Measurement Errors in Japanese

Working Paper Series

Consumer Price Index

Shigenori Shiratsuka

Working Papers Series

Research Department
(WP-99-2)

Federal Reserve Bank of Chicago

 Measurement Errors in
Japanese Consumer Price Index
Shigenori SHIRATSUKA
Bank of Japan and Federal Reserve Bank of Chicago
E-mail: [email protected]

February 1999

Abstract
In Japan, the Consumer Price Index (CPI) is widely used as a measure of inflation or
the cost of living. The CPI is constructed by using a fixed-weight Laspeyres formula.
This formula is used mainly because of its ease of calculation and comprehension, thus
limiting the total cost of constructing the statistics. However, such simplicity makes
it difficult for the CPI to reflect dynamic changes in economic activity such as changes
in consumers’ behavior between goods in response to relative price fluctuation, the
introduction of new goods, and the disappearance of old goods. As a result,
measurement errors are introduced into the CPI. In this paper, I summarize the
problems pertaining to measurement errors inherent in the Japanese CPI, and provide
some quantitative assessment. Based on currently available information, I place the
point estimate for overall bias in the CPI at about 0.90 percentage point per year.
Although this is the best estimate taking into account all information currently
available, it is true that the estimate was based on various, rather bold assumptions.
In addition, it should be noted that accuracy of the estimate is not necessarily high due
to the lack of existing studies in this field in Japan.

Key Words: Consumer Price Index, Measurement Errors, Cost of Living Index,
Quality Changes, Boskin Report
JEL Classification Code: C43, E31

________________________________________________________________________
This paper is based on the chapters on evaluating the measurement problems in the Japanese CPI of
author’s monograph written in Japanese (Shiratsuka, 1998). The author is grateful to Erwin Diewert,
David Marshall, and seminar participants at the Federal Reserve Bank of Chicago and Federal Reserve
Board for their helpful comments; and to Michio Kitahara, Joel Spenner, and Laura Kutianski for their
extensive assistance to make this paper more readable. The opinions expressed are those of author and
do not necessarily reflect those of the Bank of Japan or the Federal Reserve Bank of Chicago.
 Table of Contents

1. Introduction ............................................................................................ 1
2. Sources of Measurement Errors ........................................................... 2
(1) Limitation of the fixed-weight Laspeyres index formula .....................................2
(2) Substitution effects...............................................................................................4
(3) Effects of quality change ......................................................................................6
(4) The new goods effect............................................................................................8
(5) Technical problems in constructing the CPI statistics........................................10
3. Magnitude of the Measurement Error: A Quantitative Evaluation. 11
(1) The range of quantitative evaluation ..................................................................11
(2) Problems of index formulas ...............................................................................13
(3) Problems in aggregating individual sample prices into item level.....................15
(4) Problems in quality adjustment method .............................................................15
(a) Upward bias in durable goods ......................................................................16
(b) Quality adjustment for services in the CPI...................................................17
(c) Quality change bias in the total ....................................................................18
(5) Effects of the structural change in the retail market...........................................19
(4) The magnitude of measurement errors ...............................................................21
(a) Evaluation of upward bias in total................................................................21
(b) Reservations for the results of the estimates ................................................22
4. Comparison with the Boskin Report .................................................. 23
5. Economic Policy Implications............................................................. 24
(a) Measurement of price stability .....................................................................24
(b) Treatment of asset prices..............................................................................25
(c) The needs for exploring additional methods to gauge the underlying trend of
inflation..........................................................................................................25
(d) Fiscal balance and implication on fiscal policy............................................26
6. Conclusion............................................................................................ 27
(a) Introduction of chained CES index formula.................................................28
(b) Introduction of hedonic approach as quality adjustment method.................29
(c) Review of survey samples ............................................................................29
Appendix 1. Index Formulas ................................................................... 29
Appendix 2. Theoretical Relationship between Laspeyres Price Index
and Cost of Living Index ............................................................................ 31
Appendix 3. Application of the Hedonic Approach to Specification
Changes....................................................................................................... 35
References .................................................................................................. 36
1. Introduction

In this paper, I review issues pertaining to measurement errors inherent in the

Japanese Consumer Price Index (hereafter CPI), and quantitatively evaluate the magnitude
of its upward bias.

The CPI, which is widely used as a measure of inflation or the cost of living in
Japan, is constructed by using a fixed-weight Laspeyres formula. This formula has been
used mainly because of the simplicity of its concept, which aggregates individual price
quotations using weights fixed at the base period, thus limiting the total cost of constructing
the statistics.

At the same time, however, such simplicity makes it difficult for the CPI to reflect
the dynamic nature of economic activity such as changes in consumers’ behavior between
goods in response to relative price fluctuation, the introduction of new goods, and the
disappearance of old goods. As a result, measurement errors are introduced into the CPI.
In particular, quality changes brought in by technological innovation are a major cause of
measurement errors, and the magnitude of such biases is crucial under the current trend of
rapid technological innovation.

Moreover, accuracy of inflation measures becomes a more important problem

when considering whether to go further down from an already low inflation rate.1 In the
1970s, it is perfectly apparent whether it is desirable to cut the inflation rate because
inflation is high and prices are rising in any inflation measures. However, with recent low
and stable inflation rates in major countries, the issue of measurement problems in price
statistics becomes much more important for monetary policymakers. Measurement errors
in price indices are an especially important issue in a country like Japan where there is
controversy as to whether the country is on the verge of deflation.

1
Such concern was explicitly pointed out in the speech of Chairman Greenspan (1996) of the Federal
Reserve Board during the August 1996 Conference held by the Federal Reserve Bank of Kansas City.

1
 This paper is constructed as follows. In Section 2, I examine the sources of
measurement errors in the CPI, from the practical viewpoints of statistics compilation. In
Section 3, I make a quantitative evaluation of measurement errors in the CPI. In these two
sections, I specify four major causes of the upward bias in the CPI: (i) problems in the
index formula, (ii) problems in aggregating individual prices into item levels, (iii)
inappropriateness of the quality adjustment method, and (iv) effects of structural changes in
retail markets. Then, each cause is quantitatively assessed and aggregated to get the point
estimate of the magnitude of bias. In Section 4, I compare my estimate of the Japanese
CPI with that in the US by the “Boskin Report.”2 In Section 5, I discuss some policy
implications of measurement errors in the CPI. In Section 6, I will conclude the paper by
proposing some possible measures to improve the accuracy of the CPI.

2. Sources of Measurement Errors

In this section, I first show the limitation of the fixed-weight Laspeyres index
formula used in constructing the CPI, and then examine various causes of measurement
errors from the practical viewpoint of constructing the CPI statistics.

(1) Limitation of the fixed-weight Laspeyres index formula

The CPI is constructed by using a fixed-weight Laspeyres formula. This formula

has been adopted because (i) it is a simple formula that calculates the weighted average of
sample prices using weights fixed at those of the base period, and (ii) it is only necessary to

2
For the discussion in the US, see also Gordon (1993), Wynne and Sigalla (1994, 1996), Fixler (1993), and
Shapiro and Wilcox (1996). Moulton (1997) summarized various estimates of the magnitude of upward bias
in the US CPI in tabular form. Discussion on this issue between academia and BLS economists still
continues, and the 1998 Winter issue of the Journal of Economic Perspectives, a special issue on the
measurement errors in the US CPI, covers various contributions from members of the Boskin Report, BLS
economists, and academia (Boskin et al., 1998; Abraham et al., 1998; Deaton, 1998; Diewert, 1998; Nordhaus,
1998; and Pollak, 1998).

2
survey the prices of the reference period in order to calculate the price index, thus making it
possible to limit the total cost of constructing the statistics.

At the same time, however, the adoption of the fixed-weight Laspeyres formula to
compile the CPI is also a main cause of measurement error. Within the framework of the
Laspeyres formula, it is difficult to cope with dynamic changes in economic activity such as
changes in consumers’ choices in response to relative price changes, the introduction of
new goods, and the disappearance of old goods.

These problems would result in the introduction of measurement errors in the CPI
through its three components: (i) accuracy of sample prices, (ii) accuracy of the weights,
and (iii) appropriateness of the index formula. Specific problems which affect these
components are, as examined below: (i) substitution effects induced by relative price
changes, (ii) effects of quality changes, (iii) effects of the introduction of new products, and
(iv) technical problems in constructing the statistics.

When measurement errors in the CPI are discussed, the CPI is compared to the
cost of living index, which represents the changes of total expenditure while holding the
households’ utility level constant.3 In other words, measurement errors in the CPI can
generally be expressed as:

(Changes in the CPI) = (Changes in the cost of living index) + (measurement error).

In this case, the problem of measurement error can be analyzed from the viewpoint of its
size and variability. In the following, I focus on the size of the bias in the CPI, while also
referring to its variability by discussing its upper and lower bounds.

3
See Appendix 2 for a detailed explanation of the relationship between the cost of living index and the
CPI. The Laspeyres index indicates the upper bound of the cost of living index for the utility level at the
base period.

3
(2) Substitution effects

Since the CPI is constructed by surveying prices of specific goods and services at
specific outlets for a fixed consumption basket at a specified base period, it does not
thoroughly reflect changes in households’ purchase behavior in response to relative price
changes (substitution effects).

More specifically, substitution effects might be easier to understand when they are
divided into (i) substitution among item levels, (ii) substitution in aggregating individual
prices surveyed into item levels, (iii) substitution among brands within the same category,
(iv) substitution among the outlets, and (v) substitution induced by the emergence of new
discount outlets (outlet substitution effects).

1) Substitution among item levels: This bias is induced by the fact that the weight used
to calculate the weighted average of prices is fixed at that of the base period. For
example, fish and meat are generally thought to be substitutes, thus there will be a
shift in household expenditure from meat to fish when the price of meat increases.
Since the weights applied to meat and fish in the CPI are those of the base period, an
upward bias is introduced by the overvaluation of the price increase of meat.

2) Substitution in aggregating individual prices into item levels: This bias is

introduced at the stage of aggregating individual prices, which are at a lower level
than the publicized items, into the item levels.

3) Substitution among brands within the same category: This bias is induced by the
fact that there exist many goods other than those surveyed which are close substitutes.
For example, the current CPI adopts a color television set of 21-inch multiplex-voice
type with brand specified as the survey sample. However, electronics chain stores
and supermarkets carry various television sets, ranging from large size color
television sets with satellite tuners to compact low price television sets. Although
the CPI price surveys are conducted while considering a product’s representativeness

4
 in the market, most of the items surveyed are specified as one brand or specification,
thus making it difficult in many cases to gauge the price changes in the product
categories as a whole.

4) Substitution among outlets: This bias is induced by the inability to thoroughly grasp
consumers’ price search activities. For example, many consumers are believed to
shop around neighboring outlets such as supermarkets, department stores, and brand
shops in order to purchase the cheapest product at the time of shopping. Since the
CPI surveys the selling price on a specific date and at a specified shop, it fails to
thoroughly reflect such activities.

