ACTL2002/ACTL5101 Probability and Statistics
c Katja Ignatieva
School of Risk and Actuarial Studies
Australian School of Business
University of New South Wales
k.ignatieva@unsw.edu.au
Week 1 Video Lecture Notes
Week 2
Week 3
Week 4
Probability: Week 1
Week 6
Review
Estimation: Week 5
Week
7
Week
8
Week 9
Hypothesis testing:
Week
10
Week
11
Week
12
Linear regression:
Week 3 VL
Week 4 VL
Week 5 VL
Video lectures: Week 2 VL
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes
Links to UNSW TV
Click on the topic to go to the online recording:
Week 1: Probability space, Calculating with probabilities,
Counting.
Week 2: Bernoulli distribution, Binomial distribution,
Geometric distribution, Negative Binomial distribution.
Week 3: Numerical methods to summarize data, Graphical
procedures to summarize data.
Week 4: Sampling with and without replacement, Properties
of sample mean and variance.
Week 5: Chisquared distribution, Studentt distribution,
Snecdor’s F distribution, Distribution of sample mean and
variance.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes
Probability space
Sample Space & σalgebra
Introduction in Probability
Probability space
Sample Space & σalgebra
Probability Measure
Calculating with probabilities
Properties of the Probability Measure
Conditional Probability
Independence
Bayes Theorem
Counting
Counting Principles
Computing Probabilities
Note: A sample point is a simple event. are pairwise disjoint sets in F. E ∈ F implies E c ∈ F. Ei ∩ Ej = ∅ ∞ S (null/empty set) for all i 6= j. then Ek ∈ F. k=1 An element E of F is called an event. E1 . The set of all possible outcomes is called the sample space. . 2. 2/33 . It is nothing but a subset of Ω. the outcomes cannot be predicted with certainty in advance. . E2 . that is. An element ω of Ω is a sample point.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Probability space Sample Space & σalgebra Sample Space & σalgebra In a random experiment. . A family F of subsets of the sample space Ω is said to be a σalgebra (σfield) if the following conditions hold: 1. denoted by Ω.
C }.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Probability space Sample Space & σalgebra If there are N elements of the sample space. diamonds (D). The sample space is: Ω = {H. spades (S). a σalgebra F of events may consist of 2N elements. All the sets of possible events are a σalgebra of F. or clubs (C ). 3/33 . S. Example: Consider the experiment of selecting a card at random from a deck of four cards with four different suits. and noting its suit: hearts (H). D.
so one σalgebra has 252 elements. C ) (H. C )(= Ω) ∅ (H. S. S) (D. S. S) (H. Note that a standard deck has 52 elements in the sample space. S. D. C ) (D. D. C ) Each element in the σalgebra is called an event. D) (H.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Probability space Sample Space & σalgebra The following collection of 24 = 16 sets is a σalgebra F : (H) (D) (S) (C ) (H. C ) (S. 4/33 . D. C ) (H. C ) (D. S) (H.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Probability space Sample Space & σalgebra Sample Spaces 5/33 .
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Probability space Probability Measure Introduction in Probability Probability space Sample Space & σalgebra Probability Measure Calculating with probabilities Properties of the Probability Measure Conditional Probability Independence Bayes Theorem Counting Counting Principles Computing Probabilities .
0 ≤ Pr (Ei ) ≤ 1. 6/33 . . 2. E2 . .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Probability space Probability Measure Probability Measure A probability measure on ω is a function Pr from the subsets of ω to < that satisfies the following axioms: 1. . k=1 k=1 A random experiment is therefore described as a probability triple (or probability space): {Ω. F. P}. . Pr (Ω) = 1. For all events Ei . where Ei ∩ Ej = ∅ for every i 6= j. . = Pr (Ek ) . For any events E1 . then: ! ∞ ∞ [ X Pr Ek = Pr (E1 ) + Pr (E2 ) + . 3.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Properties of the Probability Measure Introduction in Probability Probability space Sample Space & σalgebra Probability Measure Calculating with probabilities Properties of the Probability Measure Conditional Probability Independence Bayes Theorem Counting Counting Principles Computing Probabilities .
i.e. ω ∈ C iff ω ∈ A or ω ∈ B.. ω ∈ Ac iff ω ∈ / A. Intersection of two events: C = A ∩ B is an event that both A and B occurs. Events are disjoint: if they have no outcomes in commun i..ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Properties of the Probability Measure Definitions Union of two events: C = A ∪ B is an event that both A and/or B occurs. Complement of an event: B = Ac is the event that A does not occur. A and C are disjoint iff A ∩ C = ∅. i. i..e.e..e. 7/33 . ω ∈ C iff ω ∈ A and ω ∈ B.
Distributive laws: (A ∪ B) ∩ C = (A ∩ C ) ∪ (B ∩ C ) and (A ∩ B) ∪ C = (A ∪ C ) ∩ (B ∪ C ). Associative laws: (A ∪ B) ∪ C = A ∪ (B ∪ C ) and (A ∩ B) ∩ C = A ∩ (B ∩ C ).ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Properties of the Probability Measure Useful laws Commutative laws: A∪B =B ∪A and A ∩ B = B ∩ A. 8/33 . DeMorgan’s laws: (A ∩ B)c = Ac ∪ B c and (A ∪ B)c = Ac ∩ B c .
then Pr (E1 ) ≤ Pr (E2 ). 9/33 . then: Pr (E1 ∪ E2 ) = Pr (E1 ) + Pr (E2 ) − Pr (E1 ∩ E2 ) . Null / Empty Set: Pr(∅) = 0. Additive: If E1 and E2 are any two events. Subsets: If E1 and E2 are two events such that E1 ⊆ E2 .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Properties of the Probability Measure Properties of the Probability Measure Complement: Pr(E c ) = 1 − Pr(E ).
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Properties of the Probability Measure Properties of the Probability Measure 10/33 .
