Economic Modeling for Agriculture, Food and the Environment - - PowerPoint PPT Presentation

AgreenSkills Seminar Toulouse, October 29, 2014

Economic Modeling for Agriculture, Food and the Environment

Objectives, Approaches, Frontiers Alban THOMAS

Head of the Social Science Division (SAE2), INRA

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Economic modeling

Objectives: Describe and represent behaviour of agents / actors in production, consumption activities  Analyse and predict consequences of decisions for market

• utcomes (local markets, international trade)

Methods:  To describe: statistical analysis and econometrics (Reduced form)  To test and predict structural changes: system equation, structural econometrics (Structural form)  Data:  Individual surveys, field surveys  Experimental and behavioural economics Time series (country, region, etc.)

.02

INTRODUCTION

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014r

Predictions dealing with various dimensions:

• time
• space
• population category

In the following, three examples  Food security and biofuel policy: modelling land use and energy demand for world regions  Environmental impacts of agricultural production decisions  Natural resources: management of irrigated cropping systems under water scarcity

.03

INTRODUCTION

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Application 1. Food security and biofuel policy: modelling land use and energy demand for world regions

.04

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

An application of partial-equilibrium models Purpose: predict future food price changes given biofuel policy

.05

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

US biofuel mandate Ujjayant Chakravorty, Marie-Hélène Hubert, Michel Moreaux and Linda Nøstbakken (2012), Do Biofuel Policies Really Increase Food Prices? Alberta University WP. Distribution of land quality

Land class US EU Other HICs MICs LICs World Land already under Agriculture (million ha) 1 100 100 25 300 150 675 2 48 32 20 250 250 590 3 30 11 20 243 44 350 Land available for farming (incl. fallow lands) (million ha) 1 2 11 8 21 300 300 640 3 11 8 21 500 500 1040

Sources: Eswaran et al. (2003), FAO (2008a), Fischer and Shah (2010). Notes: Land available for US, EU and

• thers HICs is not available per land class. We assume that half of the available land is class 2.

Land under agriculture and endowment of available land

Example: impact of biofuel policy

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

.06 c) Specify production technology (production costs) f) Simulation from scenarios on climate and technology a) Define world regions and commodities b) Specify region-wise demand trends

𝑛𝑛𝑛 𝑚 𝑑=1,…,𝐷 𝑚𝑑 𝑞𝑑𝑔 𝑚𝑑 + 𝜐𝑑 − 𝑠𝑚𝑑 − 𝑋

𝑑 𝐷 𝑑=1

s.c. ∑ 𝑚𝑑 ≤ 𝑀

𝐷 𝑑=1

d) Solve for equilibrium price on markets e) Derive regionwise optimal land use

.07

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014 l l

l l l

D A P w N

α β

=

Demand for final product l

2 1

) ( ) ( ln ) ( φ φ

θ

+             ∑ − − =

= s i t s i s i s

L l L t C

Land conversion cost (time t, use s)

2

1

( )

j j j j j i i i i i i

C k L k L

η

η   = ∑ ∑    

Total cost of product j in region

1 1 1

) )( 1 (

− − −

          + − + =

ρ ρ ρ ρ ρ ρ

µ µ λ

bs bf g g g e

q q q q

Production technology: energy from biofuel and gasoline

3

1 2

( ) ( ( ))

t

• il

x C x t X

=

  ∑     = +      

ϕ θ

θ ϕ ϕ

Cost of oil extraction

US EU Other HICs MICs LICs Representative crop Corn Rapeseed Corn Sugar-cane Cassava

(94%) (76%) (96%) (84%) (99%)

Unit cost of production ($/gallon) 1.01 0.55 1.10 0.94 1.30

Sources: Production costs (FAO 2008a; Eisentraut 2010); Notes: The numbers in parentheses represent the percentage of first-generation biofuels produced from the representative crop (e.g., corn).

Unit costs of first generation biofuels

.08

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014 BASE REG Weighted food price ($/ton) 2007 557 564 2022 639(15%) 746(32%) Biofuel price ($/gallon) 2007 2022 2.14 1.97 2.18 2.19 Crude oil price ($/barrel) 2007 105 106 2022 121 119

Notes: Weighted food price is the average of cereal and meat prices weighted by the share of each commodity in total food consumption. The numbers in brackets represent the percentage change in prices between 2007-

• 22. Our predictions for crude oil prices are quite close to US Department of Energy (EIA 2010, p 28) projections
• f $115/barrel in 2022.