5) Outlet substitution effects: This bias is induced by the structural change in retail
stores that have been the focus of recent attention in relation to the so-called “price
busting” phenomenon. The CPI price survey fails to cover most of the discount
stores, thus insufficiently reflecting the consumers’ shift from retail shops and
department stores to discount stores.4

These substitution effects are classified into the problem of index formulation ((i)),
the problem of aggregating sampled prices into an item level ((ii)), and the problem of
survey prices ((iii) -(v)).5

In addition, as years pass after the base period revision, the levels of the price
index for various items differ substantially. Such differences will lead to an overvaluation
of the items whose price has increased in the case of an arithmetic average index such as the

4
In order to incorporate the effects of advance in discount outlets into the CPI, it is necessary to examine
whether price differences between existing outlets and discount outlets correspond to quality difference
between them. One criterion is to observe changes in consumers’ behavior: if consumers are shifting from
existing to new discount outlets, it can be assumed that the number of consumers who felt the products in
discount outlets to be less expensive after taking account of quality differences is increasing. It should be
noted that what is referred to here as quality differences are not only the differences in the “product itself,” but
also the difference in “retail services” such as how easy it is to shop and how crowded the parking lot is.
5
In the Boskin Report, introduced in Section 4 of this paper, calls (i) “upper level substitution,” (ii)
“lower level substitution,” and (v) “new outlet substitution.”

5
Laspeyres.6

(3) Effects of quality change

The CPI surveys specific items continuously, which becomes difficult in many
cases where the products surveyed have disappeared from the market or have lost
representativeness as a result of structural changes in the economy or the development of
technological innovations. Therefore, it becomes necessary to substitute survey samples
(specifications) in line with the transition of product cycles in the market. In such cases,
quality differences between new and old specifications are adjusted so that pure price
changes are reflected in the price index: these adjustments are called specification
changes.

The current Japanese CPI mainly adopts the following three methods of
specification changes. First, when the change does not involve any difference either in
quantity or in quality, the price of the new specification is directly linked to that of the old
one (direct comparison method). Second, when there is an apparent qualitative
improvement as well as a price increase, the price index is automatically linked by
assuming that the price index of both specifications are constant (price link method).
Third, when there is no qualitative change and the difference between the new and the old
specifications is attributable to the difference in quantity, the prices are linked after
adjusting the ratio of the new and old quantities.7

As a price index is supposed to represent a price change of a product while keeping

its quality constant, its rate of change should be equal to the rate of change in product price
minus the rate of a quality change: that is,

6
See Shiratsuka (1995a) for details.
7
In the case of the Japanese Wholesale Price Index (WPI), besides the widely used “cost comparison
method,” which is adjusted for quality changes based on the difference in production cost, the “hedonic
approach” has also been used for some of the items.

6
 D(Price Index) = D(Product Price) - D(Quality),

where D(•) represents the rate of change of the variable in parentheses. This relationship is
useful for explaining the above three methods for quality adjustment in the Japanese CPI.

The direct comparison method assumes a “quality improvement rate equal to

zero,” making the price index increase rate equal to the nominal price increase rate. The
price link method, on the other hand, assumes a “nominal price increase rate equal to the
quality improvement rate,” thus making the price index increase rate zero. However, both
methods are not that realistic: in the real world, there might be quality changes, and such
changes might not be equivalent to price changes. Thus, quality differences are
appropriately adjusted only when the third method can be applied, that is, when quantity has
changed without any qualitative changes.

As a result, it is apparent that the CPI does not fully account for actual quality
changes. Therefore, the quality adjustment methods in specification changes are a most
likely source of measurement errors in the official CPI, especially in products subject to
rapid technological innovation, such as electronic products.

Shiratsuka (1995c) checks the accuracy of the quality adjustment method in the
CPI through a simulation of specification changes for automobiles, and suggests that quality
changes are likely to have caused a upward bias in the CPI.

Some 13 Toyota and Nissan models were selected as simulation samples, and their
quality changes from the models in the previous year are evaluated by the pre-estimated
hedonic functions.8 The Table 1 presents the rate of changes in product price, quality, and
quality-adjusted price indices. Toyota Corolla models, for example, changed in 1990-91
and product prices rose 20.5 percent. However, as the quality change computed by the
hedonic function increased 17.1 percent, it follows that the quality-adjusted price index rose

8
See Appendix 2 for the details of the evaluation method for quality difference used in the simulation.

7
only 3.4 percent (20.5 percent minus 17.1 percent).

In the 52 simulation samples (13 automobiles times four years), 28 cases are
deemed to have some quality changes. Among these 28 cases, it is only with respect to the
1994 Toyota Camry that the rate of change in the quality-adjusted price index is less than
two times the standard errors in the bottom row in the Table. Therefore, the other 27 cases
have experienced a statistically significant change in the quality-adjusted price index. An
increase in the quality-adjusted price index is found in 11 cases and a decrease in 16 cases.

The above simulation used relatively continuous models to ensure consistency in

terms of size and styling of the simulation samples. In practice, however, the construction
of a price index is faced with product diversification in terms of sizes and stylings such as
an increase in the number of standard and RV-type vehicles in the market. In such cases,
the present construction methodology of the CPI is less than appropriate and a bias is likely
to exist. Moreover, because quality changes are not properly taken into account, there is a
high possibility that a specification change in the sample structure may miss the best timing.

(4) The new goods effect

Since new goods and services are not brought into the CPI basket immediately, but
only after a time lag after their introduction to the markets, the impacts of the appearance of
new goods and services on the CPI are not thoroughly reflected. When new products are
introduced and come into wide use among households, they will create new demand as well
as replace old products. This phenomenon suggests that households regard new products
as relatively less expensive than old products on a quality-adjusted basis. In other words,
unless the new products are included in the survey sample, the price of items included in the
survey will become relatively more expensive than those excluded from the survey, thus
resulting in an upward bias in the CPI. Figure 1 illustrates this point.

As shown in Table 2, which lists the products newly adopted at the time of base
year revision, new products are not introduced at the appropriate time. In fact, new

8
products included in the CPI basket after a certain time lag from the time they came into
wide spread use in each household are compact cars (under 2,000cc engine displacement)
and pianos (1970); fully-automatic washing machines, stereos, and tape recorders (1975);
microwave ovens, and portable calculators (1980); room air conditioners (1985); word
processors, and camcorders (1990); medium size cars (over 2,000cc engine displacement),
and telephones (1995). In addition, personal computers, facsimile machines, and cellular
phones are yet to be included in the survey range.

Lags in introduction are also observed in the case of some services. For example,
garage rental charges and amusement park fees are included from the 1985 base, and fast
food prices such as hamburgers and rental fees for videotapes are included from the 1990
base. Telephone bills of the new telecommunication companies and rent-a-car fees are
still not included, and various financial services such as credit card fees and account
transfer fees are excluded from the survey range.9

Such lags are especially large in the case of products subject to rapid technological
innovation and short product cycles. For such products it is difficult to measure quality
changes, and this makes it very hard to construct and update quality-adjusted price indices
using conventional methods. As a result, it has been decided to postpone introduction of
these items into CPI basket until some time in the future.

In addition, it has been pointed out that, since items are subdivided into lower
disaggregation levels, newly adopted commodities are not always compared with existing
ones with similar functions in the CPI basket. For example, when personal computers are
included in the future, effects that stem from their substitution for word processors will not
be taken into account.10 This implies that the appearance of new goods affects the

9
The Corporate Service Price Index (CSPI) already includes various financial service fees such as those of
bank account transfers.
10
Of course, since many of the new products provide new functions that are not available in the old
products, there exists a limit for exact comparison. Correspondence by electronic mails on the Internet and
usage of cellular phone are regarded as substitutes for communication based on existing telephones, facsimiles,

9
accuracy of the CPI through not only the improvement in quality but also the increase in the
range of goods and services.

(5) Technical problems in constructing the CPI statistics

In addition to the aforementioned substitution bias, quality change bias, and new
product bias, there are unique technical problems pertaining to the compilation
methodology of the Japanese CPI. Such technical problems can be divided into the
problems of price survey and those of weighting methods.

There are two major problems in price survey. First, since price quotations are
collected on a specific date, irregular factors such as bargain sales and seasonal prices are
easily introduced. The CPI survey is, in principle, conducted every month on Wednesday,
Thursday, or Friday in the week which includes the twelfth day of the month, thus the
actual survey date will vary by a maximum of eight days (see Figure 2).11 As a result, the
price quotation of some items can differ substantially depending on whether or not the
survey date coincides with a special event such as a bargain sale.12 Second, monthly
changes in private rents tend to differ substantially, since their price quotations have been
collected only once every three months, and the number of samples is limited.

With regards to the weights, a major source of the problem lies in the Family
Income and Expenditure Survey (hereafter FIES), compiled by the Management and
Coordination Agency, from which the CPI weights are calculated. Mizoguchi (1992)
reviewed the past discussion of this issue and pointed out the following two points. First,

and postal services, which also have strong features as new ways of communication. See Nordhaus (1997)
for the detailed discussion on this point.
11
As an exception to the principle, hotel charges are collected every month on the weekend (Friday and
Saturday) in the week that includes the fifth day of the month. In addition, price quotations of fresh foods,
which often show big changes due to factors such as bad weather, are collected three times a month.
12
The CPI excludes, in principle, items sold at specially reduced prices from the viewpoint of collecting
price samples of goods and services regularly sold; although the survey includes items sold at such special
prices for more than 7 days at the time of the survey. As such, since the survey date is subject to differ by 8
days at the greatest, items sold at special prices will be sometimes included and sometimes excluded.

10
there is a bias in the process of selecting the household samples. Second, the survey is
insufficient to gauge the total expenditure of the household since it excludes households
with one person, and it collects information mainly through housewives who might not be
fully aware of other family members’ expenditures.13 In addition, problems pertaining to
the calculation of the weight of imputed rent have also been pointed out.14

3. Magnitude of the Measurement Error: A Quantitative Evaluation

In this section, I will present a quantitative evaluation of the upward bias in the
CPI. I will first specify, among the causes of the measurement errors we have discussed
so far, which ones are suitable for quantitative evaluation at this stage. Then, I will
calculate the point estimates for each of the individual causes specified, and estimate the
overall bias by summing them up.

(1) The range of quantitative evaluation

As a starting point, I will rearrange the four causes of measurement errors ---
substitution effect, quality change, introduction of new products, and technical problems
pertaining to the construction of the statistics --- from the viewpoint of the three
components of the CPI --- index formula, accuracy of prices surveyed, and accuracy of the
weights --- and specify the range for quantitative evaluation at this stage.

First, as previously mentioned, there are four problems with the substitution
effects: (i) the index formula for aggregating the upper level items, (ii) the substitution in

13
On September 8, 1996, the evening edition of Nihon Keizai Shinbun carried an article “Four problems of
the Household Expenditure survey,” which pointed out the problems of the survey as: (i) it requires a great
deal of time to fill it out, with small reward, and is prone to omission, (ii) it excludes households with only one
person, (iii) the sample number is small, and (iv) uncertain expenditure such as pocket money and social
expenses are increasing.
14
See Shiratsuka (1995a, 1996) for details.

