7/5/26

Business Analytics: Methods, Models, and Decisions

Third Edition

Chapter 9
Forecasting
Techniques

Universityof Windsor

Learning Objectives
• Forecasting
ForecastingTechniques
Techniques
• Qualitative
Qualitativeandandjudgmental
judgmental techniques
techniques
• Statistical
Statisticaltime-series
time-series models
models
• Types
TypesofofTime
TimeSeries
Series Forecasting
Forecasting Models
• Moving
MovingAverage
Average Models
• Exponential
ExponentialSmoothing
Smoothing Models
• Forecasting
ForecastingModels
Modelswith
withaaLinear
Linear Trend
Trend
• Double
DoubleExponential
Exponential Smoothing
Smoothing
• Regression-based
Regression-based Forecasting
Forecasting Models
• Forecasting
Forecasting Time
Time Series
Series with
with Seasonality
Seasonality
• Regression-based
Regression-based Forecasting
Forecasting Models with Seasonality
• Holt-Winters
Holt-WintersForecasting
Forecasting Models
Models with
with Seasonality
Seasonality
• Holt-Winters
Holt-WintersAdditive
Additive Seasonality
Seasonality Model
• Holt-Winters
Holt-WintersMultiplicative
Multiplicative Seasonality
Seasonality Model
• Explanatory/causal
Explanatory/causal models
models

1
 7/5/26

Forecasting Techniques
• Managers
Managers require
require good
good forecasts of future events for decision-making
For example,
• Forecasts of interest rates and economic indicators are needed for financial planning
• Sales forecasts are needed for production planning
• Consumer behaviour forecasts are needed for strategic planning

• Business
Business Analysts
Analysts use
use three
three major
major categories
categories of
of forecasting
forecasting approaches
• Qualitative and judgmental techniques
• Statistical time-series models
• Explanatory/causal models

Universityof Windsor

Qualitative and Judgmental Forecasting

• Qualitative
Qualitativeand
and Judgmental
Judgmental techniques
techniques rely
rely on
on experience and intuition

• They
Theyare
areused
used when
when historical
historical data
data is not available

• Some
Somecommon
commonJudgmental
Judgmental Forecasting
Forecasting Techniques
Techniques include
• Historical
HistoricalAnalogy
Analogy (Similarity)
(Similarity)
• Delphi
DelphiMethod
Method
• Indicators
Indicatorsand
andIndexes
Indexes

Universityof Windsor

2
 7/5/26

Historical Analogy
• Historical Analogy (Similarity) approach forecasts based on comparative analysis with
prior situations

• For
For example
example
• IfIfaanew
newproduct
productisisintroduced
introduced in
in the
the market
market
• The
Theresponse
responseofofconsumers
consumerstotomarketing
marketingcampaigns
campaigns is expected to be similar to
previous products

• AllAllfactors
factorsare
arenot
notfully
fullyconsidered
consideredin
in such
such an
an approach
• However,
However,some someinsight
insightcan
canoften
oftenbe
begained
gainedthrough
through an
an analysis
analysis of
of past
past experiences
experiences

Universityof Windsor

Example: Predicting the Price of Oil

• Early
Early 1998
1998 ––oil
oilprice
price was
was about
about $22
$22 aa barrel
• Historically OPEC (Organization of the
the Petroleum
Petroleum Exporting Countries) controls oil prices, and oil
prices keep rising with reduced supply from OPEC

• Mid-1998 – oil price dropped to $11 a barrel because of oversupply and high production
in non-OPEC regions

• So the historical analogy was wrong

• Historical analogies cannot always account for current
current realities!
realities!

