Partner Links     
 
  European Organisation for Nuclear
Research
  University of Geneva   Swiss Association
of Market Technicians
 
 

Hewlett Packard

Banque Cantonale
de Genève




Dell Computers


Institute for Nuclear Research


Dukascopy Trading Technologies

 
 
 

GRC Welcome

The Geneva Research Collaboration (GRC) is a non-profit organization dedicated to the promotion and development of innovative applications of interdisciplinary research and education in natural and social sciences. The current research focus of GRC is concentrated on mathematical and physical models applied to financial risk management and forecasting. The applications rely on new conceptual model structures which are developed into operational tools through the use of advanced computer techniques.

 

You are kindly invited to the next Geneva Research Collaboration seminar. The seminar will take place on Monday, 21 March at 17-00 at CERN, conference room  40-S2-B01.
Presentation by Dr. Antti Korhonen:

STRATEGIC FINANCIAL MANAGEMENT IN BANKING AND INSURANCE: A MULTIPLE GOAL STOCHASTIC PROGRAMMING APPROACH

Abstact: The paper discusses a multi-stage stochastic programming approach to the strategic financial management of banks, life insurance companies, pension funds and multinational financial conglomerates. The model is an extension of an earlier model designed to deal with the strategic asset and liability management problem of a multi-company group with main emphasis on non-insurance activities. Many of the model components are common to all the institutions and the paper therefore first gives an outline of these common components and the general model structure. The special features of life insurance are then discussed in more detail. The paper ends with some general notes on the use of scenario-based dynamic optimization in strategic planning.

You are kindly invited to the next Geneva Research Collaboration seminar. The seminar will take place on Wednesday, 16 February at 17-00 at CERN, AB Auditorium I inBuilding 6-2-024.
Presentation by Laurent Barras:

Asset allocation and data-snooping.

Abstact: This paper examines the impact of real-time uncertainty on the performance of mean-variance conditional asset allocation. Real-time uncertainty is modeled by implementing a large number of conditional strategies that an investor could reasonably choose from. Using data from twelve 12 domestic equity markets between January 1990 and September 2004, we find that real-time uncertainty greatly reduces the performance of conditional asset allocation and makes it highly sensitive to the specification choices. Moreover, it turns out that the different solutions used to mitigate the impact of real-time uncertainty are not effective and call for more efficient implementation of these strategies. All these findings are consistent across different levels of investor's risk aversion and transaction costs.

You are kindly invited to the next Geneva Research Collaboration seminar.
The seminar will take place on
Wednesday, 26 May at 17-15 at CERN, Room 40-S2-B01.
Presentation by Olivier Scaillet:
"A Kolmogorov - Smirnov type test for positive quadrant dependence"

Abstract:

We consider a consistent test that is similar to a Kolmogorov-Smirnov test, of the complete set of restrictions that relate to the copula representation of positive quadrant dependence. For such a test we propose and justify inference relying on a simulation based multiplier method and a bootstrap method. A Monte Carlo experiment is used to explore the finite sample behavior of both methods. A first empirical illustration is given for US insurance claim data. A second one examines the presence of positive quadrant dependence in life expectancies at birth of males and females among countries.

You are kindly invited to the next Geneva Research Collaboration seminar.
The seminar will take place on
Wednesday, 19 May at 16-00 at CERN, Room 40-S2-B01.

Presentation by Nikita Stepanov (ITEP/CERN/GRC):
"Predicting randomness" experiment: new perspectives and results from the test run.

“Predicting randomness” experiment was proposed about one year ago with the purpose of investigating the randomness of bit sequences generated from different “random” sources so as to elucidate if such sequences may be locally predictable in statistically significant sense by the human being or “intelligent” computer program.  The experiment was designed to be a large scale internet initiative which could allow thousands of participants around the world to be involved. During the past year we developed the hardware and software, a set of statistical tests, which allows one to estimate the randomness of the experiment data sources and the quality of predictions of the experiment participants. All this allowed us to launch the first “technical” run of “Predicting randomness” experiment a few months ago. Although this event was not announced officially, several human beings and artificial predictors took part in the test run. The preliminary analysis of the data accumulated has been finished recently. In this talk we would like to outline the present status of the experiment hardware and software and present some interesting observations derived from the test run data analysis; in particular, the intriguing similarity of prediction results of the human and artificial predictors. The performance tests fulfilled during this run clearly indicate that now we are ready to conduct the experiment in full scale, at the same time, we have realized that the primordial goals of the experiment and perhaps, even the whole experiment scheme have to be adjusted. We think it can be made more attractive to participants and more practically feasible, and here we would like to discuss new proposals for our initiative.

