Agent-based Modelling in Agricultural Economics Perspectives and Challenges Alfons Balmann JRSS...

Preview:

Citation preview

Agent-based Modelling in Agricultural Economics

Perspectives and Challenges

Alfons Balmann

JRSS 2012, Toulouse, December 14, 2012

View Video: http://ut-capitole.ubicast.tv/videos/agent_based-modelling/

2

Structure

• Motivation:Challenges of understanding structural change• topical• conceptual

• Agent-based modeling• Idea• Case study I: AgriPoliS • Case study II: SpAbCoM

• Conclusions

3

Challenges of Structural Change

General trends• New markets

• Food, fuel, fibres …• Rapid growth of emerging countries

• Globalization• Verticalization of supply chains• New technologies: GMO, …• Increasing knowledge intensity of agricultural production• Policy changes

Can structural change keep pace? How do society and policy respond?

4

Number of farms with livestock production in Germany2000=100%

Source: Statistisches Bundesamt, own calculations

2000 2003 2005 2007 20100%

20%

40%

60%

80%

100%

Dairy farms Hog farms

Challenges of Structural Change

- 41 %

- 58 %

5

Challenges of Structural Change

Size distribution of hog farms (Germany 2010)

Source: Statistisches Bundesamt, own calculations

1-99 100-499 500-999 1000-1999 2000-4999 > 50000%

10%

20%

30%

40%

50%

Anteil Betriebe Anteil SchweineShare of farms Share of hogs

6

Herausforderungen des Agrarstrukturwandels

Verteilung Größenklassen Schweine haltender Betriebe (2010)

Quelle: Statistisches Bundesamt, eigene Berechnungen

7

Structural change in poultry production (North-west Germany)

Source:Quelle: www.noz.de

8

Challenges of Structural Change

9

Challenges of Structural Change

Stakeholder analysis dairy production

Source: Ostermeyer 2011

FarmersExpertsPublic stakeholders

Altmark (Saxony Anhalt)

Abhängigkeit der Milchproduktion im Ostallgäu von dauerhafter

staatlicher Unterstützung

gering sehr hoch

Bed

eutu

ng d

er

Milc

hpro

dukt

ion

im

Ost

allg

äuse

hr h

och

gerin

g

Ostallgäu (Bavaria)

dependence of dairy farms on permanent subsidies

dependence of dairy farms on permanent subsidies

regi

onal

impo

rtanc

e of

dai

ry p

rodu

ctio

n low high low high

low

h

igh

regi

onal

impo

rtanc

e of

dai

ry p

rodu

ctio

n

low

h

igh

10

Agricultural structures as complex adaptive systems

manifold dimensions manifold levels:

individuals, enterprises, institutions, sectors, regions,… subjective perceptions, bounded rationality dynamics, non-linearities, discontinouities

Evolutionary process with limited foresights! Analysis requires specific (also heterodox) approaches!

Agent-Based Modelling (ABM)

interaction

time space

Challenges of Structural Change

11

(computer-)models,consisting of

artificial entities (agents), which

communicate and interactin an environment

(Ferber, 1999)interactions

Agent-based Models are …

12

Agents are …

sub-systems which perceive

parts of theirenvironmentand respond

autonomously

???

(Ferber, 1999)

13

Agent-based Modelling

Bottom-up approach allows flexible assumptions on individual level

(e.g. heterogenous agents, bounded rationality) allows for flexible frameworks

(e.g. non-convex functions, imperfect markets)

Self-organization spontanous order endogenous change

Particular perspectives for the analysis of emergence and change of structures organization and coordination problems

Discovery of "islands in the chaos"

14

Agent-based Modelling

Several predecessors in economics and social sciences Recursive Programming and Production Response (Day 1963)

Micromotives and Macrobehavior (Schelling 1978)

Nonlinear dynamics, chaos, erratic behavior (Benhabib & Day 1981)

