The Agricultural Productivity Gap Douglas Gollin Oxford
David Lagakos UCSD
Michael E. Waugh NYU
November 11, 2013
0 / 37
Agriculture Sector Across Countries
• Share of value added lower than share of employment
• True in basically every country in world
• Particularly so developing countries
1 / 37
Agriculture Sector Across Countries
100
Share of Value Added in Agriculture
90 80 70 60 LAO CAF ETH
40
MLI UZB KHM
30 20 10 0 0
BDI
SLE
50
GUY
NGA
KGZ BTN TON MNG PAKFJI BGD STP ARM NIC SWZ MDA SEN WSM BLZ BLR CPV IDN MAR CHN LKA PHL EGYHND SUR DMA TUN BOL GTM SLV MKD SRB MNE MYS THA NAM VCT DZA MHL UKR COL IRNTUR ROM BGR CRIGRD ARG LBN URY DOM MDV ISLNZL AZE GAB KAZ PAN JAM HRV CUB LCA BRA RUS VEN POL SVK MUSGRC HUN IRQ LTU CHL PRT MEX ZAF LVA EST JOR KOR FIN AUS ESP SAU SVN CZE CYP BHS BWA OMN ITA BRB ATG CAN ISR MLT FRA NLD SWE AUT IRL DNK NOR USA BRN JPN GBR CHE GER BEL BMU PRY
SYR
10
20
30
BEN GHA
NPLLBR TCD SDN PNG KEN IND HTI
40 50 60 70 Share of Employment in Agriculture
BFA MWI MDG
TZA
CIV GMB UGA GIN TJKVNM ZMB ZWE ALB LSO YEM GEO
RWA
CMR
80
90
100
2 / 37
The Agricultural Productivity Gap
We define the Agricultural Productivity Gap (APG) to be APG ≡
VAn /Ln . VAa /La
• Simple two-sector model says APG should be 1 • In practice, average APG ∼ 3 • Poorest quartile of income distribution: average APG = 5.6 !
3 / 37
The Agricultural Productivity Gap
• Taken at face value, gaps suggest misallocation
• Policy debate: encourage movement out of agriculture?
• This paper: step back and address measurement issues
4 / 37
What Do Agricultural Productivity Gaps Reflect?
• Sector differences in hours worked per worker?
Construct measures of hours worked by sector for 76 countries • Sector differences in human capital per worker?
Construct measures of human capital by sector for 124 countries • Measurement error in national accounts data?
Construct our own estimates using household survey data in 10 countries
5 / 37
What Do Agricultural Productivity Gaps Reflect?
• Sector differences in hours worked per worker?
Construct measures of hours worked by sector for 76 countries • Sector differences in human capital per worker?
Construct measures of human capital by sector for 124 countries • Measurement error in national accounts data?
Construct our own estimates using household survey data in 10 countries
5 / 37
What Do Agricultural Productivity Gaps Reflect?
• Sector differences in hours worked per worker?
Construct measures of hours worked by sector for 76 countries • Sector differences in human capital per worker?
Construct measures of human capital by sector for 124 countries • Measurement error in national accounts data?
Construct our own estimates using household survey data in 10 countries
5 / 37
What Do Agricultural Productivity Gaps Reflect?
• Sector differences in hours worked per worker?
Construct measures of hours worked by sector for 76 countries • Sector differences in human capital per worker?
Construct measures of human capital by sector for 124 countries • Measurement error in national accounts data?
Construct our own estimates using household survey data in 10 countries
5 / 37
What We Conclude
• Our adjustments reduce average APG to around two
• Still bigger in developing countries
• Large gaps also present in household survey data
• Needed: better understanding of why residual gaps so large
6 / 37
Simple Two-Sector Model • Technologies
Ya = Aa Lθa Ka1−θ
and
Yn = An Lθn Kn1−θ
• Households can supply labor to either sector. • Competitive labor markets, i.e. workers paid their marginal product. • Equilibrium:
APG ≡
VAn /Ln Yn /Ln = = 1. VAa /La pa Ya /La
7 / 37
Computing “Raw” Agricultural Productivity Gaps
Measures of VAa and VAn • Value added as defined in 1993 System of National Accounts (SNA) • Source: UN National Account Statistics
Measures of La and Ln • Employed persons working in the production of some good or service
recognized by the 1993 SNA • Source: ILO, via population censuses or labor force surveys.