.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Properties of the Probability Measure In the case of three events. B = E2 . C = E3 . ** using distributive law: (A ∪ B) ∩ C = (A ∩ C ) ∪ (B ∩ C ) with A = E1 . 11/33 * using additive rule: Pr (A ∪ B) = Pr (A) + Pr (B) − Pr (A ∩ B) with A = (E1 ∪ E2 ) and B = E3 . *** using additive rule: Pr (A ∪ B) = Pr (A) + Pr (B) − Pr (A ∩ B) with A = E1 and B = E2 & A = (E1 ∩ E3 ) and B = (E2 ∩ E3 ). we have: Pr (E1 ∪ E2 ∪ E3 ) = ∗ = Pr ((E1 ∪ E2 ) ∪ E3 ) Pr (E1 ∪ E2 ) + Pr (E3 ) − Pr ((E1 ∪ E2 ) ∩ E3 ) ∗∗ = Pr (E1 ∪ E2 ) + Pr (E3 ) − Pr ((E1 ∩ E3 ) ∪ (E2 ∩ E3 )) ∗∗∗ = Pr (E1 ) + Pr (E2 ) + Pr (E3 ) − Pr (E1 ∩ E2 ) − Pr (E1 ∩ E3 ) − Pr (E2 ∩ E3 ) + Pr (E1 ∩ E2 ∩ E3 ) .
. . k=1 k=1 which follows the additive law of probability. .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Properties of the Probability Measure Inequality rules Using the definition of probability measure one can prove the following inequalities: Boole’s Inequality: If E1 . . then: n X Pr (E1 ∩ . E2 . E2 . Bonferroni’s Inequality: If E1 . k=1 12/33 . . En are any n events. then ! n n [ X Pr Ek ≤ Pr (Ek ) . . . En are any n events. . ∩ En ) ≥ 1 − Pr (Ekc ) . . .
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Properties of the Probability Measure Inequality rules 13/33 .
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Conditional Probability Introduction in Probability Probability space Sample Space & σalgebra Probability Measure Calculating with probabilities Properties of the Probability Measure Conditional Probability Independence Bayes Theorem Counting Counting Principles Computing Probabilities .
otherwise Pr (A B ) = 0. Pr (A B ) ≥ 0. The multiplication rule immediately follows: Pr (A ∩ B) = Pr (A B ) · Pr (B) .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Conditional Probability Conditional Probability The conditional probability of A. . The following properties are also immediate: 1. . 3. A2 . then mutually ∞ ∞ S P Pr Ak B = Pr (Ak B ). are disjoint events. given B. Pr (A A ) = 1. 2. as: Pr (A B ) = Pr (A ∩ B) Pr (B) provided Pr (B) > 0. . If A1 . 14/33 k=1 k=1 .
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Conditional Probability Conditional Probability 15/33 .
we have: Pr (A) = ∞ X k=1 16/33 Pr (A Ek ) · Pr (Ek ) . E2 . . .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Conditional Probability Law of total probability Law of total probability: If E1 . . are mutually disjoint (E Si ∞∩ Ej = ∅ for i 6= j) and comprise the entire sample space ( k=1 Ek = Ω). then for any event A ∈ F. .
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Independence Introduction in Probability Probability space Sample Space & σalgebra Probability Measure Calculating with probabilities Properties of the Probability Measure Conditional Probability Independence Bayes Theorem Counting Counting Principles Computing Probabilities .
. we have: Pr (Ei1 ∩ Ei2 ∩ . Eim . . En are independent if: Pr (E1 ∩ E2 ∩ . . E2 . . En . . we say that the events are mutually independent if for any subcollection Ei1 . we have events A and B independent if: Pr (A B ) = Pr (A) and Pr (B A ) = Pr (B) . . . . . . . . . . . · Pr (En ) . Note that we say the collection of events E1 . · Pr (Eim ) . 17/33 . Ei2 . . Equivalently. . ∩ En ) = Pr (E1 ) · Pr (E2 ) · . For a collection of several events E1 . . .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Independence Independence Events A and B are said to be independent if: Pr (A ∩ B) = Pr (A) · Pr (B) . E2 . ∩ Eim ) = Pr (Ei1 ) · Pr (Ei2 ) · . .
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Independence 18/33 .
. . i. We can generalize this to several events as follows: events E1 . . A ∩ B = ∅ so that Pr (A ∩ B) = 0 and Pr (A ∪ B) = Pr (A) + Pr (B) .e. E2 .. En are said to be mutually exclusive events if no two have an element in common (mutually disjoint) and that Pr (E1 ∪ E2 ∪ .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Independence Mutually Exclusive Events: A and B are said to be mutually exclusive if they are mutually disjoint. + Pr (En ) . . ∪ En ) = Pr (E1 ) + Pr (E2 ) + . . . . 19/33 . .
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes
Calculating with probabilities
Independence
Example
Consider the experiment of selecting n cards at random from a
deck of 52 cards with 13 hearts (H), 13 diamonds (D), 13 spades
(S), and 13 clubs (C ) cards.
False For n = 2, the second card is independent of the first card.
False For n = 14, the 14th card is mutually independent of the first
13 cards.
True Previous two questions, but now when the card is put back in
the stock before the next one is selected.
True For n = 3, A=“at least 2 H” and B=“at least 2 C” are
mutually exclusive events.
False For n = 4, A=“at least 2 H” and B=“at least 2 C” are
mutually exclusive events.
20/33
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes
Calculating with probabilities
Bayes Theorem
Introduction in Probability
Probability space
Sample Space & σalgebra
Probability Measure
Calculating with probabilities
Properties of the Probability Measure
Conditional Probability
Independence
Bayes Theorem
Counting
Counting Principles
Computing Probabilities
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes
Calculating with probabilities
Bayes Theorem
Bayes Theorem
Bayes Theorem: Suppose E1 , E2 , . . . represent a complete
partitioning of the sample space Ω, then for any nonempty
event A ∈ F, we have:
Pr (A Ek ) · Pr (Ek )
,
Pr (Ek A ) = P∞
j=1 Pr (A Ej ) · Pr (Ej )
for any k = 1, 2, . . ..
