Actuarial AdvantEDGE Exam P / CAS 1 Study Manual

George Barnidge, MS, FSA

v1.0

NO RETURNS IF OPENED

Contents
1 Introductory Probability Topics 1.1 Risk and Insurance Concepts . . . . . . . . . 1.2 Summation Notation . . . . . . . . . . . . . . 1.3 Measures of Central Tendency and Dispersion 1.4 Set Notation . . . . . . . . . . . . . . . . . . 1.5 Probability Rules and Related Topics . . . . . 1.6 Counting Methods . . . . . . . . . . . . . . . 1.7 Conditional Probability and Independence . . 1.8 Bayes Rule . . . . . . . . . . . . . . . . . . . 1 1 3 10 17 23 31 39 55 63 63 70 80 83 87 100 109 116 124 132 133 133 145 152 158 168 174 194 200 205

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

2 Discrete Probability 2.1 Introduction to Discrete Probability Distributions 2.2 Expectation Properties . . . . . . . . . . . . . . . . 2.3 The Uniform Distribution . . . . . . . . . . . . . . 2.4 The Hypergeometric Distribution . . . . . . . . . . 2.5 The Binomial Distribution . . . . . . . . . . . . . . 2.6 The Geometric Distribution . . . . . . . . . . . . . 2.7 The Negative Binomial Distribution . . . . . . . . 2.8 The Poisson Distribution . . . . . . . . . . . . . . 2.9 Discrete Transformations . . . . . . . . . . . . . . 2.10 Chapter Appendix . . . . . . . . . . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

. . . . . . . . . .

3 Continuous Probability 3.1 Introduction to Continuous Probability Distributions 3.2 Expectation Properties . . . . . . . . . . . . . . . . . 3.3 The Uniform Distribution . . . . . . . . . . . . . . . 3.4 The Exponential Distribution . . . . . . . . . . . . . 3.5 The Gamma Distribution . . . . . . . . . . . . . . . 3.6 The Normal Distribution . . . . . . . . . . . . . . . . 3.7 Continuous Transformations . . . . . . . . . . . . . . 3.8 Order Statistics . . . . . . . . . . . . . . . . . . . . . 3.9 Chapter Appendix . . . . . . . . . . . . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

. . . . . . . . .

4 Multivariate Distributions 207 4.1 Introduction to Joint Probability Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . 207 4.2 Expectation Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251 4.3 Multivariate Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 5 Related Topics 279 5.1 Linear Combinations of Indendepent Random Variables and the Central Limit Theorem . . . 279 5.2 Normal Approximation of Discrete Probabilities . . . . . . . . . . . . . . . . . . . . . . . . . . 292

6 Practice Exam #1 299 6.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299 6.2 Solutions to Practice Exam #1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305 7 Practice Exam #2 313 7.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 7.2 Solutions to Practice Exam #2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319 8 Practice Exam #3 329 8.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329 8.2 Solutions to Practice Exam #3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 9 Practice Exam #4 345 9.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345 9.2 Solutions to Practice Exam #4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351 10 Practice Exam #5 361 10.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361 10.2 Solutions to Practice Exam #5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367

Total Event Decomposition: Suppose the sample space S is completely partitioned into N mutually exclusive (i.e. non-overlapping) events, with each event labeled Ai (where i = 1, 2, 3, . . . , N ). Further, suppose another event, B , intersects some (or all) of these Ai events. This setup is given in Figure 1.5.

S AN 1 A1 A2 A3 A5 B A4 ... AN

Figure 1.5: Event B interweaving the partitioned sample space S

Figure 1.5 shows that event B can be equivalently expressed as the collection (i.e. union) of its overlap (i.e. intersection) with each of the individual Ai events. Using set notation, this means:

and
N

or (1.13)

B =(B A1 ) (B A2 ) . . . (B AN ) = (B Ai )
i=1

In words, event B happens if. . .

B happens within A1 OR B happens within A2 OR B happens within A3 OR. . . OR B happens within AN

intersection union intersection union intersection intersection

Expressing B in Formula 1.13 may seem long-winded at rst, but it can be a convenient (and necessary) arrangement to solve many problems.

20

Example 1.21 (1 of 153; SOA May 2003 #1) A survey of a groups viewing habits over the last year revealed the following information: i) 28% watched gymnastics ii) 29% watched baseball iii) 19% watched soccer iv) 14% watched gymnastics and baseball v) 12% watched baseball and soccer vi) 10% watched gymnastics and soccer vii) 8% watched all three sports Calculate the percentage of the group that watched none of the three sports during the last year. Solution: Let us dene the following events. Let. . . G = watching gymnastics B = watching baseball C = watching soccer Note that I am using C rather than S for soccer since S is usually reserved to denote the sample space. If each event circle represents watching a sport and since we are asked to nd the percentage who did not watch a sport, then we need to nd the percentage who are classied outside the event circles. For this problem, we recommend starting at the most intersected region and working out to the least intersected region. The Venn diagram region dealing with the intersection of all three event circles: Since 8% refers to those who watched all three sports, we label the innermost intersection with 0.08. The Venn diagram region dealing with the intersection of two event circles: Since 10% watched gymnastics and soccer and since the 8% who watched all three sports is a subset of this 10%, then 10% 8% = 2% watched just gymnastics and soccer. Similar logic shows that 12% 8% = 4% watched just baseball and soccer and that 14% 8% = 6% watched just gymnastics and baseball. The Venn diagram region dealing with the non-intersected event circles: Since 19% watched soccer and since the 8% who watched all three sports and since the 2% who watched just gymnastics and soccer and since the 4% who watched just baseball and soccer are subsets of this 19%, then 19% 8% 2% 4% = 5% watched just soccer. Similar logic shows that 29% 8% 4% 6% = 11% watched just baseball and that 28% 8% 2% 6% = 12% watched just gymnastics. The seven percentages identied in the previous three paragraphs completely ll the event circles in the Venn diagram, summing to 0.08 + (0.02 + 0.04 + 0.06) + (0.05 + 0.11 + 0.12) = 0.48. Since 48% watched some type of sport, then Formula 1.14 shows that 1 48% = 52% did not watch a sport. The Venn diagram shows these percents. S 0.52 0.12 G 0.06 0.08 B 0.11 0.04 C 0.05 0.02

25

1.7

Conditional Probability and Independence

This section introduces some very simple yet powerful probability formulas that will be used throughout the rest of this manual. It is very important that you master these, for they are crucial to passing this exam. Denition 1.23 An unconditional probability is a probability where no additional information or conditions are known or expressed. The unconditional probability that event A happens is notated P (A). Denition 1.24 A conditional probability is a probability where additional information or conditions are known or expressed. The conditional probability that event A happens given that event B happens is notated P (A|B ). The vertical bar | is read as given. Important note: With conditional probabilities, the event before the vertical bar | is the event whose probability we are interested in nding. The event after the vertical bar is the additional information or condition that may (or may not) aect the probability of interest. To determine if a probability is unconditional or conditional, look for the phrase given (if expressed verbally) or the vertical bar | (if expressed symbolically).

Suppose we have the following tree diagram where the rst level expresses the occurrence of event A or its nonoccurrence, A. The second level expresses the occurrence of event B or its non-occurrence, B . B P (A B ) = P (A)P (B |A) A B P (A B ) = P (A)P (B |A) Start B P (A B ) = P (A)P (B |A) A B P (A B ) = P (A)P (B |A) Looking at this picture, to get to the A B event endpoint, we have to rst travel up to A (rather than down to A). Then, given that we are now standing at A, we have to nd the probability that we further travel up to B (rather than down to B ). Therefore, the probability that we observe event A B is the product of an unconditional probability, P (A), and a conditional probability, P (B |A). The same patterns hold for the other three branches. The general theme of this movement is expressed here: P (A B ) = P (A) P (B |A) A rearranged version of Formula 1.21 is given here. P (B |A) = P (A B ) P (A) (1.22) (1.21)

Example 1.26 A sample of 100 drivers was taken in which respondents were asked whether they passed or failed their drivers exam on their rst attempt. The following table shows the results from this sample, broken down by gender. Test outcome Pass Fail 25 35 30 10 55 45

Gender Male Female Total

Total 60 40 100

39

Chapter 3

Continuous Probability
3.1 Introduction to Continuous Probability Distributions

As we saw in the previous chapter, discrete random variables take on a countable number of possibilities, usually the whole numbers. Each of these possibilities has an associated probability, determined by the pmf. In this chapter, we study continuous random variables. Unlike discrete random variables, continuous random variables take on values over an interval and are therefore not countable. Denition 3.1 A generic random variable X is continuous if it takes on values over an interval1 .

