparisons.




6
 BLS estimated (1) the cost to update the expenditure weights and (2) the range of percentage effect
on the CPI if the expenditure weights were updated at 5-year intervals. BLS also provided us with its
past comparisons of old-weighted and new-weighted indexes. CBO estimated the dollar effect on the
federal budget if CPI growth were lowered 0.1 percentage point or 0.2 percentage point annually. SSA
estimated the impact on Social Security payments if the CPI were reduced annually by 0.1 percentage
point.



Page 38                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix I
Objectives, Scope, and Methodology




Our work was designed to examine the importance of updating the CPI
sooner than about once every 10 years. The results we obtained are
intended to contribute to the discussion of how often the CPI should be
updated but are not intended to represent all future effects of shortening
the updating cycle. The point of shortening the updating cycle is to have
the CPI reflect, as close as practical, the current spending patterns of
consumers, regardless of whether the index is pushed upward or
downward. Our work is also not intended to evaluate a change in the basic
formula that could address substitution bias in the CPI.




Page 39                          GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix II

Background Information on the Consumer
Price Index

                  BLS produces the CPI by measuring the average change over time in the
                  prices paid by urban consumers for a fixed market basket of consumer
                  goods and services. The market basket is determined from detailed
                  records of purchases made by thousands of individuals and families. The
                  items selected for the market basket, such as potatoes, are to be priced
                  each month at retail outlets, such as grocery stores, in urban areas
                  throughout the country. According to BLS, in 1995, field representatives
                  visited approximately 30,000 retail establishments and housing units each
                  month, with prices collected for 94,000 items.

                  The CPI is used as a measure of price changes to make economic decisions
                  in the private and public sectors. According to BLS, the CPI has three major
                  uses as follows:

              •   Economic indicator of inflation. The administration, Congress, and the
                  Federal Reserve use trends in the CPI as an aid to formulating fiscal and
                  monetary policies. Business and labor leaders, as well as private citizens,
                  use the CPI as a guide to making economic decisions.
              •   Escalator for wages, benefit payments, and tax brackets. In 1996, the CPI
                  was used by collective bargaining units to adjust the wages of 1.7 million
                  workers. It is used to adjust some federal benefit payments for inflation.
                  For example, in September 1996, as a result of changes in the CPI,
                  43.5 million Social Security beneficiaries; 6.6 million Supplemental
                  Security Income recipients; 6.4 million railroad, military, and federal
                  civilian retirees and survivors; and 25.8 million food stamp recipients had
                  their benefits adjusted for inflation. The CPI is also used to adjust the
                  federal individual income tax structure to prevent bracket creep (i.e.,
                  increases in real tax rates due solely to inflation). Some benefit payments,
                  such as those for Social Security recipients; tax deductions for personal
                  exemptions; and tax brackets are adjusted automatically by the CPI, rather
                  than on the basis of discretionary policy decisions.
              •   Deflator of selected economic statistical data series. The CPI is used to
                  adjust selected economic statistical series for price changes and to
                  translate these series into inflation-free dollars. Examples of data series
                  that are adjusted by the CPI include retail sales, hourly and weekly
                  earnings, and components of the National Income and Product Accounts.

                  The CPI was initiated during World War I, when rapid increases in the
                  prices of goods and services, particularly in shipbuilding centers, made
                  such an index essential for calculating cost-of-living adjustments in wages.
                  In 1921, BLS began regular publication of an index representing the
                  expenditures of urban wage and clerical workers, which was then called



                  Page 40                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                      Appendix II
                      Background Information on the Consumer
                      Price Index




                      the Cost-of-Living Index. The name of the index was changed to the CPI
                      following controversy during World War II over the index’s validity as a
                      measure of the cost of living. According to BLS, the CPI has always been a
                      measure of the changes in prices for goods and services purchased for
                      family living.

                      Major revisions were made to the CPI about every 10 years to update the
                      fixed market basket; the next major revision is scheduled to be released in
                      January 1998. Because consumers’ buying habits changed, new studies
                      were made of what goods and services consumers were purchasing, and
                      major revisions to the CPI were made in 1940, 1953, 1964, 1978, and 1987. In
                      the 1978 major revision, several changes were made, including the
                      publication of a new index for all urban consumers—the CPI-U. According
                      to BLS, the CPI-U, which represents the expenditures of about 80 percent of
                      the population, takes into account the buying patterns of professional
                      employees, part-time workers, the self-employed, the unemployed, and
                      retired people, as well as those previously covered in the CPI. BLS has
                      continued publication of the older index, the CPI-W, which represents the
                      expenditures of urban wage and clerical workers, or about 32 percent of
                      the population.


                      Construction of the CPI begins by selecting a collection of goods and
Construction of the   services that is usually bought by the reference population in the index.
CPI                   The collection of goods and services, called items, is known as the market
                      basket. The CPI market basket is developed from detailed expenditure
                      information that is provided by families and individuals who participate in
                      the Consumer Expenditure Survey (CEX). Altogether, about 29,000
                      individuals and families provide expenditure information for use in
                      determining the importance, or weight, of each item in the index structure.
                      These data are also used to select the categories of items from which
                      specific, unique commodity and service items are selected to be priced for
                      the CPI.

                      BLS measures price changes each month by checking the prices of the
                      items in the market basket and then comparing the aggregate costs of the
                      market basket with those for the previous month. BLS field representatives
                      obtain prices for most of the items through personal visits to
                      approximately 30,000 retail establishments and housing units.




