CPI Mismeasurements and Their Impacts on Economic Management in Korea

CPI Mismeasurements and Their Impacts on Economic Management in Korea
CPI Mismeasurements and Their Impacts on Economic Management in Korea

Chul Chung
Korea International Trade
Association
Washington, DC 20036
USA
and Korea Institute for
International Economic Policy
Seoul 137-747, Korea
cchung@kita.net

John Gibson
Department of Economics
University of Waikato
Private Bag 3105
Hamilton, New Zealand
jkgibson@waikato.ac.nz

Bonggeun Kim
Department of Economics
Seoul National University
599 Gwanak-no, Gwanak-gu
Seoul 151-746, Korea
bgkim07@snu.ac.kr

CPI Mismeasurements and Their
Impacts on Economic Management
in Korea*

Abstract
We estimate the consumer price index (CPI) bias in Korea by
employing the approach of Engel’s Law as suggested by Hamilton
(2001). Using Korean panel data (Korean Labor and Income Panel
Study) and following Hamilton’s model with a non-linear specifica-
tion correction, our estimation result shows that the CPI bias over
the sample period (2000–05) averaged at least 0.7 percent annu-
ally, which implies that about 21 percent of the inflation rate dur-
ing the sample period can be attributed to the bias. This CPI bias
has caused a substantial understatement of the growth in real
GDP and contributes to excessive transfers from younger taxpay-
ers to the elderly through indexed pension payments. We discuss
the implications of the CPI bias for economic management and
policies in Korea.

1. Introduction

A consumer price index (CPI) measures the average price
of consumer goods and services purchased by households.
Inºation is measured by the percent change in the CPI,
which is frequently used to index wages, pensions, social
security payments, or long-term contracted prices. The
CPI is also used to convert nominal variables to real ones
such as real income and real economic growth by adjust-
ing for the effects of inºation so that those nominal vari-
ables can be compared over time at the price level of the
base year.

* We are grateful to Barry Bosworth, Prema-chandra Athukorala,
Iris Claus, and the participarts in the 2008 Asian Economic
Panel Meeting in Washington, DC, for their helpful comments.

Asian Economic Papers 9:1

Institute of Technology

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

CPI Mismeasurements and Their Impacts on Economic Management in Korea

It is crucial to measure the CPI precisely for the accurate evaluation of a country’s
economic status and for international comparison. When there is a bias in the CPI,
some macroeconomic variables can be miscalculated. Those miscalculated economic
indicators may not reºect true conditions of the economy and consequently they
may lead to imprudent economic policies possibly resulting in unintended transfers
from one group to another group in the economy. For example, when there is an up-
ward CPI bias and the public pension is indexed to the inºation rate, the govern-
ment will inadvertently transfer resources from the younger generation to the older
generation.

In this context, the U.S. Treasury welcomed the report by the U.S. Advisory Com-
mission to study the CPI, known as the Boskin Report (1996), because many govern-
ment social expenditures are tied to the inºation rate. The Boskin Report raised a
great deal of attention from academic and policymaker circles to the long-standing
debate of whether the cost-of-living index (COLI) should be the objective of measur-
ing a price index.1

The CPI has four biases inherent in its calculation. They include (1) substitution bias,
(2) quality change bias, (3) outlet substitution bias, and (4) new product bias.2 Be-
cause of these biases, the inºation rate derived from the (potentially biased) CPI
tends to be overestimated and the CPI may not measure the true COLI.3 The Boskin
Commission conªrmed this by reporting that there was an overstatement of the an-
nual inºation rate by about 1.1 percentage points. This implies that the U.S. real eco-
nomic growth rates must have been higher than those recorded. The Boskin Report
also calculated that the overstatement of inºation would add US$ 148 billion to the deªcit and US$ 691 billion to the national debt by 2006. Accordingly, since 1999, the
U.S. Bureau of Labor Statistics (BLS) has modiªed the CPI calculation method as
suggested by the Boskin Report. Nevertheless, there is an ongoing debate in the

1 According to Reinsdorf and Triplett (2009), the history of evaluation of the conceptual foun-
dations and methodologies of the CPI and price indexes in general dates back to more than
90 years.

2 The substation bias occurs when a Laspeyres type index does not recognize that consumers
make economizing substitutions away from items that have become more expensive than
they were in the base period when the CPI basket of goods was formed. The outlet bias is
similar, with shoppers responding to lower prices by switching outlets while price survey-
ors do not. The quality change and new goods bias reºect the difªculty of measuring the de-
clining cost of living when there are new and improved goods available, and this is accentu-
ated when there is delayed introduction of new goods into the basket.

3 Not everyone is convinced that CPI bias always overstates the true change of prices, at least
for some elements of the CPI. For example, Gordon and vanGoethem (2005) suggest that the
CPI for rental housing in the United States, which is the most important single category of
the overall CPI, has been downwardly biased for the last century.

