Standard Error of Technical Cost Incorporating Parameter Uncertainty

Similar documents
Chapter 5: Generalized Linear Models

Delta Boosting Machine and its application in Actuarial Modeling Simon CK Lee, Sheldon XS Lin KU Leuven, University of Toronto

Introduction to Rare Event Simulation

Institute of Actuaries of India

A Practitioner s Guide to Generalized Linear Models

7 Likelihood and Maximum Likelihood Estimation

Modelling Operational Risk Using Bayesian Inference

DELTA METHOD and RESERVING

Econometric Methods and Applications II Chapter 2: Simultaneous equations. Econometric Methods and Applications II, Chapter 2, Slide 1

Estimation of Operational Risk Capital Charge under Parameter Uncertainty

SPRING 2007 EXAM C SOLUTIONS

GLM I An Introduction to Generalized Linear Models

PRICING AND PROBABILITY DISTRIBUTIONS OF ATMOSPHERIC VARIABLES

STATS 200: Introduction to Statistical Inference. Lecture 29: Course review

Does low participation in cohort studies induce bias? Additional material

SYSM 6303: Quantitative Introduction to Risk and Uncertainty in Business Lecture 4: Fitting Data to Distributions

Institute of Actuaries of India

Practice Problems Section Problems

Two hours. To be supplied by the Examinations Office: Mathematical Formula Tables and Statistical Tables THE UNIVERSITY OF MANCHESTER.

Introduction to Regression Analysis. Dr. Devlina Chatterjee 11 th August, 2017

Practice Exam 1. (A) (B) (C) (D) (E) You are given the following data on loss sizes:

Bayesian inference. Rasmus Waagepetersen Department of Mathematics Aalborg University Denmark. April 10, 2017

INSTITUTE OF ACTUARIES OF INDIA

Non-Life Insurance: Mathematics and Statistics

Bootstrapping the triangles

PL-2 The Matrix Inverted: A Primer in GLM Theory

Risk Aggregation with Dependence Uncertainty

Introduction and Overview STAT 421, SP Course Instructor

Experience Rating in General Insurance by Credibility Estimation

Ratemaking with a Copula-Based Multivariate Tweedie Model

General approach to the optimal portfolio risk management

Chapter 5. Sampling Distributions for Counts and proportions

Creating New Distributions

Censoring mechanisms

Bayesian Nonparametric Regression for Diabetes Deaths

Characterizing Forecast Uncertainty Prediction Intervals. The estimated AR (and VAR) models generate point forecasts of y t+s, y ˆ

Part 6: Multivariate Normal and Linear Models

Course 4 Solutions November 2001 Exams

Introduction: exponential family, conjugacy, and sufficiency (9/2/13)

Impacts of Frequency Contagion on Pricing of Catastrophe Excess of Loss Reinsurance for Australian Natural Perils

Statistics - Lecture One. Outline. Charlotte Wickham 1. Basic ideas about estimation

M(t) = 1 t. (1 t), 6 M (0) = 20 P (95. X i 110) i=1

Inference and Regression

Probability, Statistics, and Reliability for Engineers and Scientists FUNDAMENTALS OF STATISTICAL ANALYSIS

Stat Lecture 20. Last class we introduced the covariance and correlation between two jointly distributed random variables.

A Study of the Impact of a Bonus-Malus System in Finite and Continuous Time Ruin Probabilities in Motor Insurance

Three hours. To be supplied by the Examinations Office: Mathematical Formula Tables and Statistical Tables THE UNIVERSITY OF MANCHESTER.

