Trade Costs and Household Specialization: Evidence

from Indian Farmers

Nicholas Li (University of Toronto)

Stanford IGC Nov. 14th 2014

Subsistence farmers

Uganda 2009 Timor 2000 India 1993 India 2009 Peru 1994 Guatemala 2000

Home share food expenditure 0.53 0.43 0.45 0.36 0.39 0.30

Home share food varieties 0.34 0.36 0.17 0.11 0.26 0.12

Agricultural VA per worker 221 508 473 600 1095 1945

1 Schulz (1953) “food problem”: low productivity, food a necessity

2 Dispersed production, comparative advantage from growing

conditions, high trade costs/frictions

Consumption and production decisions linked for farmers

Indian agriculture: quantities

1987-2010: Indian aggregate home share declines from 14% to 7%

(22% to 14% for food)

1987-88 2009-10

Farmer Ag.laborer Rural other Urban Farmer Ag.laborer Rural other Urban

Share of households 0.27 0.23 0.23 0.26 0.22 0.19 0.29 0.29

Real PCE 179 133 189 269 205 156 223 364

Land (ha) 2.6 0.39 0.57 0.24 1.7 0.24 0.27 0.12

Home share 0.29 0.06 0.07 0.02 0.2 0.04 0.04 0.01

Food share 0.67 0.67 0.65 0.62 0.56 0.57 0.55 0.5

Home food share 0.43 0.09 0.11 0.03 0.36 0.07 0.08 0.01

Indian agriculture: prices

Table : District average price paid vs. farm-gate price

1987-88 2009-10

Farmer Ag.laborer Rural other Urban Farmer Ag.laborer Rural other Urban

Rice 1.01 0.98 1.04 1.14 0.95 0.76 0.91 1.11

Wheat 1.02 1.03 1.05 1.12 0.96 0.87 0.96 1.09

Chickpea 1.1 1.16 1.16 1.2 1.14 1.18 1.18 1.21

Pigeon Pea 1.1 1.12 1.16 1.22 1.13 1.15 1.17 1.21

Potato 1.3 1.29 1.33 1.36 1.29 1.3 1.3 1.33

Onions 1.26 1.27 1.28 1.26 1.38 1.42 1.4 1.39

This project: Farmers and trade costs

1 Document household-level trade patterns, specialization, consumption

frictions for Indian farmers

2 Model where internal marketing frictions and trade costs affect

non-farmer vs. farmer choice and farmer specialization

3 Estimate drivers of decline in home share and farmer-level welfare

effects of trade

Literature and contributions

1 Trade: welfare gains proportional to trade share (Arkolakis, Costinot,

Rodriguez-Clare, 2012), Ricardian models applied to agriculture in India

(Donaldson, 2013), US (Costinot and Donaldson, 2014), Peru (Sotelo, 2014) →

individual farm households

2 Macro: misallocation across production units (Adamopolous and Restuccia, 2013),

sectoral models (Lagakos and Waugh (2013), Gollin and Rogerson (2014), Adam

et al. (2012), Restuccia, Yang and Zhu (2008))→ specialization and

trade/consumption frictions

3 Development: frictions generate non-separability of consumption/production

decisions (de Janvry and Sadoulet, 2003 review article) → big picture, welfare

Focus on India

Internal policy barriers to trade: state tariffs, district-level entry taxes,

licensing for traders, barriers to external trade (Atkin, 2013)

“Natural” barriers to trade: poor roads and low productivity in

distribution/retail (Lagakos, 2013) and/or markups (Atkin and

Donaldson, 2013)

Government intervention in marketing: Food Corporation of India and

Public Distribution System

India’s Public Distribution System

Central government (Food Corporation of India): procurement (15% of rice/wheat

output), transport, storage, allotted to states at “economic cost”

State governments: distribution through 478,000 Fair Price (“ration”) shops

1997 reform: “Targeted’ PDS offering 20KG at 50% economic cost to BPL

(below poverty line) households, subsidy value doubles 0.5% to 1% of GDP

Diversion and leakage: estimated 60% of grains do not reach intended BPL

households (identified with “ration card”)

