Econ 2390 Lecture Notes Fall 2015

Accounting for Income Differences

Michael Kremer, Harvard University
September 2nd, 2015

1 Lecture 1: Accounting for Income Differences I

1.1 Introduction

1.1.1 Selected Comparisons (data from 2009 or latest year available):

Year Italy Mexico Bangladesh

GNI per capita 35,240 9,980 520
GNI per capita PPP 30,250 14,270 1,440
Population growth per year (2000-08) 0.6 1.0 1.6
Female life expectancy at birth 84 77 67
Fertility rate (births per woman) 1.3 2.2 2.3
Under five mortality rate per 1,000 4 16.8 52
Malnutrition prevalence (height for age, % of children < 5) 2.7 15.5 43.2
Adult literacy rates 99 93 55
Primary school enrollment 99 98 85
Secondary school enrollment 92 72 41
Share of household income (lowest 40%) 18 13 22
Share of Agriculture (% of GDP) 2 4 19
Share of Industry (% of GDP) 27 37 29
Share of Services (% of GDP) 71 59 52
Gross savings (% of GNI) 16 22 35
Urban population (% of total) 67.6 76.3 25.7
Exports of goods and services (% of GDP) 24 28 19

Table 1: Selected World Development Indicators (WDI)

1
1.1.2 Falling poverty across the world through 2000, but not in Africa

Figure 1: World Distribution of Income in Various Years (Sala-i-Martin (2006) Figure IV)
Economics 270c: Lecture 1 17

2
(a) World Distribution of Income 1970 (Sala-i-Martin (2006) Figure IIIa)

Economics 270c: Lecture 1 18

(b) World Distribution of Income 2000 (Sala-i-Martin (2006) Figure IIIb)

Economics 270c: Lecture 1 19

3
1.1.3 Millennium Development Goals

MDG 1: Eradicate extreme poverty

Figure 2

4
MDG 2: Achieve universal primary education

Figure 3

5
MDG 3: Promote gender equality and empower women

Figure 4

6
MDG 4: Reduce child mortality

Figure 5

7
No Unconditional
1.2 Convergence
Conditional vs unconditional convergence
Consistent with the post-war patterns presented so far.

.0 6
HKG
KOR

THA
J PN IRL
.0 4
CHN ROM MM
YSUS
a n n u a l g r o wth r a te 1 9 6 0 - 2 0 0 0

SYC PRT
BRB
IDN ESP L UX
COG PAK CPV GRC
IND AUT
IT A
SYR
GAB
TUR FIN
ISR BEL
L SO EGYDOM BRA FRA
M AR NOR
L KA
.0 2

M WI NPL PAN
BGD IRN CHL ISL
GBRDNK USA
TT O
GNB SWNL
ED
AUS
ZW E M EX CAN
UGA ECU GT MJ OR
BF A GM B CIV PHL ZAF
TZ A URY CHE
KEN
GHA PRY
COL ARG
ET H BEN GIN SL CRI
V
NZ L
CM R HND
BDI TGO J AM
RW A
BOL PER
0

COM
SEN
MMDGOZTCD
ZM B
M LI
VEN
NER
NGA NIC
- .0 2

6 7 8 9 10
log gdp per w ork er 1960

But a Di¤erent Picture Among Relatively Similar Countries

(a) Annual growth rate of GDP per worker between 1960 and 2000 versus log
Figure: Annual
GDPgrowth rate
per worker of GDP
in 1960 per worker
for the entire between
world (Source: 1960(2009)).
Acemoglu and 2000 versus log
Convergence among (original) OECD countries.
GDP per worker in 1960 for the entire world.
J PN IRL
.0 4

Daron Acemoglu (MIT) Advanced Growth Lecture 1 September 5, 2007. 18 / 48

PRT
L UX
ESP
a n n u a l g r o wth r a te 1 9 6 0 - 2 0 0 0

GRC
AUT
.0 3

IT A

FIN
BEL
FRA
NOR
.0 2

ISL
GBR USA
DNK
NL D
SW E AUS
CAN
.0 1

CHE

NZ L

9 9.5 10 10.5
log gdp per w ork er 1960

(b) Annual growth rate of GDP per worker between 1960 and 2000 versus log
Figure: Annual
GDPgrowth rate
per worker of for
in 1960 GDPcore per
OECDworker between
countries 1960 and
(Source: Acemoglu 2000 versus log
(2009)).
GDP per worker in 1960 for core OECD countries.
Figure 6: Convergence

