Big question in this lecture Does infrastructure
Stockholm Doctoral Course Program in Economics
Development Economics I — Lecture 8Infrastructure
Masayuki Kudamatsu
IIES, Stockholm University
Big question in this lecture
Does
infrastructure
promote It’s NOT easy to empirically identify the impact of infrastructure Endogenous placement of
- infrastructure
Infrastructure may be a response to
- rising economic opportunities (reverse causality) Govt. may target poor areas to
- improve their economic conditions
What’s the mechanism?
- Dams (Duflo & Pande 2007 QJE)
- Mobile phones (Jensen 2007 QJE)
- Excellent example of when DID works best
- Electricity (Dinkelman 2008)
- Careful analysis on heterogenous treatment effect
- Railroads (Donaldson 2008)
Duflo and Pande (2007, QJE)
- Using geography as instrument to
- estimate the impact of infrastructure
1-1 Research questions
What’s the impact of irrigation dams
- on agricultural production and rural poverty? What’s the distributional
- consequence of building irrigation dams?
1-2 Data
Annual agricultural production,
- 1971-1999, for 271 districts Poverty data in 1973, 83, 87, 93, 99
- for 374 districts Dams: location (nearest city) and
- date of completion from World Registry of Large Dams
Ratio to district area of river area with gradient more than 6%, 3-6%, 1.5-3% Data source: GTOPO30 (elevation
- at 30-arc second grid space) & Digital Chart of World (river drainage network) Identify GTOPO30 cells where >rivers flow Calculate gradient in such c
1-3 Empirical strategy: 1st stage
!
4
- D = ν + µ α (RGr ∗ ¯ D )
ist i st k ki st k
=2
!4
- β(M ∗ ¯ ) + α (RGr ∗ l )
i D st k ki t k =2
- ω ist D ist : # of dams in district i of state s in
- k : 2 for 1.5 to 3%; 3 for 3-6%; 4 for above 6%
- River flowing at some gradient: ideal for irrigation dams
- Very steep river: ideal for power generation dam
D st : # of dams in India in year t (Figure III) multiplied by fraction of dams in state s in 1970
- ¯
- Why not the actual # of dams in state
- i
ν : district FE
- st
µ : state-year FE (different trends
across states)
M
- i
: area, elevation, overall gradient,
river length i st
Why (M ∗ ¯ D ) included?
- l
- t : year dummies
(RGr ∗ l ) included?
ki t Why1st stage results (Table II)
Dictricts w/ more river gradient
- 1.5-3% or above 6%: more dams built Dictricts w/ more river gradient
- 3-6%: less dams built
Empirical strategy: 2nd stage
U Uy ist = γ i + η st + δD ist + δ D Z U U
ist
- Z ist δ + Z ist δ Z + ε ist U
U
w/ ˆ D ist , ˆ D , Z ist , Z ist as instruments
ist
- ist : outcome variable
y
- i : district FE
γ
- st U
η : state-year FE
D : # of dams in all upstream
Z
- U
ist i st ki t
: vector of M ∗ ¯ D , RGr ∗ l
Z
ist : vector of M i ∗ ¯ D st , RGr ∗ l
ki t for
upstream districts ˆ
- ist U
D ist ,: fitted value for D
ˆ D
- ist : the sum of fitted values for
D over all upstream districts
ist Empirical strategy: Method
Feasible optimal IV with S.E. robust
- to arbitrary covariance of the residual w/i state (see ft. 15 for how to implement this) Why?