World food, biofuel and gasoline prices (in 2007 Dollars) World weighted food prices

Application 2. Environmental impacts of agricultural production decisions

.09

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

An application of farm-level production economics Purpose: predict environmental impact of changes in price and agricultural policy

.010

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Basic framework

Lacroix, A. and A. Thomas, 2011. American Journal of Agricultural Economics.  Trade and agricultural policy scenarios  change in agricultural prices  Simulation of environmental policy  tax on nitrogen fertiliser  How to evaluate changes in farmer decisions regarding

• crop output
• land use
• fertiliser application rate

 Dynamic model for simplified representation of crop rotations  Model with multiple crops and multiple inputs  Linking with environmental simulator for nitrogen leaching (groundwater nitrate contamination)

.011

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Market prices Ouput Input Previous crops Constraint on total farm land Probability to select crops, c=1,2,…,C Pr 𝑚𝑑𝑑 > 0|𝑚𝑑,𝑑−1, 𝑄, 𝜐 Policy instruments (CAP subsidy rates per ha) Crop yield decision 𝑟𝑑𝑑 Land use decision 𝑚𝑑𝑑 Input decision 𝑋

𝑑

Environmental impact simulator Expected crop yield, land-use and input, conditional on selected crops 𝐹 𝑟𝑑𝑑| 𝑚𝑑𝑑 > 0, 𝑚𝑑′𝑑 > 0, … 𝐹 𝑚𝑑𝑑| 𝑚𝑑𝑑 > 0, 𝑚𝑑′𝑑 > 0, … 𝐹 𝑋

𝑑| 𝑚𝑑𝑑 > 0, 𝑚𝑑′𝑑 > 0, …

.012

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

The multi-crop model with multiple selection

[ ]

1 1 C K t ct ct ct ct kt kt c k

l p q r W τ

= =

Π = + −

∑ ∑

1

s.t.

C t ct c

L l

=

=∑

( , , , ) ( , , , ) ( , , , ) ; ;

t t t t t t t t t t t t t t t ct ct kt ct ct kt

p r L p r L p r L l q W p r τ τ τ τ ∂Π ∂Π ∂Π = = = − ∂ ∂ ∂

C crops, c=1,2,…,C K inputs, k=1,2,…,K lct : land for crop c at time t pct : output price for crop c at time t qct : crop yield of crop c at time t τct : subsidy (per hectare) for crop c at time t Wk : quantity of input k at time t rct : unit price of input k at time t Profit of farmer at time t Optimal land use Optimal output level Optimal input demand

.013

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

( )

( ) ( ) ( )

1

• 1. Estimation of selection equations Prob

, , , / ; 0 , 0 , 0 , , ,

ct t t t t t q l W ct ct ct ct kt ct

l p r l L E l E l E l t c k τ ε ε ε

−

> > > > ∀ ∀ ∀

1 1 2 1 3 1

(output) ( , , , ) , (land) ( , , , ) , (input) ( , , , ) .

C q ct ct t t t t cc t c C l ct t t t t cc t c C w kt t t t t kc t c

q l F p r L l F p r L W F p r L τ λ τ λ τ λ

′ ′= ′ ′= ′ ′=

 = +    = +    − = +  

∑ ∑ ∑

• 2. Parametric specification: quadratic profit function

* 1 1 1 1 1 1 1 ' ' ' ' ' ' ' ' 1 ' 1 1 ' 1 1 ' 1 1 ' 1 1 1

1 1 1 2 2 2

C C K C K C K p r pr r c c c c k k ck c k ck c k c c k c k c k C C C C C C K K pp p rr cc c c cc c c cc c c kk k k c c c c c c k k C C Lp L c c c c c

p r p r r p p p r r Lp L

τ τ ττ τ τ

β β β τ β β β τ β β τ τ β τ β γ γ

= = = = = = = = = = = = = = = = =

Π = + + + + + + + + + + +

∑ ∑ ∑ ∑∑ ∑∑ ∑∑ ∑∑ ∑∑ ∑∑ ∑ ∑

1 K Lr c k k k

Lr τ γ

=

+∑

• 3. System estimation with correction of selection bias

.014

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Nitrogen runoff Nitrogen balance WR: water repletion rate; IN: intercrop duration; fc: nitrogen application rate; aj: average nitrogen supplied by livestock j Nj: livestock population of type j; bc: average nitrogen export of yield of crop c

Predicted from model Application: 634 farmers, period 1995-2001 (2820 observations) Three French regions (Midi-Pyrénées, Pays-de-Loire, Rhône-Alpes) From STICS crop simulator (Brisson et al., 1998)

.015

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Estimation of price elasticities under various model specifications

• Aggregate data
• Random effects
• Fixed effects
• Single crop selection
• Multiple crop selection

.016

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Application 3. Natural resource management of irrigated cropping systems under water scarcity

An application of stochastic dynamic programming Purpose: evaluate strategies for adaptation to climate change

