American Economic Journal: Macroeconomics 2010, 2:1, 207–223

http://www.aeaweb.org/articles.php?doi=10.1257/mac.2.1.207

Development Accounting†

By Chang-Tai Hsieh and Peter J. Klenow*

Researchers have made much progress in the past 25 years in

accounting for the proximate determinants of income levels: physi-
cal capital, human capital, and Total Factor Productivity (TFP).
But we still know little about why these factors vary. We argue that
TFP exerts a powerful influence on output not only directly, but also
indirectly, through its effect on physical and human capital accu-
mulation. We discuss why TFP varies across countries, highlighting
misallocation of inputs across firms and industries as a key determi-
nant. (JEL E22, E23, F21, F35, O10, O40)

R esearch on income differences can arguably be classified into one or more

arrows in the following chain of causality:

Geography, Climate, Luck ⇒ Human Capital,  Physical Capital,  TFP ⇒ Income

⇓
Institutions, Culture ⇒ Human Capital, Physical Capital, TFP ⇒ Income
⇓
Policies, Rule of Law, Corruption ⇒ Human Capital, Physical Capital, TFP ⇒ Income

Our focus is on the right-most arrows, or what is sometimes called “development

accounting.” First, we describe research from the past 25 years about the proximate
role of physical capital, human capital, and TFP in accounting for income differ-
ences across countries. The current state of the debate is as follows: human capital
is important (accounting for 10–30 percent of country income differences), physical
capital also matters (accounting for about 20 percent of country income differences),
and residual TFP remains the biggest part of the story (accounting for 50–70 percent
of country income differences).
Second, we will contend there are important positive feedback effects between
human capital, physical capital, and TFP. In particular, the level of TFP of different
sectors (investment versus consumption, human capital versus final goods) can influ-
ence the incentive to accumulate physical and human capital. We will also argue that
a key determinant of aggregate TFP is the efficiency of input allocation across firms
and industries.

* Hsieh: Graduate School of Business, University of Chicago, 5807 South Woodlawn Avenue, Chicago, IL
60637 and National Bureau of Economic Research (e-mail: [email protected]); Klenow: 579 Serra
Mall, Stanford University, Stanford, CA 94305 and the National Bureau of Economic Research (e-mail: pete@
klenow.net). We gratefully acknowledge the financial support of the Kauffman Foundation, the Alfred P. Sloan
Foundation (Hsieh), and the Stanford Institute for Economic Policy Research (Klenow). We thank Chad Jones and
David Romer for detailed comments that vastly improved the paper.
†
To comment on this article in the online discussion forum, or to view additional materials, visit the articles
page at: http://www.aeaweb.org/articles.php?doi=10.1257/mac.2.1.207.

207
208 AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2010

Along the way, we will highlight some of the many open questions that remain.
In a short paper such as this, however, many important topics must be left out. In the
interest of brevity, this survey neglects the work that has been done on any of the
left-most or vertical arrows in the simple schematic laid out at the start of this paper.
That is, we neglect the growing and important literature on how geography and
climate affect institutions and culture, which, in turn, determine policies (e.g., tax
and tariff rates), corruption, and ultimately physical capital, human capital, and TFP.

I.  Proximate Development Accounting

First and foremost, the reader should see the survey by Francesco Caselli (2005).
Ours is a comparatively quick overview of this large literature. Second, we wish to
clarify the two main ways the literature has done proximate development account-
ing. Imagine the simple aggregate production function

(1) Yi  =  Ai ​K​iα​  ​ (hi L i )1−α,

where Yi represents real gross domestic product (GDP) in country i, Ai is resid-

ual TFP, Ki is real physical capital, hi is human capital per person, and Li is hours
worked per person. Dividing equation (1) by population Ni, and rearranging, yields a
conventional expression for accounting:

α
hi Li 1−α
(2) ​ __i  ​   =  Ai  a__   ​ a​ ____
Y K
​  i ​​ b​   ​​ b ​  ​ .
Ni Ni Ni

If one takes logs of both sides of equation (2), one can then linearly decompose the
differences in income levels between any two countries (say each country versus
the United States, or the ninetieth versus the tenth percentile of country incomes).
Alternatively, one could do a variance decomposition using a sample of many coun-
tries. A capital elasticity of about one-third is typically assumed based on observed
labor shares, following Douglas Gollin (2002). Compared to regression studies, this
accounting approach need not require that, say, TFP be orthogonal to physical or
human capital.
Any accounting provides the answer to a specific question. Accounting with equa-
tion (2) asks the hypothetical question: how much would output per person increase
in response to variation in one of the following factors: physical capital per person,
effective labor per person, or residual TFP, holding the other two factors fixed.
One objection to accounting with equation (2) is that physical capital per
person will endogenously increase in response to increases in effective labor
or TFP. Because investments in physical capital are final goods, unlike human
capital or TFP, any increase in output will tend to bring forth higher physical
capital. Put differently, holding fixed physical capital per person, while increas-
ing human capital or TFP, requires a decrease in the investment rate in physical
capital. It is not obvious why this is a useful thought experiment given that the
investment rate in physical capital is presumably driven by factors such as the
VOL. 2 NO. 1 HSIEH AND KLENOW: DEVELOPMENT ACCOUNTING 209

effective tax rate on capital income and the relative price of capital, but not the
level of human capital or TFP, per se.
An alternative accounting in the literature asks a different question by rearrang-
ing the production function into an intensive form. For example,

K α/(1−α)  ____
(3) ​ __i  ​   =  ​A ​i​ 
Y 1/(1−α) __ h   Li
​ a​  i ​​ b​ 
  ​a​  i  ​  b .
Ni Yi Ni

