Economic Development and the Spatial Allocation of · PDF fileEconomic Development and the...

Economic Development and the Spatial Allocation of Labor:Evidence From Indonesia

Gharad Bryan (LSE)Melanie Morten (Stanford)

May 19, 2015

Gains from Spatial Reallocation of Labor?

Remove observables US?

Questions

1. Could productivity be increased by moving people?I Not obvious: selection (could go either way).I Quantitatively, how large are the effects?

2. If so, why do people not move?I Amenity differences (traditional in urban economics).I Costs of movement (understudied).

The Paper

1. Some facts:I Show that selection and movement costs are present.

2. Theory:I Model consistent with facts.

3. Empirical:I Structurally estimate on census data (US and Ind).

4. Quantitative:I How much do movement costs and amenity differences

explain?

Main Results

I Structural Estimates:I Movement costs have decreased over time in Indonesia.I Movement costs are much higher than in the US.I Amenity is negatively correlated with productivity.

I Quantitative Results:I Over time 1995-2012:

I Reduced movement cost account for about 25% growth.I Changes in amenity account for about 25% growth.

I Comparison to the US:I higher movement costs explain about 5% of US-Indonesia gap

.

Some Related Literature

I Rural-urban income gapsYoung (2014); Lagakos and Waugh (2013); Gollin, Lagakos and Waugh (2011);

Restuccia, Yang and Zhu (2008); Caselli (2005)

I Experimental evidence: large returns to migratingBryan, Chowdhury and Mobarak (2014)

I Migration costs are largeMorten and Oliveira (2014); Kennan and Walker (2011); Topel (1986)

I Large literature on misallocation of factorsHsieh, Hurst, Klenow and Jones (2014); Hsieh and Klenow (2009); Hsieh and

Moretti (2014), Desmet and Rossi-Hansberg (2014)

(Also, Eaton and Kortum 2002)

Outline

Reduced Form Facts: Costly Migration and Selection

Model of Spatial SortingSorting Over SpaceGeneral Equilibrium

Structural EstimatesIdentificationStructural Estimates: Highlights

GE Counterfactuals: Reduce Migration Costs

Data

I IndonesiaI 1976 SUPAS; 1995 SUPAS; 2011 SUSENAS; 2012 SUSENAS

I ∼ 27% of hoh do not live in regency of birth.I ∼ 60% of these cross a provincial border.

I Main sample: male hoh who are wage workers, aged 15-65I Supplement with IFLS 1993, 1997, 2000, 2007.

I United StatesI 1990 Census (5% sample) and 2010 ACSI 40% do not live in state of birth

Rural-urban migration

Four motivating facts

1. The gravity equation.I Long history: Ravenstein (1889), Grogger and Hanson (2011)

2. Distance travelled predicts wage.I Distance as proxy for migration costI We non-parametrically estimate costs later

3. Proportion of migrants moving negatively predicts wage.

4. Distance does not predict wage controlling for proportion.

Fact 1: Gravity Equation for Migration

(1) (2) (3) (4)1976 1995 2011 2012

Dep var: Log Prop. Mig. b/se b/se b/se b/se

Log Distance -0.47*** -0.60*** -0.61*** -0.61***(0.0041) (0.0017) (0.0016) (0.0016)

Destination FE Yes Yes Yes Yes

N 43160 166899 210373 210491

ln πdo = αd + γo + β ln distdo + εdo

Fact 2: Migration Distance Predicts Wages

(1)Dep var: Log Wage b/se

Log Distance 0.038***(0.0018)

Destination FE YesOrigin FE Yes

N 58882Correlation

Year is 1995; Regency; Every-one.

ln wido = αd + β ln distdo + εido

Fact 3: Migration Proportion Negatively Corr with Wage

(1) (2)Dep var: Log Wage b/se b/se

Log Distance 0.038***(0.0018)

Log Proportion -0.064***(0.0018)

Destination FE Yes YesOrigin FE Yes No

N 58882 58882Correlation

Year is 1995; Regency; Everyone.

ln wido = αd + βπdo + εido

Fact 4: Distance only through Proportion Migrating

(1) (2) (3)Dep var: Log Wage b/se b/se b/se

Log Distance 0.038*** -0.0044**(0.0018) (0.0021)

Log Proportion -0.064*** -0.068***(0.0018) (0.0024)

Destination FE Yes Yes YesOrigin FE Yes No No

N 58882 58882 58882Correlation -0.750

Year is 1995; Regency; Everyone.

ln wido = αd + β1distdo + β2πdo + εido

Table: 1976 Table: 2011 Table: Urban 1995 Table: Mig Only 1995

Outline

Reduced Form Facts: Costly Migration and Selection

Model of Spatial SortingSorting Over SpaceGeneral Equilibrium

Structural EstimatesIdentificationStructural Estimates: Highlights

GE Counterfactuals: Reduce Migration Costs

Outline

Reduced Form Facts: Costly Migration and Selection

Model of Spatial SortingSorting Over SpaceGeneral Equilibrium

Structural EstimatesIdentificationStructural Estimates: Highlights

GE Counterfactuals: Reduce Migration Costs

Human Capital and Wages

I N locations.