- Autocorrelation at state level • Feasible GLS: more efficient than • OLS with S.E. clustered ⇒ Small effect more likely to be detected (Power of test ր)
IV estimates: treatment effect for
- compliers (“Local Average Treatment Effect”) cf. Angrist and Imbens (1994), Imbens (2007) Dinkelman (2007) takes this issue
- seriously
1-4 Results 1: Impact on agriculture (Table III)
1 additional dam in upstream ⇒ Irrigated areas ր by 0.33%
- Production/Yield of 6 major crops
- ր by 0.34/0.19% Production of water-intensive crops
- ր by 0.47% No significant change in
- non-water-intensive crops
Rainfall shocks (deviation from 1971-99 mean) on agricultural production: Mitigated if dams built upstream
- Amplified if dams built in own
- districts
⇐ Water use restricted by govt to keep reservoir full Results 3: Impact on rural welfare (Table VIII)
Head count ratio: 0.77% pt ր by 1 more dam in own
- district 1.5% pt ց by 1 more dam upstream
- No impact on district-level
- population or in-migration (Table
VII)
Results 4 (Table IX)
Impact of dams on poverty in own districts: mitigated if tax collection in colonial days done by farmers, not by landlords cf. Banerjee and Iyer (2005): non-landlord districts ⇒ public goods ր
- agricultural productivity ր
- ⇒ Compensation for losers works w/
1-5 Taking Stock
Use geography interacted with
- nation-wide trends & inter-state variation in infrastructure-building to credibly estimate the impact of infrastructure Distributional consequences of
- infrastructure and how losers can be compensated: Important topic
Dolandson (2008)
- Moving beyond reduced-form
- evidence
2-1 Research questions
Did the expansion of railroads in
- colonial India promote agricultural development? If so, was it due to gains from trade
- caused by reduced trade costs?
2-2 Data
Sample: 239 districts in colonial
India Annual panel, 1861-1930
- Outputs & retail prices of 17
- principal crops Bilateral trade flows for 85
- commodities, 1880-1920 Daily rainfall from 3614 stations,
- 1891-1930
2-3 Background
Transportation means in colonial India Bullocks on roads (<20-30km/day)
- River (65km/day downstream,
- 15km/day upstream) Coast (>100km/day) >Railroads (600km/
2-4 Model (Eaton-Kortum 2002)
D districts, each denoted by d or o
- K commodities, each w/ a
- continuum of varieties Unit mass of identical agents in
- each district Each owns L
- d units of land,
immobile & supplied inelastically Land: only factor of production
- Land rental rate r
- d
k
dj $
= 1
k
µ
k
where %
σk σk −1
σk −1 σk
Model: Preference
(j))
k d
(C
1
µ k ln " #
K ! k =1
ln U d =
- Cobb-Douglas over commodities ⇒ µ
⇒ CES over varieties (j) of each k
- Indirect utility per acre, W , is given
d
by (cf. equation (9) on p.15): ! µ r
k d
ln W d = µ k ln
k
˜ p
d k !
r d = ln µ ln µ + & k k k
µ k
(˜ p )
k d
" $ k 1 '1 1 k k 1 −σk −σ Model: Production k
z (j): amount of variety of j of
d
commodity k produced by 1 unit of land in district d Follows type-II extreme value
- distribution k
−θk k z −A d
(z) = e F
d k
- d
A : how likely productivity is high
- k
θ : how variable productivity is Model: Commodity market
- Many competitive firms w/i district
⇒ Each firm makes zero profit ⇒ Domestic price of k (j) produced at home: p
k dd
(j) = r
d
/z
k d
(j) Model: Trade
- k
To export 1 unit of k from district o
≥ 1 units must be to d , T
od produced in o (iceberg trade cost). k k k od md k ≤ T om T T
- T = 1 (normalization) • oo k Railroads reduce T • od
⇒ Import price of k from o:
k k k k k
p (j) = T p (j) = r T /z (j)
o oo o od od od Model: Trade (cont.)
Agents: indifferent about where
- each k (j) is made
k
⇒ They pay the cheapest p (j),
od k
denoted by p (j)
d
⇒ Its distribution is given by " $ % D k k
−θk θk A T p − (r ) o o od k o=1
G (p) = 1 − e
d Model: Trade (cont.)
