27 Pages
Gain access to the library to view online
Learn more

L'indice des prix à la consommation surestime-t-il l'inflation ? (version anglaise)


Gain access to the library to view online
Learn more
27 Pages


Les débats autour des problèmes de mesure de l'inflation ne sont pas nouveaux. Ils sont revenus récemment dans l'actualité avec la publication, par une commission du Sénat américain présidée par M.J. Boskin, d'un rapport affirmant que l'indice des prix à la consommation (IPC) américain surestimait l'inflation de 1,1 % par an. Le présent article montre que le biais qui pourrait affecter l'indice des prix en France serait d'un ordre de grandeur très nettement inférieur. Résumer l'évolution d'une multitude de prix en un seul chiffre est forcément difficile. Même dans le cas théorique le plus favorable où l'on considère un consommateur unique face à un choix budgétaire, les problèmes de traitement des substitutions entre produits existants sont importants. On sait néanmoins assez bien caractériser les propriétés des diverses approches envisageables et les procédures statistiques utilisées en France mettent notre indice largement à l'abri des critiques sur ce point. L'apparition de nouveaux produits (entendus au sens large : produits réellement nouveaux sur le marché, produits déjà vendus ailleurs mais apparaissant dans un nouveau point de vente, en remplacement ou pas d'un produit ancien) soulève des difficultés non encore complètement résolues que ce soit aux États-Unis, en France ou dans les autres pays. La commission Boskin a affirmé qu'il existerait, du fait des nouveaux produits, une surestimation de 0,6 % par an dans l'IPC des États-Unis. Cette affirmation repose sur des estimations fragiles et probablement exagérées. Cette conclusion rejoint l'avis exprimé par un certain nombre d'experts américains.



Published by
Reads 96
Language English


Does the French Consumer
Price Index Overstate Inflation?*
François The debate over the issue of measuring inflation has recently been revived
Lequiller** by the publication of a report by an Advisory Commission to the U.S. Senate.
The Commission, chaired by Michael J. Boskin, found that the U.S. consumer
price index (CPI) overestimates inflation by 1.1 percentage points a year.
This article shows that the potential bias in the French CPI is on a far lower
order of magnitude.
Clearly, it is hard to summarize the changes in a multitude of prices with
a single figure. Even in the best case theoretical scenario—a single consumer
faced with a spending decision—the treatment of substitutions between existing
products raises important problems. Nevertheless, it is possible to provide a fairly
accurate description of the various possible alternatives, and the statistical
** At the time of writing,
François Lequiller was procedures used in France largely shield the country’s index from criticism
deputy head of INSEE’s on this point.
Department of Consumer
Prices, Income, and
The introduction of new products creates serious difficulties that have not beenLiving Conditions, which
is in charge of the entirely resolved in the United States, in France, or elsewhere ("new products" is
French consumer price
used here in the broad sense to denote (1) genuinely new products on the marketindex.
and (2) products already sold elsewhere but introduced in a new sales outlet,
The author wishes to
replacing existing products or not). The Boskin Commission estimates the upwardthank all those who
have taken the time to bias in the U.S. CPI due to new products at 0.6 percentage points per year.
improve this article. He
is especially grateful to The Commission’s claim rests on fragile and probably exaggerated estimates.
the unnamed referee Our conclusion converges with the opinion of several U.S. statisticians.
chosen by the editor of
Économie et Statistique
(where the original French
version is being published),
but also to his successor,
T. Lacroix, and to L. Viglino, I. Introduction report in December 1996, in which it argued
D. Temam, M. Glaude,
that the U.S. consumer price index (CPI) would
and G. Laroque. Special
be overestimated by 1.1 percentage points perthanks to J. Mandelbaum
for a magnificent translation. purred by several years’ debate over a year after 1996 and had been overestimated
This paper is available S possible overestimation of U.S. inflation, by 1.3 percentage points a year before 1996.
in the INSEE Studies
the U.S. Senate appointed a commission of In other words, the Boskin Commissionseries on INSEE’s web
site (www.insee.fr). economists chaired by Professor Michael J. claimed that "true" inflation at the
Boskin of Stanford University to examine thehousehold consumption stage would be 1.1
Names and dates issue. The commission submitted its final percentage points lower than the figure
shown in brackets refer
the reader to the
bibliography at the end
of the article * Originally published as "l’indice des prix à la consommation surestime t il l’inflation?", Économie et Statistique , No 303, 1997 3.
INSEE STUDIES No 2, September 1997 1published by the Bureau of Labor Statistics principles of French CPI determination meet
(BLS) in the years to come. For example, if the the highest international standards.
U.S. CPI rises 3.0% in 1998, the true rise,
according to the Commission, should be
reckoned at 1.9%.