11
aggregating prices surveyed into item levels, (iii) the substitution among brands within the
same category, (iv) the substitution among existing outlets, and substitution due to the
appearance of new outlets. Among these, the first two problems, i.e. the problems in
index formula for aggregating the upper level items, and the substitution in aggregating
prices surveyed into item levels, can be evaluated as problems in the index formula. The
last problem of substitution due to the introduction of new discount outlets is relevant to the
accuracy of price information. However, the other problems magnify monthly variability,
but do not have a significant impact on the direction of measurement bias.

Second, I will quantify the impact of quality change and the introduction of new
goods and services together as problems in quality adjustment methods which substantially
affect the accuracy of prices surveyed. Since quite limited quality adjustment methods
have been adopted in Japan, the problem of quality adjustment is deemed to be significant.

In the previous section, I have discussed quality changes and the introduction of
new products separately: the former as the improvement of the quality of goods and
services in the survey, and the latter as the introduction of new goods and services into the
price survey. However, the difference between quality change and the introduction of new
products depends heavily on how finely disaggregation levels of commodity classifications
are subdivided. Thus, from the practical viewpoint of constructing statistics, it is quite
difficult to separate them in an explicit way. In addition, the official CPI does not cover
all products, because quality changes in some products are difficult to measure using the
conventional methods.

Finally, technical problems in constructing the statistics are excluded from the
range of quantitative evaluation in this paper. This is because it is assumed that they may
not have a substantial impact on the accuracy of the index when we consider a longer term
such as a yearly average.15 Although these problems can be major causes of measurement

15
With regard to the accuracy of the weights, it is true that the possibility of remaining bias even on a yearly
average basis cannot be denied. For example, if there is an item that shows a smaller increase than that of the

12
error, many of the technical problems are factors that lead to a magnification of the monthly
variability of the price index.

In summary, what I can quantitatively estimate at this stage are the effects of index
formula; aggregating individual price samples into the item level; quality adjustment
methods; and structural changes in the retail market. Figure 3 illustrates the relationship
between causes of measurement error and quantitative evaluations to be shown in this paper.

(2) Problems of index formulas

As in previous studies, I will evaluate the problem of index formula by comparing

fixed-weight Laspeyres index to the superlative indices such as the Fisher and Törnqvist
indices (see Appendix 2 for the details on the index formula). To this end, I picked up,
from the CPI data series, the lowest classification (88 commodities) available continuously
from 1970 up to now; estimated the corresponding weights annually by using the FIES16
and composed the fixed-weight Laspeyres price index (corresponds to the CPI), as well as
the chained Törnqvist and Fisher price indices.

Figure 3 reports the estimation result: the fixed-weight Laspeyres index, the
chained Törnqvist index, and the chained Fisher index (1970=1) are 3.011, 2.923, and 2.923,
respectively. When converted into annual change rates, each index level corresponds to
the annual inflation rate of 4.167, 4.053, and 4.053. Therefore, the fixed-weight Laspeyres
index has upward biases over both the Törnqvist and Fisher indices in the rate of 0.114

total index and the weight of such an item is undervalued, then the total index incorporates an upward bias.
In the Japanese CPI, weights of durable goods, which decline more sharply than the overall index, are deemed
to be undervalued because of the low coverage of the expenditure pattern of family members other than the
housewife. As such, it can also be considered from the viewpoint of the accuracy of weights that the
Japanese CPI possibly inherits an upward bias.
16
Each index formula is composed by using the smallest 88 specifications for which a continuous series is
available since 1970 in the CPI data, and by estimating the CPI weight from the FIES (Family Income and
Expenditure Survey). It should be noted that imputed rents are excluded due to the difficulty of calculating
the weight of each year. For the fixed base series, weights are modified every five years taking into account
that the base year is revised every five years.

13
percent from 1970 to 1997. These figures are somewhat smaller than the US’s estimate of
0.2 percent, although bias can be expected to become larger once specifications are further
divided into more detailed items.17

In Table 3, I further divide the series into time periods of five years and compare
the divergence of the chained Törnqvist and Fisher price indices from the fixed-weight
Laspeyres price index. It shows that the variability of biases varies according to periods
and index formula. However, the divergence of chained Törnqvist and Fisher price
indices are 0.025 and 0.030 percentage points per year, respectively, in the 1990s, which
illustrate that substitution effects are almost negligible at the moment.

One possible explanation for the recent decline in the impact of substitution effects
is that the variability of relative prices was reduced under the low inflation rates, and, as a
result, consumers might have had less scope for substitution recently. To test this
hypothesis, following Shapiro and Wilcox (1997), I calculated an index of the cumulative
change in relative prices, Jt, defined as follows:

æp ö
J t = å wi 0 lnçç it ÷÷ - ln P0Gt , (1)
i è pi 0 ø

where wi0 is the expenditure share in the base period, pit is the price of item i at time t, and
PG0t is the fixed-weight geometric mean index that is defined in Appendix 1. The bottom
row in Table 3, previously appeared, shows changes in the index J from the previous year.
Except for the period from 1970 to 1975, when the first oil crises occurred, both the
magnitude of substitution effects and the pace of relative price drift are mild.

Taking into account the results, the size of upward bias caused by the index
formula is in the range of 0.00-0.25 percent, although it differs according to the period

17
This point is suggested in Aizcorbe and Jackman (1993), employing the smallest specification CPI data in
the US (44 regions and 207 item strata).

14
analyzed and the index formula adopted. In addition, such bias is deemed to be negligible
for the latest period.

(3) Problems in aggregating individual sample prices into item level

Since the Management and Coordination Agency does not release the price index
of those lower than the item level, problems in aggregating individual sample prices into
item levels have not been estimated.18 However, taking into account the fact that (i) the
increase in the Japanese CPI is now at a low rate, thus biases caused by the index formula
are deemed to be almost negligible, and (ii) as the classification of Japanese CPI items is
more detailed than the item strata used in the US, it can be safely assumed that biases
caused in the process of aggregation of individual prices into item level are considerably
smaller than the US estimate of 0.25 percent.19

In this paper, I will assume the bias stemming from the process of aggregating
individual prices to item level to be 0.10 percent, a figure derived as the difference between
upper level substitution and lower level substitution, which were both estimated in the
Boskin Report.

(4) Problems in quality adjustment method

My previous research (Shiratsuka, 1995b, c, Shiratsuka and Kuroda, 1995)

estimates hedonic functions for the quality-adjusted prices for durable goods such as
personal computers, automobiles, and camcorders, and shows that their quality-adjusted

18
With respect to the WPI, the Bank of Japan (1998) begins to compile and release a reference index using
a geometric mean formula since June 1998. In the reference index, geometric mean is partly used in the
lower level aggregation, from “sample price” to “commodity class,” and Laspeyres formula is used in the
upper level of aggregation, from “commodity class” to “all commodities.” Bank of Japan (1998) shows that
the Laspeyres index has upward bias over the geometric mean index in the rate of 0.3 percent annually from
1995 to 1998.
19
See Advisory Commission to Study Consumer Price Index (1996) and Bureau of Labor Statistics (1977)
for the details of the estimation.

15
prices indicate declining trends. In the official CPI, however, such quality adjustment is
not sufficiently implemented, and thus appears to introduce an upward bias.

(a) Upward bias in durable goods

Shiratsuka (1995b, c), and Shiratsuka and Kuroda (1995) have estimated hedonic
price indices for certain durable goods and calculated the upward bias by replacing the CPI
item indices with their estimated indices. The results are summarized in Table 4.

When hedonic price indices for automobiles, camcorders, and personal computers
are included, the level of the overall CPI is lowered by 0.01, 0.01, and 0.02 percentage
points respectively; and that of CPI durables is lowered by 0.16, 0.09, and 0.36 percentage
points. By just adding up these figures, the upward bias reaches 0.04 percentage points for
the overall index, and 0.6 percentage points for the durable goods index. Considering the
fact that the relative importance in the CPI basket of these three goods totals just two
percent, contribution of durable goods as a whole to the overall bias can well reach a
substantial amount when such hedonic estimates are obtained for various other
microelectronic products.

It should also be noted that the magnitude of quality change bias also changes over
time. For example, in the case of automobiles, the upward bias for the total CPI has
increased from 0.01 to 0.02 percent during 1993 and 1994.

In addition, among durable goods, microelectronic products such as personal

computers are subject to rapid technological innovation and have short product cycles.
This suggests that such products are subject to the problem of quality adjustment at the time
of specification change. The prices of these products are difficult to track on a continual
basis, and normally accompany lags in their inclusion in the CPI basket. In fact, personal
computers are yet to be included in the CPI, and word processors are included only from the
1990 base CPI.

16
 (b) Quality adjustment for services in the CPI

With respect to services in the CPI, there have been many unresolved problems
such as the difficulty in specifying “one unit for standard service,” and it is thought to be an
area where price accuracy can be greatly improved.20 In Japan, however, existing studies
in this area are limited.21 In the following, taking rent, the cost of privately owned houses,
and medical care as examples, I will point out the problems inherent in quality adjustment
in the CPI services.

In the case of the CPI rent, the current compilation method is likely to introduce an
upward bias because it fails to take account for the recent improvement in the structure and
convenience of houses. The CPI surveys average rent per residential area on three
classifications of houses: (i) wooden small size houses (residence area less than 30 square
meters), (ii) wooden medium size houses (residence area over 30 square meters), and (iii)
non-wooden houses. Table 5 illustrates how the structure and facilities of houses have
developed over time. It shows that the structure of the houses has undergone a major shift
from wooden to non-wooden, with an increasing share of reinforced-concrete and steel
frame construction. In the wooden house category, that of fireproof houses has
substantially increased. In addition, the state of facilities seems to have improved as the
ratios of those houses with flush toilets in bathrooms increase year after year.

With respect to the cost of privately owned houses, the Japanese CPI covers it as

20
For example, see Griliches (1992) and Kroch (1991).
21
With respect to the relation between changes in service price and changes in quality, Sawa et al. (1989)
estimated a hedonic price index for hotel charges and showed that the increase rate of the CPI was bigger than
that of the hedonic index. In addition, there are estimates such as Ito and Hirono (1993), Kasuga (1996) for
house rent, and Nanbu et al. (1994) for medical expenses, although both studies are deemed as problematic for
use in examining the impacts on the measurement errors in the CPI. Ito and Hirono (1993) derived price and
specification data from the new contract rent carried in housing magazines, and those data are deemed to
overestimate the average rent which should be gauged in the CPI. Nanbu et al. (1994) stated that explanation
variables in the hedonic function were not sufficient to adjust for quality changes brought by technological
innovation, and a substantial effect of quality changes was mingled in the estimate parameter of annual dummy
variable.

