What Can Economists Learn from Ants? Regina Martinez Economics Department, GWU Nov. 8, 2012

Transcription

1 What Can Economists Learn from Ants? Regina Martinez Economics Department, GWU Nov. 8,

2 What this talk is about Sharing what I learned at Stanford University about Networks and Complex Systems (SFI) Inspiring about innovative ways of understanding Economics Intellectual excitement and fun Not original research Not trying to convince you 2

3 THE GREAT RECESSION The recent crisis has put classical economic thinking under huge pressure Some decision makers in a state of shocked disbelief (Alan Greenspan) But, exciting for new ideas! Institute of New Economic Thinking (INET) to re-thinking about the Economics profession Created/funded by Soros 6 Nobel Laureates on its Board 3

4 Few economists saw our current crisis coming, but the predictive failure was the least of the field s problems. More important was the profession s blindness to the very possibility of catastrophic failures in a market economy (Krugman, The New York Times, Sep.6, 2009) 4

5 A black swan is an event that is improbable Yet, it causes massive consequences. The book argues that, despite these events explain almost everything about our world, we are blind to them. Nassim Taleb (2007) 5

6 6

7 LOOKING AT THIS, ONE WONDERS: 1. Are agents really rational? 2. Clearly, the behavior of the crowd is fundamentally different from that of a single individual. Then, does the representative agent reflect the reality? 7

8 THE ECONOMY IS COMPLEX The theory of perfectly rational representative agent does not capture the complexity. Other complex physical systems in nature (earthquakes, avalanches, etc.) Physicists found that the dynamics come from collective effects 8

9 This means that: Individual components have a simple behavior, but interactions between them lead to new complex phenomena The whole is fundamentally different from any of its sub-parts 9

10 LIKE ANTS! Single ant Ant bridge 10

11 LIKE ANTS! Main question: How is it possible that systems with: Simple components Non-linear interactions No central control give rise to complex, adaptive, intelligent behavior? 11

12 ARE ECONOMIC AGENTS LIKE ANTS? Individuals are relatively simple When they create complexity is in their interaction with others (markets, etc.) Perhaps the dynamics of economic systems have the same underlying mechanisms? Some suggest an alternative approach in which the strong classical assumptions not needed 12

13 ALTERNATIVE APPROACH Economic agents are heterogeneous What matters is the interactions among them Macroeconomics from the bottom up (Farmer) Use alternative models in which macroeconomics emerge from the microscopic interactions of individuals 13

14 AGENT BASED MODELS Definition: ABM are computational models in which a large number of agents interact through prescribed rules Advantages: ABM can handle non-linear behavior It does not assume the economy is at a steadystate Allows to keep track of the many interactions over time 14

15 AGENT BASED MODELS Result: ABM are capable of generating complex dynamics even with simple behavioral rules Does it remind you to something? Question: Are ABM able to model the complex economy? 15

16 HISTORY OF ABM IN ECONOMICS The earliest ABM were developed was in 1960s by Thomas Schelling (Nobel Prize). But, there has been limited interest in developing these approaches 16

17 WHEN ARE ABM USEFUL? ABM are well-suited for modeling: 1. Trajectory over time is important 2. Heterogeneity of actors 3. Processes with bounded rationality and imperfect information 4. The system does not have stable equilibria 17

18 DISADVANTAGES OF ABM Quality control: not much guidance yet Choosing the decision rules Highly data intensive Huge number of parameters It demands serious computing power Multidisciplinary collaboration 18

19 EXAMPLES OF ABM Geanakoplos et al. (2012) Goal: Test 2 causes of the housing bubble and crash Low interest rates Too much leverage Motivation: traditional methods give inconclusive results Methodology: build ABM using data from DC housing market to choose parameter values 19

20 Model fit: Results: With frozen int. rates housing boom With frozen leverage NO boom Both frozen: No boom 20

21 FROM ABM TO NETWORKS Since ABM is about the inter-linkages of agents, they are based on a network structure Network Science 21

22 WHAT IS A NETWORK? A network is a representation of interactions between units Units are the nodes People Firms Computers Webpages Interactions : links Friendship Trade Hyperlinks Data transfer 22

23 Network science studies all kinds of Social Interactions interactions: 23

24 Relationship Network Source: Chains of Affection: The Structure of Adolescent Romantic and Sexual Networks (Bearman, Moody, and Stove, 2004) 24