Universityof Windsor

3
 7/5/26

Delphi Method
• Delphi
DelphiMethod
Methoduses
usesaapanel
panel of
of experts
experts to
to make
make a forecast
• A questionnaire about the forecast is circulated 2-3 times in order to reach a convergence of
opinion on the forecasted variable

• After each round of responses, every expert’s opinions help to reinforce those in agreement and
further refine the questions about disagreement

• The process is repeated till aa consensus

consensus is reached

Universityof Windsor

Indicators and Indexes

• Indicators
Indicators influence the behavior of a variable that we wish to forecast

– For example, economic indicators include

• Gross domestic product (GDP)
• Consumption
• Investment
• Interest rates
• Prices

4
 7/5/26

Indicators and Indexes

• The index is a single measure that weights multiple indicators, thus providing a measure
of overall expectation
– For example,
• Index (a
Consumer Price Index (a measure
measure of
of inflation
inflation –– consumer’s capacity)
• Index (a
Producer Price Index (a measure
measure of
of selling
selling price
price –– producer’s capacity)

Universityof Windsor

Statistical Forecasting Models

• Statistical
StatisticalForecasting
ForecastingModels
Modelsare
arethe
thebuilding
buildingblocks
blocks of
of any
any planning
planning process

• Statistical
Statisticaltime-series
time-seriesmodels
modelsare
arevery
verygood
goodfor
forshort-range
short-range forecasting
forecasting problems

Time series forecasts are based on the assumption that the future is an extrapolation of the past

Universityof Windsor

10

5
 7/5/26

Time Series Components

Time Series Components include:

• Trend: is a gradual upward or downward movement

• Cyclical effect: up and down over a longer time frame

• Seasonal effect: repeats at fixed intervals of time (month or week)

• Random effect: occurs due to uncertainty in parameters

11

Example: Identifying Trends in a Time Series

• Total
TotalEnergy
Energy Consumption
Consumption
– There
– Thereisis aa general
general upward trend from 1949 to 1979, and then from 1982 to 2006
–– There
There isis aa short
short downward
downward trend from 2006 to 2021

Universityof Windsor

12

6
 7/5/26

Example: Cyclical Effects

• Federal
Federal Funds
Funds rate
rate shows a general 20-year cycle

• 1960s
1960s (Low),
(Low), 1980s
1980s (High),
(High), 2000s (Low)

Universityof Windsor

13

Example: Seasonal Effects

• Natural
Natural Gas
Gas Use
– Natural
– Natural gas
gas use is high in Winter months (furnaces are working 24 hours)
– Natural
– Natural gas
gas use is low in the Summer
Summer months

Universityof Windsor

14

7
 7/5/26

Types of Time-Series Forecasting Models

• Common
Common types
types of
of Time
Time Series
Series Forecasting
Forecasting Models include

• Moving Average Models: the next observation is the mean of every past observation
• Exponential Smoothing Models:
–– In simple moving average, the past observations are weighted equally
–– In exponential smoothing, exponentially decreasing weights are assigned for events that
occurred too much in the past

Universityof Windsor

15

Moving Average Models

• The
Thesimple
simplemoving
moving average
average method is
is a smoothing method based on averaging random
fluctuations in the time series

• The
Thesimple
simplemoving
moving average
average forecast
forecast for the next period is computed as the average of the
most recent k observations

(9.1)
Ft+1 = Forecast value in period
period (t+1)
(t+1)

Universityof Windsor

16

8
 7/5/26

Example 1: Moving Average Forecasting

• Open
Openthe
the Excel
Excelfile:
file:Moving
Moving Average

– The file contains the number of Computer units sold over the past 17 weeks
– Use a three-period moving average window to forecast the number of units sold for week 18

Universityof Windsor

17

Example 1: Moving Average Forecasting

• Open
Open the
the Excel
Excel file:
file: Moving
Moving Average
Average
• First,
First,make
makeaaline
linegraph
graph for
for the
the computer
computer units sold over the past 17 weeks

The line graph does not show any obvious trends

18

9
 7/5/26

Example 1: Three-period Moving Average Forecast

• Three-period
Three-periodmoving
movingaverage
average forecast for week 4
– In the cell C7, use the function = =AVERAGE(B4:B6)
–
– Forecast for week
– week 44 using
using three-period
three-period moving
moving average = 64 units
–– Copy paste cell C7 from C8:C21