.017

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

.018

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Small-scale irrigation systems in India

Context:  (Very) small land plots (< 1 ha)  Individual irrigation sources: borewells & tubewells  Continuous depletion of groundwater  Hardrock aquifers of south India (Karnataka, Kerala)  Electric pumps with limited but free electricity (3 hours / day)

.019

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Expectation on future rainfall (R) Invest in upgraded irrigation Crop choice Expectation on future prices (P) Irrigation calendar Harvest Plot A Plot B No Yes Borewell Borewell capacity

• Account for

borewell depreciation

• Credit capacity from harvest

1 year / season

.020

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Period-specific restriction on water 1 if crop c planted on plot b at time τ Crop yield Random rainfall Water requirement for crop c Unit irrigation cost Investment cost (new well, reboring, etc.) Random price

Static case (one year)

.021

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Dynamic programming framework

 State variable: groundwater availability (year)  Control variables:

• Investment in upgrading irrigation equipment
• Crop choice (share of total land from crop 1, 2, etc.

𝑋

𝑑 = 𝑔 𝑋 𝑑−1, 𝐽𝑑, 𝑆𝑑

𝑛𝑛𝑛 𝐽,𝐷 𝐹𝑆

𝐹𝑄

1 (1 + 𝑠)𝑑 𝑚𝑑𝑑 𝑞𝑑𝑑𝑧𝑑 𝑋

𝑑𝑑, 𝑆𝑑, 𝑌𝑑 − 𝐷𝑋 𝑑(𝑢, 𝑋 𝑑) − 𝑠𝐽𝑑𝐽𝑑 𝐷 𝑑=1 𝑈 𝑑=1

 Irrigation unit cost increases over time (but decreases with upgrading investment)  Investment irrigation has to be reimbursed yearly (loan)  Future stream of benefits is discounted (with rate r) 𝑋

𝑑 ≥ 𝑋 𝑑𝑑 𝐷 𝑑=1

where 𝑋

𝑑 = 𝑔 𝑋 𝑑−1, 𝐽𝑑, 𝑆𝑑

and

.022

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Infinite-horizon case: Bellman equation for value function 𝑊 𝑋 = 𝑛𝑛𝑛𝐽 𝜌 𝑋, 𝐽, 𝑆 + 𝜀𝐹𝑑𝑊 𝑔(𝑋, 𝐽, 𝑆)

 Value function iteration  Policy iteration  Parameterized expectations algorithm (PEA) Methods for solving a discrete-time continuous-state stochastic dynamic problem: Collocation method:  Discretize state variable, 𝑋

𝑑 = 𝑋 1𝑑, 𝑋 2𝑑, … ,

 Specifiy N basis functions (Chebychev polynomials, cubic splines, etc.)  Solve, for each discretized value of control variable 𝜚𝑘 𝑋 𝑑

𝑘 = 𝑛𝑛𝑛𝐽𝑗

𝜌 𝑋, 𝐽, 𝑆 + 𝜀𝐹𝑆,𝑄 𝜚𝑘(𝑔(𝑋))𝑑

𝑘 𝑂 𝑘=1 𝑂 𝑘=1

 Get optimal value of control variable, I

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

All three examples:  Strong assumptions on technology and agents’ preferences  Issue: robustness to misspecification?  All based on observed outcomes, market values  Models based on representative agents  Unobserved individual heterogeneity added afterwards Crucial assumption: policy or market changes will not affect model parameters Most models are  devoted to evaluate policy impacts  while controlling for agents’ heterogeneity

Major issues with  selection of observations (non random sample selection)  endogeneity (specification of models with random terms)  heterogeneity (many unobserved factors) This implies some assumptions difficult / impossible to test for

.024

Frontiers

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Major differences with life sciences: For most analyses, no experimental data Exceptions: experimental economics, natural experiments, randomized controlled experiments

.025

Frontiers

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014

Context variables Time 1 X1 Time 2 X2 Preferences Attitudes Decision rules 𝜄, 𝑌, 𝑄 ⇒ 𝑧 Policy 1 at time t=1 P1t Policy 2 at time t=2 P2t Observed decisions 𝑧(𝜄, 𝑌1, 𝑄

1)

𝑧(𝜄, 𝑌2, 𝑄2) 𝜄 How to infere the impact of policy change, given θ (unobserved) and X (exogenous, observed)? t=1 t=2

Control group Treatment group

: observed Treatment effect : 𝐹 𝑧22 − 𝑧21 − 𝑧12 − 𝑧11

Change in treatment group Change in control group

THANK YOU FOR YOUR ATTENTION

thomas@toulouse.inra.fr

Alban Thomas – AgreenSkills Seminar 29 / 10 / 2014