You can have a look at presentation materials. (MS PowerPoint ~600kb)

You are kindly invited to the next Geneva Research Collaboration seminar.

The seminar will take place on Wednesday,  5 May  at 15.30 in GRC office at the following address: Rue de Veyrot 12, CH-1217 Meyrin-Geneva (click here for GRC location map).

 

Presentation by Couderc Fabien on "Time-to-Default : Life Cycles, Global and Industry Cycle Impacts"

 

Abstract:

 

 Investigating once again time-to-default of individual firms, from our semi-parametric framework this study proposes a comprehensive analysis of the determinants of default. We assess whereas rating agencies succeed to achieve their goal or not, and further shed light on the need for conditioning with respect to economic conditions even for through-the-cycle ratings. We exhibit various effects bearing strong implications in credit risk management depending on investment horizon as well as rebalancing frequency.

In addition extensive studies of log-linear parametric models of instantaneous probabilities reveal that both firms life cycles within a given risk class and the global business cycle matter. Concerning the debate between structural and reduced form models, we overall point out how lag information can substantially improve performance of simple models. Indeed, we show that some factors convey anticipations of future bad default state, even if differences can be observed among studied grades and industries, as well as the magnitude of their impacts. These underline importance of ratings in financial decisions.
However this semi-parametric setting indicates that the business cycle along with the credit cycle are far from being sufficient to extract all joint movements in default probabilities. Investigating industry default processes, we illustrate strong discrepancies between industries which may partially fill this gap. In particular our results suggest that the default and business cycles are not always procyclical which can not be caught through economic factors. Further the default cycle both leads and persists after global cycles which may only be captured by endogenous modelling of aggregate components of the default cycle. Therefore, this "snow ball" effect appears to be an important ingredient to include in a good predictive model. 

You are kindly invited to the next Geneva Research Collaboration seminar.

The seminar will take place on Thursday, 19 February at 15.30 in GRC office at the following address: Rue de Veyrot 12, CH-1217 Meyrin-Geneva (click here for GRC location map).

Presentation by Couderc Fabien on : A Default Duration Model in a Multivariate Setting Consistent with Stylized Facts

 

 

Abstract:

This paper studies time-to-default of individual firms in a multivariate setting. We concentrate our efforts on analysis and modelling of default intensities. Hence, this research can be classified into the so-called reduced form class but our framework exploits information contained in structural factors (macro-economic, business and firm specific).
Using a Standard & Poor's database, non-parametric and semi-parametric studies of intensities show that both firms' life cycles and economic conditions matter. In particular, the largest movements in intensities are due to economic cycles whereas short and mid term trends are driven by aging effects. Moreover significant factors varies with the horizon, the studied grades and the industry, as well as the magnitude of their impacts. Analysing aggregated default processes, we observe that they can be used as an indicator of local business conditions, which exhibits markovian properties.
As most of previous works consider proportional hazard specifications with constant baseline, we propose to extend these models and further use random scale changes (i.e. speed varying hazard models) showing especially the lack of power of non trivial proportional models at short horizons. Further, our specification allows to achieve realistic correlations of default through auto-correction from the aggregated default process. Finally, duration predictions can be run out through simulations thanks to the multivariate hazard construction method.

You are kindly invited to the next Geneva Research Collaboration seminar.
The seminar will take place on
Wednesday, 12 November at 16-30 at CERN, Room 40-S2-B01.