• Three-period
Three-periodmoving
movingaverage
average forecast for week 18

Forecast for week 18 using three-period moving average = 68 units

Universityof Windsor

19

Example 1: Three-period Moving Average Forecast

• Insert
Insertaaline
lineplot
plotfor
forboth
boththe
theoriginal
original data
data and
and the
the forecast
forecast data

• Although
Althoughthe
theforecast
forecast tracks
tracks the actual data fairly well, notice that
– The
– The forecast overestimates the actual value when the trend is down
– While
– While it underestimates the actual value when the trend is up

This is because it uses past data, and thus lags the changes in the data

Universityof Windsor

20

10
 7/5/26

Example 2: Using Excel’s Moving Average Tool

• Open
Open the
theExcel
Excel file
file Example
Example 1:
1: Moving
Moving Average
Average

– Use a three-period moving average window to forecast the number of units sold for week 18,
using the Data Analysis pack

Universityof Windsor

21

Example 2: Using Excel’s Moving Average Tool

• Open
Open the
theExcel
Excel file
file Example
Example 1:
1: Moving
Moving Average
Average
– Data>Data Analysis> Moving Average
– Input range is units sold
– Interval = 3 (number of periods of moving average window)

22

11
 7/5/26

Example 2: Using Excel’s Moving Average Tool

• Note
Notethat
thatthe
theforecast
forecastfor
for week
week18
18isis aligned
aligned with
with the
the actual
actual value
value for week 17 on the chart

Universityof Windsor

23

Error Metrics and Forecast Accuracy

• Forecast Accuracy: The quality of a forecast depends on how accurate it is in predicting future
values of a time series

• Error
Erroror
orResidual
Residual in
in aa forecast
forecast is the difference between the forecast and the actual value

• Three
Threeerror
errormetrics,
metrics,which
whichcompare
compare forecasts
forecasts with
with actual
actual observations
observations are commonly used

– Mean absolute deviation

– Mean square error and Root means square error
– Mean absolute percentage error

Universityof Windsor

24

12
 7/5/26

Error Metrics
– Mean absolute deviation (M A D) (9.2)

– Mean square error (M S E) (9.3)

– Root mean square error (R M S E) (9.4)

– Mean absolute percentage error (M A P E)

(9.5)
At is the actual value at time t, Ft is the forecast value at time t, and n is the number of observations

Universityof Windsor

25

Example 3: Using Error Metrics to Compare Moving Average Forecasts

• Open
Openthe
theExcel
Excel file
fileExample
Example 1:
1: Moving
Moving Average
Average

• Use
Useerror
errormetrics
metricsto
tocompare
compare 2-period,
2-period,3-period,
3-period,and
and4-period
4-periodmoving
moving average
average models

26

13
 7/5/26

Example 3: Using Error Metrics to Compare Moving Average Forecasts

• Open
Openthe
theExcel
Excelfile
fileExample
Example 1:
1: Moving
Moving Average
Average

• For
For2-period
2-periodMoving
Moving Average
Average (k = 2)
– InIn the cell C6, use the function = =AVERAGE(B4:B5)
–
– Copy
– Copy and paste cell C6 from C7:C20
–– Error
Error = Actual – Forecast (cell D6 = B6 – C6)
– Absolute
– Absolute Deviation
Deviation (cell E6 = ABS(D6)), MAD (cell E21 = AVERAGE(E6:E20))
–– Squared Deviation (cell
Squared Deviation (cell F6
F6 == D6^2),
D6^2), MSE
MSE (cell
(cell F21 = AVERAGE(F6:F20))
–– Absolute
Absolute % error
error (cell
(cell G6 == ABS(D6/B6)*100),
ABS(D6/B6)*100), MSE (cell G21 = AVERAGE(G6:G20))

• For
For3-period
3-periodMoving
Moving Average
Average (k = 3)
– InIn the cell H7, use the function = =AVERAGE(B4:B6)
–
– Rest
– Rest ofof the
the procedure
procedure is the same

