Week 5 – Predictive Analytics: Tools for Decision Making

Summary Transcripts

Making Decisions Under Uncertainty

How to compare and evaluate decisions when we do not know their impact on our profits or cost?
Simulation toolkit is a helpful tool to tackle complex business decisions under uncertainty. Each
decision results in outcomes, in terms of objective function value and key performance indicators.

Comparing Decisions
The difference between making decisions in low uncertainty settings and high uncertainty settings is
given below:
Low uncertainty settings High uncertainty settings

In low uncertainty settings, companies know In high uncertainty settings, irrespective of how
exactly how much profit they will make and how you model random demand values, it leads to
much resources it will consume. random profit values.
Making decisions in low uncertainty settings: Making decisions in high uncertainty settings:
• Calculate the objective function value and • Calculate the distribution for any key
determine if the decision is feasible performance indicator
(R,N) → 150*R + 160*N • Compare distributions of outcomes
• Select the best objective function value
max 150*R + 160*N

Roadmap for Making Decisions

Here’s the roadmap for making decisions in high uncertainty environments:

High uncertainty settings

Decide on reward and risk measures

For each competing decision, use simulation to

estimate reward and risk measures
Use optimization with reward as an objective
function and risk measure as constraints

Making Decisions in High Uncertainty Situations

Example: Wireless Data Plan

One of our consultants has to make a decision about the family data plan based on usage and cost.
Her options are:

Business Analytics: From Data to Insights Page 1 of 7

 Old Plan:

$10 per GB of data in a given month

Alternate Plan:

$20 per up to 20GB of data per month

• $15 per GB exceeding 20GB provided

• There is no discount or roll over to the next month for unused data

Reward and Risk

Identify key performance indicators that may be affected by decisions; use one of them as the
objective and the rest of them as constraints. The expected value of profit or cost is a measure of the
attractiveness of a decision.

The major risk in the data plan example is the standard deviation of the actual monthly payment. A
decision-maker may prefer lower values of the standard deviation of monthly payment; however, risk
may be different for different decision makers.

Monthly Payment for Any Value of Data Usage

Under alternate plan (new data plan), the monthly payment value can be expressed as:

Usage U (in GB) Monthly payment value P (in $)

U≤20 160

U>20 160 + 15*(U-20)

We can combine both cases using one Excel formula:

Final Cost = Base Price + IF (Condition, Choice 1, Choice 2)

P = 160 + IF (U>20,15*(U-20),0)

Expected Value of Monthly Payment

Directly substituting the expected value will only work when P is a linear function of U (as in the old
data plan).
For example, if we substitute the expected value of U = 23 in the expression, P = 160 + IF (U>20,15*(U-
20),0), we get:
P = 160 + 15*(23-20) = $205

This is the wrong way to go.

Suppose U takes only two values:
• 18GB with a probability of 50%
• 28GB with a probability of 50%
• The expected value of U is 23GB and standard deviation is 5GB

Business Analytics: From Data to Insights Page 2 of 7

Business Analytics: From Data to Insights Page 3 of 7
 In this case, P can be calculated as follows:
Usage U (in GB) Monthly payment value P (in $)
18 160

28 160 + 15*(28-20) = 280

Expected value of P = 0.5*160 + 0.5*280 = 220

Introduction to Simulation
What are the reward and risk measures that describe our new plan?

How do we estimate them?

Simulation is a tool for converting probability distributions of random factors we cannot perfectly
control (data usage) into probability distributions for outcomes we're interested in (monthly
payment).

Use the probability distribution of U as input, run the simulation, to get the probability distribution of
P as the output.

Concept of Simulation

A random instance of the input variable is generated in each step of a simulation. Simulation runs can
be repeated as many times as necessary to generate sample distribution of output values.

Sample distribution can be analyzed to determine estimates for the expected value and standard
deviation. Excel can be used for both running the simulation and for analyzing the result.

Simulation Toolkit

• Part of standard Excel installation on Windows

• On Mac (see https://support.microsoft.com/en-us/kb/2431349), users can use a similar free
software called StatPlus: mac LE, (http://www.analystsoft.com/en/products/statplusmacle/)
• On Google Sheets: An equivalent add-on called XLMiner Analysis ToolPak is available

Steps for Running a Simulation

1. Set up your Excel sheet with the data you already have.
2. Under the Data tab, click Data Analysis on the extreme right.
3. Select Random Number Generator from the list.
4. Fill out the details required and click OK (for Excel sheet, Data_0.xlsx, we had the data input
as shown on the left).

Comparing Results for Different Number of Simulation Runs

The more simulation runs we conduct, the closer the sample mean and standard deviation will be to
the true value.

