Projekt

Zurück zur Übersicht

Mathematical modeling of credit and equity risk beyond homogeneity and stationarity assumptions: statistical factor models and high-performance data mining

Gesuchsteller/in Horenko Illia
Nummer 140829
Förderungsinstrument Projekte
Forschungseinrichtung Facoltà di scienze informatiche Università della Svizzera italiana
Hochschule Università della Svizzera italiana - USI
Hauptdisziplin Mathematik
Beginn/Ende 01.09.2012 - 31.03.2017
Bewilligter Betrag 319'696.00
Alle Daten anzeigen

Alle Disziplinen (2)

Disziplin
Mathematik
Volkswirtschaftslehre

Keywords (8)

high-performance numerical methods, non-stationary time series analysis, data mining, nonlinear factor models, credit risk, massively parallel computing, Markov processes, equity risk

Lay Summary (Englisch)

Lead
Lay summary

Analysis and prediction of financial risks, such as market and credit risks, is one of the central problems in modern economy. The task of adequate mathematical description of the available risk data in its very complex nature (resulting from the presence of different temporal and spatial, i.e. regional, sectorial and global, scales) becomes more and more important in the context of the recent evolutions of the world economy. Important questions thereby are: (i) an investigation of the mutual influence of different risks and their spatial (e.g., regional) and temporal (e.g., associated with the business cycle) evolution, (ii) identification of the most important impact factors that play a role in their dynamics.
Accumulation of sufficiently detailed economic time series has led to the creation of huge databases, implicitly containing hidden information that may enhance our understanding of the complex processes underlying the economy. However, the extraction of this essential information is hindered by the  non-stationary (in time) and non-homogeneous (geographically) nature of the analyzed data. Non-stationarity is understood here as the change over time of some model parameters, while non-homogeneity refers to the inherently different dynamics across statistical units (e.g., assets, corporations). Main problems arising from these issues most prominently manifest themselves in the following areas : (i) in the applicability of standard statistical methodology; (ii) in adequateness of the standard mathematical description of the underlying processes and (iii) in the practical computational implementation of the data analysis algorithms on modern supercomputer architectures.

The main aim of this project will be to address these problems in a specific context of credit and market risk analysis and modeling. It is planned to develop new methods of time series analysis for parameterizing the risks as a spatially coupled non-stationary and non-homogeneous stochastic process under the influence of global and local impact factors (e.g., gross national product, level of depth, stock market indicators, etc.). Conceptual development of new mathematical models and statistical analysis methods for market and credit risks will go hand-in-hand with implementation and comparison on high-performance platforms at the Swiss National Supercomputing Center (CSCS) in Lugano, Switzerland. Resulting methods and algorithms should be applied to transparent analysis of the available financial data bases for identification of statistically significant regional and global inter-dependencies and extraction of the significant external factors influencing Swiss economy.

This project bundles the expertise of the three PIs in non-stationary time series analysis (I. Horenko), financial econometrics/computational finance (P. Gagliardini, I. Horenko), credit risk models (P. Gagliardini) and high-performance architectures and algorithms (W. Sawyer).

Direktlink auf Lay Summary Letzte Aktualisierung: 21.02.2013

Verantw. Gesuchsteller/in und weitere Gesuchstellende

Mitarbeitende

Publikationen

Publikation
Discrete non-homogenous and nonstationary logistic and Markov regression models for spatio-temporal data with unresolved external influences
Wiljes J. de, Putzig L., Horenko I. (2014), Discrete non-homogenous and nonstationary logistic and Markov regression models for spatio-temporal data with unresolved external influences, in Communications in Applied Mathematics and Computational Science (CAMCoS), 9(1), 1-46.
Time-Varying Risk Premium in Large Cross-Sectional Equity Data Sets
Gagliardini P., Ossola E., Scaillet O. (2016), Time-Varying Risk Premium in Large Cross-Sectional Equity Data Sets, in Econometrica, 84(3), 985-1046.
Efficiency in large dynamic panel models with common factors
Gagliardini P., Gourieroux C. (2014), Efficiency in large dynamic panel models with common factors, in Econometric Theory, 30(5), 961-1020.
On inference of statistical regression models for extreme events based on incomplete observation data
Kaiser O., Horenko I. (2014), On inference of statistical regression models for extreme events based on incomplete observation data, in Communications in Applied Mathematics and Computational Science (CAMCoS), 9(1), 143-174.
Statistical regression analysis of threshold excesses with systematically missing covariates
Kaiser O., Igdalov D., Horenko I. (2014), Statistical regression analysis of threshold excesses with systematically missing covariates, in SIAM J. of Multiscale Modeling and Simulation, 13(2), 594-613.

