LECTURE 5

MONTE CARLO SIMULATION BY EXCEL

(PART 1)

1
AGENDA

 Random number generation by Excel: uniform distribution

 Examples in Excel

2
RANDOM NUMBER TABLE

3
RANDOM NUMBERS IN EXCEL

 Generation of random numbers has the following properties:

 Uniformly distributed, greater than or equal to 0 and less than 1.
 Probabilistically independent.
 Easy to generate another set of random numbers for a new simulation.
 Useful as the “building block” of a simulation.

4
RANDOM NUMBERS IN EXCEL:
RAND FUNCTION

 Formula for generating a single random number:

 =RAND( )
 Parentheses (with nothing between them) are to indicate function without
arguments.
 When this is copied to another set of cells, a different set of random numbers is
generated.
 See Excel “Random Numbers” worksheet

5
EXPERIMENTING WITH RAND FUNCTION

 Enter “= RAND( )” in one cell.

 Copy it to 499 other cells.
 The numbers change whenever you change the spreadsheet.
 You can obtain a new set of random numbers by pressing “F9”.

6
EXPERIMENTING WITH RAND FUNCTION

 Develop a frequency table to see how the 500 random number are
distributed.
 Enter a column of “bin values” (the upper ends of the categories): 0.1, 0.2,
up to 0.9
 Highlight an adjacent range one cell longer than the bin range.
 Enter formula =FREQUENCY(data,bin) and press Ctrl-Shift-Enter.

7
EXPERIMENTING WITH RAND FUNCTION

 This table shows the frequencies (number of values) in each of several

categories.
 The first frequency is the number of values less than or equal to 0.1.
 These were found with the FREQUENCY function (an array function).

8
FREEZING THE RANDOM NUMBERS

 Copy the random numbers on the original sheet to a new sheet with the Paste
Special/Values option.
 The pasted copy is now frozen (does not change when F9 key is pressed).
 Freezing random numbers is sometimes useful in simulation models, but not
always.
 See “Freezing Numbers” worksheet.

9
SIMULATING VALUES FROM PROBABILITY
DISTRIBUTIONS

 RAND() generates random number in [0,1), such that the probability of

generating a number within [𝑎, 𝑏] is equal to 𝑏 𝑎 (0 𝑎 𝑏 1).
 This is somewhat like drawing 00-99 out of a black box, except that we can also
draw numbers like 0.5155, 0.56564, etc.
 Goals:
 Divide [0,1) into intervals that reflect our probability distribution (i.e.
random number assignment).
 Use Excel functions to identify the random observation based on the
generated random number.

10
EXAMPLE 1: MONTE CARLO SIMULATION
REVISIT WITH EXCEL

 Last class: Higgins Plumbing and Heating

 Use simulation to estimate the average weekly sales.

Heater Sales Probability

4 0.12
5 0.10
6 0.18
7 0.24
8 0.16
9 0.14
11
10 0.06
Total 1.00
STEP 1: ASSIGN RANDOM NUMBER INTERVALS

 When we specify the random number intervals, it will seem as if the

upper limit of one row overlaps with the lower limit of the next row.
 However, the probability that we get any specific number like 0.1, 0.5,
etc with the RAND() function is essentially 0.
 Don’t need to worry about RAND() being equal to the lower/upper
limits.

12
STEP 2: DETERMINE RANDOM OBSERVATION
(LOOKUP)

 LOOKUP(lookup_value, lookup_vector, result_vector)

 Look for lookup_value in lookup_vector and return a value from the same position
in result_vector.
 lookup_value = Our random number
 lookup_vector = Random number assignment
 result_vector = Observations
 Remark:
1. The values in lookup_vector must be placed in ascending order
2. If the LOOKUP function can't find lookup_value, Excel rounds DOWN to
match the closest value in lookup_vector .
13
 Implication of Remark 2  Use Lower Limit for lookup_vector .
HIGGINS - LOOKUP

 See “Higgins - Lookup” tab.

14
 STEP 2: VLOOKUP (ALTERNATIVE)

 Alternatively, we can use the VLOOKUP function.

 VLOOKUP(lookup_value, table_array, col_index_num, range_lookup)
 Searches for a value in the leftmost column of a table, and then returns a value in the
same row from a column you specify in the table.
 Remark:
1. Lookup_value: Excel lookups for this in the FIRST column of the table.
2. Table_array is the table of information.
3. Col_index_num of x means that we return values from the x-th column of the table.
4. Range_lookup is either TRUE (approximate match) or FALSE (exact match).
• If “TRUE” or omitted, then VLOOKUP works like LOOKUP and rounds down lookup_value.
• If “FALSE”, VLOOKUP will find an exact match, or return the error value #N/A if not found. 15
 STEP 2: VLOOKUP (ALTERNATIVE)

 Alternatively, we can use the VLOOKUP function.