21/33
a.01 · 0.5 = 0.005 Pr (C ) = Pr (C L)·Pr (L)+Pr (C M)·Pr (M)+Pr (C H)·Pr (H) = 0.5 + 0.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Bayes Theorem Example An insurance company classifies its policyholders according to three risk classes: L (low risk). For each of the risk classes.02 · 0.02 for M.3 + 0.04 · 0.04 for H. 0. Solution: Let C = “be the event that there is a claim”. where Pr (L ∩ C ) = Pr (C L) · Pr (L) = 0. and 0. what is the probability that it is from a L (low risk) policyholder? 22/33 a.01 · 0. .019 (using LTP). M (medium risk) and H (high risk). Question: If a claim occurs. The proportion of H policyholders is 20% and the proportion of L policyholders is 50%. the probability of a claim is 0. then we have: ) Pr (LC ) = Pr(L∩C Pr(C ) = 26%.2 = 0.01 for L.
what is the probability that it is from a M (medium risk) policyholder? c.2·0. . Pr(C M)·Pr(M) ) = Pr (MC ) = Pr(M∩C Pr(C ) = Pr(C ) c.3·0.019 0.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Calculating with probabilities Bayes Theorem Example (cont.04 0.02 0. = 42%. what is the probability that it is from a H (high risk) policyholder? b. Solution: Similar to a.) b.019 = 32%. Question: If a claim occurs.. Question: If a claim occurs. and b. Solution: Similar to a. ) Pr(C H)·Pr(H) Pr (HC ) = Pr(H∩C = Pr(C ) = Pr(C ) 23/33 0..
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Counting Principles Introduction in Probability Probability space Sample Space & σalgebra Probability Measure Calculating with probabilities Properties of the Probability Measure Conditional Probability Independence Bayes Theorem Counting Counting Principles Computing Probabilities .
nm . · nm . . . . . Permutation: The number of ways of arranging n distinct objects is given by: n! ≡n · (n − 1) · (n − 2) · . S2 . . · 2 · 1 n! =n · (n − 1)! 0! ≡1. . . Sm are m sets with respective number of elements n1 . . . n2 . 24/33 . .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Counting Principles Counting Principles Multiplication Rule: Suppose S1 . The number of ways of choosing one element from each set is given by: n1 · n2 · . . .
. ≡ r ! · (n − r )! r Multinomial: The number of ways that n objects can be grouped into r classes with nk in the k th class. n2 . where r P k = 1. . where r ≤ n. . is given by: k=1 n n1 . nr 25/33 ≡ n! . n1 ! · n2 ! · . . .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Counting Principles Combination: The number of ways of choosing r objects from n distinct objects. . is given by: n n! . . r and nk = n. . 2. . · nr ! . .
and 8 flights daily from Honolulu to Los Angeles. with n2 = 8. Let: S1 = “Flight from Sydney to Honolulu”. The airline offers no direct flight from Sydney to Los Angeles. how many different flight arrangements can the airline offer from Sydney to Los Angeles? Solution: Use the multiplication rule. 26/33 . with n1 = 6. S2 = “Flight from Honolulu to Los Angeles”. If the flights are to be made on separate days. then we have n1 · n2 = 6 · 8 = 48 different flight arrangements.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Counting Principles Example Question: An airline has 6 flights daily from Sydney to Honolulu.
Solution: We use combinations & multiplication. there are no additional restrictions. two of the chosen academics must be the two female members of the group of 8. possible number of ways: to8choose n = = 70. possible number of n 5 ways: r = 2 = 10.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Counting Principles Question: A committee is to consist of 4 academics and 2 practitioners. How many ways can you form a committee if: a. For the academics (“S1 ”) we have n = 8 distinct ones and we need r = 4 academics. r 4 For the practitioners (“S2 ”) we have n = 5 distinct ones and we need to choose r = 2 practitioners. 27/33 a. to be selected from a larger group of 8 academics and 5 practitioners. . b. Total ways of forming a committee: n1 · n2 = 70 · 10 = 700.
28/33 Note: probability of having two female academic members is: 150/700 = 3/14 ≈ 0. Solution: We use combinations & multiplication again.For the male academics (“S2 ”) we have n = 6 distinct ones and we need to choose r 6= 2 male academics. possible n number of ways: r = 2 = 15.For the female academics (“S1 ”) we have n = 2 distinct ones and we need to choose r 2= 2 female academics.214 . Total ways of forming a committee: n1 · n2 · n3 = 1 · 15 · 10 = 150. possible n number of ways: r = 2 = 1. possible number of ways: nr = 52 = 10. .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Counting Principles b.For the practitioners (“S3 ”) we have n = 5 distinct ones and we need to choose r = 2 practitioners. . .
. with n3 = 4.class three be “letter s”. n4 29/33 different letter arrangements. with n4 = 2. hence there are n n! = 34650 = n1 ! · n2 ! · n3 ! · n4 ! n1 .class two be “letter i”. with n2 = 4. Note: we have n = n1 + n2 + n3 + n4 = 11.class four be “letter p”. n3 . with n1 = 1. . . using all the letters? Solution: Let .class one be “letter m”.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Counting Principles Question: How many different letter arrangements can be obtained from the letters of the word mississippi. . n2 . We use multinomial.
For the last project we need to choose r = from 4 actuaries the n = 7 left distinct actuaries. hence nr = 12 = 220 ways. How many different ways are there to assign the actuaries to the projects? 30/33 Solution: We have 15 distinct actuaries. 3 For the third project we need to choose r = 2 actuaries from n 9 the n = 9 left distinct actuaries. 000. hence nr = 15 3 = 455 ways. The company currently has 15 actuaries. . and the other requires 4 actuaries. Use combinations.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Counting Principles Question: An actuarial consulting company has four projects to do. For the second project we need to choose r = 3 actuaries from the n = 12 left distinct actuaries. hence nr = 74 = 35 ways. The total number of ways to assign the actuaries to the projects is 455 · 220 · 36 · 35 = 126. hence r = 2 = 36 ways. For the first project we need to choose r= 3 actuaries from n = 15 distinct actuaries. one requires 2 actuaries. Two of the projects require 3 actuaries. 126. of which 5 are females.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Computing Probabilities Introduction in Probability Probability space Sample Space & σalgebra Probability Measure Calculating with probabilities Properties of the Probability Measure Conditional Probability Independence Bayes Theorem Counting Counting Principles Computing Probabilities .
31/33 Solution: Let M = “at least one matching pair among 6 socks”.3467. N Question: A drawer contains 10 pairs of socks. 16 14 12 10 Pr (M c ) = 1 · 18 19 · 18 · 17 · 16 · 15 = 0.6533. where M c = “no matching pair among 6 socks”.3467 = 0. Hence. If 6 socks are taken at random and without replacement. We have Pr (M) = 1 − Pr (M c ).ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Computing Probabilities Computing Probabilities If the elements of the sample space Ω all have equal probability and assuming: (a) there are N elements in Ω and (b) the event A can occur in any of n mutually exclusive ways. Pr (M) = 1 − Pr (M c ) = 1 − 0. compute the probability that there is at least one matching pair among these 6 socks. . then: n Pr (A) = .