The Evolution of Cooperation (Axelrod 1984)

Positive Feedbacks in the Economy (Arthur 1990)

Computational Economics driven by curiosity on complex social and economic processes

driven by increasing power and availability of computers

15

Agent-based Modelling

Clock rate

3,8 GHz

60 MHz

0,7 MHz

x 1000 000x 5000

16

Agent-based Modelling

ABM examples from social sciences and economics Santa Fe Artificial Stock Markets (Palmer et al. 1993)

Genetic Algorithm and the Cobweb Model (Arifovic 1994)

Growing Artificial Societies (Sugarscape) (Epstein and Axtell 1996)

The Complexity of Cooperation (Axelrod 1997)

Understanding complexity of markets and society Study complex games and economics Relaxing micro-economic assumptions (convexity, rationality)

17

Agent-based Modelling

In agricultural economics used since some 20 years CORMAS

rather a modeling platform with focus on common pool resource management developed at CIRAD by Francios Bousquet et al. since early 1990s participatory approach (compagnion modelling)

linking ABM with role-playing games broad international community of users

AgriPoliS developed in Göttingen, Berlin and at IAMO since 1991 somehow in the tradition of recursive programming models (Day 1963) focus on analysis of structural change several clones and extensions exist, e.g. MAS-MP (Berger 2001)

18

AgriPoliS

Policy

Environment

Markets (labor, capital, land, quotas)

Actions

Maximizationof profit/

household income

Perceptions

Resources

Interactions

Leave?

Invest?

Grow?

19

Agricultural Policy Simulator• agent-based (bottom-up)• spatial (land market)• dynamic (farm growth, exits, investments)

Farm level adjustments are considered! Endogenous structural change!

AgriPoliS

20

Effects of capping direct paymentson the Altmark region• ~ 270.000 ha UAA• ~ 980 farms > 10 ha• Ø farm size ~ 280 ha• > 45 % of land in farms > 1000 ha• 1,4 WU / 100 ha

AgriPoliSCase study: EU CAP after 2013

21

ScenarioREF

CAPPING

DescriptionNo modulation after 2013Devision of direct payments (352 €/ha) in base payment (70%) and greening component (30%)

Like REF, but:Capping base payment after deduction of wage costs (20,000 €/WU):150.000-200.000 € : 20%200.000-250.000 € : 40%250.000-300.000 € : 70%>300.000 € : 100%

AgriPoliSCase study: EU CAP after 2013

22

150

200

250

300

350

€/ha

2013 2015 2017 2019 2021 2023 2025year

payment REF CAPPINGeconomic land rent REF CAPPINGprofit REF CAPPING

Development of direct payments, profits and land rents

AgriPoliSCase study: EU CAP after 2013

23

0

10

20

30

%

farms not affected farms affected

0-30 h

a

31-100 h

a

101-200

201-500 h

a

501-1,0

00 ha

>1,000 h

a

201-500 h

a

501-1,0

00 ha

>1,000 h

a

Share of UAA in 2025 by farm size class of 2025

REF CAPPING

AgriPoliSCase study: EU CAP after 2013

Distribution of land in 2025 according to farm size classes

24

AgriPoliSCase study: EU CAP after 2013

Effects of capping on the Altmark region• Only a few large farms affected

• Adjustments allow to avoid capping• Adjustments cause negative long-term effects on efficiency and profitability• Some large farms even benefit

• Hardly any benefits for small and medium-sized farms (10-200ha)• Regional effects

• Just marginal losses of direct payments!• Losses in efficiency and profitability higher!

25

AgriPoliSConclusions

Contribution to understanding of structural change and policies• Powerful opportunities and broad scope for scenario analyses

• on structural change• on distributional issues (sizes, incomes, rents)• on productivity and efficiency Opportunity to use it also for participatory stakeholder interaction!