17
an imputed rent. However, since there is a large difference in facilities and comfort
between privately owned houses for rent and owner-occupied houses, the quality
adjustment of the surveyed price becomes important. As shown in Figure 5, distribution
of the amount of floor space, which can be deemed as a proxy for comfort of the house,
differs substantially between privately-owned houses for rent and owner-occupied houses
regardless of construction materials (wooden or non-wooden). Therefore, it has been
questioned whether the current CPI appropriately gauges the actual owner’s equivalent rent,
which are houses with large residential space.

With regard to medical care in the CPI, gauging advances in medical technology
has been a big problem.22 In Japan, however, the survey items are quite limited, and there
has been great doubt whether the survey reflects overall medical expenditure appropriately
(see Table 6). For example, only standard medicines sold over the counter in a drugstore
are included, while medicines on provided prescriptions in hospitals are not. In regards to
hospitalization expenses, the survey includes that for childbirth but not for others such as
general medical treatment or operations.

It should also be noted that the weight for medical expenses is underestimated, and
the quality adjustment in this area can result in a large impact on the overall bias. The
weight used in the CPI is calculated based on the FIES, which surveys medical expenses
directly paid by the households. Considering the fact that most of such expenses are paid
in an indirect way, through health insurance, the weight calculated only on the basis of
direct payment would probably result in an underestimate of actual medical expenses.

(c) Quality change bias in the total

Table 7 shows the items that are thought to be difficult to adjust for quality
changes, and their weights add up to about 30 percent of the overall CPI.

22
Shapiro and Wilcox (1996) examine the measurement issues in the medical service in detail.

18
 In order to estimate the average upward bias for these items, I take an annual
average of 1.0 to 1.5 percent for total durable goods (estimate in Gordon, 1990) as a starting
point.23 Then, I assume the upper limit of the bias to be approximately three percent,
taking into account the following three factors. First, the Japanese CPI does not employ
the hedonic approach at all, and the quality adjustment methods for the current Japanese
CPI do not account for the actual quality changes appropriately. Second, survey items of
the Japanese CPI are fairly subdivided and their specifications are stipulated in detail, thus
the introduction of formerly uncovered new products into the survey is limited at the time
of the base year revision. Third, there is a time lag before the inclusion of products subject
to rapid technological innovation in the survey sample. As a result, the impact of quality
adjustment is estimated to be somewhere in the range of 0.30 to 0.90 percent. I assume
the median to be about 0.70 percent, slightly higher than the mean value of the range.24

(5) Effects of the structural change in the retail market

The effects of sampling are difficult to quantify because of the lack of appropriate
statistical evidence. Admitting its extreme nature, I have made an examination of the case
of the Great Southern Hyogo Earthquake that took place in January 1995.25 The CPI
showed a very peculiar movement before and after the earthquake, which could be used to
obtain a hint of the magnitude of the effects of the prevalence of discount stores on the CPI.

23
The declining pace of the price of durable goods is not very different between the US and Japan.
According to the estimate of Shiratsuka (1995b), the hedonic price index for personal computers has been
declining annually at the pace of about 30 percent, and this is almost equivalent to that in the US, estimated by
Berndt and Griliches (1993). However, it should be noted that these estimates assume that the CPI for
services inherits the same size of upward bias as that of durable goods. As aforementioned, the CPI for
services is deemed as problematic from the viewpoint of price accuracy, although the size of such a problem is
yet to be empirically examined.
24
The estimate here specifies the range affected by quality changes, and assumes an average upward bias
within the range, and no bias without the range. This is an unavoidable treatment due to the lack of existing
studies in judging to what extent individual goods and services contain upward bias. However, it is also
reasonable to think that the size of upward bias differs for goods and services. This question awaits future
study.
25
The Great Southern Hyogo Earthquake was the worst natural disaster in Japan in 70 years. More than
5,000 people died, and about 2 million people, including foreign residents in Japan, suffered from the disaster.

19
 In February 1995, the CPI of Hyogo prefecture, where Kobe city is situated,
decreased by 2.3 percent from the previous month, falling substantially lower than the CPI
decrease of 0.4 percent in the Tokyo Metropolitan area. Since the Great Southern Hyogo
Earthquake broke out between the dates of CPI price survey in January and February, such
irregular changes in prices may suggest some connection with the Earthquake. In practice,
it is reported that reasons behind this phenomenon are (i) the price of fresh foods which
need fire to cook have decreased, (ii) damaged shops made a discount sale of their stocks or
pulled down the price of their products, and (iii) supermarkets which carried rather cheaper
products were substituted for department stores and shops which were shut down.26

Table 8 compares the price movements of commodity groups between Hyogo

prefecture and Tokyo Metropolitan area for the period before and after the earthquake.
The table lists 19 commodity groups in which the Hyogo CPI declined two percentage
points more than the Tokyo CPI in February 1995, and placed them in reverse order.
Items normally believed to be discounted deeper in the discount stores such as liquors (beer,
whiskeys, etc.), recreational goods (toys, sporting goods, etc.), and clothing accessories (ties,
belts, etc.), rather than fresh foods, are ranked in the top class. The size of the decline
compared with the previous month is far larger than the average change of the previous
three years. Furthermore, if three categories --- fish, meat, eggs, and daily products, which
are regarded as fresh foods --- are excluded from the above 19 commodity groups and the
weighted average of the difference in the monthly change rate between the two areas is
calculated using the Tokyo Metropolitan area weights, it is estimated that the overall index
will be lowered by about 1.7 percentage points.

Based on the above result, the downward bias caused by the substitution of the
outlets surveyed is deemed as significant, taking into account the finding that commodity
groups with large divergence were thought to be greatly affected by adding discount outlets

26
For example, see evening edition of Nihon Keizai Shinbun on March 3, 1995.

20
to the survey.27 Of course, it is true that this result should be interpreted carefully, since
the divergence of the two areas is due partly to Hyogo’s unique factors such as stock
clearance sales by shops which had suffered damage during the earthquake (for example,
Japanese clothing).

It should be noted that the expansion of new and low-priced outlets such as
discounters and road-side shops, sometimes represented by the development expressed as
“price busting,” does not progress at a constant pace. In particular, recent price
development and consumer behavior suggest that the shift from department stores and
specialty shops to discount outlets has largely subsided, and price differences between these
outlets has settled down to a level consistent with the difference in retail service quality
provided by them. This phenomenon implies that measurement errors induced by
structural changes in the retail market have been diminishing in recent years.

Bearing these points in mind, I will assume the upward bias of the CPI to be 0.10
percentage points for the median, 0.05 percentage points for the lower limit, and 0.60
percentage points for the upper limit, which corresponds to one-third of the above
estimation result.

(4) The magnitude of measurement errors

(a) Evaluation of upward bias in total

As discussed above, the measurement biases are introduced by way of index

formula, aggregation of individual prices into item index, quality adjustment method, and
price survey sampling. If I sum my point estimates measurement biases in those four
sources, the total bias is, at this moment, judged to be some 0.90 percentage points, as

27
Of course, price differences between existing retail outlets and discount stores partly reflect the difference
in the retail services provided, which also needs to be adjusted in the CPI.

21
shown in Table 8.28 However, it should be noted that a possible range of estimates will be
as wide as from 0.35 to 2.00 percentage points, according to various conditions.29

(b) Reservations for the results of the estimates

It should be noted that the method adopted in this paper, that is, the individual
examination of the problems inherent in the Japanese CPI and the simple adding of the
results, has the following limitations:

1) At present, available research results are quite limited in Japan. With respect to the
effects of quality adjustment, for example, it has been proven that there is an upward
bias for certain durable goods, especially for microelectronic products which are
subject to rapid technological innovation. However, for non-durable goods and
services, there is no accumulation of studies in Japan, which lead to an indecisive
conclusion with respect to the impacts of quality adjustment on the price index.

2) The question of whether effects of the problems inherent in the CPI with regard to its
accuracy can be correctly estimated by this simple adding has been noted. As
already shown, sources of measurement errors in the CPI and the sources which
introduce such error are mutually correlated and quite complicated. Adding them
without any adjustment means that I assume no correlation among sources.30

3) I have shown that the point estimate of the upward bias in the Japanese CPI is judged

28
The size of measurement error is, as described later, less than one percent on an annual basis, although it
is quite important to adjust appropriately for such errors taking into account the cumulative effect on assessing
the general price level and productivity.
29
In this paper, I update the estimation results of the upward bias in the Japanese CPI shown in Shiratsuka
(1998). The point estimate remains unchanged, while upper limit was lowered from 2.35 to 2.00 percentage
points per year, based on the revised estimation results on the upper substitution bias.
30
In order to solve the problem of mutual dependence of sources of measurement errors, and that of
estimates of measurement errors and credibility range, Shapiro and Wilcox (1996) specified the (subjective)
probability distribution of biases for each source, and calculated the probability distribution of overall
measurement errors taking into account such mutual relationships.

22
 to be 0.9 percentage points on an annual basis with a possible range of 0.35-2.00
percentage points. However, this range does not refer to a statistical confidence
interval. The point estimate itself is the most reliable “best shot” based on all the
available information to date, although it is true that the calculation is also based on
many assumptions. Therefore, it should be noted that the estimates shown in this
paper are not necessarily ones with a high accuracy.

4. Comparison with the Boskin Report

In the US, the Senate Finance Committee’s Advisory Commission for studying the
CPI presented a report “Toward a more accurate measure of the cost of living” (the so-
called “Boskin Report”) in December 1996.31 The Report specified four sources of
measurement error: (i) upper level substitution, (ii) lower level substitution, (iii) new
products/quality change, and (iv) new outlets. The Report examined past studies in detail
for the above four sources, and presented its best estimate of the size of the upward bias as
1.10 percentage points per year (see Table 10 for details).

If I roughly compare the four sources pointed out in this report with those of our
study, they will correspond (i) to the index formula problem, (ii) to the problem of
aggregation of individual prices, (iii) to quality adjustment methods problem, and (iv) to
survey sample problem. Among these sources, lower level substitution has not been
studied in Japan due to data availability, thus the estimate for this cause should be regarded
as preliminary.32

31
Since the publication of the Boskin Report, there has been a lot of discussion on support for, and criticism
of their estimation results. However, Boskin et al. (1998) takes the position that there are no reasons to
change the original estimate of a 1.1 percentage points per annum upward bias in the US CPI.
32
As the range of plausible values of the upward bias in the Japan CPI, I have set 0.35 to 2.00 percentage
points (0.8 to 1.6 in case of the Boskin Report) around the point estimate of 0.90 percentage points. Due to
the lack of similar studies in Japan, the estimate is bound to be quite preliminary, thus taking quite a wide
range.

23
5. Economic Policy Implications

In this section, I will discuss the policy implication of measurement errors in the
CPI.

(a) Measurement of price stability

As the measured inflation rate approaches zero, it is generally believed that

measurement error portion increases within the observed inflation rate. Therefore,
although the accuracy of the price index will not be that problematic in the process of
lowering the inflation rate from, say, ten percent to three percent, it will become crucial in
considering bringing down the rate from three percent to zero.