Example 3.1 The following are examples of continuous random variables: a) Let X measure the amount of rainfall (in inches) over the next year. Since X can be not only 1 or 2 or 3, etc. of rainfall but also all the values in between, such as 0.71 or 3.14 of rainfall, X is a continuous random variable. b) Let X measure the weight (in pounds) of a person during his next life insurance underwriting process. Since X can be not only 185 pounds or 197 pounds but also all the values in between, such as 186.53 pounds, X is a continuous random variable.

Denition 3.2 A probability density function (pdf), f (x), is a function that assigns probabilities to continuous random variables2 . A pdf can be presented as a formula or graph and is the analog of the discrete pmf3 .

Example 3.2 The following are equivalent versions of the same pdf: As a formula: f (x) = As a graph:
is a loose denition of a continuous random variable but one that should suce for our purposes. is a loose denition of a probability density function but one that should suce for our purposes. 3 A pdf, unlike pmfs, is never presented in table form because continuous random variables do not take on a countable set of possibilities (i.e. they cant be listed).
2 This 1 This 3x 2

where 1 x 3.

133

f (x) 1

In Example 3.2, notice that the area under f (x) above the x axis is 1. Also notice that f (x) 0. These two properties, given here, will be true for any pdf and is analogous to Formula 2.1:

1)

f (x) dx = 1 (3.1)

2) f (x) 0 To calculate probabilities associated with continuous random variables, the main theme is. . . Probabilities are areas!!! Areas are probabilities!!! Recall from your calculus class that integration allows us to calculate areas under curves. Probabilities associated with continuous random variables are calculated by integrating the pdf over the given interval, shown here:

P (a X b ) = f (x)

b a

f (x) dx

The main theme from the previous picture is summarized here:

b

P (a X b ) =
a

f (x) dx

(3.2)

In the previous chapter on discrete random variables, it was common to calculate probabilities associated with a single possibility of the discrete random variable. For example, if X is a discrete random variable, it was common to calculate positive probabilities for expressions such as P (X = 7) or P (X = 0) (just 7 or just 0). This does not translate over to continuous random variables. Recalling, from the continuous perspective, that Probabilities are Areas, and Areas are Probabilities, this next picture shows why probabilities associated with a single constant value, call it a, are 0.

P (X = a) = f (x )

a a

f (x ) d x = 0

(an innitely-thin line has no area and therefore no probability)

134

The main theme from the previous picture is summarized here: P (X = a) = 0 (3.3)

Formula 3.3 may seem somewhat counterintuitive at rst since X = a is clearly within the range of possibilities. But although X = a is a possibility, it has no probability since it has no area (!!!). The answer therefore is 0. Continuous random variables have positive probability only over intervals, not single values. Therefore, with continuous random variables, the probability P (X > a) and P (X a) would have the same answer (since they dier only in the inclusion or exclusion of the single value x = a). The same can be said about P (X < a) and P (X a). This next picture shows how P (a X b) can be expressed as the dierence of two probabilities.

P (a X b ) = f (x )

b a

f (x) dx f (x) =

P (X b ) =

f (x) dx f (x )

P (X a) =

f (x) dx

The previous picture can be expressed from a slightly dierent perspective, given here:

P (a X b ) = f (x )

b a

f (x) dx f (x) =

P (X a) =

f (x) dx f (x)

P (X b) =

f (x) dx

The main theme from the previous two pictures is summarized here: P (a X b) = P (X b) P (X a) = P (X a) P (X b) (3.4)

For continuous distributions, Formula 3.4 is true regardless of the inequality signs that appear in P (a X b). For example, one could apply Formula 3.4 even if the original problem was P (a < X b) or P (a X < b) or P (a < X < b). These all give the same answer due to Formula 3.3. Although the logic of Formula 3.4 can also apply to discrete random variables, the inequality sign can impact the answer. Therefore, more care must be taken when applying Formula 3.4 to a discrete random variable. Denition 3.3 The cumulative distribution function (CDF) of a random variable X , denoted FX (x), is the probability that X takes on a values less than or equal to x. Symbolically, this means that FX (x) = P (X x).

The next picture shows the graphical relationship between the pdf and the cdf.

135

P (X t ) = f (x )

f (x) dx = FX (t) F X (x ) 1

t the density function, f (x)

the cumulative distribution function, FX (x)

In the previous picture, notice that the graph of FX (x) (the right graph) begins with a height of 0. This happens because the graph of f (x) (the left graph) has not yet accumulated any area when t is to the left of the leftmost possible x value. Also notice that FX (x) approaches a height of 1 as t moves to the right. This happens because the total accumulated area under f (x) is 1 (see Formula 3.1) once t moves to the far right. Finally, notice that FX (x) never decreases, which happens because f (x) 0 (see Formula 3.1). In summary, cdfs from continuous distributions always begin at 0 and end at 1. These ideas are summarized here and are the analog of Formula 2.2:
x

FX (x) = P (X x) =

f (t) dt (3.5)

FX () = 0 FX () = 1 Example 3.3 Refer to Example 3.2. a) Find FX (x) using Formula 3.5. b) Check that FX (x) satises the two conditions at the bottom of Formula 3.5. c) Graph FX (x) d) Find P (X 2.5) e) Find P (X 1.75) f) Find P (1.75 X 2.5) g) Find P (X = 2)
x

a) FX (x) = P (X x) =

3t dt = 2

x 1

3t 3 t2 dt = t 2 2 4

=
1

6t t 2 4

=
1

6x x2 5 4

In the previous algebra, notice how the lower bound of integration was changed from (as appears in Formula 3.5) to 1. This is because the lowest value that X can assume in this example is x = 1. b) When x = 1 (the leftmost bound for x), FX (1) should be 0. FX (1) =
6(1)(1)2 5 4

=0

When x = 3 (the rightmost bound for x), FX (3) should be 1. FX (1) =

6(3)(3)2 5 4

1895 4

4 4

=1

c) The graph of FX (x) is given here:

136

F (x) 1

d) P (X 2.5) = FX (2.5) =

6(2.5)2.52 5 4

= 0.9375
0.6094

e) P (X 1.75) = 1 P (X < 1.75) = 1 FX (1.75) = 1

F3.4

6(1.75) 1.752 5 = 0.3906 4

f) P (1.75 X 2.5) = P (X 2.5) P (X 1.75) = 0.9375 0.6094 = 0.3281 g) P (X = 2) = 0 (see Formula 3.3)

Formula 3.5 shows that FX (x) is calculated by integrating f (x). Since integration and dierentiation are inverse processes, then the pdf is calculated by dierentiating FX (x), given here: f (x ) = d FX (x) dx (3.6)

Just as the previous chapter dedicated separate sections to specic brand name discrete random variables (i.e. binomial, geometric), so too will we devote separate sections in this chapter to specic brand name continuous random variables.

