Forecasting

2
 Objectives

• Give the fundamental rules of forecasting

• Calculate a forecast using a moving average,

weighted moving average, and exponential
smoothing, Winter, regression analysis,
multiple regression, cyclical.

• Calculate the accuracy of a forecast

3
 Demand Management

• Make-to-order (MTO) /Job order

• Make-to-stock (MTS)/mass
production/continuous production
Uncertainty, need forecasting

4
 Definition (Tersine)
??
• Forecasting is the prediction, projection, or
estimation of the occurrences of uncertain future
events or levels of activity or to predictor estimate
future demand
• Fortunately, many item produced by an organization
do not need forecasts.
• Dependent demand items such as components,
subassemblies can be calculated from the forecasts
for the end item. Forecasts should be made only
for end items (independent demand) that have an
uncertain demand.
5
Why is forecasting important?
• Demand for products and services is usually uncertain.
• Forecasting can be used for…
• Strategic planning (long range planning)
• Finance and accounting (budgets and cost controls)
• Marketing (future sales, new products)
• Production and operations

• For MTS, forecasting can be used for:

• Acquisition of plant and equipment
• Manpower planning
• Purchasing of materials
• Scheduling of production
6
What is forecasting all about?

Demand for Mercedes E Class We try to predict the

future by looking back
at the past

Predicted
demand
looking
Time back six
Jan Feb Mar Apr May Jun Jul Aug months
Actual demand (past sales)
Predicted demand

7
 Some general
characteristics of forecasts

• Forecasts are always wrong

• Forecasts are more accurate for
groups or families of items or
aggregated products
• Forecasts are more accurate for
shorter time periods
• Every forecast should include an
error estimate
• Forecasts are no substitute for
calculated demand.
8
Forecasting Time Horizons (Narasimhan)
 Short-range forecast
 Up to 1 year, generally less than 3 months
 Purchasing, job scheduling, workforce levels, job
assignments, production levels
 Medium-range forecast
 3 months to 3 years or 1-3 years
 Sales and production planning, budgeting
 Long-range forecast
 Over 5 years
 New product planning, facility location, research
and development
9
Key issues in forecasting

1. A forecast is only as good as the information included in

the forecast (past data)
2. History is not a perfect predictor of the future (i.e.: there is
no such thing as a perfect forecast)

REMEMBER: Forecasting is based on the assumption

that the past predicts the future! When forecasting,
think carefully whether or not the past is strongly
related to what you expect to see in the future…

10
Example: Mercedes E-class vs. M-class Sales
Month E-class Sales M-class Sales
Jan 23,345 -
Feb 22,034 -
Mar 21,453 -
Apr 24,897 -
May 23,561 -
Jun 22,684 -
Jul ? ?

Question: Can we predict the new model M-class sales based on

the data in the table?

Answer: Maybe... We need to consider how much the two

markets have in common
11
 Forecasting Steps

• What needs to be forecast?

– Level of detail, units of analysis & time
horizon required
• What data is available to evaluate?
– Identify needed data & whether it’s available
• Select and test the forecasting model
– Cost, ease of use & accuracy
• Generate the forecast
• Monitor forecast accuracy over time

12
 Types of Forecasting Models

Qualitative Quantitative
(technological) (statistical) methods:
methods: • Forecasts generated
• Forecasts generated through mathematical
subjectively by the modeling, rely on data
forecaster and analytical techniques

13
Qualitative Methods

14
 Quantitative Method
Intrinsic methods Time
Series Models:
– Assumes the future will
follow same patterns as the
past
– Last period demand,
Arithmetic average
(average methods), Moving
average, Exponential
smoothing, regression
analysis
Extrinsic methods/Causal
Models:
– Explores cause-and-effect
relationships
– Uses leading indicators to
predict the future
– Simple Linear Regression,
Multiple Linear Regression
– E.g. housing starts and
appliance sales
15
 Time Series
Forecasting

• Anything that is
observed sequentially
over time is a time series
• Examples of time series
data include:
– Daily IBM stock
prices
– Monthly rainfall
– Quarterly sales
results for Amazon
– Annual Google
profits

16
 Causal Method

• Review on Industrial Statistics 1 and 2

• See on given exercises (do it in a group as a
report)

17
Linear regression in forecasting

Linear regression is based on

1. Fitting a straight line to data
2. Explaining the change in one variable through changes
in other variables.

dependent variable = a + b  (independent variable)

By using linear regression, we are trying to explore which

independent variables affect the dependent variable
18
Example: do people drink more when it’s cold?