                      Page 41                         GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                      Appendix II
                      Background Information on the Consumer
                      Price Index




Components            BLS classified all CEX expenditure items into 206 item strata, which are
                      arranged into 7 major components: (1) food and beverages; (2) housing;
                      (3) apparel and upkeep; (4) transportation; (5) medical care;
                      (6) entertainment; and (7) other goods and services, such as haircuts,
                      college tuition, and bank fees. Taxes that are directly associated with the
                      prices of specific goods and services, such as sales and excise taxes, are
                      also included.1


Expenditure Weights   Expenditure weights are used to give proportionate emphasis for price
                      changes of one item in relation to other items in the CPI. Expenditure
                      weights allow the CPI to distinguish between items that have a major
                      impact on consumers and to provide appropriate emphases to price
                      changes associated with these items.

                      The weight of an item in the CPI market basket is derived from consumers’
                      expenditures as reported in the CEX. To compute the weight, BLS first totals
                      the amount spent on an item stratum, such as white bread, by CEX
                      respondents during the base weighting period. BLS then divides that total
                      by the number of CEX responding units, which results in an average
                      expenditure per unit. Next, the average expenditures per unit are weighted
                      with data from the decennial census to represent the U.S. urban
                      population. To do so, the average expenditure amounts are multiplied by
                      certain factors to represent the geographic dispersion of the urban
                      population. Finally, these nationwide urban expenditures on the market
                      basket items are totaled into an aggregate amount. The 206 expenditure
                      weights are the percentages of this aggregate amount that are spent on
                      each of the 206 item strata (e.g., white bread).

                      On the basis of average expenditures during the reference period,
                      expenditure weights remain fixed, or constant, until the next major
                      revision of the CPI and serve as a benchmark from which price
                      comparisons are calculated. The weights of the components for the last
                      major revision in 1987 are those as derived from the 1982 through 1984 CEX
                      (see fig. II.1).




                      1
                       The CPI excludes taxes not directly associated with the purchase of consumer goods and services,
                      such as income and Social Security taxes. The CPI does not include investment items, such as stocks,
                      bonds, real estate, and life insurance, because they relate to savings, not daily living expenses.



                      Page 42                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                                        Appendix II
                                        Background Information on the Consumer
                                        Price Index




Figure II.1: Expenditure Weights Used
for the 1987 CPI Revision
                                                                                            6.5%
                                                                                            Apparel and upkeep

                                                                                            5.1%
                                                                                            Other goods and services

                                                                                            4.8%
                                                                                            Medical care

                                                                                            4.4%
                                                                                            Entertainment


                                                               •
                                                           •

                                                       •
                                                  •
                                                                      42.6% •               Housing

                                             17.8%
                                                •


                                                           18.7%
                                                              •


                                                                                            Transportation

                                                                                            Food and beverages




                                        Note: Percentages do not total to 100 percent because of rounding.

                                        Source: BLS.




Pricing of Market Basket                Each month, BLS field representatives visit or call thousands of retail
Items                                   stores, service establishments, rental units, and doctors’ offices all over
                                        the United States. For the entire month, they record the prices of about
                                        94,000 items. To determine which retail outlets its representatives should
                                        visit to obtain its monthly price quotations, BLS sponsors the
                                        Point-of-Purchase Survey (POPS), which is conducted by the Bureau of the




                                        Page 43                               GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix II
Background Information on the Consumer
Price Index




Census.2 The survey respondents are asked, by item categories such as
doctors, whether they made specific purchases and, if so, the names and
locations of all places of purchases and the expenditure amounts. BLS uses
the results from the survey to select outlets for pricing. This survey is
conducted in approximately 20 percent of a sample of urban areas each
year; as a result, the entire nonshelter sample is updated every 5 years.

BLS  field representatives visit each selected outlet to initially select items
that will be priced either monthly or bimonthly. For each outlet, categories
of items are selected for pricing. Using probability selection methods that
are based on revenues and volume information that is provided by the
retail outlet, BLS field representatives use a table of random numbers to
select for pricing a unique item within the specified categories. The
monthly price changes for the same item (e.g., cigarettes) that are
collected by BLS field representatives in urban areas throughout the United
States are averaged, weighted, and published. Because the concepts BLS
uses to measure medical care and shelter costs are different than those
used for the items previously described, the pricing of these items is
approached in a different manner.




2
 The POPS also is used to establish the point-of-purchase weights, which are used to combine the
prices of the 94,000 items into the 206 item strata.



Page 44                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix III

Analysis of Price Indexes With Alternative
Expenditure Weights

                      BLS reported that historical evidence suggests that a 5-year update of the
                      market basket could reduce the rate of growth of the CPI by between 0
                      (zero) and 0.2 percentage point per year. Regarding the effect of a 1991
                      update to the CPI, BLS cited one specific source of evidence that shows the
                      rate of growth would be lower by 0.11 percentage point. In addition, BLS
                      states that the effect of updating the expenditure weights in 1998 will
                      likely be a reduction of 0.1 or 0.2 percentage point. As a result of this and
                      other information, we chose 0.1 percentage point, the midpoint of BLS’
                      range, for the purposes of our calculations.

                      Evidence on the possible impact of more frequent updates of expenditure
                      weights on the rate of growth of the CPI includes overlap studies
                      performed by BLS at the time of major revisions and other CPI index
                      comparisons using specialized databases. BLS has been performing overlap
                      studies for more than 50 years, and these studies consistently have shown
                      a difference between indexes computed with old and new weights. In most
                      of these cases, indexes computed with the old weights show a higher rate
                      of growth than indexes computed with the new expenditure weights.
                      Other evidence includes alternative CPI index series, which were computed
                      using databases that allow comparisons between old and new weights.
                      Indexes computed with new expenditure weights almost always produce
                      different results than indexes with old weights. These same data also
                      suggest that indexes relying on older expenditure weights typically show a
                      higher rate of growth than indexes computed with newer expenditure
                      weights. An upward bias in indexes computed with older expenditure
                      weights is consistent with economic theory and other evidence. In our
                      analysis, we applied an economic concept that an index was more
                      accurate if the expenditure weights used to compute it represented, as
                      much as practical, current consumer spending.