In this sense, mismeasurement in the CPI matters a lot for the conduct of monetary
policy. The existence of upward bias in the CPI means that pursuing a zero inflation rate
is to conduct a deflationary policy, thus possibly resulting in a loss of economic welfare.
This suggests that true price stability will not necessarily correspond to zero measured
inflation.

In addition, the time-varying nature of this problem in the short-run suggests that it
is difficult to interpret movements in the measured inflation rate. In other words, to accept
a certain inflation rate as an upward bias may also lead to the loss of economic welfare,
since the magnitude of measurement error is deemed to change according to economic
conditions. Bearing this point in mind, how measurement errors in the price index change
over time in relation to the business cycle is an important issue awaiting solution.

Furthermore, an overestimation of inflation is, to put it differently, an

underestimation of productivity growth or economic growth. For example, if price decline
is brought about by a downward shift of the aggregate supply curve as a result of an
increase in productivity, it is possible to argue that such downward pressure on the general

24
price level is acceptable (Figure 6). Even if the price index incorporates an upward bias of
the same magnitude, the implication for monetary policy will differ according to the source
of such bias.

(b) Treatment of asset prices

As far as monetary policy tries to achieve the medium- to long-run sustainable

price stability, it is not sufficient to monitor only the fluctuation of current price indices.
Therefore, it is important for policy judgement to take into account the asset prices that
implicitly reflect the future development of goods and service prices as well as current price
indices. From the viewpoint of considering the dynamic nature of consumer behavior, it is
important to extend the current price index concept in order to trace intertemporal changes
in cost of living, depending on the future path of consumption.

However, asset price information can only be used as a supplementary measure to

price indices in making policy judgements on price developments because of the following
two reasons. First, current asset prices are affected by various factors other than price
expectations for future goods and services, thus current changes in asset prices do not
necessarily indicate future changes in prices of goods and services. Second, the accuracy
of such asset prices is quite low compared with those of existing price indices. Therefore,
it is deemed as difficult to attach a central role to price indices, including asset prices, in the
judgement of monetary policy.33

(c) The needs for exploring additional methods to gauge the underlying
trend of inflation

In order to conduct a preemptive monetary policy, it becomes necessary to gauge

appropriately the changes in underlying inflation trends. To make use of the asset prices
mentioned above can be one way, and to reexamine information contained in relative price

33
Treatment of asset prices in price indices is discussed in detail in Shiratsuka (1996).

25
changes across the various goods and services is another.

What makes it difficult to trace the underlying inflation trends are not only the size
and variability of biases in price indices, which both stem from measurement errors, but
also the adjustment method pertaining to transitory or temporary fluctuations in the prices
of individual items. In order to cope with the latter issue and to supplement the judgement
of underlying inflation trends, central banks employ various devices. For example, Japan
uses the CPI series that excludes fresh food items; New Zealand and the UK sometime
employ a limited influence estimator, an index that excludes items located on both tails of
the cross-sectional distribution of inflation.

The limited influence estimator provides important information concerning

changes in underlying inflation trends, and could well be a quite useful and powerful
indicator for policy judgement. Shiratsuka (1997) shows that such an indicator is
applicable to the Japanese economy, and points out that the indicator helps to (i) reveal the
underlying trend in price changes by adjusting for temporary disturbance such as rapid yen
appreciation and sudden rises in oil prices, and (ii) clarify the magnitude of upward pressure
on prices by making use of monthly and yearly changes in the index.

(d) Fiscal balance and implication on fiscal policy

The major incentive for compiling the Boskin Report was that upward bias in the
CPI had a great impact on the fiscal budget. In the US, upward bias in the CPI has been a
major source of the increase in the federal budget, since about 30 percent of fiscal
expenditure (such as Social Security and pension payments) and 45 percent of fiscal
revenue (income tax) are tied to the CPI. According to the Boskin Report, the
Congressional Budget Office (CBO) has estimated that if the change in the CPI was brought
down by an average of 1.1 percentage points for the next decade, it would slash as much as
$148 billion from projected federal deficits by the year 2006 (see Advisory Commission to

26
Study Consumer Price Index, 1996).34

When the fiscal system and inflation indexation are discussed, however,
examination of the validity of the price basket is also important. In the case of pensions, it
could well be the case that the average consumption basket of all households and that of
pension recipients are quite different.

6. Conclusion

In this paper, I have summarized the problems pertaining to measurement errors

inherent in the Japanese CPI, and have provided some quantitative evaluations. Based on
currently available information, I presented the point estimate of about 0.90 percentage
points as the size of measurement error. Although this is the best estimate taking into
account all information currently available, it is true that the estimate was based on various,
rather bold assumptions. In addition, it should be noted that the accuracy of the estimate is
not necessarily high due to the lack of existing studies in this field in Japan.

Despite the efforts of many statistical institutions in constructing accurate statistics,

measurement errors are unavoidable to some extent. Thus, the most important point is to
ascertain whether they are small enough to be safely ignored in practice or serious enough
to mislead users. The debate about the accuracy of CPI should therefore be aimed at
investigating the sources of measurement errors and the extent to which they affect
accuracy. Unfortunately, research on price index measurement errors has been limited in
Japan, and further research in the area is necessary.

In order to deal with the three problems quantitatively estimated in this paper ---

34
Pension payments constitute the only item in Japan’s fiscal budget which is tied to the CPI, and its weight
is, in fiscal year 1994, about 13 percent of the total expenditure of the general government (current
expenditure + total capital formation + acquisitions less disposal of land) in the System of National Accounts.

27
the index formula problem, the effects of quality changes, and the effects of structural
change in retail markets --- can be somewhat clarified by adopting the following methods.

(a) Introduction of chained CES index formula

The superlative indices such as the chained Törnqvist and Fisher indices have
more desirable features than the fixed-weight Laspeyres index formula currently used in the
CPI as a measure of the cost of living. This is because the chained Törnqvist and Fisher
indices reflect the substitution effect appropriately. However, such indices cannot be
computed on as timely a basis as the current CPI due to the delays in the availability of the
required expenditures data.

Therefore, it is important to produce an approximation of the chained Törnqvist

and Fisher indices with data available to the statistical agency when they compile the CPI.
One possible answer to this question is application of the CES index formula, proposed by
Shapiro and Wilcox (1997). More specific, I construct the version of the CES index that
is defined as:

1 /(1-s )
1-s
é æ pit ö ù
Pt CES = êå wit çç ÷÷ ú , (2)
êë i è pi 0 ø úû

where s is the elasticity of substitution between items (assumed to be identical for all
possible pairs of goods and services in the CPI).

Table 11 compares the annual rates of change in the cost of living indices,
computed by the eight different index formulas. The first two rows use the Törnqvist and
Fisher index formulas. The remaining rows apply Equation (2) for six different
assumptions about the elasticity of substitution, that is, s = 0.3, 0.4, 0.5, 0.6, 0.7, and 1.0
(identical to the geometric mean formula with one-year lagged weight). This table shows
that the elasticity of substitution, on average, lies between 0.4 and 0.5 from 1970 to 1997,

28
and between 0.5 and 0.6 in the 1990s. This estimate is a little lower than that in the
United States of 0.7, reported in Shapiro and Wilcox (1997).

(b) Introduction of hedonic approach as quality adjustment method

Bias induced by quality changes can be made small enough by the introduction of
the hedonic approach. I propose to adopt a framework of using a pre-estimated hedonic
function in adjusting the quality difference between new and old products (for the details,
see Appendix 3). This framework can avoid the cumbersome process of estimating the
hedonic function every time the price index is constructed. In addition, this framework is
deemed to be highly feasible since it is relatively compatible with the methodology of
surveying specific prices every month. In fact, the Japanese WPI employs this framework
for some computer items and has been constructing a price index on a monthly basis.

(c) Review of survey outlets

Whether the development of price busting has been appropriately reflected in the
CPI is still an open question. However, judging from the case study of the Kobe
earthquake, I believe that the effect of the expansion of discount outlets on price survey is
substantial. Therefore, the review of survey outlets is worth considering in order to
improve the accuracy of the CPI.

Appendix 1. Index Formulas

In Appendix 2, I practically summarize some basic price index formulas applied in

the paper.

The basic components of a price index are the price of item i in time t, denoted pit,
and the quantity of this item purchased in time t, xit. Then, the fixed-weight version of
Laspeyres (PL0t), Paasche (PP0t), Fisher (PF0t), Törnqvist (PT0t), geometric mean (PG0t)

29
indices are defined as follows:

n n n
pit
P0Lt = å pit xi 0 å pi 0 xi 0 = å wi 0 ´ , (A-1)
i =1 i =1 i =1 pi 0
-1
n n
æ n p ö
P = å pit xit
P
0t å pi 0 xit = çç å wit ´ it ÷÷ , (A-2)
i =1 i =1 è i =1 pi 0 ø

P0Ft = P0Lt ´ P0Pt , (A-3)

( wi 0 + wit 2 )
æp ö n
P0Tt = Õ çç it ÷÷ , (A-4)
i =1 è pi 0 ø
wi 0
æp ö n
P = Õ çç it ÷÷
G
0t . (A-5)
i =1 è pi 0 ø

The chained index formula first aggregates individual prices in period-to-period

basis to compute an intermediate period indices, and, then, chains these intermediate period
indices to obtain a long-term index. In general, chained price index (CPk0t) is defined as
follows:

t -1
CP0kt = P01k ´ P12k ´ L´ Pt -k1,t = Õ Psk,s +1 (for k = L, P, F, T, G). (A-6)
s =0

Therefore, chained Laspeyres (CPL0t), Paasche (CPP0t), Fisher (CPF0t), Törnqvist (CPT0t),
geometric mean (CPG0t) indices are written as:

t -1 t -1 n pi ,s+1
CP0Lt = Õ PsL,s+1 = Õå wis ´ , (A-7)
s =0 s =0 i =1 pis
-1
t -1
æ n
t -1
p ö
CP = Õ P
P
0t
P
s ,s +1 = Õ çç å wis ´ i ,s +1 ÷÷ , (A-8)
s =0 s =0 è i =1 pis ø
t -1 t -1
CP0Ft = Õ PsF,s +1 = Õ PsL,s +1 ´ PsP,s +1 = CP0Lt ´ CP0Pt , (A-9)
s =0 s =0
( wis + wi ,s +1 2 )
t -1
æp ö
t -1 n
CP = Õ P
T
0t
T
s , s +1 = ÕÕ çç i ,s +1 ÷÷ , (A-10)
s =0 s =0 i =1 è pis ø

30
 wi 0
t -1 t -1 n
æp ö
CP = Õ P
G
0t
G
s ,s +1 = ÕÕ çç it ÷÷ . (A-11)
s =0 s =0 i =1 è pi 0 ø

Appendix 2. Theoretical Relationship between Laspeyres Price Index and

Cost of Living Index

The purpose of the CPI is to measure the average change in the prices paid by
households for a fixed basket of marketable goods and services while keeping quality
constant. According to the theory of consumer behavior, the CPI can be considered as an
approximation to the “cost-of-living index” by the Laspeyres price index.35 The cost-of-
living index is defined as the ratio of the minimum expenditure required to achieve a
particular level of satisfaction, or utility level, between two points of time.