Asian Economic Papers

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

CPI Mismeasurements and Their Impacts on Economic Management in Korea

United States and in other countries about the continuing extent of the CPI bias
problem.

This debate has been aided by a new approach to estimating the CPI bias using
Engel curves, which were introduced by Hamilton (2001). The basis of the approach
is that Engel’s Law shows the inverse relationship between food’s budget share and
household real income. Because food’s budget share may reºect real living stan-
dards, its movements can serve as an indicator of movements in real income. There-
fore, Hamilton estimates the degree to which the CPI is biased by comparing the im-
plied movements in real income that are inferred from movements in food’s budget
share with directly measured movements in real income that have nominal income
deºated by the CPI. This method has an advantage over other approaches for the
following three reasons. First, the relationship between food’s budget share and
household real income is one of the most established empirical regularities. Second,
Engel curves should not drift over time if consumer preferences are stable. Third,
any drift over time in the food Engel curve can be detected rather easily because as a
necessity, food’s demand is income inelastic and its budget share moves as real in-
come changes by much more than the budget shares of goods with unitary income
elasticities.

This paper builds on the work of Hamilton (2001) to investigate CPI mismeasure-
ments in Korea using the Korea Labor Income Panel Survey (KLIPS) data from
2000–05.4 Further, we modify the Almost Ideal Demand System, which is widely
used to estimate food demand, to a nonlinear functional form for more precise esti-
mation of the CPI bias. Our results show that the CPI upward bias in Korea is at
least 0.7 percent annually, which implies that about 21 percent of the inºation rate
during the sample period can be attributed to the bias. With our estimated bias, we
discuss its implications for economic management in Korea. For example, the CPI
bias has caused undue transfers from the younger generation to the elderly through
various indexed pension programs in Korea. We extend our discussion to incorpo-
rate changes in exchange rates. In doing so, we also conªrm Gibson’s (2008) ªnding
of a substantial CPI bias in a small open economy, which is likely to be more respon-
sive to trade shocks that result in volatile relative prices and a stronger substitution
bias than occurs in large economies like the United States. Finally, we discuss the
implications of the CPI bias for economic theories and policies related to business
cycles and we conclude with our suggestions.

4 Chung, Kim, and Park (2007) is a ªrst-ever attempt to estimate the Korean CPI bias. This pa-
per reªnes its estimation method and investigates the implications of the CPI bias for eco-
nomic management and policies in Korea.

Asian Economic Papers

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

CPI Mismeasurements and Their Impacts on Economic Management in Korea

The structure of the paper is as follows. Section 2 describes the food Engel curve es-
timation method. We describe the KLIPS data and provide empirical results in sec-
tion 3. Section 4 discusses implications of CPI bias for economic management in Ko-
rea, the relationship between being a small open economy and the size of the bias,
and the implications of CPI bias for economic theories on business cycles.

2. Engel curve method

We adopt an empirical framework used in Hamilton (2001), Costa (2001), and Gib-
son, Stillman, and Le (2008) to measure CPI bias from a food Engel curve estimated
on panel data. This method starts with the Leser-Working form of the Engel curve,
where the budget share is a linear function of the logarithm of real household in-
come and a relative price term:5

wi,j,t

(cid:2) (cid:3) (cid:4) (cid:5)(lnPF,j,t

(cid:6) lnPN,j,t) (cid:4) (cid:7)(lnYi,j,t

(cid:6) lnPj,t) (cid:4) X(cid:8)(cid:9) (cid:4) ui,j,t ,

(1)

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

where wi,j,t is the budget share of food for household i in region j and time period t,
PF,j,t, PN,j,t, and Pj,t represent the true but unobserved prices of food, non-food, and all
goods, Y is the household’s total income, X is a vector of individual household char-
acteristics, and u is the disturbance. The true cost of living is treated as a geometric
weighted average of food and non-food prices:

lnPj,t

(cid:2) (cid:10)

t lnPF,j,t

(cid:4) (1 (cid:6) (cid:10)

t)lnPN,j,t

In addition, it is assumed that prices of a good G (either food, non-food, or all
goods) are measured with error, which is assumed not to vary geographically,

lnPG,j,t

(cid:2) lnPG,j,0

(cid:4) ln(1 (cid:4) (cid:11)

G,j,t) (cid:4) ln(1 (cid:4) EG,t)

(2)

(3)

G,j,t represents the cumulative percentage increase in the true price

In equation (3), (cid:11)
of good G from period zero to period t and EG,t is the period-t percent cumulative
measurement error in the CPI since the base period. By inserting equations (2) and
(3) into (1), we have:

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

wi,j,t
(cid:4) (cid:5)[(cid:13)
F,j,t) (cid:2) (cid:12)

F,t

(cid:2) (cid:3) (cid:4) (cid:5)[(cid:12)
(cid:6) (cid:13)

