Summer Class | Management Science

FORECASTTING TECHNIQUES

Forecasting

Forecasting is the process of making predictions of the future

based on past and present data. This is most commonly by
analysis of trends. Prediction is a similar, but more general
term.

Sales Forecast

A sales forecast which is the estimate of how much the com-

pany will actually sell. It is an estimate of expected sales rev-
enue within a specific time frame, such as quarterly, monthly,
or yearly. Seasonal

Seasonality refers to regular fluctuations in demand that occur

Forecasting Horizons
annually for a large number of products. Seasonality is quite
important when it comes to trends and outdoor items.
 Long term forecasting tends to be completed at high
levels in the organization.
 Medium term forecasting tends to be several months
up to 2 years into the future and is referred to as interme-
diate term.
 Short term forecasting is daily up to months in the fu-
ture.

Forecasting Methods

 Qualitative forecasting techniques are subjective,

based on the opinion and judgment of consumers and ex-
perts.
 Quantitative forecasting models are used to forecast
future data as a function of past data.

Time Series Analysis Irregular Variations

Events or a sequence of events that are not anticipated to oc-

 A specific way of analyzing a sequence of data points col-
cur again can frequently have an impact on demand.
lected over an interval of time.
 Time is a crucial variable because it shows how the data
Random Variations
adjusts over the course of the data points as well as the
final results.
The remaining unexplained changes in demand after all other
factors have been taken into account are known as random
Demand Chart
variations. This is commonly referred to as noise.

Demand Chart is a graphical representation of historical de-

Naive
mand data for a given time frame. It helps forecast future de-
mand by visualizing trends, seasonal patterns, and possible
 The simplest forecasting method.
cyclical behavior in the data.
 The forecast for the next period is set at the actual de-
mand for the previous period.
Elements of Demand Chart
Formula
Trends

It is a constant shift of the demand either up or down. This

Y t =Y t −1
could be connected to the life cycle of the product.
Example

Month Actual
Jan (1) 200
Feb (2) 300
Mar (3) 200
Apr (4) 400
May (5) 500
Jun (6) 600
Jul (7) -

Moving Average

A technique that averages a number of recent actual values,

updated as new values become available.
Cycle
Formula
A data pattern that appears frequently and lasts longer than a
year is called a cycle. The political situation, consumer confi- A t−n +… At −2+ A t−1
dence, interest rates, and other market conditions are fre- F t=MA n =
quently linked to them. n
Note: n is the number of periods

Example

Week Demand
1 125
2 175
3 150
4 150
5 10

Solution
Summer Class | Management Science
 Moving Average Example data has both linear trend and seasonal pattern. This method
 Assume n=2 is also called Holt-Winters exponential smoothing.

Week Demand Exponential Smoothing

1 125
2 175 This method uses a combination of the last actual value and
3 150 (125 + 175)/2 = 150 the last forecast to produce the forecast for the next period.
4 150 (175 + 150)/2 = 162.5
5 10 (150 + 150)/2 = 150 Exponential Smoothing Formula
(150 + 160)/2 = 155
There are two versions of the same formula for calculating the
Weighted Moving Average
exponential smoothing:

More recent values in a series are given more weight in com-

puting the forecast.  F t=F t + α ( A t−F t )
 F t=α ( At ) +(1−α ) Ft
Formula

t Where:

∑ ( X i∗W i )  α = Alpha/Smoothing constant

WMA= i=t −n  F t = Forecast value
n
At = Actual value
∑Wi 

i =1
Sample Problem

 t = total number of observations Exponential Smoothing α =0.2

 i=t−n = series starts a total – window size At Ft
 X i = value of i previous day
th

