Introduction to Agent Based Modelling Tommaso Ciarli SPRU, - - PowerPoint PPT Presentation

Introduction to Agent Based Modelling

Tommaso Ciarli

SPRU, University of Sussex t.ciarli@sussex.ac.uk

Advances in Economic Dynamics and Development: Economics and Complexity Third meeting PPGDE, UFPR Curitiba – June 13-17, 2016

aAcknowledgement: some of the material draws on presentations given by Alan

Kirman and Giorgio Fagiolo

0 / 96

Introduction Learning outcomes The problem: evidence

Introduction

The financial crisis, and the related real crisis, were unpredictable, and are only partially understood, using available economic models Where unpredictable refers to Knightian uncertainty, where risk cannot be measured, rather than a wrong guess on a probability distribution Why? Some fundamental features of economic systems network structure – banks, users and countries individual’s beliefs and expectations – satisficing, partly adaptive, and heterogeneous interdependent behaviour – contagion ⇒ Features of a complex system

1 / 96

Introduction Learning outcomes The problem: evidence

Network: Cross border financial flows

(a) To advanced economies (b) To emerging and developing

economies Source: Minoiu and Reyes (2011)

2 / 96

Introduction Learning outcomes The problem: evidence

Network: Cross border banking network: core-periphery

Source: Minoiu and Reyes (2011)

3 / 96

Introduction Learning outcomes The problem: evidence

Behaviour: Markets are made up of rational individuals optimising in isolation?

Source: Kirman slides

4 / 96

Introduction Learning outcomes The problem: evidence

Contagion: herding behaviour

Source: Kirman (2010)

5 / 96

Introduction Learning outcomes The problem: evidence

Contagion: information cascade

Source: The Economist, Nov 1, 1997

6 / 96

Introduction Learning outcomes The problem: evidence

Contagion: information cascade

Decision are taken sequentially: market signals adjust through time We do not jump from one equilibrium to another Information is mediated locally by other actors (people), not only centrally by prices “After a sufficient time the cumulated actions of other actors contain so much information that an individual will have an incentive to ignore his or her own information and a ‘cascade’ will start” (Kirman, 2010) Choice of a restaurant comparing public and private information Adoption of technologies, diffusion and lock-in (e.g. Arthur, 1989; Cowan and Gunby, 1996)

7 / 96

Introduction Learning outcomes The problem: evidence

Characteristics of a complex system (Page, 2015)

Interaction structure (facebook, finance) Interdependent: people influence each other (fads) Learning and adaptation: change, behaviour, connections and interdependency Selection (and variation) Heterogeneity: initial, and as a process of adaptation, or innovation

8 / 96

Introduction Learning outcomes The problem: methodological approach

Why is this relevant?

9 / 96

Introduction Learning outcomes The problem: methodological approach

Why is this relevant?

Because this is the system that we want to analyse and understand. Particularly one in which innovation is crucial, a driver of change in both high and low income countries We need tools to model such systems, and these models may look different from the ones that we, as economists, are accustomed to analyse (where the agent is one, representing all) Because it is crucial to understand he relation between complicated dynamics such as growth of income, development and structural change, and environmental sustainability.

10 / 96

Introduction Learning outcomes The problem: methodological approach

The role of models in the crisis

‘[T]here is also a strong belief, which I share, that bad or rather

• ver-simplistic and overconfident economics helped create the crisis. There was a

dominant conventional wisdom that markets were always rational and self-equilibrating, that market completion by itself could ensure economic efficiency and stability, and that financial innovation and increased trading activity were therefore axiomatically beneficial” Adair Turner, Ex Chairman of the Financial Services Authority, U.K

11 / 96

Introduction Learning outcomes The problem: methodological approach

Main problems in standard macroeconomics

Focussed on atomistic behaviour with no interactions Use of static equilibrium Little investigation on the evolution towards equilibrium Information transmission (e.g. contagion) assumed away Holds on two crucial assumptions: rationality of individuals and aggregate behaving like a “rational individual”.

12 / 96

Introduction Learning outcomes The problem: methodological approach

Economic models need “sound microfoundations” (Kirman, 2010)

Rational optimising behaviour of the individuals in the market or economy has been widely criticised from Simon onwards. Yet, this is at the heart of the General Equilibrium Model Aggregate behaviour modelled as resulting from such an individual model. Economic structure is lost under aggregation

13 / 96

Introduction Learning outcomes The problem: methodological approach

The aggregate differs from individuals

“The problem of a rational economic order is determined precisely by the fact that the knowledge of the circumstances of which we make use never exists in concentrated or integrated form, but solely as the dispersed bits of incomplete and frequently contradictory knowledge which all the separate individuals possess. The problem is thus in no way solved if one can show that all of the facts, if they were known in a single mind (as we hypothetically assume them to be given to the

• bserving economist), would uniquely determine the solution; instead

we must show how a solution is produced by the interactions of people each of whom possesses only partial knowledge.” (Hayek, 1945)

14 / 96

Introduction Learning outcomes Methodological solution?

Opening to other tools?

“First, we have to think about how to characterise the homo

• economicus at the heart of any model. The atomistic, optimising agents

underlying existing models do not capture behaviour during a crisis period. We need to deal better with heterogeneity across agents and the interaction among those heterogeneous agents. We need to entertain alternative motivations for economic choices. Behavioural economics draws on psychology to explain decisions made in crisis circumstances. Agent-based modelling dispenses with the optimisation assumption and allows for more complex interactions between

• agents. Such approaches are worthy of our attention.” Trichet (2010) – Ex

President of the European Central Bank

15 / 96

Introduction Learning outcomes Methodological solution?