N,t] (cid:6) (cid:7)(cid:13)
F,j,t, lnYi,j,t

F,j,t

(cid:6) (cid:12)
(cid:4) (cid:5)(pF,j,0

(cid:2) yi,j,t , ln(1 (cid:4) EG,t) (cid:2) (cid:13)

(cid:6) pN,j,0) (cid:6) (cid:7)pj,0

(cid:6) ui,j,t ,
G,t , lnPG,j,0

(cid:2) pG,j,0

N,j,t] (cid:4) (cid:7)[yi,j,t

(cid:6) (cid:12)

j,t] (cid:4) X(cid:8)(cid:9)

(4)

ln(1 (cid:4) (cid:11)

5 This functional form provides the basis of the Almost Ideal Demand System of Deaton and
Muellbauer (1980). Results when a quadratic in log income is used are also described subse-
quently.

Asian Economic Papers

CPI Mismeasurements and Their Impacts on Economic Management in Korea

An empirical version of equation (4) can be estimated if a database can be con-
structed from a panel household expenditure survey, and a temporal and cross-
sectional CPI for food, non-food, and all consumption:

w
i

j t
,

(cid:2)

(cid:3)

[

(cid:2)
(cid:3)
f g p
T
(cid:7)

d
D
t

(cid:4)
(cid:7)

F j t
,
,
J
(cid:3)

]
N j t
,
,
(cid:3)
D u
j

(cid:2)

j t
,

(cid:3)

b
y
[

j t
,

(cid:4)

]

j t
,

(cid:3)

(cid:5)
X (cid:6)

(5)

t and (cid:14)

where Dt is a dummy variable equal to 1 in period t, Dj is a dummy equal to 1 for re-
gion j, (cid:14)
the coefªcients of the omitted time and region dummies. The time dummy variables
are crucial to the measurement of CPI bias because

(cid:2)f is the intercept from equation (4), plus

j are their coefªcients, and

(cid:14)
t

(cid:2) (cid:5)((cid:13)

F,t

(cid:6) (cid:13)

N,t) (cid:6) (cid:7)(cid:13)

(6)

If the CPI bias in food and non-food is equal, then Hamilton (2001) notes that equa-
tion (6) can be reduced to:

(cid:8)
t

d
t(cid:9)
(cid:4)
b

(7)

That is, the cumulative percentage CPI bias at period t, Et, is given by a simple ratio
of estimated coefªcients from equation (5), 1 (cid:6) exp((cid:6)(cid:14)

t/(cid:7)).

3. Empirical analysis

3.1 Data
It is helpful to provide some background on how the CPI is constructed in Korea to
better understand any CPI mismeasurements.6 The ªrst CPI in Korea for the whole
country, named the Whole Country Retail Price Index, was released in April 1949.
The retail price index for Seoul was released earlier in 1936. Since 1990, the Korea
National Statistical Ofªce (KNSO) has produced and released the CPI in four differ-
ent categories, such as the basic category, which is the traditional benchmark case
classiªed by categories according to Classiªcation of Individual Consumption by
Purpose of the United Nations and three special categories, namely commodity
property, fresh food, and purchasing frequency. The CPI started to include owner-
occupied housing as a supplementary index in 1995.

The index was revised every ªve years to reºect the changes in consumption struc-
ture of urban households. In January 2002, the results of the revised index as of 2000

6 This section heavily depends on the information available at the KNSO’s web site:
www.nso.go.kr and the Korean CPI data can be downloaded from that source.

Asian Economic Papers

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

CPI Mismeasurements and Their Impacts on Economic Management in Korea

were released. In April 2003, the nationwide chain index as of 2002 was released as a
supplementary index. The indexes have been revised on a 5-year basis to reºect the
changes in consumption structure of urban households. In December 2006, the re-
sults of revised indexes were released with 2005 prices based.

Year 2005 is the current base year for the Korean CPI. The Korean CPI is produced
using prices of 489 items that have a higher than 1/10,000 share of the average ur-
ban household’s monthly expenditure, worth 185 Korean won. Agricultural, ªshery,
and livestock products are surveyed three times per month on a weekday of three
weeks, which contain the 5th, 14th, and 23rd day of the month. Meanwhile, infor-
mation on industrial products is collected once a month on two weekdays of the
week containing the 14th day of the month and service once a month on two week-
days of the week containing the 23rd day of the month.

KNSO ofªcials, who visit approximately 21,000 retail stores and 10,000 renting
households in 38 cities, conduct the survey. When some items have unusually high
or low prices due to natural disasters, a store closing sale, bulk trading, and so on,
they are excluded from the survey. The index is calculated on a monthly basis using
the Laspeyres method and released in the ªrst week of the next month.