• For
For4-period
4-periodMoving
Moving Average
Average (k = 4)
– InIn the cell M8,
– M8, use
use the function = =AVERAGE(B4:B7)
– Rest
– Rest ofof the
the procedure
procedure is the same

Universityof Windsor

27

Example 3: Using Error Metrics to Compare Moving Average Forecasts

• The
The2-period
2-periodmodel
modelhas
has the
the lowest
lowest error
error metric
metric values
values and
and is the best model
A 8 C D M N P
Tablet Computer Sales
k=2 K=3 k=
Week Units Sold Forecast ErrorAbsolute Squared AbsoluteForecast ErrorAbsolute Squared Absolute Forecast Error Absolute Squared Absolute
1 88 Deviation Error %Error Deviation Error %Error Deviation Error %Error
2 44
3 60 66.006.00 6.00 36.00 10.00
97
4 56 52.004.00 4.00 16.00 7.14 64.00-8.00 8.00 64.00 14.29
5 70 58.0012.00 12.00 144.00 17.14 53.3316.67 16.67 277.78 23.81 62.00 8.00 8.00 64.00 11.43
67 91
54
63.0028.00
80.50-28.50
28.00
26.50
784.00
702.25
30.77
49.07
62.0029.00
72.33-18.33
29.00
18.33
841.00
336.11
31.87
33.95
57.50
69.25
33.50
-15.25
33.50 1122.25
15.25 232.56
36.81
28.24
8 60 72.50-12.50 12.50 158.25 20.83 71.67-11.67 11.67 136.11 19.44 67.75 -7.75 7.75 60.06 12.92
12 48 57.009.00 9.00 81.00 18.75 68.33-20.33 20.33 413.44 42.36 68.75 -20.75 20.75 430.56 43.23
13 10 35 54.00-19.00 19.00 361.00 54.29 54.00-19.00 19.00 361.00 54.29 63.25 -28.25 28.25 798.06 80.71
11 49 41.507.50 7.50 56.25 15.31 47.671.33 1.33 1.78 2.72 49.25 -0.25 0.25 0.06 0.51
12 44 42.002.00 2.00 4.00 4.55 44.00 0.00 0.00 0.00 0.00 48.00 -4.00 4.00 16.00 9.09
13 61 46.5014.50 14.50 210.25 23.77 42.6718.33 18.33 336.11 30.05 44.00 17.00 17.00 289.00 27.87
14 68 52.5015.50 15.50 240.25 22.79 51.3316.67 16.67 277.78 24.51 47.25 20.75 20.75 430.56 30.51
15 82 64.5017.50 17.50 306.25 21.34 57.6724.33 24.33 592.11 29.67 55.50 26.50 26.50 702.25 32.32
16 71 75.004.00 4.00 16.00 5.63 70.330.67 0.67 0.44 0.94 63.75 7.25 7.25 52.56 10.21
17 50 76.50-26.50 26.50 702.25 53.00 73.67-23.67 23.67 560.11 47.33 70.50 -20.50 20.50 420.25 41.00

2 18 60.50 13.63 254.38 23.63

MAD MSE MAPE
67.67 14.86 299.84 25.37
MAD MSE MAPE
67.75 16.13 355.25 28.07
MAD MSE MAPE

28

14
 7/5/26

Exponential Smoothing Models

• Simple Exponential Smoothing Model: In order to forecast the value (Ft+1) in period (t+1), we give
different weights to forecast the value in period t (Ft) and the actual value in period t (At)

(9.6)

• a is a constant between 0 and 1 called the Smoothing Constant

• To begin, set F1 and F2 equal to the actual observation in period 1, (A1)

Universityof Windsor

29

Example 4: Using Exponential Smoothing to Forecast Sales

• Open
Openthe
theExcel
Excelfile
fileExample
Example 1:
1: Moving
Moving Average
Average