II. Substitution effects
As the upward biases would add up over theand new products
years, the combined total would automatically
11. This extrapolation reach 11.5 percentage points after 10 years . In
assumes a constant the United States, the CPI is used to index s may easily be imagined, however, it is no
bias over time–an
welfare benefits and income tax brackets. AnAeasy task to determine an indicator that isassumption that has
not been fully explored. upward bias on either item would swell the supposed to summarize in a single figure the
It also supposes that federal deficit, as indexed benefits would be changes in prices of the billions of transactions
BLS will not change
exaggerated and tax revenues would be involved in household purchases. The processits methods in the
future. diminished by an over rapid rise in the raises many problems, such as the coverage,
income tax brackets. The Commission reckons representativeness, and accuracy of the index
that this mechanism alone would add abou(tsampling), and the recording of actual
$1 trillion to the public debt by 2008. Amid the transaction prices (followed by that of
present raging debates over the federal deficit,unadvertised discounts: see box 2). This article
the Commission did not hesitate to describe the addresses two very specific issues: (1) the
CPI bias as the "fourth largest federal program, coverage of households’ purchasing behavior
after social security, health care and defense".(substitution effects between existing
products), and (2) the introduction of new
Although the subject has long been discussed in products (and/or quality changes).
Europe and the U.S., the controversy over the
Boskin Commission findings offers an Both issues have long been at the heart of the
opportunity to review the French situation onCPI debate. Many economists, including the
22. Appendix 1 tabulates this issue. members of the Boskin Commission, stress that
the results of other the methods for calculating price indexes are
studies similar to the
The French CPI is unquestionably one of thebased on the hypothesis of fixed consumptionBoskin Commission’s,
for the U.S. and other most closely watched of INSEE’s economic structures and on equilibrium assumptions. As
countries. 3indicators . It serves as the main gauge of a result, it is claimed, the methods are not good
3. INSEE, the French
inflationary pressures in monetary and fiscal at describing an ever changing economy madeNational Institute of
Statistics and Economic policy, and, as such, is directly involved in thof imbalances, nee w products, and new agents.
Studies, is the French assessment of one of the convergence criteriaIn particular, as we shall see in §III, the
counterpart to BLS for
defined by the Maastricht Treaty. The detailedbalanced market hypothesis plays a central rolethe calculation of the
Consumer Price Index. price indexes that compose it are used as in the treatment of new products in the price
household consumption "deflators" in the index. This hypothesis, it is argued, does not
national accounts. As household consumption allow for possible temporary imbalances, such
represents 60% of GDP, the CPI is crucial to as a "price wthe ar" triggered by the introduction of
determination of the economic growth rate. Innew products, or the emergence of new, more
France, the CPI is used less directly and efficient producers.
systematically than in the U.S. for determining
social benefits and tax brackets. However, it is
the direct indexing instrument for the minimum
wage (called SMIC), pensions, and family 4. The official procedures automaticfor CPI indexing apply to the
minimum wage, retirement and disability pensions, and familybenefits. It is also the indirect instrument
benefits. The law no. 93 936 of July 22, 1993, provided that, for a
for indexing wages, tax brackets, selected five year period from January 1, 1994, retirement and disability
social benefits, and many types of private pensions would first be indexed on the forecast change in
4 consumer prices, then adjusted to the actual change. The law no.contracts .
94 629 of July 27, 1994, defined the same mechanism for
indexing family benefits, with effect through December 31, 1999.
For all these reasons, the determination of the In the sphere of private contracts, alimony settlements are among
the most commonly indexed to the CPI. It should also be notedCPI—in France and the U.S.—rests on the
that indexing in France is regulated by the Neiertz Law, which bans
strict application of standardized procedures, linkages to indexes that include tobacco. Note: this article does
and INSEE, like BLS, devotes substantial not discuss the actual principle of indexing, which is the subject
5 of debate both in France and the U.S. (see, especially, Grilichesresources to the task (box 1) . The harmonization
of European price indexes, in which France is5. The CPI’s suitability to its various applications is discussed by
heavily involved, has confirmed that the basic Glaude 1997.
2 INSEE STUDIES No 2, September 1997The novelty in the Boskin Commission the Boskin Commission did. Section IVIn , we
approach is that, for the first time, an officialspecifically discuss the issue of outlet
commission has ventured to produce a substitution. In Section V, we examine the
quantitative estimate of the alleged upward quantification of the potential upward bias
bias. Like our BLS colleagues, we believe the in the French CPI compared with the U.S.
6estimate is questionable. We shall see how index. Although we refuse to produce a figure
statisticians tackle these issues. for France to be set alongside the Boskin
Commission’s 1.1 percentage points for the
Section II examines issues relating to productU.S., our conclusion is that, if an
substitution. We look at the weightings and overestimation exists in France, it is probably
mathematical formulas used to offset the far weaker.
substitution bias. Our conclusion is that the
statistical methods used in France differ from
those of the U.S., and that the French methods 6. This adjective is not intended to be polemical. We must stress
that the Boskin Commission managed to maintain a highlargely shield the French CPI from the kind of
standard of discussion in its report, notably avoiding all polemic.criticism voiced by the Boskin Commission.