Zusammenarbeit

Gruppe / Person Land
Felder der Zusammenarbeit
Olivier Scaillet, HEC Geneva Schweiz (Europa)
- vertiefter/weiterführender Austausch von Ansätzen, Methoden oder Resultaten
- Publikation
Prof. Dr. Dirk Becherer/ Humboldt University of Berlin Deutschland (Europa)
- vertiefter/weiterführender Austausch von Ansätzen, Methoden oder Resultaten

Wissenschaftliche Veranstaltungen

Aktiver Beitrag

Titel Art des Beitrags Titel des Artikels oder Beitrages Datum Ort Beteiligte Personen
Seminar of the Department of Economics Einzelvortrag Is Industrial Production Still the Dominant Factor for the US Economy? 29.03.2017 University of Surrey, Surrey, Grossbritannien und Nordirland Gagliardini Patrick
Seminaire du Laboratoire de Finance Assurance Einzelvortrag A DIAGNOSTIC CRITERION FOR APPROXIMATE FACTOR STRUCTURE 23.02.2017 CREST, Paris, Frankreich Gagliardini Patrick
TUM-AIS Focal Period 2017, International symposium "Machine Learning Challenges in Complex Multiscale Physical Systems" Poster Towards a computationally-tractable maximum entropy principle for non-stationary financial time series 10.01.2017 TUM Institute for Advanced Study (TUM-IAS), Garching, Deutschland Marchenko Ganna
Machine Learning Challenges in Complex Multiscale Physical Systems Vortrag im Rahmen einer Tagung On Some Aspects of a Data-driven Model Reduction in Multiscale Systems 09.01.2017 TUM Institute for Advanced Study (TUM-IAS), Garching, Deutschland Horenko Illia
Seminar of the Department of Economics Einzelvortrag A Diagnostic Criterion For Approximate Factor Structure 19.12.2016 University of Bern, Bern, Schweiz Gagliardini Patrick
10th International Conference on Computational and Financial Econometrics Vortrag im Rahmen einer Tagung Towards a computationally-tractable maximum entropy principle for nonstationary financial time series 09.12.2016 University of Seville, Seville, Spanien Marchenko Ganna
Seminar: Different Mathematical Perspectives on Description of Unresolved Scales in Multiscale Systems Einzelvortrag Invited speaker 20.11.2016 Oberwolfach Institute, Oberwolfach , Deutschland Horenko Illia
Seminar of the Research Center for Statistics Einzelvortrag Diagnostic Criterion for Approximate Factor Structure 28.10.2016 University of Geneva, Geneva, Schweiz Gagliardini Patrick
Workshop "Mathematical and Algorithmic Aspects of Data Assimilation in the Geosciences" Vortrag im Rahmen einer Tagung On a direct data-driven reduction of Bayesian models 02.10.2016 Oberwolfach Institute, Oberwolfach , Deutschland Horenko Illia
The annual workshop of the Center for Computational Sciences Einzelvortrag Towards a direct data-driven identification of causality-preserving reduced dynamical models 29.09.2016 IMB, Mainz, Deutschland Horenko Illia
Workshop "Multiscale Interactions in Geophysical Fluids" Vortrag im Rahmen einer Tagung Causality or correlation? Challenges in data-driven modeling of multiscale geophysical systems 14.08.2016 Oberwolfach Institute, Oberwolfach , Deutschland Horenko Illia
Seminar of the Institute of Finance Einzelvortrag A Diagnostic Criterion For Approximate Factor Structure 12.04.2016 EPFL, Lausanne, Schweiz Gagliardini Patrick
Seminar of the Tinbergen Institute Einzelvortrag Invited speaker 08.10.2015 Tinbergen Institute, Amsterdam, Niederlande Gagliardini Patrick
DMV 2015 Vortrag im Rahmen einer Tagung Causality or correlation? Multiscale inference and applications to geoscience 25.09.2015 Hamburg University, Hamburg, Deutschland Horenko Illia
Workshop "Causality in Turbulence" (organizer P. Koumoutsakos) Einzelvortrag Analysis of non-stationary time series data 09.06.2015 ETHZ, Zurich, Schweiz Horenko Illia
SoFiE Summer School on the Econometrics of Option Pricing 2015 Einzelvortrag Invited Lecturer 01.06.2015 ULB, Bruxelles, Belgien Gagliardini Patrick
