Business Simulation

through Excel
Topics Covered
• Session 1:
• Introduction to Simulation
• Monte Carlo Simulation
• Pratap Car Wash
• Salesmen planning at bargain store
• Session 2: Structured knowledge as per chapters
• Chapter 1: Introduction to Modeling and Simulation
• Chapter 2: Statistical Concepts For Discrete Event Simulation
• Session 3: Chapter 3 to 5
Important Statistical Concepts
• Parameter
• Sample
• Statistics
• Statistical descriptors of performance typically fall into three categories:
• (1) descriptors that help locate the center of a distribution, such as the mean;
• (2) descriptors that measure the spread of a distribution, such as the variance;
and
• (3) descriptors that indicate relative standing within a population, such as the
probability of occurrence or quartile location.

• Estimation: Statistical procedures, similar to the confidence interval

estimate of mean, exist for estimating population variance, probability of
occurrence, and quartile location
OUTPUT DATA ANALYSIS
important concepts
• Confidence Interval Estimate of the Mean

• Precision refers to the accuracy of an estimate. In a confidence interval, the

precision of the estimate is indicated by the width of the confidence interval.

The smaller the confidence interval, the higher is the precision of the estimate.

• Lower the standard error higher the precision

Discrete-Event
Simulation
 Queuing Simulation
Algorithm 1.1 — How to develop a model:
1) Determine the goals and objectives
2) Build a conceptual model
3) Convert into a specification model
4) Convert into a computational model
5) Verify
6) Validate
Typically
Entities an iterative
can represent process
either customers or objects
servers can represent persons or production stations that treat or interact with the entity,
and queues are the holding or waiting position of entities.
The queue size may be assumed to be either finite or infinite.
Requires awareness

• The number of queues and servers present in the real - world

system
• patterns of behavior for entities and servers.
• These behaviors patterns can be derived from our
understanding of historical data
• the past behavior of entities (e.g., probabilistic or deterministic arrival
times)
• servers (e.g., probabilistic or deterministic service times) within the system.

• When historical data are not avail able,

• the researcher has the options of either engaging the real - world system OR
• relying on subject matter experts to express the behavior of entities and
servers.
Verification vs. Validation
• Verification
• Computational model should be consistent with
specification model
• Did we build the model right?
• Validation
• Computational model should be consistent with the
system being analyzed
• Did we build the right model?
• Can an expert distinguish simulation output from
system output?
• Interactive graphics can prove valuable
• State Variables State variables are those measures that characterize a system at a particular
moment or state. Thus, the state of the queue at time t (i.e., Q(t)) may be characterized by the
number of entities in the queue, while the state of the server at time t (i.e., Q(t)) may be
characterized as either busy or idle.
• Events are the arrival to, or departure from, the system by entities or objects. The termination, or
end, of the simulation is also a form of an event, sometimes referred to as a pseudo event since it
is not marked by either the arrival or the departure of an entity.
• Entities and servers are described by attributes.
• Attributes are components of the system state. A server can have an attribute that describes its
state as either busy or idle.
 Single Server

Multi Server Queue

In this system, we
assume that all
servers are
identical, i.e. there
is no difference
which server is
chosen for which
item.
 MODELING
CONTINUOUS
SYSTEMS
In civil engineering for the construction of dam embankment and tunnel
constructions.
In military applications for simulation of missile trajectory, simulation of fighter
aircraft training, and designing & testing of intelligent controller for underwater
vehicles.
etc
Softwares
• There are also mathematical programming
environments, such as MATLAB (produced by The
MathWorks, Inc.)
• IDL (produced by ITT Visual Information Systems),
which are custom - made for just this type of work
Oximoron We are asked to subtract two very similar numbers in the
numerator and divide that result by a tiny number.
This experiment is custom - made to exploit the rounding
errors of digital computation.

• Digital computers cannot, in general, represent continuous (real) numbers, so

how can we possibly use computers to simulate continuous systems?
• Imagine we are simply taking the derivative of t2 at t = 1. The limit demands
that we use ever smaller values of Δt to get better and better estimates of the
derivative.
MONTE CARLO
SIMULATION
four - step process
Step 1 Define a distribution of possible inputs for
each input random variable.
Step 2 Generate inputs randomly from those
distributions.
Step 3 Perform a deterministic computation using
that set of inputs.
Step 4 Aggregate the results of the individual
computations into the final result.
The necessary components for
Four Step Process
1. probability distribution functions (pdfs) for each random variable
2. a random number generator
3. a sampling rule— a prescription for sampling from the pdfs
4. scoring— a method for combining the results of each run into
the final result
5. error estimation— an estimate of the statistical error of the
simulation output as a function of the number of simulation runs
and other parameters.
 Example: Computing Product
Earnings
• earning = (unit price) × (unit costs) - (variable cost s + fixed costs)
• The last 5 years of data for these variables are
shown in Table 5.1 .

An appropriate distribution representation would be a triangular distribution, which is

typically used when only a small amount of data is available to characterize the input
variables
https://magesblog.com/img/magesblog/distributions.png
https://www.johndcook.com/blog/distribution_chart/
 Constructing a triangular
distribution requires three
values:
a minimum (a) ,
a maximum (b), and
2
most likely or mode (c).
(𝑏 − 𝑎)
The max values were chosen as
best guess estimates using subject
matter experts intuitively familiar
with how these variables are likely
to behave or through averaging

a c b
the probability density function
Use MS Excel
• Using these triangular distributions, 10,000 samples
were generated and used to compute predicted
future earnings.

• Using Excel : https://www.youtube.com/watch?v=LOl6GeM5xN4

https://blog.learningtree.com/monte-carlo-methods-excel-part-4-triangle-distribution/
SENSITIVITY ANALYSIS
• The study of how uncertainty in a model’s output
can be assigned to the various sources of input
uncertainty.
• Sensitivity analysis is important as it can help
uncover model errors and identify important
bounds on input variables
Derivative - based approach
• Given ∂Yj/∂Xi where Yj is a output random variable
and Xi is a input random variable, one can see that this
partial derivative can be interpreted as the change in Yj
with respect to Xi, which is consistent with our
definition of sensitivity analysis.
• Derivative - based approaches are only valid at the
point that they are computed. This is acceptable for
linear systems but would be of little value for systems
exhibiting nonlinear behavior.
Visual Assessment
• This method employs a scatter plot where each
input variable in the Monte Carlo simulation is
individually plotted against the output variable and
the resulting pattern is analyzed. The more
structured the output pattern, the more sensitive is
the output variable to that input variable.
• we had four random input variables that
contributed to computing the earnings
random output variable. If we plot each of the
10,000 randomly generated inputs against the
corresponding output using a scatter plot

The scatter plots show that the variable

earnings is more sensitive to the product
of unit_price and unit_sales than the fixed
and variable costs because of its
structured pattern.

From the results of this analysis, one could

set the fixed_costs and variable_costs
inputs to their mean values and rerun the
Monte
Carlo simulation using only unit_sales and
unit_price as random input variables with
little loss in accuracy.
Previous result with 4 input variable result with unit_price and unit_sales
as input variable

Can you calculate CI for mean for the

Confidence Interval for the mean is
results above
[75,223, 75,869]
Sigma- normalized derivatives

The larger the result of this computation, the

more sensitive the output is to this input.

As one can see, the product of unit_sales and unit_price is the most sensitive
parameter and bears out the results of the scatter plot analysis.