We should also salute the contribution of BLS (which has been
Section III addresses the issue of new extremely forthcoming despite the difficult context) to improving
public knowledge of CPI statistical methods. Indeed, many of theproducts. We describe the statistical methods
Boskin Commission calculations rely on BLS studies.used, without hiding their limitations or the
Consequently, when we refer to "Boskin Commission figures" for
need for improvements. We show, however, that the sake of brevity, the reader should understand that those
figures are often based on studies by BLS itself, even if the contexterrors can occur in both directions and that no
is, at times, a questionable extrapolation. One of the mainconvincing studies have yet been produced that
controversies over the Commission findings centers on the
make it possible to identify and quantify an difficulty of avoiding double counting when assessing the various
biases.upward bias—although that is precisely what
Box 1
The French consumer price index is a complex introduction in 1998, along with prices of selected
statistical tool based on the strict application of domestic and legal services, fees for government
standardized procedures. More than 160,000 price services, and second hand motor vehicles. As regards
quotations are collected each month in about a non market services, INSEE is examining ways to
hundred urban areas representative of towns and include hospitals, clinics, and welfare expenditures.
cities of more than 2,000 inhabitants. All types of
outlets are covered, including hypermarkets, The weightings of the index groupings and their
supermarkets, hard discounters, superettes, specialized component products are reviewed each year. The
retailers, "mom and pop stores," street markets, etc. index uses an annually chained Laspeyres formula.
Thirty thousand outlets are visited each month by Every year, the statisicians in charge of the broadt
the CPI’s 170 full time or part time price collectors. index sectors (food, durable goods, clothing,
One hundred and thirty other INSEE personnel, services, etc.), assisted by field workers, conduct a
including several dozen statisticians, are responsible thorough examination of the characterisics of thet
for computing the index. The Institute’s price 1,000 detailed families of products ( variétés or
collectors are specially trained for their assignment. sub-items) that make up the index. These families
Quality control procedures are applied at every step are modified where necessary to preserve CPI
of the process. The initial results for the representativeness. One example of such changes
month—known as the "provisional" index—are concerns the meat sub items whose consumption
published toward the 10th of the following month, has fallen as a result of the mad cow crisis. New
i.e., sometimes within seven working days of the sub items are introduced each year as additions
close of price collection. The index covers the entire or substitutes for existing sub items that have
range of goods and services included in the definition become less representative. A recent example was
of household consumption: food, beverages, tobacco; the introduction of computer floppy disks. The current
clothing and footwear; housing, heating, and index is known as the 1990 base index (1990 = 100).
electricity; furnishings,ehold equipment, and This base is France’s sixth generation CPI since the
routine maintenance of the house; health; start of the century. Methods and coverage have
transportation and communications; recreation, improved with each generation. Estimations of
entertainment, culture, and education; personal care, sampling errors on the latest generation of the
restaurants, hotels, and other services to households. French CPI show that the twelve month change
The CPI thus covers about 90% of households’ fluctuates in a very narrow range of +0.1 to 0.1
market consumption. Insurance is scheduled for percentage points (Ardilly and Guglielmetti 1993).
INSEE STUDIES No 2, September 1997 3one would simply pick one product andII.1. Product substitution:
measure its price change. However, we knowthe issue of weightings
that the average rise (or fall) in prices masks
The problem raised by product substitution fordifferent variations from one product to
CPI statisticians is that households change their another. In other words, relative prices change
consumption basket at the same tim as pricee s at the same time as the general upward or
vary. As we shall see, a CPI whose weightingsdownward movement in prices. To calculate
rested on obsolete information may tend to our consumer’s price index, we therefore need
overstate inflation. to track all the products he or she consumes, or
at least a representative sample of those
Let us begin by examining the case of asingle products; we then have to determine the
consumer and try to calculate his or her pricavere age of the changes in prices for the
index between a base period and the currentproducts in the sample. This immediately raises
period. We define the consumer’s price indexone question: what weighting should be
between the two periods as the growth rate of assigned to each of the products included in the
his or her budget outlays that enables the average? The only possible weighting, of
consumer to maintain the same utility level a course, is based on the quantity consumed. t The
current prices in the current period as in the price index is therefore the result of an average
base period. This is the concept known as the of price changes weighted by values of
77. For some economists, preservation of "purchasing power" . quantities consumed. This definition is
the term "purchasing
inadequate, though, as there are many ways to
power" implies, by
If all product prices varied proportionally, suchcompute the average. In particular, prices areconstruction, that the
price index uses the a price index would be very easy to calculate: not the only variables to have changed between
quantities consumed in
the base period. Here,
however, we use the
Box 2term in a broader
Criticism of the French CPI, especially in recent the phenomenon. This would require access to
years, has often focused on the following issue: the retailers’ bills—which are regarded by retailers as
CPI, it is argued, does not take into account the highly sensitive data. However, using the CPI
"unadvertised" and "private" discounts agreed upon weighting structure and some conventional
1between buyers and sellers. assumptions, we can show that the impact of
the bias on the overall index is probably rather
To clarify the discussion, it should be pointed out that limited. First, the types of product to which discounts
the French CPI is now tracking all advertised are most commonly applied—automobiles, clothing
discounts, special offers, and promotions on all and footwear, household appliances, and
products. INSEE price collectors are instructed to furnishings—account for only 14% of the total index.