Alcohol Sales

Which line best

fits the data?

Average Monthly
Temperature
19
The best line is the one that minimizes the error

The predicted line is …

So, the error is …

Where: ε is the error

y is the observed value
Y is the predicted value

20
Least Squares Method of Linear Regression

The goal of LSM is to minimize the sum of squared errors…

Then the line is defined by

21
What does that mean?

Alcohol Sales ε ε
ε

So LSM tries to
minimize the distance
between the line and
the points!

Average Monthly
Temperature
22
 Time Series Forecasting

• Set of evenly spaced numerical data

• Obtained by observing response variable at
regular time periods
• Forecast based only on past values, no other
variables important
• Assumes that factors influencing past and
present will continue influence in future

23
 Time Series Components

Trend Cyclical

Seasonal Random

24
 Components
Trend
component
Demand for product or service

Seasonal peaks

Actual
demand

Average
demand over
Random four years
variation
| | | |
1 2 3 4
Year
25
 Trend Component (linear, non linear)
• Persistent, overall upward or downward pattern
• Changes due to population, technology, age,
culture, etc.
• Typically several years duration
Sales

Time Time
Downward linear trend Upward nonlinear trend
26
 Seasonal Component (additive,
multiplicative)
• Regular, relatively short-term pattern of up and
down fluctuations
• Due to weather, customs, etc.
• Occurs within a single year

Number of
Period Length Seasons
Week Day 7
Month Week 4-4.5
Month Day 28-31
Year Quarter 4
Year Month 12
Year Week 52
27
 Cyclical Component
• Repeating up and down movements
• Affected by business cycle, political, and economic factors
• Multiple years duration
• Often causal or associative relationships
• normally not included in short-term forecasting models
but do play a role in long-range forecasting.

1 Cycle
Sales

Year 28
 Irregular/Random Component
• Erratic, unsystematic, ‘residual’ fluctuations
• Due to random variation or unforeseen events
• Short duration and nonrepeating

M T W T F
29
 Time Series Plot
A time-series plot (time plot) is a two-dimensional
plot of time series data

• the vertical axis

measures the
variable of interest
• the horizontal axis
corresponds to the
time periods

30
Time Series: Moving average
• The moving average model uses the last t periods in order to
predict demand in period t+1 by averaging the actual demand
for the last N time periods
• There can be two types of moving average models: simple
moving average and weighted moving average
• The moving average model assumption is that the most accurate
prediction of future demand is a simple (linear) combination of
past demand.
• The choice of the value of N should be determined by
experimentation and often lies within the range of 3 to 8
(Tersine) and 5 to 7 (Smith)
Large N for stable demand (response to change will be slow);
Small N for demand subject to frequent significant change
(response to change will be fast) 31
 Naïve Model/Last Period Demand

• Uses recent past as the best indicator of the

future.

• The error associated with this model is

computed as:

32
Example for the Naïve Model

Week Sales (in $1,000) Forecast

1 9 -
2 8 9
3 9 8
5 12 9
6 9 12
7 12 9
8 11 12
9 ? 11
33
 Example for the Naïve Model
Week Sales (in Forecast Error Absolute Squared
$1,000) Error Error
1 9
2 8 9 −1 1 1
3 9 8 1 1 1
5 12 9 3 3 9
6 9 12 −3 3 9
7 12 9 3 3 9
8 11 12 −1 1 1
Sum 2 12 30
Mean 0.33 2 5
34
Time series: Simple moving average

Similar to the naïve model, this model uses part of the historical
data to make a forecast.
In the simple moving average models the forecast value is

𝑑 𝑡 − 1 + 𝑑𝑡 − 2 +… . + 𝑑𝑡 − 𝑁 ∑ 𝑑𝑡 − 𝑖
dt ′= = 𝑖=1
𝑁 𝑁

Where: = forecasted demand for period t

= actual demand in period t-i
N = number of time periods included in moving average

35
Example: forecasting sales at Kroger

Kroger sells (among other stuff) bottled spring water

Month Bottles
Jan 1,325
Feb 1,353
Mar 1,305 What will
the sales be
Apr 1,275
for July?
May 1,210
Jun 1,195
Jul ?