5 4 Pr (MMFF ) = . {FMMF }.e. Number of combinations: E = {{MMFF }. 5 8 = 0. {MFMF }. What is the probability that the committee formed has both sexes equally represented? Solution: Let E = “sexes are equally represented in the committee”..0758.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Computing Probabilities Question: A committee of four individuals is to be formed from a group of 5 males and 6 females. . {FMFM}. {FFMM}}. n = 4 and r = 2: nr = 42 = 6.0758 = 5 11 . {MFFM}.e. Each combination has equal probability.. . = Pr (FFMM) = 11 · 10 · 96 · Combing we have: 5 4 Pr (E ) = 42 · 11 · 10 · 32/33 6 9 · 5 8 = 6 · 0. . i. i.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Video Lecture Notes Counting Computing Probabilities Odds and probabilities Sometimes. the odds are a to b that an event will occur. For example. the odds that E will occur is: Pr (E ) Pr (E ) = . c Pr (E ) 1 − Pr (E ) Often. That is. odds are quoted in terms of positive integers. a+b . provided neither probability is zero. Then they mean that the probability it will occur is: Pr (E ) = 33/33 a . if E is the event. The odds that an event will occur is the ratio of the probability that the event will occur to the probability it will not occur. we get confused between odds and probabilities.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 ACTL2002/ACTL5101 Probability and Statistics c Katja Ignatieva School of Risk and Actuarial Studies Australian School of Business University of New South Wales k.au Week 1 Week 3 Week 4 Probability: Week 2 Week 6 Review Estimation: Week 5 Week 7 Week 8 Week 9 Hypothesis testing: Week 10 Week 11 Week 12 Linear regression: Week 2 VL Week 3 VL Week 4 VL Video lectures: Week 1 VL Week 5 VL .edu.ignatieva@unsw.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 101/144 .
ACTL2002/ACTL5101 Probability and Statistics: Week 1
Introduction
Course introduction
Moments and measures of dispersion
Introduction
Course introduction
Introduction in probability
Exercise
Mathematical methods
Random variables and distributions
Measures of location: probability distributions
Measures of dispersion
r th central/noncentral moments
Generating functions
Summary
Summary
ACTL2002/ACTL5101 Probability and Statistics: Week 1
Introduction
Course introduction
Course overview
Week 1
Week 24
Week 56
Week 79
Week 1012
102/144
General introduction in probability.
Distribution functions:
 univariate & multivariate special distributions;
 joint distributions;
 functions of distributions.
Parameter estimation.
Hypothesis testing.
Linear regression.
ACTL2002/ACTL5101 Probability and Statistics: Week 1
Introduction
Course introduction
This week
Describing distribution using measures of location and
dispersion:

mean;
variance;
skewness;
kurtosis.
Calculating these measures:
 central moments;
 noncentral moments;
 generating functions.
103/144
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Introduction in probability Exercise Moments and measures of dispersion Introduction Course introduction Introduction in probability Exercise Mathematical methods Random variables and distributions Measures of location: probability distributions Measures of dispersion r th central/noncentral moments Generating functions Summary Summary .
b. 3}. {1. two. 1. {0. {3}. {2. 4}. 3}. Give the σ−algebra. 2}. {1. Solutions: a. {2}. or no claim in a given quarter. 1. b. {0. 3}. 1. 2}. 2. 2. {1}. {{0}. {0. E ∈ {{0}. Ω = {0. c. {1. three. {3}. {2}. 2. 2. 3}. 3}. 2. 3}. . 104/144 c. {0. 2}. Give the sample space. 1. 3}. {0. {0. ∅}. {1}. ∅}. 3}. 3}. 2}. 3}. {0. 3}. 1}. 2}. 3}. {1. it is not possible to issue more than three claims. {0. {0. {0. {2. {0. State all possible events. 2. 2}. 3}. {0. 1. 1. 1. 2. For simplicity. {0. {1. {1. {0.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Introduction in probability Exercise Sample spaces An insurer offers health insurance. 3}. Questions: a. An individual can have file either one. 1}.
and the law of total probability. b. Questions: a. and Bayes theorem find the probability of odd number of claims given that the number of claims is two or three. Pr(C = 2) = 8/90. Using a. Pr(C = 1) = 0.8. Pr (even) = 8/9. Pr (odd) = 1/9.. c. Are “odd number of claims” and “two or three claims” independent? d.1.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Introduction in probability Exercise Calculating with probabilities We have the following probabilities: Pr(C = 0) = 0. and Pr(C = 3) = 1/90. Same for even number of claims. Using a. What is the probability of 2 or 3 claims given that there are odd number of claims. 105/144 . find Pr(C ≥ 2).
c. 3}) = 1/10 = Pr ({2. 3}C = {0.3}) = = 1/10.3})·Pr(C ={1. Pr (C = {1. Pr (C = {2. 3}C = {1. 3}C = {2. and Pr (C = {2. 3}) = Pr(C ={2.3}) Pr(C ={2.3}C ={0. 3} odd) · Pr (odd) + Pr ( {2. 2}) = Pr(C ={2}) (Pr(C ={0})+Pr(C ={2})) = 1/10. d. Pr ({2. 3}) = Pr ( {2. 3} even) · 1 1 1 Pr (even) = 91 · 10 + 89 · 10 = 10 . 3}C = {1. Pr (C = {2. 3}). 3}) = Pr(C ={3}) (Pr(C ={1})+Pr(C ={3})) Pr(C ={2.3}C ={1.3}) Pr(C ={1.3})·Pr(C ={1.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Introduction in probability Exercise Solutions: a.3}C ={1.2}) 1 1 1 10 · 9 1 1 1 8 = 9.2})·Pr(C ={0. thus independent.3}∩C ={1. b.3})+Pr(C ={2. 10 · 9 + 10 · 9 106/144 = .
