OM725 / OM726

Creating Value through Operations and

Supply Chain Management

Session 3: Forecasting

OM725 / OM726 Session 3 Page 1

 Project Teams: A-Section
3 students haven’t yet signed up for a team…
Teams should select a company to analyze and email their proposal to me by Feb 4th
• Only 1 person needs to email the team’s proposal to me, but please remember to cc your entire
team on the email

Review OSCM Session #1 slides for details on project requirements, which companies
can not be selected, etc.

OM725 / OM726 Session 3 Page 6

 Project Teams: B-Section
Confirming everyone has a team slot?
Teams should select a company to analyze and email their proposal to me by Feb 4th
• Only 1 person needs to email the team’s proposal to me, but please remember to cc your entire
team on the email

Review OSCM Session #1 slides for details on project requirements, which companies
can not be selected, etc.

OM725 / OM726 Session 3 Page 7

 Forecasting

OM725 / OM726 Session 3 Page 8

 How Easy is it to Forecast?
“The stock market has successfully predicted nine of the last five
recessions.” Paul Samuelson

“There are three kinds of lies: lies, damned lies, and statistics.” Mark Twain?

Benchmarking from PRTM: Best-in-class forecasting

accuracy (actual sales turned out to be + / - 10% of
the forecasted figure) 3 months out?

OM725 / OM726 Session 3 Page 9

 Why Forecast?

Ikea Couch (MTS) Yanko Design Couch (MTO)

Forecasting is required when demand is unknown and supply lead-time (sourcing,
production, and/or delivery lead-time) is longer than the lead-time customers expect
to receive product
• In turn, this requires that you manufacture or position product in advance of the sale
The greater the ability to quickly react to changes in demand, the less accurate
your forecasting needs to be, and the less you need to forecast further out in time
Note: Forecasting is used to predict independent demand, not dependent demand
In the business world, less ability to forecast accurately is generally “solved” with
inventory, but that strategy can have limitations…
OM725 / OM726 Session 3 Page 10
 Forecasts are a Balancing Act

Costs of forecasts that are too high Costs of forecasts that are too low
▪ Money lost holding excess inventory ▪ Lost sales
▪ Obsolescence cost of inventory that is never ▪ Unhappy customers – who may purchase a
sold competitor’s product
▪ Capacity spent making products no one
wants to buy

OM725 / OM726 Session 3 Page 11

 Forecasting Techniques
Five Basic Techniques:
• Moving Average
• Weighted Moving Average
• Exponential Smoothing
• Linear Regression
• Seasonality Analysis

Which technique should you use?

How do you know which technique works best?

OM725 / OM726 Session 3 Page 12

 Moving Average

Simple
Moving dt + dt -1 + dt -2 + ... + dt - n
Ft +1 =
Average n

4 Days 3 Days 2 Days 1 Day

4-Day Moving Average
Ago Ago Ago Ago
4-day Moving = ( 137.1 + 123.6 + 134.9 + 160.0 ) / 4
Average Forecast for 137.1 123.6 134.9 160.0
Thursday = 138.9
4-day Moving = ( 123.6 + 134.9 + 160.0 + 140.4 ) / 4
Average Forecast for 123.6 134.9 160.0 140.4
Friday = 139.7
OM725 / OM726 Session 3 Page 13
 Weighted Moving Average

Weighted
Moving Ft +1 = at dt + at-1dt-1 + at-2dt-2 + ... + at-ndt-n
Average

How do you determine the weights?

4 Days 3 Days 2 Days 1 Day

4-Day Weighted Moving Average
Ago Ago Ago Ago

4-day Weighted = (137.1 * 0.1) + (123.6 * 0.2) +

137.1 123.6 134.9 160.0 (134.9 * 0.2) + (160.0 * 0.5)
Moving Average
(Sun) (Mon) (Tues) (Weds)
Forecast for Thursday = 145.4

4-day Weighted = (123.6 * 0.1) + (134.9 * 0.2) +

123.6 134.9 160.0 140.4 (160.0 * 0.2) + (140.4 * 0.5)
Moving Average
(Mon) (Tues) (Weds) (Thurs)
Forecast
OM725 /for Friday
OM726 Session 3 = 141.5 Page 14
 Forecast Bias

Month Forecast Actual Demand

January 5000 6000
February 4500 5400
March 6667 8000
April 7000 8400

Is there anything wrong with this forecast?

Is there anything good about this forecast?

OM725 / OM726 Session 3 Page 15

 Exponential Smoothing
A weighted average method which uses the forecast error (actual vs forecast)
from previous periods to adjust this period’s forecast
• “The most used of all forecasting techniques” (?)
• Integral part of automated forecasting
• Alpha (“α”) must be between 0 and 1
• Alpha is generally a small number (0.1 to 0.3), but no hard and fast rules

OM725 / OM726 Session 3 Page 16

 Example: Exponential Smoothing
Sunrise Bakery: Bread Production
Use exponential smoothing to forecast sales this week:
• Last week’s forecast was 22,000 loaves, and
• Only 21,000 loaves were actually demanded

Alpha = 0.10

Ft = Ft-1 + α * (At-1 – Ft-1)

Forecast = 22,000 + 0.10 * (21,000 - 22,000)
Forecast = 22,000 + (0.10 * -1000)
Forecast = 22,000 + -100
Forecast = 21,900 loaves
OM725 / OM726 Session 3 Page 17
 Forecasting using SLOPE and INTERCEPT
slope Y-axis
Use excel functions: intercept
INTERCEPT and SLOPE Y = mx + b
Dependent Independent
variable variable