8 / 37
Raw Agricultural Productivity Gaps
Quartile of Income Distribution All Countries
Q1
Q2
Q3
Q4
10th Percentile
1.3
1.0
1.3
1.0
1.2
Median
2.6
1.7
2.7
2.8
4.3
Mean
3.5
2.0
3.2
3.4
5.6
90th Percentile
6.8
4.0
6.6
7.1
12.5
Number of Countries
151
38
38
38
37
9 / 37
“Simple” Measurement Error in National Accounts Data?
1. Understate agricultural VA by excluding home production? • In principal: No, it is included as per SNA. • Accepted practice: output of particular crop = area planted X yield
2. Overstate agricultural employment, by including all rural persons? • In principal: No, only economically active persons included per SNA. • We find national accounts consistent with household surveys.
10 / 37
Our Adjustments 1. Improved measures of labor input by sector • Sector differences in hours worked per worker. • Sector differences in human capital per worker.
2. Alternative measures of value added by sector • Reconstruct national accounts data from household survey data.
11 / 37
Our Adjustments 1. Improved measures of labor input by sector • Sector differences in hours worked per worker. • Sector differences in human capital per worker.
2. Alternative measures of value added by sector • Reconstruct national accounts data from household survey data.
11 / 37
Improved Measures of Labor Input by Sector
12 / 37
Sector Differences in Hours Worked Average hours worked per worker might differ across sectors We construct average hours worked per worker by sector for 76 countries • Population census micro data or labor force surveys • All employed persons 15+ years old • Industry of primary employment • Hours worked in reference period (usually one week)
13 / 37
Sector Differences in Hours Worked
55
2.0
1.5
ZWE
1.0
BGD NIC
KOR GTM
Hours Worked in Non−Agriculture
50
RWA
KEN
ARM CIV
LBR ALB SLV
ZMB PHL LSO SYR
45 NPL
40
FJI
LVA ROM LKA VNM LTU BGR JAM LCA MUS
ETHMWI POL TON
EGY PRY
JOR
TUR
SVN GHA FIN
HND HUN ESTGRC TZA
SWZ BOL SLE NGA MEXPAK KHM IDN ZAF BTN BRAVEN CHL BWA
TJK PRT SWE ITA FRA ISR BEL
GBR
IRQ ESP
PAN
RUS
USA GER
AUS
CHE CAN
35 NLD
30
25 25
30
35
40 45 Hours Worked in Agriculture
50
55
14 / 37
Sector Differences in Hours Worked
Agricultural Productivity Gaps and Hours Adjustment Measure
Raw APG
Hours Adjustment
5th Percentile
1.4
1.6
Median
3.2
2.7
Mean
3.1
2.9
95th Percentile
5.7
4.2
Number of Countries
76
76
Note: All statistics are weighted with each country weighted by its population.Only countries with hours data shown.
Differences in hours worked contribute a factor of 1.1.