2record prices net of all such discounts. Consequently, if we make the extreme assumption
that 10% of all purchasers of these products are
It is true, however, that INSEE price collectors can offered an additional unadvertised discount of 2% on
only monitor advertised discounts and not actual the previous year’s price, this would have an impact
transactions if the listed prices are changed through of 0.10 x 0.02 x 0.14 = 0.03 points on the annual
direct bargaining between buyers and sellers. INSEE average of the overall index. This upward bias would
can hardly require its field workers to simulate a be offset by a negative bias if the suppliers’ position
purchase every month, as this would violate basic improved thanks to a consumption rebound.
ethical principles. In all likelihood, though, bargaining
over advertised prices became more common in
France in 1993 owing to the consumption slowdown.
The French CPI may therefore have overestimated 1. Discount coupons, which have become very popular in recent
inflation during a specific period by failing to take intoyears, are another variant of this practice.
account the greater incidence of unadvertised 2. Except for discounts applied for a few hours only ("surprise"
3discounts . But if we accept that inflation may have special offers announced in stores), clearance sales, and
consumer credit rebates. Our position on clearance sale pricesbeen overstated during consumption slumps, we
is that–by definition–the old product cannot be replaced whenshould also admit the converse assumption, namely,
its price can no longer be observed. As for consumer creditthat inflation may have been understated during
rebates, CPI coverage does not extend to interest rate
consumption recoveries, when unadvertised
discounts tend to become less common. In other
3. In this connection, it should be remembered that the factor
words, this discount bias is not a structural bias, like that would influence the price index change is the variation in
those described in this article, but a cyclical bias. unadvertised discounts and not their absolute level. If their level
did not vary, the price index change would be identical
Estimating the discount bias is a particularly delicate regardless of whether or not the unadvertised discounts were
included.task. One major obstacle is the high cost of tracking
4 INSEE STUDIES No 2, September 1997the base period and the current period: the for example, statisticians work at a fairly high
quantities of products consumedh ave changed level of aggregation for the annual weighting
as well. Since that is so, what quantities should updates based on data from year n 2. The other,
we use to establish the price index weightings? more detailed levels are not treated identically.
The quantities consumed in the base period? In this section, we describe the three stages of
Those of the current period? Or an average of the aggregation process used to obtain the
the quantities consumed in the two periods?overall index from the individual price
Index theory (appendix II, §1) does not offer a quotations. The first stage, hereafter referred to
single, final answer—not even for a single as the "lower level" ( niveau détaillé), is the
consumer and, still less so, for multiple calculation stage that yields the highly detailed
consumers. The theory does tell us, however,indexes (often called "micro indexes") for
that—under certain hypotheses—one of the highly disaggregated product categories
9best proxies of an ideal index is a Fisher inde(x:variétés or "sub items") for a specific
this is an average of an index using base period geographic region ( agglomération or "urban
10weightings (called a Laspeyres index) and anarea") . The micro-indexes are obtained from
index using current period weightings (known price quotations collected from sales outlets in
as a Paasche index). The theory also shows that the urban areas examined. The second stage,
the Laspeyres index usually overestimates thehereafter designated as the "intermediary
11Fisher index, while the Paasche index usuallylevel" (niveau intermédiaire), yields the
underestimates it. The mechanism behind thisindexes for product groupings from the
result is simple: the Laspeyres index micro indexes. In third and final stage, called
overweights the products whose prices register the "upper level" (niveau agrégé), the overall
the steepest increases, whereas, by logic, theprice index is calculated from the grouping
share of these products in consumer indexes. In each stage, a Laspeyres formula is
expenditures will decline once consumers used (or was used, as we shall see). Upper-level
accept a degree of substitution between weightings are taken from the national
12products while maintaining constant utility. accounts . Intermediate level weightings are
usually taken from public or private surveys of
In practice, however, a Fisher index is household expenditures, or from other sources
impossible to calculate, at least during the year such as private panel surveys of distributors,
and within the short deadlines required for an production and import statistics, etc. At the
index such as the CPI. The reason is that the lower level, the lack of information has led to
Fisher index cannot be established without thethe convention of assigning an identical, fixed
13current period weightings, which take a longweight to each product . At each of these
time to determine. To calculate the Fisher indestages, x an overestimation—described hereafter
for 1997 against a 1990 base, for example, we as "substitution bias"—may occur. We shall
would need to know, in particular, the therefore examine each level separately to
quantities consumed annually in 1997. assess the possible size of that bias.
Naturally, those quantities cannot be
determined during 1997. That is why all
countries calculate the CPI as a Laspeyres
8. The French termposte , translated here as "grouping," denotes
index, that is, using fixed weightings derived the broad product c egorat ies that constitute the first level oenchf Fr
from the base period. The "age" of the base yearCPI publication (the U.S. equivalents of "groupings" are the item
strata). Examples of groupings include "fresh fruits,"used to estimate the weightings varies from one
"automobiles," and "men’s hairdr essers." There are 265
country to another. In France, the weightingsgroupings in the current index, known as the "base 1990" index.
used to obtain the overall index from The "base 1980" index had 295 groupings.
9. Equivalent to the "entry level items" in the U.S. CPI."grouping" indexes are updated every year
10. Roughly equivalent to the "primary sampling unit" (PSU) of8from recent data. In the U.S., the weighting the U.S. CPI.
base is much older. 11. The Boskin Commission does not distinguish an intermediary
level, but amalgamates it into the "lower level."