End-of-Section Practice Problems

Problem 3.1.1 Let f (x) = 1/10, where 7 x 3. Find. . . a) the median. b) the 85th percentile. Problem 3.1.2 The lifetime X (in years) of a machine has pdf f (x) = 5) = 0.01, nd .
2x 2

exp[(x/)2 ], where x > 0. If P (X >

Problem 3.1.3 The annual demand X for hand soap (in thousands of gallons) for a school district has the pdf f (x) = 2x exp(x2 ), where x > 0. If 3,000 gallons are in stock at the beginning of each year (with no additional amount provided during the year), what is the probability that the soap will run out before the year is over? Problem 3.1.4 Suppose that X assumes the density function f (x) = cx, where 0 x 4 and c is a constant. a) Find the value of c that makes f (x) a pdf. b) Find FX (x). c) Find P (1 X 2) using FX (x). Problem 3.1.5 A grain supplier for farms has a 250-gallon tank that is lled at the beginning of each month. The monthly demand for grain shows a relative frequency behavior that linearly increases up to 100 gallons and then is level between 100 and 250 gallons. If X denotes that monthly demand in hundreds of gallons, the relative frequency

137

3.5

The Gamma Distribution

A generic gamma random variable, X , is often (but not always) used to measure the waiting time for (alpha) observations of some event, where is a measure of the average waiting time between event observations7 . Since waiting time can be any positive number, the range of possible values is X > 0. The symbols and are considered the parameters of the gamma distribution since they are the only values needed to fully dene the probability space. Shorthand notation: For convenience, rather than tediously writing X is distributed as a gamma random variable with parameters and , we will condense this information into the equivalent expression X Gamma(, ). The symbol should be interpreted as is distributed as. Relationship to the negative binomial random variable: Recall from Section 2.7 that a negative binomial trials random variable counts the number of trials needed to observe r successes. Since the gamma random variable measures the waiting time for observations of some event, it can be seen as the continuous analog of the discrete negative binomial trials random variable. The pdf of a gamma random variable X is

f (x) = where

1 x1 exp(x/) () (3.33)

x>0 >0 >0 () = ( 1)! when is an integer

The graph of several gamma pdfs is given here:

f (x)
1

=1 =1

=2 =3 x

Probability calculations involving the gamma random variable will require us to integrate the pdf from Formula 3.33. This requires the use of integration by parts. To save time, the following formulas show, for a given value of , the

7 Compare this to the description of the exponential random variable in Section 3.4. The exponential random variable measures the waiting time for the rst observation whereas the gamma random variable measures the waiting time for observations, where is a positive whole number.

168

integration of the gamma pdf for = 1, 2, 3 For = 1: 1 1 (1) For = 2: 1 2 (2) For = 3: 1 3 (3)
a b b a b a

x11 exp(x/) dx = ex/

b a

x21 exp(x/) dx = ex/ 1 +

b a

(3.34)

x31 exp(x/) dx = ex/

1+

x x2 + 2 2

b a

Example 3.11 Cars arrive at a border crossing at a mean rate of 180 cars every hour. Assume that it is the beginning of the day, and no cars have yet arrived. Find the probability that the border station guard will have to wait longer than 1 minute for the second car to arrive at the station. Solution: Let X measure the waiting time for the second car to arrive. Since X measures the waiting time, we want to express the given rate in terms of time, not cars. Since the given rate is 180 cars every 60 minutes, then this is the same as 60 minutes every 180 cars, or 60/180 = 1/3. In other words, a car arrives, on average, every 1/3 minute; so = 1/3. Since we are waiting until the second car arrives, = 2. Therefore, X Gamma( = 2, = 1/3). We are being asked to nd P (X > 1), given here:
F3.34 where = 2 and = 1/3 =0

P (X > 1) =

3x

(1 + 3x)
1

= lim e
x

3x

(1 + 3x) e3(1) (1 + 3(1)) = 0.1991

The mean and variance of a gamma random variable are: X =E(X ) =

2 X =V(X ) = 2

(3.35)

The moment generating function (mgf) of a gamma random variable is: MX (t) = (1 t) Example 3.12 If MX (t) = (1 7t)15 , nd. . . a) the pdf f (x) b) X = E (X ) 2 c) X = V (X ) Solution: Matching MX (t) with Formula 3.36, we see that = 15 and = 7. Therefore X Gamma( = 15, = 7). a) Formula 3.33 gives f (x) =
1 x151 715 (15)

(3.36)

exp(x/7).

b) Formula 3.35 gives X = E (X ) = 15 7 = 105.

2 c) Formula 3.35 gives X = V (X ) = 15(7)2 = 735.

169

3.6

The Normal Distribution

A generic normal random variable is a continuous random variable that resembles the bell curve. It is the most important distribution (for reasons we will see in Section 5.1 and since a lot of natural phenomenon is approximately normally distributed). The symbols (mu) and (sigma) represent the mean and standard deviation, respectively, of the distribution and are considered the parameters of the normal distribution since they are the only values needed to fully dene the probability space. Shorthand notation: For convenience, rather than tediously writing X is distributed as a normal random variable with parameters and , we will condense this information into the equivalent expression X N(, ). The symbol should be interpreted as is distributed as. The pdf of the normal random variable X is

f (x ) = where

1 1 exp 2 2

<x< << 0<<

(3.39)

The graphs of several normal pdfs are given in Figure 3.1. Notice how the dierent values of (a location parameter) and (a dispersion parameter) aect the curves center and spread, respectively.

W X 0 5

W N(W = 5, W = 0.8) X N(X = 5, X = 1.7) Y N(Y = 15, Y = 1.7) Y 10 15 20

Figure 3.1: pdf of several normal distributions

Denition 3.4 A standard normal random variable is a normal random variable with parameters = 0 and = 1. It is given the special symbol Z . Therefore, we write Z N(Z = 0, Z = 1).

There is nothing mathematically special about a standard normal random variable compared to any other normal random variable. However, the standard normal random variable comes up very often in practice, so it has been assigned the special symbol Z . The graph of the standard normal random variable, Z , is given here. Notice how it is centered at its mean Z = 0 with a spread of Z = 1:

174

f (z )

Z = 0

1 Z = 1

2 Z = 1

3 Z = 1

The parameters of the normal distribution are themselves the mean and variance: E (X ) = V (X ) = 2 (3.40)

The z -chart: Unlike the other continuous distributions, we will never be asked to integrate the normal pdf8 . Instead, we will use a probability chart for a standard normal random variable (from here on simply called the z -chart) to equivalently determine areas under the normal pdf. The z -chart that you will receive on exam day is constructed according to the following picture. You may have used a z -chart that was constructed dierently in your undergraduate statistics class. If so, you will want to become familiar with the version given during the exam (given just before the End-of-Section Practice Problems in this section and at the beginning and end of this manual). The rst column of the z -chart gives you the ones and tenths places of a positive value along the horizontal z -axis. The rst row of the z -chart give you the hundredths place of this same z value. The intersecting cell of the row (coming from the ones and tenths places) and column (coming from the hundredths place) give you the area (i.e. probability) under the z curve, from up to the positive z -value. This can be seen in the following picture.

f (z )

0 P(Z < 0.00) = 0.5 P(Z < +z ) > 0.5

+z

When z = 0.00, for example, the chart gives you the area under the left half of the curve. Since the curve is symmetrical, then this left half is 0.5000 (verify this 0.5000 by looking up z = 0.00 in the top-left cell of the z -chart). Although the domain of the normal distribution is technically (, +), the vast majority of the data falls between 3 z 3, which is why the z -chart does not give probabilities signicantly past z = 3.00. Any area in the extreme tails beyond values given in the z -chart is, for practical purposes, 0. Example 3.14 Let Z N( = 0, = 1). Find P (Z < 1.52).
8 The

pdf of the normal distribution does not have an antiderivative.

175

X +k2 X Z +k2 Z =k2

Most problems are phrased using the X scale but are answered after converting to the Z scale. f (z )dz Equal areas X N(X , X )

f (x)dx =
Z +k1 Z =k1

X +k1 X

where f () is the pdf of the normal distribution

X 2 X X X X

X 1 X

X + 1 X X

X + 2 X

181 Z 1 Z = 1 Z = 1 Z = 0 Z = 1 Z = 1

X = X + Z X

Z N(Z = 0, Z = 1) Z=

X X X

Z 2 Z = 2

Z + 1 Z = 1 Z = 1

Z + 2 Z = 2

Figure 3.2: Normal Curve Transformation

Problem 4.1.14 (77 of 153) A device runs until either of two components fails, at which point the device stops running. The joint density function of the lifetimes of the two components, both measured in hours, is: f (x, y ) = x+y 8 for 0 < x < 2 and 0 < y < 2.

What is the probability that the device fails during its rst hour of operation? Problem 4.1.15 (88 of 153; SOA May 2001 #22) The waiting time for the rst claim from a good driver and the waiting time for the rst claim from a bad driver are independent and follow exponential distributions with means 6 years and 3 years, respectively. What is the probability that the rst claim from a good driver will be led within 3 years and the rst claim from a bad driver will be led within 2 years? Problem 4.1.16 (89 of 153; SOA November 2000 #20) The future lifetimes (in months) of two components of a machine have the following joint density function: f (x, y ) = 6 (50 x y ) 125, 000 for 0 < x < 50 y < 50.

