CHAPTER 3

Forecasting
Operations Management
Bus Adm 370

1
Agenda

Forecasting
How do we forecast?
Qualitative methods
Delphi …
Quantitative methods
Time Series Data Analysis
Associative Models
Forecast Accuracy
MAD, MSE, MAPE

2
What is Forecasting?

3
Forecasting overview

Forecast: A statement about the future

4
Forecasting in Taco Bell

30% of every sales dollar goes to labor

Labor requirements vary greatly throughout
the day
1988, started to solve the labor allocation problem
Forecasts the arrival rate by 15-minute
Forecasting method: 6-week moving averages
1993-1996, saved $40.34 million in labor costs

5
Applications

We focus on the forecasting of demand

Forecasts are also used to predict
Profits/Revenues
Costs
Prices
Interest rates
Movements of key economic indicators
And more . . .
Concepts and techniques are the same
6
Common Features

Assumes same causal system: past ==> future

Be alert to unplanned occurrences
Forecasts rarely perfect because of randomness
Forecasts more accurate for I see that you will
get an A this semester.
groups vs. individuals
Forecast accuracy decreases
as time horizon increases

7
How do we Forecast?

8
Forecasting Process

“The forecast”

Step 6 Monitor the forecast

Step 5 Make the forecast
Step 4 Gather and analyze data
Step 3 Select a forecasting technique
Step 2 Establish a time horizon
Step 1 Determine purpose of forecast
9
Forecasting Approaches

Qualitative Methods Quantitative Methods

Judgmental forecasts Time series forecasts
Involves intuition, & Associative models
experience Involves mathematical
e.g., forecasting sales on techniques
Internet
e.g., forecasting sales of
Used when situation is color televisions
vague & little data exist Used when situation is
New products ‘stable’ & historical data
New technology exist
Existing products
Current technology
10
Judgmental Forecasts

Executive opinions
A small group of upper-level managers
(marketing, operations and finance) may meet
and collectively develop a forecast.
Sales force opinions
The sales staff or the customer service staff is
often a good source of information. Sales staff
may under-report to reduce quota.
11
Judgmental Forecasts (Con’t)

Consumer surveys
Consumers ultimately determine demand.
Expensive.
Outside opinion
Industrial reports, press.
Delphi method
• Opinions of managers and staff
• Achieves a consensus forecast

12
 Quantitative Forecasting Methods

 Complete formulas
Quantitative
available in Table 3.7 on
Forecasting
page 104 of textbook

Time Series Associative

Forecasting Models

Naïve Averaging Trend Seasonality  Linear regression

 Curvilinear
regression (not req.)
 Multiple regression
 Moving Average  Linear trend  Additive (not req.)
 Weighted Moving – linear trend equation  Multiplicative
Average
– trend-adjusted
 Exponential exponential
smoothing smoothing (no req.)
 Nonlinear trend 13
(not req.)
Time Series Data

14
Time Series Forecasts

Time series : A time series is a time-ordered

sequence of observations taken at regular
intervals over a period of time (hourly, daily,
weekly, monthly, quarterly, annually).
Data : measurements of demand (sales,
earnings, profits, ……)
Example
Year: 1998 1999 2000 2001 2002
Sales: 78.7 63.5 89.7 93.2 92.1

Assumption : future will be like the past

15
Time Series Behaviors

Figure 3.1
Stable

Irregular
variation

Trend

Cycles

90
89
88
Seasonal variations
16
Time Series Behaviors

Trend - A long-term upward or downward movement

in data.
Seasonality - Short-term regular variation related to
calendar or time of day.
Cycle – Wavelike variations lasting more that one year.
Irregular variations - Caused by unusual circumstances,
not reflective of typical behavior.
Random variations - Residual variations after all other
behaviors are accounted for and caused by chance

17
Time Series Data
- Naïve Method
18
Naïve Forecast

Uh, give me a minute....

We sold 250 wheels last
week.... Now, next week we
should sell....

The forecast for any period

=
The previous period’s actual value.

