www.ccss.ethz.ch

Phone: +41 44 632 8880 ● ●e-mail: dhelbing@ethz.ch www.soms.ethz.ch

ETH Competence CenterCoping with Crises in Complex Socio-Economic Systems (CCSS)

Many observed macroscopic phenomena in humansociety, such as residential and ethnic segregation,integration or outbreaks of violence are attributed tothe actions taken at the level of individuals or groups.In order to better understand the link between themicro-scopic individual actions and the emergentmacro-scopic outcomes, we have developed theframework of migration games to describe thestrategic interactions among individuals moving inspace. In migration games, one can do both, move tonew positions that promise higher benefits, and/oradapt the interaction strategy.Our model reproducesvarious observed social phenomena resulting fromindividual interactions, for example, different de-grees of integration (Figs. A and B) or spatial segre-gation patterns (Figs. C and D), which determine thelevel of cooperation or conflict between individuals. Urbanization and economic development play a pro-

minent role for transportation infrastructures. Roadnetworks experiencing high traffic volumes andheavy congestion waste time and fuel, and pollutethe environment. By applying mathematical modelsthat capture behavioural and dynamical features oftraffic flows, we study control-strategies that lead tosmaller average travel times. Our control approachesfocus on the flexible selforganization and coor-dination between neighboring trafficlights to avoidcongestion spreading.By means of reliable computersimulations we test various mechanisms to copewith traffic problems more successfully.The figure tothe left shows a small section of our simulation ofBerlin s traffic. Pink parts of road segments illustratequeued traffic that causes delays for drivers. Ourcurrent research studies the crucial determinants ofcongestion spreading, considering cascading effectssimilar to what is known from disaster spreading.

'

Prof. Dr. Dirk HelbingChair of Sociology,in particular of Modeling and Simulation

Spatial Social Interactions

Traffic Management

A B

C D

www.ccss.ethz.ch

Phone: +41 44 632 83 50 e-mail: fschweitzer@ethz.ch www.sg.ethz.ch● ●

ETH Competence CenterCoping with Crises in Complex Socio-Economic Systems (CCSS)

The recent and still on-going financial crisisconfronts us with fundamental questions: Althoughfinancial engineering is supposed to improve theability of financial intermediaries to spread andmanage risk, financial crises appear to be more fre-quent, stronger and more global. Is this a necessaryprice to pay in order for economies to be sustainable?Or, if not, is there anything we can do about it?Endogenous financial crises like the current one (i.e.without a clearly identifiable shock from the outside)are not well accounted for by standard neo-classiceconomics. Our aim is to contribute to the debate onsuch questions from a complex systems perspective.On the one hand, the globalization of capital marketsplays an important role. On the other hand, it is clearthat, if all global investors would invest in the sametype of overvalued asset (i.e. mortgage-backed se-curities and alike), then a bubble would hit all in-vestors at once, with global consequences. However,this view does not explain the intensity of the currentcrisis. We include in our analysis two importantaspects that have been overlooked so far. The first isthat the financial robustness of an agent (as mea-sured for instance by its equity ratio) is affected by

the financial robustness of its neighbors and, thus,depends on the position of the agent in the networkof contracts.As the second aspect we emphasize thatthe dynamics of financial robustness is intrinsicallysubject to positive feedbacks (financial distress to anagent is more likely to lead to additional financialdistress in the future and vice versa). We havemodeled the economy as network of inter-dependent units connected by financial contracts(left figure), and we have shown analytically thatthese two aspects already lead to an unexpectedresult: A more connected financial system is notnecessarily more robust, as most of the economicliterature predicts. On the contrary we have foundthat the individual probability to fail, as well as theprobability of large cascades of failures, are both U-shaped as a function of the risk diversification(bottom figure). Thus, our model offers a simpleframework to investigate the relation between thestructure of financial networks and systemic risk.Our next goal is two-fold: On the one hand we willmodel more explicitly the contracts and the choicesof the agents. On the other hand, we will test and

Prof. Dr. Dr. Frank SchweitzerChair of Systems Design

validate model with empirical data.

www.ccss.ethz.ch

Phone: +41 44 632 89 17 ● ●e-mail: dsornette@ethz.ch www.er.ethz.ch

ETH Competence CenterCoping with Crises in Complex Socio-Economic Systems (CCSS)

1/2

All collective social phenomena arise from theaggregate behavior of many individuals. Uncoveringrules governing collective human behavior is a diffi-cult task, because of the myriad of factors that influ-ence an individual's decision. Investigations into thetiming of individual activity, as a basis for under-standing more complex collective behavior, havefound statistical evidence that human actions rangefrom random to highly correlated ones. While mostof the time, the aggregated dynamics of ourindividual activities create seasonal trends or simplepatterns, sometimes our collective action results inblock-busters, best-sellers, and other large-scaletrends in financial and cultural markets. Usingepidmic-type models and a massive database of theviewing activity regarding millions of videos onYouTube, we have shown that it is possible to under-stand collective behavior and capture how bursts ofactivity evolve over time as a result of the influence ofmedia and word-of-mouth spreading. The predictedtypes of behavior, which we have identified, aresummarized in the upper figure.