Week Sales Forecast F t=α ( At ) +(1−α ) Ft

 W i = weight assigned to i trailing days’ price
th

1 39 39.00 F t+ 1=0.2 ( A t ) + 0.8(F ¿¿ t)¿

 i=t = the sum of all weights
2 44 39.00 F t=0.2 ( 39 ) +0.8 (39.00)
Example 3 40
4 45
 Assume n=2, W 1=0.7, W 2 =0.3 5 38
6 43
7 39
Week Demand
1 125
2 175 Exponential Smoothing α =0.2
3 150 At Ft
4 150
5 10 Week Sales Forecast F t=α ( At ) +(1−α ) Ft
Solution 1 39 F t+ 1=0.2 ( A t ) + 0.8(F ¿¿ t)¿
2 44 39.00 F 2=A t
 Assume n=2, W 1=0.7, W 2 =0.3
3 40 40.00 F 3=0.2 ( 44 ) +0.8(39.00)
Week Demand 4 45 40.00 F 4=0.2 ( 40 ) +0.8(40.00)
1 125
2 175 5 38 41.00 F 5=0.2 ( 45 ) +0.8 (40.00)
3 150 (0.7)(175) + (0.3)(125) = 160
4 150 (0.7)(150) + (0.3)(175) = 157.5 6 43 40.40 F 6=0.2 ( 38 ) +0.8(41.00)
5 10 (0.7)(150) + (0.3)(150) = 150
(0.7)(160) + (0.3)(10) = 157
7 39 40.92 F 7=0.2 ( 43 )+ 0.8(40.40)
8 40.54 F 8=0.2 ( 39 ) +0.8(40.92)
Exponential Smoothing
Forecasting Accuracy and Evaluation
There are three main methods to estimate exponential
smoothing. They are: Forecast accuracy is the measure of how accurately a given
forecast matches actual sales. Forecast bias describes how
 Simple or single exponential smoothing much the forecast is consistently over or under the actual
 Double exponential smoothing sales. Common metrics used to evaluate forecast accuracy in-
 Triple exponential smoothing clude Mean Absolute Percentage Error (MAPE), Mean squared
error (MSE) and Mean Absolute Deviation (MAD).
Simple or Single Exponential Smoothing
Importance:
 If the data has no trend and no seasonal pattern, then
this method of forecasting the time series is essentially Forecasting accuracy is crucial because it underpins efficient
used. This method uses weighted moving averages with operations, informed decision-making, and strategic growth
exponentially decreasing weights. initiatives. By improving the reliability of predictions about the
 This forecasting method is most widely used of all fore- future, businesses can enhance their competitive advantage
casting techniques. It requires little computation. and adapt more effectively to changing market conditions.

[ ]
|( e t )|
Double Exponential Smoothing
Period
Actual De-
mand
Fore-
cast et |(e t )| et
2
×100 %
 This method is also called exponential smoothing. Holt's Dt
trend corrected or second-order exponential smoothing.
1 63 68
 This method is used for forecasting the time series when
2 59 65
the data has a linear trend and no seasonal pattern. The
3 54 61
primary idea behind double exponential smoothing is to
4 65 59
introduce a term to take into account the possibility of a
series showing some form of trend. This slope component
Here are what need to do:
is itself updated through exponential smoothing.

Triple Exponential Smoothing  Step 1: Calculate the error as e t =Dt−F t (the differ-
ence between the actual demand and the forecast) for
In this method, exponential smoothing applied three times. any period t and enter the values in the table above.
This method is used for forecasting the time series when the
Summer Class | Management Science
 Step 2: Calculate the absolute value of the errors calcu- Time consuming and expen-
Easy to understand
lated in step 1 [i.e., |(e t )|], and enter the values in the ta- sive
Provide only estimates or
ble above. Flexible
approximations
2
 Step 3: Calculate the squared error (i.e., et ) for each pe- Convenient
Does not guarantee an opti-
mal solution
riod and enter the values in the table above.
 Step 4: Calculate ¿ for each period and enter the value
Monte Carlo Simulation
under its column in the table above.

