Simulation

and the Shahi Dawat Case

Overview of the Case

What is the decision problem presented in the case?

What are the issues Rajat must consider in deciding among the alternative choices?

Accounting Model of Shahi Dawats Monthly Earnings What are the monthly fixed costs? L = labor costs (random, between `3,58,400 and `4,36,800) U = rent, utilities, other unavoidable costs, `3,99,500 What are the monthly variable costs? F = total food costs M = number of meals served in month F = `1,100 * M What is the monthly total cost? L + U + F = L + 3,99,500 + 1,100 * M
2

Shahi Dawats Monthly Earnings, continued

What is the monthly revenue? R = monthly revenue, P = price of meal R = P *M What are the monthly earnings? X = monthly earning = revenue costs = P * M ( L + 3,99,500 + 1,100 * M ) = (P 1,100 ) * M L 3,99,500 Which of these quantities are random variables? P = price of prix fixe meal M = number of meals sold L = labor cost X = monthly earnings (X is a function of random variables, so it is a random variable)
3

Assumptions Regarding the Random Variables

P = price of the prix fixe meal We assume that P obeys the following discrete probability distribution.

M = number of meals sold per month We assume that M obeys a Normal distribution with m = 3,000, s = 1,000. L = labor cost per month We assume that L obeys a uniform distribution with a minimum of `3,58,400 and maximum of `4,36,800
4

The Behavior of the Random Variables, continued

X = earnings per month We do not know the distribution of X. It depends on P,M, L: X = (P 1,100 ) * M L 3,99,500

Always ask the following questions in any management analysis: How realistic is this model? How good are the assumptions?

What question do we want to ask & answer about the random variable X? What is the shape of the probability distribution of X? What is E(X) ? What is SD(X) ? What is P(X `75,000) ? What is P(X `2,50,000) ? Other questions to ask? And at the end: What would you do if you were Rajat?
6

Simulation of Shahi Dawat Monthly Earnings for February

Generate a value of M (number of meals served) from the probability distribution N (3,000, 1,000)

Generate a value of P (price of prix fixe meal) that obeys the given discrete probability distribution
Generate a value of L (labor cost) that obeys the uniform distribution in the range [504000 , 686000] Calculate the monthly earnings X: X = (P 1,100 ) * M L 3,99,500 Run this n times. We used n = 1,000 for illustration purpose This will generate as output n numbers x1, x2, ..., xn.

What shall we do with this output?

7

What to do with output created by the simulation? We have as output the n numbers x1, x2, ..., xn

Create an estimate of the shape of the underlying probability distribution of X (earnings). Create an estimate of the mean m of X Create an estimate of the standard deviation s of X Create an estimate of P(X `75,000)

Create an estimate of P(X `2,50,000)

How to generate a random variable with a given distribution?

Example: Generate the price P according to the given distribution: P 1800 1650 1450 1300 Probability 0.25 0.35 0.30 0.10

We first generate a uniformly distributed random number in [0, 1], and then assign a value to P based on this number: [0, 1] uniform random number falls in range 0.00 --- 0.25 falls in range 0.25 --- 0.60 falls in range 0.60 --- 0.90 falls in range 0.90 --- 1.00 assigned price 1800 1650 1450 1300 Probability 0.25 0.35 0.30 0.10
9

Generating a sequence of prices Trial 1 2 3 4 5 6

o o o

Random Number .662 .923 .300 .377 .362 .499

o o o

Price of Prix Fixe Meal (R) 1350 1200 1550 1550 1550 1550
o o o

Assignment Table: [0, 1] uniform random number 0.00 --- 0.25 0.25 --- 0.60 0.60 --- 0.90 0.90 --- 1.00

P 1800 1650 1450 1300

10

How to Generate Random Numbers from a

given Continuous Probability Distribution

Most simulation software packages can generate random numbers from a variety of continuous distributions, such as the Normal distribution, the uniform distribution, etc. The user needs to specify the type of distribution and the parameters (m and s for the Normal, a and b for the uniform) However, it is worthwhile to point out how the computer accomplishes this task

11

The Probability Density Function f(y) of the Random Variable

f(y)
0.3 0.2 0.1 0 0.0 2.0 4.0

6.0

8.0

10.0

12

Cumulative Distribution Function F(y) of the Random Variable

1.0

F(Y)
0.8 0.6 0.4 0.2 0.0 0.0 2.0 4.0

6.0

8.0

10.0
13

Steps to generate a random number that

follows a given CDF F(y)

1. First, use a random number generator to generate a number u that obeys a uniform distribution between 0.0 and 1.0. 2. Place the number u on the vertical axis of the graph of the CDF F(y) of the given distribution. Then find the point y on the horizontal axis whose CDF value F(y) is equal to u. 3. The number y generated this way has the desired CDF F(y).