Here, the thought experiment is a change in effective labor per person or resid-
ual TFP, allowing capital per person (but not the capital-output ratio) to change in
response. As pointed out by N. Gregory Mankiw, David Romer, and David N. Weil
(1992), this question is compatible with the steady state of a neoclassical growth
model in which the level of human capital or TFP has no direct effect on the steady
state capital-output ratio. For comparing large, persistent differences across coun-
tries, a steady-state assumption may be a good approximation. The bigger exponents
on residual TFP (i.e., 1/(1−α) instead of 1) and on effective labor input (1 rather than
1 − α) in equation (3) reflect the impact of these variables on output both directly
and indirectly through capital per worker.
A valid objection to accounting with equation (3) is its asymmetry. It asks how
much output per person differs when the capital-output ratio changes, holding fixed
the levels of human capital and residual TFP. One could, instead, argue that physical
capital is an important input to accumulation of human capital and investments in
higher TFP. In this spirit, Mankiw, Romer, and Weil (1992) carried out accounting
that incorporated an effect of physical capital on human capital. This channel is
weaker, however, if investments in human capital and TFP are intensive in human
capital more than physical capital. And, in contrast to the well-understood endogene-
ity of physical capital in the neoclassical growth model, the determinants of human
capital and TFP are much less well understood.1 Still, we will describe the results of
accounting with both equations (2) and (3), and then return to the issue of what may
be driving differences in human capital and TFP.

A. Physical Capital

Using equation (3), Klenow and Rodríguez-Clare (1997) and Robert E. Hall and
Charles I. Jones (1999) attribute about 20 percent of variation in income per worker
to variation in capital-output ratios. Their variance decomposition evenly splits the
covariance terms between any two of the following: physical capital, human capital,
and TFP. Caselli (2005) uses equation (2), which gives more weight to physical capi-
tal, but also looks at relative variances rather than assigning the covariance terms.
These differences roughly offset each other so that he also credits about 20 percent
of income variation to physical capital. Despite this broad agreement, open questions
remain about measuring the quality of physical capital across countries given pos-
sibly differing vintage (e.g., Roc Armenter and Amartya Lahiri 2006) and efficiency

1
See Klenow and Andrés Rodríguez-Clare (2005) and Juan Carlos Córdoba and Marla Ripoll (2008) for
models in which TFP endogenously responds to the levels of human and physical capital per worker.
210 AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2010

of the component due to government infrastructure investments (e.g., Lant Pritchett

2000).

B. Human Capital

To measure levels of human capital, Mankiw, Romer, and Weil (1992) used the
secondary school enrollment rate as a proxy for the investment rate in human capital.
They attributed about 50 percent of income differences to human capital differences
in their 1985 sample of 98 nonoil countries. But primary and tertiary schooling must
matter as well. And attainment of the workforce—as opposed to the enrollment
rate of the school-age population—should be what matters for the human capital of
workers.
Klenow and Rodríguez-Clare (1997) use years of schooling attainment from
Robert J. Barro and Jong-Wha Lee (1993), which covers primary, secondary, and
tertiary schooling. They also propose that a useful way to capture the impact of
attainment is to look at how wages vary with schooling within countries. Based
on evidence in the labor literature, they assume a Mincerian log linear relationship
between years of schooling and human capital:

hit   =   Bit ​e​ φit Sit​ .

Here, Sit is years of schooling in country i in year t; φit is the “Mincerian return” to
a year of schooling; and Bit captures factors such as the quality of schooling, human
capital accumulated in early life, and human capital accumulation on the job. The
Mincerian return may also be affected by the quality of schooling or, for that matter,
early childhood stimulation, nutrition, and so on.
Given data on Sit , we need to know φit and Bit to implement the Mincerian
approach to estimate human capital across countries. Klenow and Rodríguez-Clare
(1997) impose φit = 0.095 for all country-years, and allow Bit to vary across coun-
tries as a function of estimated teacher human capital and differences in capital-
output ratios. If human capital production is as intensive in physical capital as the
production of other goods, they find human capital differences explain about 30
percent of income differences. If, instead, they put more weight on student time and
teacher human capital, and less weight on physical capital inputs, they find human
capital differences explain only about 10 percent of income differences.
Caselli (2005) goes further in incorporating nonstudent inputs in measuring
human capital, including teacher-pupil ratios, human capital of teachers, human
capital of parents, classroom materials per student, experience, and even health and
nutrition.2 He finds that such factors could, in principle, allow human capital to
explain the full 80 percent of income variation not attributable to physical capital.
But each time he finds that the required elasticity of human capital with respect to
these inputs (and/or the amount of variation in these inputs) is much larger than the

2
See Weil (2007) for creative use of micro evidence to try to measure the impact of health differences on
income differences across countries.
VOL. 2 NO. 1 HSIEH AND KLENOW: DEVELOPMENT ACCOUNTING 211

limited micro evidence available. He concludes that human capital explains between
10 percent and 30 percent of income variation.
Many estimates, including those by Klenow and Rodríguez-Clare (1997), Lutz
Hendricks (2002), and Caselli (2005), explicitly attempt to incorporate differences
in the quality of schooling that do not show up in the Mincerian return. That not-
withstanding, recent papers by Rodolfo Manuelli and Ananth Seshadri (2007);
Andrés Erosa, Tatyana Koreshkova, and Diego Restuccia (2007); and Benjamin F.
Jones (2008) emphasize the shortcomings of using the Mincerian approach alone to
infer variation in human capital across countries. Imagine the log of human capital
(of individual or country i ) is a function of years of schooling si and schooling inputs
xi : ln hi  =   f   (si ,  xi ). Then the Mincerian return to schooling is