                      BLS began performing overlap studies for its work on the first revision of
BLS Overlap Studies   the CPI in 1940. In these overlap studies, BLS computed two indexes for the
Consistently Show     same period. One index was calculated with the original weights, and one
Differences Between   was computed with more recent weights. As a result, these overlap
                      indexes provide evidence on the effects of updating weights over a long
Indexes Calculated    period and under different economic conditions.
With Old and New
                      In those cases where there were no important changes in other procedures
Weights               or other anomalies, the difference between these two indexes can be
                      attributed to the change in weights. However, BLS has often instituted new
                      procedures and improvements as a part of the revisions, along with the



                      Page 45                       GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                                     Appendix III
                                     Analysis of Price Indexes With Alternative
                                     Expenditure Weights




                                     updates to the expenditure weights. These improvements could include
                                     changes in geographic coverage, adoption of probability sampling
                                     methods, and other changes. Those changes and any unusual economic
                                     conditions at the time of the revision limit the applicability of the overlap
                                     findings to current questions regarding the likely effects of updating
                                     expenditure weights.

                                     In all of the overlap comparisons, differences were found between indexes
                                     calculated with the new weights and indexes calculated with the old
                                     weights. In five of the seven comparisons, the indexes with the new
                                     weights recorded a lower rate of inflation than those with the old weights.
                                     In the two instances where the new weights resulted in higher rates of
                                     growth, changes in price collection methodology and the aftermath of the
                                     wartime economy might have had an effect on those results.


Studies for the 1940                 As a part of the first revision of the CPI in 1940, BLS conducted three
Revision                             comparisons that used expenditure weights that were derived from 1917
                                     through 1919 Consumer Expenditure Survey (CEX) data (old weights) and
                                     weights from 1934 through 1936 CEX data (new weights). Comparisons
                                     were conducted for three different periods using the old and the new
                                     weights: 1925 through 1929, June 1930 through March 1935, and
                                     March 1935 through December 1939. Differences were reported for all
                                     three comparisons, and, in each case, the CPI with the old weights had a
                                     higher rate of growth than the index computed with new weights. In the
                                     periods beginning 1925 and 1935, the difference between the two indexes
                                     was small, but the difference between the overlapping indexes for the
                                     period beginning in 1930 was more than 1 percent (see table III.1).

Table III.1: Results From 1940 CPI
Revision Overlap Tests               Numbers in percent
                                                                     Rate of growth          Rate of growth
                                     Calculation period           with new weights          with old weights                Difference
                                     1925 - 1929a                                 –3.7                     –3.6                      –0.1
                                     June 1930 - March                           –18.7                   –17.5                       –1.2
                                     1935a
                                     March 1935 -                                   1.8                     1.9                      –0.1
                                     December 1939
                                     a
                                      During periods of deflation, the indexes computed with the old weights showed less deflation
                                     than indexes computed with the new weights.

                                     Source: BLS.




                                     Page 46                               GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                                          Appendix III
                                          Analysis of Price Indexes With Alternative
                                          Expenditure Weights




Overlap Studies for 1953                  In response to a presidential request, BLS also conducted an overlap study
Through 1987 Revisions                    related to the 1953 revision. In this case, BLS applied the weights used in
                                          the 1940 CPI revision to January 1953 through June 1953 price data, the first
                                          months following the implementation of the 1953 revision. A comparison
                                          of the indexes computed with these two sets of weights showed that,
                                          unlike the previous 1940 revision studies where the old weights produced
                                          a higher rate of growth than the new weights, the index with the old
                                          weights showed growth of 0.5 percentage point less than the index using
                                          the new weights (see table III.2).1 These results may be an anomaly related
                                          to the use of the 1947 through 1949 CEX data as the base period for the 1953
                                          revision. Consumption in those years reflected the purchases of
                                          consumers following World War II, and the change may reflect unusual
                                          changes in consumer preferences or changes in the availability of various
                                          goods and services.

Table III.2: Estimates of Difference in
Annual Rates of Growth Between Old        Numbers in percent
and New Weights, by Major Revision        Revision year                                                                              Difference
                                          1953a                                                                                              –0.5
                                          1964                                                                                                  0.4
                                          1978                                                                                               –0.2
                                          1987                                                                                                  0.8
                                          1998b                                                                                           0.1/0.2
                                          Note: A negative number indicates that the price index using old weights was lower than the
                                          index using new weights.
                                          a
                                              The months of comparison were January through June 1953.
                                          b
                                              BLS forecast.

                                          Source: BLS.



                                          In the more recent series of overlap studies, BLS calculated the CPI with
                                          both the old and new weights for 6 months following a major revision.2 In
                                          the 1964 and the 1987 revisions, indexes computed with the old weights
                                          produced higher growth rates than indexes computed with the new
                                          weights. Overlap indexes computed for the 1978 revision showed the

                                          1
                                           The 1940 revision expenditure weights are considered the old weights in this section; whereas the
                                          1953 revision weights are considered the new weights. In each of the following discussions, the old
                                          weights are those that were used through December before the introduction of the revision under
                                          discussion; the new weights are those that were introduced in January of the revision year under
                                          discussion.
                                          2
                                           The results from these studies are reported in a research paper written by a BLS official (Greenlees,
                                          op. cit.).