12. In the French national accounts, the public surveys on
household expenditures and other statistical sources on
II.2. A three level decomposition consumption are collated in sufficient detail to serve as a basis
for CPI weightings. In the U.S. and most other European
countries, the weightings are derived simply from public surveys
The situation is not quite so simple, however,on household expenditures.
since a price index as complex as the CPI is 13.the Thanks to the computerization of the wholesaling/retailing
trade and the standardization of bar codes, explicit weightings atresult of successive aggregations of indexes;
this level of detail–which seemed like statistical science fiction
each aggregation level has its independent only a few years ago–may become reality in the foreseeable
weightings, with different "ages." In France, future.
INSEE STUDIES No 2, September 1997 5market-research firms would be desirable. TheII.3. No overestimation at the upper level
French index may thus be subject to anin the French index
upward bias due to the obsolescence of some
The methods used to calculate the French indeintermediate weightingsx . Unfortunately, there
protect it from upward bias at tupper lehe ve.l is no way to measure that bias directly. Some
Since the early 1970s, the weightings of CPIindications, however, can be used to provide a
"groupings" are reviewed annually against very rough range. If a substitution bias exists at
national accounting data from year n 2: in the intermediate level, it probably exceeds the
other words, the 1997 monthly price index is upper-level bias, which, as we saw, would be
based on 1995 weightings. The "instantaneous"equal to 0.11 percentage points per year in
French index—that is, the monthly changes inFrance if we did not calculate a chained
the current year—is therefore based on very Laspeyres index. This assumption is based on
recent upper-level weightings. The "long term"the notion that, as the level of index detail
French index, which tracks price changes overincreases, so does the substitutability between
several years, is based on the chaining of thes the e index components. On the other hand, many
instantaneous indexes. The French index is thus sub item weightings are regularly updated,
14referred to as a "chained" Laspeyres inde . x even if not annually, and this would tend to
reduce the potential upward bias. On balance,
In sum, the age of the French CPI groupinga mild upward bias in the range of 0.05 0.10
weightings never exceeds two years. We can percentage points a year cannot be ruled out.
therefore conclude that the upper-level
substitution bias is, in practice, negligible or
15 II.5. The geometric mean eliminates non existent in France . In the U.S., by
comparison, the Boskin Commission estimates the risk of upward bias at the lower
the upper-level substitution bias at 0.15 level in France
percentage points per year based on the present
weighting structure, which dates from For its lower-level indexes, France previously
1982 84—that is, well over a decade ago. The used a method based on implicitly fixed
Commission relied on simulations conducted weightings. As we shall see, the current method
in the U.S.: these showed that, in present is based on a formula that allows for a degree of
conditions, a Laspeyres index whose substitutability and therefore removes the risk
weightings are approximately ten years old of upward bias. For the past three years
increases by about 0.1 0.3 percentage points a European statisticians have focused—among
year more than the same index calculated using other methodological issues (box 3)—on the
a Tornqvist formula, a close equivalent offormulas used by EU member States to
the Fisher formula (Aizcorbe and Jackman calculate "micro indexes." It emerged that the
1993). In France, we compared the official use of geometric means as micro indexes had
index with an unchained index for the periodtwo advantages over the conventional
1980 90. As expected, the unchained index waars ithmetic mean formulas (appendix II, §2):
found to riseo n average by 0.11 percentage (1) the geometric mean allows for possible
points more than the official index (Viglino and
16Montiel 1995) .
II.4. A mild residual upward bias
14. In Europe, two countries use a nearly identical system: Britain
at the intermediate level and Sweden. Germany updates its weightings every five years.
The other European countries update somewhere in between
these two frequencies.
By contrast with the upper level, the weightings
15. The chaining method is not necessarily a universal cure. In
of the French intermediate level index—i.e., particular, it can lead to positive biases in case of sharp swings
in relative prices, as shown in appendix II, §2. However, most ofthe weightings of the "sub items" in the
these swings do not occur at a level as aggregated as the product
"groupings"—are not updated each year. Of
groupings. Indeed, all the numerical simulations for the 1980 95
course, as far as the CPI budget allows, INSEE period show that chaining makes the Laspeyres index more
comparable to a Fisher index than before.gathers and uses all the detailed information
16. The figure of 0.11 percentage points is an average. The
published in wholesale/retail trade journals or
difference is not necessarily identical in all years and many even
obtained from trade associations and be negative in some years. Another illustration of this type of gap,
but in the other direction, is offered by the consistent differencemarket research firms. Admittedly, there is
between the CPI and the Paasche indexes of
room for improvement at this level. For
household consumption implicit deflators in the French national
example, a more routine use of data from accounts (box 5).