What is the probability that both components are still functioning 20 months from now? Problem 4.1.17 (90 of 153; SOA May 2000 #10) An insurance company sells two types of auto insurance policies: Basic and Deluxe. The time until the next Basic Policy claim is an exponential random variable with mean two days. The time until the next Deluxe Policy claim is an independent exponential random variable with mean three days. What is the probability that the next claim will be a Deluxe Policy claim? Problem 4.1.18 (91 of 153; SOA November 2000 #36) An insurance company insures a large number of drivers. Let X be the random variable representing the companys losses under collision insurance, and let Y represent the companys losses under liability insurance. X and Y have joint density function: f (x, y ) = 2x + 2 y 4 for 0 < x < 1 and 0 < y < 2.

What is the probability that the total loss is at least 1? Problem 4.1.19 (92 of 153; SOA November 2001 #28) Two insurers provide bids on an insurance policy to a large company. The bids must be between 2,000 and 2,200. The company decides to accept the lower bid if the two bids dier by 20 or more. Otherwise, the company will consider the two bids further. Assume that the two bids are independent and are both uniformly distributed on the interval 2,000 to 2,200. Determine the probability that the company considers the two bids further. Problem 4.1.20 (93 of 153; SOA May 2003 #20) A family buys two policies from the same insurance company. Losses under the two policies are independent and have continuous uniform distributions on the interval from 0 to 10. One policy has a deductible of 1 and the other has a deductible of 2. The family experiences exactly one loss under each policy. Calculate the probability that the total benet paid to the family does not exceed 5. Problem 4.1.21 (108 of 153; SOA November 2001 #37) A device containing two key components fails when, and only when, both components fail. The lifetimes, T1 and T2 , of these components are independent with common density function f (t) = et , t > 0. The cost, X , of operating the device until failure is X = 2T1 + T2 . Write an expression for the density function of X for x > 0.

219

y (0, 2) (1, 2) y 1x (0, 1) y =1x

(0, 0)

(1, 0)

Looking at the shaded region, integrating over y rst, we see that y goes as low as the line y = 1 x and is as high as y = 2. Once we integrate over y , we then integrate over x, which goes as far left as x = 0 and as far right as x = 1. P (X + Y 1) = = =
1 4 1 4 1 0 1 0 1 4 1 0 2 (2x 1x

+ 2 y ) d y dx =

1 4

1 0

2xy + 2y 1 y2 2
1 4 5 (13 6

y =2

dx
y =1x

1 2 [2 (1 x)2 ] dx 2x[2 (1 x)] + 2[2 (1 x)] 2 5 2 x 2

+ 3x +

1 2

dx =

1 4

5 3 x 6

1 +3 x2 + 2 x 2

x=1 x=0

3 03 ) + 2 (12 02 ) + 1 (1 0) = 2

17 24

Solution 4.1.19 Let X and Y be the insurers bids. The company considers the two bids further if the dierence between the two bids is less than 20. Mathematically, this happens when |x y | < 20 20 < x y < 20 y > x 20 and y < x + 20. The shaded region over which we need to integrate to nd the answer is given in the following picture. y = x + 20 (2000, 2200) (2180, 2200) (2200, 2200) (2200, 2180) y = x 20 (0, 2000) (2200, 2000)

y (0, 2200) (2000, 2020)

(0, 0)

(2000, 0) (2020, 2000)

(2200, 0)

Since X and Y are each uniformly distributed over the interval (2000, 2200) and since x and y are independent, then Formula 4.7 says that the joint pdf is the product of the marginal pdfs: f (x, y ) = 1 2, 200 2, 000
F3.20

1 2, 200 2, 000
F3.20

1 2002

A quick, geometric approach for the area of the shaded region uses the complement property. For this, recall that 1 the area of a triangle is 2 (base height). P (|x y | < 20) = 1 P (volume above two unshaded triangles)
symmetry of triangles and f (x, y )

= 1 2P (volume above one unshaded triangles)

236

=12

1 (2, 200 2, 020)(2, 180 2, 000) 2

area of triangle volume above one unshaded triangle

1 2002
f (x,y )

= 0.19

Solution 4.1.20 Let X and Y be the loss amounts from the policies with deductibles of 1 and 2, respectively. Since X and Y are each uniformly distributed over the interval (0, 10) and since x and y are independent, then Formula 4.7 says that the joint pdf is the product of the marginal pdfs: f (x, y ) = 1 10 0
F3.20

1 10 0
F3.20

1 100

The following graph will be helpful to determine the region where the total post-deductible benet is at most 5. y (0, 10) (0, 7) A (0, 2) B (0, 0) (1, 0) (6, 0) C x (10, 0) (10, 10) (y 2) + (x 1) = x + y 3 5 y 8 x D

The letters indicate the four regions where the total post-deductible benet is at most 5: Region Region Region Region A: B: C: D: The payment is y 2 because x < 1 (no payment) and y > 2. The payment is 0 because x < 1 (no payment) and y < 2 (no payment). The payment is x 1 because x > 1 and y < 2 (no payment). The payment is (y 2) + (x 1) = x + y 3 because x > 1 and y > 2.

All the other unshaded regions represent situations where the total post-deductible benet exceeds 5. Let us use geometric formulas for the area of a rectangle and triangle to nd the answer: P (post-deductible benet 5) = P (volume above shaded region)
A B C D f (x,y )

1 = (5 1) + (2 1) + (5 2) + (5 5) 2

1 100

= 0.295

Solution 4.1.21 We are told that T1 Exp( = 1) and T2 Exp( = 1). Further, since T1 and T2 are independent, then f (t1 , t2 ) = [exp(t1 )][exp(t2 )] for t1 > 0 and t2 > 0. We are being asked to nd g (x), the pdf of X = 2T1 + T2 .
F3.24 F3.24

Let G(x) be the cdf of X .

F3.6 X

d G(x) = g (x) = dx

d P (X dx

x) =

d P (2T1 dx

+ T2 x) =

d P (T2 dx

x 2T1 ).

The following picture shows the integration region needed to calculate P (T2 x 2T1 ):

237

T2 (0, x) T2 = x 2T1

(0, 0)

(1 x, 0) 2

T1

Looking at the shaded region, integrating over T2 rst, we see that T2 goes as low as T2 = 0 and is as high as the line T2 = x 2T1 . Once we integrate over T2 , we then integrate over T1 , which goes as far left as T1 = 0 and as far right as T1 = 1 x. Therefore. . . 2 g (x) = = = =
d dx d dx d dx d P (T2 dx
1x 2

x 2T1 ) =

x d 2 dx 0 F3.27

exp(t1 )

x2t1 0

exp(t2 ) dt2 dt1

1x 2

exp(t1 ) 1 exp[(x 2t1 )] dt1 =

1x t1 = 2

d dx

exp(t1 ) exp(t1 x) dt1

exp(t1 ) exp(t1 x)

t1 =0

d dx

exp( 1 x) exp(0) exp( 1 x x) exp(0 x) 2 2

1 1 2 exp( 1 x) + exp(x) = exp( 2 x) exp(x) 2

Solution 4.1.22 Let P and C be the premiums and claims, respectively. We are told that P Exp( = 2) and C Exp( = 1). Further, since P and C are independent, then f (p, c) = [(1/2) exp(p/2)][exp(c)] for p > 0 and
F3.24 F3.24

c > 0. We are being asked to nd g (x), the pdf of X =

F3.6 X d P (X dx

C . P

Let G(x) be the cdf of X . P x).

g (x) =

d G(x) = dx

x) =

d P dx

C x = P

d P (C dx

The following picture shows the integration region needed to calculate P (C P x): C C = Px

C Px

(0, 0)

Looking at the shaded region, integrating over C rst, we see that C goes as low as C = 0 and is high as the line C = P x. Once we integrate over C , we then integrate over P , which goes as far left as P = 0 and as far right as P = . Therefore. . . g (x) = =
d dx d P (C dx 1 0 2

P x) =

d dx

1 0 2 F3.27

exp(p/2)

px 0

exp(c) dc dp
1 0 2

exp(p/2) 1 exp(px) dp =

d dx

exp(p/2)

1 2

exp

p(1+2x) 2

dp

238

For constants c and d and random variables X and Y , we have: Cov(c, X ) = 0 and Cov(cX, dY ) = cdCov(X, Y ) (4.23) Whereas the covariance of two random variables X and Y can be negative, zero, or positive of any magnitude, the correlation coecient (from now on, simply called correlation) of two random variables X and Y , dened here and often denoted X,Y , is a standardized number between -1 and 1: X,Y = Cov(X, Y ) X Y (4.24) (4.22)