The data used in this paper are drawn from KLIPS, which is an ongoing nationally
representative longitudinal household survey ªelded since 1998 by the Korea Labor
Institute. KLIPS collects data on an exhaustive list of individual and household
characteristics including detailed income and expenditure data. We use six rounds
of KLIPS data from 2000 to 2005,7 and combine these with the annual CPI for food
and non-food that is calculated for each of the 16 regions of Korea. We use a sample
of two-adult families, which are headed by a man, with or without children, where
the adults are between 20 and 65 years old, because more precise estimates of the
under-reporting parameter may be obtained by focusing on a fairly homogeneous
group. We drop the households which had experienced changes in their composi-
tion during the sample period to remove the effects of food consumption changes
due to newly added members or exits of original members. We restrict attention to
urban households, because measured food shares for rural households may be dis-
torted if the survey has difªculty in capturing consumption from own production,
which is likely to be more important in rural areas. The samples are further re-
stricted to those households whose food-at-home shares are in the 0.01–0.99 interval.
The resulting sample size is 5,134 households.

7 The collection of data on food expenditure at home starts only in 2000, so earlier waves of

KLIPS data cannot be used in this study.

Asian Economic Papers

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

CPI Mismeasurements and Their Impacts on Economic Management in Korea

Table 1. Descriptive statistics (KLIPS, 2000–05), obs.

5,134

Variable

w(Food Expenditure Share at Home)
Xres(Food Expenditure Share our of Home)
ln(Y)(Log Transformed Household Income)
ln(Y/P)(Log Transformed Household Real Income)
Age of Householder
Age of Spouse
Education Years of Householder
Education Years of Spouse
Yearly Hours of Work of Householder
Yearly Hours of Work of Spouse
2000 Sample
2001 Sample
2002 Sample
2003 Sample
2004 Sample
2005 Sample

Mean

0.1865
0.0258
17.1069
17.0317
41.40
38.40
12.68
11.86
2,609.98
1,254.46

0.2056
0.1891
0.1799
0.1474
0.1474
0.1306

S.D.

0.1164
0.0300
0.6243
0.6139
6.62
6.47
3.02
2.67
1,018.28
1,415.14

0.4015
0.3916
0.3842
0.3545
0.3545
0.3371

Min

0.0108
0
14.3461
14.3107
23
20
0
0
0
0
0
0
0
0
0
0

Max

0.96
0.5
19.8557
19.7808
65
65
27
25
8,400
8,400
1
1
1
1
1
1

The dependent variable is the budget share for food consumed at home, and control
variables include real total income (deºated by the CPI with a 2005 average base),
relative food price changes, and demographic, educational, and employment char-
acteristics. The model also includes the budget share for food out of the home. This
form of consumption is not part of the dependent variable because it is assumed
that restaurant meals are not perfect substitutes for food-at-home. Ideally, the substi-
tution possibilities between restaurants and home cooking would be captured by in-
cluding the relative price of restaurant meals, but this is not available. Therefore, we
follow the practice in the literature that uses Engel curves to measure CPI bias, and
we use the budget share for restaurant meals as an explanatory variable in place of
the required price.