Daron Acemoglu (MIT) Advanced Growth Lecture 1 September 5, 2007. 19 / 48

8
Figure 7: Source: Human Development Report (2005)

Economics 270c: Lecture 1 26

Figure 8: Source: WDI 2014

9
• Growth over time
India Egypt South Korea Japan
Growth rate per capita GNP 1960-1980 1.4 3.4 7.0 7.1
Growth rate per capita GNP 1980-2000 2.9 3.2 5.6 2.1

Table 2: Source: Lucas (1988) and Jones & Klenow (2010)

10
Lucas (1988):

I do not see how one can look at figures like these without seeing them as

representing possibilities. Is there some action a government of India could

take that would lead the Indian economy to grow like Indonesia’s or Egypt’s?

If so, what, exactly? If not, what is it about the ‘nature of India’ that makes

it so? The consequences for human welfare involved in questions like these

are simply staggering: Once one starts to think about them, it is hard to think

about anything else.

11
Figure 9

12
1.3 Stylized facts

1.3.1 Kaldor facts.

In 1961, Nicholas Kaldor stated six now famous “stylized” facts, using them to

summarize what economists had learned from their analysis of twentieth-century

growth:

• Labor productivity has grown at a sustained rate.

• Capital per worker has also grown at a sustained rate.

• The real interest rate, or return on capital, has been stable.

• The ratio of capital to output has been stable.

• Capital and labor have captured stable shares of national income.

• Among the fastest growing countries of the world, there is an appreciable

variation in the rate of growth “of the order of 2-5 percent.”

13
1.3.2 “New” Kaldor facts.

Chad Jones and Paul Romer (2010) provide a new list of stylized facts:

1. Increases in the extent of the market. Increased flows of goods, ideas, fi-

nance, and people—via globalization, as well as urbanization—have increased

the extent of the market for all workers and consumers.

2. Accelerating growth. For thousands of years, growth in both population

and per capita GDP has accelerated, rising from virtually zero to the relatively

rapid rates observed in the last century.

3. Variation in modern growth rates. The variation in the rate of growth

of per capita GDP increases with the distance from the technology frontier.

4. Large income and total factor productivity (TFP) differences. Dif-

ferences in measured inputs explain less than half of the enormous cross-

country differences in per capita GDP.

5. Increases in human capital per worker. Human capital per worker is

rising dramatically throughout the world.

6. Long-run stability of relative wages. The rising quantity of human cap-

ital, relative to unskilled labor, has not been matched by a sustained decline

in its relative price.

14
1.4 Other stylized facts

• Population growth rates are negatively correlated with the level of income?

• Both skilled and unskilled workers migrate from poor to rich countries?

• Kuznets Curve: Inequality first increases, then decreases with income?

• Rich countries are are more politically stable?

• Rich countries are more secular?

• Greater division of labor - household work? In Kenya, for instance, people

grow the corn, grind it, gather water for fire, gather wood to burn, use pur-

chased cooking pot, and eat it. In the US, food is grown by farmers and

processed into microwaveable stuff. Electricity and water are delivered, just

microwave it.

• Relationships in rich countries have less of a tendency to be multi-stranded

(e.g. landlord & creditor; sharecropping, etc.)? Family and business relation-

ships more separated in rich countries?

• Rich countries are more politically democratic? Luxury good. Rich countries

tend to be nation-states. Borders of culture = borders of nation? Unified

culture of TV rather than vaguely blending over into next area.

• Rich countries have more intergenerational social mobility?

• Rich countries have an ideology of equality, rather than hierarchy, as in aris-

tocracies?

15
• Rich countries have an ideology of rule of law, rather than personalistic rule?

• Rich countries have more geographic mobility?

• Education tends to be run by the state in rich countries? In poor countries,

run by religious institutions, or people hire private tutors?

• Assignment of jobs is done in part by educational qualifications?

• Investment and education is correlated with growth (see figures below)?