⇒ k (j)’s expected price in district d: " $ − ! D 1
k k k k
−θ θk k
E [p (j)] = λ A (r T )
o d 1 o od
o
=1 k
- d
E [p (j)] is also the average price of
commodity k varieties, denoted by
k
p
d This is the price of each commodity
- observed in the data.
- (see fn. 16 of Eaton-Kortum)
Prob. for district d to import k (j)
from o:
k k −θ k
A (r T )
o
o
k odπ =
od % D k k −θ k
A (r o T )
o o od
=1
k- od
π is also the fraction of varieties of
k that district d imports from o Eaton-Kortum’s result no. 2
- k
Price of a variety that district d
imports from o: distributed by G (p)
d See ft. 17 of Eaton-Kortum
- ⇒ District d’s expenditure for imports from o: same across o for each k
k k k
⇒ π = X /X where
od od d k od
X : Trade flow from o to d for
- commodity k k X : d ’s total expenditure on • d
- k k
Land: inelastically supplied
- o od k
If A UP or T DOWN
⇒ π UP & demand for land in o UP
od
⇒ Rental price r o should go up Land rental prices r d ’s solve the
- following system of equations ! !
k
= π µ , ∀o ∈ {1, ..., D} r o L o k r d L d
od k d
2-5 Taking Model to Data
6 empirical steps to estimate the impact of railroads:
1. Trade costs
2. Trade flows
3. Market integration
4. Mean income
5. Income volatility
6. Quantitive assessment of the model Prediction 6
Indirect utility per acre for agents in
- d , W , is given by
d ! !
µ µ
k k k k
ln W = Ω+ ln A − ln π
d d dd
θ k θ k
k k k
- dd
π : (inverse of) trade openness
Trade costs & other districts’
- productivity and land affect welfare
k
only via π Prediction 6 can be used for identifying the mechanism of the railroad impact on welfare % µ Regress ln W
- dt on RAIL dt w/ k
k
A as control (reduced-form
k d θ k
estimation) % µ
k
k Then add π as additionalk dd
θk
regressor Extent to which coeff. on RAIL
- dt
gets small: how much the model Testing Prediction 6 ! !
µ k µ k
k k
ln W = Ω + ln A − ln π
d d dd
θ θ
k k k k
- d : real agricultural income per
W
acre
Observed from each commodity’s
- yield per acre and price & land areas
- k : k ’s consumption share
µ
Observed from outputs and trade
µ k µ k
k k
ln W = Ω + ln A − ln π
d d dd
θ θ
k k k k
- k k
We need to estimate unobserved
A & π as functions of exogenous
d dd
variables We also need to estimate θ
- k
- od
Estimate the trade cost T in the
model Check whether railroads really
- reduced the trade cost
Remember average price of
- commodity k in d is
D 1 " $ ! − k k k k −θ θk k
p = λ A (r o T )
d 1 o od o =1 k
- od
We can infer T from commodity k
produced only in one district Denote this commodity by o. Then
ln p = β + β + φ t
d ot dt od o
- δ ln TC(R ) + ε
t odt odt
Commodity o: salt produced only in
- a particular district
R
- t
: Railway network in year t Step 1: Specification (cont.) o o o o
ln p = β + β + φ t
d ot dt od o
- δ ln TC(R ) + ε
t odt odt
Prediction 1 tells us:
o o
- ot ot
β = ln p
- t ) odt : time-variant
δ ln TC(R
component of trade cost btw. o & d
o Step 1: Measuring TC (R ) t odt