WHY?
14

Suppose the [0, 1] uniform random number we get Which y values are more likely to be picked? happens to be u=0.826

1.0 0.8 0.6

F(y)
0.4 0.2 0.0
y = 6.851

10
15

1000 trials of Shahi Dawat monthly simulation

Trial Random Price of Random Number 1 2 3 4 5 6 7 999 1,000 Number 0.6616 0.9233 0.3003 0.3774 0.3621 0.4994 0.0554 0.3012 0.9151 Meal Number 1350 1200 1550 1550 1550 1550 1700 1550 1200 0.1024 0.0971 0.1691 0.2059 0.8036 0.2903 0.2164 0.2901 0.292 Meals Random Served 1,732.13 1,701.95 2,042.38 2,179.43 3,854.52 2,447.60 2,215.43 2,446.91 2,452.37 Number 0.803 0.649 0.262 0.609 0.409 0.208 0.288 0.627 0.365 Labor Costs 6,500.94 6,221.59 5,517.08 6,147.86 5,784.60 5,417.72 5,564.93 6,181.58 5,704.71 Monthly Earnings -969.23 -3,408.81 5,805.77 6,202.87 19,129.32 8,944.26 10,378.95 8,175.24 109.75 10,526.49
16

Sample Mean

Discussion of Simulation Model Output

From n trials, we get n monthly earnings: x1, x2, ..., xn They are the realized values of random variable X

Armed with these numbers, we want to estimate: the shape of the probability distribution of X the mean m of X the standard deviation s of X

the probability that X will lie in a given range, i.e., P( a X b) for given values of a and b

17

Discussion of Simulation Model Output, continued

A histogram of the trial values x1, x2, ..., xn is a good estimate of the shape of the probability distribution of X

Suppose that we want to estimate P(a X b) for some given values of a and b
Let m be the number of values in the n observations x1, x2, ..., xn that are in the range between a and b. Let
p m . Note that p is simply the fraction of the n

observations that are in the range between a and b. Then p is an estimate of the probability P (a X b).

18

Histogram of Monthly Earnings from the Shahi Dawat Simulation Model

0.12

0.10

Probability

0.08

0.06

0.04

0.02

Monthly Earnings (R 00s)

19

Estimating the mean m and the standard deviation s

Estimating m

x1 x2 xn x n

For example, from the spreadsheet we obtain

Estimating s

( x1 x )2 ( x2 x )2 ( xn x )2 s n 1
For example, from the spreadsheet we obtain

20

Simulation of Monthly Earnings at Shahi Dawat using Crystal Ball

An input random variable is modeled in Crystal Ball as an "assumption" cell For each assumption cell, the user must choose the type of distribution and the parameters The output quantity we are interested in (i.e., the monthly earnings) is modeled as a "forecast" cell Crystal Ball will automatically compute the histogram, sample mean, sample standard deviation, and all sorts of other very useful information for the cell
21

Crystal Ball Output: Monthly Salary

Crystal Ball Report

22

Interpretation of Simulation Results: Monthly Earnings at Shahi Dawat

What is the shape of the distribution of monthly earnings? What is the expected monthly earnings? estimate = R6,63,196 The standard deviation? estimate = R5,81,378

What is the probability that Rajat will earn less than R75,000 in a given month? estimate = 23% What is the probability that Rajat will earn more than in consulting (R2,50,000) in a month? estimate = 70%

What would you do if you were Rajat?

23

Simulation of Annual Earnings at Shahi Dawat

Instead of looking just at monthly earnings at Shahi Dawat, it would be better to look at annual earnings

What should the annual model assume?