A description of the dependent and explanatory variables is shown in Table 1. The
dependent variable, which is the expenditure share for food consumption at home,
averages 18.6 percent for the sample period. Control variables include relative food
price changes, demographic and educational characteristics, hours of work, and the
expenditure share for food out of home. The share of food out of home averages
2.6 percent. Reported real total household income including labor income and
ªnancial income averages 3,400 million Korean won, which is approximately equal
to US$ 30,000 in 2003. On average, the household head is 41.4 years old and has 12.7 years of schooling whereas the spouse has 1 year less and is about 3 years younger. To show how our main variables like food shares and household incomes have changed over time, the beginning-, middle-, and end-period averages of those vari- ables are reported in Table 2. The ªrst row of Table 2 shows that the average food-at- 7 Asian Economic Papers l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / e d u a s e p a r t i c e – p d / l f / / / / / 9 1 1 1 6 8 2 9 9 1 a s e p 2 0 1 0 9 1 1 p d . . . . . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 CPI Mismeasurements and Their Impacts on Economic Management in Korea Table 2. Trend of main variables over time (KLIPS, 2000–05), obs. 5,134 Variable w (Food Expenditure Share at Home) Xres (Food Expenditure Share out of Home) ln(Y) (Log Transformed Household Income) ln(Y/P) (Log Transformed Real Household Income P,t (Consumer Price Index) PFj,t (Food CPI) PN,,t, (Non-food CPI) 2000 0.2254 0.0264 16.885 16.885 1 1 1 2003 0.1776 0.0236 17.176 17.076 1.105 1.154 1.087 2005 0.1553 0.0241 17.319 17.157 1.176 1.284 1.146 home share in Korea fell by about 7 percentage points from 22.5 percent in 2000 to 15.5 percent in 2005. Over the same period, the CPI increased by 17 percent, nominal household income grew by 63 percent, and its real value adjusted by the CPI grew about 40 percent. This fall in the food share is relative to the measured growth in real income, which is consistent with the existence of a substantial CPI bias in Korea. Figure 1 illustrates the persistent downward shift of Engel curves over the time. We attribute this downward shift to understated real income, which in turn is due to the overstated CPI. 3.2 Estimation results The method we use, which follows several previous studies, is critically dependent on the assumption of stable consumer preferences over time. To provide some assur- ance that this assumption is reasonable we adopt several precautionary procedures, which advance those used in previous studies. First, we drop the households who had experienced changes in their composition during the sample period to remove the effects of food consumption changes due to newly added members or exits of original members.8 Second, we report results both in a linear and a quadratic form of log real income models. If the relationship is non-linear, then the linear model es- timates in the previous studies will be biased due to an omitted variable bias and the resulting relationship could be time-varying due to the different level of real in- come over time. Third, because our data are a genuine panel we avoid potentially time-varying relationships caused by a different sample composition in each year (with potentially different consumption tastes), which could result if we had fol- lowed a common strategy in this literature of relying on repeated cross-sectional samples. Equation (5) is estimated for a sample of two-adult families, with or without chil- dren, where the adults are between 20 and 65 years old and several other sample se- 8 This restriction removed approximately 10 percent of the sample. 8 Asian Economic Papers l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / e d u a s e p a r t i c e – p d / l f / / / / / 9 1 1 1 6 8 2 9 9 1 a s e p 2 0 1 0 9 1 1 p d . . . . . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 CPI Mismeasurements and Their Impacts on Economic Management in Korea Figure 1. Downward shift of Engel curve (KLIPS) l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / e d u a s e p a r t i c e – p d / l f / / / / / 9 1 1 1 6 8 2 9 9 1 a s e p 2 0 1 0 9 1 1 p d . . . . . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 lection criteria are adopted here such as only using households where the food bud- get share is in the interval 0.01–0.99. These restrictions are similar to those employed by Hamilton and other studies. The Engel curve relationship should hold for any group of people properly controlling for taste variables and thus a better estimate of CPI bias can be obtained by focusing on a fairly homogeneous group. Furthermore, we also use other control variables to exclude the other sources of the shift of Engel curves over time. Table 3 contains the results of estimating equation (5) in both a linear and a qua- dratic form of log real income. Because our data are a genuine panel we can use household ªxed effects regression to control for time-invariant unobserved hetero- geneity in consumption preferences across the households. The negative coefªcients on real income indicate that food budget shares fall as households become richer, which means that food’s income elasticity is less than one and this is precisely why food is used as the indicator good here. If the relationship is non-linear, then our lin- ear model estimates will be biased due to an omitted variable bias, and in fact, the linear relationship between log real income and food share is strongly rejected in our data, so we mainly focus on the quadratic model in Table 3. The estimation results indicate a persistent and substantial downward drift in the food Engel curves. We attribute this drift to unmeasured growth in real expenditures. 9 Asian Economic Papers CPI Mismeasurements and Their Impacts on Economic Management in Korea Table 3. Food Engel curve estimations of Korea (KLIPS, 2000–05), obs. 5,134 Variable Intercept Log (Real Household Income) (Log (Real Household Income))2 Log (Food CPI/Non-food CPI) Number of children under 15 years old in the household Education Years of Householder Education Years of Spouse Yearly Hours of Work of Householder Yearly Hours of Work of Spouse Food Expenditure Share out of home 2001 Year Dummy 2002 Year Dummy 2003 Year Dummy 2004 Year Dummy 2005 Year Dummy 2001 Cumulative CPI bias 2002 Cumulative CPI bias 2003 Cumulative CPI bias 2004 Cumulative CPI bias 2005 Cumulative CPI bias R2 Linear 2.8340 (0.0776)c (cid:6)0.1544 (0.0037)c — 0.1744 (0.1497) 0.0036 (0.0034) 0.0009 (0.0025) (cid:6)0.00004 (0.0025) (cid:6)3.95e(cid:6)09 (1.58e(cid:6)09)b (cid:6)2.99e(cid:6)09 (1.38e(cid:6)09)b 0.0883 (0.0503)a (cid:6)0.0112 (0.0043)c (cid:6)0.0212 (0.0078)c (cid:6)0.0231 (0.0098)b (cid:6)0.0397 (0.0169)b (cid:6)0.0415 (0.0176)b 0.070 0.128 0.139 0.226 0.235 0.4132 Quadratic 15.3146 (0.6875)c (cid:6)1.6256 (.0806)c 0.0431 (0.0023)c 0.3270 (0.1431)c 0.0024 (0.0032) 0.0019 (0.0024) (cid:6)0.0017 (0.0024) (cid:6)4.48e(cid:6)10 (1.52e(cid:6)09) (cid:6)1.91e(cid:6)09 (1.32e(cid:6)09) 0.1375 (0.0480)c (cid:6)0.0133 (0.0041)c (cid:6)0.0289 (0.0074)c (cid:6)0.0338 (0.0093)c (cid:6)0.0596 (0.0162)c (cid:6)0.0622 (0.0168)c 0.008 0.017 0.020 0.036 0.037 0.4664 Note: a. Statistically signiªcant at the 10 percent level. b. Statistically signiªcant at the 5 percent level. c. Statistically signiªcant at the 1 percent level. Results on region dummies are omitted. We have no reason to believe that the nominal income estimates from the KLIPS are becoming increasingly understated in later years of the sample period. Hence, this miscalculation of real expenditures is attributed to CPI bias. We estimate that over the 6 years from 2000 to 2005, the cumulative bias in the Korean CPI is be- tween 0.037 and 0.235, depending on the estimation method. Both of these esti- mates are statistically signiªcantly different from zero. This range corresponds to an average annual bias of between 0.7 and 4.5 percentage points. The lower and more reliable values come from the non-linear estimates. Based on the lower value, at least one-ªfth of the measured rise in the cost of living can be attributed to CPI bias. 10 Asian Economic Papers l D o w n o a d e d f r o m h t t p : / / d i r e c t . m i t . / e d u a s e p a r t i c e – p d / l f / / / / / 9 1 1 1 6 8 2 9 9 1 a s e p 2 0 1 0 9 1 1 p d . . . . . f b y g u e s t t o n 0 8 S e p e m b e r 2 0 2 3 CPI Mismeasurements and Their Impacts on Economic Management in Korea 4. Implications of CPI bias for economic management and theories 4.1 CPI bias on economic management in Korea It is crucial to measure the CPI precisely for the accurate evaluation of a country’s economic progress. When there is an upward bias in the CPI, some macroeconomic variables, such as real GDP as an indicator of living standards, can be understated. Those mismeasured economic indicators may not reºect true conditions of the econ- omy, and consequently, they may lead to imprudent economic policies. Examples in- clude ineffective inºation targeting for monetary policy and possibly social payment programs that result in unintended and undue transfers from one group to another group in the economy. For monetary policy, if there are overstated changes in prices, policymakers are more often forced to pursue tight policies to reduce the rate of inºation. Indeed the possibility of a CPI bias is one of several reasons for inºation-targeting central banks to have a positive rather than zero target rate of inºation. However, because there have not been previous measures of the CPI bias for Korea, it has not been possible to debate whether monetary policy efforts to reduce measured CPI inºation have been excessive and might induce undue sacriªces like a higher unemployment rate, at least according to the view from the short-run Phillips curve. For public transfers, when there is an upward CPI bias and the public pension is in- dexed to the inºation rate, the government will inadvertently transfer resources from the younger generation to the elderly. Currently, measured inºation (as the change in the CPI) is used to index pension payments in Korea. In 2005, the various pension payments in Korea are up to US$ 9 billion per year and the share of GDP for
these payments is rising rapidly due to the large baby boom generation, which, as in
the United States, is reaching retirement age. In fact, the total pension payment is ex-
pected to double by 2015. For the next 5 years, the estimated bias in the Korean CPI
will induce undue pension payments of more than 0.01 percent of GDP annually.

