Chapter 11: Monte Carlo Simulation: Appendix 11.

1 Common Probability Distributions for Simulation

Book Title: Business Analytics, Third Edition
Printed By: Jordan Miles ([email protected])
© 2019 Cengage Learning, Cengage Learning

Chapter Review
Appendix 11.1 Common Probability Distributions for Simulation

Selecting the appropriate probability

distribution to characterize a random Simulation software such as Analytic Solver,
variable in a simulation model can be a Crystal Ball, or @RISK automates the
critical modeling decision. In this generation of random values from an even
appendix, we review several probability wider selection of probability distributions
distributions commonly used in than is available in native Excel.
simulation models. We describe the
native Excel functionality used to
generate random values from the
corresponding probability distribution.

Continuous Probability Distributions

Random variables that can be many possible values (even if the values are discrete) are
often modeled with a continuous probability distribution. For common continuous random
variables, we provide several pieces of information. First, we list the parameters which
specify the probability distribution. We then delineate the minimum and maximum values
defining the range that can be realized by a random variable that follows the given
distribution. We also provide a short description of the overall shape of the distribution
paired with an illustration. Then, we supply an example of the application of the random
variable. We conclude with the native Excel functionality for generating random values from
the probability distribution.

Normal Distribution

Parameters: mean (m), standard deviation (s)

Range:

Description: The normal distribution is a bell-shaped, symmetric distribution centered at

its mean m. The normal distribution is often a good way to characterize a quantity that is
 the sum of many independent random variables.

Example: In human resource management, employee performance is often well

represented by a normal distribution. Typically, the performance of 68% of employees is
within one standard deviation of the average performance, and the performance of 95%
of the employees is within two standard deviations. Employees with exceptionally low or
high performance are rare. For example, the performance of a pharmaceutical company’s
sales force may be well described by a normal distribution with a mean of 200 customer
adoptions and a standard deviation of 40 customer adoptions.

Native Excel: NORM.INV(RAND(), m, s)

Beta Distribution

Parameters: alpha (α), beta (β), minimum (A), maximum (B)

Range: A to B

Description: Over the range specified by values A and B, the beta distribution has a very
flexible shape that can be manipulated by adjusting and . The beta distribution is
useful in modeling an uncertain quantity that has a known minimum and maximum
value. To estimate the values of the alpha and beta parameters given sample data, we
use the following equations:
 Example: The boom-or-bust nature of the revenue generated by a movie from a polarizing
director may be described by a beta distribution. The relevant values (in millions of
dollars) are , , , and . This particular distribution is U-
shaped and extreme values are more likely than moderate values. The figures in the left
margin illustrate beta distributions with different values of α and β, demonstrating its
flexibility. The first figure depicts a U-shaped beta distribution. The second figure depicts
a unimodal beta distribution with a positive skew. The third figure depicts a unimodal
beta distribution with a negative skew.

Native Excel: BETA.INV(RAND(), α, β, A, B)

Gamma Distribution

Parameters: alpha (α), beta (β)

Range: 0 to +`

Description: The gamma distribution has a very flexible shape controlled by the values of
α and β. The gamma distribution is useful in modeling an uncertain quantity that can be
as small as zero but can also realize large values. To estimate the values of the alpha
and beta parameters given sample data, we use the following equations:

Example: The aggregate amount (in $100,000s) of insurance claims in a region may be
described by a gamma distribution with and .

Native Excel: GAMMA.INV(RAND(), α, β)

Exponential Distribution

Parameters: mean (m)

Range: 0 to

Description: The exponential distribution is characterized by a mean value equal to its

standard deviation and a long right tail stretching from a mode value of 0.

Example: The time between events, such as customer arrivals or customer defaults on
bill payment, are commonly modeled with an exponential distribution. An exponential
random variable possesses the “memoryless” property: the probability of a customer
arrival occurring in the next x minutes does not depend on how long it’s been since the
last arrival. For example, suppose the average time between customer arrivals is 10
minutes. Then, the probability that there will be 25 or more minutes between customer
arrivals if 10 minutes have passed since the last customer arrival is the same as the
probability that there will be more than 15 minutes until the next arrival if a customer just
arrived.

Native Excel: LN(RAND())*(−m)

Triangular Distribution
Because the exponential distribution with
Parameters: minimum (a), most mean m is equivalent to the gamma
likely (m), maximum (b) distribution with parameters α = 1 and β =
(1/m), an exponential random variable can
Range: a to b also be generated with GAMMA.INV(RAND(),
1, 1/m).
Description: The triangular
distribution is often used to
subjectively assess uncertainty when
 little is known about a random variable besides its range, but it is thought to have a single
mode. The distribution is shaped like a triangle with vertices at a, m, and b.