[ ]
Monte Carlo Simulation – is a technique for selecting num-
|( e t )| bers randomly from a probability distribution (i.e., “sampling”)
Period
Actual De-
mand
Fore-
cast et |(e t )| et
2
×100 % for use in a trial (run by a computer) of a simulation. The
Dt Monte Carlo technique is not a type of simulation model but
1 63 68 -5 5 25 7.94% rather a mathematical process used within a simulation.
2 59 65 -6 6 36 10.17%
3 54 61 -7 7 49 12.96%  a mathematical technique that generates the range of
4 65 59 6 6 36 9.23% possible outcomes for an uncertain event
 Monte Carlo Simulation Technique: The more you sample,
Solution: the more accurate the sampling range and better estima-
tion.
Calculation for Accuracy measure:
True Random Numbers – produced only through physical
 MAD = The average of what we calculated in step 2 (i.e., process e.g. spinning the wheel
the average of all the absolute error values) =
Pseudorandom Numbers – computer-generated using nu-
(5+6 +7+6) 24
= =6 merical technique
4 4
 MSE = The average of what we calculated in step 3 (i.e., History
the average of all the squared error values) =
(25+36 +49+36) 146  named by Hungarian mathematician John Von Neumann
= =36.5 (working with bombs during World War II)
4 4  Monte Carlo Simulation’s name was derived from a casino
 MAPE = The average of what we calculated in step 4 = place in Monaco.

(7.94 %+10.17 % +12.96 % +9.32 %) 40.3 %

= =10.07 %The artificially created random numbers must have the
4 4 following characteristics:

SIMULATION MODELING  The random numbers must be uniformly distributed.

 The numerical technique for generating random numbers
should be efficient.
What is Simulation Model?
 The sequence of random numbers should not reflect any
pattern.
 One of the most widely used quantitative approaches to
business decision making.
Monte Carlo Application (but not limited to):
 A computer-based technique used to model the operation
1. Risk Analysis
of a system or process so that experimentation can be
2. Decision Support
conducted to evaluate the consequences of alternative
decisions. This technique employs computer programs to
3 Basic Steps on How to Run Monte Carlo Simulation
model the operation and perform simulation computa-
tions.
 Provides the decision maker with the opportunity to ex- 1. Identifying the dependent variable (to be predicted or to
periment with certain parts of a decision problem and an- be measured) and independent variable (causes an effect
alyze the likely consequences of alternative decisions. on dependent variable).
2. Assign range of likely values for each of the independent
variables through historical data or subjective judgment.
Common Applications of Simulation
3. Run Simulation.
1. New product development
Application in Risk Analysis and Decision Support
2. Reservation systems
3. Inventory systems
4. Queuing systems Risk analysis is the process of predicting the outcome of a
5. Flows Simulation decision in the face of uncertainty. It involves identifying and
assessing factors that could negatively affect the success of a
business or project. It is concerned with both the probability
Types of Simulation
of a loss and the magnitude of a loss.
 Static Simulation Modes
Decision Support System (DSS) is a computer program ap-
 Dynamic Simulation Model
plication used to improve a company's decision-making capa-
 Discreet-Event Simulation Model
bilities. It analyzes large amounts of data and presents an or-
ganization with the best possible options available.
Static Simulation Models
Terms to remember:
Models that are based on independent trials in which the re-  Parameters – values that are set constant, a set of facts
sults for one trial do not affect what happens in subsequent or a fixed limit that establishes or limits how something
trials. In this sense, the system being modeled does not can or must happen or be done
change or evolve over time e.g., porta com project and inven-  Probabilistic/Variable Inputs – values or inputs that
tory simulation. are not known with certainty; information that is based on
relational patterns and the likelihood of a certain outcome
Dynamic Simulation Model  Equally Likely – Equally Likely to Occur, each outcome
of an experiment occurs with equal probability
Simulation models that must take into account how the sys-  Probability Distribution – a statistical function that de-
tem changes or evolves over time. e.g., queuing models. scribes all the possible values and probabilities for a ran-
dom variable within a given range
Discreet-Event Simulation Model  Random Numbers – randomly chosen; Computer-gen-
erated random numbers are called pseudorandom
numbers. The difference between random numbers and
In situations where the arrivals and departures of customers
pseudorandom numbers is primarily philosophical, and
are events that occur at discrete points in time, this simula-
we use the term random numbers regardless of whether
tion is used e.g., ATM queuing models.
they are generated by a computer.