The following picture shows the values of X,Y for various (x, y ) scatter-plots:

x X,Y = 1 y 0 < X,Y < 1

x X,Y = 1 y

x 1 < X,Y < 0

x X,Y 0 X,Y 0

Correlation, much like covariance, measures the strength of the linear relationship between X and Y . Of these two measures, correlation is a much more interpretable metric because it is constrained between -1 and 1. On one extreme, the correlation is exactly 1 when an increase of x is associated with an exact linear increase in y . This is seen in the rst graph on the top row. Correlation is positive (but not exactly 1) when an increase of x is generally associated with a linear increase in y . This is seen in the second graph on the top row. Similar arguments can be made for negative correlations, as seen in the last two graphs in the top row. The graphs in the second row show correlations around 0. In the rst of these two graphs, x and y appear to be randomly distributed. Although the second graph shows a strong parabolic relationship between x and y , it is important to recall that covariance and correlation measure the strength of the linear relationship between two variables. Since none of these bottom two graphs exhibit a strong linear relationship, their correlations are around 0. The sum of a xed number of random variables For random variables Xi and Yj and constants ai , bj , m, and n, let us dene the following linear combinations:
m i=1 ai Xi n j =1 bj Yj

= a1 X1 + a2 X2 + + am Xm = b1 Y1 + b2 Y2 + + bn Yn

253

Problem 4.2.19 (104 of 153; SOA May 2001 #7) A joint density function is given by: f (x, y ) = kx where k is a constant. What is Cov(X, Y )? Problem 4.2.20 (105 of 153; SOA May 2003 #6) Let X and Y be continuous random variables with joint density function: 8 f (x, y ) = xy for 0 x < 1, x y 2x. 3 What is Cov(X, Y )? Problem 4.2.21 (107 of 153; SOA November 2001 #7) Let X denote the size of a surgical claim and let Y denote the size of the associated hospital claim. An actuary is using a model in which E (X ) = 5, E (X 2 ) = 27.4, E (Y ) = 7, E (Y 2 ) = 51.4, and V (X + Y ) = 8. Let C1 = X + Y denote the size of the combined claims before the application of a 20% surcharge on the hospital portion of the claim, and let C2 denote the size of the combined claims after the application of that surcharge. Calculate Cov(C1 , C2 ). Problem 4.2.22 (121 of 153; SOA May 2003 #24) Let X represent the age of an insured automobile involved in an accident. Let Y represent the length of time the owner has insured the automobile at the time of the accident. X and Y have joint probability density function: f (x, y ) = 1 (10 xy 2 ) 64 for 2 x < 10 and 0 y 1. for 0 < x < 1, 0 < y < 1,

Calculate the expected age of an insured automobile involved in an accident. Problem 4.2.23 (122 of 153; SOA November 2000 #40) A device contains two circuits. The second circuit is a backup for the rst, so the second is used only when the rst has failed. The device fails when and only when the second circuit fails. Let X and Y be the times at which the rst and second circuits fail, respectively. X and Y have joint probability density function: f (x, y ) = 6ex e2y What is the expected time at which the device fails? Problem 4.2.24 (137 of 153) Let X and Y be identically distributed independent random variables such that the moment generating function of X + Y is: M (t) = 0.09e2t + 0.24et + 0.34 + 0.24et + 0.09e2t Calculate P (X 0). Problem 4.2.25 (138 of 153) A machine consists of two components, whose lifetimes have the joint density function: f (x, y ) = 1 50 for x > 0, y > 0, x + y < 10. for < t < . for 0 < x < y < .

The machine operates until both components fail. Calculate the expected operational time of the machine. Problem 4.2.26 (148 of 153) The number of hurricanes that will hit a certain house in the next ten years is Poisson distributed with mean 4. Each hurricane results in a loss that is exponentially distributed with mean 1,000. Losses are mutually independent and independent of the number of hurricanes. Calculate the variance of the total loss due to hurricanes hitting this house in the next ten years. Problem 4.2.27 (106 of 153; SOA May 2000 #20) Let X and Y denote the values of two stocks at the end of a ve-year period. X is uniformly distributed on the interval (0, 12). Given X = x, Y is uniformly distributed on the interval (0, x). Determine Cov(X, Y ) according to this model.

262

kx

Solution 4.2.19 Since f (x, y ) = g (x) h(y ) can be reexpressed as the product of a function of just x and a function of just y and since the (x, y ) space is a product space, then X and Y are independent random variables (see Formula 4.7). Therefore, Formula 4.21 gives Cov(X, Y ) = 0. You can verify this by calculating Cov(X, Y ) = E (XY ) E (X )E (Y ). Solution 4.2.20 The (x, y ) space is given here: y y = 2x (1, 2) x y 2x (1, 1) y=x (0, 0) (1, 0) x

Looking at the shaded region, integrating over y rst, we see that y goes as low as the line y = x and is as high as the line y = 2x. Once we integrate over y , we then integrate over x, which goes as far left as x = 0 and as far
8x 3

right as x = 1. We are being asked to nd Cov(X, Y ) = E (XY ) E (X )E (Y ). Although f (x, y ) = g (x) h(y ) can be expressed as the product of a function of just x and a function of just y , the (x, y ) integration space is not rectangular. Therefore, unlike Problem 4.2.19, we cannot assume that X and Y are independent nor that Cov(X, Y ) = 0. We must therefore calculate the various moments in the covariance formula:
8 xy 3

E (XY ) = =
8 9 1 0

1 0

2x x

xy f (x, y ) dy dx =
56 9 1 0

8 3

1 0

x2
56 9

2x x

y 2 dy dx =
x=1

8 3

1 0

x2

1 3 y 3

y =2x

dx
y =x 56 54

x2 (2x)3 x3 dx =
8 xy 3

x5 dx =
8 3 1 0

1 6 x 6

=
x=0

56 54

16 06 = x2
1 2 y 2

E (X ) = =
8 6 1 0

1 0

2x x

x f (x, y ) dy dx =
24 6 1 0

x2

2x x 24 6

y dy dx =
1 5 x 5 x=1

8 3

1 0

y =2x

dx
y =x 24 30

x2 (2x)2 x2 dx =
8 xy 3

x4 dx =
1 0

=
x=0 8 3

24 30

15 05 =
1 3 y 3 y =2x

E (Y ) = =
8 9 1 0

1 0

2x x

y f (x, y ) dy dx =
56 9 1 0

8 3

2x x 56 9

y 2 dy dx =
1 5 x 5 x=1

1 0

dx
y =x 56 45

x (2x)3 x3 dx =

x4 dx =

=
x=0

56 45

15 05 =

Therefore, Cov(X, Y ) = 56/54 (24/30)(56/45) = 0.0415. Solution 4.2.21 Since the 20% surcharge only applies to Y , then C2 = X + 1.2Y . We can use the rst two moments of X and Y to nd V (X ) and V (Y ): V (X ) = E (X 2 ) [E (X )]2 = 27.4 [5]2 = 2.4 V (Y ) = E (Y 2 ) [E (Y )]2 = 51.4 [7]2 = 2.4
8 2.4 2.4 ?

Formula 4.27 gives: Therefore. . .