MATHICSE Seminar (organiser A. Abdulle) Einzelvortrag Challenges of data analysis in a multiscale context: causality inference and unresolved scales 19.04.2015 EPFL, Lausanne, Schweiz Horenko Illia
SFB1114 Seminar (organiser R. Klein) Einzelvortrag Challenges of data analysis in a multiscale context 05.02.2015 FU Berlin, Berlin, Deutschland Horenko Illia
DBTA Workshop "Big Data: Stream Processing" Einzelvortrag Invited speaker 03.12.2014 Bern, Schweiz Horenko Illia
Seminar of the Department of Economics Einzelvortrag Identification by Laplace Transforms in Nonlinear Panel or Time Series Models with Unobserved Stochastic Dynamic Effects 14.10.2014 UCL, London, Grossbritannien und Nordirland Gagliardini Patrick
SoFiE Summer School on the Econometrics of Option Pricing 2014 Einzelvortrag Invited lecturer 28.07.2014 Harvard University, Cambridge, Vereinigte Staaten von Amerika Gagliardini Patrick
Partnership for Advanced Scientific Computing (PASC'14) Vortrag im Rahmen einer Tagung Distributed Memory Compression of Climate Time Series Data 02.06.2014 ETH, Zurich, Schweiz Sawyer William
CECAM conference "Long time dynamics from short time simulations" (organiser M. Parinello) Vortrag im Rahmen einer Tagung On extensions of Markov models towards multiscale setting: unresolved scales, data-driven models and application to MD 12.03.2014 USI, Lugano, Schweiz Horenko Illia
CAOS Colloquium Einzelvortrag Time series analysis and data-driven model discrimination beyond usual assumptions: mathematical ideas, algorithmic methods, HPC and application examples 04.02.2014 Courant Institute of Mathematical sciences, New York, Vereinigte Staaten von Amerika Horenko Illia
NDNS+ Workshop "Stochastic Modeling of Multiscale Systems" (organiser J. Frank) Einzelvortrag Time series analysis and data-driven model discrimination beyond usual assumptions: ideas, methods and examples 04.12.2013 Eindhoven Multiscale Institute, Eindhoven, Niederlande Horenko Illia
Seminar of the Informatics Faculty (invited by D. Saupe) Einzelvortrag Time series analysis and data-driven model discrimination beyond usual assumptions: ideas, methods and examples 20.11.2013 Konstanz University, Konstanz, Deutschland Horenko Illia
Seminar of the ICMA Center Einzelvortrag Survival of Hedge Funds: Frailty vs. Contagion 07.02.2013 University of Reading, Reading, Grossbritannien und Nordirland Gagliardini Patrick
Workshop "Mathematical and Algorithmic Aspects of Atmosphere-Ocean Data Assimilation" (organisers A. Griewank, S. Reich, I. Roulstone, A. Stuart) Vortrag im Rahmen einer Tagung Nonstationarity and nonhomogeneity in GFD: data analysis, model discrimination beyond the standard probabilistic framework and missing data assimilation 06.12.2012 Oberwolfach Institute, Oberwolfach, Deutschland Horenko Illia
Seminar of Finance Einzelvortrag Survival of Hedge Funds: Frailty vs. Contagion 19.11.2012 ESSEC Business School, Cercy-Pontoise, Frankreich Gagliardini Patrick
HPC Service Provider Community meeting "Handling huge amount of data for HPC" Einzelvortrag Data compression and HPC: lossy or lossless? Methodological viewpoint and implications 24.10.2012 CSCS, Lugano, Schweiz Horenko Illia
Seminar of Finance Einzelvortrag Time-Varying Risk Premium in Large Cross-Sectional Equity Datasets 27.09.2012 University of Luxembourg, Luxembourg, Luxemburg Gagliardini Patrick
Seminar of Oeschger Center (invited by O. Romppainen-Martius) Einzelvortrag Statistical analysis of weather/climate data beyond the usual assumptions: theory, numerics and instructive examples 26.09.2012 Bern University, Bern, Schweiz Horenko Illia
CECAM conference "Machine Learning in Atomistic Simulations" (organiser M. Parinello) Vortrag im Rahmen einer Tagung MD Data Analysis beyond usual assumptions: theory, numerics and instructive examples 10.09.2012 USI, Lugano, Schweiz Horenko Illia