Presentation by Prof. Olivier Scalliet
on:

Risk Management Using Econometrics


Abstract:

The objective is to continue the development of econometric tools for an improved assessment and monitoring of financial and insurance risk. The proposed econometric tools are both of parametric and non-parametric nature and share the aim of improving the modeling of the distribution of risk and of the dependencies that can occur between different sources of risk. The non-parametric approach and the techniques to be investigated are based on nonparametric estimators based on asymmetric kernels which avoid the bias that occurs when a random variable is bounded. The asymmetric nature of these estimators allows the modeling of distributions that have fat-tails or of distributions that are bounded at some point. It is envisaged to extend significantly the empirical investigations and to improve certain technical aspects of these estimators. The econometric tools under current development may be used in many areas of finance. For example, they can be applied to various types of data such as interest rates, exchange rates, stock returns. As such they should allow a better understanding how to control financial losses for banks and insurance companies. The efforts will therefore be oriented towards financial econometrics modeling with relevant data coming from economics, finance and insurance fields. The research is oriented theoretically as well as empirically. Methods tailored to risk management issues are to be developed theoretically, to be compared numerically with alternative methods using Monte Carlo experiments, and to be applied to empirical data to get a better understanding of risk behavior. We plan to extend the current results to direct modeling of Value at Risk and expected shortfalls as well as to the multivariate case.

You can have a look at the full text of the research paper as well as at presentation materials.



You are kindly invited to the next Geneva Research Collaboration seminar.

The seminar will take place on Tuesday, 30 September at 17.00 in GRC office at the following address: Rue de Veyrot 12, CH-1217 Meyrin-Geneva (click here for GRC location map).

Presentation by Couderc Fabien on Credit Risk Dynamics:
Forecasting Rating Transitions Matrices


Abstract:

Nowadays, as one of the most productive area of modern finance, credit risk research has still an explanatory prospect. Nevertheless, new regulatory requirements and the growing investments in credit derivatives induce needs for predictive measurements.

Therefore, this project consists in the development of a coherent methodology to provide forecasts of the well-known rating transition matrices at various horizons. In addition, our forecasts aim to be used as inputs of dynamic credit risk models so as to bring consistency in the current way to proceed. Indeed, numerous recent dynamic models of credit use static and time-homogenous transition matrices as inputs so as to draw dynamic outputs. Our methodology will supply for time-heterogeneous matrices directly and exclusively from rating event data, avoiding problems of event scarcities and allowing smart aggregations.

Instead of looking only straight transitions to default our dynamic model of migrations take care of both probabilities’ evaluation and evolution. This study will be empirically supported by rating data from Standard & Poor’s CreditPro database The model will also allow us to investigate several issues such as the importance of timing versus aging effects, namely to determine if macro-economic conditions are more informative than firms’ life cycles. Moreover, for prediction purposes, conditional and unconditional moments of the whole transition matrices will be available so as to predict usual inputs of traditional portfolio models by simulations.

Click here to access full text of research paper

August 19, 2003

Dear friends,

GRC is proud to announce the start of its research grant programme for 2003-2004. Details of the programme can be found here.

You are kindly invited to the next Geneva Research Collaboration seminar

Tuesday, 17 June, 16.30-17.30 hours
PS conference room 2-024 building 6, CERN, Meyrin
Note: Participants from outside CERN, please confirm participation with name and affiliation by e-mail, at the latest one day in advance.

Dr. A. Din, Geneva Research Collaboration

"Trend analysis of financial time series"

Abstract:

A large part of the many claims for succesfully predicting the directional movement of financial markets is based on some kind of scheme for "riding the right trend". A problem with such claims is the lack of general agreement about the definition of what constitutes a "trend" and how good it is. The talk discusses financial time series in terms of a discrete change representation which allows for a rather straightforward definition of a trend process and a range process associated with the time series.
.
Trend analysis is first carried out for random walks so as to determine a certain reference probability distribution for the range process. It is shown how it may be possible to identify a statistically significant signal for trendy behaviour of a financial time series through the appearance of enhancements of the lower tail of the cumulative range probability distribution function. This distribution function leads in a natural way to the definition of a Trend Index a Range Index.

As an example, the range properties of daily prices for a few currency crosses, world indices and blue chip stocks are investigated. Results are presented for these time series which show distinct trend characteristics of importance for better understanding the dynamics of financial forecasting.