                                          Page 47                                 GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                         Appendix III
                         Analysis of Price Indexes With Alternative
                         Expenditure Weights




                         reverse (see table III.2). BLS suggested, however, that the lower rates of
                         inflation produced by the older weights may have been due to a 1978
                         change in methods used to select items for monthly pricing (see app. II).
                         According to BLS, an upward bias could have been reflected in the index
                         calculated with the new weights, and could have caused the higher rates of
                         inflation using the new weights. In addition, BLS estimated that, for the
                         upcoming 1998 revision, measured inflation will likely be reduced by 0.1 or
                         0.2 percentage point.


                         Additional evidence on the effect of more frequent updates of expenditure
Alternative CPI Series   weights is available from comparisons of index series computed with old
Computed With Old        and new weights. These calculations are performed on databases that have
and New Weights Also     been constructed to allow comparisons of various index weights and
                         methodologies. Although these databases are not available for the
Show Differences         historical periods covered by the overlap indexes, they make it possible to
                         compare a number of alternative CPI series that are based on various
                         combinations of 1- and 3-year weights for 1982 through 1995. For example,
                         one database has been used to compare the actual CPI, which is based on
                         1982 through 1984 weights, with indexes that were computed with 3-year
                         weights from the following periods: 1987 through 1989, 1988 through 1990,
                         1989 through 1991, and 1990 through 1992. Other databases have been
                         used to compare CPI rates of growth that are based on various
                         combinations of 1-year weights.

                         Evidence from these databases suggests that changing the weights used in
                         the computation of the CPI will usually change the rate of growth of the CPI.
                         These rate of growth differences vary according to the database, the years
                         used for the older and newer weights, and the specific time frame and
                         methodology used for computing the index. As in the case of the overlap
                         studies, most of the evidence suggests that older weights typically produce
                         a higher rate of growth in the CPI than indexes computed with newer
                         weights. For example, a comparison of indexes computed with the 1982
                         through 1984 base period and the indexes computed with weights that
                         were based on the 1987 through 1989 periods indicated that the older
                         weights overstated inflation by approximately 0.1 percentage point per
                         year over a 5-year period. BLS cited this and other evidence to support its
                         statement that a 5-year update would reduce the measured rate of inflation
                         by between 0 and 0.2 percentage point per year.3



                         3
                          Greenlees, op. cit.



                         Page 48                            GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                           Appendix III
                           Analysis of Price Indexes With Alternative
                           Expenditure Weights




                           As in the case of the overlap studies, BLS noted that the difference between
                           the two indexes cannot be attributed only to changes in the weights. In
                           those instances when there are important changes in procedures or
                           anomalies in the data, the differences may not be an accurate reflection of
                           the changes in the weights. BLS also noted that the differences may be
                           affected by the overall rates of inflation, in that differences may be larger
                           during periods when the overall inflation rates are higher.


Alternative Indexes With   BLS  research that compared the 1982 through 1984 expenditure-based CPI
3-Year Base Periods        (official) with alternative 3-year expenditure base periods indicated that
                           the official CPI rose slightly faster, on average, than the alternative indexes
                           with more current weights. For example, comparisons of the actual CPI
                           with indexes computed with 3-year base periods starting with 1987
                           through 1989 and ending with 1990 through 1992 showed that the
                           increases in the official CPI were often higher than the alternative indexes
                           with more recent base periods, but there was no consistent finding across
                           all of the 3-year combinations. In fact, price changes for 1994 for two of
                           the more recent base-period indexes (1987 through 1989 and 1989 through
                           1991) were larger than the official CPI for that year.


Alternative Indexes With   Two separate databases were also created that allowed additional
1-Year Base Periods        comparisons to be made between the CPI rates of growth, using a number
                           of alternative index calculation methods and 1-year base periods. The
                           following different expenditure base periods were available in those
                           databases: (1) 1986 through 1995 and (2) 1982 through 1995. These two
                           databases allow a variety of comparisons to be made of indexes, using
                           different base-period expenditure weights.

                           BLS used a regression analysis to summarize the effect that the age of the
                           weights has on price indexes.4 The analysis that was based on the first
                           database provided no evidence that a price index calculated with more
                           current weights will produce a lower rate of inflation than an index
                           calculated with older weights. However, BLS’ regression analysis with the
                           second database found evidence that more current base weights yield
                           smaller estimates of price change.5 In other words, indexes based on older

                           4
                            In many of these regressions, BLS also estimated the difference between a chained Fisher index and a
                           1-year Laspeyres index estimate.
                           5
                            BLS suggested that the incorporation of the 1980 census-based geographic samples into the CEX in
                           1986 could explain the difference between their two regression analyses. BLS also looked at spending
                           distributions during the period before the change in samples and found that no particular market
                           basket item (such as energy or high-tech consumer goods) contributed to the differences.



                           Page 49                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                                          Appendix III
                                          Analysis of Price Indexes With Alternative
                                          Expenditure Weights




                                          expenditure weights tended to show a higher rate of inflation than indexes
                                          with more recent weights.

                                          These data were also used to more directly address the question of the
                                          impact of a 5-year update on the rate of measured inflation. For this
                                          purpose, BLS compared inflation estimates obtained with the 1982 through
                                          1984 weights with the inflation rates that would have been produced by
                                          1987 through 1989 weights (see table III.3). This comparison indicated
                                          that, on average, inflation was lower by 0.11 percentage point with the
                                          1987 through 1989 average than with the 1982 through 1984 average (i.e., a
                                          5-year update would have reduced inflation by an average of about
                                          0.11 percentage point per year). BLS noted, however, that the difference
                                          varied widely from year to year.