• Use is very demanding!• programming (AgriPoliS 3.0)• adaptation and calibration (regions, scenarios, maintainance, updates)• validation• analysis of results (not just pushing a button – apply theory and statistics!)• communication of assumptions and results

26

SpAbCom

• agricultural products (raw milk) are spatially distributed

• transport is costly many producers face few but also

spatially distributed processors

Location of milk processors in Germany

27

SpAbCom

Location of milk processors in Germany

milk market: uniform delivered pricing (udp) (Alvarez et al. 2000)

farmers receive the same price irrespective of location to dairy

price discrimination

What determines different spatial price strategies in agricultural markets?

28

SpAbCom Spatial price theory (monopsony)

local price p(r):• mill price m less a portion

of the transport costs tr• = (m, ) is the spatial

pricing strategy of a firmdistance r

price p(r)

odp

fob

udp

trmrp

Rfob,udp Rodp

zpl… zero profit liner ... distance to processors location fob ... free-on-board pricingudp ... uniform delivered pricingodp ... optimal dicriminatory pricingR ... market radius of the processor t ... transport rate

1

(=1)

(=0)

(=1/2)

zpl

= local prices differ by transport costs

= farmers receive same price irrespective of distance= local prices differ by less than transport costs

29

SpAbCom Spatial competition (duopsony)

price p(r)

odp

fob

udp

1zpl

distance rA B

zplA zplB

odp

fob

udp

price p(r)

standard assumptions (Espinosa 1992, Zhang/Sexton 2001):

distance AB=1 linear supply at each location

q(r)=p(r) price of the finished good is 1 linear transport rate t

What are the optimal strategies in terms of m and under spatial

competition?

1

30

SpAbCom Spatial competition

normalized transport costs (t)0 0.5 1.0 1.5 2.0

Perfect competition

Perfect competition

distance rA B

t=0

31

Local Monopsony

t>2

SpAbCom Spatial competition

normalized transport costs (t)0 0.5 1.0 1.5 2.0

Perfect competition

distance rA B2.0

Local Monopsony

32

0<t≤1

SpAbCom Spatial competition

normalized transport costs (t)0 0.5 1.0 1.5 2.0

Spatial competition Local MonopsonPerfect competition

distance rA B2.0

Local Monopsony

Spatial Competition

33

1<t≤2

SpAbCom Spatial competition

normalized transport costs (t)0 0.5 1.0 1.5 2.0

Spatial competition Local MonopsonPerfect competition

distance rA B2.0

Local Monopsony

Spatial Competition

34

SpAbCom Prior studies

Perfect com-

petitionSpatial competition Local

Monopson

t 0 0.4 0.6 1.1 4/3 5/3 2

ZS fob/fob udp/fob fob/udp udp/udp udp

or fobany combination

of fob and udp

ZS = Zhang and Sexton (2001) J IND ECON

Spatial competitionPerfect competition

Local Monopsony

35ZS = Zhang and Sexton (2001) J IND ECON

SpAbCom Prior studies

Perfect com-

petitionSpatial competition Local

Monopson

t 0 0.4 0.6 1.1 4/3 5/3 2

ZS fob/fob udp/fob fob/udp udp/udp udp

or fobany combination

of fob and udp

Prior studies only consider fob (α=1) and udp (α=0) as pricing

options but nothing in-between!!!

What comes out for 0 < α<1?

36

SpAbCom Methodology

Agent-based modeling farmers processors

Genetic algorithm (GA) one GA per agent selection of most profitable

strategies

1zplBzplA

distance rA B

F0

Agents of type „farmer“: max p(r)

F1 F3 F9 F10F6F5 F7F4F2

m

m

Generation 1 Generation 1

best in population

best in population

Agents of type „processor“: max PROFIT(ΓA, ΓB)

1

p(r)

p(r)

37

Generation 2 Generation 2

SpAbCom Methodology

Agent-based modeling farmers processors

Genetic algorithm (GA) one GA per agent selection of most profitable

strategies creation of new strategies

(recombination, mutation)