Let u R be the constant utility level of a consumer and C (p t , u R ) be the amount

of minimum expenditure necessary to realize this utility level under the price vector p t .
Then the cost-of-living index at time 1 relative to time 0 is defined as

C (p 1 , u R )
P(p 1 , p 0 ; u R ) = . (A-12)
C (p 0 , u R )

The price index for a given consumption vector q R , which minimizes the
household’s expenditure under the price vector of a reference period, while attaining the
utility level u R , is defined as

p 1q R
P(p 1 , p 0 ; q R ) = . (A-13)
p 0q R

35
See Deaton and Muelbauer (1980), Diewert (1987), Pollak (1989), and Morita (1989) for theoretical
background for the price index.

31
 Laspeyres and Paasche price indices are defined respectively as

p 1q 0
P(p 1 , p 0 ; q 0 ) = , (A-14)
p0q0

p1q1
P(p 1 , p 0 ; q 1 ) = . (A-15)
p 0 q1

With these equations in hand, the relationship between these three price indices is
derived. Between the Laspeyres and cost-of-living indices, the following equation holds,

p 1q 0 > C (p 1 , u 0 )
P(p 1 , p 0 ; q 0 ) = 0 0 = 0 0
= P(p 1 , p 0 ; u 0 ) . (A-16)
p q C (p , u )

First, the denominators of both sides of equation (A-16) are identical by definition.
Second, looking at the numerators of both sides of equation (A-16), it is found that p 1q 0 is
greater than or equal to C (p 1 , u 0 ) . This is because that: (i) on the one hand, the
consumption vector q0 does not necessarily minimize the expenditure under the price
vector p 1 , although it attains the utility level u 0 ; (ii) on the other hand, C (p 1 , u 0 ) is the
amount of minimum expenditure necessary to realize the same utility level u0 under the
price vector p 1 .

By similar arguments, the following equation between the Paasche and the cost-of-
living indices are derived,

p1q1 < C (p1 , u1 )

P(p , p ; q ) = 0 1 =
1 0 1
0 1
= P(p1 , p 0 ; u1 ) . (A-17)
pq C (p , u )

Figure A-1 and Figure A-2 make these relationships among three price indices
even more intuitive in a two-goods problem. First, Figure A-1 shows the relationship
between the cost-of-living index and the Laspeyres index. At time 0, a consumer

32
maximizes his or her utility at E, where the budget line AB and the indifference curve for
the utility level u 0 are tangent to each other. When the price of x1 increases, the budget
line shifts to AC. Now consumer equilibrium moves to F, which brings the lower utility
level u 1 . In order to attain the initial utility level u 0 under this new set of relative prices
with minimum expense, the combination G must be realized, and the new budget constraint
intersects the y-axis at point J. On the other hand, if we were to realize the combination E
to attain the same utility level u 0 under the new set of relative prices, the budget constraint
can intersect the y-axis at point K. Since the price of good x2 is held constant, the ratio
of distances to origin OJ/OK is the ratio of expenditure to realize the utility level u 0 with
and without minimizing living expenses.

In other words, the Laspeyres index and the cost-of-living index are respectively
defined as

p 1q 0 OK
P(p 1 , p 0 ; q 0 ) = = , (A-18)
p 0 q 0 OA

C (p 1 , u 0 ) OJ
P(p 1 , p 0 ; u 0 ) = = , (A-19)
C (p 0 , u 0 ) OA

and yielding

OK OJ
³ . (A-20)
OA OA

Hence the Laspeyres index is larger than or equal to the cost-of-living index.

As for the relationship between the cost-of-living index and the Paasche index in
Figure A-2, and these two index formula are respectively defined as,

p1q1 OA
P(p1 , p 0 ; q1 ) = = , (A-21)
p 0q1 OM

33
 C (p1 , u1 ) OA
P(p 1 , p 0 ; u 1) = = , (A-22)
C (p 0 , u1 ) OL

and yielding

OA OA
£ . (A-23)
OM OL

Hence the cost-of-living index is larger than or equal to the Paasche index.

However, the combination of equations (A-20) and (A-23) does not necessarily
mean that the cost-of-living index is located between the Laspeyres and Paasche indices.
This is because there is no generally accepted relationship between the cost of living indices
P(p1 , p 0 ; u 0 ) and P(p1 , p 0 ; u 1 ) at time 0 and 1 respectively. Impacts of changes in
relative prices vary with the utility level.

If preferences are homothetic, the cost-of-living index situated between the

Laspeyres and Paasche indices.36 This implies that the income expansion path or Engel
curve --- the resulting locus of utility-maximizing bundles when prices are held constant
and income is allowed to vary --- must be a straight line through the origin. If this is the
case, then the minimum expenditure function C (p, u ) will be written as a (p) ´ b(u ) .
This implies that the minimum expenditure function is separable with respect to prices and
utility, and depends only on the price vector at a constant utility level. Then the cost of
living index can be written as

36
It should be noted that Diewert (1983) shows that the unobserved true cost of living index lay between the
observable Paasche and Laspeyres price indices without assuming homothetic preferences.

34
 C (p1 , u 0 ) a(p1 ) ´ b(u 0 ) a(p1 )
P(p1 , p 0 ; u 0 ) = = =
C (p 0 , u 0 ) a(p 0 ) ´ b(u 0 ) a(p 0 ) (A-24)
,
a(p1 ) ´ b(u1 ) C (p1 , u1 ) 1 0 0
= = = P(p , p ; u )
a(p 0 ) ´ b(u1 ) C (p 0 , u1 )

thus giving

< P(p1, p0 ; u1 ) = P(p1, p0 ; u 0 )<

P(p1, p0 ; q1 ) = 1 0 0
= P(p , p ; q ) . (A-25)

Appendix 3. Application of the Hedonic Approach to Specification Changes

In Appendix 3, I present a practical method for constructing a reliable CPI by

adjusting quality changes with the hedonic approach. This method is something of a
compromise between the following two requirements. On the one hand, it is necessary to
employ the hedonic approach to account for quality changes more adequately. On the
other hand, it is also necessary to sustain the conventional methodology of surveying
specific prices every month. Thus, it is not appropriate to use the anti-logarithm of the
estimated parameter for the year dummy as a quality-adjusted price index.

Figure A-3 describes the method for applying the hedonic approach to specification changes
in the case of one performance characteristic (functional form for the hedonic function is
assumed to be a semi-log linear). The x-axis measures the characteristic, and the y-axis
measures the logarithm of the product price. A straight line with a constant (a) and a slope
(b) represents the pre-estimated hedonic function. Let XO and XN represent the
characteristic values of the exiting and new products, respectively. Then the theoretical
price (that is, the estimated price based on the hedonic function) is given by the anti-
logarithm of ln P = a + bX, that is, P̂O and P̂N for the old and new products respectively.
The quality change between the existing product and the new product is measured by the
difference in the theoretical prices between the existing product and the new product, which
is depicted by AB in Figure A-3. Let n be the observed prices for old and new products

35
PO and PN respectively. Then, CD measures the change in the product price. Therefore,
the difference between CD and AB corresponds to the change in the quality-adjusted price
index. In this particular example, the quality-adjusted price index rises with an
introduction of the new product because CD > AB.

With this methodology, the following relationships hold among rates of change in
terms of product price, quality, and the quality-adjusted price index.

D(Product Price) > D(Quality) ==> D(Price Index) > 0

D(Product Price) = D(Quality) ==> D(Price Index) = 0
D(Product Price) < D(Quality) ==> D(Price Index) < 0

References

Abraham, Katharine G., John S. Greenlees, and Brent R. Moulton, “Working to Improve
the Consumer Price Index,” Journal of Economic Perspectives, 12(1), 1998, pp. 27-
36.
Advisory Commission to Study the Consumer Price Index, Toward a More Accurate
Measure of the Cost of Living: Final Report, 1996.
Aizcorbe, Ana M., and Patrick C. Jackman, “The Commodity Substitution Effect in CPI
Data, 1982-91,” Monthly Labor Review, December 1993.
Bank of Japan, “Release for a Reference Wholesale Price Index Using Geometric Mean
Formula,” Bank of Japan Quarterly Bulletin, 6(3), 1998.
Bank of Japan Price Index Study Group, Bukka no Chishiki, (Basics of Price Index), Nikkei
Bunko, Nihon Keizai Shinbun Sha, 1992 (in Japanese).
Berndt, Ernst R., and Zvi Griliches, “Price Indices for Microcomputers: An Explanatory
Study,” in M. F. Foss, M. E. Manser, and A. H. Young, eds. Price Measurements and
Their Uses, 1993.
Boskin, Michael J., Ellen R. Dulberger, Robert J. Gordon, Zvi Griliches, and Dale W.
Jorgenson, “Consumer Prices, the Consumer Price Index, and the Cost of Living,”

36
 Journal of Economic Perspectives, 12(1), 1998, pp. 3-26.
Bureau of Labor Statistics, “The Experimental CPI Using Geometric Mean (CPI-U-XG),”
1997.
Deaton, Angus, “Getting Prices Right: What Should Be Done?” Journal of Economic
Perspectives, 12(1), 1998, pp. 37-46.
______, and John Muellbauer, Economics and Consumer Behavior, Cambridge
University Press, 1980.
Diewert, W. Erwin, “Exact and Superlative Index Numbers,” Journal of Econometrics, 4(2),
1976, pp. 114-145.
______, “Superlative Index Numbers and Consistency in Aggregation,” Econometrica,
46(4), 1978, pp. 883-900.
______, “The Theory of the Cost of Living Index and the Measurement of Welfare
Change,” in W. E. Diewert and C. Montmarquette, eds., Price Level Measurement,
1983, pp. 767-780.
______, “Index Numbers,” in John Eatwell, Murray Milgate, and Pete Newman eds., The
New Palgrave: A Dictionary of Economics, vol. 2, 1987, pp. 767-780.
______, “Index Number Issues in the Consumer Price Index,” Journal of Economic
Perspectives, 12(1), 1998, pp. 47-58.
Fixler, Dennis, “The Consumer Rice Index: Underlying Concepts and Caveats,” Monthly
Labor Review, December, 1993.
Gordon, Robert J., The Measurement of Durable Goods Prices, University of Chicago Press,
1990.
Greenspan, Alan, “Opening Remarks,” in Achieving Price Stability: A Symposium
Sponsored by the Federal Reserve Bank of Kansas City, 1996.
Griliches, Zvi, ed., Output Measurement in the Service Sector, Studies in Income and
Wealth, vol. 56, University of Chicago Press, 1992.
Ito, Takatoshi and Keiko Nosse Hirono (1993), “Efficiency of the Housing Market,” Bank
of Japan Monetary and Economic Studies, 11(1), 1993.
Kasuga, Yoshiyuki, “Shutoken ni okeru Juutaku no Kousatsu: Mikuro Deita ni yoru Jutaku
Shijou no Kenshou,” (Housing Problems in Metropolitan Area: Examination of
Housing Market by Micro Data), Chousa No. 211, Japan Development Bank, 1996