4.2 CPI bias in a small open economy
The GDP deºator measured by the ratio of nominal GDP to real GDP provides dif-
ferent information about the change in true price levels in the economy. In general,
CPI as a Laspeyres index is likely to overstate the true increase in the cost of living
and the GDP deºator, such as a Paasche index, is likely to understate it. One of the
key differences is that the GDP deºator does not include the imported goods con-
sumed by consumers. For the last decade, imports into Korea from China have in-
creased immensely, especially in agricultural products. These imports have had sub-
stantial impacts on the Korean economy. In particular, the prices of food products

Asian Economic Papers

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

CPI Mismeasurements and Their Impacts on Economic Management in Korea

have been maintained at low levels due to cheap agricultural products from China
and the imported goods might bring deºation for some products as in the phenome-
non of Japan (Broda and Weinstein 2007). Here, we extend our discussion to incor-
porate changes in exchange rates into the analysis. In doing so, we investigate the
relationship between the size of CPI bias and whether an economy is small and
open, and hence, more responsive to trade shocks, which results in volatile relative
prices and a larger substitution bias in the CPI than for that of large economies like
the United States. Since 2000, the Korean CPI has increased by 23.5 percent and GDP
deºator increased by only 11.2 percent.9 The pattern conªrms the predicted biases
from the Laspeyres and Paasche indexes, but the size of the difference is much
greater than those of other large countries like the United States. This large differ-
ence also justiªes our discussion on CPI bias in a small open economy.