Auszeichnungen

Titel Jahr
Mercator Fellowship, SPP 1114 2014

Verbundene Projekte

Nummer Titel Start Förderungsinstrument
162633 New Econometric Methods for Big Data 01.09.2017 Projekte
162488 Corporate Default Risk in the Long-Run: Evidence from Switzerland, 1883-2015 01.09.2016 Projektförderung (Abt. I-III)

Abstract

The modeling of financial risks, such as market and credit risks, is one of the central problems in modern computational finance. The task of adequate mathematical description of the available risk data in its multi-scale nature (resulting from the presence of different temporal and spatial, i.e. regional, sectorial and global, scales) becomes more and more important in the context of the recent evolutions of the world economy. Important questions thereby are: (i) an investigation of the mutual influence of different risks and their spatial (e.g., regional) and temporal (e.g., associated with the business cycle) evolution, (ii) identification of the most important exogenous impact factors that play a role in their dynamics. Accumulation of sufficiently detailed economic time series has led to the creation of huge databases, implicitly containing hidden information that may enhance our understanding of the complex processes underlying the economy. However, the extraction of this essential information out of the data is hindered by the multidimensional, non-stationary (in time) and non-homogeneous (in the phase space) nature of the analyzed signal. Non-stationarity is understood here as the change over time of some model parameters, while non-homogeneity refers to the inherently different dynamics across statistical units (e.g., assets, corporations). Main problems arising from these issues most prominently manifest themselves in the following areas : (i) in the applicability of standard statistical time series analysis methodology; (ii) in adequateness of the standard mathematical modeling of the underlying dynamics and (iii) in the practical computational implementation of the data analysis algorithms on modern supercomputer architectures. In the context of statistical data mining, most of the existing time series analysis methods (e.g., Kalman filter, VARX, MVAR, ARCH/GARCH, etc.) impose restrictive stationarity assumptions on the data and are not suitable for describing multivariate data in very large dimensions, unless strong homogeneity assumptions are introduced. Therefore these methods are in principle not directly applicable to realistic non-stationary and non-homogeneous risk data. In context of econometrics and mathematical finance the same problem arises: strong implicit mathematical and statistical assumptions restrict the applicability of the standard models in describing realistic processes. For instance, in credit risk modeling the benchmark specifications assume typically time-constant model parameters as well as homogeneous effects of systematic risk factors on the individual risks in a portfolio, while in reality these stationarity and exchangeability conditions can be far from being satisfied. In the context of practical computational implementation, the unfavorable scaling restrains the applicability of the majority of the existent computational algorithms to a relatively small number of assets and short time series, a full analysis of realistic databases with several thousands of financial items and/or millions of the resolved time ticks (as available for stock and bond markets) still remains beyond their scope. The main aim of this project will be to address these problems in a specific context of credit and market risk analysis and modeling. It is planned to develop the new methods of time series analysis for parameterizing the risks as a spatially coupled non-stationary and non-homogeneous discrete process (Bernoulli and/or Markov) with external impact factors. Conceptual development of new statistical analysis methods and mathematical models for market and credit risks will go hand-in-hand with the high-performance implementation, investigation and comparison of the methods on the supercomputer architectures of the Swiss National Supercomputing Center CSCS (Manno, Switzerland). Resulting methods and algorithms should be applied to transparent analysis of the available financial data bases for identification of statistically significant regional and global interdependency and external factors influencing Swiss economy. This project bundles the expertise of the three PIs in non-stationary time series analysis (I. Horenko), financial econometrics/computational finance (P. Gagliardini, I. Horenko), credit risk models (P. Gagliardini) and high-performance architectures and algorithms (W. Sawyer).