EXAMINATIONS OF THE ROYAL STATISTICAL SOCIETY

green green green/green green green yellow green/yellow green yellow green yellow/green green yellow yellow yellow/yellow yellow

Confidence Intervals. CAS Antitrust Notice. Bayesian Computation. General differences between Bayesian and Frequntist statistics 10/16/2014

Hypothesis Testing. Part I. James J. Heckman University of Chicago. Econ 312 This draft, April 20, 2006

Lecture 7. Testing for Poisson cdf - Poisson regression - Random points in space 1

Reducing Model Risk With Goodness-of-fit Victory Idowu London School of Economics

Tail Conditional Expectations for Extended Exponential Dispersion Models

Approximation around the risky steady state

Copula Regression RAHUL A. PARSA DRAKE UNIVERSITY & STUART A. KLUGMAN SOCIETY OF ACTUARIES CASUALTY ACTUARIAL SOCIETY MAY 18,2011

Elicitability and backtesting

Content Preview. Multivariate Methods. There are several theoretical stumbling blocks to overcome to develop rating relativities

Econ 2148, spring 2019 Statistical decision theory

Ruin Theory. A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science at George Mason University

Reinsurance and ruin problem: asymptotics in the case of heavy-tailed claims

Department of Agricultural Economics. PhD Qualifier Examination. May 2009

Modeling Uncertainty in the Earth Sciences Jef Caers Stanford University

Parameter Estimation

exp{ (x i) 2 i=1 n i=1 (x i a) 2 (x i ) 2 = exp{ i=1 n i=1 n 2ax i a 2 i=1

A Note On The Erlang(λ, n) Risk Process

Quality Control Using Inferential Statistics In Weibull Based Reliability Analyses S. F. Duffy 1 and A. Parikh 2

University Of Calgary Department of Mathematics and Statistics

Exam C Solutions Spring 2005

MONOTONICITY OF RATIOS INVOLVING INCOMPLETE GAMMA FUNCTIONS WITH ACTUARIAL APPLICATIONS

STA 216: GENERALIZED LINEAR MODELS. Lecture 1. Review and Introduction. Much of statistics is based on the assumption that random

Statistical inference

On the Importance of Dispersion Modeling for Claims Reserving: Application of the Double GLM Theory

Stat 710: Mathematical Statistics Lecture 12

Operational risk modeled analytically II: the consequences of classification invariance. Vivien BRUNEL

Mixture distributions in Exams MLC/3L and C/4

Bayesian Computation

Part 8: GLMs and Hierarchical LMs and GLMs

ABSTRACT KEYWORDS 1. INTRODUCTION

Polynomial approximation of mutivariate aggregate claim amounts distribution

Course 1 Solutions November 2001 Exams

Research Article A Nonparametric Two-Sample Wald Test of Equality of Variances

Comparing two independent samples

GB2 Regression with Insurance Claim Severities

green green green/green green green yellow green/yellow green yellow green yellow/green green yellow yellow yellow/yellow yellow

Statistics Boot Camp. Dr. Stephanie Lane Institute for Defense Analyses DATAWorks 2018

Discrete Distributions Chapter 6

You must continuously work on this project over the course of four weeks.

SCHOOL OF MATHEMATICS AND STATISTICS. Linear and Generalised Linear Models

(a) (3 points) Construct a 95% confidence interval for β 2 in Equation 1.

Basic Concepts of Inference

if n is large, Z i are weakly dependent 0-1-variables, p i = P(Z i = 1) small, and Then n approx i=1 i=1 n i=1

Logistic regression: Miscellaneous topics

Inferring from data. Theory of estimators

HB Methods for Combining Estimates from Multiple Surveys

An Introduction to Spectral Learning

Confidence Intervals of the Simple Difference between the Proportions of a Primary Infection and a Secondary Infection, Given the Primary Infection

Corrections to Theory of Asset Pricing (2008), Pearson, Boston, MA

f X (x) = λe λx, , x 0, k 0, λ > 0 Γ (k) f X (u)f X (z u)du

LECTURE 11: EXPONENTIAL FAMILY AND GENERALIZED LINEAR MODELS

Transcription:

Standard Error of Technical Cost Incorporating Parameter Uncertainty Christopher Morton Insurance Australia Group This presentation has been prepared for the Actuaries Institute 2012 General Insurance Seminar. The Institute Council wishes it to be understood that opinions put forward herein are not necessarily those of the Institute and the Council is not responsible for those opinions.

