27
Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto – LABORatorio R. Revelli ESSA 2007, September 10-15, 2007, Toulouse

Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

Embed Size (px)

Citation preview

Page 1: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

Talk, but speak clearImproving the rigour in agent-based social simulation

Matteo RichiardiUniversità Politecnica delle Marche

and Collegio Carlo Alberto – LABORatorio R. Revelli

ESSA 2007,

September 10-15, 2007, Toulouse

Page 2: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

2

Outline

What is Agent-based Computational Economics (ACE)

Main features

When to use an agent-based model

The dual problem of micro-macro relation

A dynamic system representation of agent-based models

The notion of equilibrium in agent-based models

Analysis of the models

Estimation / calibration of the parameters

Description of the models

Page 3: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

3

References

• Leombruni, Richiardi, Saam, Sonnessa (2006), “A Common Protocol for Agent Based Social Simulation”, JASSS, 9(1)

• Richiardi, “Agent-based Computational Economics. A Short Introduction”, mimeo

• ….

downloadable from

www.dea.unian.it/richiardi/richiardi.htm

www.dea.unian.it/richiardi/courses/phd_ace.htm

Page 4: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

4

What is ACE

Agent-based computational models are models in which:

(i) a multitude of objects interact with each other and with the environment

(ii) the objects are autonomous, i.e. there is no central, or “top down” control over their behavior;

(iii) the outcome of their interaction is numerically computed.

“ACE is the computational study of economic processes modeled as dynamic systems of interacting agents.” (L. Tesfatsion)

Page 5: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

5

A methodological remark

Methodological individualism: explaining society as the aggregation of decisions by individuals (Austrian School of Economics). Reductionism: the whole is nothing but the sum of its parts.

ACE

Holism: the proprierties of a system cannot be deduced by the properties of its components alone. Indeed, the system as the whole determines how the parts behave: the whole is more than the sum of its parts (Aristotle, Metaphysics). Organicism (Ritter, 1919): the organization, not the composition, of organisms is what counts.

Page 6: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

6

Main features of ACE models

Heterogeneity

Explicit space

Local interaction

Cognitive foundations:

Bounded rationality

Limited / asymmetric information

Non equilibrium dynamics

Page 7: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

7

When to use ACE

To get a quick intuition of the dynamics a given system is

able to produce (scrap paper use)

To thoroughly investigate models that are not susceptible of a

more traditional analysis, or are susceptible of a more

traditional analysis only at too a high cost:

1. Numerical computation of analytical models

2. Robustness analysis of analytical models

3. Stand-alone simulation models

Page 8: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

8

When to use ACE

Analytical models: little interaction, little dynamics,

sophysticated behaviors (often, not always)

Simple behaviors can produce complex patterns

(e.g. Langton’s ants) [GO]

Simple choices can lead to complex and diversified behaviors

(e.g. El Farol bar problem, W. Brian Arthur, “Inductive

reasoning and bounded rationality”, American Economic

Review, n. 84, p. 406, 1994)

Page 9: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

9

The dual problem

The dual problem of the micro-macro relation:

a)FROM MICRO TO MACRO: Find the aggregate implications of given individual behaviors

b)FROM MACRO TO MICRO: Find the conditions at the micro level that give raise to some observed macro phenomena - ACE as generative social science: “If you didn’t grow it, you didn’t explain it” (Epstein, 1999)

Page 10: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

10

Our real problem

• a growing community of modellers

• a disappointing publication rate

• audience is more often interested in the methodology than in

the topic autoreferentiality

ABM papers seem to be confined only to specialized journals like

the Journal of Economic Dynamics and Control, the Journal

of Artificial Societies and Social Simulation, Computational

Economics, the Journal of Economic Interaction and

Coordination, Advances in Complex Systems and few others

Page 11: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

11

The real problem

How can

the rate of acceptance

of papers with agent-based methodology

in the top journals

be increased?

Talk to the mainstream !!!

Break the auto-referentiality circuit !!!

Page 12: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

12

Traditional analytical modelling

Traditional analytical modelling relies on a very well established,

although implicit, methodological protocol,

both with respect to:

• the way models are presented and to

• the kind of analyses that are performed.