 VLOOKUP(lookup_value, table_array, col_index_num, range_lookup)
 Searches for a value in the leftmost column of a table, and then returns a value in the
same row from a column you specify in the table.
 Remark:
1. Lookup_value: Our random number.
2. Table_array: First column of table should be the lower limit.
3. Col_index_num: Column with heater sales.
4. Range_lookup: TRUE

16
VLOOKUP FUNCTION: EXAMPLE
 See “Higgins - Vlookup” worksheet
 Observe that “Upper limit” = “Cumulative probability” in the last
example  Don’t need two different columns.

17
HIGGINS – SALES AND OUT-OF-STOCK

 Suppose Jerry restocks and starts each week with 8 heaters.

 Run a simulation over 100 weeks to compute the average weekly
demand.
 On average, how many customers are unable to buy a heater
because Jerry is out-of-stock?
 IF(logical_test, value_if_true, value_if_false)
 Logical_test: demand <= 8
 Value_if_true: 0 (everyone buys a heater)
 Value_if_false: demand – 8 (# of unhappy customers)
 How can we modify the IF function if we only want to count the
number of weeks that Jerry is out-of-stock?
18
EXAMPLE 2: ART’S NEWSSTAND

 Art sells newspaper. He cannot foretell the daily demand 𝐷 (i.e. the number of
newspapers that he can sell in any day).
 Each morning, Art orders 𝑞 newspapers from his wholesaler at $3 each.
 Art sells the newspaper at $5 each. If he has unsold units at the end of the day,
he receives $0.50 per scrap (i.e. an unsold piece).
 Art wants to maximize his average daily profit. How many newspapers, 𝑞, should
he order each morning?

19
EXAMPLE 2: ART’S NEWSSTAND

 Decision Variable: 𝑞 = daily order quantity.

 Art’s random daily profit:
 If 𝐷 𝑞, then Art earns $5 𝑞 3 𝑞.
 If 𝐷 𝑞, then Art earns $5 𝐷 0.50 𝑞 𝐷 3 𝑞.
 Combining the two scenarios yields Art’s random daily profit:
5 min 𝑞, 𝐷 0.5 max 𝑞 𝐷, 0 3 𝑞.

Revenue Scrap value, if any Purchase cost

20
EXAMPLE 2: ART’S NEWSSTAND

 Art wants to determine the optimal order quantity to maximize his average
daily profit.
 Why focus on average profit?
 Taking expectation on Art’s random daily profit yields Art’s average daily profit,
namely:
E5 min 𝑞, 𝐷 0.5 max 𝑞 𝐷, 0 3 𝑞
 Maximizing Art’s average daily profit is expressed by
max E 5 min 𝑞, 𝐷 0.5 max 𝑞 𝐷, 0 3 𝑞

 How can we determine the optimal order quantity?

21
EXAMPLE 2: ART’S NEWSSTAND

 Art cannot control the daily demand.

 Hence, the daily demand 𝐷 can be described by a random variable.
 An idea without computer:
 Order 80 newspaper over the next month.
 Compute the average daily profit when Art orders 𝑞 80 newspapers.
 Repeat for different values of 𝑞.

22
EXAMPLE 2: ART’S NEWSSTAND

 Alternatively, determine the probability distribution of past demand and

run simulations.
 Q1: Suppose he orders 80 newspaper each morning. Can we compute his
average profit?
 Be careful! Our goal is to compute Art’s average profit, NOT average
demand.

23
EXAMPLE 2: ART’S NEWSSTAND

 Art tells us that the daily demand follows a uniform distribution between 30
pieces and 150 pieces, inclusive.
 That is, 𝐷 ~ 𝑈 30,150 and every integer between 30 and 150 occurs with
EQUAL probability.
 When we have a uniform distribution between a and b, we can use the
RANDBETWEEN(a,b) function as a shortcut!

24
 EX 2, Q1: COMPUTING AVERAGE PROFIT
Average over profits
in Col C

Profit
=5*MIN($B$1,B7)
+0.5*MAX(0,$B$1-B7)
-3*$B$1

25

Simulated Demand
=RANDBETWEEN(30,150)
EXAMPLE 2: ART’S NEWSSTAND

 Average Profit as a Function of the Number of Observations, given that the daily
order quantity is 𝑞 80.

26
EX Q2: OPTIMAL ORDER QUANTITY

 Q2: How many newspaper should Art order daily to maximize his
average profit?
 Method 1: Data table
 Method 2: Scenario Manager

27
SUMMARY

 Monte Carlo simulation via Excel

 Computing an optimal decision using simulation

28