∂ f ∂ f ___∂ x

​ _____    =   ​ ___  ​   +  ​ ___  ​ 
d ln hi
 
 ​    ​  i ​   
.
dsi ∂ si ∂ xi ∂ si

Micro-Mincer regressions that run across individuals within a country should cap-
ture not only the direct effect of a higher quantity of schooling, but also the indirect
effect of the higher quality of schooling of those individuals who tend to get more
education within countries. But the macro-Mincer relationship that growth econo-
mists are interested in can be different. The quality of schooling may covary more
with schooling attainment across countries than it does within countries. This is what
happens in Manuelli and Seshadri (2007) and Erosa, Koreshkova, and Restuccia
(2007) because TFP in producing school inputs is higher in rich countries, but all
individuals within countries face the same price of school inputs.3 They posit lower
prices of school inputs (relative to the wage) in rich countries because some school
inputs are books, equipment, and buildings. Whereas the micro-Mincer semi-elas-
ticity may be 10 percent, the macro-Mincer semi-elasticity may be 20 to 30 percent.
These papers have the potential to overturn the conventional wisdom described
here, that human capital explains only 10–30 percent of income differences.
Open questions remain about this approach, namely estimating key parameters
for human capital production at home, at school, and on the job—including the
importance of nonteacher inputs in human capital accumulation. But the appar-
ently large wage gains to immigrants from poor to rich countries, as documented by
Hendricks (2002), are damaging to the view that human capital varies so much that
there is little residual variation in TFP across countries. We return to some of this
work when we ask why human capital is higher in rich countries.

C. Hours Worked per Person

A much smaller literature has investigated differences in hours worked per person
across countries. Olivier Blanchard (2004) and Edward C. Prescott (2004) attribute
the bulk of G7 income differences in the mid-1990s to differences in hours worked
per person. But Caselli (2005) does not find people work systematically more in rich

3
Although richer families may send their children to higher quality schools, public funding may also tend to
equalize school quality more within countries than across countries.
 212 AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2010

PPP capital-output ratio in 1996 4

3 JPN
CHE

FIN NOR
SGP
FRA
AUT
KOR
ESP GER
ROM THA GRC BEL
ISL
CAN
ITA
SWE DNK
TZA NLD
POL
LSO HUN NZL AUS
ISR
PRT
2 JAM HKG
PER CYP GBR USA
MEX
ECUDZAPANBRA MYS
GNB GUY IRN VEN ARG
ZWE CRI IRL
ZMB TUR
PHLJOR TWN
BWA
NPL COGHND CHN IDN FJI
TUN CHL
DOM COL URY BRB
PNG
LKA PRY ZAF TTO MUS
1 ZAR MWI PAK
CMRIND BOL SYR
MLI
CAF
TGO BGD
KEN
NER GMBBEN SEN SLV
GTM
GHA
RWA EGY
SLE
MOZ HTI
UGA

0
1/64 1/32 1/16 1/8 1/4 1/2 1

PPP GDP per worker relative to the United States in 1996

Figure 1. Capital-Output Ratios versus Income Levels

versus poor countries. Across countries, the labor force participation rate increases
mildly with income, the unemployment rate is unrelated to income, and the work-
week falls with income.
An important caveat to Caselli’s (2005) results is that there is little data on hours
worked per worker outside the manufacturing (or at least urban) sector in develop-
ing countries. So it remains an open question whether, say, limited market work
contributes importantly to limited market income per worker in rural areas of poor
economies. Stephen L. Parente, Richard Rogerson, and Randall Wright (2000) make
the case that distortions would predict such low market activity. A related issue is
how much informal and home production is captured in income statistics.

D. Residual TFP

The upshot is that 50 percent or more of cross-country income variation appears

to remain unexplained by a combination of physical capital, human capital (includ-
ing health), and hours worked. This broad conclusion is not sensitive to whether the
accounting uses equation (2) or equation (3).

II.  Why Does Physical Capital Vary?

We now turn to work on the underlying causes of factor differences, starting with
physical capital. Figure 1 shows purchasing power parity (PPP) capital-output ratios
across 97 countries in 1996, derived from data in Penn World Table 6.1 (see Alan
 VOL. 2 NO. 1 HSIEH AND KLENOW: DEVELOPMENT ACCOUNTING 213

1996 percent investment rate at domestic prices (US = 1)

60 COG TKM

50

THA
40 KNA
JOR SVK KOR SGP
CZE
JAM SYR ATG HKG
IDN GRDBLZ
30 ALB JPN
VNM SWZ
TJK EST
VCTDMA HUN
AZE TUR BWATUN
LKA PHLECU LCA
RUS
ROM
PAN CHL PRT AUT
MLI NPL MDA KGZ UKR LTU
BLRHRV GAB
IRNMEX MUS
BHR
SVN ISR NOR
ZWE
UZB PER LBN NZL
QATGER LUX
BGD
20 KEN ARM PAK MKD
MAR POL BRA ARG GRCESP AUSCHE
NLD
ZMB SEN GIN LVA BRB
BEN CAN
OMN
ISL IRL
USA
BEL
ITA
KAZ DNK
FRA
YEM VEN FIN
GBR
TZA CMR CIV BOL SWE
NGA EGY TTO BHS BMU
MWI
MDG GEO URY
FJI
10
BGR
SLE