• More civil wars, ethnic conflict & violence in poor countries?

• Less division of labor and diversity of products in poor countries?

• Better developed financial markets & access to credit in rich countries?

• Richer countries have more macro stability (inflation, GDP growth, etc.)?

• Better social safety net in rich countries?

• Social networks & kin financing more important in poor countries?

• Government revenue sources: commodity taxes & donors in poor countries?

• Fewer posted prices in poor countries?

• Gender gap more pronounced in poor countries?

• Less contractibility & more renegotiation in poor countries?

16
2 Lecture 2: Accounting for Income Differences II

2.1 Calibrating the Solow Model

• Assume a production function

Y (t) = K(t)α (A(t)L(t))1−α 0<α<1 (1)

• Assume that L and A grow exogenously at rates n and g:

L(t) = L(0)ent A(t) = A(0)egt (2)

Hence, the number of effective units of labor grows at rate n + g.

• Define capital per effective worker as

K
k≡ (3)
AL

and output per effective unit of labor as

Y
y≡ = kα. (4)
AL

17
• Assume savings rate s, depreciation rate δ, and n and g constant.

• The evolution of capital per effective worker is governed by

k̇(t) = sk(t)α − (n + g + δ)k(t) (5)

which implies that k converges to a steady state value k ∗ :

k̇(t) = 0 ⇒ sk(t)α = (n + g + δ)k(t) (6)

 1
 1−α
s
⇒ k∗ = (7)
n+g+δ

• In steady state, we have

 
Y (t)
log = log(A(t)) + log(k ∗ )α (8)
L(t)
α α
= log(A(0)) + gt + log(s) − log(n + g + δ) (9)
1−α 1−α

18
2.1.1 Mankiw, Romer, Weil (1992)

• Assuming log(A(0) = a + , where a is a constant and  is a country-specific

shock, MRW arrive at the following estimating equation1 :

 
Yi α α
log =a+ log(si ) − log(ni + g + δ) + i
Li 1−α 1−α

• MRW plug in α = 31 , and run a cross-country regression, expecting to find an

elasticity of income per capita with respect to savings rate of about one half,

and an elasticity of income with respect to n + g + δ of about minus one half

(see Table I).

1
with t = 0 for simplicity, g + δ is assumed to be 0.05 for all countries

19
414 QUARTERLY JOURNAL OF ECONOMICS

TABLE I
ESTIMATION
OF THETEXTBOOK
SOLOWMODEL

Dependent variable: log GDP per working-age person in 1985

Sample: Non-oil Intermediate OECD

Observations: 98 75 22
CONSTANT 5.48 5.36 7.97
(1.59) (1.55) (2.48)
ln(I/GDP) 1.42 1.31 0.50
(0.14) (0.17) (0.43)
ln(n + g + 8) -1.97 -2.01 -0.76
(0.56) (0.53) (0.84)
H2 0.59 0.59 0.01
s.e.e. 0.69 0.61 0.38
Restricted regression:
CONSTANT 6.87 7.10 8.62
(0.12) (0.15) (0.53)
ln(I/GDP) - ln(n + g + 8) 1.48 1.43 0.56
(0.12) (0.14) (0.36)
1?2 0.59 0.59 0.06
s.e.e. 0.69 0.61 0.37
Test of restriction:
p-value 0.38 0.26 0.79
Implied a 0.60 0.59 0.36
(0.02) (0.02) (0.15)

Note. Standard errors are in parentheses. The investment and population growth rates are averages for the
period 1960-1985. (g + 8) is assumed to be 0.05.

Three aspects of the results

Figure Tablethe
support
10: MRW I Solow model. First,
the coefficients on saving and population growth have the predicted
signs and, for two of the three samples, are highly significant.
Second, the restriction that the coefficients on ln(s) and
ln(n + g + 8) are equal in magnitude and opposite in sign is not
rejected in any of the samples. Third, and perhaps most important,
differences in saving and population growth account for a large
fraction of the cross-country variation in income per capita. In the
regression for the intermediate sample, for example, the adjusted
R2 is 0.59. In contrast to the common claim that the Solow model
"explains" cross-country variation in labor productivity largely by
appealing to variations in technologies, the two readily observable

about 0.03 or 0.04. In addition, growth in income per capita has averaged 1.7
percent per year in the United States and 2.2 percent per year in our intermediate
sample; this suggests that g is about 0.02.
20
• MRW interpret their Table I as supportive of the Solow model:

– The coefficient on log(s) and log(n + g + δ) are significant, roughly equal

in magnitude, and have the predicted signs.