Assumption in this week lectures

No hyper rationality... Heterogeneity Interaction Beyond stable preferences How do collective outcomes emerge from the interaction between individuals each of whom has only a local vision of the system (‘snowflakes’)? Self organising markets ̸= equilibrium

16 / 96

Introduction Learning outcomes Lectures

Aims of lectures

What shall we learn, or learn to problematise:

1 Methodological limitations with standard economics that we

want to address with ABM

2 The use of Agent based simulation models (ABM)

Interaction, networks and complexity

3 Modelling structural change: the interactions between different

aspects of structural change

Growth and inequality Consumer dynamics Market concentration Growth regimes Experimental design in ABM

17 / 96

Introduction Learning outcomes Laboratory for Simulation Development

Aims of the hands-on lectures: LSD

1 Learning how to design, run and analyse an (Agent Based)

simulation model

2 Learning a user friendly modelling application 3 Reproduce existing models and replicate results 4 Basic notions to start developing one own model 18 / 96

Introduction Learning outcomes Outline

Outline

Part I Agent based method Limitations of standard economic theory Economy as an evolving system Agent-Based Computational Economics: ACE Short discussion Short introduction to Laboratory of Simulation Development (LSD) Part II: Applications Structural interactions and long run growth

Experimental design applied to ABM

19 / 96

Introduction Learning outcomes Outline

Literature

A rich repository for Agent Based Models economics:

http://www2.econ.iastate.edu/tesfatsi/ace.htm

Main references we use (more references at the end of slides) Complexity

Arthur, W. B. (2013), ’Complexity Economics: A Different Framework for Economic Thought’, Working Paper 2013-4-2012, Santa Fe Institute. Kirman, A. (2010) Complex Economics: Individual and Collective Rationality Taylor & Francis Page, Scott E. 2015. What Sociologists Should Know About

• Complexity. Working Paper, mimeo.

ABM

Tesfatsion, L. & Judd, K. (ed.) Handbook of Computational Economics: Agent-Based Computational Economics Elsevier, 2006, 2

20 / 96

Introduction Learning outcomes Outline

Literature

Evolutionary modelling

Dawid, H. (2005), Agent–Based Models of Innovation and Technical Change, in Leigh Tesfatsion & K. L. Judd, ed., Handbook of Computational Economics, Volume 2: Agent-Based Computational Economics, North–Holland Safarzyńska, K. & van den Bergh, J. C. J. M. (2010), ’Evolutionary models in economics: a survey of methods and building blocks’, Journal

• f Evolutionary Economics 20(3), 329-373

Macro

Silverberg, G. & Verspagen, B. (2005) Evolutionary Theorizing on Economic Growth in Dopfer, K. (ed.) The Evolutionary Foundations of Economics, Cambridge University Press, 506-539 Verspagen, B. (2006), Innovation and Economic Growth, in Jan Fagerberg; David C. Mowery & Richard R. Nelson, ed., The Oxford Handbook of Innovation, Oxford University Press, Oxford, pp. 487-513.

21 / 96

Introduction Learning outcomes Outline

Literature

ABM Macro

Nelson, R. R. & Winter, S. G. (1982), An Evolutionary Theory of Economic Change, Harvard University Press, Cambridge, MA. Ch 12 & 14 Ciarli, T. (2012), ’Structural interactions and long run growth: An application of Experimental Design to Agent Based Models’, Revue de l’OFCE, Debates and policies 124(5), 295-345. Lorentz, A., T. Ciarli, M. Savona, and M. Valente. 2016. ‘The Effect of Demand-Driven Structural Transformations on Growth and Technological Change.’ Journal of Evolutionary Economics 26(1):219–246. Ciarli, T., and M. Valente. 2016. ‘The complex interactions between economic growth and market concentration in a model of structural change.’ Structural Change and Economic Dynamics forthcomin. Ciarli T., A. Lorentz, M. Savona, and M. Valente. 2016. ‘Growth Regimes and Structural Change’ mimeo. Colander, D.; Howitt, P .; Kirman, A.; Leijonhufvud, A. & Mehrling, P . (2008), ’Beyond DSGE models: toward an empirically based macroeconomics’, American Economic Review 98(2), 236–240. Dosi, G.; Fagiolo, G. & Roventini, A. (2010), ’Schumpeter Meeting Keynes: A Policy–Friendly Model of Endogenous Growth and Business Cycles’, Journal of Economic Dynamics and Control 34, 1748-1767.

22 / 96

Introduction CS CS Properties ACE Discussion LSD

Part I

Agent Based Modelling in the social sciences

23 / 96

Introduction CS CS Properties ACE Discussion LSD Standard economics

Main features of standard economic theory

Rationality: fully rational agents Infinite computation and memory Know the model of the world Know that all other agents are also fully rational No need for learning Interactions Information, knowledge and goods flow through individuals: all connected, no frictions Macro: no interactions Game theory: interactions with all other individuals

24 / 96

Introduction CS CS Properties ACE Discussion LSD Standard economics

Main features of standard economic theory

Heterogeneity Homogeneous Heterogeneity does not change results Average behaviour = behaviour of the average (normal/symmetric distribution) Time and aggregation Economy always in equilibrium: all micro and macro forces compensate, in short and long run No crucial time dimension: infinitely lived agents Repeated static models

GE , Microfoundations , Game Theory 25 / 96

Introduction CS CS Properties ACE Discussion LSD Definition

Characteristics of a complex system (Page, 2015)

Interaction structure (facebook) Interdependent: people influence each other (fads) Learning and adaptation: change, behaviour, connections and interdependency Selection (and variation) Heterogeneity: initial, as a process of adaptation, or innovation

26 / 96

Introduction CS CS Properties ACE Discussion LSD

What does empirical evidence suggests about Information and uncertainty Rationality Interactions Heterogeneity Time and dynamics Emergence?