• Use
Usethe
theexponential
exponential smoothing (𝛼==0.10
smoothing model (𝛼 0.10𝑡𝑜
𝑡𝑜 1.0
1.0)) to forecast sales for weeks 1 to 17

30

15
 7/5/26

Example 4: Using Exponential Smoothing to Forecast Sales

• Open the Excel file Example 1: Moving Average

• Forecasting Sales (𝛼 = 0.1)

– Cells C4 and C5 are the same as cell B4
– Cell C6 =(1-C$3)*C5+C$3*$B5
– Copy cell C6 to cells C7:C21

• Forecasting Sales (𝛼 = 0.2 to 1.0)

– Repeat the process for 𝛼 = 0.1
• We can also calculate MAD, MSE, MAPE for each value of 𝛼

Universityof Windsor

31

Example 4: UsingL Exponential Smoothing

6 to Forecast Sales
Tablet Computer Sales
Smoothing Constant
Week Units Sold 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
1 88 88 88 88 88 88 88 88 88 88 88
2 44 88 88 88 88 88 88 88 88 88 88
3 60 83.60 79.20 74.80 70.40 66.00 61.60 57.20 52.80 48.40 44.00
4 56 81.24 75.36 70.36 66.24 63.00 60.64 59.16 58.56 58.84 60.00
5 70 78.72 71.49 66.05 62.14 59.50 57.86 56.95 56.51 56.28 56.00
6 91 77.84 71.19 67.24 65.29 64.75 65.14 66.08 67.30 68.63 70.00
7 54 79.16 75.15 74.37 75.57 77.88 80.66 83.53 86.26 88.76 91.00
8 60 76.64 70.92 68.26 66.94 65.94 64.66 62.86 60.45 57.48 54.00
9 48 74.98 68.74 65.78 64.17 62.97 61.87 60.86 60.09 59.75 60.00
10 35 72.28 64.59 60.45 57.70 55.48 53.55 51.86 50.42 49.17 48.00
11 49 68.55 58.67 52.81 48.62 45.24 42.42 40.06 38.08 36.42 35.00
12 44 66.60 56.74 51.67 48.77 47.12 46.37 46.32 46.82 47.74 49.00
13 61 64.34 54.19 49.37 46.86 45.56 44.95 44.70 44.56 44.37 44.00
14 68 64.00 55.55 52.86 52.52 53.28 54.58 56.11 57.71 59.34 61.00
15 82 64.40 58.04 57.40 58.71 60.64 62.63 64.43 65.94 67.13 68.00
16 71 66.16 62.83 64.78 68.03 71.32 74.25 76.73 78.79 80.51 82.00
17 50 66.65 64.47 66.65 69.22 71.16 72.30 72.72 72.56 71.95 71.00
18 64.98 61.57 61.65 61.53 60.58 58.92 56.82 54.51 52.20 50.00

32

16
 7/5/26

Example 5: Using Excel Exponential Smoothing Tool

• Open
Openthe
theExcel
Excelfile
fileExample
Example 1:
1: Moving
Moving Average
Average

• Use
Usethe
theExcel
Excelexponential
exponential smoothing (𝛼==0.10
smoothing tool (𝛼 0.10)) to forecast sales for weeks 1 to 17

Universityof Windsor

33

Example 5: Using Excel Exponential Smoothing Tool

• Open the Excel file Example 1: Moving Average
• Data>Data Analysis>Exponential Smoothing
• Input range is units sold
• Damping factor = 0.10

34

17
 7/5/26

Example 5: Using Excel Exponential Smoothing Tool

• For 𝛼==0.10
For ((𝛼 ), sales
0.10), sales forecast
forecast for
for weeks 1 to 17

Universityof Windsor

35

Forecasting Models for Time Series with a Linear Trend

• For time series with a linear trend but no significant seasonal components, the following two models
are used for forecasting

• Double exponential smoothing models

• Regression-based models

• These
Theseare
arebased
based on
on the
the linear
linear trend
trend equation

Ft+k =a,+b,f (9.7)