I Location specific human capital hd drawn from

F(h1, . . . , hN) = exp

[

N

∑i=1

h− θ

1−ρ

d

]1−ρ

I ρ ↑: people are good at everything.

I θ ↓: more dispersion within each destination (θ = θ1−ρ )

I Wage earned at location d

waged = wdhd

Hsieh, Hurst, Klenow and Jones (2013)

Utility Maximization

I Utility: location amenity, consumption and time with family:

Udo = αdcβt1−β

I Maximize subject to

t ≤ T, c = wdhd(T − t), t = t(1− τdo)

I t: time away from workI T: total hours available.I τdo: time to travel o to d (symmetric).I (specific interpretation here, but could be more general)

I Solution:

Ud =

(βT(α

1β

d ((1− β)(1− τdo))1−β

β wdhd

)β

= (wdohd)β

Selection

I Results on Frechet distribution imply:

1. Migration: πdo =wθ

do∑N

s=1 wθso

2. Sorting: E(hdo) = Γ(

1− 1θ(1−ρ)

)π− 1

θdo

Relation to Motivational Evidence

I From the above equations we get

ln(πdo) = θ ln wd +θ

βlog αd − γo +

θ(1− β)

βlog(1− τdo) gravity

ln wagedo = ln Γ− 1β

ln αd +1θ

γo −1− β

βln(1− τdo) dist wage

ln wagedo = ln Γ + ln wd −1θ

ln πdo prop wage

I Consistent with the motivational evidence.

Other Implications

I Wage ratio does not depend on productivity!

wagedowaged′o

=αd′(1− τd′o)

αd(1− τdo)

I Why? Three effects of w ↑:I Increase in in wages of those already there.I Movement of migrants in – less selectedI Other home becomes more selected.

I Without amenity differences or movement costs, no spatialvariation in average wages.

I Seems a natural benchmark.

I Productivity still important because migrants more productive.

Outline

Reduced Form Facts: Costly Migration and Selection

Model of Spatial SortingSorting Over SpaceGeneral Equilibrium

Structural EstimatesIdentificationStructural Estimates: Highlights

GE Counterfactuals: Reduce Migration Costs

Closing the Model: Demand Side and Agglomeration

I Location decisions (labor supply) determined by:I Base wages: wd.I Amenity: αd.

I Demand side: Wages

Πd = pd Adhd − wdhd

wd = pd Ad

Ad = AdHγd

I Ad: natural advantage of location d.I γ ≥ 0: agglomeration externalities.I pd: prices (from next slide).

Allen and Arkolakis (2014)

Closing the Model: Prices

I Prices, from Armington Assumption:

Y =

(N

∑i=1

qσ−1

σ

d

) σσ−1

qd = AdHd

I First order condition for maximisation gives

pd =

(Yqd

) 1σ

I So:I High wage wd leads people to move to d.I Increases production from d.I Pushes down wage.I Reduces migration into d.

Closing the Model: Amenities and Congestion

I We assumeαd = αdLλ

d

I λ ≤ 0.I α “natural amenity”.

I Amenities are subject to a congestion cost:I More people chasing little land.I Need for institutions in dense cities.I Pollution.

General Equilibrium

Definition

An equilibrium consists of:

I Prices pd.

I Base wages wd.

I Labor supply Ld (from selection equations).

I Human capital Hd (from selection equations).

Such that:

1. Consumers maximize utility

2. Producers maximize profit

3. Labor markets clear

4. Goods markets clear

Outline

Reduced Form Facts: Costly Migration and Selection

Model of Spatial SortingSorting Over SpaceGeneral Equilibrium

Structural EstimatesIdentificationStructural Estimates: Highlights

GE Counterfactuals: Reduce Migration Costs

Outline

Reduced Form Facts: Costly Migration and Selection

Model of Spatial SortingSorting Over SpaceGeneral Equilibrium

Structural EstimatesIdentificationStructural Estimates: Highlights

GE Counterfactuals: Reduce Migration Costs

Estimating the Spatial Model

1. Estimation from sorting equations:I Base wages (endogenous) wd ,I Amenities (endogenous) αdI Exogenous migration costs τdoI Frechet parameters: θ, ρ

2. GE solution yields exogenous amenity and productivity:I GE solution gives us pricesI wd = f (pd, Ad, Hd)I αd = f (αd, Ld)

Parameters set exogenously:

I Spillover parameters: amenities (λ) and productivity (γ)

I CES parameter σ

How do the Equations Identify the Model (1)?