19
Naïve Forecasts

Simple to use
Virtually no cost
Quick and easy to prepare
Data analysis is nonexistent
Easily understandable
Cannot provide high accuracy
Can be a standard for accuracy

20
Uses for Naïve Forecasts

Stable time series data

• Ft = At-1
Seasonal variations
• Forecast for this season = Actual value from last season
Data with trends
• Ft = At-1 + (At-1 – At-2)

where
Ft = Forecast for period t.
At-1 = Actual demand or sales for period t-1.

21
Example 1: Naïve Forecasts

Forecast for period i is the actual value for

period i-1: Fi = Ai-1.
Period i Actual Demand Forecast
1 42
2 40
3 43
4 40
5 41
6 39
7 46
8 44
9 45
10 38
11 40
12 - 22
Solution to Example 1

Period i Actual Demand Forecast

1 42 -
2 40 42
3 43 40
4 40 43
5 41 40
6 39 41
7 46 39
8 44 46
9 45 44
10 38 45
11 40 38
12 - 40

23
Time Series Data
- Averaging
24
Techniques for Averaging

Moving average
Weighted moving average
Exponential smoothing

25
Moving Average

Moving average – A technique that averages

a number of recent actual values, updated as
new values become available. n
 At-i
Ft = MAn = i=1

Where
n
i = an index that corresponds to periods.
n = Number of periods (data points) in the moving
average period.
Ai = Actual value in period i.
MAn = Forecast based on most-recent n periods.
Ft = Forecast for time period t. 26
Example 2: Moving Average

Find moving average with n = 5.

Period i Actual Demand Forecast
1 42
2 40
3 43
4 40
5 41
6 39
7 46
8 44
9 45
10 38
11 40
12 -
27
Solution to Example 2

Start from F6 (forecast for period 6).

Period Actual Forecast
i Demand
1 42 -
2 40 -
3 43 -
4 40 - 43+40+41+39+46
5 41 - =
5
6 39 41.2
7 46 40.6
8 44 41.8
9 45 42.0
10 38 43.0
11 40 42.4 46+44+45+38+40
=
12 - 42.6 5 28
Smooth and Lag

29
Lag Increases with Periods

30
Weighted Moving Average

Assigns more weight to recent observed

values
More responsive to changes
Selection of weights is arbitrary, but weights
must add to one. The values for the weights
are always given.

31
Example 3: Weighted Moving Average

Find weighted moving average using

Fi =0.4Ai-1 + 0.3Ai-2 + 0.2Ai-3 + 0.1Ai-4.
Period i Actual Demand Forecast
1 42
2 40
3 43
4 40
5 41
6 39
7 46
8 44
9 45
10 38
11 40
32
12 -
Solution to Example 3

Start from F5 (forecast for period 5).

Period Actual Forecas

i Demand t
1 42 -
2 40 -
3 43 -
4 40 -
5 41 41.1 = 0.1(42)+.2(40)+.3(43)+.4(40)
6 39 41.0
7 46 40.2
8 44 42.3
9 45 43.3
10 38 44.3
= 0.1(39)+.2(46)+.3(44)+.4(45)
11 40 42.1
12 - 40.8 33
Shown solutions of Example 2 and 3

48

46

44

42

40

38

36
Observed
34 MA
32 WMA

30
1 2 3 4 5 6 7 8 9 10 11 12
34
Exponential Smoothing

Current forecast = Previous forecast + α(Actual -

Previous forecast)

Ft = Ft-1 + (At-1 - Ft-1)

where
Ft = Forecast for period t
Ft-1 = Forecast for period t-1
α = Smoothing constant
At-1 =Actual demand or sales for period t-1
35
Exponential Smoothing (Cont.)

Premise: The most recent observations might

have the highest predictive value.
• Therefore, we should give more weight to the more
recent time periods when forecasting.