27 / 96

Introduction CS CS Properties ACE Discussion LSD Knightian uncertainty

Uncertainty

Risk: when we know the probability distribution of future events Incremental innovation Knightian Uncertainty: when the risk cannot be measured Radical innovation: future directions and trajectories of technologies? E.g. environmental impact of innovations Returns from investment in innovations? e.g. pharma before biotech ⇒ procedural, bounded rationality to make decisions (incremental): routinised behaviour ⇒ no Bayesian agent with a clear set of possible outcomes (radical): innovation as a guess, requires intuition, animal spirits

28 / 96

Extreme losses and gains (with non-negligible probability)

Source: Levy (1998)

S&P 500 1 minute rate of return distribution (90-95)

Technology per se drives uncertainty

Source: Little green blog

“dishwasher versus hand washing”: Aproximadamente 855.000 resultados (0,39 segundos)

Introduction CS CS Properties ACE Discussion LSD Rationality

Micro entities with simple and routinised behaviour

Experimental evidence from cognitive Psychology: Kahneman, Tversky, Gigerenzer, etc Difference between Risk and Uncertainty Inherent difficulty in dealing with uncertainty and probability Different risk aversion for gains and losses Bayesian VS frequentist approaches Cognitive biases People take decisions in a relative way, comparing local options

31 / 96

Introduction CS CS Properties ACE Discussion LSD Rationality

Risk aversion

Problem A (win): an individual is given $1,000 A1: Win $1,000 with 50% probability (0 otherwise) A2: Win $500 with certainty Problem B (loss): an individual is given $2,000 B1: Loose $1,000 with 50% probability (0 otherwise) B2: Loose $500 with certainty Rational choice In both cases the expected outcome is $1500 Depending on risk aversion, if the rational individual chooses A1(A2), she should also choose B1(B2)

32 / 96

Introduction CS CS Properties ACE Discussion LSD Rationality

Risk aversion

Lab experiment: A statistically significant majority of individuals choose A2 and B1 ⇒ Individuals are risk lovers for losses and risk averse for gains ⇒ The structure of the problem (decision making) affects the choice

33 / 96

Introduction CS CS Properties ACE Discussion LSD Rationality

Relative decision making

Economist.com subscription: real world experiment Consumers have the following choices

1 Internet only option: $59 2 Print only option: $125 3 Print and Internet option: $125

Result: 16% (1), 0% (2), 84% (3) Consumers have the following choices

1 Internet only option: $59 2 Print and Internet option: $125

Result: 68% (1), 32% (2)

34 / 96

Introduction CS CS Properties ACE Discussion LSD Rationality

Micro entities with simple and routinised behaviour

⇒ Difficult to maximise: individuals are not ready to make all necessary calculations leading to the optimal choice, even if they had all the required information. Adaptive trial and error behavioural rules (Gigerenzer heuristics: simple heuristics more efficient to resolve complex problems) Individuals tend to use first known routines, and if these are not successful they will use calculations.

35 / 96

Introduction CS CS Properties ACE Discussion LSD Rationality

Expectations

Use of imperfect knowledge on the past Experimental evidence on adaptive expectations ⇒ difficult to be rational and predict the preference of other individuals (preferences do change over time)

36 / 96

Introduction CS CS Properties ACE Discussion LSD Heterogeneity

Heterogeneity

Distribution of Consumers: wealth, income, wages, preferences Firms: size, growth rates, productivity Markets: prices, institutions, organisation, peers All scale free distributions (Pareto): variance tends to infinite Meaning of an average?

47 / 96

Introduction CS CS Properties ACE Discussion LSD Heterogeneity

World income distribution

Source: Ortiz and Cummins (2011) 48 / 96

Introduction CS CS Properties ACE Discussion LSD Heterogeneity

Skewed distribution of innovation by size

Not all innovations are equally relevant Citations, value, returns to investment Returns to innovation are also fat tailed (high kurtosis): variance is not finite ⇒ SO is the risk of of returns

49 / 96

Introduction CS CS Properties ACE Discussion LSD Heterogeneity

Computed Tomography scanner patents

20 40 60 80 Citations 50 100 150 200 Patents

(a) Linear scale

50 100 150 200 Patents 20 40 60 80 Citations

Quantile-Quantile Plot

(b) Q-Q Plot

Source: Trajtenberg (1990)

50 / 96

Introduction CS CS Properties ACE Discussion LSD Heterogeneity

Innovation size distributions (Pareto Plots) based on patent citations

Source: Gerry Silverberg

EPO 1989 patent citations (left) and USPTO 1989 patent citations (right)

51 / 96

Pareto distributions are a feature of complex system

Source: Mitzenmacher (2004)

Introduction CS CS Properties ACE Discussion LSD Heterogeneity

Heterogeneity in firms and innovation

Distribution of firm size in an industry is heavily skewed (Simon and Bonini, 1958) Firms’ heterogeneity persists through time (technology, productivity, profits, growth) (Dosi et al., 2010b) Large differences across sectors and small differences across countries within sectors in firm’s demography (Breschi et al., 2000) Size of innovation is also drawn from a very skewed distribution (Silverberg and Verspagen, 2007)

53 / 96

Introduction CS CS Properties ACE Discussion LSD Heterogeneity

Heterogeneity is a relevant property

Many economic phenomena are driven by heterogeneity: diffusion curves, business fluctuations, pricing

54 / 96

Introduction CS CS Properties ACE Discussion LSD Heterogeneity

Heterogeneity in the aggregate?

Heterogeneity does not cancel out in the aggregate, unless characteristics are normally distributed. Some components dominate (e.g. the financial crisis does not

• ccur because a couple of small firms fail)

Imitation and avalanches in networks ⇒ Structural change ⇒ Changing shares of populations Aggregate properties may differ from individual properties: many times impossible to predict from individual behaviour Path dependency and non ergodic systems – stochastic process is time variant Cumulative process: persistent heterogeneity (e.g. Diamond, 1997)

55 / 96

Introduction CS CS Properties ACE Discussion LSD Interactions

Interactions and networks

Social sciences are all about interactions Trade Information and knowledge Expectations Social relations: e.g. six degrees of separation Neighbourhoods Most natural, technological and social interactions can be described as networks Most real networks have very similar properties: complex systems

