Sessional Research Event: Extreme Events Working Party Paper
The Extreme Events working party will present an afternoon workshop followed by a formal sessional meeting. In order to attend both parts of the event you need to make two bookings, please follow the links below:
Workshop
Please book now for the workshop
Please download the following excel spreadsheets developed by the working party and pre-load to your laotop before the workshop (Please note: you are not required to review the spreadsheets before the workshop):
13:30 - 15:00 Session I: Errors in proxy models
Section 5 of the paper gives numerical examples of the use of proxy models, that is, approximate formulas for assets and liabilities which are quicker and easier to calculate than full models. This workshop shows in detail how these examples were constructed, based on the Excel spreadsheets the working party has developed, which will be distributed electronically to delegates.
Delegates explore:
- How to construct proxy models of bonds, annuities, term assurance and investment guarantees, fitted to stress tests
- Linear fits, separable quadratic, cross terms and higher order terms
- Fit measures: R-squared, Akaike and Bayes information, testing parameter significance with F-tests
- Monte Carlo simulation of exact and approximate loss functions in percentile calculations
- Comparing mean squared error to worst case error
- Constructing minimax fits and identifying Achilles nodes where the fit is especially challenging.
- How the fit improves as the numbers of fitting points and basis functions increase.
- Discussions of how these may relate to other products ]
15:00 -15:10 Break
15:10 - 16:30 Session II: Errors in model choice and parameter estimates
Section 4 of the paper shows ways of allowing for uncertainty in models and parameters. The second half of the workshop shows applications of these tools using the spreadsheets which the working party developed. The presentation explores:
- Frequentist definitions of interval size, bias and efficiency
- Algorithms for interval construction using Monte Carlo construction
- Computation of percentile estimates based on method of moments, probability weighted moments and maximum likelihood
- Understanding why different methods produce different answers and understanding the attributes of data sets for which the differences are largest.
- Metropolis Hastings algorithms for calculating Bayesian posterior distributions
- Calculating capital charges for parameter and model risk
- Implications for statistical quality standards and model validation
- Communication of statistical test results
Excel spreadsheet #3 (Macros required)
To gain maximum benefit from the workshop, delegates should bring their own laptops and upload the sample spreadsheets provided in advance. Please ensure you have at least two hours of battery life. The models in Session II use VBA macros so please ensure your security settings allow macros.
Sessional Meeting
Please book now for the sessional meeting
Download the paper by Ralph Frankland (chair) and other members of the Extreme Events working party.
16:30 - 17:00 Meet the Authors
17:00 - 17:30 Registration
17:30 - 19:00 Sessional meeting
Please note, that as a result of member feedback, we will be trialling a new ‘enhanced networking initiative’ at this sessional research meeting.
The purpose of this initiative is to improve integration and networking among members, with particular focus on making those who are newly qualified or new to events, feel encouraged to approach more experienced members and become integrated into the actuarial community.
At each meeting, members who have particular experience in the topic area or those who are happy to be approached by new members and share their experience will be asked to wear a yellow ribbon on their name badge.
We would invite all those members who are happy to wear a yellow ribbon on their name badge to signal that they encourage any member to approach them and ask questions about the sessional meeting, chat or even just to introduce themselves to get in touch with Event Management.