Comment on your results. number of ways choosing objects: 20 = 15. r1 = 15. r3 = 3.504. r = 5. Thus. Use Combination. number of 20! = 155. Use Multinomial. probability is 1/15. 2 quarters with 1 claim and 15 quarters without a claim. . 5 b. 040. 504. Questions: a. Solutions: 107/144 a.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Introduction in probability Exercise Counting principles The history of an individual’s claiming record is had he had 3 quarters with 2 claims. r2 = 2.040. What is the probability that the insured had first 15 quarters without a claim and then 2 quarters with one claim and then 3 quarters with two claims? c. n = 20. What is the probability that the insured had first 15 quarters without a claim and then 5 quarters with at least one claim? b. n = 20. Thus. probability is ways choosing objects: 15!·2!·3! 1/155.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Random variables and distributions Moments and measures of dispersion Introduction Course introduction Introduction in probability Exercise Mathematical methods Random variables and distributions Measures of location: probability distributions Measures of dispersion r th central/noncentral moments Generating functions Summary Summary .
The survival function of X is defined by: SX (x) = 1 − FX (x) = Pr(X > x). That is. It is a mapping from the sample space Ω to the set of real numbers <.f. for all x. of X is defined by: FX (x) = Pr (X ≤ x) .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Random variables and distributions Random variables and distributions A random variable is generally a quantity X whose value depends on the outcome of a random experiment. The cumulative distribution function.. 108/144 .d. X : Ω → <. abbreviated c.
Types of Random Variables: 109/144 . →0 3.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Random variables and distributions Distribution function Properties of a distribution function: 1.e. FX (+∞) = 1. lim+ FX (x + ) = FX (x). that is for all x. FX (x1 ) ≤ FX (x2 ) whenever x1 ≤ x2 . 4. .. FX (·) is rightcontinuous.continuous. FX (·) is a nondecreasing function.mixed. i.discrete. . . FX (−∞) = 0. 2.
Not rightcontinuous: C. Not F (∞) = 1: 10 E. Not F (−∞) = 0: D. Not nondecreasing: A.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Random variables and distributions Which one is a distribution function? 1.6 0.2 1 A B C D E F F(x) 0.2 0 −10 110/144 −5 0 x 5 Which one is/are distribution functions? Solution: B and F. .4 0.8 0.
m.f.d. k=0 p.) is defined by: fX (x) = Note that: p. satisfies ∞ P ∂ FX (x) .f.f. 111/144 . ∂x pX (xk ) = 1. requires Rthe rightcontinuous property. satisfies −∞ fX (x) dx = 1.m. ∞ p.f.d.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Random variables and distributions The probability mass function (p.f. The probability density function (p.) is defined by: pX (xk ) = Pr (X = xk ) = FX (xk ) − FX (xk−1 ) .m.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of location: probability distributions Moments and measures of dispersion Introduction Course introduction Introduction in probability Exercise Mathematical methods Random variables and distributions Measures of location: probability distributions Measures of dispersion r th central/noncentral moments Generating functions Summary Summary .
with pdf: 1.v. The expected value of X is: X x · pX (x) . with pdf pX (if discrete) or fX (if continuous). ZX µX = E[X ] = ∞ x · fX (x) dx. −∞ if X is continuous.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of location: probability distributions Measures of location: probability distributions Let X be a r. Solution:R 1 R1 ∞ E [X ] = −∞ x · fX (x)dx = 0 x · fX (x)dx = 12 x 2 0 = 12 .v. . else. 112/144 Exercise: Calculate E [X ] when X is a r. all x if is discrete. fX (x) = 0. if 0 ≤ x ≤ 1.
1. X h (x) · pX (x) . 9 and zero otherwise. 113/144 . 1 Exercise: Let X be a r. Let X be a r.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of location: probability distributions Mathematical expectation of functions of a r. ZX E[h(X )] = ∞ h (x) · fX (x) dx. 2. . with pdf pX (if discrete) or fX (if continuous) and let h(X ) be a realvalued function. . −∞ if X is continuous. .v. Solution: P P 1 E [h(X )] = all x h(x) · pX (x) = 9x=0 x 2 10 = 28.v.v. .5. with pdf: pX (x) = 10 for x = 0. let h(X ) = X 2 . all x if is discrete. Calculate E [h(X )].
E[X · Y ] = E[X ] · E[Y ].: (* using X and Y independent) X X E[mX + b] = x (mx + b) pX (x) =m x (xpX (x)) + b =mE[X ] + b XX X E[X + Y ] = (x + y ) pX . y ) = E[X ] + E[Y ] x. b ∈ <.y = X x xpX . . y ) = x.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of location: probability distributions Properties of the expected value operator: Let X and Y be random variables. E[X + Y ] = E[X ] + E[Y ].Y (x.y = y XX x (x · pX (x)) · X y (x · pX (x)) · (y · pY (y )) y (y · pY (y )) = E[X ] · E[Y ]. y ) + ypX .Y (x.Y (x. we have: E[mX + b] = mE[X ] + b.Y (x.Y (x. only if X . y ) x. y ) + X x.y E[X · Y ] = X X x 114/144 ypX . and m.v. y ) = xpX .Y (x. Proof for discrete r.y ∗ (x · y ) · pX . Y independent.
115/144 d. Question: Assumes that there are 150 insured. The claim value is $1. The required capital will depends on investment return.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of location: probability distributions Example An insurance company offers motor vehicle insurance. 20% and a decrease of 5% occurs w. The probability that an insured files a claim is 20%. Question: What is the expected value the insurer needs to cover the claims? . What is the expected number of claims. Assume that the insured files not more than one claim. Question: What is the probability mass function? b. 15%. a.p. Question: What is the expected number of claims. c.p. The claims will be paid at the end of the year.000 for each claim. A 10% increase in asset value occurs w.
2. .2 · 1 + 0.95 =1000 · 0.2.65 · 1 + 0.8 · 0 = 0. d. Let Y be the random variable of the asset value.65.1.15. E [150 · X ] = 150 · E [X ] = 150 · 0. if y = 1.1 0. 0. if x = 1. P b. We have: 0.89. and zero otherwise. ∗ E [(150 · X ) · (1000 · Y )] =E [150 · X ] · E [1000 · Y ] 116/144 =30 · 996. if y = 1/0. if x = 0.2 = 30. if y = 1/1.99689 = 996.2 · + 0.95.8.15 · 1. 0. E [1000 · Y ] =1000 · E [Y ] 1 1 =1000 · 0.89 = 29. c.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of location: probability distributions Solution 0.2. pX (x) = and zero otherwise. pY (y ) = 0. E [X ] = all x pX (x) · x = 0. a. 907.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Moments and measures of dispersion Introduction Course introduction Introduction in probability Exercise Mathematical methods Random variables and distributions Measures of location: probability distributions Measures of dispersion r th central/noncentral moments Generating functions Summary Summary .