0
1/32 1/16 1/8 1/4 1/2 1

1996 PPP GDP per worker (US = 1)

Figure 2. Domestic Price Investment Rates versus Income Levels

Heston, Robert Summers, and Bettina Aten 2002). The ratios differ by a factor of
about four across rich and poor countries, from a low of about three-fourths in the
poorest countries to a high of about three in the richest countries. Behind this graph
(Figure 1) is one of the strongest relationships established in the empirical growth
literature: the positive correlation between the investment rate in physical capital and
the level of output per worker (see Ross Levine and David Renelt 1992 and Xavier
X. Sala-i-Martin 1997).
Why is the PPP investment rate higher in richer countries? A first thought is that
the savings rate is simply lower in poorer countries. Parente and Prescott (2000), for
example, document lower savings rates in poorer countries. Possible reasons include
subsistence savings needs, financial underdevelopment (low returns to savers and
high cost of borrowing for investment), and high implicit tax rates on capital income
(taxes, expropriation, corruption) in poorer countries. A shallow local savings pool and
inability to tap foreign pools may limit opportunities to finance domestic investment.
As noted, there is evidence of lower savings rates in poorer countries. But if low
savings rates explain the low PPP investment rates in poor countries, then we expect
the investment rate to be low in poor countries at domestic prices. These terms are
related as follows:

PPP Investment Rate in Country j  =   ​ __  ​  ;

Ij

Yj

PI ,  j Ij
Domestic Price Investment Rate in Country j  =      ​ _____   ​ 
.
PY ,  j Yj 
 214 AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2010

50% SLV

PRY
CIV
40%
MUS
EGY BWA
Naïve MPK in 1996

COG
MAR
BOL
30% ZAF
URY
BDI COL
ECU
TUN CHL
PHL TTO
VEN
JOR
MEX
20% LKA PER DZA
JAM MYS
CRI SGP IRL
ZMB
PAN NZL AUS
PRT KOR ESP ISR NLD
CAN NOR USA
GBR
DNK
AUT ITA
10% JPNFIN BEL
SWE FRA
GRC
CHE

0%
0 10,000 20,000 30,000 40,000 50,000 60,000

PPP GDP per worker in 1996

Figure 3. NaÏve MPK versus Income Levels

Ij and Yj denote the value of investment goods and output, respectively, in country j
valued at a common set of international (PPP) prices.4 PI, j and PY, j are the domestic
prices investment and output, respectively, in country j, relative to the international
prices of investment and output. Figure 2, taken from Hsieh and Klenow (2007),
shows that in 1996 there was no tendency for richer countries to invest a higher frac-
tion of their GDP at domestic prices. Foreign capital inflows (including official aid)
must have fully offset low domestic savings in poorer countries. Thus, the key to
understanding the low PPP investment rates in poor countries is to understand their
high domestic relative price of investment (compared to the international relative
price of investment).
A corollary to Figure 2 is that the marginal product of capital looks higher in
poor countries at international prices, but not at domestic prices. Caselli and James
Feyrer (2007) make this point, and further adjust the marginal product of reproduc-
ible capital (equipment and structures that go into K) by subtracting payments to
nonreproducible capital (land, minerals). In their definitions,

αj Yj αj PY ,   j Yj  −  rentsj
Naïve MPKj   =    ​ ____ ​    Corrected MPK   ≡    ​ ____________   
 ​    .
Kj PK, j Kj

4
PPP prices are a weighted average of prices in different countries (Heston, Summers, and Aten 2002).
 VOL. 2 NO. 1 HSIEH AND KLENOW: DEVELOPMENT ACCOUNTING 215

50%

40%
Corrected MPK in 1996

30%

20%

SLV
BWA SGP
URY MUS
ISR IRL
10% KOR
MEXCHL ESPGBR NLD USA
PHL PER
MAR
PRY JOR ZAF PRT JPNFIN DNK ITA
BEL
SWE AUT
AUS NOR
FRA
CAN
JAM COL PAN TUN MYS CHE
LKA GRC NZL
BOL EGY VEN
CIV CRIDZA
ECU TTO
ZMBCOG
BDI
0%
0 10,000 20,000 30,000 40,000 50,000 60,000

PPP GDP per worker in 1996

Figure 4. Corrected MPK versus Income Levels

Figures 3 and 4 use data from Caselli and Feyrer (2007) on the marginal product of
capital in 52 countries in 1996. In Figure 3, naïve marginal products hover around
10 percent in the OECD countries, and range from 15 percent to almost 50 percent
in poorer countries. So it looks as though capital flows fail to equalize marginal
products. It is as if risk or implicit tax payments mandate higher marginal products
in poorer economies.
Figure 4 displays marginal products for the same 52 countries, only corrected
for local differences in the price of capital, relative to output, and for payments to
nonreproducible capital. Strikingly, marginal products now appear lower in most
countries outside the OECD. Figure 4 suggests that rates of return are far more
equalized than previously imagined. This is because the price of investment relative
to output is higher in poor countries, and a higher fraction of capital income goes to
nonreproducible capital in poor countries (e.g., in land-intensive agriculture).
The relative price adjustment conjures a second hypothesis for low PPP invest-
ment rates in developing countries—expensive investment goods relative to rich
countries.5 Jonathan Eaton and Samuel Kortum (2001) establish that most devel-
oping countries import most of their equipment, so transportation costs and trade
barriers (tariffs and nontariff barriers alike) might make investment more expensive
in poor countries. Hsieh and Klenow (2007) show that this is not the case if one
compares Penn World Table prices of investment goods across countries. Figure 5,
taken from Hsieh and Klenow (2007), plots the price of equipment (relative to the

5
The high relative price of investment in poor countries has been well known since at least Barro (1991).
 216 AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2010

MKD
SYR

GAB
1996 equipment price (US = 1)