Where:
Y = dependent variable (the FORECAST; the “y” range in excel)
X = independent variable (the DATA USED TO FORECAST; this is the “x” range in
excel, time period “1” to current time period “n”)

Excel will then calculate:

• b = Y intercept
• m = Slope of the line

OM725 / OM726 Session 3 Page 18

 Example: Forecasting by Fitting a Line to the Data
One Independent Variable

Month 1 2 3 4 5
Demand 700 760 780 790 ?
Use Excel Formulas “Slope” and “Intercept”

=SLOPE (demand cell range, time cell range “month”) = 29

=INTERCEPT (demand cell range, time cell range “month”) = 685

Forecast for Month “5” (May) = y-intercept + (month * slope)

Forecast for Month “5” (May) = 685 + (5 * 29) = 830

Note: The months in this equation are essentially arbitrary, but the intervals between the months are not

OM725 / OM726 Session 3 Page 19

 Linear Regression with One Independent Variable

slope Y-axis
intercept
Y = mx + b
Dependent Independent
variable variable

The underlying formula is the same formula as the basic line-slope formula
• Regression models just tell you how well you’ve fit a line to the data

Regression results describe how well the x-variable(s) explain the y-variables
OM725 / OM726 Session 3 Page 20
 Multiple Regression Techniques

Multiple Regression:
Often, more than one independent variable can / should be used to predict
future demand
In this case, you can use multiple regression
• Analogous to linear regression analysis, but with multiple independent variables
• Multiple regression is supported by statistical software packages, including Excel
Essentially the same as the “simple” line-slope formula (Y = mX + b), however in
this case:
• X, T, Z, etc. are the multiple independent variables used to forecast Y
• m, n, p, etc. are the coefficients calculated by the regression formula
• b is the intercept calculated by the regression formula

Equation: Y = b + (m * X) + (n * T) + (p * Z) + ….

Independent variables
OM725 / OM726 Session 3 Page 21
 Multiple Regression – Example

Mark Price, the new productions

manager for Speakers and
Company, needs to find out which UNIT SALES PRICE $/UNIT ADVERTISING
YEAR
variable most affects the demand 1 386 287 598
for their line of stereo speakers. He
2 686 215 826
is uncertain whether the unit price
3 889 211 1,115
of the product or the effects of
4 1,308 213 1,408
increased marketing are the main
drivers in sales and wants to use 5 1,151 215 1,200

regression analysis to figure out 6 1,199 204 1,308

which factor drives more demand 7 889 232 902

for its particular market. Mark 8 1,115 219 1,115

collected information over the past 9 986 215 703

12 years and created the following 10 1,239 211 902

table showing his data: 11 921 224 703

12 815 241 688

OM725 / OM726 Session 3 Page 22

 Multiple Regression – Example
Problem 3-30
Mark Price, the new productions manager for Speakers and Company, needs to find out which variable
most affects the demand for their line of stereo speakers. He is uncertain whether the unit price of the
product or the effects of increased marketing (advertising) are the main drivers in sales and wants to
use regression analysis to figure out which factor drives more demand for its particular market.
Pertinent information was collected by an extensive marketing project that lasted over the past 12
years ….
c. Predict average yearly speaker sales for Speakers and Company based on the regression
results if the price was $297 per unit and the amount spent on advertising (in thousands) was
$896.

OM725 / OM726 Session 3 Page 23

 Multiple Regression – Example
Problem 3-30

Note that the coefficient for

price is negative

OM725 / OM726 Session 3 Page 24

 Multiple Regression – Example
Problem 3-30 R Square: How well the regression line
approximates the real data, essentially how
well the dependent variable (”Y”) is
explained by the independent variable(s).

Ideally this should be at least 0.6

The lower the P-Value, the higher the likelihood that the coefficient is valid
• Example: P-Value of 0.036 indicates there is only a 3.6% chance that the result occurred by chance
OM725 / OM726 Session 3 Page 25
 Forecasting – Sensitivity Analyses

What’s the one thing you know Forecast

about a forecast? 250

200

150

100

Solution: Don’t rely on just one 50

forecast; run a set of forecasts 0

2018 2019 2020 2021 2022 2023 2024 2025
and determine the impact – High Base Low

“sensitivity” – of your solution

OM725 / OM726 Session 3 Page 28

 Dell Case

OM725 / OM726 Session 3 Page 29

 Forecasting – Summary
Which forecasting technique should you use?
Forecasting is a fundamental step in planning independent demand
• Forecast effort should be proportional to the magnitude of the decision being made
Many ways to forecast – average, weighted average, exponential smoothing,
regression, etc.
Some tips:
• Plot your data
• Watch for forecast bias and seasonality or other trends
• Back-cast to see how well your (proposed) forecast model works
• More data is better
• Watch for outliers – keep or delete, but have a good reason for doing what you do
• Try to understand the true drivers of your forecast – aim for causation (not correlation)
• All forecasts have errors – run sensitivity analyses especially if the stakes are high

OM725 / OM726 Session 3 Page 30

 Forecasting – Summary (continued)
Whether your business needs to forecast or not depends on:
• How flexible and fast your Supply Chain can react to demand (and changes in demand)
• Customer expectations for delivery lead-time once they have placed their order
• How much can / does the customer customize the product

The best way to win the forecasting game?

OM725 / OM726 Session 3 Page 31

 Next Class

Read: Chapter 11 (Inventory) and the section on “Measuring Sourcing

Performance” in Chapter 13

Homework:
• Q3-3, 3-12, and 3-13 in MH Connect

OM725 / OM726 Session 3 Page 32