56 / 96

Introduction CS CS Properties ACE Discussion LSD Interactions

Social networks

(a) Facebook connections (b) Twitter followers

57 / 96

Business collaborations: internet industry

Source: http://www.orgnet.com/netindustry.html

250 companies: announced joint ventures, strategic alliances, other partnership

Business informal relations

Source: http://blog.kiwitobes.com

400 largest US companies: shared board members

Combined knowledge in research: citation networks

Source: (Rafols et al., 2012)

Disciplines of study (publications) at the Institute of the Study of Science Technology and Innovation (ISSTI), Edinburgh

Introduction CS CS Properties ACE Discussion LSD Interactions

Network and complexity

Network as constraints Behaviour of a node depends on all others: interdependence Networks define the order of a complex system

61 / 96

The aggregate effect of social influence

Source: Salganik et al. (2006)

A: songs ordered randomly. C: songs ordered by downloads ⇒ Path dependence and cumulation

Introduction CS CS Properties ACE Discussion LSD Interactions

Example 1: NK (Kauffman and Levin, 1987)

Complexity is the product of interactions The fitness of a system F depends only on the interaction structure among its N nodes (and their mutation strategy) Each node i ∈ N is connected to K other nodes Each node i has a fitness contribution f given by a string of K + 1 binary values

Independent from other nodes (K = 0) Dependent on other nodes’ string value (K > 0)

K (interactions) defines complexity (product decomposability (Simon, 2002))

63 / 96

Introduction CS CS Properties ACE Discussion LSD Interactions

Example 2: Business Fluctuations (Ciarli and Valente, 2007)

See also Weisbuch and Battiston (2005) and Carvalho (2014). How do micro shocks affect the system fluctuation Production: Input-Output structure Identical firms with iid shocks Consecutive decisions in adjusting a small shock in the final demand generates fluctuations

⇒ Attempts to smooth business cycles generate business cycles Fast adjustments in the demand for inputs create hysteresis Micro volatility is always smaller than aggregate volatility

64 / 96

Introduction CS CS Properties ACE Discussion LSD Interactions

Diffusion of goods and technologies

Is influenced by social networks “In 1953, a young female Macaque monkey in the south of Japan washed a muddy sweet potato in a stream before eating it. This

• bvious improvement in food preparation was imitated quickly by
• ther monkeys and in less than 10 years it became the norm in her

immediate group; by 1983, the method had diffused completely” (Hall, 2006, p. 459) ⇒ Some contagion effect: learn about the new technology from somewhere ⇒ Time to diffuse

65 / 96

Introduction CS CS Properties ACE Discussion LSD Interactions

Contagion is the simpler way to explain diffusion

Source: VisializingEconomics

Contagion models: each user contacts a non adopter with some probability (Geroski, 2000)

66 / 96

Introduction CS CS Properties ACE Discussion LSD Equilibrium

Equilibrium

Evolutionary dynamics Each time period a process of adaptation, change in population, emergence of novelty Path dependent cumulative causation Even assuming equilibrium, need to know how it is reached: shocks, transitions, adjustments?

67 / 96

Introduction CS CS Properties ACE Discussion LSD Summing up

Comparing Economic models and Complex Systems

Standard Economic models Complex Evolving Systems Individuals 1,2 or infinite, fully rational, sophisticated learning N large but finite, simple en- tities, adaptive, routinary be- haviour Interactions Extreme cases, trivial pat- terns (full or empty/star graphs) Non trivial patterns, local in- teractions with subset of other agents Diversity Possibly heterogeneous, but diversity does not matter for aggregate dynamics Persistently heterogeneous, diversity matters for aggre- gate dynamics Time and Aggregate Dynamics Static (not truly dynamic) models,

• nly equilibrium

states count Truly dynamic systems, equilibria possibly irrele- vant, meta-stable states and emergent (self-organized) properties

68 / 96

Introduction CS CS Properties ACE Discussion LSD Summing up

Empirical evidence: complex system features of social systems

No evidence of purely rational behaviour Plenty of evidence of direct interactions within and between different populations Plenty of evidence on heterogeneity Little evidence of equilibria, only temporary (a particular condition) Macro as aggregation of micro properties and dynamics

69 / 96

Introduction CS CS Properties ACE Discussion LSD Summing up

Given the knowledge gained from empirical observations in different sciences

“Can one do good science based on models whose assumptions are clearly at odds with empirical evidence?” (Dick Day, 2003) Wouldn’t economics make more empirical sense if it were based on how do people actually behave, interact, etc, rather than how people should behave, interact, etc? (Miller and Page, 2007) “Water which is too pure has no fish”

70 / 96

Introduction CS CS Properties ACE Discussion LSD Why

Why simulations?

In order to analyse complex social problems (i.e. development and environmental sustainability) we need a different class of models that can Embed realistic assumptions into micro and macro models: uncertainty, procedural decision, heterogeneity, local interactions, non-equilibrium Replicate some of the empirical evidence discussed Include structural changes Do not assume macro dynamics

71 / 96

Introduction CS CS Properties ACE Discussion LSD Why

Why simulations?

Interaction of objects (agents) as a complex problem → no analytical solution Social interaction as a complex problem with individual behaviour (less straightforward then physical behaviour) No closed system Interaction of heuristics and reaction heuristics “I can calculate the motion of heavenly bodies, but not the madness of people” (I. Newton) Simple interactions can lead to complex outcomes (Arthur, 1994; Schelling, 1971) Minority games, urban segregation, choice of a technology The place where you are sitting now

72 / 96

Introduction CS CS Properties ACE Discussion LSD Why

Schelling segregation model: explaining segregation

Micromotives and Macrobehaviour (Schelling, 1978): segregation can be explained by simple individual choices (no racism...)