Vast differences across states in implementation (Bihar vs. Tamil Nadu)

Outline

1 Stylized facts

2 Model

3 Estimation and welfare

Outline

1 Stylized facts

2 Model

3 Estimation and welfare

Data

NSS consumption and employment data (1987-2010)

Approx. 100,000 hh per survey round (10-12 per village)

Match districts over time district (about 300, median district 7500 sq.km, 116th ranked

MSA in the US)

Farmer=main household income is “self-employed in agriculture”

Household-level recall of 30-day consumption at item level (18 grains, 12 pulses, 53 fruits

and vegetables)

Record quantities and value and source (gives unit value 6= price)

Home-producible goods (77% of food exp.): value imputed at farm-gate price (exclusive

of distribution costs)

ICRISAT VDSA “meso” (district-level) data

Harvest prices, output and area sown for 11 major commodities (50% of food exp.)

Trade costs: road density, retail/wholesale/transport sector employment (NSS)

Production inputs, technology

Stylized facts

1 Home shares of farmers and districts: variable, declining over time

Dispersion/decline in food home shares

District mean farmer home shares

Stylized facts

1 Home shares of farmers and districts: variable, declining over time,

and correlated

2 Specialization: farmers and districts becoming more specialized

Farmer and district speciaization rising

Stylized facts

1 Home shares of farmers and districts: variable, declining over time,

and correlated

2 Margins of import shares and specialization: extensive margin

(variety) specialization and relative quantities important

3 Farm-gate vs. market prices and geographic dispersion: market/farm

wedge variable and declining, geographic market price dispersion not

declining

Within-district market/home wedge dispersion/decline

Across-district rural market prices

Stylized facts

1 Home shares of farmers and districts: variable, declining over time,

and correlated

2 Margins of import shares and specialization: extensive margin

(variety) specialization and relative quantities important

3 Farm-gate vs. market prices and geographic dispersion: market/farm

wedge variable and declining (big role for PDS), large relative to

geographic market price dispersion (not declining)

4 Home shares positively correlated with farmer size, negatively

correlated with trade costs

Farmer characteristics and home share

District trade costs and home share

Outline

1 Stylized facts

2 Model

3 Estimation and welfare

Ingredients

Lagakos and Waugh (2013): heterogeneous ability in farm vs. market

(urban) sector (within location)

Eaton and Kortum (2002): comparative advantage across “locations”

for farm goods

Conventional trade frictions between locations, marketing friction

within “location”

Highlights role of local “marketing costs” in allocation across sectors (farm

vs. non-farm) and specialization within farming

Preferences

Continuum of households with measure 1, Stone-Geary preferences over

agriculture (necessity) and market good (luxury)

U = (qa)α(qm + γ)1−α (1)

subject to Paqa + pmqm ≤ I .

Agricultural consumption is CES aggregate of varieties:

qa ≡(∫ 1

0qσ−1σ

i di

) σσ−1

(2)

with∫ 10 qaipaidi ≤ Paqa, CES price index Pa

Production

Fraction of households in agriculture denoted by F ; households choose between

agriculture or market production

Households have market productivity drawn from distribution S

Average and marginal market productivity given proportion of farmers denoted

S(F ) and S∗(F ); both are increasing functions of F .

Agricultural productivity for each variety (Ai ) is same for households in a location but

varies across locations.∫ 1

0yi/Aidi ≤ 1

Drawn from Frechet distribution with parameters T and θ (Pr(Ai < z) = e−Tz−θ )

Frictions

For simplicity assume N symmetric locations

Iceberg trade cost across locations for agriculture varieties: d

Iceberg “marketing cost” within location paid for non-home produced

agricultural varieties: τ

This gives agric. price index for agric. households:

Paa = T 1/θw

(1−θ + N(dτ)−θ

)−1/θ(3)

and for non-agric. households:

Pma = T 1/θw

(τ−θ + N(dτ)−θ

)−1/θ(4)

Non-farmer/farmer price wedge: W ≡ Pma

Paa> 1

Equilibrium

Welfare of marginal worker equalized between sectors Ua(F ) = Um(F ):(PaaT

1/θ

(∑

d−θi )−1/θ+ γ

)W α = (S∗(F ) + γ) (5)

m good market clears:

(1 − α)

(F

PaaT

1/θ

(∑

d−θi )−1/θ+ (1 − F )S(F )

)− γα = (1 − F )S(F ) (6)

Numeraire pm = 1, two equations in two unknowns: F and Paa .