6 INSEE STUDIES No 2, September 1997substitutions between products, whereas the II.6. A wider difference between
arithmetic mean assumes fixed weightings; (2) geometric mean and arithmetic mean
the geometric mean, by construction, avoids in the United States
what has been called "formula bias." Formula
bias is a positive bias that occurs when an The size of the correction procured by the
arithmetic mean is used without precautions inintroduction of the geometric mean in France
a chaining operation. The 15 EU member States (0.10 percentage points) is well below the
therefore agreed unanimously to extend the use Boskin Commission’s figure for the U.S. Yet
of the geometric mean. This decision—now the latter was obtained with the same type of
embodied in a European regulation—will notcalculation—i.e., comparing (1) an index
only improve the calculation methods in eachprepared from micro-indexes consisting of
country, but make it much legitimate to geometric means with (2) an index based on
compare the price indexes of the 15. arithmetic means. For the U.S., the difference is
put at 0.50 percentage points, which leads the
France is therefore gradually substituting Boskin Commission to estimate a formula bias
geometric means for those of its micro indexof 0.25 percentage points and a lower-level
formulas that relied on arithmetic means of
17price ratios . The geometric mean is
17. See appendix II, §2. The substitution will be applied to aboutwell suited to French micro indexes, which
12,500 out of 25,000 micro indexes. It began in January 1997 for
cover highly specific categories of products
one half of the micro indexes concerned and will be complete by
exhibiting similar technical characteristics and January 1999. To preserve the time comparability of data series,
France preferred a gradual introduction–which neverthelesswhose prices are collected from the purchasing
complies with the European regulation–to a massive,18area known as "urban area" ( agglomération) .
instantaneous switch. It should also be noted that the other
The geometric mean would not necessarily 12,500 micro indexes are based on a price sums ratio (PSR)
formula. Such micro indexes have been used for years in the CPIwork well with larger product families, for
for so called "homogeneous" sub items and are accepted under
which the substitution elasticities might be
the European regulation. The results of a PSR simulation show
lower. Simulations performed on recent years that, although it can lead to divergences from the geometric mean
for individual sub items, the calculations on several dozenshowed that—all other things being equal—the
sub items converge toward the same figures as the geometric
switch would lower the index by an estimated
mean. PSRs are not vulnerable to formula bias.
average of about 0.10 percentage points per 18. When the geometric mean was introduced in France, French
statisticians analyzed the impact on individual sub items. One ofyear for the period examined. The switch will
the main consequences of the analysis was a redefinition of some
thus gradually eliminate the bulk of lower-level
sub items to achieve greater precision and make the products
substitution bias and formula bias in France. more homogeneous.
Box 3
The CPI has been chosen as the main indicator for micro-indexes, (2) transition to monthly price
the inflation criterion of the Maastricht Treaty. collections and improvement of the treatment of
Member States applying for monetary union must missing values, (3) acceleration and coordination of
show an inflation rate no more than 1.5 points above new product introductions, (4) improvement of
the average of the three best performing countries. weighting updates, (5) improvement of the treatment
The criterion’s precision calls for a harmonization of of quality changes, and (6) harmonization of index
methods to ensure maximum comparability between coverage. In compliance with (6), France has
1country indicators . This harmonization drive has incorporated insurance prices into its HICP in 1997
made great progress since its launch three years (but not yet in its national CPI). Eurostat will inspect
ago. A framework regulation (#2494/95) was adopted the HICPs of each country to ensure compliance
by the Council in October 1995. A first major step with the rules jointly approved by all Member States.
was the publication of an initial version of the Beyond its strictly regulatory aspects, harmonization
Harmonized Index of Consumer Prices (HICP) in has fostered common research programs,
early 1997. Harmonization has both increased the particularly on quality change.
comparability of national indexes and improved each
individually. Far from confining themselves to a
"common methodological denominator," the Fifteen
have sought to incorporate the best method for each 1. See Buchwald and Saglio (1994) for an example of the
non comparability of the French and German non harmonizedfield into the implementing regulations, which cover
CPIs.the following areas: (1) geometric mean and other
INSEE STUDIES No 2, September 1997 723
substitution bias of 0.25 percentage points asIII. New products
well. The difference between the U.S. and
French results can be explained in two ways.
ne of the main difficulties in constructing
(1) The U.S. formula bias was, by construction,Oprice indexes lies in the contradiction
1919. We use the past much greater than the French one . The reason between (1) the fixity-of-products principle,
tense because–in the is that 20% of the component products in 70%which is crucial to the comparison of prices in
U.S. as in France–the
of the entry level items in the U.S. CPI are two different periods, and (2) economic reality,formula bias is a thing of
the past, since BLS systematically replaced at random each year. in which new products are constantly launched
started correcting it in This highly sophisticated procedure eliminatesand obsolete products disappear. The
1996. Rather than
product selection bias and provides an discussion may gain in clarity from aintroducing a geometric
mean, as in France, the automatic 1/5 renewal of products tracked bydistinction between (1) "product replacement,"
Bureau corrected the 20the index . However, the procedure was in which a product tracked by the index is
implicit weightings of
vulnerable to a formula bias similar to the one replaced by a similar product, and (2) newthe chained index.