All assumptions for the monthly model, plus: there are twelve distinct random variables M1, ... , M12 for the twelve distinct number of meals served in each month these random variables will be independent and identically distributed (i.i.d.) the random variable P (price of prix fixe meal) remains the

same over the entire year

the random variable L (labor cost in month) remains the same over the entire year
24

Simulation of Annual Earnings at Shahi Dawat, continued.

Generate random values of P , L , and M1, M2, according to their respective distributions Calculate the monthly earnings, for month i: Xi = (P 1100 ) * Mi L 3,99,500

..., M12

for i = 1, ..., 12

A = annual earnings = X1 + X2 + . . . + X12

How is this model different from the monthly model?

25

Simulation Output from Crystal Ball Annual Proprietor Salary

26

Interpretation of Simulation Results: Annual Earnings under Proprietorship

What is the shape of the distribution of the annual earnings?

What is the expected annual earnings? The standard deviation of the annual earnings? Estimate of mean = R__________ Estimate of SD = R__________ How do these annual statistics compare to the monthly statistics?
What is an estimate of the probability that Rajat will earn less than R9,00,000? ______ What is an estimate of the probability that Rajat will earn more than R30,00,000? ______ What would you do if you were Rajat?
27

Summary Table of Some Relevant Estimates

CHOICE Estimate of Mean Estimate of SD 95% Confidence Interval of Mean Estimate of Pr(Salary R30,00,000) Estimate of Pr(Salary R30,00,000) CONSULTING PROPRIETORSHIP R_________

R30,00,000
R0 R30,00,000 0.0

R__________ R________ ____

100%

_____%

0%

____%

28

Simulation of Annual Earnings at Shahi Dawat Under a Partnership

What are the terms of the partnership?

29

Rajat's Monthly Salary Under Financial Partnership (R)

Rajat's Monthly salary (R00)

10,000
8,000

6,000
4,000 2,000 0 0 2,000

6000 8000 10000 12000 14000 16000 Shahi Dawat Restaurant Monthly Earnings (R00)
30

4,000

Simulation Output from Crystal Ball Annual Partnership Salary

31

Interpretation of Simulation Results: Annual Earnings at Shahi Dawat (Partnership)

What are the mean and SD of annual earnings under partnership? Estimate of mean = `-----------estimate of SD = `------------Under partnership, what is the probability that Rajat will earn less than `9,00,000? estimate = _____% Under partnership, what is an estimate of the probability that Rajat will earn more than `30,00,000? estimate = _____% What would you do if you were Rajat?
32

Summary Table of Some Relevant Estimates

CHOICE Estimate of Mean Estimate of SD 95% Confidence Interval of Mean Estimate of Pr(Salary R80,000) Estimate of Pr(Salary $60,000) CONSULTING R30,00,000 PROPRIETORSHIP R_______ PARTNERSHIP R________

R0

R_______

R________

R30,00,000 0.0

R_______ ________

R_________ ________

100%

_______%

______%

0%

_______%

______%
33

Comparison: Partnership versus Proprietorship

What is the likelihood that the proprietorship outperforms the partnership? estimate = _______% Should Rajat go for Shahi Dawat alone, or in partnership with his aunt?

34

Simulation Output from Crystal Ball Annual Proprietor Premium

35

Discussion of the Estimates How good are these estimates?

Intuition suggests that when the number of trials is very large, then the estimates are very reliable
However, to give a more precise answer to this question, we will need to learn sampling theory

36

Some Lessons of Simulation

Simulation attempts to evaluate things that average case analysis and simple formulas cannot Simulation can handle very complex systems and dynamics

Compared with some other tools such as decision tree,

simulation analysis is not very transparent

The results that one can obtain in a simulation are not

precise, due to the inherent randomness in a simulation. Care must be used in interpreting simulation results

37

Some Lessons of Simulation, continued

The output of a simulation model are estimates of the (unknown) true quantities. One can supplement these estimates with confidence intervals and other inferences The question of how many trials or runs of a simulation can become a complex statistical issue Fortunately, with todays computing power, this is not a paramount issue for most problems

38

Some Lessons of Simulation, continued

In practice, one should recognize that gaining managerial confidence in a simulation model will depend on at least three factors:

1. A good understanding of the nature of the management problem 2. Ones ability to use the concepts of probability and statistics correctly 3. Ones ability to communicate these concepts effectively

39