Advantages And Disadvantages of Using Simulation

Static Models

Advantage Disadvantage
1. PortaCom Project
Summer Class | Management Science

Project: PortaCom’s product design group developed a proto-

type for a new portable printer. The new printer features an in-
novative design and has the potential to capture a significant
share of the portable printer market.

Case: PortaCom would like an analysis of the first-year profit

potential for the printer. Because of PortaCom’s tight cash
flow situation, management is particularly concerned about
the potential for a financial loss.

Preliminary marketing and financial analyses provided the for

lowing selling price, first-year administrative cost and first The cost of direct labor, the cost of parts and the first-year de-
year advertising cost, variable inputs were also provided: mand for the printer are not known with certainty. At this
stage of the planning process, PortaCom’s best estimates of
Parameter (constant) In- Probabilistic or Valuable these inputs are $45 per unit for the direct labor cost, $90 per
puts Inputs
unit for the parts cost and 15 000 units for the first-year de-
Selling Price Per Direct Labor
$249 $45 mand.
Unit Cost/Unit
Administrative
$400,000 Parts Cost/Unit $90
Cost A. Base-Case Scenario
First Year De-
Advertising Cost $800,000 15,000
mand (Units)
Profit=( 249−C1 −C2 ) x−1,000 , 00 0
Profit=( 249−45−90 ) (15,000)−1,000 ,00 0
Profit=$ 710,000
Thus, the base-case scenario leads to an anticipated
profit of $710,000.

Flowchart for the PortaCom Simulation

Direct Labor Cost per
Probability
Unit
$43 0.1
$44 0.2
$45 0.4
$46 0.2
What-If Analysis: With a selling price of $249 per unit and $47 0.1
administrative plus advertising costs equal to $400 000 +
$600 000 = $1 000 000, the PortaCom profit model is:

Profit=( $ 249−Direct labor cost per unit−Pa r ts cost per unit )( Demand )−$ 1,000 ,00 0
Letting:
C 1 = direct labor cost per unit
C 2 = parts cost per unit
x = first year demand
the profit model for the first year can be written as follows:

B. Worst-Case Scenario
Profit=( 249−C1 −C2 ) x−1,000 , 00 0
Using the what-if approach to risk analysis, we selected values
Profit=( 249−C1 −C2 ) x−1,000 , 00 0
for direct labor cost per unit (c1), parts cost per unit (c2) and Profit=( 249−4 7−10 0 ) (1 , 500)−1,000 , 00 0
first-year demand (x), and then calculated the resulting profit.
To generate such values, we must know the probability distri- Profit=$ 710,000
bution for each variable.
So, the worst-case scenario leads to a projected loss of
Flowchart for the PortaCom Simulation $847,000.
Direct Labor Cost per
Probability
Unit What-If Analysis
$43 0.1 Worst Best
$44 0.2 Direct Labor Cost/Unit $47 $43
$45 0.4 Parts Cost/Unit $100 $80
$46 0.2 First Year Demand (Units) 1,500 28,500
$47 0.1
Estimate First Year Probabil-
ity

C. Best-Case Scenario

Profit=( 249−C1 −C2 ) x−1,000 , 00 0

Profit=( 249−4 3−8 0 ) ( 28 ,500)−1,000 , 00 0
Profit=$ 2,591 , 000
So, the best-case scenario leads to a projected profit of
$2,591,000.

Note: The what-if analysis indicates that either a sub-

stantial loss or a substantial profit is possible. But the
problems with What-If is it does not indicate the likelihood
of the various profit or loss values. In particular, we do
not know anything about the probability of a loss.