C1

V (X + Y ) = V (X ) + V (Y ) +2 Cov(X, Y ) Cov(X, Y ) = 1.6

C2

F4.29

Cov(C1 , C2 ) = Cov(X + Y , X + 1.2Y ) = Cov(X, X ) + Cov(X, 1.2Y ) + Cov(Y, X ) + Cov(Y, 1.2Y )

273

2.8

The Poisson Distribution

A generic poisson random variable, X , counts the number of events occurring in a xed interval of time (or space, distance, volume, or some other measure), where the average observation rate for the given interval is denoted . The symbol is considered the parameter of the poisson distribution since it is the only value needed to fully dene the probability space. The range of possible values is X = 0, 1, 2, . . . Shorthand notation: For convenience, rather than tediously writing X is distributed as a poisson random variable with parameter , we will condense this information into the equivalent expression X Pois(). The symbol should be interpreted as is distributed as. Example 2.24 Assume that copper wire is known to have irregularities, on average, once every 12 feet. Let X count the number of defects in 12 feet of copper wire. Assume that X is distributed as a poisson random variable. a) Express the poisson random variable using shorthand notation. b) Adjust your answer to a) if X counts the number of defects in 30 feet of copper wire. c) Adjust your answer to a) if X counts the number of defects in 1 foot of copper wire. Solution: a) We are told that, on average, 1 irregularity shows up every 12 feet. Since X counts the number of defects in 12 feet of wire, then X Pois( = 1). b) If there is, on average, 1 copper irregularity every 12 feet, then there must be, on average, 2.5 copper irregularities every 30 feet. Therefore, X Pois( = 2.5). This 2.5 is calculated using proportions, given here: 1 irregularity x irregularities = 12x = 30(1) x = 2.5 12 feet 30 feet c) If there is, on average, 1 copper irregularity every 12 feet, then there must be, on average, 1/12 copper irregularity every 1 foot. Therefore, X Pois( = 1/12). This 1/12 is calculated using proportions, given here: 1 irregularity x irregularities = 12x = 1(1) x = 1/12 12 feet 1 feet Note: In a), b), and c), the possible values that X can assume are X = 0, 1, 2, . . . This is true for any poisson random variable, regardless of how is adjusted using proportions.

Each of the possible X values has an associated probability, given by the following pmf: P (X = x) = where x = 0 , 1, 2, . . . >0 Note: The exp phrase in Formula 2.37 simply refers to the irrational number e = 2.718 . . . The mean and variance of a poisson random variable are given here: X = E(X ) =
2 X = V(X ) =

x exp() x! (2.37)

(2.38)

Formula 2.38 shows that the parameter of the poisson distribution, , is also the mean and variance of the distribution.

116

Example 2.25 Customers arrive at a grocery store at a mean rate of 3 per hour. Assuming that the number of arrivals per time period has a Poisson distribution, nd the probability that at least 1 customer arrives in the next. . . a) hour. b) 10 minutes. Solution: In a) and b), we are being asked to nd P (X 1). Formula 1.14 gives P (X 1) = 1 P (X = 0). a) Let X count the number of customers arriving per hour. X Pois( = 3). Formula 2.37 gives: P (X 1) = 1 P (X = 0) = 1
30 exp(3) 0!

Since customers arrive at a rate of 3 per hour,

= 1 0.0498 = 0.9502.

b) Let X count the number of customers arriving every ten minutes. Because is presented to us as a per hour rate and because b) phrases this problem over a 10 minute interval, we need to use proportions to adjust accordingly: 3 customers x customers = 60x = 10(3) x = 0.5 60 minutes 10 minutes This implies that = 0.5 customer arrives, on average, every 10 minutes. Therefore, X Pois( = 0.5). P (X 1) = 1 P (X = 0) = 1
0.50 exp(0.5) 0!

= 1 0.6065 = 0.3935.

The following picture shows the graphical probability distribution of X Pois( = 3). Since there are innitely-many values that X can assumed (X = 0, 1, 2, . . .), only the rst ve are shown. The height of each rectangle represents the probability that X assumes that particular value. Also the mean is labeled.

Probability 0.25 0.20 0.15 0.10 0.05 1 2 3 4 X = E ( X ) = = 3

The moment generating function (mgf) of a poisson random variable is: MX (t) = exp (et 1) Example 2.26 In the context of Example 2.25 part a), identify the mgf of X . Solution: Substituting = 3 into Formula 2.39 gives MX (t) = exp 3(et 1) . (2.39)

Sums of poisson random variables Let Xi Pois(i ) for i = 1, 2, . . . , m represent m independent poisson random variables. Because is subscripted, this implies that it can be dierent from one Xi to the next. Let X = m i=1 Xi = X1 + X2 + + Xm be the sum of

117

these m random variables. Then X Pois( = 1 + 2 + + m ). In words, the sum of m poisson random variables is a poisson random variable whose parameter is the sum of the individual parameters. This is summarized here: Let Xi Pois(i ) where i = 1, 2, . . . , m be m independent poisson random variables. Let X = X1 + X2 + + Xm . Then. . . X Pois( = 1 + 2 + + m ) (2.40)

End-of-Section Practice Problems

Problem 2.8.1 Let X have a Poisson distribution with a mean of 3. Find. . . a) P (2 X < 5) b) P (X < 2) c) P (X 2) Problem 2.8.2 Let X have a Poisson distribution with a standard deviation of 4. Find P (X = 18). Problem 2.8.3 The random variable X has a Poisson distribution such that P (X = 3) = 0.60 P (X = 2). Find P (X = 5). Problem 2.8.4 Flaws in linen produced at a certain textile factory appear at a rate of 3 in every 204 square feet. If we assume the Poisson distribution, nd the probability of at most one aw in 80 square feet. Problem 2.8.5 The mean of a Poisson random variable X is E (X ) = X = 3. Find P (X X X < X + 2X ). Problem 2.8.6 The opening day of a golf season at a country club gets postponed if it rains on that day. The club loses $15,000 for each consecutive day that it rains. Say X , the number of consecutive days that it rains at the beginning of the golf season, has a Poisson distribution with a mean of 0.3. What is the expected loss before the opening golf day? Problem 2.8.7 Let X denote a random variable that has a Poisson distribution with mean 3. Find. . . a) P (X 3) b) P (X 1) c) P (X 3|X 1) Problem 2.8.8 The number of errors that an actuarial intern makes when designing spreadsheets has a Poisson distribution with an average of ve errors per spreadsheet. If more than two errors appear in a given spreadsheet, the intern must do 100 pushups. What is the probability that the intern does not have to do 100 pushups for her design of a certain spreadsheet? Problem 2.8.9 The number of defects per square foot of sheet metal, X , is assumed to have a Poisson distribution with a standard deviation of 1.35. The prot, P , per square foot when the sheet metal is sold is P = 100 4X X 2 . Find the expected prot per square foot. Problem 2.8.10 Let X Pois(). Use Formula 2.39 to. . . a) Show that the rst moment is E(X ) = . b) Show that the second moment is E(X 2 ) = 2 + . c) Use your answers to a) and b) to show that V(X ) = . Note: This is not an exam-type problem but is important because it shows the link between Formulas 2.38 and 2.39.

118

Problem 2.8.11 The moment generating function of a random variable is MX (t) = exp 12(et 1) . a) Identify the distribution of X , including its parameters. b) Calculate P (X = 15). Problem 2.8.12 (30 of 153; SOA May 2000 #24) An actuary has discovered that policyholders are three times as likely to le two claims as to le four claims. If the number of claims led has a Poisson distribution, what is the variance of the number of claims led? Problem 2.8.13 (50 of 153; SOA November 2000 #23) A company buys a policy to insure its revenue in the event of a major snowstorm that shuts down business. The policy pays nothing for the rst such snowstorm of the year and 10,000 for each one thereafter, until the end of the year. The number of major snowstorms per year that shut down business is assumed to have a Poisson distribution with mean 1.5. What is the expected amount paid to the company under this policy during a one-year period? Problem 2.8.14 (67 of 153; SOA November 2001 #19) A baseball team has scheduled its opening game for April 1. If it rains on April 1, the game is postponed and will be played on the next day that it does not rain. The team purchases insurance against rain. The policy will pay 1,000 for each day, up to 2 days, that the opening game is postponed. The insurance company determines that the number of consecutive days of rain beginning on April 1 is a Poisson random variable with mean 0.6. What is the standard deviation of the amount the insurance company will have to pay? Problem 2.8.15 (153 of 153) Let X represent the number of customers arriving during the morning hours and let Y represent the number of customers arriving during the afternoon hours at a diner. You are given: i) X and Y are Poisson distributed. ii) The rst moment of X is less than the rst moment of Y by 8. iii) The second moment of X is 60% of the second moment of Y . Calculate the variance of Y .

End-of-Section Practice Solutions

Solution 2.8.1 Formula 2.38 shows us that the mean of the Poisson random variable, , is also the parameter of the Poisson distribution. Since the mean is 3, then X Pois( = 3). a) We are being asked to nd P (X = 2) + P (X = 3) + P (X = 4). Formula 2.37 gives: P (X = 2) = P (X = 3) = P (X = 4) =
32 exp(3) 2! 33 exp(3) 3! 34 exp(3) 4!

