The Actuarial Profession is running dependency modelling courses which aim to bridge the gap between industry practice and the latest thinking on dependency modelling. The course combines the latest theory and practical thinking in the areas of multivariate distributions, and calibration of copulas with examples that demonstrate how to apply these techniques to real world problems. The course will take place from 18-20 November 2009.
Who should attend?
Those who will benefit from attending the workshop are quantitative analysts, practising actuaries, other industry experts who deal with Risk Management, Asset Liability Modelling, Pricing, Value at Risk (VAR) or Funding Level at Risk (FLAR) estimates, and Econometrists interested in techniques to better capture market information.
Key benefits
These courses will provide attendees with Tools to broaden their modelling intellectual capital; Practical knowledge on how to make
better a priori decisions about risk management; The ability to recognise and better model unusual dependency structures.
Continuing Professional Development
Members of The Actuarial Profession may find the seminar a useful contribution to their personal professional development.
The amount of verifiable CPD hours recorded is left to the discretion of the individual. For each half day course you can record up to 2.5 hours of Technical CPD.
Format
It is recommended that you attend the half day overview course before attending the practical sessions. The overview session will cover a high level view of multivariate distributions, multivariate time series and alternative approaches. It will be followed by four half-day practical sessions, which will deliver rigorous training. Each session will be designed to contain practical examples of implementations with the view of delegates being able to implement the techniques at the workplace straight away. All sessions can be booked separately.
Session 1: Multivariate Gaussian distribution and Multivariate t-distribution
The course will show what the mutivariate Gaussian distribution is, how it arises, and some of the tests based on it. The multivariate t-distribution will be described
in the same way, including the case where the marginals do not necessarily have the same numbers of degrees of freedom.
Session 2: Multivariate negative binomial
distribution and D distribution
The context of the negative binomial distribution will be described, first classically and then in the case of non-integer parameter values, and tests using and related to this distribution will be described. The Ddistribution, a simple and flexible way to model a priori dependencies between variables will be described, first in terms of looking at correlation as a paired number, and then using this concept to build multivariate
samples with very different dependency structures for forward looking analysis.
Session 3: Empirical distributions and
alternative approaches
When there is no theoretical model for a distribution, one has to work simply with the observed data and the resulting
empirical distributions. Methods of treating these will be described, including copula function methods, and including warnings of some of the pitfalls that can arise.
Session 4: Multivariate time series
Often data arrive in the form of time series, and so rather than successive values being independent there is a time-structure that is an important part of the process and has to be represented in the model. Multivariate time-series models will be
described including ARMA models, the multivariate Yule-Walker equations for fitting such models to data, and some of the issues that arise in making those fits.
Book online at: http://www.actuaries.org.uk/members/transactions/conference_booking.