Random Numbers and Generating Probabilistic Values

Random numbers and the probability distributions associated

with each variable are used to generate representative values.
Summer Class | Management Science
parts cost. With a uniform probability distribution, the fol-
To illustrate how to generate these values, we need to intro- lowing relationship between the random number and the
duce the concept of computer-generated random numbers. associated value of the parts cost is used.
Computer generated random numbers are randomly selected
decimal numbers from 0 up to, but not including, 1. The com- PartsCost =a+r (b−a)
puter-generated random numbers are all equally likely and are
uniformly distributed over the interval from 0 to 1.
Where:
r = random number between 0 and 1
a = smallest value for parts cost
b = largest value for parts cost

PartsCost =80+ r ( 100−80 )

PartsCost =80+ r ( 20 )
Random generation of 10 values for the parts cost per unit
Direct Labor Cost
Trial Random Number
($)
1 0.6983 45
2 0.0082 43
3 0.6799 45
4 0.8898 46
5 0.6515 45
6 0.3976 45
7 0.0642 43
8 0.0377 43
9 0.5739 45
10 0.5827 45

For the first trial, let’s try 0.3125:

PartsCost =80+ 0.3125 ( 20 )

PartsCost =86.25
C. In Terms of First-Year Demand

Because first-year demand is normally distributed with a

mean of 15 000 units and a standard deviation of 4500,
we need a procedure for generating random values from
a normal probability distribution.

¿ NORMINV ¿
¿ NORMINV ¿
For example:
A. In Terms of Labor Costs  If RAND() function generates the random number
0.7005, the Excel function shown in equation (12.4)
Random number intervals for generating values of direct labor will provide a first-year demand of 17,366 units.
cost per unit
Direct Labor Probabil- Interval of Random  If RAND() generates the random number 0.3204,
Cost per Unit ity Cost ($) equation (12.4) will provide a first-year demand of
$43 0.1 0.0 but less than 0.1 12,900.
$44 0.2 0.1 but less than 0.3
$45 0.4 0.3 but less than 0.7 Random generation of 10 values for first-year demand
$46 0.2 0.7 but less than 0.9 Direct Labor Cost
Trial Random Number
$47 0.1 0.9 but less than 1.0 ($)
1 0.7005 17,366
An interval of random numbers is assigned to each possi- 2 0.3204 12,900
ble value of the direct labor cost in such a fashion that 3 0.8968 20,686
the probability of generating a random number in the in- 4 0.1804 10,888
terval is equal to the probability of the corresponding di- 5 0.4346 14,259
rect labor cost. 6 0.9605 22,904
7 0.5646 15,732
Random generation of 10 values for the direct labor cost per 8 0.7334 17,804
unit 9 0.0216 5,902
Direct Labor Cost 10 0.3218 12,918
Trial Random Number
($)
1 0.6983 45 Note: Random numbers less than 0.5 generate first-
2 0.0082 43 year demand values below the mean and that random
3 0.6799 45 numbers greater than 0.5 generate first year demand
4 0.8898 46 values greater than the mean.
5 0.6515 45
6 0.3976 45 Running The Simulation Model
7 0.0642 43
8 0.0377 43
Running the simulation model means implementing the se-
9 0.5739 45
quence of logical and mathematical operations described in
10 0.5827 45
the flowchart:
Clearly with only ten trials we cannot simulate the labor
Flowchart of the PortaCom Simulation
cost probability distribution accurately. But if we were to
repeat these trials a sufficiently large number of times
then we would be able to accurately simulate this distri-
bution.

B. In Terms of Part Costs

Because this random variable has a different probability

distribution than direct labor cost, we use random num-
bers in a slightly different way to generate values for
Summer Class | Management Science
Net Profit: Gross Profit −Holding Cost
Case 1: D > Q

When demand is greater than the replenishment level (D >

Q), Q units are sold, and a shortage cost is imposed for each
of the D - Q units of demand not satisfied.