20. There might be a described in appendix II, §2: whenever the products that have few if any equivalents
product-selection bias initial price of the new product selected at among past products.
if, for example, all price
random was low (because of special offers orcollectors chose the
same brand of milk seasonal factors), the subsequent price rises in
chocolate to represent III.1. Statistical methods for dealingthe U.S. index were automatically overstated.
the "milk chocolate"
Conversely, when the initial price was too high, with product replacemententry level item. The
U.S. procedure avoids the subsequent decreases were automatically
this problem thanks to its understated (Moulton 1996). This error proved Product replacement has a heavy impact on the
probabilistic product-
to be particularly large—an estimated 0.25 CPI. In the French index, 30% of productsselection method.
percentage points—for the U.S. index of freshwhose prices are tracked in conventional
24fruits and vegetables. In the French index, thoeutlets are replaced in any given year. In
problem affected only one third of the CPI absolute terms, that makes more than 30,000
weighting: one notable exception was, product replacements a year. A A statistic,
precisely, fresh products, for which another calculated for the U.S. index but probably
(unchained) formula is used. Nor was the applicable to France, gives an idea of the CPI’s
French index affected by special offers, sinceupward or downward sensitivity to these
new products are never introduced when theytreatments, at least in periods of mild inflation.
are on special offer. All the factors that In 1995, the U.S. CPI excluding housing (and a
accentuated the bias due to the use of arithmetic few other minor items) was recalculated by
means in a chained index in the U.S. were canceling the statistical procedures used by
21therefore absent in France . In 1996, BLS
undertook a specific correction to eliminate the
22bias with effect from 1996 .
21. For the sake of completeness, however, we should note the(2) The method used by the Boskin
recent discovery that the problem did exist in the French indexCommission to estimate the lower-level
at an intermediate level of aggregation (Poinat 1996). The
substitution bias effectively consisted, as in resulting bias proved to be a negligible 0.01 percentage points
or so per year.France, in comparing the official index—all
22. BLS did not, however, revise the index series. The 0.25 pointother things being equal—with the same index
formula bias, therefore, still applies to the years prior to 1996. This
in which arithmetic mean formulas were explains the difference between the Boskin Commission’s
1.3 point bias for the years prior to 1996 and its 1.1 point bias forreplaced by geometric mean formulas. Unlike
the years after 1996.in France, however, the replacement was made
23. For brevity’s sake, we will refer to products"new " in this section
not only at the lower level but also at the as a shorthand for the more accurate phrase "new goods and
services." In fact, most "new products" probably consist of "newintermediate level. The use of geometric means
services."at the intermediate level is not always
24. When stated in terms of price quotations collected during the
advisable, as BLS itself has pointed out year, the figure is obviously smaller, since quotations are
gathered twelve times a year. This banal observation puts into(Moulton 1996). Indeed, the substitution
perspective the issue of product replacement in high inflationelasticity at the intermediate level may, in
situations. N.B.: The figures and percentages for product
some cases, be closer to 0 than to 1. Hencereplacement given in this section are based exclusively on the
price quotations recorded directly in stores by INSEE priceBLS caution about implementing the
collectors. They account for some 90% of total price quotationsBoskin Commission’s overly systematic
used in the index. The sectors excluded from these statistics are
recommendation of the geometric mean those for which price quotations are collected from a central
source (automobiles, utilities, rail and air transportation, etc.).(Abraham 1997).
8 INSEE STUDIES No 2, September 1997BLS in 1995 to eliminate quality change Second, the recording procedures are designed
values. The recalculated index rose 2.5 to enable price collectors to replace the deleted
percentage points more than the original indexproduct with the closest possible substitute, in
(Abraham 1997). order to minimize the size of the quality change
value and hence the potential error in its
estimation. Having defined this framework, let
III.2. Comparing prices us now examine the four methods used to deal
"at constant quality" with quality change:
Whenever a product is replaced, the price of the(1) Chaining (also known as "linking" or
new product must be compared with the price "splicing"): This method is probably the most
of the deleted one. To measure this price widely used in price indexes, because it is
change, the price ratio needs to be adjusted for simple and has some basis in economic fact. It
any difference in quality between the old andis used for 60% of product replacements
25 2625. French CPI new products . For example, if a car model observed in stores in France . That percentage
statisticians call this the without air conditioning is replaced by the would be even higher if we extended the
"treatment of quality
same model with air conditioning, a direct concept of replacement to the periodicchange value" (traitement
de l’effet qualité) . comparison between the prices of the two carreconstruction s of the sample that takes place
is obviously impossible. We will have to every December and to the sectors for which
estimate the "price" of the air conditioning, for price quotations are collected from a central
example from the price in the automaker’s source. In its pure form, the method assumes
catalog when air conditioning was optional; we the old and new products are observed on the
will then have to subtract that price from the market during the same period (as we shallt
price of the new model to obtain the change isee, though, this is not actually the case). Then
price "at constant quality." In this example, theprice change for the old product is used to
procedure seems fairly easy, since the value ofmeasure the change in the overall index
the "air conditioning" option is rather simple to between t 1 (and earlier periods) and t. Once
estimate. It is easy to see, however, that thethe old product is deleted, the change in the
procedure may prove much harder in other price of the new product will serve for the
cases: the concept of "quality" is often elusive,overall index between t and t+1 (and
and its quantitative estimation even more so. subsequent periods). In other words, the price
changes for the new and old products are
Price index statisticians readily admit the "chained"—hence the method’s name. In
importance of this problem. The simple theorypractice, the method implies that the price
of indexes outlined earlier is not very difference between the two products in t is
enlightening on this point, since it assumes, byexactly equal to the quality differenc. Te his
definition, that the same products exist in the identity somehow validates the price difference
base period and the current period. Here, we are recorded in the market, which is assumed to be
27dealing with precisely the opposite case. in equilibrium at the time . To take an
However, the theory’s reference to the principleexample, let us imagine that the washing
of maintaining the consumer’s utility level machine Y is replaced by a new, more efficient
provides a useful guideline for the statistician.model, Y’, sold for FF500 more. The price of Y
When a product is replaced by another, the has been tracked in the index until April 1997.
price variation should be calculated after the That same month, we also collect the price of Y’
two products have been "restated" in terms of in the same store. As the two machines are sold
equal utility. Admittedly, this concept is hard to at the same time in the same market, we assume
define for a single consumer—and all the more that the FF500 difference measures the value
so for millions of consumers. consumers attach to the difference in efficiency.
