Shining a light on the Indonesian oil palm and development debate with big data
-
Upload
anucrawfordphd -
Category
Economy & Finance
-
view
219 -
download
0
description
Transcript of Shining a light on the Indonesian oil palm and development debate with big data
Shining a light on the Indonesian oil palm and development debate, with ‘Big Data’ Ryan B. Edwards Arndt-Corden Department of Economics Crawford School of Public Policy College of Asia and Pacific
Crawford PhD Conference 2014 18 November 2014
Source: Wikipedia
Does palm oil production make people in Indonesia better off?
2
Indonesia now exports over 16 times the palm oil it did in 2000…
3 Source: The Atlas of Economic Complexity, 2014 (http://www.atlas.cid.harvard.edu/ )
.. to all over the world…
4 Source: The Atlas of Economic Complexity, 2014 (http://www.atlas.cid.harvard.edu/ )
.. and is now the largest supplier (by over 13,000,000,000,000 tonnes!)
5
Source: Indexmundi (2014), http://www.indexmundi.com/agriculture/?commodity=palm-oil&
But palm oil can have high ecological costs..
6
..leading to boycotts, divestment, and more
7
… while poverty remains widespread..
8 Source: World Bank Indonesia (2014), http://www.worldbank.org/content/dam/Worldbank/Feature%20Story/EAP/Indonesia/Poverty%20infographic%20revised.png
..and we have no systematic empirical evidence on the development effects of Indonesia’s recent palm oil expansion.
9
10
What I do
Use the World Bank’s first public sub-national public database to estimate the short-run household welfare effects
of palm oil production in Indonesia, at the district level
11
What I find
Q: Does palm oil production make people in Indonesia better off?
A: Yes, on average, in the short run
But, it likely depends on who, how
12
Introducing the data: INDONESIA DATABASE FOR
POLICY AND ECONOMIC RESEARCH (DAPOER)
13
14
Main variables
• Per capita palm oil production, by district – Since 1997 from Tree Crop Estate Statistics of Indonesia, Ministry of
Agriculture, via DAPOER
– Denominated by district population, for consistency (e.g., reformasi)
– Non-producing districts coded as zeros; kept level to retain control
15
16
National variation in palm oil production intensity, by province, 2013
Source: http://www.simreg.bappenas.go.id/
Main variables
• Per capita palm oil production, by district – Since 1997 from Tree Crop Estate Statistics of Indonesia, Ministry of
Agriculture, via DAPOER
– Denominated by district population, for consistency (e.g., reformasi)
– Non-producing districts coded as zeros; kept level to retain control
• Per capita household monthly expenditures, by district (IDR) – Based on district aggregates in the National Socioeconomic Survey
(SUSENAS) from Statistics Indonesia, via DAPOER
– Reasonable proxy for average household welfare in each district, i.e., not equal to local GDP, but increasing nationwide 1997-2010
– Put into natural logarithms for appropriate form and easier semi-elasticity interpretations
17
18
11
12
13
14
15
16
0 1 2 3 4 5Per capita palm oil production, district (tons)
Log per capita HH expenditure Fitted values
A naïve comparison reveals a positive correlation
Source: DAPOER
IDENTIFICATION CHALLENGES
19
Time trend
IDENTIFICATION CHALLENGES
20
Time trend Reverse causality
Expenditure may influence decisions and ability to produce
IDENTIFICATION CHALLENGES
21
Time trend Reverse causality
Expenditure may influence decisions and ability to produce
Time-invariant omitted variables
District and regional; observable and unobservable
IDENTIFICATION CHALLENGES
22
Time trend Reverse causality
Expenditure may influence decisions and ability to produce
Time-invariant omitted variables
District and regional; observable and unobservable
Time-varying omitted variables
Common shocks; District- or region-specific
Estimated equation
𝑙𝑜𝑔 𝑦𝑖,𝑡 = 𝛽1 + 𝛽2 𝑃𝑖,𝑡 + 𝛽3𝑇 + 𝑣𝑖 + 𝑒𝑖,𝑡
𝑦𝑖,𝑡 = outcome of interest, district i, time t
𝑃𝑖,𝑡= per capita palm oil production
𝑇 = time trend
𝑣𝑖 = district fixed effect
𝑒𝑖,𝑡 = district-clustered robust error term
23
Estimated equation
𝑙𝑜𝑔 𝑦𝑖,𝑡 = 𝛽1 + 𝛽2 𝑃𝑖,𝑡 + 𝛽3𝑇 + 𝑣𝑖 + 𝑒𝑖,𝑡
𝑦𝑖,𝑡 = outcome of interest, district i, time t
𝑃𝑖,𝑡= per capita palm oil production
𝑇 = time trend
𝑣𝑖 = district fixed effect
𝑒𝑖,𝑡 = district-clustered robust error term
Main estimators (i.e., within, fixed effects) focus on the yearly changes within each district (i.e., short-run)
24
Identifying assumption
Within-district palm oil production changes are exogenous to changes in average household
expenditures in the same district, conditional on time-varying common factors and district-
specific time-invariant factors
25
Plausibility of the identifying assumption
• Production takes many years, i.e., cannot be contemporaneously endogenous to household spending
26
Plausibility of the identifying assumption
• Production takes many years, i.e., cannot be contemporaneously endogenous to household spending
• District variation in palm oil production is mostly affected by ‘random’ centralized land use decisions and climatic conditions
27
Plausibility of the identifying assumption
• Production takes many years, i.e., cannot be contemporaneously endogenous to household spending