– Regression has high adjusted R2 . More than half the cross-country varia-

tion in income per capita can be explained by saving rates and population

growth.

21
• Issue: The estimated impacts of saving and labor force growth are much

larger than the model predicts, and the value of α implied by the coefficients

should equal capital’s share, which is roughly one third.

The estimates, however, imply a much higher α. For instance, in the con-

strained regression of column 1, we get:

α 1.48
= 1.48 ⇒ α = ≈ 0.6
1−α 2.48

22
• MRW augment the Solow model to incorporate human capital H:

Y (t) = K(t)α H(t)β (A(t)L(t))1−α−β (10)

• This yields the following equation to estimate (see paper for derivation):

 
Y (t) α α β
log = log(A(0)) + gt + log(sk ) − log(n + g + δ) + log(h∗ )
L(t) 1−α 1−α 1−α

where sk is the fraction of income invested in physical capital, and h∗ is the

steady state human capital per effective unit of labor (see Table II).

23
Figure 11: MRW Table II

24
• Incorporating human capital into the Solow model improves its performance.

– Including the human capital measure greatly reduces the size of the co-

efficient on physical capital.

– The three variables together explain almost 80 percent of the cross-

country variation in income per capita.

– The values for α and β implied by the coefficients in the restricted re-

gression are about one third.

• Mankiw, Romer, & Weil (1992) argue that an augmented Solow model that

includes accumulation of human as well as physical capital provides an excel-

lent description of the cross-country data. MRW estimate that physical and

human capital explain 78% of cross-country variance in output.

25
• Alwyn Young (1995) finds that the East Asian growth miracle was due to

accumulation of capital and increase in labor rather than rising productivity.

– Translog production function

– Splits capital and labor into translog indices of subinputs

– Labor

∗ measured using census

∗ factor share estimated using wage data, by sector

∗ finds rising participation rates, improving levels of education

– Capital

∗ estimated using perpetual inventory approach with geometric depre-

ciation

∗ factor share estimated as one minus labor share

∗ finds increasing investment

– Intersectoral transfers of labor (from agriculture to manufacturing) also

found to be important

26
• Chang-Tai Hsieh (1999)

• “The central point of this paper is that, if East Asia’s growth was largely

driven by capital accumulation with little technological progress, the return

to capital should have fallen dramatically as capital accumulation encounters

diminishing returns”

• Considers returns to factors rather than stocks of factors.“Dual” growth ac-

counting vs. “primal” growth accounting. See problem set for more details.

– SRprimal = Ŷ − sk K̂ − sl L̂

– SRdual = sk r̂ + sl ŵ

• “except in the case of Korea, real wages have increased rapidly in East Asia

without a corresponding decline in the return to capital, implying a high rate

of dual TFPG.”

27
• Problem: Assumption that technology differences across countries are orthog-

onal to all other variables may be untenable.

– Omitted variable bias and reverse causality. Causation could go from

high A to (high) income to (low) fertility, (high) savings. We know that

the savings rate changed over time within the US and other countries.

– If we think there is some causation in this direction, then we have an

upper bound of the effect of savings on output. This would mean that

MRW’s estimate of the capital share is likely to be an upper bound.

28
• MRW show that, assuming particular parameter values, the Solow model can

fit the data quite well, but we need additional facts to pin down the rate of

return to human capital.

• Model implies rates of return higher in poor countries. MRW argue that

capital does not flow because of bad capital markets and risk of expropriation

(we will return to this issue).

29
2.1.2 Klenow & Rodriguez-Clare (1997)

• Klenow and Rodriguez-Clare redo MRW’s exercise, using micro-data on re-

turns to education rather than estimating these returns from cross country

regressions.

• They use data on primary and tertiary schooling, which had not been avail-

able at the time of MRW’s study, and argue that the (Solow) residual A is

important in explaining cross-country income differences.