(a) Stay if at least 1/3 of

neighbours are ‘kin’

(b) Move to random location otherwise

Source: L-E Cederman

73 / 96

Introduction CS CS Properties ACE Discussion LSD Why

Schelling segregation model: set-up

N agents located on a 2-dimensional grid (torus) of LxL cells. Types: Each agent can be either RED or GREEN Only a percentage p of cells is occupied: N < LxL Agents are initially located on the grid at random In each time period, agents may be happy or unhappy Agent cares about the proportion q of other agents of same colour in its Moore neighbourhood of radius 1 Agents are unhappy if q is below a certain critical threshold (parameter of the model) and happy otherwise In each iteration of the model one unhappy agent is randomly selected to move to a random empty cell in the lattice

74 / 96

Introduction CS CS Properties ACE Discussion LSD Why

Schelling segregation model: experiments

Source: Izquierdo et al. (2009)

http://jasss.soc.surrey.ac.uk/12/1/6/appendixB/ Schelling1969.html

75 / 96

Introduction CS CS Properties ACE Discussion LSD Why

Schelling segregation: Chicago 1940

Source: Möbius and Rosenblat (2001)

Percentage of blacks: 1-5% yellow; 5-10% pink; 10-25% orange; 25-50% red; 50-75% dark red; 75-95% brown; > 95% black 76 / 96

Introduction CS CS Properties ACE Discussion LSD Why

Schelling segregation: Chicago 1950

Source: Möbius and Rosenblat (2001)

Percentage of blacks: 1-5% yellow; 5-10% pink; 10-25% orange; 25-50% red; 50-75% dark red; 75-95% brown; > 95% black 77 / 96

Introduction CS CS Properties ACE Discussion LSD Why

Schelling segregation: Chicago 1960

Source: Möbius and Rosenblat (2001)

Percentage of blacks: 1-5% yellow; 5-10% pink; 10-25% orange; 25-50% red; 50-75% dark red; 75-95% brown; > 95% black 78 / 96

Introduction CS CS Properties ACE Discussion LSD What

ACE: Definition

ACE Agent–Based Computational Economics: “the computational study of economic processes modelled as dynamic systems of interacting agents” (L. Tesfatsion) Modeller constructs a virtual economic world populated by various agent types (economic, institutional, social, biological, physical) Modeller sets initial world conditions Modeller then steps back to observe how the world develops over time (no further intervention by the modeller is permitted) World events are driven by agent interactions

79 / 96

Introduction CS CS Properties ACE Discussion LSD What

Main properties of ACE

Population of heterogeneous (economic) ‘agents’ Agents live in complex systems evolving through time (Kirman, 1998). True dynamics: non reversible “Hyper-rationality” not viable (Dosi et al., 1996): internal states, rules

• f behaviour, and adaptive expectations

Agents are autonomous or semi–autonomous Agents interact with one another and possibly with an environment (local/social interactions) Endogenous and persistent novelty (technological change):

• pen-ended spaces

Aggregate structure emerges from agent interactions (Tesfatsion, 1997) Generations of agents emerge from the interactions of their ancestors (selection, retention, innovation → evolution) (Nelson and Winter, 1982)

80 / 96

Introduction CS CS Properties ACE Discussion LSD What

Micro and macro interactions

Source: Page (2015)

81 / 96

Introduction CS CS Properties ACE Discussion LSD Implementation and analysis

Structure of ABM

Time t = 0, 1, 2, ..., (T) Discrete Sets of Agents It = 1, 2, ..., Nt Often Nt = N Sets of Micro States i → xi,t Firm’s output Vectors of Micro-Parameters i→θi

• Res. Wage

Vector of Macro-Parameters Θ ∈ ℜm

• Min. Wage

Interaction Structures Gt ∈ ℘(It) Networks Micro Decision Rules Ri,t(·|·) Innovation rule Aggregate variables Xt = f(x1,t, ..., xNt,t) GNP

82 / 96

Introduction CS CS Properties ACE Discussion LSD Implementation and analysis

Implementation of AB computations

Each agent is an object instance variables (representing internal states) and methods (representing behavioural routines) The population of agents is also an object → Upstream: hierarchies (emergent properties and interactions) Topology of interaction, e.g., a spatial environment or a social network, a market Mechanisms for activating agents: incentives & routines Usually stochastic processes (uncertainty)

83 / 96

Introduction CS CS Properties ACE Discussion LSD Analysis

Realisations

Highly parametrised: analysis Parameters of interest: functional analysis Whole space / reasonable space Stochastic processes Uncertainty: sequence of stochastic events can have a strong effect on the outcome (e.g. percolation) Analyse distribution of each output variable Each realisation a scenario (consistent with the model and in probability) Analysis of plausible scenarios

84 / 96

Introduction CS CS Properties ACE Discussion LSD Analysis

Robustness

Calibration Abstract model vs explanation of a phenomenon vs foresight Reproducing empirical evidence, under given parameter values (validation) Robust assumptions

85 / 96

Introduction CS CS Properties ACE Discussion LSD Analysis

Procedure of ABM

Initial Conditions: ( xi,0 ) Micro & Macro Pars: (θi ), Θ Generate Time-Series through Simulation {( xi,t ), t =1,…,T} { Xt , t =1,…,T} Compute a Set of Statistics S= {s1, s2 , … }

• n micro/macro Time-Series

Repeat M ind. times

Generate Montecarlo Distribution for each Statistics in S= {s1, s2 , …} Studying how Montecarlo Distributions of Statistics in S= {s1, s2 , …} behave as initial conditions, micro and macro parameters change Statistical Tests for difference between moments

Source: G. Silverberg 86 / 96

Introduction CS CS Properties ACE Discussion LSD Examples

Some applications in economics and business

Evolutionary-Games: P . Young, Kandori et al., Blume, Ellison (Local) Interaction Models: Kirman, Weisbuch, Lux Endogenous Network Formation: Vega-Redondo, Cowan, Goyal, Jackson-Watts...) Innovation (Polya-Urn Schemes): Arthur, Dosi, Kaniovski, Lane, Marengo Complexity: Frenken, Valente, Marengo Strategy and organisations: Carley and Pietrula, Lomi and Larsen Technological modularity, firm and industry organisation: Ethiraj et al. (2007); Frenken et al. (1999); Kauffman et al. (2000); Marengo and Dosi (2005); Ciarli et al. (2008)