As a small open economy, Korea often faces large variations in exchange rates.
When there are some discrepancies between the prices of imported goods (which
are highly responsive to varying exchange rates) and local goods, consumer-
purchasing decisions in the market may change faster than changes occur in the bas-
ket of goods used in the calculation of the CPI. Hence, these discrepancies could be
a source of Korea’s substantial CPI bias. Indeed, the size of the CPI bias appears to
be sensitive to having real income adjusted by the exchange rate, with a consider-
able fall in the calculated bias (Table 4). As in Table 4, when we re-estimate equa-
tion (5) with real income adjusted by the changes in exchange rates, the cumulative
bias in the Korean CPI during the sample period is 0.028 and 0.15, depending on the
estimation method. Results indicate approximately one-third of the CPI bias can be
attributed to some discrepancies between the changes in imported goods and prices
and the changes in goods and prices used in formation of CPI. This conªrms Gib-
son’s (2008) ªnding of a substantial CPI bias in a small open economy like New Zea-
land, which is likely to be more responsive to trade shocks and have a stronger sub-
stitution bias than that of large economies like the United States.

4.3 Implications of CPI bias on economic theories of business cycles
If the estimated CPI bias in the previous section is to overstate the true increases in
prices, then the growth of real GDP is to be understated. Analogously, real wage
growth is also to be understated. Therefore, we consider implications of this under-
stated real wage growth on economic theories, which play an important role in de-
termining appropriate economic policies for business cycles.

9 For the three decades before 2000, GDP deºator increased more rapidly than CPI and the

pattern is reversed after 2000. This interesting pattern needs an explanation, and we plan to
investigate it in an extended work.

Asian Economic Papers

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

CPI Mismeasurements and Their Impacts on Economic Management in Korea

Table 4. Food Engel curve estimations of Korea with real income adjusted by exchange rates
(KLIPS, 2000–05), obs.

5,134

Linear

Quadratic

Adjusted real income
by exchange rate

2001 Cumulative CPI bias
2002 Cumulative CPI bias
2003 Cumulative CPI bias
2004 Cumulative CPI bias
2005 Cumulative CPI bias

0.070
0.128
0.139
0.226
0.235

0.085
0.099
0.085
0.145
0.151

Note: The share of imported goods is assumed to be 0.5.

0.008
0.017
0.020
0.036
0.037

0.009
0.014
0.014
0.027
0.028

As an example of the type of economic theories that can be affected by the under-
stated real wage growth, we can consider a ªxed effects model of earnings transi-
tions, as in the following:10

(cid:2)

P
it

(cid:3)

T
it

(cid:4)

(cid:2)

(

(cid:3)

2
it

)

(cid:3)

b
(

(cid:3)

(cid:8)
it

)

(cid:4)

(8)

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

i represents the combined effect of time-invariant characteristics of

P is the permanent
T is the transitory component, which can be affected by a busi-

where yit is the ith worker’s log real annual earnings in year t, y it
component, and y it
ness-cycle or just individual speciªc transitory events, and pt is log CPI in year t, the
ªxed effect (cid:10)
worker i, Xit is worker i’s years of work experience as of year t, Ut is a business-cycle
indicator such as the unemployment rate or real GDP, and (cid:13)
it is an individual transi-
tory ºuctuation. If we are interested in the transitions of real annual earnings due to
either the permanent factor or the transitory factor, then we may ªrst-difference
equation (5) to get

(cid:15)yit

(cid:2) (cid:14)

(cid:4) (cid:14)

1Xit

(cid:4) (cid:7)(cid:15)Ut

(cid:4) (cid:15)(cid:13)

it .

(9)

When interested in whether earnings vary counter-cyclically, non-cyclically, or pro-
cyclically with the business cycle, one can investigate the sign of (cid:7). However, in-
stead of true real wage growth, we use dependent variables with non-classical er-
rors-in-variables due to the CPI bias like:

D
y

*
it

(cid:2)

D
y

p
it

(cid:3)

D
y

t
it

(cid:4)

D
p

*
t

(cid:2)

D
y

p
it

(cid:3)

D
y

t
it

(cid:4)

p
t

(cid:4)

(cid:8)
t

(cid:2)

D
y

(10)

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

where (cid:12)
t to period t (cid:4) 1 and (cid:13)

t represents the cumulative percentage increase in the true price from period

t is the percent cumulative measurement error in the CPI.

10 This ªxed effect model follows the speciªcation used in Kim and Solon (2005).

Asian Economic Papers

CPI Mismeasurements and Their Impacts on Economic Management in Korea

What does this imply for the estimation of cyclicality in real earnings? Substituting
equation (10) into equation (9) yields

D
y

*
it

(cid:2)

l d
(
0

(cid:3)

)

(cid:3)

(cid:8) .