• The
Theforecast
forecast for
for kk periods
periods into the future (Ft+k) is a function of the level at and the trend bt

Universityof Windsor

36

18
 7/5/26

Double Exponential Smoothing

• Estimates
Estimates of
of the
the parameters
parameters (at and bt) are obtained from the following equations:

(9.8)

• Initial
Initial values are chosen for a1 as A1 and b1 as (A2 – A1)
values are

• The
Theforecast
forecast for
for kk periods
periods beyond the last period (period T) is

Fr+k = ar +br(k) (9.9)

Universityof Windsor

37

Example 6: Double Exponential Smoothing

• Open the Excel File Example 6: Double Exponential Smoothing
–
– The file shows data for the first ten years of Coal Production
–
– Use double exponential smoothing model to forecast coal production for the year 11, using
a = 0.6 and b = 0.4
–
– Calculate the Mean Absolute Deviation
Deviation (MAD)
(MAD)
–
– Plot the actual and forecast values

Universityof Windsor

38

19
 7/5/26

Example 6: Double Exponential Smoothing

• First
First calculate
calculate the
the parameters
parameters (at and bt) using the following equations:

(9.8)

• Initial
Initial values are chosen for a1 as A1 and b1 as (A2 – A1)
values are
• For
For a1, cell C8 = B8
• For
For b1, cell D8 = B9 - B8
• For
For a2, cell C9 = $B$4*B9+(1-$B$4)*(C8+D8)
• For
For b2, cell D9 = $B$5*(C10-C9)+(1-$B$5)*D9
• Copy
Copy and
and paste
paste cell C9 to C10:C17
• Copy
Copy and
and paste
paste cell D9 to D10:D17

Universityof Windsor

39

Example 6: Double Exponential Smoothing

• Forecast
Forecast for
for kk periods
periods beyond the last period (period T) is

Fr+k = ar +br(k) (9.9)

• KK == 11 for
for the
the next
next period

• Forecast
Forecast
• Cell
Cell E10
E10 == C9
C9 + D9
• Copy
Copy and
and paste cell E10 to E11:E18

• The
The forecast
forecast for year 11 = 588,716,009 tons

Universityof Windsor

40

20
 7/5/26

Example 6: Double Exponential Smoothing

Mean Absolute Deviation (MAD)
– First, calculate the absolute error in column F
– Cell F10 = ABS(B10 – E10)
– Copy paste F10 to F11:F17

– For MAD, Cell F20 = AVERAGE(F10:F17)

– MAD = 26,404,946

Universityof Windsor

41

Example 6: Double Exponential Smoothing

• Line plot for Actual
Actual and
and Forecast values using Double Exponential Smoothing

42

21
 7/5/26

Regression-Based Forecasting for Time Series with a Linear Trend

• Simple
Simplelinear
linearregression
regression can
can be
be applied
applied to
to forecasting
forecasting using time as the independent variable

• Example:
Example: Coal
Coal Production
Production with
with aa linear trendline

– Regression equation can predict the value of coal production for any year (x) within the data
– Not used for future forecasts
Universityof Windsor

43

Forecasting Time Series with Seasonality

• When
Whentime
timeseries
seriesexhibit
exhibitseasonality,
seasonality, the
the following
following techniques
techniques provide
provide better forecasts:

– Multiple Regression Models with categorical variables for the seasonal components

– Holt-Winters Models are similar to exponential smoothing models

• Smoothing constants are used
used to
to smooth out variations in the seasonal patterns over time

Universityof Windsor

44

22
 7/5/26

Example 7: Regression-Based Forecasting for Natural Gas Usage

• Open
Openthe
theExcel
Excel File
File Example
Example 7: Gas & Electric Use
– The
– The file
file shows
shows monthly
monthly gas
gas usage for the past two years
– Estimate
– Estimate the
the regression
regression equation
equation to
to forecast
forecast gas
gas usage for any month