87 / 96

Introduction CS CS Properties ACE Discussion LSD Examples

Some applications in economics and business

Growth: Nelson and Winter (1982), Silverberg, Verspagen, Dosi, Howitt, Llerena and Lorentz (2004); Dawid and Fagiolo (2008); Dosi et al. (2010a); Ciarli et al. (2010); Ciarli (2012); Ciarli et al. (2012); Fagiolo and Roventini (2012) Firms location: David et al. (1998) Firms and technological change: Dawid (2006); Teitelbaum and Dowlatabadi (2000); Yildizoglu (2002) Markets: Axtell, Epstein, Tesfatsion, Kirman and Vriend (2000) Electricity markets: Tesfatsion Sectoral studies: Malerba et al Environmental economics: van den Bergh, Safarzynska, Windrum et al. (2009a,b)

88 / 96

Introduction CS CS Properties ACE Discussion LSD Examples

Some applications in economics and business

Industrial life cycle cycles: Windrum and Birchenhall (2005), Malerba et al Labour market: Tesfatsion, Fagiolo et al. (2004), Richiardi and Leombruni Financial markets (a huge number): Delli Gatti et al. (2004), Delli Gatti and Stiglitz, Cont, econophisycs Macro instability: Bak et al. (1993); Dosi et al. (2006), Weisbuch and Battiston, Ciarli and Valente (2007) Macro: Howitt, Duffy, Arifovic Firms coalition and network formation: Cowan and Jonard, Ozman, Page, Huberman, Axtell, Vega-Redondo, Jackson, Watts Foresight: Lempert Other social sciences: Politics (state cooperation, conflict), Sociology, Anthropology, ...

89 / 96

Introduction CS CS Properties ACE Discussion LSD Some critical aspects

Do we really need simulations?

It depends on the phenomenon under study See introductory discussion on the crisis, and empirical regularities of complex systems: societies are complex. How reasonable and helpful are the assumptions for what we want to study?

90 / 96

Introduction CS CS Properties ACE Discussion LSD Some critical aspects

Critical aspects of simulations

See discussions in Windrum (2007); Valente (2007); Windrum et al. (2007); Leombruni and Richiardi (2005) Determinacy of results

• Sim. models produce all and only whatever you code into them

True, but computers help to understand exactly the implications

• f the assumptions. Think, for example, of models of weather
• forecasting. The basic physics is trivial, but the aggregate effect is

impossible to derive by analytical means, and computers help to fill the gap between the hypotheses (e.g. basic physics) and their implications (forecasting).

91 / 96

Introduction CS CS Properties ACE Discussion LSD Some critical aspects

Critical aspects of simulations

Empirical validation Are results confirmed empirical observations? (Windrum, 2007) Data replication is useless without understanding their meaning. Worse, there are always a large number of different ways to replicate a data set, only some of which may make sense. ֒ → First need to have robust evidence on assumptions on micro behaviour Results come in the form of distributions, depending on the randomness of initial conditions and on interactions

92 / 96

Introduction CS CS Properties ACE Discussion LSD Some critical aspects

Critical aspects of simulations

Testing randomness and parameter space Random models/models with many parameters must be adequately tested for the robustness of results: a single random run may be an exceptional case in a distribution Crucial, open issues Pushing policy and design exercises Fostering empirical validation techniques

93 / 96

Introduction CS CS Properties ACE Discussion LSD A short intro

Outline for the hands on simulation course

Laboratory for Simulation Development:

thttp://www.labsimdev.org/Joomla_1-3/

Last version: https://github.com/marcov64/Lsd Objective: learning how to use simulations implemented in LSD to make research in Economics Plan Introduction: goals and plan of the course Definitions: a normal form of a simulation model. Introduction to LSD: equations, structures and configurations of models. Tutorials: implementation of increasingly complex example models.

94 / 96

Introduction CS CS Properties ACE Discussion LSD Broad methodological issues

Simulation programme

Using a standard programming language the most difficult task is not the coding of the model. Rather it is the coding of ancillary tools necessary to declare the model’s elements, assign initial values, export results, etc. Using LSD, conversely, the modeller focuses only on the model, and the system automatically generates professional tools to control and access any aspect of the model.

95 / 96

Introduction CS CS Properties ACE Discussion LSD Hands on

Topics of the course

During the course we will approach the following topics: Design: decide what the model should look like, for it to contribute to a research project. Implementation: turning an abstract idea into a working simulation program. Interpretation: extracting knowledge from simulation models.

96 / 96

Backup slides References Introduction

General Equilibrium in a slide

Key Assumptions Full rationality (fully informed optimizing agents) No time and no dynamics Only equilibrium analysis No interactions (Star Network) Positive questions: equilibrium Existence and Uniqueness Stability

97 / 96

Backup slides References Introduction

General Equilibrium: limitations

Back

Meaning of existence of a social equilibrium (observation?) How can an equilibrium be established (Walrasian auctioneer)? What happens out of equilibrium (a part from instantaneous attraction)? What (or who) is an auctioneer ? Interactions ? How does an economy move from an equilibrium to another one? What happens in between ? Assumptions on micro behaviour and predictive power (e.g. crisis)

98 / 96

Backup slides References Introduction

Microfounded macroeconomic models in a slide

Example: economic growth Representative firm with a production function: F(Lt, AtKt) Representative household with utility function: U = ∫ ∞

t=0 e−ρtu(C(t)) Lt H dt

Both firms and households are fully rational (maximizing) agents, with unbounded computational skills

99 / 96

Backup slides References Introduction

Microfounded macroeconomic models: limitations

Back

Rationality assumptions No space for radical innovation and structural change Aggregate change in GNP should be interpreted as a transition through equilibria (but the model is static) (Kirman 1989) Heterogeneity and aggregation: Preferences of a RA over available choices may be very different from those of the society Reaction of a RA to shocks need not reflect how individual react to shocks