(11)

The coefªcient of (cid:15)Ut is not the original wage cyclicality parameter (cid:7), but rather (cid:7)
rescaled by the CPI bias parameter (cid:16). More broadly, the mean reversion in depend-
ent variables like our model will tend to make estimated regression coefªcients too
small in magnitude, which is contrary to the textbook case where errors in the de-
pendent variable cause no bias in slope coefªcients. This (mean-reverting) measure-
ment error in real wage growth may lead to as much as a 25 percent underestima-
tion of pro-cyclicality of real wages. This result could misguide policymakers in
their understanding of business cycles and might lead them to an inadequate choice
of economic policies for business cycles.

4.4 Policy suggestions
To improve the quality of the Korean CPI, KNSO may consider what the BLS of the
United States has done already. Subsequent to the Boskin Report, the BLS intro-
duced methodological changes that can address the substitution, quality, and new
goods issues. According to Johnson, Reed, and Stewart (2006, p. 10),

These include the following: 1) the introduction of the geometric means formula
to account for lower-level substitution, 2) the introduction of the Chained Con-
sumer Price Index for All Urban Consumers (C-CPI-U) to provide an index that
accounts for upper-level substitution, 3) expansion of the use of hedonic models
to improve the measurement of quality change, and 4) the institution of proce-
dures to introduce new goods into the index more quickly by more frequent up-
dates to the item samples.

The estimates of CPI bias for Korea reported in the current paper, and the various
economic implications of this bias, suggest that it is a matter of some urgency for
improvements to be made to the Korean CPI so that it provides a better basis for
evaluating economic progress in Korea.

References

Boskin, Michael J., Ellen R. Dulberger, Robert J. Gordon, Zvi Griliches, and Dale W. Jorgenson.
1996. Final Report of the Advisory Commission to Study the Consumer Price Index. Washington, DC:
U.S. Government Printing Ofªce.

Broda, Christian, and David E. Weinstein. 2007. Exporting Deºation? Chinese Exports and Jap-
anese Prices. Working paper. Chicago, IL: University of Chicago.

Asian Economic Papers

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3

CPI Mismeasurements and Their Impacts on Economic Management in Korea

Chung, Chul, Bonggeun Kim, and Myungho Park. 2007. CPI Bias in Korea. Journal of Interna-
tional Economic Studies 11 (2):261–284.

Costa, Dora L. 2001. Estimating Real Income in the United States from 1888 to 1994: Correcting
CPI Bias Using Engel Curves. Journal of Political Economy 109 (6):1288–1310.

Deaton, Angus S., and John Muellbauer. 1980. An Almost Ideal Demand System. American Eco-
nomic Review 70 (3):312–326.

Gibson, John K. 2008. Using Engel Curves to Estimate CPI bias in a Small, Open, Inºation-
Targeting Economy. Working paper. Hamilton: University of Waikato.

Gibson, John K., Steven Stillman, and Trinh Le. 2008. CPI Bias and Real Living Standards in
Russia During the Transition. Journal of Development Economics 87 (1):140–160.

Gordon, Robert J., and Todd vanGoethem. 2005. A Century of Housing Shelter Prices: How Big
is the CPI Bias? NBER Working Paper No. 11776. Cambridge, MA: National Bureau of Eco-
nomic Research.

Hamilton, Bruce W. 2001. Using Engel’s Law to Estimate CPI Bias. American Economic Review
91 (3):619–630.

Johnson, David S., Stephen Reed, and Kenneth J. Stewart. 2006. Price Measurement in the
United States: A Decade After the Boskin Report. Monthly Labor Review 129 (5):10–19.

Kim, Bonggeun, and Gary R. Solon. 2005. Implications of Mean-Reverting Measurement Error
for Longitudinal Studies of Employment and Wages. Review of Economics and Statistics 87
(1):193–196.

Korea National Statistical Ofªce. www.nso.go.kr

Reinsdorf, Marshall, and Jack E. Triplett. 2009. A Review of Reviews: Ninety Years of Profes-
sional Thinking about the Consumer Price Index and How to Measure It. In Price Index Con-
cepts and Measurement, Studies in Income and Wealth, vol. 70, edited by W. Erwin Diewert,
John Greenlees, and Charles R. Hulten, 17–84. Chicago, IL: The University of Chicago Press for
the National Bureau of Economic Research.

Asian Economic Papers

D
o
w
n
o
a
d
e
d

f
r
o
m
h

t
t

:
/
/

d
i
r
e
c
t
.

i
t
.

e
d
u
a
s
e
p
a
r
t
i
c
e
–
p
d

f
/

9
1
1
1
6
8
2
9
9
1
a
s
e
p
2
0
1
0
9
1
1
p
d

b
y
g
u
e
s
t

o
n
0
8
S
e
p
e
m
b
e
r
2
0
2
3 CPI Mismeasurements and Their Impacts on Economic Management in Korea image

Download pdf