Physics of Risk
This project contributes to our research line:
Duration: 27 months (September 2006 - November 2008)
Funding source: Swiss State Secretariat for Education and Research SER ( C05.0148)
Project partners: ETH Zürich, Chair of Systems Design (Prof. Frank Schweitzer), ETH Zürich, Institute for Operations Research (Prof. Hans-Jacob Lüthi), Zürich University of Applied Sciences Winterthur,
This project is related to the EU COST Action P10: "Physics of Risk". It is structured in 5 smaller projects carried out by 5 different institutions across Switzerland.
Project 1: Network structure, robustness and adaptivity of organizations
The network structure of a social or economic organization directly relates to system properties such as robustness against perturbations, or adaptivity in a changing environment. For organizations, the risk to fail thus can be seen as coming from these different sources, namely lack of robustness and lack of adaptivity. The project aims at quantifying the relations between network structure, robustness and adaptivity of organizations by means of both theoretical and empirical investigations.
Project 2: Failure risk propagation in economic and supply networks
Failure risk propagation is a phenomenon with great impact in today's society. This project aims at understanding how failures and bankrupcies of firms affecting other connected firms, might impact global dynamics, i.e. leading to cascading of failures or not. We use the well studied paradigm of multiplicative processes to model firm growth, but in contrast to previous works, we couple the budget growth of firms, creating a connected network of firms. Indeed, it is widely known that network topology has a huge impact in the process that occurs on it, and we intend to study this influence in the global dynamics of the system. We will investigate how different mechanisms of failure cost absorption by neighboring firms influences the global dynamics taking into account the topology. The results will provide valuable insight into how policies and individual hedging strategies affect the vulnerability of the whole system, in other words, the systemic risk, an issue poorly addressed by current theories.
Project 3: Robust fitting technique in financial and non-financial time series
Generalized hyperbolic (GH) distributions will be used in a univariate and multivariate context to fit financial and non-financial time series. This class of distributions is very flexible and allows one to statistically describe extreme events. To apply this method to short series, a robust fitting technique will be developed. Different type of time series will be fitted:
(i) High-frequency univariate returns of the so-called GOP (growth optimal portfolio) which has been constructed in a previous work.
(ii) Daily multivariate returns to the GOP expressed in different numeraires.
(iii) Performance measure Sharpe-Omega for monthly hedge fund returns. Since these time series are pretty short, this requires robust techniques.
(iv) Returns distributions generated by multi-agent models of financial markets.
(v) Suitable events generated by multi-agent models outside financial markets. This will give some means to characterize distributions of extreme events. The work is important in view of risk quantification, measurement and management of extreme events.
Project 4: Collaborative information filtering for the internet
Our project aims to devise agent models to describe socio economic interactions taking place on networks and to reduce their inherent risk. This means for us to try and understand how efficiently resources are used and how inefficiencies can be avoided. The waste of resources, from a physicists' perspective, can be interpreted as entropy production, as the presence of large fluctuations around an equilibrium point, or as the frustration of a spin system. As a simple to specify, yet generally applicable model of interacting agent preferences and competition, statistical physics tools have proven most relevant to studying frustrated dynamics of large numbers of competing agents. Steady progress has been made in modeling financial markets with multi agent systems, Minority Games being a prominent example. Here information scarcity and information asymmetries are the main sources of risk. The physicists' community has also been successful in modeling information networks, like the Internet. Here the problem to tackle is that of information overload, where not all the information sources are trustworthy. In this respect we want to devise efficient information filters, which can be regarded as improved search engines. They should be able to rank information sources in order of increasing reliability. For some information, though, sources reliability is not the only requirement taste plays an important role. In this case personalized recommendations highly improve the service provided. Their mathematical modeling and optimization will be the ultimate aim of our research, thus reducing the local source of risk. We stress that the theories we are developing are based on agent models, the dynamics of which is intended to be performed on scale-free network topologies.
Project 5: Impact of severe environment changes on population dynamics
Our main goal is to investigate the impact of severe environmental changes on the micro and macro-evolution in the framework of population dynamics models. A difficult question in modeling such systems is the choice of the level of description. One is looking for the simplest level of description which reproduces the correct qualitative and quantitative properties of the system. The different levels of description are typically mean-field like, reaction-diffusion like, or individual based model (IBM). There is no general understanding about how to decide a priori what is the good level of description. Accordingly, it would be very important to have a better understanding of this generic question and to formulate some general criteria. This is the first aspect of the project.
The second aspect of our project concerns the study of several agents based models describing both micro and/or macro-evolution of populations. We studied simple models, exhibiting nontrivial generic behaviors. We intend to study similar models in cases where the changes in the environment are due to several ecological factors which have not yet been considered as for example pollution, natural catastrophes, migrations. These changes may be time dependent and thecompetition between the different timescales may lead to new cooperative phenomena. Moreover this kind of problems have been, until now, mostly studied on regular lattices. An interesting question concerns the role played by the topology of the underlying network. Accordingly, we intend to investigate the role the network topology in several of the problems described above.
EPJ-B Topical Issue on Risk, Markets, Games and Networks
The complete contents of the special issue can be accessed here.