Technical Cost: the best estimate of all the costs from underwriting a policy Key input into: Pricing decisions Introduction Performance and Monitoring Reports However, The variance is not commonly used 2

Motivation 3

Motivation Multiple claim types combine to form a technical cost: Technical Cost = ClaimFreq AvgClaimSi What distribution does this quantity have? t t ze t Which of the above policies is more risky? Just some of the questions I wanted answered 4

Motivation Review of Actuarial and Statistical literature found some answers, but: othing specific about GLMs Used simulation techniques, not a closed form solution Did not always consider parameter uncertainty So, this paper was written. 5

Today s Presentation Considering variance is Significant Useful Parameter uncertainty Is present when using models eeds to be quantified 6

Scope Fundamental Equation of Insurance Why Include Parameter Uncertainty? Practical Applications Conclusion 7

Fundamental Equation of Insurance Y = X i= 1 i Y = Claims Cost for a specific claim type where, = Random number of claims Xi = Random size of claim i This equation is: Fundamental to the study of Ruin Theory The basis of insurance. and X could belong to any distribution. We will assume: is Poisson distributed X is Gamma distributed 8

Why Include Parameter Uncertainty? Rank correlation is ~97%. 9

Why Include Parameter Uncertainty? Frequency and sizes of a person s claims are randomly distributed Each person has a unique distribution If we knew their true distribution, then our jobs would be significantly easier! However, Random process is observed only through data we collect 10

Why Include Parameter Uncertainty? Generalised Linear Models allow a convenient framework to link data and distributions together: T ( ˆ ) X ˆ β ~ Poisson µ T ˆ µ X X i X ˆ X β X ~ iid. Gamma ˆ φx, ˆ φx Because the parameters, β, are estimated by Maximum Likelihood we know they have an asymptotic distribution: = X T 1 ˆ X T X T ( X T ˆ η β ~ β, φ WX ) X This uncertainty in the estimate, is called Parameter Uncertainty 11

Why Include Parameter Uncertainty? ˆ η = ( X T WX ) X X T 1 ˆ X T ~, X T β β φ Parameter Uncertainty is affected by the specification of the GLM, X T Fitting low exposure or insignificant parameters will increase uncertainty Model fit is important 12

Further Details First condition on our data, D = X T β: Var ( Y ) = ED[ Var( Y D) ] + VarD [ E( Y D) ] = ED{ E[ Var( Y D, ) D] + Var [ E( Y D, ) D] } + VarD{ E[ E( Y D, ) D] } 2 = E Var( X D) E ( D) + E( X D) Var ( D) + Var [ E X D E D ] D [ ] ( ) ( ) i Var(Y) is now a function of the moments of X D and D Solving for the moments: 2 Var( Y ) = exp( η + σ 2ˆ η 2) exp( η + 2) X σ 2ˆ ηx {( ˆ ˆ 1 φ + φ ) exp( σ 2ˆ ) + exp( η + σ 2ˆ 2) [ exp( σ 2ˆ + σ 2ˆ ) 1] } X η X i η η D η X i 13

Further Details Morton C, 2012, Standard Error of Technical Cost Incorporating Parameter Uncertainty www.actuaries.asn.au/gis2012/ 14

Practical Applications Variance can be used together with the expected claims cost to: Add an additional dimension to analyses Graphically compare two different pricing scenarios based on Profit and Standard Error of Technical Cost 15

Pricing Scenario Heat Map Intensity of colour represents expected number of policies at that price based on elasticity models 16

Practical Applications Confidence Intervals Allocate capital down to an individual policy and use as an input into DFA modelling Risk based KPIs 17

Conclusion Variance of technical cost Has practical uses, including adding an additional dimension to analyses Variance can only be estimated from data we collect and analyse Which, necessarily, adds parameter uncertainty 18

Questions? 19