One measure of dispersion: Mean absolute deviation: MAD(X ) = E [X − µX ] .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Measures of dispersion X − µX gives deviation from the expected value. Question: Can we use expected deviation as measure of dispersion: E [X − µX ]? Solution: No. . 117/144 Note: the mean absolute deviation minimized when µX is the median of the distribution. we have that: E [X − µX ] = E [X ] − E [µX ] = µX − µX = 0.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Variance of X MAD is not easily to calculate. Another measure of dispersion: Let X be a random variable. 118/144 . The function E (X − α)2 is minimized when α = µX . the variance is given by: σ 2 = Var (X ) =E (X − µX )2 =E X 2 − µ2X . and has not the nice properties (not related to moments of distribution). The standard deviation of X is given by: p σ = Var (X ).
c ∈ <. Var (X + Y ) = Var (X ) + Var (Y ) .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Properties of the variance: Let X and Y be random variables. b. we have: Var (c) = 0. and m. Proof: Var (c) =E (c − µc )2 = E (c − c)2 = 0 Var (mX + b) =E (mX + b − µmX +b )2 = E (mX + b − mµX − b)2 =m2 · E (X − µX )2 = m2 · Var (X ) h i Var (X + Y ) =E ((X − µX ) + (Y − µY ))2 =E (X − µX )2 + (Y − µY )2 + 2(X − µX ) · (Y − µY ) ∗ =Var (X ) + Var (Y ) . Var (mX + b) = m2 · Var (X ) . only if X . Y independent. . 119/144 * using independence between X and Y .
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Exercises The claims of the motor vehicle insurance are itself stochastic.. Question: Find the expected value of the claims size. 120/144 . Question: Find the variance of the claims size. Question: The price of an insurance contract is the expected value plus half the standard deviation. but now for the 150 contract together. c. if x < 0. fX (x) = λ 5 · exp (−λ · x) . Question: Same as c. d. Find the price of the MVI contract. a. pX (0) =0. if x > 0.8. The distribution of the claim value of an insured is: 0. b.
E [X ] = 0 · 0.22 λ . Price = E 121/144 · (−1) = ∞ λ exp (−λ · x) · x 2 dx E X 2 =02 · 0. 1 5·λ .5 · Var ( i=1 Xi ) = √ p 3· 150 150·E [X ]+0. hP 150 i=1 √ 913. .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Solution R∞ a.5 · p Var (X ) = 1 5·λ + 0.8 + 0 λ5 exp (−λ · x) · xdx i∞ h = λ5 · 0 − · (−λ · x − 1) = 0 + λ5 · exp(−λ·x) (−λ)2 c. 0 1 λ2 = 1 2·λ .5 · 3 5·λ q i P150 Xi + 0.5· 150 · Var (X ) = 150 +0. = · 0− 3 5 −λ 5 · λ2 2 1 9 2 Var (X ) =E X 2 − (E [X ]) = 2 − = .5· 5·λ 5·λ = d. 5λ 25λ2 25λ2 Z b.5 λ ≈ 30. Price = E [X ] + 0.8 + 5 0 2 ∞ λ x 2x 2 =0 + · exp(−λ · x) · − + 5 −λ (−λ)2 (−λ)3 0 2 2 λ = .
= sign(m) · γX . if the distribution of X is symmetric. The skewness of X is given by: " # X − µX 3 .d. if X . . γX = E σX Coefficient of skewness: α3 = E X3 (E [X 2 ])3/2 . Y are independent. = γX ·σX3 +γY ·σY3 σX3 +Y . of X + Y ): γX γm·X +b γX +Y 122/144 = 0. Properties of skewness (σX +Y is the s.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Skewness Let X be a random variable.
. Question: Distribution ii) has a positive skewness. c. 0 x distribution iii) 1 −4 x −1 b. Question: Distribution i) has a positive skewness. Question: Distribution iii) has a smaller skewness than distribution ii).ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Which statements are true? Note: E[X ] = 0 and σ = 1 for i) and ii) and E[X ] = −4 and σ = 3 for iii). Solution: False: both have the SAME skewness. Solution: False: skewness is 1. a. Solution: True: skewness is 1. c. b. Pr(X=x) Pr(X=x) Pr(X=x) distribution i) 1/2 1/3 1/6 0 1/2 1/3 1/6 0 −2 1/2 1/3 1/6 0 −10 123/144 −1 0 x distribution ii) 2 a.
9 cumulative density function probability density function 0.6 0.4 0. v2= 40 2 x Question: Positive/negative skew (skewed to the right/left)? Solution: Positive skew (skewed to the right).2 0.2 0.3 0.d.7 0.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Snedecor‘s F p.5 0. 4 .f.1 0 0 124/144 2 x 4 0 0 v1= 20.8 0.f.6 0. v = 40 1 1 2 0. Snedecor‘s F c. 1 v = 20.d.8 0.4 0.
1 a= 4.3 0. 1 .8 1.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Beta p.5 x 1 0 0 0.9 cumulative density function probability density function 0.2 0. Beta c. b= 2 0 0 125/144 0.5 0.5 0.1 a= 4.4 0.5 1 0. b= 2 2 0.d.f.6 0.5 x Question: Positive/negative skew (skewed to the right/left)? Solution: Negative skew (skewed to the left).f.d.7 0.