I Identify Frechet parameter:

log(wagedo) = γd −1θ

log(πdo)

I The Economics:I Small increase in proportion moving leads to large drop in

wage implies a lot of heterogeneity in skill.

How do Equations Identify Model (2)?

I Separate absolute from comparative advantage

I Frechet distribution:var(wdo)

mean(wdo)2 =Γ(1− 2

θ(1−ρ))

Γ(1− 1θ(1−ρ)

)− 1.

How do the Equations Identify the Model (3)?

I Identify migration costs, amenity and base wage.

log(πdo) = θ ln wd +θ

βln αd︸ ︷︷ ︸

γd

+θ(1− β)

βln(1− τdo)︸ ︷︷ ︸

γdo

−γo

log(wagedo) = ln γ− 1β

ln αd︸ ︷︷ ︸γd

− 1− β

βln(1− τdo)︸ ︷︷ ︸

γdo

+1θ

γo

Outline

Reduced Form Facts: Costly Migration and Selection

Model of Spatial SortingSorting Over SpaceGeneral Equilibrium

Structural EstimatesIdentificationStructural Estimates: Highlights

GE Counterfactuals: Reduce Migration Costs

1a. Migration Costs Reduced over Time

1b. US Has Lower Movement Costs

1c. Migration Costs Correlated with Distance

1d. Less Correlation in the US

2e.Costs Correlate with Language and Religion Differences

2a. Amenities and productivity negatively correlated

1212

.513

13.5

Log

exog

pro

duct

ivity

-.3 -.2 -.1 0 .1 .2Log relative exogenous alpha

Coeff = -2.51; Std. Error = 0.04; t-stat = 62.75.

Indonesia, 1995

66.

57

7.5

Log

exog

pro

duct

ivity

-.3 -.2 -.1 0 .1 .2Log relative exogenous alpha

Coeff = -2.26; Std. Error = 0.03; t-stat = 75.33.

United States, 1990

All years

2b. Amenity negatively correlated with pollution

Outline

Reduced Form Facts: Costly Migration and Selection

Model of Spatial SortingSorting Over SpaceGeneral Equilibrium

Structural EstimatesIdentificationStructural Estimates: Highlights

GE Counterfactuals: Reduce Migration Costs

Quantifying the “why we care”

With our model, interesting counterfactuals:

1. Changing over time:I Movement costsI Amenities.

2. Comparison to US:I Movement cost.

Exogenous agglomeration parameters

I Amenity spillover: set to 0Albouy (2012)

I Productivity spillover: set to 0.05Rosenthal and Strange (2004): 0.03-0.08; Redding and Sturm (2014):

0.07; Kline and Moretti (2014): 0.2

I Sigma (CES production function): set to 8Allen and Arkolakis (2014)

I Beta (share consumption in utility function) : set to 0.6Hsieh, Hurst, Klenow and Jones (2014)

1a. Change migration costs

(1) (2) (3)Cut by 50% Cut by 100% 2012 costs

1976 1.316 1.434 1.1721995 1.366 1.462 1.0882011 1.396 1.512 0.9812012 1.408 1.525 1.000

Share cons. utility 0.600 0.600 0.600Amenity spillover 0.000 0.000 0.000Prod spillover 0.050 0.050 0.050CES parameter 8.000 8.000 8.000

Within model per capita GDP growth:

I 1976-1995: 118%

I 1995-2012: 38%Without agglomeration

1b. Change amenities

(1) (2)Equalize amenities 2012 amenities

1976 1.168 1.3141995 1.076 1.1022011 1.083 1.0332012 1.086 1.000

Share cons. utility 0.600 0.600Amenity spillover 0.000 0.000Prod. spillover 0.050 0.050CES parameter 8.000 8.000

Within model per capita GDP growth:

I 1976-1995: 118%

I 1995-2012: 38%Without agglomeration

2. Rescale to United States

2. Indonesia productivity if had US migration costs

(1) (2)Rescaling Parametric

1995 1.542 1.5312011 1.529 1.5112012 1.558 1.535

Share cons. utility 0.600 0.600Amenity spillover 0.000 0.000Productivity spillover 0.050 0.050CES parameter 8.000 8.000

I 1995 Indonesia-US per cap GDP gap: 15.2

Conclusion

I Wages are heterogeneous across space

I If due to costly migration or amenity difference may representopportunity.

I Tractable spatial equilibrium modelI WagesI AmenitiesI Comparative advantageI Agglomeration and congestion spillovers

I Migration costs quantitatively important:I If had 2012 costs in 1976: GDP 20% higherI 4% of US-Indonesia gap