NGA

COG
2 YEM

CIV EGY
UZB FIN
BMUCHE
DNK
SEN JOR ATG JPN
SWE
GER
BEN
UKR HRVLBN GRC QAT
PRT AUS
FRA
IRL
AUT
NOR
LUX
MDA GIN
BOL ALB PER
ECU
RUS
EST SVN NZL
THA VENSVKHUN
CZE
GBR
ESP
ISL
OMN ITA
NLDBEL
1 BGR
LVALTUFJITUR IRN
URY
BWA USA
TZA ZMB
KEN VNM BLR PAN POL CHLARG
BRA
MEX CAN
MWI
MDG CMR AZE KAZROM KNA ISR
LCA BHRBHS
GEO
JAM DMA
GRD TUN MUS BRB SGP
SLE MAR TTO
MLI TJK KGZ
ARM KOR HKG
IDN VCT
PHL SWZ BLZ
LKA
NPL
1/2 ZWE
PAK
BGD

TKM

1/4
1/32 1/16 1/8 1/4 1/2 1

1996 PPP GDP per worker (US = 1)

Figure 5. Equipment Prices versus Income Levels

Source: Hsieh and Klenow (2007)

US price) against relative income across countries in 1996. The price of equipment
varies more among poor countries, but appears no higher in poor countries on aver-
age. Any import barriers seem to be offset by lower mark-ups or lower local distribu-
tion costs.6
If investment is not particularly expensive in poor countries, then consumption
must be cheap. Figure 6, also from Hsieh and Klenow (2007), confirms that PPP
consumption prices are lower in poor than rich countries. As an explanation, we pro-
posed that poor countries have lower TFP in producing investment goods (relative to
consumption goods). This relative TFP explanation can simultaneously rationalize
why poor countries have lower PPP investment rates, similar domestic price invest-
ment rates, similar investment good prices (at least for tradable investment), and
lower consumption good prices. The upshot is that relative TFP can exert a powerful
indirect effect on income differences through its impact on capital accumulation.

III.  Why Does Human Capital Vary?

Figure 7 combines Penn World Table 6.1 incomes with Barro and Lee (2000) data
on educational attainment. Average schooling attainment ranges from about 3 years
in the poorest countries to about 12 years in the richest countries. Earlier, we dwelled

6
An important caveat regards data quality. PPP prices are supposed to be quality-adjusted prices. Eaton and
Kortum (2001) suggest this might not be the case, given that many developing countries produce some equipment
but rarely export it, as if it is not competitively priced for the global market.
 VOL. 2 NO. 1 HSIEH AND KLENOW: DEVELOPMENT ACCOUNTING 217

MKD
2
CHE
JPN
SWE
BMUDNKNOR
1996 consumption price (US = 1)

FINFRA
GERAUT
ISLNLDBEL LUX
NGA SYR SGP
NZL
GBR ISR IRL
1 ESP AUS ITA
USA
GRC
ATG PRT KORCAN HKG
YEM SVN BHS
BRA ARG
URY BHR
JAM LBN
LCA
HRV
DMA KNA
PER FJI CHL
ECUVCT GRD VENTTO
BLZ
1/2 THA
PAN
JOR
COG TURPOL MEX QAT
ZMB EST HUN OMN
TZA BOL RUS GAB CZE
MDG
BEN LVA IRN
SEN LTU SVK
CIV PHL
MLI ALB IDN
PAK MAR
BWA BRB
MWI CMR
1/4 LKA EGY
KAZROM
KEN BGD GEO BGR TUN MUS
AZE
ARMUZB UKR
ZWE
KGZ BLRSWZ
VNM MDA GIN
SLE
NPL
TJK

1/8
TKM

1/16
1/32 1/16 1/8 1/4 1/2 1

1996 PPP GDP per worker (US = 1)

Figure 6. Consumption Prices versus Income Levels

15

NORUSA
12 NZL CAN
SWE
1996 years of schooling attainment

KOR AUS
CHE
GER
POL FIN
ROM ISR DNK
NLDHKG
GBR
CYPIRLBELJPN
HUN
9 PAN ARGGRC TWN ISL
BRB
PHL FJI AUT
PER TTO
URY
CHL FRA
MEX ESP ITA SGP
LKAJOR VEN MYS
CHN ECU PRY
GUY THA
ZAF
CRIBWA MUS
6 ZMB BOL SYR PRT
COG ZWE EGY
JAM DZA COL TUR
IND SLV
DOM
IDN IRN
TUNBRA
HND
LSO
KEN PAK
GHA
UGA
TGO CMR GTM
ZAR MWI HTI
TZA CAF PNG
3 RWA SLEBEN BGD
SEN
GMBNPL

MOZ
NER
GNB
MLI

0
1/64 1/32 1/16 1/8 1/4 1/2 1

1996 PPP GDP per worker (US = 1)

Figure 7. Schooling Attainment versus Income Levels

 218 AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2010

School fees relative to 1996 level 6

Secondary
5
Primary

0
1996 1997 1998 1999 2000 2001 2002 2003 2004

Year

Figure 8. Public School Fees in China from 1996 to 2004

Source: Hsieh and Klenow (2007)