15 / 37
Not Artifact of Seasonality
Malawi Hours Worked
Hours Worked
Ghana 50 40 30 20 2
4
6
8
10
50 40 30 20
12
2
4
50 40 30 20 2
4
6
6
8
10
12
8
10
12
Uganda Hours Worked
Hours Worked
Tanzania
8
10
12
50 40 30 20 2
4
6 Month
Hours Worked
Vietnam Non−Agriculture Agriculture
50 40 30 20 2
4
6 Month
8
10
12
16 / 37
Not Artifact of Secondary Jobs
Sector of Hours Worked Country
Worker Classification
Agriculture
Cote d’Ivoire (1988)
Agriculture
35.1
1.0
Non-agriculture
0.7
49.2
Agriculture
28.8
3.7
Non-agriculture
2.0
30.6
Agriculture
47.6
1.3
Non-agriculture
0.8
49.1
Agriculture
26.4
1.4
Non-agriculture
2.3
38.2
Agriculture
39.5
0.1
Non-agriculture
0.1
39.3
Agriculture
18.7
2.1
Non-agriculture
1.8
43.3
Ghana (1998) Guatemala (2000) Malawi (2005) Tajikistan (2009) Uganda (2009)
Non-agriculture
17 / 37
Sector Differences in Human Capital
Average human capital per worker could differ across sectors (Caselli & Coleman, 2001; Vollrath, 2009; Herrendorf & Schoellman, 2013) We construct human capital per worker by sector for 124 countries • Years of schooling measured directly when available • Impute years of schooling using educational attainment otherwise • Baseline: Assume 10% rate of return on year of schooling
hj,i = exp(sj,i · 0.10)
18 / 37
Sector Differences in Schooling
15
Years of School in Non−Agriculture
2.0
10
5
NLD
1.5
ARM UKRGEO RUS USA CAN KGZGER AUS ROM KAZ MDA IRL SVN KOR AUT MNG JAM HUNUZB NOR BLR BGR ALB GRC CUB SWE DNK SRB GUY AZE GBR CHL FRA TJK TON ISR PRY EGY ISL PHL LVA ITA MHL PAN JOR EST LTU ARG MKD LKA ESP TUR ZAFBLZ IRN FJI SWZ BOL ZWE CHN NAM CRI URY COL NGA VNM MEX VEN ZMB DOM NIC THA IDN SUR HND MDV LSO UGA SLV MYS IND PNG SYR BRA CMR YEM GHA PRT IRQ BWA GTM MDG MWI BDI GAB TZA KEN HTI LAO LCA LBR PAK ETH BTN RWA KHM MAR STP CAF NPLGMB CIV BFA BGDBEN
1.0
CHE
TCD SLE SEN SDN GIN MLI
0 0
5
10
15
Years of School in Agriculture
19 / 37
Sector Differences in Human Capital
4.5 2.0
Human Capital in Non−Agriculture
4
3.5
3
2.5
2
1.5
1 1
1.5
NLD
1.0
ARM RUSUSA UKR GEO CAN KGZ AUS GER ROM KAZ MDA SVN IRL KOR AUT MNG JAM HUN UZBNOR BLR ALB BGR GRC CUB SWE DNK SRB CHL GBR TON GUY FRAAZE ISR PRY TJK EGY ISL PHL LVA ITA PAN JOR LTU EST MHL ARG MKD LKA ESP TUR IRN ZAF SWZ BLZ FJI BOL CHNNAMZWE CRI URY COLNGA VNM MEX VEN DOM NIC ZMB THAIDN SUR MDV LSO HND UGA SLV MYSIND PNG SYR CMR BRA YEM GHA PRT IRQ BWA MDG GTM MWI TZA BDI GAB LAO KEN HTI LCA LBR PAK ETH MAR RWA KHMBTN STP CAF GMB NPL CIV BGD BEN BFA
CHE
TCD SLE SEN SDN GIN MLI
1.5
2
2.5 3 Human Capital in Agriculture
3.5
4
4.5
20 / 37
Sector Human Capital Differences
Agricultural Productivity Gaps and Human Capital Adjustment Measure
Raw APG
Human Capital Adjustment
5th Percentile
1.4
1.2
Median
3.6
2.6
Mean
3.8
2.7
95th Percentile
6.4
4.7
Number of Countries
124
124
Note: All statistics are weighted with each country weighted by its population. Only countries with schooling data shown.
Human capital differences contribute a factor of 1.4.
21 / 37
Quality Adjustments to Schooling Data
• Rural schools often of lower quality than urban schools
(Williams, 2005; Zhang, 2006) • Health, parental inputs may be lower in rural areas
• Potentially overestimate human capital among agriculture workers
• We use literacy data to adjust for this
22 / 37
Uganda: Literacy by Years of Schooling Completed
("
Literacy Rate
!#'" !#&" !#%" !#$" !" !"
$"
%"
&"
'"
(!"
Years of Schooling )*)+,-".*/01/2"
,-".*/01/2"
23 / 37
Measuring Quality Differences in Schooling
• Given literacy rates by years of schooling: ℓni (s) and ℓai (s) for s = 1, 2, ...