Zipf's law describes a remarkably universal rank sizedistribution in natural and socio-economic systems.For almost a century, various efforts have beenunderway to discover the mechanism(s) of Zipf's law,the most famous one being "proportional growth",also known under the name "preferential attach-ment". In spite of lacking empirical verification, themodel has been widely used to explain stochasticgrowth in complex systems, such as the dynamics ofcity or firm sizes, or the power-law connectivitydistribution of nodes in a network. For the first time,the proportional growth mechanism underlyingZipf's law has now been verified in detail along threeempirical properties in a real-world complex system,namely in the dynamical evolution of opensourcesoftware distribution The result is highly relevant forresearchers in a variety of fields,ranging from physicsand biology to economics,sociology and the Internet.This work emphasizes the role of interactionsbetween human developers/users and computers,with important applications to the understanding ofupside and downside risks, such as in virtual worldeconomies and cyber-risks.

Prof. Dr. Didier SornetteChair of Entrepreneurial Risks

Revealing Robust Dynamic Classes byMeasuring the Response Function of aSocial System (R. Crane)

Empirical Verification of ProportionalGrowth of Packages in Debian Linux(T. Maillart)

[R. Crane and D. Sornette, PNAS 105, 15649 (2008)]

[T. Maillart et al., PRL 101, 218701 (2008)]

www.ccss.ethz.ch

Phone: +41 44 632 89 17 ● ●e-mail: dsornette@ethz.ch www.er.ethz.ch

ETH Competence CenterCoping with Crises in Complex Socio-Economic Systems (CCSS)

2/2

Oil prices exhibited a record rise followed by aspectacular crash in 2008. The peak of $145.29 perbarrel was set on July 3, 2008, and a recent low of$40.81 was scraped on December 5, 2008, a level notseen since 2004. On May 27, 2008, we addressed thequestion of whether oil prices were exhibiting abubble-like dynamics, which may be symptomatic ofspeculative behavior, using our techniques based onmethods from statistical physics and complexitytheory. Our methodology aims at detecting transientphases, where positive feedbacks operating in somemarkets or asset classes create local unsustainableprice run-ups. The mathematical signature of thesebubbles is a log-periodic power law (LPPL). A morethorough post-crash analysis of our May 27, 2008experiment indicates a 'predicted' crash date of June11, 2008, i.e. just 21 days before the actual crash. Weare using the same techniques to predict bubbles

While trading on the stock market, taking a decisionof whether to sell, to buy or just to hold your actualposition may be tough. Being subject to many di-fferent forms of information, it is hard for the traderto make up his mind on what source of informationhe should trust. We simulate this situation in anartificial stock market with traders, who base theirtrading decision on three different sources of infor-mation: (i) the news (generated by white noise), (ii)private information and (iii) the decisions of collea-gues and in the surrounding. In our model the tradersadapt their level of trust in each of these informationsources to their past predictive power. In this way, weshow that rallies and crashes can occur as ampli-fications of random lucky or unlucky news, while noextraordinary events are needed. It is the trader'sattempt to dynamically optimize their strategy,which pushes the whole market into a state of eu-

Prof. Dr. Didier SornetteChair of Entrepreneurial Risks

and crashes in other financial time series.[D. Sornette et al. arXiv.0806.11 70]

phoria or fear, resulting in a rally or crash.[G. Harras and D. Sornette, arXiv.0806.2989]

Prediction of an Oil Price Bubble(R. Woodard)

Endogenous versus exogenous originsof financial bubbles and crashes(G. Harras)

www.ccss.ethz.ch

Phone: +41 44 633 39 43 ● ●e-mail: axhausen@ivt.baug.ethz.ch www.ivt.ethz.ch

ETH Competence CenterCoping with Crises in Complex Socio-Economic Systems (CCSS)

In urban environments, the functioning of the trans-port infrastructure is not only dependent on theengineering of the built components and thetechnical implementation of traffic control, but to alarge extent also on the behavior of the travellers.This is especially true, if we look at unforeseen eventswith severe outages of the transport infrastructure.How will people react in such situations? How cantheir decisions be positively influenced by usingavailable control mechanisms? And if training can beapplied as a precaution to such situations, howshould people be trained? We investigate suchquestions by developing and applying a multi-agentintegrated transport model for very large scenarios(lower figure: simulated greater Zurich area). Thesimulation consists of a population of agents per- the world, are carefully constructed such that they

correspond to real scenarios and policies. Each agentin this microscopic simulation moves around andperforms activities according to a plan, which he orshe has optimized by maximizing a given utilityfunction. The resulting travel demand can be beunderstood as a side effect of the agents' desire toperform activities at different places. This has theadvantage that the agents do not produce trafficmerely to match data from real life – as is often donein standard traffic models – but rather behave mean-ingfully while trying to perform as good as possiblein a holistic way. Our current research goes into twodirections: 1. understanding and predicting the im-mediate reaction of a consistently loaded trafficsystem, when exposed to severe failure, and 2. thepotential of training of the traffic participants, andproviding them with up-to-date information toalleviate the severity of such situations.