on whether the Mincerian return across countries was closer to 10 percent (so that
schooling contributes a factor of 2.5 to income differences) or 30 percent (implying
income differences closer to a factor of 15). Here, we ask why schooling attainment
is higher in richer countries. We hasten to add that this section is more speculative
than the previous one on physical capital. As we note below, we will omit many
factors that could be critical for explaining schooling differences across countries.
A first candidate is that richer countries may provide greater public subsidies to
education. Consider, for example, the case of Mexico. From 1991 to 2004 the share
of government expenditures devoted to schooling rose from 15 percent to 26 percent.
Public spending per student rose from 5 percent of GDP per capita to 15 percent of
GDP per capita. This more generous public funding may have helped boost Mexico’s
secondary school enrollment rate from 45 percent in 1991 to 67 percent in 2004.7
Now consider China, which reduced public support for education at the same
time. The share of public expenditures for education fell from 21 percent in 1994 to
15 percent in 2004, forcing students to finance more of their education out of tuition
and fees, which rose from 7 percent to 32 percent of school spending over the same
period. These ratios understate the decline in support for primary and secondary
schooling in China because at the same time China increased public support for uni-
versities. Figure 8 plots fees (in constant prices) for public primary and secondary
schools in China from 1996 to 2004. Despite the increasing cost of public schooling,
the enrollment rate in secondary schooling in China rose from 50 percent to 70 per-
cent from 1991 to 2004—almost parallel to the increase in Mexico.8
Clearly, other factors besides government funding are important for schooling
attainment in China, and presumably elsewhere. To hint at some explanations for

7
The source of the numbers cited in the paragraph for school enrollment and spending in Mexico is the World
Development Indicators.
8
Figure 8 and the share of tuition and fees in total spending are from Emily Hannum et al. (2008). Public
spending on schools as a share of government spending and secondary school enrollment are from the World
Development Indicators.
VOL. 2 NO. 1 HSIEH AND KLENOW: DEVELOPMENT ACCOUNTING 219

schooling attainment in China, imagine that the marginal cost and benefit of addi-
tional schooling are

MCi  = Direct Costi  +  Opportunity Costi   =   Pi (Si )

φi Wi (Si )
MBi  =   ​ ____________       ​.
δ (ri , gi ,  Ti  −  Si )

The direct cost is tuition, and the opportunity cost is foregone earnings, both of
which are presumably increasing with the level of schooling S. The marginal ben-
efit is the present discounted value of additional earnings. Here φi represents the
Mincerian return to schooling. Wi (Si ) is the wage for an individual with school-
ing Si , and δ is an appropriate discount rate. The effective discount rate should be
increasing in the real interest rate ri , decreasing in the growth rate of wages gi , and
increasing in the number of years working Ti − Si , where Ti is the retirement age.
For simplicity, let Ti  =  ∞ so that δi  =  ri  − gi . Then, equating the marginal cost
and marginal benefit of schooling yields
φi Wi (Si ) Pi (Si ) φi
Pi (Si )  =   ​ _______   _____
ri  −  gi ​ 
  ⇒   ​    =   ​ ______
 ​   
     ​  
.
Wi (Si ) r i   −  gi

Although not necessary for the qualitative points we wish to make, we can obtain a
closed-form solution if we assume the cost of schooling relative to the wage is

Pi (Si )
​ _____ =   ​ ___i  ​  Siβ ,
P
 
 ​   
 
Wi (Si ) Wi

where Pi is the price of inputs to schooling, Wi is the wage, and β > 1. Combining
the previous two equations yields

φi 1/β
​S​i*​ ​  =  a​ _____________
      ​​b​  ​ .
(Pi /Wi )  ·  (ri  −  gi )

So, optimal years of schooling are increasing in the Mincerian return (φi),
decreasing in the relative price of schooling (Pi /Wi ), and decreasing in the effec-
tive discount rate (δi  =  ri − gi ). Do rich countries have higher schooling because of
higher Mincerian returns, say, due to skill-biased technology? Abhijit V. Banerjee
and Esther Duflo (2005) say no. They conclude that Mincerian returns are relatively
flat across countries with high versus low schooling attainment. Might rich countries
have lower discount rates because of better financial systems or better protection of
property rights? If so, then the marginal product of capital should be lower in rich
countries, contrary to the evidence in Caselli and Feyrer (2007). Related, if dis-
count rates are lower in rich countries, then we might expect to see lower Mincerian
returns in rich countries, contrary to the Banerjee and Duflo (2005) evidence. See
David Card (2001) and Todd Schoellman (2009) for expositions closely tying the
Mincerian return to discount rates and the relative price of schooling. It is possible,
of course, that discount rates and skill-biased technology differ in offsetting ways
220 AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2010

across countries. That is an open question for future research, as is the potential role
for heterogeneity in discount rates across individuals.
Might the relative price of schooling fall with development? The Mexico versus
China comparison notwithstanding, perhaps rich countries subsidize schooling at a
higher rate. Caselli (2005) finds little evidence for this in the Barro and Lee (2000)
data on schooling expenditures. In China, growth in TFP may be reducing the price
of nontime inputs to schooling (books, equipment, buildings) relative to the wage per
unit of human capital. This is precisely the thesis of Manuelli and Seshadri (2007)
and Erosa, Koreshkova, and Restuccia (2007). Another possibility is that teacher
salaries are lower relative to average wages in richer countries.
One would need a richer model to take to the data. This simple model does not
incorporate many factors that surely affect schooling decisions, such as liquidity
constraints, the ability to learn (presumably endogenous to early childhood invest-
ments), life expectancy, schooling costs not tied to the general level of wages, any
direct consumption value of schooling, and cultural and institutional differences (in
attitudes toward girls, for example).9 But, as of now, we lack information on critical
parameters even for this simple model. What does the human capital production
function look like? What is the price of schooling across countries? What is the
curvature parameter β? We leave these vital questions for future research in growth,
labor, and development fields.