. Kurtosis coefficient: α4 = 126/144 E X4 (E [X 2 ])2 .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Kurtosis Let X be a random variable. σX Measures the peakedness (positive) or flatness (negative) of a random variable. The (excess) kurtosis of X is given by: " # X − µX 4 κX = E − 3.
a. 1/18 −3 Pr(X=x) Pr(X=x) distribution i) 8/9 0 x distribution ii) 1/8 0 x distribution iii) 2 3/4 1/8 −10 127/144 a. Solution: False: both have the SAME excess kurtosis. 3/4 −2 Pr(X=x) 3 −4 x 2 c. c.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Which statements are true? Note: E[X ] = 0. . σ = 3 and γ = 0 for iii). Question: Distribution ii) has a smaller excess kurtosis than distribution i). Solution: True: excess kurtosis for i) is 6 and for ii) is 1. σ = 1. Question: Distribution iii) has a smaller excess kurtosis than distribution ii). and γ = 0 for i) and ii) and E[X ] = −4.
negative: blue excess kurtosis.3 0.8 0.1/√ 2) 0.1 0 −2 0 x 2 0 −2 0 x Question: Positive/none/negative excess kurtosis? Solution: Positive: green. none: red.1 0. Var (X ) = 1 and γX = 0 for all distributions.d.4 0.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion p. 128/144 Note: E[X ] = 0. 1 0.7 0.4 0.9 cumulative density function probability density function 0.6 0.6 0.2 0. c.d.√ 3) Normal(0.5 0.f.3 0.2 Uniform(−√ 3.7 0. 2 .f.1) Laplace(0.5 0.
a. II. The regulator requires that insurers have enough reserves in order to reduce the probability of ruin to 0. with mean $50 and standard deviation of $10. with mean $100 and standard deviation of $5 and a skewness of $5. Would the insurer company prefer claims with a distribution I. with mean $100 and standard deviation of $5. II.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Exercises An insurance company is pricing its policies using the standard deviation pricing principle. Holding capital is a cost. b. with mean $100 and standard deviation of $5 and a skewness of $5. .5%. Would the insurer company prefer claims with a distribution 129/144 I.
with mean $100 and standard deviation of $5. with mean $100 and standard deviation of $5 and a skewness of $0 and a kurtosis of $5. d. Would the insurer company prefer claims with a distribution I. a positive skewness and a negative kurtosis. a negative skewness and a positive kurtosis. d. Solution: a. II. . II.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Measures of dispersion Exercises c. c. with mean $100 and standard deviation of $5. Cannot say from the question. Would the insurer company prefer claims with a distribution I. b. with mean $100 and standard deviation of $5 and a skewness of $0 and a kurtosis of $2. 130/144 II. II. II.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods r th central/noncentral moments Moments and measures of dispersion Introduction Course introduction Introduction in probability Exercise Mathematical methods Random variables and distributions Measures of location: probability distributions Measures of dispersion r th central/noncentral moments Generating functions Summary Summary .
all x if is discrete. −∞ if X is continuous. 131/144 .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods r th central/noncentral moments r th central moments Let X be a r.. Relation central & noncentral moments: r h i X r r ∗ E [(X − µX ) ] = · E X k · (−µX )r −k . r ZX E [(X − µX ) ] = ∞ (x − µX )r · fX (x) dx. k k=0 * using binomial expansion.v. the rth central moment is given by: X (x − µX )r · pX (x) .
2 + (0 − 0. Consider the Motor Vehicle insurer from slide 115. ..ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods r th central/noncentral moments (Noncentral) Moments Let X be a r.2 · 12 + .08 · 02 ) − 0.v. on slide 116.2)2 · 0. a. all x if is discrete. Recall the mean of the claim size from b. Solution: Var (X ) = E hX 2 − (E [Xi ])2 = (0. Question: Find the second central moment for an insured. the rth (noncentral) moment is given by: X x r · pX (x) . ZX E [X r ] = ∞ x r · fX (x) dx. −∞ if X is continuous.22 = 0.2)2 · 0.8 = 0.16. a.16 or 132/144 E (X − µX )2 = (1 − 0.
2 · (1 − 0.8 = 0.2 + 02 · 0.1024 − 0. Solution: Start with the third central moment: h i E (X − µX )3 =0.008=0.5. Solution: h i E (X − µX )3 = E X 3 − 3µX X 2 + 3µ2X X − µ3X = E X 3 − 3µX E X 2 + 3µ2X E [X ] − 3µ3X =0. h i # " 3 3 E (X − µ ) X 0.2)3 =0.163/2 c. 133/144 . 2 the second 2 c.096 (X − µX ) = = = 1.024 − 0.0064=0. γX =E 3 3 σ σ 0. Question: Find and third noncentral moments. d. d.2. Question: Find the skewness using the third central moment.2 + 03 · 0.2)3 + 0.2 and 3 E X = 13 · 0.2 − 0.8 · (0 − 0.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods r th central/noncentral moments Exercises b. Solution: E X = 1 · 0.12 + 0.096.096.8 = 0. b. Question: Find the skewness using only noncentral moments.
e... every distinct distribution function has an unique set of moments. Theorem: The complete set of all moments is required to characterize an arbitrary distribution.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods r th central/noncentral moments Knowledge of the mean. with x = X · t. Proof: Using Taylor series: t2 t3 1 + E [X ] · t + E X 2 · + E X3 · + . = E 1 + X · t + X2 · 2! 3! h i ∗ = E e Xt . .. .. 2! 3! t2 t3 + X3 · + . . skewness and kurtosis can provide useful knowledge of the distribution... * using exp(x) = 1 + x + 134/144 x2 2! + x3 3! + . variance. i.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions Moments and measures of dispersion Introduction Course introduction Introduction in probability Exercise Mathematical methods Random variables and distributions Measures of location: probability distributions Measures of dispersion r th central/noncentral moments Generating functions Summary Summary .
. Mm·X +b (t) = MX (m · t) · e b·t .g. The moment generating function of a r. Y are independent. 2.. MX +Y (t) = MX (t) · MY (t). 135/144 . only if X . X is defined as: h i MX (t) =E e X ·t t3 t2 + E X3 · + . =1 + E [X ] · t + E X 2 · 2! 3! Properties of m. for constants m.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions Moment generating function (mgf) of a r.v. 1. b.f..: (r ) MX (0) = E [X r ] .. . .v. 3. . for r = 0.
using the relationship: .f.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions Use of moment generating function Relation m.f. k! k=0 Generating noncentral moments using the m. and noncentral moments: we can write the m.g.f.f.g.g.g.: we can generate the moments from the m. as an infinite series of the moments as follows: ∞ h i X tk MX (t) = E e X ·t = µk · .
.