36
What if we use a 3-month simple moving average?

dJun + dMay + dApr

FJul = = 1,227
3

What if we use a 5-month simple moving average?

dJun + dMay + dApr + dMar + dFeb

FJul = = 1,268
5

37
 5-month
MA forecast

3-month
MA forecast

What do we observe?

5-month average smoothes data more;

3-month average more responsive
38
 Moving Average Lags a Trend

The choice of using a smaller or a larger number of observations

(n) in computing the moving average has implications for the
forecast. Smaller n, the more weight is given to recent periods. 39
Stability versus responsiveness in moving averages

40
 Last Period Demand Arithmetic Average Moving Average (2)
Month Demand Forecast Absolut Forecast Absolut Forecast Absolut
Demand Deviation Demand Deviation Demand Deviation
1 34 - - - - - -
2 44 34 10 - - - -
3 42 44 2 39 3 39 3
4 30 42 12 40 10 43 13
5 46 30 16 38 8 36 10
6 44 46 2 39 5 38 6
7 56 44 12 40 16 45 11
8 50 56 6 42 8 50 0
9 38 50 12 43 5 53 15
10 44 38 6 43 1 44 0
11 36 44 8 43 7 41 5
12 46 36 10 42 4 40 6
13 42 46 4 43 1 41 1
14 30 42 12 42 12 44 14
15 52 30 22 42 10 36 16
16 48 52 4 42 6 41 7
17 58 48 10 43 15 50 8
18 54 58 4 44 10 53 1
19 46 54 8 44 2 56 10
20 48 46 2 44 4 50 2
21 40 48 8 44 4 47 7
22 50 40 10 44 6 44 6
23 58 50 8 44 14 45 13
24 60 58 2 45 15 54 6
25  - 60 -  46  - 59  -
190 166 160
41
160/22=7,2
MAD= 190/23=8,26 MAD= 166/22=7,55 MAD= 7
 Data Plot
70

60

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

LPD Arithmetic MA(2) Demand

42
 Time series: Weighted Moving Average

Used when trend is present, Older data usually less important

∑ (weight for period n)

Weighted x (demand in period n)
moving average = ∑ weights
𝑁

𝐶 𝑡 − 1 𝑑𝑡 −1+ 𝐶 𝑡 − 2 𝑑𝑡 −2 +… . +𝐶 𝑡 − 𝑁 𝑑 𝑡 − 𝑁
∑ 𝐶 𝑡 −𝑖 𝑑 𝑡 −𝑖
′ 𝑖 =1
dt = =
∑𝐶 ∑𝐶
t is the current period.
Ft+1 is the forecast for next period
n is the forecasting horizon (how far back we look),
A is the actual sales figure from each period.
w is the importance (weight) we give to each period 43
Why do we need the WMA models?
Because of the ability to give more importance to what
happened recently, without losing the impact of the past.

Demand for Mercedes E-class Actual demand (past sales)

Prediction when using 6-month SMA
Prediction when using 6-months WMA

For a 6-month
SMA, attributing
equal weights to all
past data we miss
Time the downward trend
Jan Feb Mar Apr May Jun Jul Aug

44
Example: Kroger sales of bottled water

Month Bottles
Jan 1,325
Feb 1,353
What will
Mar 1,305
be the sales
Apr 1,275 for July?
May 1,210
Jun 1,195
Jul ?

45
6-month simple moving average…

AJun + AMay + AApr + AMar + AFeb + AJan

FJul = = 1,277
6

In other words, because we used equal weights, a slight

downward trend that actually exists is not observed…

46
What if we use a weighted moving average?

Make the weights for the last three months more than the first
three months…
Average demand of the first three-months = 1,328
Average demand of the last three-months = 1,227

6-month WMA WMA WMA

SMA 40% / 60% 30% / 70% 20% / 80%

July 1,277 1,267 1,257 1,247

Forecast

The higher the importance we give to recent data, the more we

pick up the declining trend in our forecast.
47
How do we choose weights?