100 / 96

Backup slides References Introduction

Game theory in a slide

N agents with choice set A Payoffs: Πi(ai; aj, j ̸= i) (stage-game matrix) Agents are fully rational Analysis: Nash equilibria Interactions: Complete Network (vs. Star network) Example: Oligopoly with N firms (Bertrand Model)

101 / 96

Backup slides References Introduction

Game theory: limitations

Back

Rational assumptions Computational abilities of an agent (backward inductions? assumptions on the other agents behaviour?) Tractability when N is large and games are repeated a finite number

• f times (regress problem)

Predictive power in infinitely repeated games

102 / 96

Backup slides References Introduction

References I

Arthur, W. B. (1989). Competing technologies, increasing returns and lock-in by historical events. Economic Journal, 99:116–131. Arthur, W. B. (1994). Inductive reasoning and bounded rationality. American Economic Review, 84:406. Bak, P ., Chen, K., Scheinkman, J., and Woodford, M. (1993). Aggregate Fluctuations from Independent Sectoral Shocks: Self–Organized Critically in a Model of Production and Inventory

• Dynamics. Ricerche Economiche, 47(1):3–30.

Breschi, S., Malerba, F., and Orsenigo, L. (2000). Technological Regimes and Schumpeterian Patterns of Innovation. The Economic Journal, 110(463):388–410. Carvalho, V . M. (2014). From Micro to Macro via Production

• Networks. Journal of Economic Perspectives, 28(4):23–48.

103 / 96

Backup slides References Introduction

References II

Ciarli, T. (2012). Structural interactions and long run growth: An application of experimental design to agent based models. Revue de l’OFCE, Debates and policies, 124(5):295–345. Ciarli, T., Leoncini, R., Montresor, S., and Valente, M. (2008). Technological change and the vertical organisation of industries. Journal of Evolutionary Economics, forthcomin. Ciarli, T., Lorentz, A., Savona, M., and Valente, M. (2010). The effect

• f consumption and production structure on growth and
• distribution. A micro to macro model. Metroeconomica,

61(1):180–218.

104 / 96

Backup slides References Introduction

References III

Ciarli, T., Lorentz, A., Savona, M., and Valente, M. (2012). The role

• f technology, organisation, and demand in growth and income
• distribution. LEM Working Papers 2012/06, Laboratory of

Economics and Management, San’Anna School of Advanced Studies, Pisa. Ciarli, T. and Valente, M. (2007). Production Structure and Economic

• Fluctuations. LEM Working paper Series 2007/02, Laboratory of

Economics and Management Sant’Anna School of Advanced Studies, Pisa. Cowan, R. and Gunby, P . (1996). Sprayed to Death: Path Dependence, Lock-in and Pest Control Strategies. The Economic Journal, 106(436):521–542.

105 / 96

Backup slides References Introduction

References IV

David, P . A., Foray, D., and Dalle, J.-M. (1998). Marshallian Externalities and the Emergence and Spatial Stability of Technological Enclaves. Economics of Innovation and New Technology, 6:147–182. Dawid, H. (2006). Agent–Based Models of Innovation and Technical

• Change. In Tesfatsion, L. and Judd, K. L., editors, Handbook of

Computational Economics, Volume 2: Agent-Based Computational Economics, chapter 25, pages 1235–1272. North-Holland. Dawid, H. and Fagiolo, G. (2008). Agent-based models for economic policy design: Introduction to the special issue. Journal of Economic Behavior and Organization, 67(2):351–354.

106 / 96

Backup slides References Introduction

References V

Delli Gatti, D., Di Guilmi, C., Gaffeo, E., Giulioni, G., Gallegati, M., and Palestrini, A. (2004). A New Approach to Business Fluctuations: Heterogeneous Interacting Agents, Scaling Laws and Financial Fragility. Journal of Economic Behavior & Organization, 56:489–512. Diamond, J. (1997). Guns, Germs, and Steel. W. W. Norton. Dosi, G., Fagiolo, G., and Roventini, A. (2006). An Evolutionary Model of Endogenous Business Cycles. Computational Economics, 27:3–34. Dosi, G., Fagiolo, G., and Roventini, A. (2010a). Schumpeter Meeting Keynes: A Policy–Friendly Model of Endogenous Growth and Business Cycles. Journal of Economic Dynamics and Control, 34:1748–1767.

107 / 96

Backup slides References Introduction

References VI

Dosi, G., Lechevalier, S., and Secchi, A. (2010b). Introduction: Interfirm heterogeneity–nature, sources and consequences for industrial dynamics. Industrial and Corporate Change, 19(6):1867–1890. Ethiraj, S. K., Levinthal, D. A., and Roy, R. R. (2007). The Dual Role

• f Modularity: Innovation and Imitation. Management Science,

Forthcomin. Fagiolo, G., Dosi, G., and Gabriele, R. (2004). Matching, Bargaining, and Wage Setting in an Evolutionary Model of Labor Market and Output Dynamics. Advances in Complex Systems, 14:237–273. Fagiolo, G. and Roventini, A. (2012). Macroeconomic Policy in DSGE and Agent-Based Models. Revue de l’OFCE, 124(5):67.

108 / 96

Backup slides References Introduction

References VII

Frenken, K., Marengo, L., and Valente, M. (1999). Interdependencies, {N}early-{D}ecomposability and {A}daptation. In Brenner, T., editor, Computational Techniques for Modelling Learning in Economics. Kluwer, Boston Dordrecht and London. Geroski, P . (2000). Models of technology diffusion. Research Policy, 29(4-5):603–625. Hall, B. H. (2006). Innovation and Diffusion. In Fagerberg, J., David

• C. Mowery, and Nelson, R. R., editors, The Oxford Handbook of

Innovation, chapter 17, pages 459–484. Oxford University Press, Oxford. Hayek, F. A. (1945). The Use of Knowledge in Society. The American Economic Review, 35(4):519–530.

109 / 96

Backup slides References Introduction

References VIII

Izquierdo, L. R., Izquierdo, S. S., Galán, J. M., and Santos, J. I. (2009). Techniques to Understand Computer Simulations: Markov Chain

• Analysis. Journal of Artificial Societies and Social Simulation, 12(1):6.