Universityof Windsor

45

Example 7: Regression-Based Forecasting for Natural Gas Usage

• Open
Openthe
theExcel
Excel File
File Example
Example 7: Gas & Electric Use
• The
Thefile
fileshows
showsmonthly
monthlygas
gas usage
usage for the past two years
• Since
Sincethe
thedata
dataexhibits
exhibitsaaseasonal
seasonal pattern,
pattern, we
we add
add aa seasonal
seasonal categorical variable
variable (Time)
(Time) with k = 12 levels

• We
Wehave
haveto
toadd
add(k(k−−11==11)
11)dummy
dummyvariables
variables (Feb-Dec),
(Feb-Dec),with
withJanuary
Januarybeing
being the
the reference
reference month
• The
Thevalues
valuesinindummy
dummy variables
variables are
are 00 except
except for that month, which is 1

Universityof Windsor

46

23
 7/5/26

Example 7: Regression-Based Forecasting for Natural Gas Usage

• The
Theregression
regression equation
equation is
sLuonenbauoissaiaa

gas usage =+ time+February+March

+ 4April+ May + 6 June+ , July
+ August+ September + o October
+ 1, November + , December

• Run
Runthe
theregression
regressionmodel
modelwith
withgas
gas usage
usage as the Y variable and rest as the input variables

– We find that both time and February are insignificant variables

– We run the regression model without these variables in the model now

Universityof Windsor

47

Example 7: Regression-Based Forecasting for Natural Gas Usage

• Run the regression model with Gas Use as the Y variable Regression
and Mar-Dec as the input variables Input
OK
Input Y Range: $8$3:$8$27 Cancel
Input X Range: $E$3:$N$27

• Because the data show no clear linear trend, the variable Labels Constant is Zero

Time could not explain any significant variation in the data

ta
Confidence Level: 95 %
Output options

SB$30
• The dummy variable for February was probably O Output Range:
ONew Worksheet Ply:

insignificant because the historical gas usage for both O New Workbookl

Residuals
January and February were very close to each other
Residuals Residual Plots
Standardized Residuals Line Fit Plots

Normal Probability

Normal Probability Plots

Universityof Windsor

48

24
 7/5/26

Example 7: Regression-Based Forecasting for Natural Gas Usage

• Regression
Regression Results

• The regression equation can be used to forecast gas usage for any month (use 1 for the month for
which gas usage is being estimated, and 0 for all others)

gas usage = 236.75-36.75 March-99.25April

192.25 May-203.25June-208.25July
209.75August-208.25 September
196.75 October-149.75 November
43.25December

UniversityofWindsor

49

Holt-Winters Models for Forecasting Time Series with Seasonality and No Trend
• The Holt-Winters additive seasonality model with no trend applies to time series with relatively stable
seasonality and is based on the equation

(9.10)
• The Holt-Winters multiplicative seasonality model with no trend applies to time series whose
amplitude increases or decreases over time and is

(9.11)
A plot of the time series should be carefully viewed first to identify the appropriate type of model to use

Universityof Windsor

50

25
 7/5/26

Holt-Winters Additive Seasonality Model with No Trend

• The level and seasonal factors are estimated as

Level componenta=aA-S-s+1-a
Seasonal component:S=yA-a)+(1-y)S- (9.12)

• The forecast for the next period is

Universityof Windsor

51

Example 8: Using Holt-Winters Additive Seasonality Model with No Trend

• Open the Excel File Example 7: Gas and Electric Use
– Use
– UseHolt-Winters
Holt-WintersAdditive
AdditiveSeasonality
SeasonalityModel
Modelto
topredict
predictGas
Gas Use
Use for
for next
next Jan
Jan and
and Feb
– Use
– Use 𝛼𝛼 = 0.4 and 𝛾
𝛾 = 0.9
–
– Plot
Plotthe
theactual
actualand
andforecast
forecastgas
gas use
use values