1. Depending on the importance that we feel past data has

2. Depending on known seasonality (weights of past data
can also be zero).

WMA is better than SMA

because of the ability to
vary the weights!

48
 Another Example: WMA (4)
Mo(t) Demand (dt) Forecast Demand(dt') WMA (4)
1 280  
2 270  
3 330  
4 250  
5 340 {(4)(250) + (3)(330)+ (2)(270) + (1)(280)} / 10 = 284
6 330 {(4)(340) + (3)(250)+ (2)(330) + (1)(270)} / 10 = 304
7 350 {(4)(330) + (3)(340)+ (2)(250) + (1)(330)} / 10 = 317
8 300 {(4)(350) + (3)(330)+ (2)(340) + (1)(250)} / 10 = 332
9 330 {(4)(300) + (3)(350)+ (2)(330) + (1)(340)} / 10 = 325
10 350 {(4)(330) + (3)(300)+ (2)(350) + (1)(330)} / 10 = 325
11 270 {(4)(350) + (3)(330)+ (2)(300) + (1)(350)} / 10 = 334
12 290 {(4)(270) + (3)(350)+ (2)(330) + (1)(300)} / 10 = 309
13   {(4)(290) + (3)(270)+ (2)(350) + (1)(330)} / 10 = 300
14   {(4)(?) + (3)(290)+ (2)(270) + (1)(350)} / 10 = ?
49
Time Series: Double Moving
Average

• Used when we have a

linear trend in the data.
• Two different moving
averages are computed in
this model.
• The idea is to remove the
trend.

50
 Procedure
1.Calculate Single Moving Average (SMA) as St’
2.Calculate the second moving average using
moving average St’ and denotated as St’’
3. Calculate difference between 2 MAs (at):

4. Compute trend bt :

5. Forecast for m periods (dt’) into the future:

51
 Example (DMA (4))
t Demand MA (4) MA at bt Forecast Demand(dt’)
(dt) S t' (4x4) Ft+m = at + bt (m)
St''
1 140          
2 159          
3 136          
4 157 148        
5 173 156,25        
6 131 149,25        
7 177 159,5 153,25 165,75 4,17  d7’=F6+1=a6+b6(1)=-
8 188 167,25 158,06 176,43 6,13  d8’=F7+1=a7+b7(1)=165,75+4,17(1)=169,92
9 154 162,5 159,62 165,37 1,92  d9’=F8+1=a8+b8(1)=176,43+6,13(1)=182,56
10 179 174,5 165,93 183,06 5,71  d10’=F9+1=a9+b9(1)=165,37+1,92(1)=167,29
11 180 175,25 169,87 180,62 3,58  d11’=F10+1=a10+b10(1)=183,06+5,71(1)=188,77
12 160 168,5 170,12 166,37 -1,25  d12’=F11+1=a11+b11(1)=180,62+3,58(1)=184,2
13            d13’=F12+1=a12+b12(1)=166,37-1,25(1)=165,12
14           d14’=F12+2=a12+b12(2)=166,37-1,25(2)=163,87

52
Time Series: Exponential Smoothing (ES)

Main idea: The prediction of the future depends mostly on the

most recent observation, and on the error for the latest
forecast
The model relies on the assumption that the data are
stationary with a slowly varying mean.
Most recent observations play a more important role than the
distant past..

Smoothing
constant
Denotes the importance
alpha α of the past error

53
Time Series: Exponential Smoothing (ES)

• The model depends on three pieces of data:

– Most recent actual
– Most recent forecast
– Smoothing constant.
• The value of alpha assigned as a smoothing constant
is critical to the forecast.
– If α is low: there is little reaction to differences.
– If α is high: there is a lot of reaction to differences.
• The best alpha should be chosen on the basis of
minimal sum of error squared.

54
Time Series: Exponential Smoothing (ES)
• Several approaches are followed in selecting the
smoothing constant.
– If a great amount of smoothing is desired, a small
alpha should be chosen.
– The choice of alpha is affected by the characteristics
of the time series. If sharp ups and downs are
noticed in the data, the best smoothing constant is
0.1. That is alpha chosen should equal 0.1.
– If the data show that the past is very different from
the present, then alpha of 0.9 is appropriate.