Full text of research paper

Background information:

The Geneva Research Collaboration (GRC) is a non-profit Swiss foundation dedicated to the support of interdisciplinary research in natural and social sciences, to the development of novel economic applications of this research, and to contributing to make Geneva a node of excellence in an international scientific and economic research network.
The current research of GRC is concentrated on mathematical and econophysics models applied to financial forecasting, data processing, and risk management. The applications endeavour to use new conceptual model structures which are developed into operational tools through the integration of advanced computing techniques


For further information please consult the GRC web site: www.genevaresearch.org

We mourn the death of our friend, the Nobel prize winner Ilya Prigogine

During CERN lecture...

with his wife Marina...

with us...


Article about Ilya Prigogine

Dear friends,

We are proudly presenting an Internet page dedicated to the Deep-Trader Experiment, a new initiative of Geneva Research Collaboration. To visit the page, you can either click on this link or go directly to http://grc.dukascopy.org. All your comments and suggestions are welcome.

Dear Colleague,

You are kindly invited to the next meeting of the GRC time series forecasting research group.

The meeting will take place on Wednesday, 21 May at 16.30 hours

at CERN, Building 40 Room R-C 10

Presentation by Allan Din on
:

Forecasting of interest rates using neural network models

Abstract:

During the past couple of years, interest rates around the world have declined to very low levels and there is a quest in the financial community for better understanding the dynamics of this situation. In the paper, the development of the monthly Libor and Swap rates are investigated over a 10 year period (1993-2002) in relation to a number of potentially important factors. For the case of Swiss interest rates, these factors include a range of rates for different maturities, gross domestic product, inflation rate and exchange rates.
Interest rate forecasting models are developed, which endeavour to include the statistically most significant factors defined in terms of time lag changes of individual time series as well as of relative movements of different maturity rates. The models are constructed on the basis of backpropagation neural networks involving one hidden layer of 5-10 nodes. Model validation is done for a period covering the past 5 years with the model retraining carried out every 3 months.
The results for forecasting the Libor 3 month rate and the Swap 5 year rate on a forecasting horizon of 12 months show quite a good reliability. The directional movement is forecasted correctly in 83% of the months of the validation period. The quality of the models for the Libor 3 month rate and the Swap 5 year rate on a forecasting horizon of 3 months is somewhat lower, with only 61-63% of accuracy in the directional forecasting.

Click here to access a corresponding research paper.

Dear Colleague,

You are kindly invited to the next meeting of the GRC time series forecasting research group.

The meeting will take place on Thursday, 20 February at 16.00 hours

at the CERN, Room 304-1-0001A

Presentation by Nikita Stepanov, Andre Duka, Nikolaj Krasnikov on
:

Predicting randomness - Outline of an experimental design

Paper abstract:

We propose conducting an experiment to investigate the randomness properties of bit sequences generated from different “random” sources, including computerized, physical and economic ones, so as to elucidate if such sequences may be locally predictable in statistically significant sense. A sign of local predictability may be said to be demonstrated, for example, by virtue of a capacity by a human being and/or a computer to recognize and exploit certain patterns or other repeatable behavior of specific digital data sources.

According to conventional wisdom and modern science, one would expect a negative result of such an experiment, that is, the absence of any statistically significant predictive capacity in relation to any truly “random” process. Nevertheless, we believe that it is of interest to conduct such an experiment for two good reasons: first, it is a fact that it has proven to be difficult to give a positive definition of a truly “random” process, and many commonplace bit sequences are in reality only pseudo-random. Then, there is a multitude of real life claims of predictable behavior of economic time series representing so-called efficient markets which supposedly should be random.

To read the entire paper click here.

Dear Colleague,

You are kindly invited to the next meeting of the GRC Time Series Forecasting research group.

The meeting will take place on Thursday, 23 January at 16.00 hours

at the GRC office, Rue de Veyrot 12, 1217 Meyrin

Agenda:

Presentation by Prof. Michel Maignan, Banque Cantonale de Genève on:

"Lowest level of interest rates since decades
Bearish behaviour of -33% per year
Highest exchange rate since 4 years:

New quantitative methods for financial analysis and forecasting?"