Universityof Windsor

52

26
 7/5/26

Example 8: Using Holt-Winters Additive Seasonality Model with No Trend

• Level (a)
–
– Cell D4 to D15 =AVERAGE(B4:B15)
–
– Cell D16 =$D$2*(B16-E4)+(1-$D$2)*D15
–
– Copy past D16 to D17:D27
• Seasonality (s)
–
– Cell E4 = B4 – D4, copy paste to E5:E7
–
– Cell E8 =$E$2*(B8-D8)+(1-$E$2)*E4
–
– Copy past E8 to E9:E27
• Forecast
–
– Cell F16 = D15+E4
–
– Copy past F16 to F17:F29
• Mean Annual Deviation
Deviation (MAD)
(MAD)
–
– Cell G16 = =ABS(B16-F16)
–
– Copy past G16 to G17:G29
–
– Cell G31 + AVERAGE(G16:G27)

Universityof Windsor

53

Example 8: Using Holt-Winters Additive Seasonality Model with No Trend

• Plot of Actual and Forecast Values

54

27
 7/5/26

Holt-Winters Multiplicative Seasonality Model with No Trend

• The multiplicative seasonal model has the same basic smoothing structure as the additive
seasonal model with some key differences:

Levelcomponenta=aA/S+1-aa
Seasonal componentS=yA/a+1-yS (9.15)

• The forecast for the next period is

Universityof Windsor

55

Selecting Appropriate Time-Series-Based Forecasting Methods

No Seasonality Seasonality
No trend Simple moving average Holt-Winters additive or multiplicative
or simple exponential seasonality models without trend or
smoothing multiple regression
Trend Double exponential Holt-Winters additive or multiplicative
smoothing seasonality models with trend

Universityof Windsor

56

28
 7/5/26

Regression Forecasting with Causal Variables

• In many forecasting applications, other independent variables besides time may influence
the time series

• Explanatory/causal models (Econometric models) seek to identify factors that explain

statistically the patterns observed in the variable being forecast

Universityof Windsor

57

Example 9: Forecasting Gasoline Sales Using Simple Linear Regression

• Open
Openthe theExcel
Excel file
file Example
Example 9:
9: Gasoline
Gasoline Sales
–– Plot the gasoline sales over time (weeks)
–– Insert a trendline
–– Run the
the regression
regression model using Gallons sold as Y-variable, and week and price/gallon
as the independent variables
–– Predict gasoline
gasoline sales for week 11
11 using
using the regression model results

58

29
 7/5/26

Example 9: Forecasting Gasoline Sales Using Simple Linear Regression

• Plot
Plotaaline
linegraph
graph and
and insert a trend line

Universityof Windsor

59

Example 9: Forecasting Gasoline Sales Using Simple Linear Regression

• Data>Data Analysis>Regression
• Use Gallons sold as Y-input
• Use Week and Price/Gallon as X-input
• The multiple regression model is
Sales = b0 + b1 week + b2 price/gallon

60

30
 7/5/26

Example 9: Forecasting Gasoline Sales Using Simple Linear Regression

• The
Theregression
regression model
model results
results are
Toaerresutsare
SUMMARY OUTPUT

Regression Statistics
Multiple R 0.93052853
R Square 0.86588334
Adjusted R S0.8275643
Standard Err1235.40033
Observations 10
ANOVA
df SS MS Significance F
Regression 268974748.734487374.322.59668370.00088347
Residual 710683497.81526213.97
Total 979658246.5

Coefficients tandard Erroi tStat P-valueLower95%Upper 95%Lower 95.0%Upper 95.0%

Intercept 72333.084521969.92273.292368640.0132592320382.4725124283.69620382.4725124283.696
Week 508.66814168.1770863.024598360.01926086110.992523906.343756110.992523906.343756
Price/Gallon 16463.199 5351.0824 3.0766110.0179004-29116.498-3809.8998-29116.498-3809.8998

Predicted sales for week 11 = 72,333 + 508.7x(11) – 16,463x(3.80) = 15,368 gallons

UniversityofWindsor

61

31