Kauffman, S. A. and Levin, S. (1987). Towards a general theory of adaptive walks on rugged landscapes. Journal of Theoretical Biology, 128(1):11–45. Kauffman, S. A., Lobo, J., and Macready, W. G. (2000). Optimal search on a technology landscape. Journal of Economic Behavior & Organization, 43(2):141–166. Kirman, A. (2010). Complex Economics: Individual and Collective

• Rationality. Graz Schumpeter lectures. Taylor & Francis.

110 / 96

Backup slides References Introduction

References IX

Kirman, A. and Vriend, N. J. (2000). Learning to be Loyal. A Study of the Marseille Fish Market. In Delli Gatti, D., Gallegati, M., and Kirman, A., editors, Interaction and Market Structure. Essays on Heterogeneity in Economics (Lecture Notes in Economics and Mathematical Systems 484), pages 33–56. Springer, Berlin. Leombruni, R. and Richiardi, M. (2005). Why are economists sceptical about agent–based simulations? Physica A, 355:103–109. Levy, S. (1998). Wealthy People and Fat Tails: An Explanation for the Levy Distribution of Stock Returns. University of California at Los Angeles, Anderson Graduate School of Management. Llerena, P . and Lorentz, A. (2004). Cumulative Causation and Evolutionary Micro-founded Technical Change: On the Determinants of Growth Rates Differences. Revue Economique, 55(6):1191–1214.

111 / 96

Backup slides References Introduction

References X

Marengo, L. and Dosi, G. (2005). Division of labor, organizational coordination and market mechanisms in collective problem-solving. Journal of Economic Behavior & Organization, 58(2):303–326. Miller, J. H. and Page, S. E. (2007). Complex Adaptive Systems: An Introduction to Computational Models of Social Life. Princeton Studies in

• Complexity. Princeton University Press, Princeton, New Jersey.

Minoiu, C. and Reyes, J. A. (2011). A network analysis of global banking: 1978฀ 2009. IMF Working Paper 11/74, IMF. Mitzenmacher, M. (2004). A brief history of generative models for power law and lognormal distributions. Internet Math., 1(2):226–251. Möbius, M. M. and Rosenblat, T. S. (2001). The Process of Ghetto Formation: Evidence from Chicago. Working paper mimeo, Harvard University.

112 / 96

Backup slides References Introduction

References XI

Nelson, R. R. and Winter, S. G. (1982). An Evolutionary Theory of Economic Change. Harvard University Press, Cambridge, MA. Ortiz, I. and Cummins, M. (2011). Global inequality: Beyond the bottom billion. a rapid review of income distribution in 141

• countries. Social and economic policy working paper, UNICEF,

New York. Page, S. E. (2015). What Sociologists Should Know About Complexity. Rafols, I., Leydesdorff, L., O’Hare, A., Nightingale, P ., and Stirling, A. (2012). How journal rankings can suppress interdisciplinary research: A comparison between Innovation Studies and Business & Management. Research Policy, 41(7):1262–1282.

113 / 96

Backup slides References Introduction

References XII

Salganik, M. J., Dodds, P . S., and Watts, D. J. (2006). Experimental study of inequality and unpredictability in an artificial cultural

• market. Science, 311(5762):854–6.

Schelling, T. C. (1971). Dynamic Models of Segregation. Journal of Mathematical Sociology, 1:143–186. Schelling, T. C. (1978). Micromotives and Macrobehavior. W. W. Norton and Co., New York, NY . Silverberg, G. and Verspagen, B. (2007). The size distribution of innovations revisited: An application of extreme value statistics to citation and value measures of patent significance. Journal of Econometrics, 139(2):318–339. Simon, H. A. (2002). Near decomposability and the speed of

• evolution. Industrial and Corporate Change, 11(3):587–599.

114 / 96

Backup slides References Introduction

References XIII

Simon, H. A. and Bonini, C. P . (1958). The Size Distribution of Business Firms. The American Economic Review, 48(4):607–617. Teitelbaum, D. and Dowlatabadi, H. (2000). A Computational Model

• f Technological Innovation at the Firm Level. Computational and

Mathematical Organization Theory, 6(3):227–247. Trajtenberg, M. (1990). A Penny for Your Quotes: Patent Citations and the Value of Innovations. RAND Journal of Economics, 21(1):172–187. Valente, M. (2007). Qualitative Simulation Modelling. mimeo, Università dell’Aquila. Weisbuch, G. and Battiston, S. (2005). Production Networks and Failure Avalanches. Working Paper mimeo, Ecole Normale SupErieure.

115 / 96

Backup slides References Introduction

References XIV

Windrum, P . (2007). Neo-Schumpeterian simulation models. In Hanusch, H. and Pyka, A., editors, The Edward Elgar Companion to Neo–Schumpeterian Economics. Edward Elgar, Cheltenham. Windrum, P . and Birchenhall, C. (2005). Structural change in the presence of network externalities: a co-evolutionary model of technological successions. Journal of Evolutionary Economics, 15:123–148. Windrum, P ., Ciarli, T., and Birchenhall, C. (2009a). Consumer heterogeneity and the development of environmentally friendly

• technologies. Technological Forecasting and Social Change, 76(4):533–551.

Windrum, P ., Ciarli, T., and Birchenhall, C. (2009b). Environmental impact, quality, and price: Consumer trade-offs and the development of environmentally friendly technologies. Technological Forecasting and Social Change, 76(4):552–566.

116 / 96

Backup slides References Introduction

References XV

Windrum, P ., Fagiolo, G., and Moneta, A. (2007). Empirical Validation of Agent-Based Models: Alternatives and Prospects. Journal of Artificial Societies and Social Simulation, 10(2):8. Yildizoglu, M. (2002). Competing R&D Strategies in an Evolutionary Industry Model. Computational Economics, 19:51–65.

117 / 96