= 0.2240 = 0.2240 = 0.1680

Therefore P (2 X < 5) = 0.2240 + 0.2240 + 0.1680 = 0.6160. b) We are being asked to nd P (X < 2) = P (X = 0) + P (X = 1). Formula 2.37 gives: P (X = 0) = P (X = 1) =
30 exp(3) 0! 31 exp(3) 1!

= 0.0498 = 0.1494

119

Therefore P (X < 2) = 0.0498 + 0.1494 = 0.1992. c) P (X 2) = 1 P (X < 2) = 1 0.1992 = 0.8008 Solution 2.8.2 If the standard deviation is 4, then the variance is 42 = 16. Formula 2.38 shows us that the variance of the Poisson random variable, , is also the parameter of the Poisson distribution. Therefore, X Pois( = 16). Formula 2.37 gives: P (X = 18) =
1618 exp(16) 18!

= 0.0830.

Solution 2.8.3 We rst need to use the given equality to solve for . Using Formula 2.37 gives: P (X = 2) = P (X = 3) =
2 exp() 2! 3 exp() 3!

Substituting these expressions into the given equality yields:

P (X =3) P (X =2) 0.602 2

$ $ $ $ $ $ 3$ exp( ) 2 exp( ) 2 = 0.60 $ & = 6 3! 2!

P (X = 5) =
1.85 exp(1.8) 5!

& = 1.8 X Pois( = 1.8)

= 0.0260

Solution 2.8.4 Let X count the number of aws that appear in 80 square feet. We are told the rate for every 204 square feet. We need to nd , the number of aws every 80 square feet. Using proportions gives: 3 aws aws 3 = = = (80) = 1.1765 204 square feet 80 square feet 204 Therefore, there are 1.1765 aws every 80 square feet. Therefore X Pois( = 1.1765). We are being asked to nd P (X 1) = P (X = 0) + P (X = 1). Formula 2.37 gives: P (X = 0) = P (X = 1) =
1.17650 exp(1.1765) 0! 1.17651 exp(1.1765) 1!

= 0.3084 = 0.3628

Therefore, P (X 1) = 0.3084 + 0.3628 = 0.6712. Solution 2.8.5 Formula 2.38 shows that is both the mean and variance of a Poisson random variable. Therefore 2 = 3, then SD(X ) = X = 3 = 1.7321. = E (X ) = V (X ) = 3. Therefore, X Pois( = 3). Since V (X ) = X Substituting X = 3 and X = 1.7321 into the expression gives: P (X X X < X + 2X ) = P (3 1.7321 X < 3 + 2(1.7321)) = P (1.2679 X < 6.4642). Because X can only assume integer values, the previous expression simplies to P (2 X 6). Formula 2.37 gives: P (X = 2) = P (X = 3) = P (X = 4) = P (X = 5) = P (X = 6) =
32 exp(3) 2! 33 exp(3) 3! 34 exp(3) 4! 35 exp(3) 5! 36 exp(3) 6!

= 0.2240 = 0.2240 = 0.1680 = 0.1008 = 0.0504

Therefore P (2 X 6) = 0.2240 + 0.2240 + 0.1680 + 0.1008 + 0.0504 = 0.7672

120

Solution 2.8.6 Let X count the number of consecutive days of rain at the beginning of the golf season. Formula 2.38 shows us that the mean of the Poisson random variable, , is also the parameter of the Poisson distribution. Since the mean is E (X ) = 0.3, then X Pois( = 0.3). Let Y measure the total loss due to postponing the opening day due to rain, so Y = 15, 000X . We are being asked to nd E (Y ): E (Y ) = E (15, 000X ) = 15, 000E (X ) = 15, 000(0.3) = 4, 500. Solution 2.8.7 Formula 2.38 shows us that the mean of the Poisson random variable, , is also the parameter of the Poisson distribution. Since the mean is E (X ) = 3, then X Pois( = 3). We will use Formula 2.37 for the probability calculations. a) P (X 3) = 1 P (X < 3) = 1 P (X = 0) P (X = 1) P (X = 2). P (X = 0) = P (X = 1) = P (X = 2) =
30 exp(3) 0! 31 exp(3) 1! 32 exp(3) 2!

= 0.0498 = 0.1494 = 0.2240

Therefore, P (X 3) = 1 0.0498 0.1494 0.2240 = 0.5768. b) P (X 1) = 1 P (X = 0) = 1 0.0498 = 0.9502. c) Formula 1.22 expands P (X 3|X 1) =
P [(X 3)(X 1)] P (X 1)

P (X 3) P (X 1)

0.5768 0.9502

= 0.6070

Solution 2.8.8 Let X count the number of errors that the intern makes on a given spreadsheet. Formula 2.38 shows us that the mean of the Poisson random variable, , is also the parameter of the Poisson distribution. Since the mean is E (X ) = 5, then X Pois( = 5). Since the intern does not have to do pushups if at most 2 errors are made, we are being asked to nd P (X 2) = P (X = 0) + P (X = 1) + P (X = 2). Formula 2.37 gives: P (X = 0) = P (X = 1) = P (X = 2) =
50 exp(5) 0! 51 exp(5) 1! 52 exp(5) 2!

= 0.0067 = 0.0337 = 0.0842

Therefore P (X 2) = 0.0067 + 0.0337 + 0.0842 = 0.1246. Solution 2.8.9 Formula 2.38 shows us that the variance of the Poisson random variable, , is also the parameter of the Poisson distribution. Since the standard deviation of X is SD(X ) = X = 1.35, then the variance of X is 2 = 1.352 = 1.8225. Therefore, X Pois( = 1.8225). We are being asked to nd E (P ): V (X ) = X E (P ) = E (100 4X X 2 ) = 100 4E (X ) E (X 2 ) Let us now nd the rst two moments of X , using Formula 2.38: E (X ) = X = = 1.8225 2 V (X ) = X = E (X 2 ) [E (X )]2 = E (X 2 ) [1.8225]2 = 1.8225 E (X 2 ) = 5.1440
F2.10

Substituting these values gives E (P ) = 100 4(1.8225) 5.1440 = 87.5660 Solution 2.8.10 Formula 2.15 gives us the patterns to nd the rst two moments.
F2.39 u v 1) 1

a)

d MX (t) dt

d dt

e(e

1)

= e(e

et

E (X ) =

d MX (t) dt

= e(e
t=0

1)

e0 =

b) To nd the second moment, we have to dierentiate MX (t) twice. Using the dierentiation rule for products,

121

where u and v are identied above, we have (uv ) = u v + uv :

u d2 MX (t) dt2 v u t (et 1) v

=
2

d dt

d MX (t) dt

d dt

e
1

(e 1)

=e

(e 1)

e e + e

et = et e(e

1)

(et + 1)

+1
0

E (X 2 ) =

d MX (t) dt2

= e0 e(e
t=0

1)

(e0 + 1) = ( + 1) = 2 +

2 c) Formula 3.14 gives V (X ) = X = E (X 2 ) [E (X )]2 = 2 + []2 =

Solution 2.8.11 a) Matching MX (t) with Formula 2.39 shows that = 12. Therefore, X Pois( = 12).
F2.37

b) P (X = 15) =

12

15

exp(12) = 0.0724 15!

Solution 2.8.12 Let X represent the number of claims led. We are told that X is distributed as a Poisson random variable but are not given its parameter . Therefore, we have X Pois(). We are being asked to nd the variance of X , which Formula 2.38 tells us is V (X ) = . We are told that P (X = 2) = 3P (X = 4). Substituting the pmf into this expression gives:
P (X =2) P (X =4)

$ $ $ $ exp( )
2!

=3

$ $ $ $ exp( ) 4!

