Transcript Economic and Social systems modelling

Issues with
Economic and
Social systems
modelling
Mariam Kiran
University of Sheffield
Future Research Directions in Agent Based Modelling
June 2010
Talk Agenda
 Agent-based modelling for socio-economic
systems as compared to the traditional methods.
 Case study: EURACE model
 Useful results
 Issues raised
 Case study: Social Capital Model
 Conclusions
Modelling of socioeconomic systems
 Traditional approaches involve using differential
equations
 Use game theory models, commonly with a
maximum 5 number of players in the model
 Large number of exaggerated assumptions
 Rational people making rational decisions
 Small populations
 Complete knowledge
ABMs can overcome some of these issues, like
populations, heterogeneity , etc
Case Study: EURACE
 FLAME Framework
 The first attempt for economic modellers to
merge more than one market together to
represent a complete economy.
 Each individual is considered as a agent like
households, firms or more.
 Various computer scientists and economists
worked together to achieve the goals of this
project (8 universities)
EURACE markets and their
interactions
Eurace dot file
EURACE Modelling issues
 Large complicated agents and large
concentrations.
 Too much of communication overhead for
agents communicating with each other.
 Economists had very little programming
experience.
Agent
Numbers
(x30,
total=
31,110)
Firms
24
IGFirm
1
Households
1000
Malls
4
Bank
4
ClearingHouse
1
Government
1
Central Bank
1
Eurostat
1
Libmboard – FLAME
message board library
 Uses distributed memory model Single Program
Multiple Data (SPMD).
 Synchronisation helps prevent deadlocks.
 Uses Message Passing Interface to communicate
messages
 Using filters and added iterators have helped
quicken message parsing for agents.
Simulation time results
Num
processors
HAPU
NWGrid
Hec
tor
2
92.3
43.2
-
4
43.3
32.4
29.8
6
76.9
29.3
26.6
8
63.6
30.8
24.1
10
72.1
37.3
22.9
12
34.6
36.5
22.0
14
82.5
40.5
22.1
16
45.0
41.0
21.7
Comparing Economic
policies for EU
average price level (€)
0.95
FT
QE
0.9
0.85
0.8
0.75
0.7
0
6
12
18
24
30
36
42
48
54
60 66
months
72
78
84
90
96
102 108 114 120
average wage level (€)
2
1.9
1.8
1.7
1.6
1.5
0
FT
QE
6
12
18
24
30
36
42
48
54
60 66
months
72
78
84
90
96
102 108 114 120
The effect of
Fiscal
tightening
(FT) and
Quantitative
Easing (QE)
on price and
wage levels
Effects of technology,
innovation and skill for old
and new EU members
Specific skill
levels of
workers when
the labour
markets are
open or closed.
Germany
Poland
Energy shocks to the
markets
The effect on
GDP growth
with and
without energy
crisis
EURACE results
 Predicts that not increasing taxes will allow UK to
recover from the recession.
 Opening borders across the EU benefits all
countries for the labour market.
 Energy shocks to the system. System came back
to an equilibrium when this happened.
Case Study: Social Capital
Modelling
-Replication of mathematical
model
-Calculations of numbers of
transitive relationships,
reciprocated ties, incomplete
transitive ties
-Looping, Bottlenecks
-Flame group is currently
working on overcoming these
issues
1
45
89
133
177
221
265
309
353
397
441
485
529
573
617
661
705
749
793
837
881
925
969
0.3
0.2
0.05
0.1
0
0
1
45
89
133
177
221
265
309
353
397
441
485
529
573
617
661
705
749
793
837
881
925
969
centralisation
1
0.9
0.4
clustering_coef
1.2
0.8
0.6
15
0.4
10
0.2
1
45
89
133
177
221
265
309
353
397
441
485
529
573
617
661
705
749
793
837
881
925
969
1
45
89
133
177
221
265
309
353
397
441
485
529
573
617
661
705
749
793
837
881
925
969
-2.5 outdegree
1.00reciprocity
0.55transitivity
0.45similarity
Initial Structure = In-Star
Density1
0.4
0.35
0.8
0.3
0.7
0.25
0.6
0.2
0.5
0.15
0.1
avg_geodesicpathlength
1
25
20
5
0
0
Comparing Geometric and
Round robin partitioning
Geometric partitioning is when
agents are distributed across
processors based on their x and
y coordinates
Round robin partitioning is when
agents are distributed evenly
across processions
Centralised versus
Decentralised Models
Time increases as
number of nodes are
increased
Cournot Model
Time is unchanged
with nodes
Sugarscape + IPD Model
Conclusions
 Think about the kind of models.
 Initial distribution of agents on processors.
 Is the model correct? Run the model till we reach
equilibrium.
 Copying files across for data analysis. GB of data
can take hours to copy across.
 Communication problems between computer
scientists and economists, sociologists.
 Different time expectations between disciplines.
More information:
 Documentation can be found:
www.eurace.org
www.eurace.groups.shef.ac.uk
 Other current models our group
is working on:
Ant Phermone trails
Social Networks
Sperm behaviour
E-Coli behaviour
Epithelium Tissue
 FLAME Website:
www.flame.ac.uk
Move to reality using ABMs
 Collection of unique individuals.
 Experimenting with different populations.
Companies
 Most assumptions are being overcome.
 Each individual is different, represents
heterogeneous collection.
Others
Others
Others
Others
 Each has different properties, different functions,
different memories.
 There can be a million representative of the same
individuals or a million others in the system.
Banks
Shops