Weighted averaging method based on previous

forecast plus a percentage of the forecast error

A-F is the error term,  is the % feedback

36
Example 4: Exponential Smoothing

Period (t) Actual (At) Ft (α = 0.1) Error (A-F) Ft ( α = 0.4) Error (A-F)
1 42
2 40
3 43
4 40
5 41
6 39
7 46
8 44
9 45
10 38
11 40
12

37
Solution to Example 4

Ft = Ft-1 + (At-1 - Ft-1) = (1 – ) Ft-1 +  At-1

• For example: α = 0.1
• A1 = 42 → F2 = 42 (Naïve)
• A2 = 40 → F3 = F2 + α (A2 - F2)
= 42 + 0 .1 × (40 - 42) = 41.8
• A3 = 43 → F4 = F3 + α (A3 - F3)
= 41.8 + 0 .1 × (43 - 41.8) = 41.92

38
Solution to Example 4 (Cont.)

Ft = Ft-1 + (At-1 - Ft-1)

Period (t) Actual (At) Ft (α = 0.1) Error (A-F) Ft ( α = 0.4) Error (A-F)
1 42
2 40 42 -2.00 42 -2
3 43 41.8 1.20 41.2 1.8
4 40 41.92 -1.92 41.92 -1.92
5 41 41.73 -0.73 41.15 -0.15
6 39 41.66 -2.66 41.09 -2.09
7 46 41.39 4.61 40.25 5.75
8 44 41.85 2.15 42.55 1.45
9 45 42.07 2.93 43.13 1.87
10 38 42.36 -4.36 43.88 -5.88
11 40 41.92 -1.92 41.53 -1.53
12 41.73 40.92

39
Picking a Smoothing Constant α

Actual
50
  .4
  .1
45
Demand

40

35
1 2 3 4 5 6 7 8 9 10 11 12

Period
40
Picking a Smoothing Constant α

Using judgment or trial and error

Balancing smoothness and responsiveness

Usually range from 0.05 to 0.5

low α when stable

high α when susceptible to change

41
Time Series Data
- Trend
42
Techniques for Trend

Linear trend
linear trend equation
Trend-adjusted exponential smoothing (Double
exponential smoothing), not required in exams

Nonlinear trend (not required in exams)

43
Linear Trend Equation

Ft

Ft = a + b t

0 1 2 3 4 5 t

Ft = Forecast for period t

t = Specified number of time periods from t = 0
a = Value of Ft at t = 0
b = Slope of the line
44
Calculating a and b

n  (ty) -  t  y
b = 2 2
n  t - (  t)

 y - b t
a =
n

n = Number of periods
y = Value of the time series
t = Specified number of time periods from t = 0

45
Example 5:

Calculator sales for a California-based firm over the

last 10 weeks are shown in the following table.

Week (t) y yt t2
1 700 700 1
2 724 1448 4
3 720 2160 9
4 728 2912 16
5 740 3700 25
6 742 4452 36
7 758 5306 49
8 750 6000 64
9 770 6930 81
10 775 7750 100
 55 7407 41358 385 46
Solution to Example 5

1. Plot the data, and visually check to see if a

linear trend line would be appropriate.
n  (ty) -  t  y  y - b t
b = 2 2 a =
n t - ( t) n
2. n = 10,  t = 55,  y =7407,  ty = 41358, t2 = 385

b= 10(41358) - 55(7407) 413580 - 407385

= ≈ 7.51
10(385) - 55(55) 3850 - 3025
7407 - 7.51(55)
a = ≈ 699.40
10
y = 699.40 + 7.51t
47
Solution to Example 5 (Cont.)