• Assume that each year in rural school is worth γ years in urban school
• For each country i, solve for γi that solves
min γ
s¯ X
ℓ˜ni (γs) − ℓ˜ai (s)
2
s=1
where ℓ˜ni (·), ℓ˜ai (·) are polynomial interpolations of ℓni (·), ℓai (·) for s ∈ [0, s¯].
24 / 37
2.0
1.5
1.0
3
CHL PAN ARG PHL BOL
2
THA VNM MEX VEN UGA MYS GHA BRA TZA RWA
GIN MLI
1
Human Capital in Non−Agriculture
4
Sector Differences in Quality-Adjusted Human Capital
1
2
3
4
Human Capital in Agriculture
25 / 37
Human Capital: Other Issues • Country-specific returns to schooling?
- Doesn’t change much • Quality-adjusted returns to schooling from Schoellman (2012)?
- Modestly decreases importance of human capital • Human capital from experience?
- Returns somewhat lower in agriculture (Lagakos, Moll, Porzio, Qian, 2013; Herrendorf & Schoellman, 2013) - Increases importance of human capital - Limitation: returns estimated for only 20 countries
26 / 37
Adjusting the Raw APG numbers Recap: • Differences in hours worked contribute a factor of 1.1. • Differences in human capital contribute a factor of 1.4.
Now, put them all together and construct “adjusted” APGs.
27 / 37
Adjusted Agricultural Productivity Gaps
Table 4: Agricultural Productivity Gaps and All Adjustments All Adjustments by Quartile Measure
Raw APG
All Adjustments
Q1
Q2
Q3
Q4
10th Percentile
1.3
1.0
0.8
1.2
0.7
1.3
Median
3.1
1.9
1.4
2.0
2.1
2.3
Mean
3.5
2.2
1.7
2.1
1.9
3.0
90th Percentile
6.4
4.3
3.3
2.8
4.3
5.6
Number of Countries
72
72
18
16
18
20
28 / 37
Raw vs Adjusted Gaps
7 1.0 MWI
0.50
6 CHE
IRQ
Adjusted APG
5
TZA BWA
ZMB
ALB TJK
4
RWA
ZWE
3
SVN VNM IDN ROM JAM GRC MEX CHL BRA HND GTM BGD GER LCA PAK LBRBOL LKA KEN LVA CIV RUS BTN EGY ARM ETH NPL PAN TUR PHL GBRITA ZAFLTUGHA SLE FRA KOR USA ESP AUS CAN VEN NGA SWE TON NIC EST SWZ PRT SLV KHM BGR HUN JOR ISR SYR PRY NLD FJI
2
1
2
4
UGA
6
8
10
12
14
Raw APG
29 / 37
Alternative Measures of Value Added by Sector
30 / 37
Comparing Macro and Micro Data on Sector Value Added The idea: • Cross check “macro” value added data (from national accounts) with
“micro” data from household income/expenditure surveys. The data: • Use World Bank’s Living Standards Measurement Surveys (LSMS) • Explicit goal of LSMS: household income and expenditure measures
31 / 37
Measuring Value Added from Micro Data Agriculture: VAa
=
X
SE ya,i +
=
J X
L ya,i +
X
K ya,i ,
i
i
i
SE ya,i
X
pj xihome + ximarket + xiinvest − COSTSa,i , ,j ,j ,j
j=1
Non-agriculture: X
VAn
=
X
SE yn,i
=
REVn,i − COSTSn,i .
i
=
household
SE yn,i +
i
L yn,i +
K yn,i ,
i
i
and
X
j = agriculture commodity.