Of course, it would be wrong to conclude that
there are no appropriate statistical methods. Let
26. It is the only method used for heterogeneous sub items.
us begin by clearing away the most obvious 27. An alternative description of the method may, at first sight,
seem more neutral in regard to the quality change valuation butmisunderstandings. First, the CPI never
is ultimately equivalent. We can say that the price index is built
compares prices of products of objectively
by chaining monthly links that consist solely of products existing
different quality. To return to our earlier in two consecutive months. But it would be misleading to believe
that this description of the method provides a miracle solution toexample, every step will be taken to avoid a
the problem of quality change valuation. The method merely
direct comparison between the air conditioned
assumes that the entire price difference measured at the instant
model and the non air conditioned model. t is due to a quality difference.
INSEE STUDIES No 2, September 1997 9The following month, Y has been deleted, andto apply. Like the Boskin Commission,
the price change for Y’ is introduced into the however, it is worth examining whether—on
index formula. The inflation for Y’ between balance—these methods consistently produce a
April and May is therefore measured by bias in one direction or whether the errors can
comparing the May price of minus FF500Y’ go either way.
with the April price of Y.
(2) "Direct comparison" (or "equivalent III.3. Chaining
replacement"): This method consists in finding
a new product that can be regarded as For the chaining method to work properly, even
"equivalent" to the old product. The entire price when applied correctly, we must assume a
difference between the two products is stable equilibrium for the product market. This
consequently treated as a "pure" price changeis not always the case. In practice, the new
(i.e., the "quality" of the two products is product may capture market share from the old
considered identical). For the so called one—a possible sign that the instantaneous
homogeneous sub items, INSEE price price difference understates the quality
collectors are thus practically obliged, when difference. In our washing machine example,
the product whose price they recorded in thewe may assume that if the old model Y is gone
previous month has been deleted, to find from the stores in the following month, it is
another product whose technical characteristics precisely because the new machine Y’ is sold at
28can be regarded as identical . This method is a price that has eliminated the old one from the
applied to 40% of replacements of market. The theoretical equilibrium price of Y’
store tracked products in the French index. is therefore more than FF500 above that of Y. To
value the quality change at FF500 would
(3) "Deletion": The discontinued old product understate that change and therefore overstate
is not replaced. Subsequent price movements of the price change, since the price would actually
its product family are tracked via the have fallen rather than remained stable. This
"remaining" products. This method is rarely reasoning would hold not only if the price of the
used in the French index. new product was higher than the price of the old
one (as with our washing machine), but also if
(4) "Explicit quality change valuation": the new product price was equal to or even
This method consists in estimating the qualitylower than the old product price. Let us take
change value through direct methods such asanother example: Firm F’, a more efficient (or
option costs, estimations of the cost of quality aggressive) competitor of firm F, sells a more
improvement by producers, and econometric powerful personal computer than that of firm F
methods often referred to as "hedonic at the same price, capturing market share from
regression methods." Properly applied, these F in the process. The chaining method would
methods are the most convincing approach, innot recognize the price of the F’ computer as
statistical terms, for measuring quality change. lower than that of the F computer, despite the
Unfortunately, they are very expensive: for thefact that the F’ computer is more powerful than
moment, statistical agencies—both in the U.S. the identically priced F computer. Again, the
and in France—cannot afford to implement result would be an overstatement of inflation.
them except on a very modest scale. In France, That is indeed what statistical studies of the PC
they are applied mainly to automobiles, owingmarket show. The difference between a
to the grouping’s importance in the CPI and the chaining approach and an approach taking the
high quality of information available in the
industry, such as catalogs listing numerous
features, option prices, and so on. 28. Of the 1,000 sub items of the French CPI, about half are
so called homogeneous sub items; the other half are known as
heterogeneous sub items. Homogeneous sub items areNone of these four methods is perfect. In any
composed of products whose characteristics are defined with
case, it would be a fallacy to believe that theregreat precision and whose price levels are therefore very similar.
are both theoretical answers to all the issuesAs a matter of principle, replacements in homogeneous
sub items are made with the direct comparison method.raised by new products and the practical means
Heterogeneous sub items consist of products whose
to apply those theoretical solutions, even characteristics are defined less precisely. They may therefore
assuming they did exist. It should also be contain more disparate items. Although collection instructions
advise direct comparison replacements in every case, theremembered that the price index relies on
guideline is hard to enforce in practice, and there are many more
decentralized collection, and that replacement chaining based replacements than direct comparison based
procedures must therefore be simple and easyreplacements in heterogeneous sub items.
10 INSEE STUDIES No 2, September 1997