Agent-based Modelling of Systemic Risk: A Big-data Approach

•  Model individual agents and their individual decisions - decentralized decision making

•  Emergent patterns from micro-processes to macro level

•  Account for local interaction networks between agents - parallel computing can be used

•  Based on micro-foundations - big-data can be included

•  Very large models that incorporate low level details possible - need for supercomputing

Sebastian Poledna1,3 Michael Gregor Miess1,2, Stefan Schmelzer1,2, Elena Rovenskaya1,6, Stefan Hochrainer-Stigler1, and Stefan Thurner1,3,4,5 1International Institute for Applied Systems Analysis, Schlossplatz 1, 2361 Laxenburg, Austria 2Institute for Advanced Studies, Josefstädter Straße 39, 1080 Vienna, Austria 3Section for Science of Complex Systems, Medical University of Vienna, Spitalgasse 23, 1090 Vienna, Austria 4Santa Fe Institute, 1399 Hyde Park Road, Santa Fe, NM 87501, USA 5Complexity Science Hub Vienna, Josefstädter Straße 39, 1080 Vienna, Austria and 6Faculty of Computational Mathematics and Cybernetics,Lomonosov Moscow State University (MSU), Moscow, Russia

•  The quality and quantity of economic data is expanding rapidly

•  Shift from small-sample surveys to datasets with universal or near-universal population coverage

•  Enables empirical calibration of agent-based models

Agent-based Models (ABMs)

Big-data

Local Interactions and Networks

Parallel Computing and Supercomputing

IIASA’s ABM of a national economy •  Models the complete national economy of Austria with all institutional sectors

(households, non-financial corporations, financial corporations, and a general government)

•  Includes all economic activities (producing and distributive transactions) as classified by the European system of accounts (ESA)

•  Includes all economic entities: all juridical and natural persons are represented by agents (at a scale of 1:10)

•  Empirically calibrated to actual macro and micro data (national accounting, census and firm-level data)

•  Simultaneously fits observed macroeconomic variables, stylized facts, and (some) observed distributions between agents on the micro-level

Application: Systemic Risk Triggered by Natural Disasters

1 2 3 4 5 6 7 8 9 10year

-5

-4

-3

-2

-1

0

1

cum

ulativ

e ch

ange

in G

DP g

rowt

h ra

te [p

p]

1 2 3 4 5 6 7 8 9 10year

0

2

4

6

8

10

12

14

16

18

20

chan

ge in

gov

ernm

ent d

ebt-t

o-G

DP ra

tio [p

p]

•  An economy is a network of interacting firms, households, banks, etc.

•  Out-of-equilibrium dynamics of economic networks can be explicitly modeled using ABMs

Banking network of Austria in 2008 – based on data provided by the Austrian Central Bank (OeNB)

Ownership network of 170.000 Austrian firms in 2013 – based on data from the company register of Austria

Firm size distribution by output for model simulations

Comparison of output (gross value added) and inflation for model simulations and observed data of Austria

Macroeconomic effect on GDP and debt of flooding affecting only productive capital of a 1500 year event in Austria – based on the ABM simulations

•  Economic ABMs are intrinsically massively parallel computational systems

•  Very large populations of agents perform complicated local computations (decentralized decision making)

•  Agents have precise positions on physical and conceptual networks (local interactions)

•  Natural and necessary to run large-scale ABMs on supercomputers

A01A02

A03 B

C10-C12

C13-C15C16

C17C18

C19C20

C21C22

C23C24

C25C26

C27C28

C29C30

C31_C32C33 DE36

E37-E39 FG45

G46G47

H49H50

H51H52

H53 IJ58

J59_J60J61

J62_J63K64

K65K66 L

M69_M70M71

M72M73

M74_M75N77

N78N79

N80-N82 O PQ86

Q87_Q88

R90-R92R93

S94S95

S96

Sector

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

Outpu

t [Euro

]

×1010

intermediate consumptionwagessocial contributionstaxes less subsidiescapital consumptionoperating surplus

Distribution of output and cost structure by sector used as initial values for model simulations from observed data of Austria

106 107 108 109 1010

Firm size (output)

10-5

10-4

10-3

10-2

10-1

100

Prob

abilit

y2.6

2.8

3

3.2 ×105 Gross value added (GVA), million euro

modeldata

2010 2011 2012 2013 2014 2015100

105

110 Inflation (implicit GVA deflator), 2010=100

Systemic Risk and Network Dynamics An EEP-ASA-RISK crosscutting project