Table III.3: Estimates of Difference in
Annual Rates of Growth Between 1982       Numbers in percent
Through 1984 Weights and 1987                                          Average rate of      Average rate of
Through 1989 Weights                                                      growth with          growth with
                                          Revision year               1982-84 weights      1987-89 weights           Difference
                                          1991                                     4.23               4.04                 0.20
                                          1992                                     3.02               2.95                 0.07
                                          1993                                     2.94               2.82                 0.13
                                          1994                                     2.64               2.59                 0.05
                                          1995                                     2.84               2.73                 0.11
                                          Average                                  3.13               3.02                 0.11
                                          Source: BLS.




                                          Page 50                            GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix IV

Price Index Formulas


              A consumer price index may be computed with one of several index
              formulas. The purpose of this appendix is to illustrate several of those
              formulas.

              BLSconstructs the CPI with a modified Laspeyres index formula.1
              According to the economist George Stigler, a Laspeyres index formula
              produces an “upper” bound of the cost of living; whereas, the Paasche
              index formula produces a “lower” bound of the cost of living.2 A
              superlative index, such as the Fisher Ideal index formula, is regarded by
              economists as providing a good approximation to a cost-of-living index.

              To illustrate the three basic index formulas, we use information from
              hypothetical weekly grocery bills of a single woman who eats breakfasts
              and five dinners at home; the rest of her meals are eaten away from home
              (see table IV.1). In the first and second weeks, she purchased the same
              identical items, with price increases occurring in the second week for
              some items. In the third week, the prices of some items increased, and she
              altered her market basket by purchasing a different fruit. In all three
              weeks, we assumed that she attained the same level of satisfaction from
              the consumption of these food items.




              1
               Price index formulas are named for the persons who developed them. The index formulas and the
              theoretical discussions about them can be found in W.E. Diewert, “The Theory of the Cost-of-Living
              Index and the Measurement of Welfare Change” Price Level Measurement, ed. W.E. Diewert (New
              York: North Holland, 1990), pp. 79-147.
              2
               This description is true as long as consumer preferences for goods and services are homothetic (i.e.,
              expenditure shares among goods and services stay constant when income changes). If preferences are
              not homothetic, there are different cost-of-living indexes corresponding to different income and price
              situations, and the Laspeyres and Paasche indexes do not serve as upper and lower bounds on the
              same cost-of-living index.



              Page 51                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                                            Appendix IV
                                            Price Index Formulas




Table IV.1: Hypothetical Shopper’s Grocery Bills Over a 3-Week Period
                                  Week 1                              Week 2                                                  Week 3
                                      Price per   Weekly                          Price per      Weekly                           Price per   Weekly
Grocery item              Quantity         item     cost            Quantity           item        cost               Quantity         item     cost
Bananas                          5       $0.20        $1.00                   5        $0.20        $1.00                    0       $0.20        $0
Blueberries                      0        1.99             0                  0          1.99            0               1 box         1.49     1.49
Cereal                       1 box        3.89         3.89             1 box            3.89        3.89                1 box         4.19     4.19
Milk                        1 quart         .99          .99          1 quart            1.05        1.05               1 quart        1.05     1.05
Orange juice, frozen   2 small cans         .95        1.90     2 small cans              .95        1.90       2 small cans            .99     1.98
Peas, frozen                 1 bag        1.29         1.29             1 bag            1.35        1.35                1 bag         1.35     1.35
Prepared dinner,
frozen                           1        2.49         2.49                   1          2.49        2.49                    1         2.49     2.49
Dinner rolls            1 package         1.49         1.49       1 package              1.50        1.50            1 package         1.50     1.50
Leaf lettuce                1 head          .69          .69           1 head             .69          .69              1 head          .69      .69
Tomato                           1          .89          .89                  1           .89          .89                   1          .99      .99
Chicken breasts         1 package         3.59         3.59       1 package              3.59        3.59            1 package         3.59     3.59
Hamburger                  1 pound        1.59         1.59          1 pound             1.59        1.59             1 pound          1.59     1.59
Potatoes                         2          .25          .50                  2           .25          .50                   2          .25      .50
Rice, flavored          1 package         1.79         1.79       1 package              1.89        1.89            1 package         1.89     1.89
Total                     19 items         n/a     $22.10           19 items              n/a     $22.33              15 items          n/a   $23.30




                                                      Price change or change in item purchased from previous week.

                                            Legend: n/a = Not applicable

                                            Source: Hypothetical example developed by GAO.


                                            The values for three price index formulas—Laspeyres, Paasche, and Fisher
                                            Ideal—that would be derived for our hypothetical illustration are provided
                                            in table IV.2.

Table IV.2: Laspeyres, Paasche, and
Fisher Ideal Formulas Index Values for                                                  Index values using three basic formulas
Hypothetical Illustration                   Week                                      Laspeyres                       Paasche            Fisher Ideal
                                            1                                                100.0                       100.0                 100.0
                                            2                                                101.0                       101.0                 101.0
                                            3                                                103.2                       100.9                 102.0
                                            Source: GAO.




                                            Page 52                                    GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                  Appendix IV
                  Price Index Formulas




                  The following descriptions are simplified to show how the indexes differ
                  conceptually. Although the basic concepts presented are accurate, the
                  actual calculations would be substantially more complex.


                  An index calculated with a Laspeyres index formula measures price
Laspeyres Index   changes in relation to the base period’s market basket and thereby “fixes”
Formula           the market basket by holding the items in it constant. It calculates what
                  that market basket would cost in later periods, even if some of the items
                  were no longer purchased.

                  For our hypothetical shopper, the Laspeyres index formula uses what she
                  purchased in the first week as the base of the calculation—the fixed
                  market basket. All comparisons are made with respect to the quantities of
                  items she purchased and the prices she paid for them in the first week.
                  Since she purchased the same items in the second week, a Laspeyres
                  index would divide her grocery bill for the second week by the first week’s
                  bill and obtain an index value of 101.0, as shown in table IV.2.