37
 (in Japanese).
Kroch, Eugene, “Tracking Inflation in the Service Sector,” Federal Reserve Bank of New
York Quarterly Review, 16(2), 1991.
Lebow, David E., John N. Roberts, and David J. Stockton, “Monetary Policy and ‘The Price
Level’,” mimeo, Board of Governors of the Federal Reserve System, August, 1994.
Mizoguchi, Toshiyuki, Wagakuni Toukei Chousa no Gendaiteki Kadai, (Current Problems
for Japanese Statistical Surveys), Iwanami Shoten, 1992 (in Japanese).
Morita, Yuuzou, Bukka Shisuu Riron no Tenkai, (Development of Price Index Theory),
Touyou Shinpou Sha, 1989 (in Japanese).
Moulton, Brent R., “Bias in the Consumer Price Index: What is the Evidence,” Journal of
Economic perspectives, 10(4), 1996.
Nanbu, Tsuruhiko et al., “Saabisu no Shitsu no Keisoku II: Byouin Saabisu no Hinshitsu
Henkaritsu no Keisoku,” (Quality Measurement of Services: Estimation of the Rate of
Changes in Hospital Services), Financial Review, vol. 31, Institute for Fiscal and
Monetary Policy, Ministry of Finance, 1994 (in Japanese).
Nordhaus, William D., “Do Real-Output and Real-Wage Measures Capture Reality?”
Chapter 1 in T. F. Bresnahan and R. J. Gordon eds., The Economics of New Goods,
1997, pp. 29-66.
______, “Quality Change in Price Indices,” Journal of Economic Perspectives, 12(1), 1998,
pp. 59-68.
Ohta, Makoto, Hinshitsu to Kakaku (Quality and Prices), Soubun Sha, 1980 (in Japanese).
Pollak, Robert A., The Theory of the Cost-of-Living Index, Oxford University Press, 1989.
______, “The Consumer Price Index: A Research Agenda and Three Proposals,” Journal of
Economic Perspectives, 12(1), 1998, pp. 69-78.
Reinsdorf, Marshall, “The Effect of Outlet Price Differentials on the U. S. Consumer Price
Index,” Chapter 7 in M. F. Foss, M. E. Manser, and A. H. Young, eds. Price
Measurements and Their Uses, 1993.
Shapiro, Matthew D., and David W. Wilcox, “Mismeasurement in the Consumer Price
Index: An Evaluation,” NBER Working Paper No.5590, 1996.

Shiratsuka, Shigenori, “Shouhisha Bukka Shisuu to Keisoku Gosa,” (Consumer Price Index

38
 and Measurement Errors: Their Causes and Improvement Measures), Kin’yu Kenkyu,
14(2), 1995a (in Japanese).
______, “Effects of Quality Changes on the Price Index: A Hedonic Approach to the
Estimation of a Quality-adjusted Price Index for Personal Computers in Japan,” Bank
of Japan Monetary and Economic Studies, 13(1), 1995b.
______, “Automobile Prices and Quality Changes: A Hedonic Price Analysis of the
Japanese Automobile Market,” Bank of Japan Monetary and Economic Studies, 13(2),
1995c.
______, “Shisan Kakaku to Bukka Shisuu,” (Asset Price Fluctuation and Price Index),
Kin’yu Kenkyu, 16(5), 1996 (in Japanese).
______, “Inflation Measures for Monetary Policy: Measuring the Underlying Inflation
Trend and Its Implication for Monetary Policy Implementation,” Bank of Japan
Monetary and Economic Studies, 15(2), 1997.
______, Bukka no Keizai Bunseki (Economic Analysis of Inflation Measures), University of
Tokyo Press, 1998 (in Japanese).
______, and Sachiko Kuroda, “Bideo Kamera no Hedonikku Bunseki” (Hedonic Analysis
for camcorder Prices), Kin’yu Kenkyu, 14(4), 1995 (in Japanese).
______, and ______, “Apareru Seihin Kakaku to Hinshitsu Sa: CPI Apareru no Kakaeru
Mondaiten to Hedonic Aprochi ni yoru Kaizen no Kanousei” (Apparel Product Prices
and Quality Differences), Kin’yu Kenkyu, 15(1), 1996 (in Japanese).
Suzuki, Kenji, and Makoto Ohta, “A Hedonic Analysis of Land Prices and Rents in the
Bubble: Kanagawa Prefecture in Japan for 1986-1988,” Economic Studies Quarterly,
45(1), 1994.
Turvey, Ralph, Consumer Price Indices: An ILO Manual, International Labor Organization,
1989.
Wynne, Mark A., and Fiona D. Sigalla, “The Consumer Price Index,” Federal Reserve
Bank of Dallas Economic Review, 1994.
______, and ______, “A Survey of Measurement Biases in Price Indices,” Journal of
Economic Survey, 10(1), 1996.

39
 Table 1. Simulation for Specification Changes
1991 1992 1993 1994
(TOYOTA)
Carolla Product price 20.5 0.0 1.4 0.7
Quality 17.3 0.0 0.0 0.0
Quality-adjusted price 3.2 0.0 1.4 0.7
Carina Product price 0.0 -9.5 0.0 9.1
Quality 0.0 -15.6 0.0 14.3
Quality-adjusted price 0.0 6.1 0.0 -5.2
Corona Product price 0.0 8.6 0.0 0.9
Quality 0.0 18.0 0.0 0.0
Quality-adjusted price 0.0 -9.5 0.0 0.9
Camry Product price 0.0 11.9 0.0 -4.0
Quality 0.0 9.7 0.0 -4.3
Quality-adjusted price 0.0 2.2 0.0 0.3
Mark II Product price 0.0 0.0 10.4 2.6
Quality 0.0 0.0 50.9 -6.2
Quality-adjusted price 0.0 0.0 -40.5 8.8
Crown Product price 7.8 0.0 1.8 0.0
Quality 14.1 0.0 -1.0 0.0
Quality-adjusted price -6.3 0.0 2.8 0.0
Celsior Product price 0.0 5.3 0.2 0.7
Quality 0.0 0.0 0.0 2.5
Quality-adjusted price 0.0 5.3 0.2 -1.8
(NISSAN)
Sunny Product price 3.0 2.9 5.9 -4.4
Quality 0.0 0.0 21.8 -16.5
Quality-adjusted price 3.0 2.9 -15.9 12.1
Primera Product price 3.0 4.7 0.8 1.7
Quality 9.8 9.7 0.0 5.0
Quality-adjusted price -6.8 -5.0 0.8 -3.3
Bluebird Product price 1.3 0.8 1.8 4.1
Quality 9.2 -8.9 0.0 11.3
Quality-adjusted price -7.9 9.7 1.8 -7.1
Skyline Product price 11.6 0.0 -1.9 14.0
Quality 9.7 0.0 15.1 23.0
Quality-adjusted price 1.9 0.0 -17.0 -9.0
Cedric Product price 13.8 0.0 5.8 0.0
Quality 27.1 -11.0 12.1 0.0
Quality-adjusted price -13.3 11.0 -6.3 0.0
Cima Product price 0.0 -0.7 9.5 0.0
Quality -7.0 -8.7 24.2 0.0
Quality-adjusted price 7.0 7.9 -14.6 0.0
Inadequate quality adjustment case 7 7 6 7
Increase in quality-adjusted price (shaded) 3 5 1 2
Decrease in quality-adjusted price (squared) 4 2 5 5
Standard errors for hedonic price index 0.7 0.7 0.7 0.7
Source: Shiratsuka (1995c)
Note: Squared areas, and squared and shaded areas are case of increase and decrease in quality-
adjusted prices, respectively. Crossed area indicates that quality difference is insignificant.
 Table 2. Items Newly Introduced in the Base-Year Revision

Base-year Durable goods Services

1970 Automobiles (661cc to 2,000cc Lesson fees (driving school), Fire
engine displacement), Pianos, Room insurance premiums
air-conditioners, Color TV sets
1975 Washing machines (fully automatic School lunches, Expressway tolls
type), Stereo phonograph sets, Tape
recorders, Gas water heaters
1980 Microwave ovens, Portable electronic Lunch plate for children, Women’s
calculators hairdressing charges
1985 Room air-conditioners, Video tape Garage rental charges, Amusement
recorders park fees, Automotive insurance
premiums (optional), Sewerage
charges
1990 Word processors, Camcorders Hamburgers, Video rental fees
1995 Automobiles (over 2,000cc in engine Pizzas, Karaoke fees
displacement), Telephones
Excluded items Personal computers and peripherals Telephone bills for new
(e.g. printers), Faxes, Cellular telecommunication companies, Rent-
telephones a-car fees, Financial services, Internet

41
 Table 3. Comparison of Laspeyres, Törnqvist, and Fisher Price Indices

70-75 75-80 80-85 85-90 90-95 95-97 70-97 90-97

(Annual changes)

Fixed-weight Laspeyres (a) 11.379 6.297 2.604 1.095 1.155 0.892 4.167 1.079

Chained Törnqvist (b) 11.052 6.194 2.534 0.994 1.137 0.849 4.053 1.055

Chained Fisher (c) 11.052 6.194 2.534 1.001 1.137 0.831 4.053 1.049

(Deviations)

Chained Törnqvist (a)-(b) 0.327 0.103 0.070 0.101 0.017 0.043 0.114 0.025

Chained Fisher (a)-(c) 0.327 0.103 0.070 0.094 0.018 0.061 0.114 0.030

(Relative Price Changes) 3.207 1.191 0.745 1.114 0.953 1.346 1.442 1.065

42
 Table 4. Upward Bias in Durable Goods

Annual change (%) Contribution (%)

Weight
Hedonic
(%) CPI Difference to Durables to Overall
Index
Automobiles 1.8 0.1 -0.4 -0.5 -0.16 -0.01

Camcorders 0.1 -4.0 -9.6 -5.6 -0.09 -0.01

Personal
0.1 n. a. -24.4 n. a. -0.36 -0.02
computers
Sources: Author’s calculation based on Shiratsuka (1995b, c), and Shiratsuka and Kuroda
(1995).
Notes: 1. The estimates of upward bias are the average from 1991 to 1994.
2. The weights are on the basis of 1990.
3. Estimated on the assumption that half of the weight for word processors in CPI is
replaced by personal computers.