30
• Setup. They rewrite the production function as

Y = K α H β (AL)1−α−β (11)
α
  1−α−β β
  1−α−β
Y K H
=A ≡ AX (12)
L Y Y

• Keep α = 0.3 and β = 0.28 for comparison purposes (taken from MRW Table

II, restricted regression).

• Decomposition of variance of log(Y /L):

V ar(log(Y /L)) Cov(log(Y /L), log(Y /L)) Cov(log(Y /L), log(A)) + Cov(log(Y /L), log(X))
= =
V ar(log(Y /L)) V ar(log(Y /L)) V ar(log(Y /L))
Cov(log(Y /L), log(X)) Cov(log(Y /L), log(A))
⇒1= +
V ar(log(Y /L)) V ar(log(Y /L))
| {z } | {z }
Contribution of X Contribution of A

• Interpretation: When Y /L is 1% higher, how much higher is expectation of

X and A?

31
• Results (see Table 1)

– MRW0 is the original estimate: the contribution of X is 78%.

– MRW1 fixes the fact that the amount of time studying is not counted

in GDP, hence excluding the human capital production sector. This

decreases the fraction explained by X to 76%.

– MRW2 updates the dataset and restricts to countries that have enough

education data. The fraction of X goes back to 78%.

32
– MRW3 uses a weighted average of primary, secondary, and ter-

tiary enrollment rates weighted by the corresponding popula-

tion share to measure the rate of human capital accumulation.

As a result, the fraction of X decreases to 40%.

This does not mean that primary schooling is unproductive, but sim-

ply that primary schooling does not vary as much across countries as

secondary school.

– MRW4 adjusts for the fact that human capital production is human cap-

ital intensive. The share of X falls further to 33%.

33
 TheNeoclassical - 81
Revivalin GrowthEconomics

Table 1 THE ROLES OF A AND X IN 1985 PROSPERITYa

cov[ln(Y/L),In (Z)]/varln(Y/L)

Source• Z= Z= Z= X Z= A

MRWO .29 .49 .78 .22

MRW1 .27 .49 .76 .24
MRW2 .31 .47 .78 .22
MRW3 .29 .11 .40 .60
MRW4 .29 .04 .33 .67
aMRWO:from MRW (uses their data appendix). MRW1: MRWObut with Ky/Y instead of K/Y. MRW2:
MRW1 but with L = worker instead of working-age population, 14 countries in/out. MRW3: MRW2 but
with all enrollment rather than just secondary enrollment. MRW4: MRW3 but with (K, H, L) shares of
(0.1, 0.4, 0.5), not (0.20, 0.28, 0.42), in H production.

Our first modification of MRW's methodology is to recognize that,

contrary to (1), national income accounting measures of output do not
include the value of student time-an importantcomponent of human
capital investment.'3 To see how importantthis might be, we consider
the extreme case in which none of the human capital investment is
measured as part of total output. To do this we replace K/Yand H/Y in
equation (2) with Ky/Yand Hy/Y,since only Kyand Hy are used in the
production of Y when Y does not include human capitalinvestment. It
turns out that the MRWmeasure of IH/Y,namely LHIL, is also appropriate
for Hy/Ywhen all human capitalinvestment goes unmeasured.The same
is not true for physical capital intensity, for which we must use Ky/Y=
(K/Y)(Ly/L). As shown by the MRW1row of Table 1, this modification
results in a 76%ln(X) vs. 24%ln(A) breakdown, so this distinctiondoes
not appear to be quantitativelyimportant.
The MRW2row of Table1 reproduces the MRW1row, only with up-
dated data and a set of countries for which we have all the necessary
schooling attainment data for the remainderof this paper. Like MRW,
13. MRWcontend that this slippage between model and data is not quantitativelyimpor-
tant. Parenteand Prescott(1996)disagree,contendingthat unmeasuredhuman capital
investment must be implausiblylarge for the combined share of capital to be about
two-thirds. Parente, Rogerson, and Wright(1996)illustratethat unmeasuredinvest-
ment would have to be 25-76%of GDP.We are in closeragreementwith MRW,since,
accordingto Kendrick(1976),about half of schoolinginvestmentconsists of education
expenditures(teachers,facilities)which areincludedin measuredoutput. Accordingto
the 1996 Digest of Education Statistics
34published by the U.S. Department of Education
(1996), education expenditures averaged 7% of GDP over 1960-1990. Back-of-the-
envelope calculationssuggest unmeasuredinvestmentmight thereforebe only 13%of
• Table 1 suggests that human capital has very low explanatory power. But

if β is very high, even smaller variations in human capital can explain large

variation in output across countries. What is the right β?

