The aim of this workshop is to introduce the theory underlying HMM, Kalman Filters and Particle Filters.  The use of these methods in the calibration of dynamic state space models as well as in prediction of unobservable variables is also discussed.

• introduces and explores the theoretical aspects;
• presents the applications of state space modelling, Kalman filtering and nonlinear filtering;
• explains the practical applications of linear/non-linear filtering in financial risk management
• provides an overview on the future development of the application of filtering in finance

Practitioners at banks, risk professionals, traders, consultants and academics.

Programme:

Day One: Hidden Markov Models and Filtering Techniques

Introduction: Linear and Nonlinear Filtering in Mathematical Finance
Enza Messina, University of Milano-Bicocca

We review time series filtering and its applications in mathematical finance. A summary of recent empirical applications of linear filtering with real market data is presented for stochastic volatility modelling and yield curve modelling.  Some numerically tractable approaches to filtering of nonlinear time series are also outlined.

Hidden Markov Models
Peter Ruckdeschel, Fraunhofer ITWM

• Introduction and motivation for applications in Finance
• General form in discrete and continuous time observation processes
• Filtering
• Maximum likelihood estimation and EM algorithm
• Parameter estimation
• Forecast

Robust Statistics       
Peter Ruckdeschel, Fraunhofer ITWM

• Robustness in Filtering: Many definitions
• Outliers and their handling in filtering - Propagating and non-propagating behaviour
• Simulation study and stylized facts of outlier impact on prediction quality, on state reconstruction, on parameter estimation
• Summary of Robustness concepts:
  Neighborhoods, Minimax and Lemma5 Solutions
• AO (additive outlier) and IO (innovation outlier)-Robust Kalman filtering

A New Unscented Kalman Filter with Higher Order Moment‐Matching
Ksenia Ponomareva, Centre for the Analysis of Risk and Optimisation Modelling Applications (CARISMA)/Brunel University
A new approximate Bayesian algorithm for filtering nonlinear multivariate time series is proposed. It generates sample points and corresponding probability weights that match exactly the predicted values of average marginal skewness and average marginal kurtosis of the unobserved state variables, in addition to matching their mean and the covariance matrix. The performance of the algorithm is illustrated by an empirical example of yield curve modelling with real financial market data.

Day Two: Applications to Financial Time Series and Case Studies

Sequential Learning Methods for Tracking Stochastic Volatility
Enza Messina, University of Milano-Bicocca

Sequential learning approaches for estimating risk indicator parameters based on  particle filtering methods.
Tracking and forecasting the stochastic volatility through the bayesian estimation of the parameters of a stochastic process with jumps.

HMM-Based Investment Strategies for Asset Allocation
Peter Ruckdeschel, Fraunhofer ITWM

Investment strategies relying on hidden Markov model approaches are developed and analysed. Filtering techniques are utilized for recursive parameter estimations of vector observations. An investment strategy for investing in growth or value stocks is developed and results from an implementation on Russell 3000 data sets are presented.

Robustness Issues in HMM-based Investment Strategies for Asset Allocation  
Peter Ruckdeschel, Fraunhofer ITWM

Stability of the strategies developed in the case study stated above is checked by (a) implanting isolated outliers, spurious trends, and structural breaks into the original data set at a known time instance and (b) in a simulation study, where we quantify this effect by calculating MSE. We identify the culprits in the likelihood representation.

Dr Enza Messina is an Associate Professor in Operations Research at the Department of Informatics Systems and Communications, University of Milano-Bicocca, where she founded the research Laboratory MIND (Models in decision making and data analysis).
She received her PhD in Computational Mathematics and Operations Research from the University of Milano. in 1994.
Her research activity is mainly focused on the development of models and methods for decision making under uncertainty, in particular probabilistic models for sequential data analysis. These modelling techniques have been applied to the solution of risk management problems in different domains such as finance, supply chain and environment.

Ms Ksenia Ponomareva completed a BA in Mathematics and Certificate of Advanced Study in Mathematics course at Girton College, Cambridge University. After graduation she has been working in Quantitative Trading and Risk Management teams at a London hedge fund for two years prior to joining CARISMA as a PhD student.  Research areas she is working in are robustness and uncertainty in financial time series models. In particular two problems of interest  are robustness in filtering under perturbations and moment matching based algorithm for nonlinear filtering.

Dr Peter Ruckdeschel is a research associate at the Financial Mathematics Group at Fraunhofer ITWM, Kaiserslautern and at the University of Kaiserslautern.
He received his PhD in Statistics at University Bayreuth in 2001 with an award-winning dissertation on Robust Kalman Filtering. He is co-author of 13 R packages available on the comprehensive R archive network, ranging from integration of automatic markup to R documents, over object-oriented implementations of distributions to optimally robust estimation. Within the collaborative software development platform r-forge, he is developing—with other authors—R package robKalman. Since his affiliation to ITWM, he has been working on several industry projects covering parameter estimation in stochastic correlation models, quantification of operational risk, and pricing of loans in illiquid markets.

Industry Rate: 2 days £1025 + VAT

Thanks to our sponsors, there are a limited number of generous bursaries available for academics and research students so they can attend at reduced rates.

For more information regarding bursaries, please contact us on + 44 (0) 1895 819 488 or email info@optirisk-systems.com

Discounted rates for group bookings can be also arranged on request.