For n = 10, the simulation results show a mean of $252.93 and standard deviation of $92.19, whereas
for n = 100, the simulation results show a mean of $220.16 and standard deviation of $58.24.

Business Analytics: From Data to Insights Page 4 of 7

 Comparing Random Seed Values

Longer simulation runs produce more precise estimates for the reward and risk measures.

n = 10 Seed = 123 Seed 1826 Seed = 19104

Sample mean, GB 25.05 19.48 24.72

Sample standard deviation, GB 7.79 5.21 3.20

n = 1000 Seed = 123 Seed 1826 Seed = 19104

Sample mean, GB 23.28 23.08 23.04

Sample standard deviation, GB 4.88 4.90 4.96

Histograms for Simulation Results

Histograms of simulation results allow us to add a full picture of what happens with our random
variables. A bin is a range of values for a simulated random variable.

We create histograms using Histograms from data-analysis tool for columns of data usage as well as
monthly payment. Keep the Chart Output option checked for graphical representation of frequency
of each bin.

Visualizing Simulation Results

The histograms above tell us there is about a 25% chance the payment amount will be exactly $160
under the new plan.

In many settings, simulation is the tool that must be used to gain insights like this into the nature of
future outcomes, and histograms are useful visual complements to simulations.

Business Analytics: From Data to Insights Page 5 of 7

 Making Decisions in High Uncertainty Settings
Roadmap for making a Making decisions in the data plan example
decision
Decide on reward and risk • Reward measure = Expected monthly payment
measures • Risk measure = Standard deviation of monthly payment

For each competing decision, • Estimate of expected monthly payment = $220.20

use simulation to estimate • Estimate of standard deviation of monthly payment = $58.24
reward and risk measures

Use optimization with reward • Large and complex simulations can be run using commercial
as an objective function and software
risk measure as constraints • A compiled list of commercial simulation software is
available at:
http://www.orms-today.org/surveys/Simulation/Simulation.html

Simulation in Practice
The Wireless Data Plan example uses one random input and analyzes one random output. Simulation
can be used in the models with many random inputs and many random outputs.

Examples of successful use of simulations are available (articles published in Interfaces). Analytics
approaches, like simulation and optimization, are often used in combination to form an effective
solution.

Europa Solvent Industries (ESI): Business Context

Europa Solvent Industries:

• Supplies 35 tons of a chemical solvent each day

• Receives €10000 per ton
• Distributed random variable with a mean of 35 tons and a standard deviation of 7 tons
• Daily production cost is €9500 per ton of solvent

If ESI’s daily production exceeds 35 tons, it must sell the extra solvent to a discount retail chain at a
price of €9000 per ton.

If ESI’s daily production falls short of 35 tons, it is penalized €2000 for each ton below 35.

Europa Solvent Industries – Algebraic Expression

What is the total profit ESI will make the day its production (P) is 31 tons?

On the day production falls short of 35 tons, ESI will:

a. Sell the day’s production at a price of €10000 per ton

b. Incur a penalty of €2000 for each ton of shortfall
ESI’s regular revenue = €10000*31 = €310000
ESI’s production cost = €9500*31 = €294500
ESI’s penalty cost = €2000*(35-31) = €8000
ESI’s total profit = €310000 - €294500 - €8000 = €7500

Business Analytics: From Data to Insights Page 6 of 7

 An algebraic expression for ESI’s total daily profit value (V) as a function of its daily production (P)

If ESI’s daily production falls short of 35 tons or (P<35), ESI will:

a. Incur production cost of 9500*P

b. Earns revenue of 10000*P
c. Incurs penalty cost of 2000 * (35-P)

ESI’s total daily profit value (V) = 10000*P - 9500*P - 2000 * (35-P)

= 500*P - 2000 * (35-P)

= 500*P – 70000 + 2000*P

= 2500*P - 70000

If ESI’s daily production exceeds 35 tons or (P≥35), ESI will:

a. Incur production cost of 9500*P

b. Earns revenue of 10000*35 + 9000 * (P-35)
c. Sells the first 35 tons at the price of €10000 per ton, and the remaining P-35 tons at the
discount price of €9000 per ton

ESI’s total daily profit value (V) = 10000*35 + 9000 * (P-35) - 9500*P

= - 500*P + 35000

An algebraic expression for the total daily profit value (V):

IF(P<35, 2500*P – 70000, -500*P + 35000)

Europa Solvent Industries (ESI) - Answers

Question:

If n = 100 and seed = 123, use the simulation results to answer the following questions:

• What is the estimate of the expected value of total daily profit?

• What is the estimate of the standard deviation of total daily profit?

Answer:

If n = 100 and seed = 123, then

• Estimate of the expected daily profit = €8752

• Standard deviation of the daily profit = €9415

Business Analytics: From Data to Insights Page 7 of 7