32 / 37
Comparison of Macro and Micro APG
Agriculture Share of Country
Employment
Value Added
Micro
Macro
Micro
APG Macro
Micro
Armenia (1996)
34.2
36.8
32.8
0.9
1.1
Bulgaria (2003)
14.1
11.7
18.4
1.2
0.7
Cote d’Ivoire (1988)
74.3
32.0
42.1
4.7
4.0
Guatemala (2000)
40.2
15.1
18.7
3.8
2.9
Ghana (1998)
53.9
36.0
33.3
2.2
2.3
Kyrgyz Republic (1998)
56.9
39.5
39.3
2.0
2.0
Pakistan (2001)
46.9
25.8
22.6
2.5
3.0
Panama (2003)
27.0
7.8
11.8
4.4
2.7
South Africa (1993)
11.0
5.0
7.0
2.3
1.7
Tajikistan (2009)
41.0
24.7
30.1
2.1
1.6
33 / 37
Comparison of Macro and Micro APG
Agriculture Share of Country
Employment
Value Added
Micro
Macro
Micro
APG Macro
Micro
Armenia (1996)
34.2
36.8
32.8
0.9
1.1
Bulgaria (2003)
14.1
11.7
18.4
1.2
0.7
Cote d’Ivoire (1988)
74.3
32.0
42.1
4.7
4.0
Guatemala (2000)
40.2
15.1
18.7
3.8
2.9
Ghana (1998)
53.9
36.0
33.3
2.2
2.3
Kyrgyz Republic (1998)
56.9
39.5
39.3
2.0
2.0
Pakistan (2001)
46.9
25.8
22.6
2.5
3.0
Panama (2003)
27.0
7.8
11.8
4.4
2.7
South Africa (1993)
11.0
5.0
7.0
2.3
1.7
Tajikistan (2009)
41.0
24.7
30.1
2.1
1.6
33 / 37
Comparison of Macro and Micro APG
Agriculture Share of Country
Employment
Value Added Micro
APG
Micro
Macro
Macro
Micro
Armenia (1996)
34.2
36.8
32.8
0.9
1.1
Bulgaria (2003)
14.1
11.7
18.4
1.2
0.7
Cote d’Ivoire (1988)
74.3
32.0
42.1
4.7
4.0
Guatemala (2000)
40.2
15.1
18.7
3.8
2.9
Ghana (1998)
53.9
36.0
33.3
2.2
2.3
Kyrgyz Republic (1998)
56.9
39.5
39.3
2.0
2.0
Pakistan (2001)
46.9
25.8
22.6
2.5
3.0
Panama (2003)
27.0
7.8
11.8
4.4
2.7
South Africa (1993)
11.0
5.0
7.0
2.3
1.7
Tajikistan (2009)
41.0
24.7
30.1
2.1
1.6
33 / 37
Income and Expenditure Per Worker and APGs
Country
APG Micro
Income per
Expenditure per
Worker Ratio
Worker Ratio
Armenia (1996)
1.1
0.7
0.9
Bulgaria (2003)
0.7
1.4
1.2
Cote d’Ivoire (1988)
4.0
3.5
3.2
Guatemala (2000)
2.9
3.2
2.4
Ghana (1998)
2.3
2.0
1.9
Kyrgyz Republic (1998)
2.0
1.3
1.8
Pakistan (2001)
3.0
3.2
1.4
Panama (2003)
2.7
2.8
2.1
South Africa (1993)
1.7
1.7
1.2
Tajikistan (2009)
1.6
1.2
1.1
34 / 37
Different Labor Shares Across Sectors? Production functions with different labor shares Ya = Aa Lθa a Ka1−θa
and
Yn = An Lθnn Kn1−θn
In equilibrium APG =
Yn /Ln θa = pa Ya /La θn
Macro evidence on θa , θn • Employment share of agriculture varies a lot across countries; • Aggregate labor share of GDP doesn’t, Gollin (2002) ⇒ θa ≈ θn
Micro evidence on θa , θn • Sharecropping arrangements suggest θa ≈ 0.5 • Econometric estimates: θa ≈ 0.5 − 0.6 35 / 37
Why are Residual Gaps So Large?
• Yet more measurement error
– Herrendorf and Schoellman (2013) • Selection of more productive workers out of agriculture
– Lagakos and Waugh (2013), Young (2013) • Risk of migrating?
– Harris and Todaro (1971), Bryan, Mobarak, Chowdhury (2012), others • Much room for future work
36 / 37
Conclusion • Typical country has large agricultural productivity gap
• Particularly large in developing countries
• Better measurement reduces gap down to around two on average
• Large gaps also present in household survey data
• Needed: better understanding of why residual gaps so large
37 / 37