3. Then determine the equation of the trend

line, and predict sales for weeks 11 and 12.

y11 =699.40 + 7.51(11) = 782.01

y12 =699.40 + 7.51(12) = 789.51

48
Solution to Example 5 (Cont.)

800
780
760
740
720
700 Observed
680
Trend line
660
1 2 3 4 5 6 7 8 9 10
49
 True/False Questions

1. Naïve forecasts are essentially subjective forecasts. F

2. Forecasts for individual items tend to be less accurate
than forecasts for groups of similar items.
T
3. In single-factor exponential smoothing, the forecasted
values tend to lag behind the actual values. T
4. Forecasts accuracy decreases as the time horizon
increases.
T
5. Delphi techniques is a forecasting model that
incorporates the use of multiple regression. F
6. Forecasting techniques generally assume an existing
casual system that will continue to exist in the future. T

50
 89

Seasonal variations

Time Series Data

- Seasonality
51
Techniques for Seasonality

Example :
• Winter and summer sports equipment
• Rush hour traffic occurs twice a day
• Theaters and Restaurants often experience weekly
demand pattern
• Banks may experience daily and monthly seasonal
variation.
Seasonality is expressed as variation from
average or trend line.

52
Different Models of Seasonality

Additive model: Seasonality factor is expressed

as a quantity. Simply add or subtract from the
series average
Multiplicative model: Seasonality is expressed
as a percentage of the average (or trend) amount
Seasonal relative: amount by which overall
average is multiplied to generate forecast for this
season.
• Example: A seasonal relative of 1.20 for the
quantity of toys sold in May indicates that May
sales are 20% above the monthly average.

53
Additive Model and Multiplicative Model

Seasonal
Relative 54
Using Seasonal Relatives

Deseasonalize historical observations to get

nonseasonal component
Divide each data by its corresponding seasonal
relative
Seasonalize forecasts when demand has both
trend and seasonal components
Obtain trend estimates
Add seasonality to the trend estimates by multiplying
the trend estimates by the corresponding seasonal
relative
55
Example 6 (P 83)

A manager wants to predict the quarterly demand for

period 15 and 16, which are the 2nd and 3rd quarters of a
particular year. Demand series consists of both trend and
seasonality. The trend portion is Ft = 124 + 7.5t. Quarter
relatives are Q1 = 1.20, Q2 = 1.10, Q3 = 0.75, and Q4 = 0.95.

1. The trend values at t = 15 and t = 16:

F15 = 124 + 7.5(15) = 236.5
F16 = 124 + 7.5(16) = 244.0
2. Incorporating seasonality
Period 15: 236.5(1.10) = 260.15
Period 16: 244.0(0.75) = 183.00
56
Associative Forecasting

57
Associative Forecasting

Associative techniques rely on identification of related

variables that can be used to predict the variable of
interest (dependent variable)
Example 1: Crop yields are related to soil conditions and the
amounts and timing of water and fertilizer applications.
Example 2: Sales of beef may be related to the price per pound
(of beef) and the price of substitutes such as chicken, pork and
lamb.
Predictor (independent) variables - used to predict
values of variable of interest
Regression - technique for fitting a line to a set of points

58
Dependent and Independent Variables

59
Linear Regression

The object of linear regression is to obtain an

equation of a straight line (Least squares line) that
minimizes the sum of squared vertical deviations of
the data points from the line.

yc = a + b x
Where
yc = Predicated (dependent) variable.
x = Predictor (independent) variable.
a = Value of yc when x = 0.
b = Slope of the line.

60
Calculating Coefficients a and b

n  (xy) -  x y
b = 2 2
n  x - (  x)
 y - b x
a = =y–bx
n
where
n = Number of paired observations

61
Example 8 – Linear Model Seems Reasonable

n  (xy) -  x y
n = 12 b = 2 2
x
7
y
15
xy
105
x2
49
y2
225
n  x - (  x)
2 10 20 4 100
6 13 78 36 169
 y - b x
4 15 60 16 225 a = =y–bx
14
15
25
27
350
405
196
225
625
729
n
16 24 384 256 576
12  3529  132  271
b  1.593
12 20 240 144 400
14
20
27
44
378
880
196
400
729
1936
12 1796  132 2

15 34 510 225 1156

271  1.593 132
7 17 119 49 289
a  5.06
132 271 3529 1796 7159 12

62
 Example 8 – Linear Model Seems Reasonable
n = 12 Computed
2 2
x
7
y
15
xy
105
x
49
y
225
relationship
2 10 20 4 100
50
6 13 78 36 169
40
4 15 60 16 225
14 25 350 196 625 30