Prof. Michel Maignan Geostatistics/ Statistics Universite de Lausanne
Membre de la Direction
Retour à l'accueil



GRC Risk management research group meeting.

Wednesday, 11 December at 17.30 hours

at the GRC office, Rue de Veyrot 12, 1217 Meyrin

Presentation by Mikael Angberg on

"Implementing financial simulation models using Monte Carlo techniques and Java."

Abstract:

"In this talk I will briefly introduce the use of Monte Carlo methods in financial simulation models, as well as discuss some issues related to non-uniform pseudo-number generation, variance reduction techniques and correlation/co-movement. Some practical examples of simulation models for option pricing, mortgage-backed securities and Value at Risk implemented using Java and the Java Analysis Studio and COLT frameworks will be presented."

GRC High Performance Computing research group meeting

Wednesday, 4 December at 17.30 hours

GRC office
, Rue de Veyrot 12, 1217 Meyrin

Presentation by
Oxana Smirnova and Jukka Klem

"The NorduGrid"

The Grid is a technology to share and access seamlessly computing
resources that are not subject to a centralized control. The computing
resources are connected together through a layer of software called
the middleware, which uses standard, open, general purpose protocols
and interfaces. This middleware forms the glue binding the resources
into a virtual system.
NorduGrid project (
www.nordugrid.org) develops openly available middleware
and operates a production quality grid testbed. The NorduGrid system and
its usage will be presented.

GRC Time Series Forecasting research group meeting

Wednesday, 27 November at 17.30 hours

at the GRC office, Rue de Veyrot 12, 1217 Meyrin

Presentation by Ramo Gencay (University of Geneva/Windsor):

"Volatility on different time scales"

Wednesday, 23 October, 18.00-19.00 hours
Room SS-D01 building 40, CERN, Meyrin

Prof. Urs Luterbacher, Graduate Institute of International Studies, Geneva:

"Modeling Policy Choices in Climate Change"

Abstract:

The talk will explain how climate policy divergences between major countries can be examined with the help of a dynamic continuous time simulation model, but with parameters that can be potentially estimated empirically.
The model uses the SPARE simulation system which is based in part on software constructed at the Graduate Institute, in part on the CERN based PAW software package.

It allows the user:

1) to fit model output to historical data series with procedures based upon the CERN developed MINUIT minimization package, and

2) to apply a sequential decision making analysis, similar to game theory, in order to determine why an international solution to the climate change problem is preferred by a particular country.

The system also helps in determining possible cooperative solutions to divergent national policies.
A simulation model prototype is constructed with entities that have the characteristics of the US, the EU and a Rest of the World category. Some preliminary analysis will be presented as well as suggestions for further studies along the same lines.

Reference

Research group discussion meeting of the GRC time series forecasting group will take place on Tuesday, 15 Oct at 17.30 hours at CERN
Building 160 (near main entrance) Room 1-009
Meyrin

Prof. N. V. Krasnikov, Institute of Nuclear Research, Moscow
will lead the discussion about
"Uncertainty in the choice between economic models"

GRC seminar

Dr. Foort Hamelink
, Lombard Odier Darier Hentsch, Geneva,
Thursday, 26 September 16.30-17.30 hours
CERN room 60-6-002 building 60

Empirical factor analysis of the performance of global equity portfolios

Abstract:

Equity returns are believed to be strongly influenced by country, sector and style effects. A key issue is to be able to disentangle those various effects from one another. In particular, differences between country returns may simply reflect differences in the sector composition of country markets, which makes it clearly difficult to disassociate both effects. Similarly, from 1999-2001 the relative performance of Growth versus Value might be solely due to the striking performance of the Technology and Telecommunication sectors. For global equity portfolio managers, it is crucial to identify which factors offer the highest diversification benefits and return potential. We apply a multi-factor approach to estimate "pure" country, sector and style factor returns. Using data going back to 1990, we identify the major changes that have occurred in developed markets until 2001. Our various indicators clearly point out the growing influence of sector factors. However, country effects remain important and there is no clear-cut evidence that sector factors dominate country factors. Style factors such as Growth, Value and Size also remain significant, even once sector and country effects are deduced. Finally, we show that momentum strategies based on sector returns offer substantial gains, while momentum strategies based on country returns do not. These findings suggest that, while diversification and return benefits from sector strategies have become substantial, managers should continue to monitor carefully country as well as style rewards and risks.