Trade and home shares

Implications for home production shares and trade shares

Farmer’s home share in agriculture: λ = 11+N(dτ)−θ

Fraction farmer sells to non-farmer in own location: τ−θ

τ−θ+N(dτ)−θ

Fraction farmer sells to non-farmer in other location: (dτ)−θ

τ−θ+N(dτ)−θ

Fraction farmer sells to farmer in other location: (dτ)−θ

1+N(dτ)−θ

Location-level “local share” of agriculture

With heterogeneous farm productivity, standard trade result (“bigger”

farmers have higher home shares)

Role of trade costs

Equating marginal utilities:

(paT

1/θ

(∑

d−θi )−1/θ+ γ

)W α = (S∗(F ) + γ)

Consumption advantage of farmers W :

W ≡(τ−θ + N(dτ)−θ

1−θ + N(dτ)−θ

)−1/θ

> 1 (7)

W is increasing in internal trade cost τ and external trade cost d (provided τ > 1)

Decline in W leads to marginal household entering market sector

F and Pa decline, market/farmer income gap falls in nominal and real terms

Non-homotheticity (γ > 0) increases impact of changes in W but not essential

Reduction in τ could increase local trade share with no change in external trade

costs/price gaps (d)

Extensions

Elasticity of substitution < 1 between farm/market sector amplifies

effects of marketing/trade costs, neutral and agr. productivity (T )

determine size of farm sector

Farmer heterogeneity: income distribution effects, different abilities,

land

Three sectors? Rural farmer vs. rural non-farmer (small income/price

gaps) vs. urban (larger gaps)

Calibration and simulation

Calibration Data Baseline No mkt cost No trade cost No frictions

Farmer share 0.22-0.73 0.585 0.575 0.580 0.575

Non-farmer/farmer income 1.3-1.7 1.288 1.257 1.271 1.257

Food share farmer 0.56 0.567 0.559 0.563 0.559

Food share non-farmer 0.5 0.474 0.475 0.475 0.475

Price index for A 10.663 9.574 7.953 6.854

Non-farmer/farmer price 1-1.13 1.057 1 1.025 1

Home share farmers 0.2 0.200 0.143 0.093 0.062

Home food share farmers 0.35 0.352 0.256 0.165 0.111

Whole pop. Home share 0.07 0.104 0.074 0.048 0.032

Aggregate local food share 0.7 0.311 0.256 0.141 0.111

Baseline calibration uses θ = 2.5, N = 9, d = 1.5, τ = 1.2, γ = 45, α = 0.15, S distributed U[5, 25]

Outline

1 Stylized facts

2 Model

3 Estimation and welfare

Estimated effect of trade cost variables (1987-2010)

Farmer-level District-level

(1) (2) (3) (4)

Home share Home varieties Frac. farmers Home share

Roads -0.0345*** -0.0478*** -0.0145*** -0.0336

(0.0125) (0.0180) (0.00333) (0.0555)

Frac. PDS -0.215*** -0.00784 -0.0105** 0.233*

(0.0667) (0.0774) (0.00471) (0.124)

Irrigation 0.211* -0.135 -0.00941 0.533***

(0.117) (0.116) (0.0115) (0.193)

Fertilizer 0.000509 -0.000558 -1.27e-05 -0.000723

(0.000328) (0.000356) (2.77e-05) (0.000533)

Population 0.0309 0.00340 -0.00129 -0.450***

(0.0417) (0.0542) (0.00142) (0.0900)

District FE YES YES YES YES

Year FE YES YES YES YES

Observations 84,688 84,693 1,302 1,292

R-squared 0.223 0.273 0.975 0.848

Standard errors in parentheses (clustered by district), *** p<0.01, ** p<0.05, * p<0.1

Farmer-level regressions include quadratics in land, household size and real expenditure.