Prof. Dr. Kay AxhausenChair of Transport Planning and Systems

Agent Behavior During Transport

forming trips in a virtual world.Both,population and

www.ccss.ethz.ch

Phone: +41 44 633 27 01 ● ●hans@ifb.baug.ethz.ch www.ifb.ethz.ch/comphys

ETH Competence CenterCoping with Crises in Complex Socio-Economic Systems (CCSS)

Many complex systems such as the Internet, electriccircuits, and biological cells can retain their functionsdespite strong perturbations. The robustness ofthese systems is attributed to underlying hetero-geneous network structures. These networks haveoften power-law degree distributions P(k) k , i.e.,they are scale-free. Although scale-free networks areremarkably resistant to random failure, a targeted,malicious attack can disrupt then, when only a smallfraction of links fails. The network disruptionrepresents crisis in the networked system. Optimiz-ing the network's robustness against maliciousattacks is one of the ways of avoiding such a crisesand thus of coping with it. By studying properties ofoptimized networks, we can learn how to buildattack-robust networks, and how to increase therobustness of existing networks. We develop amethod that substantially improves the robustnessof complex networks to malicious attacks.We modelthe malicious attack as a dynamic, degree-basedattack on nodes.During the attack, the node with thehighest degree is removed from the network, and thedegree distribution is recalculated before a newnode, which now has the highest degree, is removed.

α

The way the network's structure is changing duringthe attack is used to measure its robustness. In theprocess of numerical optimization of the network'srobustness, we change the network topology, keep-ing the degree distribution invariant. We use ourmethod on a real network, the Internet at the auto-nomous system level, and on the three differenttypes of model scale-free networks, the Barabasi-Albert network, the Apollonian network, and thestatic scale-free network with random connections.During the process of optimization, the topology ofthe network is changing, which is illustrated for theexample of the Apollonian network in the Figuresabove (the size and the color of a node in the networkindicate its degree, i.e., the number of its links). Westudy the topological properties of the optimizednetworks, such as the clustering coefficients or thecorrelations of the degrees of connected nodes, com-pare these properties to those of the initial networks,and look for similarities between these properties fordifferent network types.We also study the propertiesof the optimization process itself,such as the fractionof links changed during the process.

Prof. Dr. Hans J. HerrmannChair of Building Materials

Complex Network's Robustness toMalicious Attack

-γ

www.ccss.ethz.ch

Phone: +41 44 632 67 59 ● ●e-mail: lcederman@ethz.ch www.icr.ethz.ch

ETH Competence CenterCoping with Crises in Complex Socio-Economic Systems (CCSS)

Many of conflicts in today's world occur withinstates. However, quantitative conflict research tendsto rely on highly aggregated country-level data,thereby blurring the causal mechanisms on theground. New tools are increasingly becomingavailable to explore conflict patterns in greaterdetail. Geographic information systems (GIS), inparticular,provide a flexible way to combine spatiallyorganized data about conflicts and conflict parties.The Geo-Referencing Ethnic Groups (GREG) datasetoffers information on ethnic groups and theirsettlement patterns around the world. If combinedwith GIS-measures on terrain and distances, thisdata resource tells us that ethno-nationalist civilwars are likely to involve large and remote groupsthat are excluded from state power. Created throughan online expert survey in collaboration with UCLA,the enhanced dataset on Ethnic Power Relationsdocuments even more accurately the relationshipbetween ethnicity and conflict,and is currently beinggeo-coded.

Agent-based models of conflict enable researchers tograsp complex spatio-temporal processes in a con-venient and flexible way. Some of these models aremostly theoretical and serve as virtual laboratorieswithin which counterfactual thought experimentscan be conducted in order to explore interactionsamong causal mechanisms. Inspired by non-equilibrium physics, other models serve to recon-struct macro-level data patterns, such as power-lawdistributed war sizes measuring casualties. Futurecomputational models will increasingly draw ondisaggregated data of the type described above formicro-level calibration and validation. Integratedcomputational platforms are able to let artificialagents interact within realistic,GIS-based spaces as away to explore the conflict processes in a realisticsetting. Ultimately, such analysis will lead to moresystematic and accurate risk assessment of warfarearound the world.

Prof. Dr. Lars-Erik CedermanChair of International Conflict Research

Disaggregated Measures of Conflict

Computational Models of Conflict