2 2

34 24

= V (X ) = 2

Solution 2.8.13 Let X count the number of major snowstorms that shut down business during a given year, where X = 0, 1, 2, . . .. We are told that E (X ) = 1.5 which, based on Formula 2.38, is also the value of . Therefore, X Pois( = 1.5). Let Y be the amount paid to the company under this policy. Based on the given information, we have 0 if x = 0, 1 Y = 10, 000(X 1) if x 2 We are being asked to nd E (Y ) =
=0 y

yP (Y = y ), calculated here:

E (Y ) = 0P (X = 0) + 0P (X = 1) +
x=2

10, 000(x 1)P (X = x)

=
x=2

10, 000(x 1)P (X = x)

=0

= 10, 000(0 1)P (X = 0) 10, 000(0 1)P (X = 0)

=0

+ 10, 000(1 1)P (X = 1) 10, 000(1 1)P (X = 1)

+
x=2

10, 000(x 1)P (X = x)

E [10,000(X 1)] 0

= 10, 000(0 1)P (X = 0) 10, 000(1 1)P (X = 1) +

[10, 000(x 1)]P (X = x)

x=0

Note that the overbrace expression in lines 3 and 4 above were included so as to allow the summation to go over all possible X values (from 0 to ) so as to create E [10, 000(X 1)] in the last line. Using the pmf of X , we have:
1.5) P (X = 0) = 1.5 exp( = 0.2231 0! E [10, 000(X 1)] = 10, 000E (X ) 10, 000 = 10, 000(1.5) 10, 0000 = 5, 000
0

122

Therefore, E (Y ) = 10, 000(1)(0.2231) + 5, 000 = 7, 231. Solution 2.8.14 Let X count the number of consecutive days of rain beginning on April 1, where X = 0, 1, 2, . . . We are told that E (X ) = 0.6 which, based on Formula 2.38, is also the value of . Therefore, X Pois( = 0.6). Let Y be the amount paid to the team under its insurance policy. Based on the given information, we have Y = 1, 000x 2, 000 if x 1 if x 2

We are being asked to nd SD(Y ) = V (Y ). Since V (Y ) = E (Y 2 ) [E (Y )]2 , we will need to nd the rst two moments of Y . Let us rst nd E (Y ) = y yP (Y = y ):
0

E (Y ) = 1, 000(0)

0.60 exp(0.6) 0.61 exp(0.6) +1, 000(1) + 2, 000P (X = x) 0! 1! x=2

=600 exp(0.6) + 2, 000P (X 2) =600 exp(0.6) + 2, 000[1 P (X 1)]

P (X =0) 0 1 P (X =1)

=600 exp(0.6) + 2, 000 1

0.6 exp(0.6) 0.6 exp(0.6) 0! 1!

=0.12190138

=573 Let us now nd E (Y 2 ) =

y

y 2 P (Y = y ):
0

E (Y 2 ) = [1, 000(0)]2

0.60 exp(0.6) 0.61 exp(0.6) +[1, 000(1)]2 + 2, 0002 P (X = x) 0! 1! x=2

=600, 000 exp(0.6) + 2, 0002 P (X 2) =600, 000 exp(0.6) + 2, 0002 (0.12190138) =816, 893 Therefore, V (Y ) = 816, 893 [573]2 = 488, 564. Therefore, SD(Y ) = V (Y ) = 488, 564 = 699.

Solution 2.8.15 We are told that both X and Y are Poisson random variables but with dierent parameters. Therefore, we can write: X Pois(x ) and Y Pois(y ). Formula 2.38 gives the rst moment for a Poisson random variable: E (X ) = x and E (Y ) = y . Statement (ii) tells us that x = y 8. Formula 2.10 tells us that V (X ) = E (X 2 ) [E (X )]2 E (X 2 ) = V (X ) + [E (X )]2 . Formula 2.38 tells us that 2 2 V (X ) = x and V (Y ) = y . Therefore, the second moments are E (X 2 ) = x + 2 x and E (Y ) = y + y . Statement 2 2 2 2 (iii) tells us that E (X ) = 0.60E (Y ) x + x = 0.60(y + y ). Substituting x = y 8 into the previous equation gives: (y 8) + (y 8)2 = 0.60(y + 2 y ),
x x

which is a quadratic equation. Rearranging this gives 0.42 y 15.6y + 56 = 0. Substituting a = 0.4, b = 15.6, and c = 56 into Formula 1.4 gives y = 4 and y = 35. If y = 4, then the statement from (ii) gives x = 4 8 = 4. Since the Poisson parameter must be positive, y = 4. Therefore y = 35. We are asked to nd V (Y ). Since V (Y ) = y , then V (Y ) = 35.

123

5.2

Normal Approximation of Discrete Probabilities

Sometimes we are asked to calculate a very tedious probability involving a discrete random variable. For example, if X Binom(n = 750, p) and if we are asked to calculate the probablity that X is at least 500, we would need to nd: P (X 500) = P (X = 500) + P (X = 501) + P (X = 502) + + P (X = 749) + P (X = 750), where each of the 251 probabilities on the right side must be evaluated using Formula 2.21. This would take a very long time to calculate exactly. Under certain conditions, the continuous normal distristribution provides a very close (and much faster!) approximation to the exact discrete calculation. To make these approxmations a little more accurate, we need to apply a continuity correction as given by the following table. For a generic discrete random variable X and constant c, do the following: If problem is originally phrased like this. . . P (X < c) P (X c) P (X > c) P (X c) P (X = c) Apply this adjustment for the continuity correction P (X < c 0.5) P (X c + 0.5) P (X > c + 0.5) P (X c 0.5) P (c 0.5 X c + 0.5)

(5.8)

Note: Since X assumes only one value in the last row of the previous formula, in this case, one would probably calculate the probability exactly rather than approximating it with the normal distribution.

Example 5.4 Assume X is a generic discrete random variable. Apply Formula 5.8 to the following inequalities as if we were asked to approximate a discrete probability calculation using the normal curve. a) P (X > 5) b) P (X 71) c) P (14 X ) d) P (X < 12) e) P (3 < X < 10) f) P (X = 20) Solution: a) P (X > 5.5) b) P (X 71.5) c) P (14 X ) = P (X 14) P (X 13.5) d) P (X < 11.5) e) P (3 < X ) = P (X > 3) P (X > 3.5) P (3.5 < X < 9.5) f) P (19.5 X 20.5)

You may be interested in how the normal curve (a continuous distribution) can provide a reasonable approximation to a discrete probability calculation. Under certain mathematical conditions, the probability distribution of the discrete random variable is reasonably symmetric and bell-shaped, closely mimicking the normal distribution (albeit from a discrete perspective). The following picture shows two distributions. The discrete distribution is of X Binom(n = 10, p = 0.5), whose mean and standard deviation (given by Formula 2.22) are X = 5 and X = 1.5811. The continuous normal distribution is of Y N(Y = 5, Y = 1.5811).

292

Probability 0.25 0.20 0.15 0.10 0.05 0

X Binom(n = 10, p = 0.5) Y N(Y = 5, Y = 1.5811)

10

Since the two curves in this picture closely mimic each other, we should have condence that the normal curve can be used to calculate reasonable approximations to discrete calculations. Note that the discussion so far in this section has referenced the binomial distribution. The normal curve, however, can be used to approximate any discrete probability, not just the binomial distribution. Here are the steps. . .

Step #1: Apply the continuity correction (Formula 5.8) to the discrete probability expression. Step #2: Calculate the mean and standard deviation of the discrete random variable. Step #3: Apply the z -transformation formula (Formula 3.43) to all sides of the probability expression. Step #4: Carry out the calculation from the Z perspective, calculating the answer exactly like the normal probability calculations from Section 3.6. (5.9)

Example 5.5 Assume that 15% of all college students would answer yes to the question Would you drive home from a party if you had been drinking excessively? Let X count the number of students from a random sample of size n = 250 who would say yes to this question. Suppose we are interested in P (X 45). a) How would you calculate P (X 45) exactly? b) Approximate P (X 45) using the method from this section. Solution: Assuming that responses are independent, X is a binomial random variable since it counts the number of successes (driving while being intoxicated, with probability p = 0.15) from a xed number of people (n = 250) and because any one person would eith drive home or not drive home. Here, q = 1 0.15 = 0.85. a) To calculate this probability exactly, we would tediously apply Formula 2.21 to each of the following probabilities: P (X 45) = P (X = 0) + P (X = 1) + + P (X = 44) + P (X = 45) b) Let us rst nd the mean and standard deviation of the binomial random variable X using Formula 2.22: X = E (X ) = np = 250(0.15) = 37.5 2 X = V (X ) = npq = 250(0.15)(0.85) = 31.875 X = SD(X ) = Now let us calculate the probability of interest:
F5.8 Z

V (X ) =

31.875 = 5.65

P (X 45) = P (X 45.5) = P

X X X

45.537.5 5.65

P (Z 1.42) = 0.9222

293