• District variation in palm oil production is mostly affected by ‘random’ centralized land use decisions and climatic conditions
• Remaining threat is time-variant district-specific OVB – Province-year and island-year fixed effects yield similar results
– Diff-GMM and Sys-GMM yield similar results ; instrumenting with palm oil price (weak), total arable land, and palm oil land yield similar results
– Province rainfall, humidity, and temperatures are weak IVs
– Work underway on alternative external IVs and identification strategies
28
Main results
29
Dependent variable Log per capita household expenditure (IDR)
Column (1) (2) (3) (4) (5)
Estimator OLS BE GLS RE FE FE
Per capita palm oil production
(tons)
0.09*** 0.11** 0.06*** 0.06*** 0.05**
(0.02) (0.06) (0.02) (0.02) (0.02)
Time trend (T) 0.13*** 0.1*** 0.13*** 0.13***
(0.00) (0.01) (0.00) (0.00)
Year dummy N N N N Y
District fixed effects N N N Y Y
N observations 3939 3939 3939 3939 3939
N districts 459 459 459 459 459
Avg. obs. per district 8.6 8.6 8.6 8.6 8.6
Overall F 3160 55 9955 4836 1169
R-squared (within) 0.55 0.18 0.55 0.81 0.82
LOOKING BELOW THE AVERAGE
30
Distributional results
31
0
0.05
0.1
0.15
0.2
0.25
Elasticity, percentage change from an additional ton of palm oil production per capita, district level
Log bottom 20% TOTAL household expenditure
Notes • All years, all districts • Within estimator • District fixed effects • Year fixed effects • Robust district-clustered
90 percent confidence intervals
Log poverty rate
SECTOR HETEROGENEITY (preliminary results)
32
Private and smallholder sectors are similar sized
33
0
50000
100000
150000
200000
250000
300000
350000
2005 2010
Ave
rage
dis
tric
t p
rod
uct
ion
, to
tal (
ton
s)
Total
Private
State-owned
Smallholders
Productivity is similar across sectors
34
0
500
1000
1500
2000
2500
3000
3500
4000
4500
Private State-owned Smallholder
Ave
rage
dis
tric
t yi
ed
(kg
/ha)
2007 2008
Main results, by sector
35
Smallholder N=952, D=159
Private N=516, D=120
State-owned N=207, D=53
Notes • Bars represent semi-
elasticity point estimates
• Whiskers represent 90 percent robust district clustered confidence intervals
• All years and districts where data; no ‘controls’, estimates for producing districts
• Missing data across years and districts
• No reason to suspect missing data are zeros
• Generalised least squares, with district-level random effects
• Time and province-level fixed effects -0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
Elasticity, % change in avg. household expenditures from an extra ton of palm oil production per capita
Distributional results, by sector Notes
• Bars represent semi-elasticity point estimates
• Whiskers represent 90 percent robust district clustered confidence intervals
• All years and districts where data; no ‘controls’, estimates for producing districts
• Missing data across years and districts
• No reason to suspect missing data are zeros
• Generalised least squares, with district-level random effects
• Time and province-level fixed effects
36
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6 Elasticity, % change from ton of palm oil production per capita, district level, by sector
private
state-owned
smallholder
Bottom 20% total
household
expenditure
Poverty rate
Implications so far
• In short run, increasing palm oil production in all sectors has tended to be good for average household well-being throughout the income distribution
• Slowing production is, in the short run, unambiguously harmful for average household welfare at the district level
37
Implications so far
• In short run, increasing palm oil production in all sectors has tended to be good for average household well-being throughout the income distribution
• Slowing production is, in the short run, unambiguously harmful for average household welfare at the district level
• Different sectors are likely to have differential effects at the lower end of the income distribution
• Key policy challenge environmental / human trade-off, maintaining production without adverse environmental effects
38
39
Next steps
For this paper
• Obtain ‘purer’ short-run causal estimates, i.e., improve dataset, add covariates, and add improved instruments
• Examine lagged effects and dynamics, and regional dynamics (i.e., by island)
• Disaggregate expenditure effects by expenditure type and on savings
40
Next steps
For this paper
• Obtain ‘purer’ short-run causal estimates, i.e., improve dataset, add covariates, and add improved instruments
• Examine lagged effects and dynamics, and regional dynamics (i.e., by island)
• Disaggregate expenditure effects by expenditure type and on savings
Research agenda (related to this topic)
• Medium and long-run effects
• Effects on labour markets and employment, and on other sectors
• Effects on human capital (i.e., decreased secondary participation)
• Efficiency analysis, i.e., how to increase output holding land constant?
41
Acknowledgements
42
Contact
Email: [email protected]
Twitter: @ryanbedwards
Web: https://crawford.anu.edu.au/people/phd/ryan-barclay-edwards
43