Page 13: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

13

Traditional analytical modelling

E.g., in most papers:

• detailed reference to the literature;

• the model often adopts an existing framework and extends, or

departs from, well-known models only in limited respects;

• this allows a concise description, and saves more space for the

results, which are

• finally confronted with empirical data;

• when estimation is involved measures of validity and

reliability of the estimates are always presented, in a

• very standardized way.

Page 14: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

14

Agent based modelling

E.g., in most papers:

• very limited reference to the literature;

• the model often departs radically from the existing literature

and adopts or extends no existing, well-known framework.

• This allows only a superficial description of the model, and

even leaves no space for the results, which are

• only rarely confronted with empirical data.

• When estimation is involved few measures of validity and

reliability of the estimates are presented, often in a

• very unstandardized way.

Page 15: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

15

ACE is mathematics

Computer simulation as a third symbol system, aside verbal description and mathematics (Ostrom, 1988)

“Simulation is neither good nor bad mathematics, but no mathematics at all” (Gilbert et al., 1999)

“Any theory that can be expressed in either of the first two symbol systems can also be expressed in the third symbol system.”

“There might be verbal theories which cannot be adequately expressed in the second symbol system of mathematics, but can be in the third”

Page 16: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

16

ACE models as recursive systems

behavioral rules

state variables

structural parameters

);,( ,,1,1, tittiiti xxfx

);,( ,,1,1, tittiiti xxa

Page 17: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

17

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

)),,(),...,,,((

),...,(

0,0,10,0,1

1,1,1,1,11,11,11

,,1

nnt

tntntnnttt

tntt

xxg

xxfxxfs

xxsY

I-O transformation function

initial conditions

ACE models as recursive systems

structural parameters

Page 18: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

18

2 notions of equilibrium: AT A MICRO LEVEL, when individual strategies are

constant

AT A MACRO LEVEL, when some relevant aggregate statistics of the system are stationary:

Equilibrium

),...,;,...,(lim 0,0,10,0, nnitt

e xxa

),...,;,...,(lim 0,0,10,0, nnitt

e xxgYY

Page 19: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

19

g is unknown get an inductive evidence about g, by performing multiple

(thousands, millions) runs and recording inputs and outputs of each run

Interpretation of results

),...,;,...,(lim 0,0,10,0, nnitt

e xxgYY

Page 20: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

20

sensitivity analysis around some default values of the parameters and initial conditions: keep all parameters and initial conditions fixed and change only one at a time

Local analysis of g

equivalent to exploring the partial derivatives of g

choice of the default values is an issue

Page 21: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

21

Let all parameters change between the different simulation runs

Get an estimate of g

(metamodel, response surface, compact model, emulator, ecc.)

Global analysis of g

Page 22: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

22

Thank you for your attention!

Matteo [email protected]

Page 23: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

23

“the distinction drawn between calibrating and estimating the parameters of a model is artificial at best. Moreover, the justification for what is called “calibration” is vague and confusing. In a profession that is already too segmented, the construction of such artificial distinctions is counterproductive” (Hansen and Heckman, 1996)

Choice of structural parameters: some parameters have real counterparts and their value is

known: no need for calibration/estimation otherwise: choice of the values that make the artificial data

produced via simulation as close as possible to the real data

Calibration / estimation

Page 24: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

24

Gouriereux and Monfort, 1997 Mariano et al, 2000 Train, 2003

define some target variables (e.g. moments) compute them both in the artificial and in the real data keep changing the parameters to estimate (the ) until the

distance between the values of the target variables in the artificial data and in the real data is minimized

Indirect Inference: the targets are the estimates of an auxiliary model

Simulation based estimation

Page 25: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

25

code availability the sequence of events in the simulation must be carefully

described:

pseudocode time-sequence diagrams

Replicability

Page 26: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

26

The pseudo-code

Source: Neugart (2006)

Page 27: Talk, but speak clear Improving the rigour in agent-based social simulation Matteo Richiardi Università Politecnica delle Marche and Collegio Carlo Alberto

27

Time-sequence diagrams

Source: Richiardi (2007)