District-level regressions include fraction of farmers in 4 land size categories.

Gains from trade

Within agriculture sector Arkolakis, Costinot, Rodriguez-Clare

(2012) insights apply

Gains from trade: W = (λ′/λ)1/ε

Domestic (home production) shares of consumption: λ′(=1 under

autarky) and λ

Trade elasticity ε: -5 to -10 range in trade (Anderson and Van

Wincoop, 2003), but for individuals?

Estimating home share elasticity

ln farmerimportsidt = αd + γt + ε ln τdt + ΩXidt + uidt

Think of each farmer i trading with a single entity – “the market”

Farmers in a district face the same “marketization cost” τdt ,

measured using the market/home wedge index

Farmer productivity, trade costs with other districts, unobserved trade

costs might also affect farmer import share

Estimation Challenges

ln farmerimportsidt = αd + γt + ε ln τdt + ΩXidt + uidt

Measurement error in τdt

Omitted variables

Instruments for τdt :

1 Fraction of “market” rice and wheat sold through TPDS (=0 before

1992)

2 Road construction: national 4-lane highways 200km (1988) to

10,000km (today), rural road program

Controls: District and round fixed effects, quadratics in hh size, real

expenditure and land, district irrigation and fertilizer use, pop. density

Farmer trade elasticity

(1) (2) (3) (4)

OLS IV PDS IV Roads IV both

Major commodities

Elasticity -0.04 −2.62∗∗∗ −2.52∗∗ −2.44∗∗∗

(0.06) (0.75) (1.07) (0.73)

Obs. 112190 110869 103909 103908

First-stage F-stat 14.07 7.91 7.99

OverID test p-value 0.94

All food

Elasticity −0.26∗∗∗ −1.64∗∗ −3.75∗ −1.84∗∗

(0.09) (0.74) (2.10) (0.91)

Obs. 112739 112548 105593 105593

First-stage F-stat 11.05 4.7 6.01

OverID test p-value 0.16

Standard errors in parentheses (clustered by district), *** p<0.01, ** p<0.05, * p<0.1

Regressions include quadratics in land, household size and real expenditure, district and round FE.

Similar results using state by year FE.

Welfare gains for farmers

Major comm. All food

ACR(2012) ingredients

Elasticity 2.5 1.8

1993 Home share 0.53 0.54

2009 Home share 0.41 0.43

Welfare (food only)

Gains 2009 vs. autarky 0.30 0.37

Gains 2009 vs. 1993 0.10 0.11

Changes in trade costs 1993-2009

Observed % decline in τ 0.06 0.05

Removing TPDS (through τ) 0.04 0.04

∆ roads (through τ) 0.01 0.01

Implied % decline in τd 0.15 0.18

Policy Implications

Public Distribution System

Usually think about nutritional and poverty effects (Tarozzi (2005),

Dreze and Khera (2013))

PDS reforms in the 1990s had major effects on farmer specialization

and market/home price wedges

Partly counteract “distortion” that keeps households in farming and

producing for self instead of market

One the other hand, national-level farmer support prices are having the

opposite effect

Summary

1 Documented large decline in home production shares in India, interpreted through

the lens of trade and local “marketing” costs

2 Model relating home production and farmer/non-farmer occupation choice to

farmer/non-farmer wedge and usual trade forces

3 Despite minimal liberalization of domestic trade, substantial (10%) welfare gains

to Indian farmers through “trade channels” (PDS and local roads)

Empirical: spatial price differences and seasonality, detailed farmer production data

Theory: farmer vs. rural non-farmer vs. urban, adding farmer heterogeneity and land,

better calibration

Appendix: farmer trade costs (Gollin and Rogerson, 2014)

Appendix: VDSA harvest price vs. NSS “farm gate price”

Appendix: Rice prices 2009

Appendix: Wheat prices 2009

Appendix: PDS share 2009