25 Information networks Citations Political blogs 25

26 Technology networks Internet 26

27 Transportation World Airline Route-map 27

28 28

29 The Small World Concept Experiment by Stanley Milgram (1967) Goal was to show how many degrees of separation between any 2 persons 29

30 Six Degrees of Separation Result: But IT S GETTING SMALLER... 30

31 How many steps between and Obama? 31

32 How many steps btw Art and Obama? 32

33 Aside: summary of the basics Origins in Graph Theory (Euler, 1735) Fundamental question from NS perspective: If you are given a network, how do you analyze it? Which are the main organizing principles? What do these properties tell you? 33

34 A bit of terminology Degree (k): number of connections a node has Then, 34

35 Paul Erdos and Alfred Renyi in late 1950s put the basics down From math perspective, it s very simple: Networks are nodes connected through links Let s assume they connect randomly The degree distribution of a random network is a Poisson distribution 35

36 Random (Erdos-Renyi) network Degree distribution of a random n. Is this what we find in reality? 36

37 Not quite. A very common pattern is: It follows a Power-law distribution Small number of hubs and many peripheral nodes Scale-free Network It has a power-law distribution (long-tail) 37

38 Intuitively, are networks random? NO People don t connect randomly but based on friendship, interests, language, race, etc. What about networks we did not create? Metabolic network within our cell, protein network, neuronal, etc. 38

39 Protein network Neural networks in the auditory cortex They are both scale-free networks 39

40 What s wrong with random networks? One of the problems is that networks expand according to the preferential attachment : New nodes prefer to link to highly connected nodes [Netlogo simulation] 40

41 Does it matter what type of network? YES. It matters, for example, for robustness Ability to function even when some components are not functional What happens if nodes are removed? 41

42 It depends: random removal or targeted attack? Random networks are fragile to random failure Scale-free networks are: robust to random failure but fragile to attacks 42

43 Last concept: Community How do you extract useful information in huge N? The tool developed was to break it into locally dense neighborhoods (or communities). Members of a community have similar patterns Many implications: marketing, portfolio decisions, etc. A community is a dense sub-graph: 43

44 To summarize this aside: Despite the apparent differences between them, networks have common organizing principles: Small world: short paths exists between any 2 nodes Scale-free: many nodes held together and a few hubs Preferential attachment: new nodes prefer to link to highly connected nodes Robustness: resilience against random error & attacks Communities: group members have similar patterns 44

45 WHEN ARE NETWORKS USEFUL? When you are modeling heterogeneous agents When the system structure is important Dynamics Robustness Adaptability Transmission of info, technology, disease, etc. 45

46 MAIN POINT FOR ECONOMISTS: Can we develop insights by analyzing networks? Economics is about allocation of resources Much of the allocation takes place in networked situations Micro: games among few players (oligopoly, bargaining, auctions, etc.) Markets: interactions among many anonymous players (trade, finance, etc.) Many have done that already. 46

47 IN LABOR ECONOMICS Most workers find their jobs through a social contact (Myers and Shultz, 1951) The Strength of Weak Ties.-Most people find jobs through acquaintances rather than friends (Granovetter, 1973) Which network property gives us this result? 47

48 IN FINANCE Propagation of shocks (Allen and Gale, 2000) Systemic risk Banking regulation 48

49 IN URBAN ECONOMICS What s the difference between this. Ant colony (Africa) 49

50 and this? Hong Kong (China) 50

51 Ant colonies can t grow beyond their local environmental capacity: Hou et al., PNAS

52 US cities neither. 52

53 CONCLUSION The economy is a complex system Relatively simple individuals Interact with each other forming complex structures ABM and Network science are tools to understand complex systems Alternative to the traditional models of economics with strong assumptions 53

54 TO LEARN MORE: Santa Fe Institute NECSI In our department: Prof. Joshi and Prof. Chen Come to the seminars! 54

55 PAPERS PRESENTED AT GWU SEMINARS: Diebold and Yilmaz: Network topology of variance decompositions: measuring the connectedness of financial firms (2011) Alessandra Fogli and Laura Veldkamp (NYU): Germs, Social Networks and Growth (2012) Mardie Dungey et al (2012): Ranking Systematically Important Financial Institutions Chris Carroll: The Epidemiology of Macroeconomic Expectations Maggie Chen: Intra-Network Trade, Reallocations, and Productivity Growth: Micro-level Evidence from China (2012) Gisela Rua (re-scheduled): Fixed Costs, Network Effects and the International Diffusion of Containerization (2012) 55

56 Many thanks. Any question? 56