Example: In corporate finance, a triangular distribution may be used to model a project’s

annual revenue growth in a net present value analysis if the analyst can reliably provide
minimum, most likely, and maximum estimates of growth. For example, a project may
have worst-case annual revenue growth of 0%, a most -likely annual revenue growth of
5%, and best-case annual revenue growth of 25%. These values would then serve as the
parameters for a triangular distribution.

Native Excel: IF(random < (m − a)/(b − a), a + SQRT((b − a)*(m − a)*random), b − SQRT((b
− a)*(b − m)*(1 − random))) where random refers to a single, separate cell containing
=RAND()

Uniform Distribution

Parameters: minimum (a), maximum (b)

Range: a to b

Description: The uniform distribution is appropriate when a random variable is equally

likely to be any value between a and b. When little is known about a phenomenon other
than its minimum and maximum possible values, the uniform distribution may be a
conservative choice to model an uncertain quantity.

Example: A service technician making a house call may quote a 4-hour time window in
which he will arrive. If the technician is equally likely to arrive any time during this time
window, then the arrival time of the technician in this time window may be described with
a uniform distribution.

Native Excel:
Log-Normal Distribution

Parameters: log_mean, log_stdev

Range: 0 to

Description: The log-normal distribution is a unimodal distribution (like the normal

distribution) that has a minimum value of 0 and a long right tail (unlike the normal
distribution). The log-normal distribution is often a good way to characterize a quantity
that is the product of many independent, positive random variables. The natural
logarithm of a log-normally distributed random variable is normally distributed.

Example: The income distribution of the lower 99% of a population is often well
described using a log-normal distribution. For example, for a population in which the
natural logarithm of the income observations is normally distributed with a mean of 3.5
and a standard deviation of 0.5, the income observations are distributed log-normally.

Native Excel: LOGNORM.INV(RAND(), log_mean, log_stdev), where log_mean and

log_stdev are the mean and standard deviation of the normally distributed random
variable obtained when taking the logarithm of the log-normally distributed random
variable.

Discrete Probability Distributions

Random variables that can be only a relatively small number of discrete values are often
best modeled with a discrete distribution. The appropriate choice of discrete distribution
relies on the specific situation. For common discrete random variables, we provide several
pieces of information. First, we list the parameters required to specify the distribution. Then,
we outline possible values that can be realized by a random variable that follows the given
distribution. We also provide a short description of the distribution paired with an
illustration. Then, we supply an example of the application of the random variable. We
conclude with the native Excel functionality for generating random values from the
probability distribution.

Integer Uniform Distribution

Parameters: lower (l), upper (u)

Possible values: l, l + 1, l + 2, . . ., u − 2, u − 1, u
 Description: An integer uniform random variable assumes that the integer values
between l and u are equally likely.

Example: The number of philanthropy volunteers from a class of 10 students may be an

integer uniform variable with values 0, 1, 2, . . ., 10.

Native Excel: RANDBETWEEN(l, u)

Discrete Uniform Distribution

Parameters: set of values

Possible values:

Description: A discrete uniform random variable is equally likely to be any of the specified
set of values .

Example: Consider a game show that awards a contestant a cash prize from an envelope
randomly selected from six possible envelopes. If the envelopes contain $1, $5, $10, $20,
$50, and $100, respectively, then the prize is a discrete uniform random variable with
values {1, 5, 10, 20, 50, 100}.

Native Excel: CHOOSE(RANDBETWEEN(1, k), , , . . ., )

Custom Discrete Distribution

Parameters: set of values and corresponding weights

such that

Possible values:

Description: A custom discrete distribution can be used to create a tailored distribution

to model a discrete, uncertain quantity. The value of a custom discrete random variable
 is equal to the value with probability .

Example: Analysis of daily sales for the past 50 days at a car dealership shows that on 7
days no cars were sold, on 24 days one car was sold, on 9 days two cars were sold, on 5
days three cars were sold, on 3 days four cars were sold, and on 2 days five cars were
sold. We can estimate the probability distribution of daily sales using the relative
frequencies. An estimate of the probability that no cars are sold on a given day is 7/50 =
0.14, an estimate of the probability that one car is sold is 24/50 = 0.48, and so on. Daily
sales may then be described by a custom discrete distribution with values of {0, 1, 2, 3, 4,
5} with respective weights of {0.14, 0.48, 0.18, 0.10, 0.06, 0.04}.

Native Excel: Use the RAND() function in conjunction with the VLOOKUP function
referencing a table in which each row lists a possible value and a segment of the interval
[0, 1) representing the likelihood of the corresponding value. Figure 11.25 illustrates the
implementation for the car sales example.