IV.  Why Does TFP Vary?

As described, development accounting yields sizable residual TFP terms. In the

previous two sections, we have argued that TFP can indirectly affect physical and
human capital accumulation through the relative price of physical and human capi-
tal. The natural next question is why TFP itself varies.
One can decompose aggregate TFP into the (unweighted) average of firm-level
TFPs, and the efficiency of input allocation across firms. Parente and Prescott (2000)
examine some of the many factors that affect TFP at the firm level, such as disem-
bodied TFP, work rules, government ownership, and corruption.
Here we focus, instead, on how TFP can be affected by the efficiency of resource
allocation across firms. This is an old idea, but there has been a recent surge of mod-
els and evidence on misallocation. Restuccia and Rogerson (2008) show that modest
distortions can have nontrivial effects on aggregate TFP in a Lucas span-of-control
model. Hugo Hopenhayn and Rogerson (1993) present an early model on misalloca-
tion of labor due to firing costs. Caselli and Nicola Gennaioli (2003) model misal-
location of capital due to capital market imperfections, as do Francisco J. Buera and
Yongseok Shin (2008). Nezih Guner, Gustavo Ventura, and Yi Xu (2008) analyze
the consequences for TFP of size-dependent policies. Jones (2009) demonstrates
that complementarities across industries can allow modest industry-level distortions
to have larger effects on aggregate TFP.

9
Manuelli and Seshadri (2007) and Erosa, Koreshkova, and Restuccia (2007) incorporate several of these
factors.
 VOL. 2 NO. 1 HSIEH AND KLENOW: DEVELOPMENT ACCOUNTING 221

India
0.25

0.20

0.15

0.10

0.05

0
Percent of plants

1 4 16 64 256 1,024 4,096 16,384

United States
0.25

0.20

0.15

0.10

0.05

0
1 4 16 64 256 1,024 4,096 16,384

Number of workers in plant

Figure 9. Distribution of Plant Size, India versus the United States

Regarding the magnitude of resource misallocation, Banerjee and Duflo (2005)

present suggestive evidence that India’s low TFP in manufacturing relative to the
United States could reflect misallocation of capital across plants. In Hsieh and
Klenow (2009), we compare misallocation in China and India to that in the United
States. We document much wider dispersion in the value marginal products of capi-
tal and labor within manufacturing industries in China and India. We argue that
misallocation could explain about one-third of the manufacturing TFP gap between
China or India and the United States. In China, allocative efficiency appeared to
improve over our 1998–2005 sample, as inefficient state-owned enterprises faded in
importance and manufacturing TFP surged. In India, we find no such improvement
in the period 1987–1994 despite reforms, consistent with India’s tepid manufacturing
TFP growth.
In Hsieh and Klenow (2009), we also report a thick left tail of small plants in India
even within the formal manufacturing sector. Here, we show how thick the left tail
of small plants in India is when we include both formal and informal firms. Figure
9 compares the distribution of plant size (in terms of employment) in Indian versus
US manufacturing, where Indian plants include formal and informal firms. US data
are from the 1997 Census of Manufactures, and Indian data are from ­combining
222 AMERICAN ECONOMIC JOURNAL: MACROECONOMICS JANUARY 2010

the 1999 Annual Survey of Industries (which covers larger plants) with the 1999
National Sample Survey (which covers even the smallest plants). The majority of
plants in Indian manufacturing have 5 or fewer workers, whereas most US plants
have more than 40 workers.
Major questions for future research include the following: how much of the dis-
persion is real versus a by-product of measurement error? What specific distortions
(man-made or natural) generate greater dispersion in marginal products in poorer
countries? How large are the misallocations across sectors? Why is the distribution
of firm size so left-skewed in poor countries? We hope to see progress on these ques-
tions in the years to come.

V.  Conclusion

We have learned a great deal in the last two decades about the proximate deter-
minants of income differences. Although important questions remain, particularly
about the role of human capital differences, there is a broad consensus that differ-
ences in human capital account for 10–30 percent of country income differences,
physical capital accounts for 20 percent of country income differences, and residual
TFP may be the biggest part of the story (accounting for 50–70 percent of country
income differences).
But, we have much less understanding as to why these factors differ. We suggest
that boosting TFP may not only have a direct effect on output, but may also have an
important indirect effect via physical capital and human capital by lowering the price
of capital and schooling relative to the price of output. However, we need to know
more, particularly about the production function for human capital, before we can
make more definitive statements. Finally, we suggest the misallocation of inputs across
firms and industries may be an important determinant differences in residual TFP, but
it remains to be seen what the forces behind the misallocation are.

REFERENCES
Armenter, Roc, and Amartya Lahiri. 2006. “Endogenous Productivity and Development Account-
ing.” Federal Reserve Bank of New York Staff Report 258.
Banerjee, Abhijit V., and Esther Duflo. 2005. “Growth Theory through the Lens of Development Eco-
nomics.” In Handbook of Economic Growth, Vol. 1A, ed. Philippe Aghion and Steven N. Durlauf,
473–552. New York: North-Holland.
Barro, Robert J. 1991. “Economic Growth in a Cross Section of Countries.” Quarterly Journal of
Economics, 106(2): 407–43.
Barro, Robert J., and Jong-Wha Lee. 1993. “International Comparisons of Educational Attainment.”
Journal of Monetary Economics, 32(3): 363–94.
Barro, Robert J., and Jong-Wha Lee. 2000. “International Data on Educational Attainment Updates
and Implications.” National Bureau of Economic Research Working Paper 7911.
Blanchard, Olivier. 2004. “The Economic Future of Europe.” Journal of Economic Perspectives,
18(4): 3–26.
Buera, Francisco J., and Yongseok Shin. 2008. “Financial Frictions and the Persistence of History: A
Quantitative Exploration.” Unpublished.
Card, David. 2001. “Estimating the Return to Schooling: Progress on Some Persistent Econometric
Problems.” Econometrica, 69(5): 1127–60.
Caselli, Francesco. 2005. “Accounting for Cross-Country Income Differences.” In Handbook of Eco-
nomic Growth, Vol. 1A, ed. Philippe Aghion and Steven N. Durlauf, 679–741. New York: North-
Holland.
VOL. 2 NO. 1 HSIEH AND KLENOW: DEVELOPMENT ACCOUNTING 223