(r ) r . µr = E [X ] = M X (t).
t=0 Function of random variables: week 4. 136/144 .
then it uniquely determines the probability distribution. exists for t in an open interval containing zero.f. Set t = 0 and you get the desired result.g. 137/144 .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions Proof: To prove the above result. consider the continuous case (similar proof in the discrete case): Z ∞ ∂ r h X ·t i ∂r (r ) MX (t) = r E e e x·t · fX (x) dx = r ∂t ∂t −∞ Z ∞ r ∂ x·t = e · fX (x) dx ∂t r Z−∞ ∞ = x r · e x·t · fX (x) dx −∞ h i =E X r · e X ·t . Remark: If the m.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions Example: Consider the MVI from slide 115. is: h i X MX (t) = E e Xt = pX (x) · e xt = 0.f. all x The first k noncentral moments are: .g.2 · e 1·t + 0.8 + 0. The m.2e t .8 · e 0·t = 0.
.
∂MX (t) .
.
2e t . = 0.
t=0 = 0.2 E [X ] = .
∂t t=0 h i ∂ k M (t) .
.
.
X .
E Xk = = 0.2e t .
.t=0 = 0.2.
16. h i E (X − µX )4 =E X 4 − 4µX E X 3 + 6µ2X E X 2 − 4µ3X E [X ] + µ4X 138/144 =0.16 + 0. .2 − 0.2 − 0.22 = 0. h i κX =E (X − µX )4 /Var (X )2 − 3 = 0.048 − 0.2.00032 = 0.08192. k ∂t t=0 The kurtosis is given by: Var (X ) =E X 2 − (E [X ])2 = 0.162 − 3 = 0.08192/0.0064 + 0.
Solution: We have: Z ∞ h i λ −λ·x t·x t·X 0·t MX (t) = E e =0. Question: Determine the m. . 5 · (t − λ) 139/144 * note m.f.f. a.8 − . exists if t − λ < 0. b.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions Exercises Consider the Motor Vehicle insurer from slide 120. of the claim size.8 + · e (t−λ)·x 5 · (t − λ) 0 λ ∗ =0.8 + ·e dx 5 0 ∞ λ =0. Question: Use the m.8 · e + ·e · e dx 5 Z ∞ 0 λ (t−λ)·x =0. to determine the skewness.g. a.f.g.g.
Solution: Using MX (t) = 0. Find the noncentral moments using the derivatives: .8 − λ 5·(t−λ) .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions b.
.
.
∂MX (t) .
.
−λ 1 −2 .
E [X ] = = · −1 · (t − λ) .
= .
∂t 5 5λ t=0 .
.
t=0 2 .
∂ MX (t) .
λ 2 .
= · −2 · (t − λ)−3 .
.
= 2 E X2 = .
2 ∂t 5 5λ t=0 .
.
t=0 3 .
∂ MX (t) .
6 −2λ .
E X3 = · −3 · (t − λ)−4 .
.
= 3. = .
3 ∂t 5 5λ t=0 t=0 2 1 9 2 = Var (X ) =E X − (E [X ])2 = 2 − 5λ (5λ)2 25λ2 E (X − µX )3 E X 3 − 3µX E X 2 + 3µ2X E [X ] − µ3X γX = = Var (X )3/2 Var (X )3/2 6 1·2 1·1 6 · 52 2·5 1 3 − 3 · 25λ3 + 2 · 125λ3 = 5λ = − 3 · 3 + 2 · 3 = 122/27. 3 3 (3/(5λ)) 3 3 3 140/144 .
.: .g. .g. is as follows: PY (t) = MY (log(t)) .g. is defined as: ∞ h i X PY (t) = E t Y = pY (i) · t i . 1. and m. .g.The relationship between p.g.f.f. Let Y be an integervalued random variable with Pr(Y = i) = pi for i = 0.f.v.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions Probability generating function (p. 2. the p..f. i=1 Properties of p.) of a r. .f.
.
Probabilities: Pr(Y = r ) = P Y (t). (r ) .
141/144 . .Take the k th derivative: (k) PY (1) = E [Y · (Y − 1) · (Y − 2) · . /r ! t=0 . . · (Y − k + 1)] ≡ µ[k] .
v. CX +Y (t) = CX (t) + CY (t). 5. The cumulant generating function CX (t) for a random variable is given by: ∞ X 1 i · h · κi or MX (t) = e CX (t) .g. 6.f. b. Cm·X +b (t) = CX (m · t) + b · t. (r ) 142/144 Note that CX (0) 6= E [(X − µX )r ] . Y independent. for constants m. CX (t) = log(MX (t)) = i! i=1 where κi = (i) CX (0) is the i th cumulant. .. for r = 4. .g.: κi = i th central moment. .f. X . for r = 2.) of a r. . 3.ACTL2002/ACTL5101 Probability and Statistics: Week 1 Mathematical methods Generating functions Cumulant generating function (c. Properties of c.
ACTL2002/ACTL5101 Probability and Statistics: Week 1 Summary Summary Moments and measures of dispersion Introduction Course introduction Introduction in probability Exercise Mathematical methods Random variables and distributions Measures of location: probability distributions Measures of dispersion r th central/noncentral moments Generating functions Summary Summary .
if X is discrete. all x E [h (X )] = Z ∞ h (x) · fX (x) dx. −∞ Mean: µ = E [X ].ACTL2002/ACTL5101 Probability and Statistics: Week 1 Summary Summary Mathematical Expectation The mathematical expectation of h (X ) is: X h (x) · pX (x) . Central moments: E [(X − µx )r ] refers to the r th central moment. 143/144 . if X is continuous. Moments: µr = E [X r ] refers to the r th (noncentral) moment.
" # X − µX 4 Kurtosis: κX = E − 3 refers to the kurtosis. It Skewness: γX = E σX measures the lack of symmetry in the p.d.f. it uniquely defines a density function. σX It measures the peakedness or flatness of the p. # X − µX 3 refers to the skewness.d.f. .ACTL2002/ACTL5101 Probability and Statistics: Week 1 Summary Summary Dispersion h i Variance: σ 2 = Var (X ) = E (X − µ)2 = E X 2 − µ2 ... Moment generating function: MX (t) = E e X ·t .
It is useful to calculate .
(r ) r moments: µr = E [X ] = M X (t).
. " t=0 144/144 .