55
Exponential smoothing: the method
Assume that we are currently in period t. It is calculated the
forecast for the last period (dt-1’) and we know the actual demand
last period (dt-1) …

Iniatilization of F1 (d1’) can be chosen using:

1. The Value d1 or
2. The Average of first 4 -5 month actual demand (dt)
Correlation between and N :

56
Example: bottled water at Kroger

Month Actual Forecasted a = 0.2

It was known that F1= 1,370
Jan 1,325 1,370
F2 = 0.2(1,325) + 0.8(1,370)
Feb 1,353 1,361

Mar 1,305 1,359

Apr 1,275 1,349

May 1,210 1,334

Jun ? 1,309
57
Example: bottled water at Kroger

Month Actual Forecasted  = 0.8

Jan 1,325 1,370

Feb 1,353 1,334

Mar 1,305 1,349

Apr 1,275 1,314

May 1,210 1,283

Jun ? 1,225
58
Impact of the smoothing constant

59
Trend..

What do you think will happen to a moving

average or exponential smoothing model
when there is a trend in the data?

60
Impact of trend

Sales
Actual
Regular exponential
Data smoothing will always lag
Forecast behind the trend.
Can we include trend
analysis in exponential
smoothing?

Month
61
 Double Exponential Smoothing
• Similar to the double moving average model.
• Also known as Brown’s double exponential smoothing
model.
• The model is represented as:

= forecast value x periods in the future

= the difference between the simple St’ and the
double St” smoothed values
= slope in a time series
m = number of periods ahead to be forecasted
62
 Double Exponential Smoothing
• To compute the difference between simple and double
smoothed values:

• Compute the intercept and the slope of the forecast

line as follows:

63
 Example of DES (BROWN)
t Demand St ' St'' at bt Forecast Demand
(dt) (dt’)
Ft+m = at + bt (m)
1 143 143 143    
2 152 144,8 143,36 146,24 0,36
3 161 148,04 144,3 151,784 0,936 146,6
4 139 146,23 144,68 147,781 0,387 152,72
5 137 144,39 144,62 144,148 -0,06 148,17
6 174 150,31 145,76 154,856 1,137 144,09
7 142 148,65 146,34 150,956 0,577 155,99
8 141 147,12 146,49 147,741 0,156 151,53
9 162 150,09 147,21 152,974 0,72 147,9
10 180 156,08 148,99 163,164 1,772 153,69
11 164 157,66 150,72 164,599 1,735 164,94
12 171 160,33 152,64 168,014 1,921 166,33
13           169,935
14           171,856

64
 Holt’s Exponential Smoothing (DES-Holt)
• To handle linear trend, similar to the Brown’s Method.
• The difference is that in this method we smooth the
trend and the slope in the time series by using different
constants for each.
• How do we find the best combination of smoothing
constant?
• Low values of alpha and gamma should be used when
there are frequent random fluctuations in the data.
• High values of alpha and gamma should be used when
there is a pattern such as trend in the data.

65
 Holt’s Exponential Smoothing (DES-Holt)

• The following equations are used when applying the

Holt’s method:

66
 Example DES-Holt α=0,2; γ=0,3
t Demand Pemulusan St Trend St Ramalan Permintaan (dt')
(dt) Nilai Ft+m = at + bt (m)
1 143 143 9  
2 152 152 9  
3 161 161 9 161
4 139 163,8 7,14 170
5 137 164,15 5,1 170,94
6 174 170,2 5,38 169,25
7 142 168,87 3,37 175,59
8 141 165,99 1,49 172,24
9 162 166,39 1,16 167,49
10 180 170,05 1,91 167,56
11 164 170,37 1,43 171,96
12 171 171,64 1,38 171,8
13       173,02
14       174,4

67
 Winters’ Seasonal Exponential Smoothing
• Allows for both trend and seasonal patterns to be taken
into account.
• This is an extension of the Holt’s method of smoothing.
• In computing the forecast, we add an equation for
seasonality as an index.
• The forecast model is:

68
 Winters’ Seasonal Exponential Smoothing

• The Winters’ model has the following components:

Smoothing value

Trend estimate

Seasonality estimate

α = coefficient of random factor = 0 < α < 1

β = coefficient of trend factor = 0 < β < 1
γ = coefficient for seasonal factor = 0 < γ < 1

69
 Example of Winter’s ES

  1992 1993 1994

Quarter 1 146 192 272
Quarter 2 96 127 155
Quarter 3 59 79 98
Quarter 4 133 186 219

α=0,2; β=0,1; γ=0,05 dan L = seasonal length = 4 quarter/year

70
• S1 = 0,2 (146/I­-3) + 0,8 (S0 + b0)
• Initialization b0:
– Average demand 1992 = 108,5
– Average demand 1993 = 146
– Trend difference = 146 – 108,5 = 37,5 for 1 year or for 1
quarter = 37,5/4 = 9,38 = b0
• Initialization S0:
For Quarterly: S0 = average demand – b0 (2,5)
For Monthly: S0 = average demand – b0 (6,5)

S0 = demand rata-rata – b0 (2,5)

S0 = 108,5 – (9,38) (2,5) = 85,05

71
 Determination of Initial Value
Trend Line Sales Estimates  
  1992 1993  
Quarter 1 85,05 + 9,38 = 94,43 122,57 + 9,38 = 131,95  
 
Quarter 2 94,43 + 9,38 = 103,81 131,95 + 9,38 = 141,33
Quarter 3 103,81 + 9,38 = 113,19 141,33 + 9,38 = 150,71  
Quarter 4 113,19 + 9,38 = 122,57 150,71 + 9,38 = 160,09  
 
Seasonal Index Estimates = demand/ trend line estimates
  1992 1993 Average Adjustment
Quarter 1 146/ 94,43 = 1,55 1,46 (1,55+1,46)/ 2 = 1,51 1,51 * (4/4,07) = 1,48 = I-3
Quarter 2 96/ 103,81 = 0,92 0,9 0,91 0,91 * (4/4,07) = 0,89 = I-2
Quarter 3 0,52 0,52 0,52 0,51 = I-1
Quarter 4 1,09 1,16 1,13 1,11 = I0
4,07 4

72
73
S1 = 0,2 (146/1,48) + 0,8 (85,05 + 9,38) = 95,27
b1 = 0,1 (95,27 – 85,05) + 0,9 (9,38) = 9,46
F2 = (95,27 + 9,46)*0,89 = 93,21
I1 = 0,05 (146/95,27) + 0,95 (1,48) = 1,48
 
S2 = 0,2 (96/0,89) + 0,8 (95,27+9,46) = 105,36
b2 = 0,1 (105,36-95,27) + 0,9 (9,46) = 9,52
F3 = (105,36+9,52)*0,51 = 58,59
I­2 = 0,05 (96/105,36) + 0,95 (0,89) = 0,89

74
Time Series: Regression Analysis

• Linier pattern

75
 Time Series: Regression Analysis – Cyclical pattern

N = jumlah periode dalam 1 siklus

n = jumlah data (periode) 76
How can we compare across forecasting models?

We need a metric that provides estimation of accuracy

Errors can be:

Forecast Error 1. biased (consistent)

2. random

Forecast error = Difference between actual and forecasted value

(also known as residual)
77
 Measure of Accuracy

• MAD
• Mean Squared Error (MSE)
• Standard Error of Estimation (SEE)
• Mean Absolute Percentage Error (MAPE)
• Mean Forecast Error (MFE) or Mean Error (ME) or Bias

78
Measuring Accuracy: MAD

MAD = Mean Absolute Deviation or Mean Absolut Error (MAE)

It is the average absolute error in the observations

1. Higher MAD implies worse performance.

2. If errors are normally distributed, then σε=1.25MAD

79
Measuring Accuracy: MSE, SEE

Where:
• dt = actual demand at period t
• dt’ = forecast demand at period t
• n = number of periods

f = degree of freedom (constant = 1, Linier = 2, Cyclical

Quadratic = 3, Cyclical Linier = 4). Note: SEE if is for
regression analysis

80
Measuring Accuracy: MAPE

• PE = percengahe of error

81
Measuring Accuracy: MFE/ME

MFE = Mean Forecast Error (Bias)

