Friday, May 31, 2019 — 6:00 PM EDT
MACTSC 10th Anniversary banner

In 2019, the Master of Actuarial Science (MActSc) professional degree program will be celebrating 10 wonderful years at the University of Waterloo. 

Friday, May 31, 2019 — 10:30 AM EDT
Department Logo

Worst-case risk measures and distributionally robust optimization

Distributional ambiguity refers to the situation where the probability distribution of uncertain outcomes is unknown. The question of how to account for distributional ambiguity has been of central interest in risk management, and more generally, many fields involving decision making under uncertainty. In this talk, we present a general framework of risk minimization based on distortion risk measures (also known as dual utility) and show how the worst-case risk can be evaluated when only the support and moments are known for the underlying distribution. We also show that the problem of minimizing the worst-case risk, also known as distributionally robust optimization (DRO) problem, can be solved efficiently in large scale for a large class of decision problems including portfolio optimization, production and transportation planning, among many others. Worst-case distributions, i.e. distributions attaining the worst-case risk, are characterized, which offer useful intuition about the worst-case scenarios. 

Wednesday, May 15, 2019 — 9:30 AM to Thursday, May 16, 2019 — 3:30 PM EDT

This workshop provides a crash course on using statistical methods and software when conducting data analysis in survey research. There will be a hands-on opportunity to conduct basic data analysis using SAS software. This workshop is presented by Dr. Christian Boudreau, Co-director of the Survey Research Centre (SRC), along with Grace Li from the International Tobacco Control (ITC) Project.

Friday, May 10, 2019 — 10:30 AM EDT

SURPLUS-INVARIANT RISK MEASURES ON ROBUST MODEL SPACES


In this talk, we present a systematic study of the notion of surplus invariance. In essence, the property of surplus invariance stipulates that whether or not a financial institution is adequately capitalized from a regulatory perspective should not depend the surplus profile of the company but only on its default profile. Besides providing a unifying perspective on the existing literature, we establish a variety of new results including dual representations and extensions of surplus-invariant risk measures and structural results for surplus-invariant acceptance sets. The power of our results is demonstrated in model spaces with a dominating probability, including Orlicz spaces, as well as in robust model spaces where a dominating probability does not exist.

Thursday, May 9, 2019 — 4:00 PM EDT

 Robust Q-learning


The main goal of precision medicine is to use patient characteristics to inform a personalized treatment plan as a sequence of decision rules that leads to the best possible health outcome for each patient. Q-learning is a reinforcement learning algorithm that is widely used to estimate an optimal dynamic treatment regime using both multi-stage randomized clinical trials and observational studies. Starting with the final study stage, Q-learning finds the treatment option that optimizes the desired expected outcome. Fixing the optimally-chosen treatment at the last stage, Q-learning moves backward to the immediately preceding stage and searches for a treatment option assuming that future treatments will be optimized. The process continues until the first stage is reached. Q-learning requires specifying a sequence of regression models and the validity of the concluding results relies on assuming that the models are correctly specified. Specifically, due to the nature of backward induction, the subsequent models are likely to be a complex function of covariates which may result in non-ignorable residual confounding under model misspecification. We propose a robust Q-learning method that leverages flexible machine learning techniques to reduce the chance of model misspecification, thereby while maintaining the efficiency of Q-learning, mitigating the main drawback of this method. We derive the asymptotic properties of our method and show that, under certain conditions, it will result in asymptotically linear estimators with certain influence functions.

Friday, May 3, 2019 — 3:00 PM to Sunday, May 5, 2019 — 6:00 PM EDT
Datafest Logo

Friday, May 3, 2019 — 10:30 AM EDT

Optimisation under Uncertainty


Numerical solutions to optimisation problems are of large interest for various applications. Data scarcity, measurement errors and model uncertainty are clear examples where numerical optimisations are made under uncertainty. Robust optimisation is an attempt to automatically find robust optimal solutions, but show that the degree of robustness of the actual procedure depends on the shape of the objective function when finding the optimal reinsurance contract. We first go through some examples that would explain the mechanics behind robust optimisation under uncertainty and we further extend our discussion to the classical data exploratory method, namely Multidimensional Scaling (MDS). We illustrate how the robust optimisation should be adapted in order to achieve more robust decisions.

S M T W T F S
28
29
30
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
1
  1. 2019 (45)
    1. October (2)
    2. September (1)
    3. August (1)
    4. July (2)
    5. June (2)
    6. May (7)
    7. April (7)
    8. March (6)
    9. February (4)
    10. January (13)
  2. 2018 (44)
    1. November (6)
    2. October (6)
    3. September (4)
    4. August (3)
    5. July (2)
    6. June (1)
    7. May (4)
    8. April (2)
    9. March (4)
    10. February (2)
    11. January (10)
  3. 2017 (55)
  4. 2016 (44)
  5. 2015 (38)
  6. 2014 (44)
  7. 2013 (46)
  8. 2012 (44)