LCR (R t , α) odt : lowest-cost route distance in railway-equiv. km α road river coast = (α , α , α ): trade cost
- per km relative to railroad Existing transportation network + R
- t
⇒ shortest-distance btw. o & d for each α α : estimated by NLS together with δ
Railroads did reduce trade cost per
- ˆ km ( > 1)
More than reported relative freight
- rates (α = (4.5, 3.0, 2.25)) suggest
- t odt
Over & above linear trends
Important as LCR (R , α) ↓ over
- time
Elasticity of trade cost to distance in
- rail-equiv. km: 0.247 Robust to railroad link d
Check whether railroads increased
- trade flows
k
k
o
θ Estimate A & in the model Step 2: Prediction 2
k k k
−θ k
Remember
X A (r o T )
k o
od od
π = =
od k D % k k −θ k
X (r )
d A o T o o =1 od
- k k k
So trade flow is given by
ln X = [ln A − θ ln r ] − θ ln T
k o k od o od ! D k k k −θ k
- [ln (r ) + ln X ]
A o T
o od d
- β
- β
- φ
= ln A
−θ
k d
−θ k
)
k od
(r o T
k
o
A
D o =1
= ln %
k dt
− θ k ln r o
k o
k ot
Prediction 2 suggests:
k odt
ˆ δ ln LCR(R t , ˆ α ) odt + ε
t −θ k
k od
k od
k
dt
k ot
= β
k odt
ln X
Step 2: Specification
- β
- ln X
- β
k ln T k od
- Other terms:
- k
Estimate for each k to obtain ˆ θ ’s
S.E.: bootstrapped • See Deaton (1997) for references on • bootstrap
- k k
Then we obtain
ln ˆ A = ˆ β + ˆ θ k ln r ot
o ot
where r ot is measured by nominal agri. GDP per acre Step 2: Results (Table 3)
ˆ −θ δ: significantly negative on
- k
average (column 2) ⇒ Shorter railway-equiv. distance increased trade flow
No significant heterogeneity in this
- coefficient across commodities by (1) weight per unit value & (2) railroad freight class (column 3)
Extract exogenous component in ln ˆ A :
o k k k
ˆ β + ˆ θ k ln r ot = β + β + β ot
o t ot k k
- κRAIN + ε
ot odt k
- ot
RAIN : total rainfall between
sowing and harvest dates for k in o κ ˆ : 0.441 (se: 0.082)
k
k- Check if railroads integrate markets
- Remember p
)
k o
depends more on A
k d
if T od ↓ 2. p
k d
depends less on A
k d
1. p
−θ k $ − 1 θk
k od
Step 3: Prediction 3
T
o
(r
k o
A
1 " D ! o =1
k
= λ
k d
if T od ↓
- χ
- χ
dt
k t
k d
× RAIL odt
k ot
RAIN
o ∈N d
) !
1 #N d
4 (
k ot
RAIN
o
∈N d) !
d
1 #N
(
3
dt
× RAIL
2 RAIN k dt
1 RAIN k dt
= χ
k dt
ln p
Step 3: Specification
- χ
- β
- β
- β
- ε
k dt
Prediction 3: χ
1 < 0, χ 2 > 0, χ
3 < 0, χ
4 < 0 k k kln p = χ
2 RAIN × RAIL dt dt dt dt !
1
k
- χ ( ) RAIN
3 ot
#N d
o
! ∈N d1
k
- χ ( ) × RAIL
4 RAIN odt ot
#N
d
o
∈N dk k k
- β + β + β dt + ε
d t dt Step 3: Results (Table 4) ∗∗∗
- 1 ∗∗
χ ˆ = −0.402
- 2
χ ˆ = +0.375 : railroad link reduces
the dependence of price on own district rainfall χ
3 = −0.021: w/o railroad link,
ˆ
neighboring districts’ rainfall does not affect price
∗∗∗
- 4
χ ˆ = −0.082 Step 3b: Model evaluation
- k
If model correct, predicted
commodity price ˆ p should be
dt k
close to observed p
dt
Solving the model & plugging
- estimated parameters & observed
k
exogenous variables to obtain ˆ p
dt
⇒ “out-of-sample” test Step 3b: Model evaluation (cont.)