It is the average error in the observations

Be careful, the result can be negative value

A more positive or negative MFE implies worse performance;
the forecast is biased.
82
 MFE & MAD:
A Dartboard Analogy

Low MFE & MAD:

The forecast errors

are small &
unbiased

83
An Analogy (cont’d)

Low MFE but high

MAD:

On average, the
arrows hit the
bullseye (so much
for averages!)
84
MFE & MAD:
An Analogy

High MFE & MAD:

The forecasts
are inaccurate &
biased

85
 Key Point

Forecast must be measured for accuracy!

The most common means of doing so is by

measuring the either the mean absolute
deviation or the standard deviation of the
forecast error

86
Validation: Tracking signal
The tracking signal is a measure of how often our estimations
have been above or below the actual value. It is used to decide
when to re-evaluate using a model.

Positive tracking signal: most of the time actual values are

above our forecasted values
Negative tracking signal: most of the time actual values are
below our forecasted values

If TS > 4 or < -4, investigate!

87
Example: bottled water at Kroger

Month Actual Forecast Month Actual Forecast

Jan 1,325 1,370 Jan 1,325 1370

Feb 1,353 1,361 Feb 1,353 1306

Mar 1,305 1,359 Mar 1,305 1334

Apr 1,275 1,349 Apr 1,275 1290

May 1,210 1,334 May 1,210 1251

Jun 1,195 1,309 Jun 1,195 1175

Exponential Smoothing Forecasting with trend

( = 0.2) ( = 0.8)
( = 0.5)

Question: Which one is better?

88
Bottled water at Kroger: compare MAD and TS

MAD TS

Exponential
Smoothing 70 - 6.0

Forecast
Including Trend 33 - 2.0

We observe that FIT performs a lot better than ES

Conclusion: Probably there is trend in the data which

Exponential smoothing cannot capture
89
 Validation: Residual/Error Checking
• Residuals are independent (ACF test)
• Residuals are normally distributed (normality test)

Autocorrelation Function for Residual Probability Plot of Residual

(with 5% significance limits for the autocorrelations) Normal
99
1.0 M ean 4.125
StDev 40.07
0.8 95 N 8
AD 0.228
0.6 90
P-Value 0.720
80
Autocorrelation

0.4
70
0.2

Percent
60
0.0 50
40
-0.2 30
20
-0.4
10
-0.6
5
-0.8

-1.0 1
-100 -50 0 50 100
1 2
Re sidual
Lag

Residual data are taken from Module p.55 , WMA(4)

90
 Which Forecasting Method Should You Use

• Gather the historical data of what you want to

forecast
• Divide data into initiation set and evaluation set (for
validating the forecast method)
• Use the first set to develop the models
• Use the second set to evaluate
• Compare the MADs and MFEs or MSE of each model
• Conduct the validation test
Note that the difference between MAD and MSE is that the
latter penalizes a forecast much more for extreme deviations
than it does for small ones.
91
Component/ Process Methods
Constant Last Period Demand
Arithmetic Average (Average Methods)
Single Moving Average (SMA)
Weighted Moving Average (WMA)
Single Exponential Smoothing (SME)
Regression Analysis (Pola Konstan)
Trend Double Moving Average (DMA)
Double Exponential Smoothing-Browns
Double Exponential Smoothing-Holt
Exponential Smoothing Pegels
Regression Analysis (Pola Linier)
Seasonal Tripel Exponential Smoothing (TES)-Winter
Exponential Smoothing Pegels
Regression Analysis (Pola Siklis)
Trend+Seasonal Exponential Smoothing Pegels
Regression Analysis (Pola Linier Siklis)

92
 GROUP ASSIGNMENT (use
Excel)
• Find the time series data for the
website (secondary data) for at least
60 periods (daily, weekly, quarterly or
monthly data). Provide the website
link.
• Present the data as Excel table and
draw the data pattern
• Determine the 2 appropriate
forecasting methods that fit to the
data pattern
• Evaluate accuracy of the forecasting
Note: every group should and choose the best one and do the
present the different data validation 93