1zplBzplA

distance rA B

m

m

new strategy

new strategy

1

p(r)

p(r)

38

Generation n Generation n

SpAbCom Methodology

Agent-based modeling farmers processors

Genetic algorithm (GA) one GA per agent selection of most profitable

strategies creation of new strategies

(recombination, mutation)

1zplBzplA

distance rA B

m

m

optimum

optimum

1

p(r)

p(r)

39

fierce competition(low transport costs):

high price discrimination

less competition(high transport costs):

price (discrimination) increases (diminishes)

SpAbCom Results

0 0 . 5 1 . 1 . 5 2 .

0 . 2

0 . 4

0 . 6

0 . 8

1 .

normalized transport costs (t)

m

0 0.5 1.0 1.5 2.0

0.2

0.4

0.6

0.8

1.0

Spatial competition Local MonopsonPerfect competition

2.0

Local Monopsony

udp

partial freight absorption (FA)

m,

40

SpAbCom Compared to prior studies

Perfect com-

petitionSpatial competition Local

Monopson

t 0 0.4 0.6 1.1 4/3 5/3 2

ZS fob/fob udp/fob fob/udp udp/udp udp

or fobany combination (of

fob and udp)

GBS udp/udp partial FA* odp/odp

ZS = Zhang and Sexton (2001) J IND ECON GBS = Graubner, Balmann, and Sexton (2010)* FA = freight absorption (0<α<1/2)

41

Perfect com-

petitionSpatial competition Local

Monopson

t 0 0.4 0.6 1.1 4/3 5/3 2

SpAbCom Real world observations

Perfect com-

petitionSpatial competition Local

Monopson

t 0 0.4 0.6 1.1 4/3 5/3 2

ZS fob/fob udp/fob fob/udp udp/udp udp

or fobany combination (of

fob and udp)

GBS udp/udp partial FA* odp/odp

GBS = Graubner, Balmann, and Sexton (2010)* FA = freight absorption (0<α<1/2)

e.g., on markets of: raw milk, almonds, canning peaches and pears, rice, sugar beets (in Germany A and B Quota), processing tomatoes, wine grapes, corn for ethanol

e.g., on markets of sugar beets (in Germany C-Quota), milk market? And some markets were commonly fob-pricing is assumed?

42

SpAbCom Summary of findings

Spatial pricing in agricultural markets• pricing depends on the competitiveness of the market (distance,

measured by normalized transport costs)• prevalence of spatial price discrimination if production and

processing is spatially distributed ud pricing under fierce competition partial freight absorption if competition is less intense• results are consistent with observations on many agricultural

markets

43

Agent-based ModellingSpAbCom

Contribution to agricultural economics• Micro-economic approach for problems without "closed-form solutions"

• analyses on abstract level• spatial market power, spatial allocation,…

44

Agent-based ModellingSpAbCom

Contribution to agricultural economics• Micro-economic approach for problems without "closed-form solutions"

• analyses on abstract level• spatial market power, location strategies,…• further applications and extensions

• land markets (large farms, land funds)• repeated games for analysis of collusion• real options, auctions

• Challenges• very high computational needs,

particularly if many dimensions considered (space, time, interactions)• design of experiments demanding: parametrization• validation and communication of model and results

45

Summary

Agent-based models provide broad scope for interesting analyses• Micro-economic approaches for problems without "closed-form solutions"

• behavioral foundation of agents based on computational intelligence (e.g. GA) interesting perspectives for combination with human experiments to study games

• Scenario analyses for understanding systems and for decision support• policies, prices, technologies, institutions, …• behavioral foundation of agents usually based on

• optimization • rules

calibration, scenario definition and validation can be linked to participatory analyses• e.g. Compagnion Modelling linking ABM and role-playing games (Bousquet et al. 1999)

Use is demanding! Convincing addressees of results is demanding!

Recommended