- Then estimate ln p
k dt
= β
k d
k t
- β
- β
- ω ln ˆ p
- ε
dt
k dt
k dt
which yields ˆ ω = 0.913. Step 4: Prediction 4
Solve system of equations (6) to
- obtain r for the case D = 3, K = 1
d
Then conduct comparative statics
- on W d w.r.t. T od W
- d ↑ if T od ↓: arrival of railroads
increase welfare
- ′
↓ if T ↓: railroads in other W
d oo
districts decrease welfare Step 4: Specification
ln (W ) = β + β + γRAIL
o o t ot !
1
- ψ( ) RAIL + ε
dt ot
#N
o
d ∈N oPrediction 4: γ > 0, ψ < 0
- Estimated by OLS, assuming
- exogenous railroad placements
Arrival of railroad
- ⇒ Real agri GDP per acre ↑ by 18.2% Railroads in N
- o
⇒ Real agri GDP per acre ↓
Treatment externality: ignoring this
- yields understimation (column 1) Distributional consequences of
- railroads
Step 4: validity checks
1. Placebo tests: estimate impact of proposed but never built railroads ⇒ No effect (Table 6)
2. IV estimation: 1876-78 rainfall deviation from long-run mean as IV ⇒ IV estimate: similar magnitude to
OLS (Table 7)
3. Bounds check: estimate coefficients separately for each Step 5: Prediction 5 k
1. A UP ⇒ W UP
d d k
2. T DOWN ⇒ W less responsive
d od k
to A
d
⇒ Trade cost reduction reduces volatility of income
- ψ
dt
θ k ˆ κ RAIN
ˆ µ k ˆ
k
× " !
3 RAIL
ot
k ot $
ˆ κ RAIN
ˆ θ k
k
ˆ µ
2 " ! k
RAIL
d ∈N o
) !
1 #N o
(
1
ot
) = γRAIL
o
ln (W
Step 5: Specification
- ψ
k ot $
- ψ
- β o + β t + ε ot
2 > 0, ˆ ψ 3 < 0
Step 5: Results (Table 9)
- Indeed ˆ ψ
⇒ Railroads reduce income volatility
- ψ
2 " ! k
ln ˆ π
k
ˆ µ
k ot $
ˆ κ RAIN
k
θ
ˆ µ k ˆ
k
× " !
k ot $
θ k ˆ κ RAIN
ˆ µ k ˆ
dt
RAIL
d ∈N o
) !
o
1 #N
(
1
ot
) = γRAIL
o
ln (W
Step 6: Specification
- ψ
- ψ
k oot
- η !
- β o + β t + ε ot
1 = ψ 3 = 0, ψ 2 = 1, η = −1 !
1 ln (W o ) = γRAIL ot + ψ
1 ( ) RAIL dt
#N o
d " ! $ ∈N o
µ ˆ
k k
- ψ κ ˆ RAIN
2 ot
ˆ θ k
k " ! $
µ ˆ
k k
- ψ RAIL × κ ˆ RAIN
3 ot ot
ˆ θ
k ! k
µ ˆ k
k
- η ln π ˆ + β + β + ε
o t ot
Step 6: Results (Table 10)
Once the openness term is included as a regressor, Railroad coefficients become
- insignificant and close to zero.
ˆ ψ
- 2 is close to 1, η is close to -1! ˆ
⇒ Model explains a very large portion of the railroad welfare impact seen in Steps 5 & 6
Future research in the literature
(my own view)Impact on industrial development
- Distributional consequences of
- infrastructure construction Non-economic impact of
- infrastructure
Public service delivery should be
- affected by infrastructure, too.
Political economy of infrastructure
References for the lecture on infrastructure Banerjee, Abhijit V., and Lakshmi Iyer. 2005. “History, Institutions and Economic Review Performance: The Legacy of Colonial Land Tenure Systems in India.” American Economic 95(4): 1190-1213. ! Deaton, Angus. 1997. The analysis of household surveys. World Bank Publications. ! Dinkelman, Taryn. 2008. “The effects of rural electrification on employment: New evidence from South Africa.” Donaldson, Dave. 2008. “Railroads of the Raj: Estimating the Impact of Transportation