Figure 11.25
Native Excel Implementation of Custom Discrete Distribution

Binomial Distribution

Parameters: trials (n), probability of a success (p)

Possible values: 0, 1, 2, . . ., n

Description: A binomial random variable corresponds to the number of times an event

successfully occurs in n trials, and the probability of a success at each trial is p and
 independent of whether a success occurs on other trials. When , the binomial is
also known as the Bernoulli distribution.

Example: In a portfolio of 20 similar stocks, each of which has the same probability of
increasing in value of , the total number of stocks that increase in value can be
described by a binomial distribution with parameters and .

Native Excel: BINOM.INV(n, p, RAND())

Hypergeometric Distribution

Parameters: trials (n), population size (N), successful elements in population (s)

Possible values: max{0, n + s − N}, . . ., min{n, s}

Description: A hypergeometric random variable corresponds to the number of times an

element labeled a success is selected out of n trials in the situation where there are N
total elements, s of which are labeled a success and, once selected, cannot be selected
again. Note that this is similar to the binomial distribution except that now the trials are
dependent because removing the selected element changes the probabilities of selecting
an element labeled a success on subsequent trials.

Example: A certain company produces circuit boards to sell to computer manufacturers.

Because of a quality defect in the manufacturing process, it is known that only 70 circuit
boards out of a lot of 100 have been produced correctly and the other 30 are faulty. If a
company orders 40 circuit boards from this lot of 100, the number of functioning circuit
boards that the company will receive in their order is a hypergeometric random variable
with , , and . Note that, in this case, between 10 (= 40 + 70 − 100)
and 40 (=min{40, 70}) of the 40 ordered circuit boards will be functioning. At least 10 of
the 40 circuit boards will be functioning because at most 30 (= 100 − 70) are faulty.
 Native Excel: Insert the file Hypergeometric into your Excel workbook, modify the
parameters in the cell range B2:B4, and then reference cell B6 in your simulation model to
obtain a value from a hypergeometric distribution. This file uses the RAND() function in
conjunction with the VLOOKUP function referencing a table in which each row lists a
possible value and a segment of the interval [0, 1) representing the likelihood of the
corresponding value; the probability of each value is computed with the HYPGEOM.DIST
function. Figures 11.26 illustrates the Hypergeometric file with the parameter values for
the circuit board example.

Figure 11.26
Excel Template to Generate Values from a Hypergeometric Distribution

Negative Binomial Distribution

Parameters: required number of successes (s), probability of success (p)

Possible values: 0, 1, 2, . . ., ∞
Description: A negative binomial random variable corresponds to the number of times
that an event fails to occur until an event successfully occurs s times, given that the
probability of an event successfully occurring at each trial is p. When , the negative
binomial is also known as the geometric distribution.

Example: Consider the research and development (R&D) division of a large company. An
R&D division may invest in several projects that fail before investing in 5 projects that
succeed. If each project has a probability of success of 0.50, the number of projects that
fail before 5 successful projects occur is a negative binomial random variable with
parameters and .

Native Excel: Insert the file NegativeBinomial into your Excel workbook, modify the
parameters in the cell range B2:B3, and then reference cell B5 in your simulation model to
obtain a value from a negative binomial distribution. This file uses the RAND() function in
conjunction with the VLOOKUP function referencing a table in which each row lists a
possible value and a segment of the interval [0, 1) representing the likelihood of the
corresponding value; the probability of each value is computed with the NEGBINOM.DIST
function. The following figure illustrates the implementation for the R&D project example.

Figure 11.27
Excel Template to Generate Values from a Negative Binomial Distribution
Poisson Distribution

Parameters: mean (m)

Possible values: 0, 1, 2, . . .

Description: A Poisson random variable corresponds to the number of times that an

event occurs within a specified period of time given that m is the average number of
events within the specified period of time.

Example: The number of patients arriving at a health care clinic in a hour can be modeled
with a Poisson random variable with , if on average 5 customers arrive to the store
in a hour.

Native Excel: Insert the file Poisson into your Excel workbook, modify the parameter in
cell B2, and then reference cell B4 in your simulation model to obtain a value from a
Poisson distribution. This file uses the RAND() function in conjunction with the VLOOKUP
function referencing a table in which each row lists a possible value and a segment of
the interval [0, 1) representing the likelihood of the corresponding value; the probability of
 each value is computed with the POISSON.DIST function. The following figure illustrates
the implementation for the health care clinic example.

Figure 11.28
Excel Template to Generate Values from a Poisson Distribution

Chapter 11: Monte Carlo Simulation: Appendix 11.1 Common Probability Distributions for Simulation
Book Title: Business Analytics, Third Edition
Printed By: Jordan Miles ([email protected])
© 2019 Cengage Learning, Cengage Learning

© 2020 Cengage Learning Inc. All rights reserved. No part of this work may by reproduced or used in any form or by any means -
graphic, electronic, or mechanical, or in any other manner - without the written permission of the copyright holder.