Caselli, Francesco, and James Feyrer. 2007. “The Marginal Product of Capital.” Quarterly Journal
of Economics, 122(2): 535–68.
Caselli, Francesco, and Nicola Gennaioli. 2003. “Dynastic Management.” National Bureau of Eco-
nomic Research Working Paper 9442.
Córdoba, Juan Carlos, and Marla Ripoll. 2008. “Endogenous TFP and Cross-Country Income Dif-
ferences.” Journal of Monetary Economics, 55(6): 1158–70.
Eaton, Jonathan, and Samuel Kortum. 2001. “Trade in Capital Goods.” European Economic Review,
45(7): 1195–1235.
Erosa, Andrés, Tatyana Koreshkova, and Diego Restuccia. 2007. “How Important Is Human Capital?
A Quantatitive Theory Assessment of World Income Inequality.” Unpublished.
Gollin, Douglas. 2002. “Getting Income Shares Right.” Journal of Political Economy, 110(2): 458–74.
Guner, Nezih, Gustavo Ventura, and Yi Xu. 2008. “Macroeconomic Implications of Size-Dependent
Policies.” Review of Economic Dynamics, 11(4): 721–44.
Hall, Robert E., and Charles I. Jones. 1999. “Why Do Some Countries Produce So Much More Out-
put Per Worker Than Others?” Quarterly Journal of Economics, 114(1): 83–116.
Hannum, Emily, Jere Behrman, Meiyan Wang, and Jihong Liu. 2008. “Education in the Reform
Era.” In China’s Great Economic Transformation, ed. Loren Brandt and Thomas G. Rawski. New
York: Cambridge University Press.
Hendricks, Lutz. 2002. “How Important Is Human Capital for Development? Evidence from Immi-
grant Earnings.” American Economic Review, 92(1): 198–219.
Heston, Alan, Robert Summers, and Bettina Aten. 2002. Penn World Table Version 6.1. Center for
International Comparisons at the University of Pennsylvania (CICUP). Philadelphia, October.
Hopenhayn, Hugo, and Richard Rogerson. 1993. “Job Turnover and Policy Evaluation: A General
Equilibrium Analysis.” Journal of Political Economy, 101(5): 915–38.
Hsieh, Chang-Tai, and Peter Klenow. Forthcoming. “Misallocation and Manufacturing TFP in China
and India.” Quarterly Journal of Economics, 124(4): 1403–48.
Hsieh, Chang-Tai, and Peter J. Klenow. 2007. “Relative Prices and Relative Prosperity.” American
Economic Review, 97(3): 562–85.
Jones, Benjamin F. 2008. “The Knowledge Trap: Human Capital and Development Reconsidered.”
National Bureau of Economic Research Working Paper 14138.
Jones, Charles I. 2009. “Intermediate Goods and Weak Links: A Theory of Economic Development.”
Unpublished.
Klenow, Peter J., and Andrés Rodríguez-Clare. 1997. “The Neoclassical Revival in Growth Econom-
ics: Has It Gone Too Far?” In NBER Macroeconomics Annual, Vol. 12, ed. Ben Bernanke and
Julio J. Rotemberg, 73–103. Cambridge, MA: MIT Press.
Klenow, Peter J., and Andrés Rodríguez-Clare. 2005. “Externalities and Growth.” In Handbook
of Economic Growth, Vol. 1A, ed. Philippe Aghion and Steven N. Durlauf, 817–61. New York:
North-Holland.
Levine, Ross, and David Renelt. 1992. “A Sensitivity Analysis of Cross-Country Growth Regres-
sions.” American Economic Review, 82(4): 942–63.
Mankiw, N. Gregory, David Romer, and David N. Weil. 1992. “A Contribution to the Empirics of Eco-
nomic Growth.” Quarterly Journal of Economics, 107(2): 407–37.
Manuelli, Rodolfo, and Ananth Seshadri. 2007. “Human Capital and the Wealth of Nations.” Unpub-
lished.
Parente, Stephen L., and Edward C. Prescott. 2000. Barriers to Riches. Cambridge, MA: MIT Press.
Parente, Stephen L., Richard Rogerson, and Randall Wright. 2000. “Homework in Development
Economics: Household Production and the Wealth of Nations.” Journal of Political Economy,
108(4): 680–87.
Prescott, Edward C. 2004. “Why Do Americans Work So Much More Than Europeans?” National
Bureau of Economic Research Working Paper 10316.
Pritchett, Lant. 2000. “The Tyranny of Concepts: CUDIE (Cumulated, Depreciated, Investment
Effort) Is Not Capital.” Journal of Economic Growth, 5(4): 361–84.
Restuccia, Diego, and Richard Rogerson. 2008. “Policy Distortions and Aggregate Productivity with
Heterogeneous Plants.” Review of Economic Dynamics, 11(4): 707–20.
Sala-i-Martin, Xavier X. 1997. “I Just Ran Four Million Regressions.” National Bureau of Economic
Research Working Paper 6252.
Schoellman, Todd. 2009. “Measuring and Accounting for Cross-Country Differences in Education
Quality.” Unpublished.
Weil, David N. 2007. “Accounting for the Effect of Health on Economic Growth.” Quarterly Journal
of Economics, 122(3): 1265–1306.