Math 478 / 568: Actuarial Modeling Professor Rick Gorvett Spring 2015

Syllabus Office Hours: 3-4 pm Tuesdays, 3-4 pm Wednesdays, or by appointment Textbook: Klugman, Panjer, and Willmot, 4th edition Exam dates: 3 exams, per syllabus Grades: Exams, homeworks, project

Syllabus (cont.) Graduate Students: Do Math 478, plus an extra project Project is potentially semester-long U/G Honors Students: Project alternatives will be handed out ~ half-way through the semester

Class Objectives Understand the mathematical foundations of actuarial modeling Loss modeling Model selection and parameter estimation Credibility theory and simulation Appreciate this material in a multi-disciplinary context Learn Exam 4/C material

Loss Modeling How do we represent the potential for financial consequences of events? Frequency × severity = aggregate loss Statistical distributions Frequency – e.g., Poisson Negative binomial Severity – e.g., Lognormal Exponential Gamma Pareto

Model Selection and Parameter Estimation How do we select amongst alternative models and parameters? How do we use empirical data to determine the characteristics of distributions? In what sense are some models and parameters “better” or “optimal” in a given situation?

Credibility Theory and Simulation How do we “blend”: Old and new data? Group versus individual data? E.g., { Z•New + (1-Z) •Old } Simulation How do we use models to estimate the impact of potential future scenarios?

Actuarial Science and Finance “Coaching is not rocket science.” - Theresa Grentz, former University of Illinois Women’s Basketball Coach Are actuarial science and financial mathematics “rocket science”? Certainly, lots of quantitative Ph.D.s are on Wall Street and doing actuarial- or finance-related work But….

Actuarial Science and Finance (cont.) Actuarial science and finance are not rocket science – they’re harder Rocket science: Test a theory or design Learn and re-test until successful Actuarial science and finance Things continually change – behaviors, attitudes,…. Can’t hold other variables constant Limited data with which to test theories

Two real-world examples Motivation Two real-world examples

Space Shuttle Challenger Explosion Example # 1 Space Shuttle Challenger Explosion http://www.youtube.com/watch?v=AfnvFnzs91s

Facts Leading Up to Launch… 23 successful launches prior to January 28, 1986 Previous launches at temperatures from 53°F to 81°F Challenger launch on morning of 1/28/86 was at 31°F – far below previous launches

Launch / O-Ring Information Launch vehicle configuration: Orbiter External fuel tank Two solid rocket boosters, manufactured by Morton Thiokol (MT) Sections sealed with O-rings, whose performance is sensitive to temperature But: MT’s recommendation stated that “Temperature data (are) not conclusive on predicting primary O-ring blowby.”

The Result Vehicle exploded 73 seconds after launch Cause (per Rogers Commission): gas leak in SRB, caused by failure or degredation of O-ring, led to weakening or penetration of external fuel tank Rogers Commission conclusion: “A careful analysis of the flight history of O-ring performance would have revealed the correlation of O-ring damage in low temperature.”

Statistical Analysis How predictable was it? Data:

Statistical Analysis (cont.) Or: Charts from “Risk Analysis of the Space Shuttle: Pre-Challenger Prediction of Failure,” by Dalal, et al, Journal of the American Statistical Association, December 1989

Taco Bell and The Mir Space Station Example # 2 Taco Bell and The Mir Space Station

Taco Bell and Mir Space Station Mir Taco Bell In orbit for 15 years Expected to crash back to earth on March 24, 2001, in the Pacific Ocean Size of projected debris field: 200 km × 6,000 km Taco Bell Offered a free Crunchy Beef Taco to every U.S. resident if the core of Mir hit a 144 square-meter target 15 km off Australian coast

Taco Bell and Mir (cont.) Suppose you are an actuary, working for an insurance firm Your firm has been approached by Taco Bell to insure against the potential financial loss associated with their possible Mir-related payout What’s a reasonable price for such coverage?

Taco Bell and Mir (cont.) Aggregate loss = frequency times severity What is the probability of Mir hitting the target? What will it cost Taco Bell if it does?

Taco Bell and Mir (cont.) Potential cost: Population of United States? Cost of a Crunchy Beef Taco? Potential cost? Probability of a hit? Indicated premium?

Taco Bell and Mir (cont.) Issues: Uniform distribution across debris field? How many will cash in? What about expenses / fixed costs?

Next Time Random variables Conditional probabilities and Bayes Theorem. Survival functions. Hazard rates.