Rerefence:

Country, Sector or Style: What matters most when constructing Global Equity
Portfolios? An empirical investigation from 1990-2001
by F. Hamelink , H. Harasty and P. Hillion




Research seminars at a glance:

GRC Research Group Meeting
"Modelling of univariate time series"

Tuesday, 23 July, 14:00 - 15:00
CERN, room 160/1-009


Speakers:
Dr.Allan Din (left), Geneva Research
Collaboration & CERN, Introduction, and
Dr.Nikita Stepanov (right), CERN, Main speaker

Modelling of the univariate time series.

Abstract:

After the introduction to the field of the time series forecasting and modeling which will include the formal definitions of the two essentially different kinds of problems and brief attempt to classify the modern methods applied, the speaker will try to advocate the "computer scientist`s" or "algorithmic" view of the problems defined which appears to be somewhat unusual and perhaps even unnatural for the bulk of the scientific community involved in these kinds of researches. In particular, the speaker will sketch the general theoretical tools and universal solutions of the probabilistic induction problems obtained already quite a long time ago. There is however the common drawback of the general elegant proofs, namely, often they are a bit far away from practical applications. Thus, after a rather general discussion the speaker will give an introduction to the phenomenological model (called Dukascopy by its author) which seems to be surprisingly good for the description of the short term behavior of the time series induced by the dynamics of the complex systems.

We are planning also the online Dukascopy based software demonstration. Everybody interested in this will be able to check his skill trying to forecast the behaviour of the natural stochastic process induced by the cosmic rays.

Some useful links:
http://www.cwi.nl/~paulv/papers/solomonoff.ps
http://www.idsia.ch/~juergen/toesv2/
http://www.dukascopy.com
http://www.cs.auckland.ac.nz/CDMTCS/chaitin/
http://www.cwi.nl/~paulv/

Dr. Allan Din,
"Optimization and forecasting with financial time series"
Tuesday, 25 June, 14.00-15.00
CERN, building 160, room 1-009

Abstract:
The talk reviews analytical techniques which may be applied to different types of financial time series.Optimization techniques are discussed from the perspective of risk analysis and optimal asset allocation. Forecasting is discussed within the framework of parametric and non-parametric statistical techniques, involving neural networks and genetic algorithms.

Useful references:
Return is Only Half the Equation
RiskGrades Technical Document
Time Series Analysis and Forecasting Techniques

Prof. Rosario Mantegna, University of Palermo,
Professor Mantegna is a leading expert in the field of theoretical and empirical modelling of complex systems. Since 1989 a major focus of his research has been studying financial systems using methods of statistical physics.
He is a co-author of the essential book on the subject "An Intruduction to Econophysics".
Prof. Mantegna spoke on the topic of:
"Econophysics and quantitative finance".

Abstract:
During the last years the number of physicists interested in the analysis and modeling of financial markets for academic and/or professional reasons is rapidly increased. Finance is today more and more a quantitative discipline and risk management is a multidisciplinary task needing several professionals to be performed at a top level. Physicists concur to these activities and to the modeling of a fascinating "complex system" with their professional and conceptual background.

6 May, 16.30 hours
CERN Building 40, room S2-B01

Click here for CERN access map.

Background information:
Giovanni Bonanno, Fabrizio Lillo and Rosario N. Mantegna
Levels of Complexity in Financial Markets, Physica A, 299, 16-27

J. Doyne Farmer, Santa Fe Institute
Physicists Attempt to Scale the Ivory Towers of Finance




In a lecture for physicists and bankers at CERN
NOBEL LAUREATE ILYA PRIGOGINE SPOKE ABOUT CHAOS AND ORDER

Prigogine`s lecture was delivered on the 24 of January 2002 in CERN

Prigogine`s visit in photos

Bruno Estier (left),
Banque Lombard Odier Cie, President of
Swiss Association of Market Technicians and Serge Laedermann, Geneva Finance
delivering the lecture on "The practice and challenges of Technical Analysis of financial data" at CERN, 12 December, 2001