43
 Table 5. Construction Structure and the State of Facilities
Unit: %
1973 1978 1983 1988 1993
ratio of wooden houses 86.2 81.7 77.4 73.0 68.1
(the ratio of wooden and fire-
19.7 25.4 31.3 31.7 34.0
proofed houses)
ratio of non-wooden houses 13.8 18.3 22.6 27.0 31.9

(the ratio of ferroconcrete houses) 10.5 15.2 20.0 24.5 29.0

ratio of houses with flush toilets 31.4 45.9 58.2 66.4 75.6

ratio of houses with bathrooms 73.3 82.8 88.3 91.2 93.5

Sources: Management and Coordination Agency, The Housing Survey of Japan.

44
 Table 6. List of CPI Items: Medical Care

Item Weight Specification

(Medicines)
Medicines for cold (multipurpose) 0.0011 “Shin Lulu A”
Medicines for cold (antipyretic and 0.0005 “Bufferin A”
analgesic)
Gastroenteric medicines (digestive) 0.0002 “Ohta’s Isan”
Gastroenteric medicines (combined) 0.0005 “New Pan Siron”
Vitamin preparations, multivitamins 0.0012 “Panvitan Hi”
Vitamin preparations, vitamins 0.0012 “ALINAMIN A”
compound
Health Drinks 0.0012 “Ripobitan D”
Dermal medicines 0.0005 “MENTHOLATUM”
Plasters 0.0004 “TOKUHON A” or “SALONPAS A”
Breath fresheners 0.0007 “Jintan SILVER PILLS”
Chinese medicines 0.0029 For women, decoction, “Chujoto”
(Medical services)
Medical treatment 0.0128 Rate of charges shared by the insured
Hospital charges 0.0032 Charges for ordinary delivery, 8 days hospital
treatment
Massage fees 0.0004 Except in the application of insurance, massage
from head to foot, about one hour

Source: Management and Coordination Agency, Annual Report on the Consumer Price Index.

45
 Table 7. Weight affected by quality adjustment problem

Weight
affected by
Weight quality Note
adjustment
problem
Commodities 0.51589 0.09530
Agricultural and aquatic products 0.08663 0.00000
Food products 0.13494 0.00000
Textiles 0.06544 0.04592
Clothing 0.02727 0.02727
Shirts and sweaters 0.01864 0.01864
Others 0.01952 0.00000
Durable goods 0.05462 0.03911
Domestic durables 0.00601 0.00601 Microwave ovens, refrigerators, etc.
Heating and cooling appliances 0.00446 0.00446 Room air-conditioners, etc.
Automobiles 0.01818 0.01818
Recreational durables 0.00972 0.00883 TV sets, camcorders, etc.
Toys 0.00311 0.00087 Household video game machines
Others 0.01315 0.00076 Telephones
Other industrial products 0.11315 0.01027
Medicines 0.01043 0.01043
others 0.10272 0.00000
Electricity, gas, and water charges 0.04377 0.00000
Publications 0.01734 0.00000
Services 0.48411 0.21568
Private house rent 0.03161 0.03161
Imputed rent 0.13401 0.13401
Public and personal services 0.25077 0.05007
Medical charges 0.01579 0.01541 Excluding massage fees
Airplane fares 0.00381 0.00381
Telephone charges 0.01745 0.01745
Hotel charges 0.01340 0.01340
Others 0.20031 0.00000
Eating-out 0.06773 0.00000
General 1.00000 0.31099

46
 Table 8. Comparison of CPI Movements in Hyogo and Tokyo in February
Unit: %
1992-94 1995 Difference Notes
(a) (b) (b)-(a)
Alcoholic beverages 0.0 15.1 ( -0.2 ) -15.1 ( -0.2 ) Beer, wine
Underwear 0.8 -14.0 ( -0.1 ) -14.8 ( -0.1 )
Japanese clothing -0.4 -11.4 ( -0.1 ) -11.0 ( 0.0 ) Toys and sporting goods
Recreational goods 2.8 -10.8 ( -0.3 ) -13.6 ( -0.3 )
Tutorial fees 0.0 -7.9 ( -0.1 ) -7.9 ( -0.1 )
Eating-out -0.1 -6.2 ( -0.5 ) -6.1 ( -0.5 )
Other clothing -0.7 -5.8 ( 0.0 ) -5.1 ( 0.0 ) Neckties, belts
Books and others -3.0 -5.5 ( -0.1 ) -2.5 ( 0.0 ) Newspapers, magazines
Cakes and candies -0.2 -4.8 ( -0.1 ) -4.6 ( -0.1 )
Domestic utensils 0.0 -4.6 ( 0.0 ) -4.6 ( 0.0 ) Tableware, kitchen utensils
Fish and shellfish -0.1 -3.8 ( -0.1 ) -3.7 ( -0.1 )
Personal effects 0.5 -3.4 ( 0.0 ) -3.9 ( -0.1 ) Bags, watches
Medical supplies -0.2 -3.3 ( 0.0 ) -3.1 ( 0.0 ) Disposable diapers
Personal care services -0.2 -3.3 ( 0.0 ) -3.1 ( 0.0 ) Men’s haircut charges
Medicines -0.2 -2.9 ( 0.0 ) -2.7 ( 0.0 )
Communication 0.0 -2.7 ( -0.1 ) -2.7 ( -0.1 ) Postage, telephone charges
Meat -0.7 -2.6 ( -0.1 ) -1.9 ( 0.0 )
Dairy products and eggs 0.6 -2.5 ( 0.0 ) -3.1 ( 0.0 )
Repairs and maintenance -0.2 -2.3 ( 0.0 ) -2.1 ( 0.0 )
Total ( -1.9 ) ( -1.9 )
Excl. fresh food ( -1.7 ) ( -1.7 )

Sources: Management and Coordination Agency, Consumer Price Index, Hyogo Prefecture Government, Hyogo-ken
no Shouhisha bukkasisuu sokuhou.
Notes: 1. The above figures are the difference between the rates of change from the previous year of CPI of Hyogo
prefecture and Tokyo Metropolitan-area.
2. Contribution to the percent change of the overall CPI the weights of Tokyo Metropolitan area in parentheses
3. The items for which figures exceed 2% in 1995 are listed in the above table.
4. Fresh food consists of fish and shellfish, meat, Dairy products and eggs.

47
 Table 9. Magnitude of Measurement Errors in CPI

Source of bias lower-bound mid-point upper-bound

Price index formula 0.00 0.00 0.25
Aggregation to item levels 0.00 0.10 0.25
Quality adjustment 0.30 0.70 0.90
Price sampling 0.05 0.10 0.60
Total 0.35 0.90 2.00

Table 10. Comparison with the Estimate in Boskin Report

United States Japan

Source of measurement error (Boskin Report) (our estimates)
Upper level substitution 0.15 0.00
Lower level substitution 0.25 0.10
New products / quality change 0.60 0.70
New outlets 0.10 0.10
Total 1.10 0.90
(0.80 - 1.60) (0.35 - 2.00)
Sources: Advisory Commission to Study the Consumer Price Index (1996)

48
 Table 11. Introduction of the CES Index

70-75 75-80 80-85 85-90 90-95 95-97 70-97 90-97

(Superlative indices)

Chained Törnqvist 11.052 6.194 2.534 0.994 1.137 0.849 4.053 1.055

Chained Fisher 11.052 6.194 2.534 1.001 1.137 0.831 4.053 1.049

(Chained CES indices)

s = 0.3 11.104 6.218 2.541 1.005 1.139 0.879 4.072 1.065

s = 0.4 11.078 6.197 2.539 1.000 1.135 0.865 4.060 1.058

s = 0.5 11.052 6.175 2.536 0.987 1.139 0.850 4.049 1.056

s = 0.6 11.026 6.154 2.533 0.982 1.127 0.852 4.037 1.049

s = 0.7 10.999 6.132 2.531 0.976 1.123 0.855 4.026 1.046

s = 1.0 (Geometric mean) 10.827 6.127 2.582 1.004 1.120 0.858 4.010 1.045

49
 Figure 1. Impact of the Appearance of New Goods and Services

Increase in demand
Items excluded from ô
Items excluded from the CPI survey Decrease in relative prices
the CPI survey

Decrease in demand
Items included in the Items included in the ô
CPI survey CPI survey Increase in relative prices

Figure 2. Date of Price Survey

(earliest case) (latest case)

SUN MON TUE WED THU FRI SAT SUN MON TUE WED THU FRI SAT
1 2 3 4 5 1 2 3 4
6 7 8 9 10 11 12 5 6 7 8 9 10 11
13 14 15 16 17 18 19 12 13 14 15 16 17 18
20 21 22 23 24 25 26 19 20 21 22 23 24 25
27 28 29 30 31 26 27 28 29 30 31

Note: Price quotations are collected during the shaded dates.

50
 Figure 3. Range of Quantitative evaluation

Pr i ce in d ex for mu l a

l Index formula for aggregating the

Substitution effects
upper level items
l Aggregation of individual prices into Within the range of
item index quantitative estimates

Q u a l i t y ch a n g e s Accuracy of price information

l Quality adjustment method

l Structural changes in retail markets
N e w g o o d s l Substitution among retailers and
brands
l Methodology of price survey

Without the range of

Technical problems
Accur a cy of we i g h t s quantitative estimate
in compilin g
methodology l Limited coverage of FIES
l Method in estimation of weight for
imputed rent for owner-occupied house

51
 Figure 4. Comparison of Various Price Index Formulae
(1970=1)
3.2
3.0
2.8
2.6
2.4
2.2
2.0
1.8
Fixed weight Laspeyres
1.6
Chained weight Tornqvist
1.4
1.2 Chained weight Fisher

1.0
0.8
70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97

Source: Author’s calculation from FIES and CPI Statistics.

52
 Figure 5. Distribution of the amount of floor space
(% ) (1) W o od en h ous es
50
wo o d en
wo o d en mid iu m s ize ho us e
40 s mall s ize
30 h

20 own hou s e
10 ho us e fo r rent

0
2 2 2 2 2 2
-29m 30-49m 50-69m 70-99m 100-149m 150m -

(% ) (2) Non-wood en h ous es

50

40 own hou s e
ho us e fo r rent
30

20

10

0
2 2 2 2 2 2
-29m 30-49m 50-69m 70-99m 100-149m 150m -

Sources: Management and Coordination Agency, The Housing Survey of Japan.

53
Figure 6. Impact of Productivity Increase Caused by Technological Innovation

General Price level

Aggregate Demand Curve
Aggregate Supply Curve

Productivity Increase Caused

by Technological Innovation

P’

Y Y’ GDP

54
 Figure A-1. Cost of Living Index and Laspeyres Price Index

Good x2

K
J

A
G

E
F
U(q) = u0
U(q) = u1
O Good x1
C B

Figure A-2 Cost of Living Index and Paasche Price Index

Good x2

M
L F E

U(q) = u0
H U(q) = u1
O Good x1
C B

55
 Figure A-3. Application of the Hedonic Approach to Specification Change
ln P

C
PN lnP = a + bX

^ A
PN
P rice C hange

Q uality C hange

^ B
PO

PO
D
b
¡ E stim ated P rice
a n A ctual P rice

XO XN X
O

56