                  Since the shopper bought blueberries instead of bananas in the third week,
                  the third week’s grocery bill cannot be simply divided by the first week’s
                  bill. An adjustment must be made to reconstruct the first week’s fixed
                  market basket by subtracting the cost of the blueberries and adding the
                  cost of the bananas as if she had purchased bananas in the third week.
                  This is done to make the third week’s market basket identical to the first
                  week’s market basket. A Laspeyres index value of 103.2 is obtained by
                  dividing the adjusted third week’s grocery bill by the first week’s bill.


                  An index calculated with a Paasche index formula measures price changes
Paasche Index     for a market basket containing what consumers are currently purchasing,
Formula           rather than what they purchased in a previous period. This index assumes
                  that consumers’ tastes and preferences change to maintain a constant level
                  of satisfaction and compares the cost of the consumers’ current market
                  basket with what it would have cost to buy this basket of goods and
                  services in an earlier period.

                  As shown in table IV.2 for the hypothetical illustration, the Paasche index
                  value (101.0) is the same as the value obtained with the Laspeyres index
                  formula (101.0) in the first 2 weeks because our hypothetical shopper
                  purchased the same items. However, the difference between the Paasche
                  and the Laspeyres formulas is evident in the third week when the shopper



                  Page 53                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                     Appendix IV
                     Price Index Formulas




                     purchased blueberries in place of bananas. Instead of pricing bananas as
                     was done in the Laspeyres calculation, blueberries remain in the market
                     basket for the Paasche calculation and are priced as if they had been
                     purchased in the first week. A Paasche index value of 100.9 is obtained by
                     dividing the third week’s grocery bill by an amount that reflects the third
                     week’s market basket priced with prices charged during the first week.

                     Because the items, quantities, and prices changed between the second and
                     third weeks, the Paasche index value for the third week cannot be
                     compared with the Paasche value for the second week. The difference
                     between these 2 weeks cannot be referred to as a price change because
                     the shopper changed the type and quantity of fruit she purchased. The
                     index numbers for the third week, calculated with a Paasche index
                     formula, can be compared only with the base period—the first week.
                     (Also, the index number for the second week can be compared only with
                     the base period.)


                     The Fisher Ideal index formula uses the Laspeyres and Paasche index
Fisher Ideal Index   values and, therefore, does not allow comparisons between adjacent
Formula              periods. To allow comparisons of index values between adjacent periods,
                     an adjustment—chaining—can be made to the Fisher Ideal index values.
                     In this section, we first describe a Fisher Ideal index with the illustration
                     of our shopper, and we then describe chaining with the Fisher Ideal index.
                     Both the Fisher Ideal index and its chain are superlative price indexes.

                     A Fisher Ideal index number is the square root of the product of the
                     Laspeyres index number multiplied by the Paasche index number. For
                     example, in the third week of our illustration, the Fisher Ideal index
                     number of 102.0 is the square root of the product of 103.2 (Laspeyres) and
                     100.9 (Paasche). The result of the Fisher Ideal index is a geometric mean,
                     which differs from an arithmetic mean, or average. For example, the third
                     week’s arithmetic mean of 102.1 is 103.2 (Laspeyres) plus 100.9 (Paasche)
                     divided by 2.3

                     Because the Fisher Ideal index incorporates the Paasche index value in its
                     calculation, the limitations of the Paasche also transfer to the Fisher. For
                     example, the comparison of the values for the third week with the values
                     of the second week cannot be interpreted as a price change because the
                     shopper purchased blueberries instead of bananas. As with the Paasche,

                     3
                      A geometric mean treats price increases and decreases symmetrically, whereas an arithmetic mean,
                     as used in a Laspeyres index, gives price increases more influence, thereby showing a faster rate of
                     price change.



                     Page 54                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
               Appendix IV
               Price Index Formulas




               the index numbers for the second and third weeks can only be compared
               with the base period—the first week—and not to each other.

               The chained Fisher Ideal index is the square root of the product of the
               chained Laspeyres index number multiplied by the chained Paasche index
               number. A chained index “chains” period-to-period indexes back to the
               reference period (i.e., week 1 in the hypothetical illustration). Because
               they are chained to each other, comparisons can be made between any
               sets of index number values.

               The Laspeyres and Paasche chained indexes are calculated similarly—the
               previous chained index value is multiplied by a price relative, which is a
               ratio of the previous and current unchained index values. For example, to
               chain the Laspeyres index numbers between the first and second weeks in
               our hypothetical illustration, the chained Laspeyres index number for the
               first week (100.0), which is also the base, is multiplied by the price relative
               of 1.01, which is the ratio between the Laspeyres index numbers for the
               first and second weeks (101.0 divided by 100.0). To calculate the chained
               value for the third week, the chained value for the second week (101.0) is
               multiplied by the price relative for the third week (1.02). These same
               procedures are followed to obtain the chained Paasche index values.
               Then, to obtain the chained Fisher Ideal index formula values, the square
               root of the product of the chained Laspeyres index number is multiplied
               by the chained Paasche index number. For example, the third week’s
               chained Fisher Ideal index number of 102.0 is the square root of the
               product of 103.2 (chained Laspeyres) and 100.9 (chained Paasche).


               The index value derived from the Laspeyres index formula attained the
Observations   highest value of the three index values by the third week, supporting
               economists’ views that it provides an upper bound for estimating the cost
               of living. As shown in table IV.2, the Paasche index value was the lowest in
               the third week in comparison with the first week. Assuming that the
               shopper was equally satisfied with either fruit selection, the value as
               derived from the Paasche suggests that it provides a lower bound for
               estimating the cost of living.

               The superlative index values, as represented in our hypothetical
               illustration by the Fisher Ideal index, indicate that by using the geometric
               mean of the corresponding Laspeyres and Paasche indexes, one obtains an
               index value that resides between them.