35
• Mincer regressions in 48 countries find wage gain of 9.5% for one more year of

education. K&R use Mincer regressions to estimate the human capital stock

(see Table 2).

– BK1: Exponential form from Mincer (1974) puts more weight on sec-

ondary school than primary school, so the share of X moves back towards

MRW.

– BK2 adjust for being out of steady state.

– BK3 adjusts for human capital due to experience.

36
– BK4 eliminates variation in school quality.

∗ Model assumes quality of schooling is much higher in rich countries be-

cause of better facilities (higher KH /LH ) and better teachers (higher

HH /LH ).

∗ Implied elasticity of quality with respect to Y /L is 0.95.

∗ Implied quality of education varies by factor of 34 across countries.

∗ Using census data on US earnings of immigrants from different coun-

tries, Borjas (1987) estimates an elasticity of 0.12.

∗ Immigrants with 1% higher per capita income in their country of

origin exhibit a 0.116% higher wage intercept.

∗ If less variation in school quality, need more variation in A to explain

data.

37
86 KLENOW &RODRiGUEZ-CLARE
"
Table2 THEROLESOFA ANDX IN 1985PROSPERITY
cov[ln(Y/L),In (Z)]/varIn(Y/L)

Sourcea Z Z= X Z= A
Z= =
BK1 .29 .31 .60 .40
BK2 .23 .33 .56 .44
BK3 .23 .31 .53 .47
BK4 .23 .11 .34 .66
aBK1:uses (7), i.e. Mincerevidence. BK2:calculatesyears of schoolings from Barro-Lee1985stocks
insteadof 1960-1985flows. BK3:adds averageyearsof experience.BK4:BK3but with (K,H, L)sharesof
(0, 0, 1) instead of (0.1, 0.4, 0.5) in H production.

over 1960-1985. Unfortunately,direct estimates of the 1960 KIYare not

availablefor most countries. We thereforeset, for each country,

K
) 1960 IK/Y
Y 196=g + +n

with the investment rate IK/Y,the growth rate of Y/L(g), and the popula-
tion growth rate (n) equal to the country'saveragesover either1960-1965,
1960-1970, or 1960-1985,and 8 either0.03, 0.05, or 0.07. Wealso followed
a procedureakin to King and Levine (1994)where we set g in the denomi-
natorequal to a weighted averageof own-countryand world growth. The
results were not at all sensitive to which way we calculatedthe 1960K/Y,
so we reportthe results with 1960K/Ycalculatedusing 8 = 0.03 (as in Table
1) and the country's own averages over 1960-1970 for g and n. To con-
structthe 1985H/Y,we use Barroand Lee's (1993)dataon averageyears of
schooling attained by the 25-64-year-old population in each country in
1985. We reportthe results of using this approachto obtain 1985levels of
K/Yand H/Y in the BK2row of Table2. Conditionalon 1%higher Y/Lin
one countryin 1985,we expect 0.56%higherX and 0.44%higherA in that
country.These results are not far from the (60%,40%)breakdownin BK1
with the steady-state assumption for K/Yand H/Y.19
We now modify (7) to incorporate human capital acquired through
experience:
38
= (KHILH)
1-/- (hT)(Ae(vls+
y2exp+y (9)
3exp2)/a),
• K&R also find a lot of variation in growth rates is due to variation in growth

of A.

• A is a residual measure of ignorance. Mis-measurement (Jorgenson)?

• Overall, K&R suggests that A differs across countries. Why is that the case?

– Barriers to technology adoption.

– Inappropriate technology: Basu and Weil.

– Not literal technology, but efficiency in using it.

– Political economy: Parente and Prescott.

39