15 27 405 225 729 20

16 24 384 256 576 10

12 20 240 144 400 0

0 5 10 15 20 25
14 27 378 196 729
20 44 880 400 1936
15 34 510 225 1156
7 17 119 49 289 A straight line is fitted to a set
132 271 3529 1796 7159
of sample points.
12  3529  132  271
b  1.593
12 1796  132 2
yc = 5.06 +1.593 x
271  1.593 132
a  5.06 63
12
 Correlation

A measure of the strength and direction of the relationship

between two variables. Correlation can range from -1.00 to
+1.00.
n  (xy) - x  y
r =
n x2 - ( x)2 n y2 - ( y)2
12  3529  132  271
In Example 8, r   0.917  r 2  0.84
12 1796  132 2  12  7159  2712
The square of the correlation coefficient, r2, provides a
measure of how well a regression line “fits” the data.
The higher r2 , the better. r2 >=0.8, then the independent variable is
a good indicator of the dependent variable; r2 < = 0.25 , poor
indicator; otherwise, moderate indicator. 64
Curvilinear and Multiple Regression Analysis

When nonlinear relationship are present, we

must employ curvilinear regression.
Models that involve more than one predictor
require the use of multiple regression analysis.

65
Forecast Accuracy

66
Accuracy and Control of Forecasts

Error: Difference between the actual value and

the value that was predicted for a given period
et = At – Ft
Forecast errors influence decisions in two
somewhat different ways
1. A choice between various forecasting alternatives
2. Evaluate the success or failure of a technique in use

67
Measures of Forecast Accuracy

Mean Absolute Deviation (MAD)

 Actualt  Forecastt
MAD =
n
Mean Squared Error (MSE)
2
MSE =  ( Actualt  Forecastt)
n -1
Mean Absolute Percent Error (MAPE)
Actualt  Forecastt
 × 100
Actualt
MAPE =
n 68
Example 9

Period Actual Forecast (A-F) |A-F| (A-F) 2 (|A-F|/Actual)×100

1 217 215 2 2 4 0.92
2 213 216 -3 3 9 1.41
3 216 215 1 1 1 0.46
4 210 214 -4 4 16 1.90
5 213 211 2 2 4 0.94
6 219 214 5 5 25 2.28
7 216 217 -1 1 1 0.46
8 212 216 -4 4 16 1.89
-2 22 76 10.26

MAD=  ‫׀‬e ‫׀‬/ n = 22 / 8 = 2.75

MSE =  e2/ (n-1) = 76 / 7 = 10.86
MAPE = 10.26/8 = 1.28

69
Forecast Accuracy

Two aspects of forecast accuracy can have

potential significance when deciding among
forecasting alternatives :
1. Historical error performance of a forecast
choose the one yields the lowest MAD, MSE, or MAPE

2. The ability of a forecast to respond to changes

the cost of not responding quickly to a change relative to the
cost of responding to changes that are not really there

70
 Choosing a Forecasting Technique (Table 3.4)
Forecasting Amount of Forecast Personnel
Methods Historical Data Data Pattern Horizon Preparation Time Background

Little
Moving Average 2 to 30 obs. Stationary Short Short Sophistication
Simple
Exponental
Smoothing 5 to 10 obs. stationary Short Short Little
10 to 20;
for seasonality at Short to
Trend Models least 5 per season Trend medium Short Moderate
Handle
cyclical
Enough to see 2 and seasonal Short to
Seasonal peaks and troughs Patterns medium Short to moderate Little

Casual Can handle Short, Long development

Regression 10 observations per complex medium, time, short time for Considerable
Models independent var. patterns or long implementation Sophistication
Multiple-Choice Questions

1. Which of the following techniques generates

trend forecasts?
a. Delphi method b. weighted moving average
c. Moving average d. Single factor exponential
smoothing e. None of the above

2. What is the MAD for forecast errors of +4, -2, +1,

-1?
a. 0.00 b. 0.50 c. 1.00 d. 1.50 e. 2.00
72