               Page 55                       GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix V

Comments From the Bureau of Labor
Statistics




             Page 56    GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix V
Comments From the Bureau of Labor
Statistics




Page 57                        GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix V
Comments From the Bureau of Labor
Statistics




Page 58                        GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix VI

Comments From the Council of Economic
Advisers

Note: GAO comments
supplementing those in the
report text appear at the
end of this appendix.




See comment 1.




See p. 28.
See p. 1.




See comment 2.




Now on pp. 22-24.
See p. 1.




                             Page 59   GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                 Appendix VI
                 Comments From the Council of Economic
                 Advisers




See comment 3.




See p. 28.




                 Page 60                         GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
               Appendix VI
               Comments From the Council of Economic
               Advisers




               The following are GAO’s comments on the Council of Economic Advisers’
               (CEA) letter dated August 8, 1997.


               1.To prevent any misunderstanding, we want to clarify that we are
GAO Comments   recommending that the expenditure weights be updated more frequently
               than every 10 years or so. We are not recommending a specific time
               interval (i.e., 5 years ) as CEA suggested.

               2.We agree that more frequent updating does not fix the bias arising from
               the fixed market basket structure of the CPI. However, the broad array of
               economic literature that we reviewed supported the proposition that the
               CPI would be improved by more frequent updating of expenditure weights.
               But economic theory is not available to guide the choice of a specific
               update interval for the CPI. We have added a statement explaining that we
               did not identify any theoretical guidance on how often expenditure
               weights should be updated.

               3.In our draft report, we did not acknowledge that we disagree with the
               conclusion of the Greenlees paper because, for the purposes of this report,
               BLS indicated that there was a relationship between market basket age and
               measured inflation. BLS estimated that the growth in the CPI could be
               reduced annually anywhere from 0 to 0.2 percentage point if the
               expenditure weights were updated on a 5-year basis. In reaching this
               position, BLS took into account the information and conclusions drawn in
               the Greenlees paper.




               Page 61                         GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix VII

Comments From the Office of Management
and Budget

Note: GAO comments
supplementing those in the
report text appear at the
end of this appendix.




See pp. 27 and 28.




See comment 1.




                             Page 62   GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                 Appendix VII
                 Comments From the Office of Management
                 and Budget




See comment 2.




See comment 3.




                 Page 63                        GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
               Appendix VII
               Comments From the Office of Management
               and Budget




               The following are GAO’s comments on the Office of Management and
               Budget’s (OMB) letter dated August 12, 1997.


               1.As suggested by OMB, we obtained the views of Bureau of Economic
GAO Comments   Analysis (BEA) officials on the potential consequences for BEA’s work if the
               CPI expenditure weights were updated more often. The BEA Director stated
               that the content of Personal Consumption Expenditures (PCE) and the
               sources of information used to construct it make the potential effect
               unclear. BEA primarily uses CPI data to adjust the dollar amounts of many
               items in the PCE. The Director said that the goods and services in the PCE
               do not all correspond to those in the CPI on a one-to-one basis1 and that, in
               adjusting items in the PCE, BEA uses many different price indexes in
               addition to the CPI.

               The BEA Director said he was more concerned with the potential effect of
               an experimental CPI that BLS began publishing in April 1997. To address
               lower-level substitution bias, BLS is using a different formula on an
               experimental basis to aggregate the 94,000 items for which prices are
               collected each month into the 206 item strata. (See footnote 14 in the
               background section of this report.) This experimental index does not
               involve the subject of this report—weighting the 206 item strata for
               aggregation. The Director said that if and when BLS introduces this change
               on a permanent basis, changing the aggregation of items at the lower level
               is more likely to affect BEA’s work than changing the frequency of
               expenditure weights of the CPI.

               2.We agree with the clarifications that OMB suggested and have made them
               throughout the report. The changes make clear that the 0.1 percentage
               point and the 0.2 percentage point are each an annual reduction in the
               growth of the CPI, and that the estimates of the effect of such reductions
               on the federal budget are cumulative totals.

               3.The dollar estimates that CBO provided to us and that were included in
               our draft report factored in the rounding rules for income tax brackets. As
               suggested by OMB, we discussed with CBO whether it should provide
               additional estimates without these rounding rules. In essence, the CBO staff
               with whom we spoke said that setting aside the rounding rules would not
               produce estimates that differed much from the estimates in the draft.

               1
                We described the differences in a 1996 letter to Senator Daniel Patrick Moynihan. See Alternative
               Poverty Measures (GAO/GGD-96-183R, Sept. 10, 1996). See enclosure I to the letter that compares the
               source data used to construct the PCE and the CEX, which is used to develop the expenditure weights
               in the CPI.



               Page 64                               GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix VII
Comments From the Office of Management
and Budget




According to the staff, the rounding rules would affect which year
departed from the trend over the number of years studied, but the
cumulative dollars over those years would not change. For that reason, we
did not ask CBO to produce additional estimates; therefore, the estimates
that appeared in the draft report also appear in this report.




Page 65                        GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix VIII

Comments From the Board of Governors of
the Federal Reserve System

Note: GAO comments
supplementing those in the
report text appear at the
end of this appendix.




See p. 27.




                             Page 66   GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
                     Appendix VIII
                     Comments From the Board of Governors of
                     the Federal Reserve System




Now pp. 12 and 13.




See pp. 12 and 13.



See p. 28.




See comment 1.




See comment 2.




                     Page 67                         GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
               Appendix VIII
               Comments From the Board of Governors of
               the Federal Reserve System




               The following are GAO’s comments on the Federal Reserve’s letter dated
               July 31, 1997.