Gross Profit:
Gross profit per unit × unit sold
Shortage Cost:
Shortage cost per unit ×(D−Q)
Net Profit: Gross Profit −Shoratge Cost
Example

The product is a home ventilation fan distributed by the Butler

Electrical Supply Company. Each fan costs Butler $75 and
sells for $125. Thus, Butler realizes a gross profit of $125
The following are the results of the first trials for each vari-
-$75 = $50 for each fan sold. Monthly demand for the fan
able:
is described by a normal probability distribution with a mean
of 100 units and a standard deviation of 20 units.
Direct labor cost: C 1=45
Parts Cost: C 2=86.25 Addition information:

First-Year demand: x=17,366 If monthly demand is less than the replenishment level, an in-
ventory holding cost of $15 is charged for each unit that is not
And with the profit equation: sold. However, if monthly demand is greater than the replen-
ishment level, a stock-out occurs and a shortage cost is in-
Profit=( 249−C1 −C2 ) x−1,000,000 curred. Because Butler assigns a goodwill cost of $30 for each
customer turned away, a shortage cost of $30 is charged for
each unit of demand that cannot be satisfied. Management
we obtain: would like to use a simulation model to determine the average
monthly net profit resulting from using a particular replenish-
Profit=(249−45−86.25)(17 366)−1,000,000 ment level. The company assigns a replenishment level of
100.
Profit=$ 1,044,847
 For Month 1
Here is the summary of results for ten trials:
Demand = 92

Gross Profit:
Gross profit per unit × unit sold
Holding Cost:
Holding cost per unit ×(Q−D)
Net Profit: Gross Profit −Holding Cost
Gross Profit: $ 50 ×92=$ 4,600
Holding Cost: $ 15 × ( 100−92 )=$ 120
Net Profit: $ 4,600−$ 120=$ 4,480
 For Month 2
Note: Clearly with only ten trials, we cannot expect to repli-
Demand = 120
cate the decision problem accurately. A much larger num-
ber of trials is needed to allow the results to better ap-
Gross Profit:
proximate to the probability distributions we have used. To
do this we need to use computer-based simulation. Gross profit per unit × unit sold
Shortage Cost:
2. Inventory Simulation Shortage cost per unit ×(D−Q)
Inventory Simulation describes how simulation can be used to
Net Profit: Gross Profit −Shortage Cost
establish an inventory policy for a product that has an uncer-
tain demand. Gross Profit: $ 50 ×100=$ 5 , 0 00
 Replenishment Level (Q) – refers to the point at which
Shortage Cost: $ 30 × ( 1 20−100 ) =$ 60 0
inventory or stock is replenished to maintain a steady Net Profit: $ 5 , 0 00−$ 600=$ 4 , 4 0 0
supply and avoid stockouts
 Demand – generated through the use of an excel func-  For Month 3
tion:
¿ NORMINV ¿ Demand = 80
 Holding Cost – refers to the costs associated with hold-
Gross Profit:
ing or carrying a given level of inventory
 Shortage Cost – the cost of having a shortage and not Gross profit per unit × unit sold
being able to meet demand from stock Holding Cost:

Case 1: D ≤ Q
Holding cost per unit ×(Q−D)
Net Profit: Gross Profit −Holding Cost
When demand is less than or equal to the replenishment level
(D ≤ Q), D units are sold, and an inventory holding cost is in- Gross Profit: $ 50 ×80=$ 4 , 0 00
curred for each of the Q - D units that remain in inventory.
Holding Cost: $ 15 × ( 100−80 )=$ 30 0
Gross Profit: Net Profit: $ 4 ,0 00−$ 30 0=$ 3 , 77 0
Gross profit per unit × unit sold
Holding Cost:  For Month 4
Holding cost per unit ×(Q−D) Demand = 110
Summer Class | Management Science

Gross Profit: Given Data After Simulation:

Gross profit per unit × unit sold
Total number of customers (N) : 900 customers
Shortage Cost:
Average waiting time per customer (AWT) : 1.59
Shortage cost per unit ×(D−Q) minutes
Net Profit: Gross Profit −Shortage Cost Total simulation time (T) : 2261.5
minutes

Gross Profit: $ 50 ×100=$ 5 , 0 00 Solution:

Shortage Cost: $ 30 × ( 1 10−100 )=$ 3 00
Net Profit: $ 5,000−$ 3 00=$ 4 , 7 00 AWT × N
Total Waiting Time (W) =
W =1.59 minutes/customer × 900 customers=1 , 431 minutes
 For Month 5
Average Number in Waiting Line = W /T
Demand = 100
Average Number in Waiting Line = 1 , 431/2261.50
Gross Profit: $ 50 ×100=$ 4,600 Average Number in Waiting Line = 0.63 CUSTOMERS
There’s neither holding costs nor shortage costs.
This means that, over the entire simulation period, there are
Net Profit: $ 5,000−$ 0=$ 5 ,0 00 usually less than one customer waiting in line at any given
time. Hence, the ATM operates smoothly with minimal waiting
Hold-
Gross Short- Net times for customers. This is a good indicator of efficient ser-
Month Demand Sales ing
profit age cost profit
cost vice and customer satisfaction.
1 92 92 $4,600 $120 0 $4,480
2 120 1010 $5,000 0 $600 $4,400
3 80 80 $4,000 $300 0 $3,700 Other Simulation Issues
4 110 100 $5,000 0 $300 $4,700
5 100 100 $5,000 0 0 $5,000 1. Computer Implementation
$23,6 $22,2
TOTAL 502 472 $420 $900
00 80
$4,72 $4,45
The use of spreadsheets for simulation has grown rapidly
AVER-
AGE 100.4 94.4 $84 $180 in recent years, and third-party software vendors have
0 6
developed spreadsheet add-ins that make building simu-
Discrete-Event Model lation models on a spreadsheet much easier. These add-in
packages provide an easy facility for generating random
values from a variety of probability distributions and pro-
Discrete-event simulation (DSS) is a method used to model
vide a rich array of statistics describing the simulation
the operation of a system as a sequence of events in time. It
output.
allows for the analysis of complex systems where events oc-
cur at discrete points in time. Discrete-event simulation, or
Two popular spreadsheet add-ins are Crystal Ball from
DES, is intended to simulate systems where events occur at
Decisioneering and @RISK from Palisade Corpora-
specific, separable instances in time. DES contrasts with a
tion. Although spreadsheets can be a valuable tool for
continuous simulation where events are tracked continuously.
some simulation studies, they are generally limited to
smaller, less complex systems.
The defining characteristics of discrete-event simulation are
as follows:
2. Verification and Validation
 Entities – Objects that move through the system (e.g.,
Verification is the process of determining that the com-
customers, products, vehicles).
puter procedure that performs the simulation calculations
 Evens – Actions that change the state of the system
is logically correct. Verification is largely a debugging task
(e.g., arrival, departure, service completion).
to make sure that no errors are in the computer proce-
 State – The collection of variables that describe the sys-
dure that implements the simulation.
tem at any point in time.
 Clock – A variable that keeps track of the current simula-
Validation is the process of ensuring that the simulation
tion time.
model provides an accurate representation of a real sys-
 Event List – A list of events that are scheduled to occur,
tem.
often organized in a priority queue based on event time.
 Random Variable – Used to model uncertainty in event
 A long period of real time is represented by a short
timings and other aspects of the system.
period of simulated time.
The general steps to perform a DES are:

Initialization: Set the initial state of the system and initialize

the event list.

Event Handling: Process events in chronological order. For

each event the corporation must update the system state.
Schedule future events as needed. Collect statistical data if
necessary.

Termination: Continue processing events until a stopping

condition is met (e.g., a certain time has elapsed or a specific
number of events have been processed).

Output Analysis: Analyze the collected data to draw conclu-

sions about the system's performance.

1. Queuing Simulation Mode (ATM Queuing)

ATM Waiting Line Simulation Model

We are simulating an ATM where customers arrive to withdraw

money. We want to see how long they have to wait and how
many people are usually in line.

The primary goal of the ATM waiting line simulation is to un-

derstand and optimize the performance of the ATM system.
Specifically, we are interested in:

Average Number of Customers in the Waiting Line – This

tells us how crowded the ATM typically is, indicating how well
the system manages customer flow.