               1.We agree with the Federal Reserve’s comment that the recent paper by
GAO Comments   Shapiro and Wilcox1 supported the conclusion in the Greenlees paper that
               more frequent updating of expenditure weights will not reduce upper-level
               substitution bias. However, there was a certain data limitation identified in
               the Greenlees paper, which is probably applicable to the Shapiro and
               Wilcox paper. This data limitation raised questions in our minds about the
               conclusions Greenlees and Shapiro and Wilcox drew from their
               comparisons derived from these data. Each paper compared a price index
               that was based on 1982 through 1984 expenditure data with a price index
               that was based on 1986 and later expenditure data. We believe that what
               Greenlees identified as a potential explanation for the contrasts between
               the two price indexes—the incorporation of the 1980 census-based
               geographic samples into the CEX in 1986—could have also affected the
               Shapiro and Wilcox results, which was a factor their paper did not
               consider. We should also point out as well that BLS, for purposes of this
               report, estimated that the growth in the CPI could be reduced from 0 to
               0.2 percentage point per year with a 5-year update of expenditure weights.
               In addition, Shapiro and Wilcox indicated that the trends they found in
               their research study were not the trends they expected. Referring to these
               as “elusive empirical puzzles,” they called for additional research in this
               area of bias in the CPI.

               2.We talked with BEA’s Director and its Chief Statistician about the Federal
               Reserve’s comment concerning the revision history of its chained price
               index. We specifically asked about BEA’s use of the “Laspeyres tail” in its
               chained price index and if BEA had any estimates that are based on its
               experience with this methodology that would indicate how much routine
               updating of expenditure weights might matter. (Since the index with the
               Laspeyres tail used fixed weights and the revised chained price index used
               a Fisher Ideal formula, which is more reflective of current consumer
               spending, the amount of difference between the two price indexes could
               indicate the level of effect that would occur with a more frequent updating
               of expenditure weights.)

               The Chief Statistician said that he had looked at the difference between
               the two price indexes and that using the Laspeyres tail appeared to have a

               1
                Matthew D. Shapiro and David W. Wilcox. “Alternative Strategies for Aggregating Prices in the CPI,”
               February 10, 1997.



               Page 68                                GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix VIII
Comments From the Board of Governors of
the Federal Reserve System




very small effect, but he had not calculated the size of its effect apart from
other factors that were also changed when the index was revised. He
believed these other factors probably had made a greater contribution to
the difference between the two indexes than changing the price index
formula.




Page 69                         GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix IX

Major Contributors to This Report


               Kathleen K. Scholl, Senior Economist
               Anthony Assia
               Richard Krashevski
               Loren Yager




               Page 70                    GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Appendix IX
Major Contributors to This Report




Page 71                             GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Related GAO Products


              Consumer Price Index: Cost-of-Living Concepts and the Housing and
              Medical Care Components (GAO/GGD-96-166, Aug. 26, 1996).

              Economic Statistics: Status Report on the Initiative to Improve Economic
              Statistics (GAO/GGD-95-98, July 7, 1995).

              Economic Statistics: Measurement Problems Can Affect the Budget and
              Economic Policymaking (GAO/GGD-95-99, May 2, 1995).

              Prescription Drug Prices: Official Index Overstates Producer Price
              Inflation (GAO/HEHS-95-90, Apr. 28, 1995).

              Developing a Consumer Price Index for the Elderly (GAO/T-GGD-87-22,
              June 29, 1987).

              Stabilizing Social Security—Which Wage Measure Would Best Align
              Benefit Increases With Revenue Increases? (GAO/IMTEC-85-13, Aug. 27, 1985).

              Funds Needed to Develop CPI Quality Control System (GAO/GGD-83-32, Apr. 1,
              1983).

              A CPI for Retirees Is Not Needed Now But Could Be in the Future
              (GAO/GGD-82-41, June 1, 1982).

              A Consumer Price Index for Retirees and Alternatives for Controlling
              Indexing (Testimony, Apr. 20, 1982).

              Measurement of Homeownership Costs in the Consumer Price Index
              Should Be Changed (GAO/PAD-81-12, Apr. 16, 1981).

              Alternatives for Modifying the Indexation of Federal Programs
              (Testimony, Mar. 10, 1981).




(410072)      Page 72                      GAO/GGD/OCE-98-2 Market Basket Expenditure Weights
Ordering Information

The first copy of each GAO report and testimony is free.
Additional copies are $2 each. Orders should be sent to the
following address, accompanied by a check or money order
made out to the Superintendent of Documents, when
necessary. VISA and MasterCard credit cards are accepted, also.
Orders for 100 or more copies to be mailed to a single address
are discounted 25 percent.

Orders by mail:

U.S. General Accounting Office
P.O. Box 37050
Washington, DC 20013

or visit:

Room 1100
700 4th St. NW (corner of 4th and G Sts. NW)
U.S. General Accounting Office
Washington, DC

Orders may also be placed by calling (202) 512-6000
or by using fax number (202) 512-6061, or TDD (202) 512-2537.

Each day, GAO issues a list of newly available reports and
testimony. To receive facsimile copies of the daily list or any
list from the past 30 days, please call (202) 512-6000 using a
touchtone phone. A recorded menu will provide information on
how to obtain these lists.

For information on how to access GAO reports on the INTERNET,
send an e-mail message with "info" in the body to:

info@www.gao.gov

or visit GAO’s World Wide Web Home Page at:

http://www.gao.gov




PRINTED ON    RECYCLED PAPER
United States                       Bulk Rate
General Accounting Office      Postage & Fees Paid
Washington, D.C. 20548-0001           GAO
                                 Permit No. G100
Official Business
Penalty for Private Use $300

Address Correction Requested