3a.

Exponential Smoothing

 Assumes the most recent observations have

the highest predictive value
 gives more weight to recent time periods

FFt+1
t+1
=
= F
F tt
+
+ (A
(A tt
-
- F
Ft)
t)
et

Ft+1 = Forecast value for time t+1 Need

Needinitial
initial
At = Actual value at time t forecast
forecastFFt t
to
tostart.
 = Smoothing constant start.
3a. Exponential Smoothing – Example 1
FFt+1
t+1
=
= F
F tt
+
+ (A
(A tt
-
- F
Ft)
t)
i Ai
Week Demand
1 820 Given
Given the
the weekly
weekly demand
demand
2 775 data
data what
what are
are the
the exponential
exponential
3 680 smoothing
smoothing forecasts
forecasts for
for
4 655 periods
periods 2-10 using =0.10?
2-10 using =0.10?
5 750
6 802 Assume
Assume FF11=D
=D11
7 798
8 689
9 775
10
3a. Exponential Smoothing – Example 1
FFt+1
t+1
=
= F
F tt
+
+ (A
(A tt
-
- F
Ft)
t)
i Ai Fi
Week Demand  = 0.1 0.6
1 820 820.00 820.00
2 775 820.00 820.00
3 680 F2815.50
= F1+ (A793.00
1–F1) =820+(820–820)
4 655 801.95 725.20=820
5 750 787.26 683.08
6 802 783.53 723.23
7 798 785.38 770.49
8 689 786.64 787.00
9 775 776.88 728.20
10 776.69 756.28
3a. Exponential Smoothing – Example 1
FFt+1
t+1
=
= F
F tt
+
+ (A
(A tt
-
- F
Ft)
t)
i Ai Fi
Week Demand  = 0.1 0.6
1 820 820.00 820.00
2 775 820.00 820.00
3 680 815.50 793.00
F3 = F2+ (A2–F2) =820+(775–820)
4 655 801.95 725.20
5 750 787.26 683.08=815.5
6 802 783.53 723.23
7 798 785.38 770.49
8 689 786.64 787.00
9 775 776.88 728.20
10 776.69 756.28
3a. Exponential Smoothing – Example 1
FFt+1
t+1
=
= F
F tt
+
+ (A
(A tt
-
- F
Ft)
t)
i Ai Fi
Week Demand  = 0.1 0.6
1 820 820.00 820.00
2 775 820.00 820.00
3 680 815.50 793.00
4 655 801.95 725.20
5 750 787.26 683.08
6 802 783.53 723.23This process
7 798 785.38 770.49 continues
8 689 786.64 787.00
through week 10
9 775 776.88 728.20
10 776.69 756.28
3a. Exponential Smoothing – Example 1
FFt+1
t+1
=
= F
F tt
+
+ (A
(A tt
-
- F
Ft)
t)
i Ai Fi
Week Demand  = 0.1  = 0.6
1 820 820.00 820.00
2 775 820.00 820.00
3 680 815.50 793.00
4 655 801.95 725.20
5 750 787.26 683.08 What if the
6 802 783.53 723.23  constant
7 798 785.38 770.49 equals 0.6
8 689 786.64 787.00
9 775 776.88 728.20
10 776.69 756.28
3a. Exponential Smoothing – Example 2
FFt+1
t+1
=
= F
F tt
+
+ (A
(A tt
-
- F
Ft)
t)
i Ai Fi
Month Demand  = 0.3  = 0.6
January 120 100.00 100.00
February 90 106.00 112.00
March 101 101.20 98.80
April 91 101.14 100.12
May 115 98.10 94.65 What if the
June 83 103.17 106.86  constant
July 97.12 92.54 equals 0.6
August
September
 3a. Exponential Smoothing – Example 3

Company
Company A, A, aa personal
personal computer
computer producer
producer
purchases
purchases generic
generic parts
parts and
and assembles
assembles themthem toto
final
final product.
product. Even
Even though
though mostmost ofof the
the orders
orders
require
require customization,
customization, they
they have
have many
many common
common
components.
components. Thus,
Thus, managers
managers of of Company
Company A A need
need
aa good
good forecast
forecast ofof demand
demand so so that
that they
they can
can
purchase
purchase computer
computer parts
parts accordingly
accordingly to to minimize
minimize
inventory
inventory cost
cost while
while meeting
meeting acceptable
acceptable service
service
level.
level. Demand
Demand datadata for
for its
its computers
computers for for the
the past
past 55
months
months isis given
given in
in the
the following table..
following table
3a. Exponential Smoothing – Example 3
FFt+1
t+1
=
= F
F tt
+
+ (A
(A tt
-
- F
Ft)
t)
i Ai Fi
Month Demand  = 0.3  = 0.5
January 80 84.00 84.00
February 84 82.80 82.00
March 82 83.16 83.00
April 85 82.81 82.50
May 89 83.47 83.75 What if the
June 85.13 86.38  constant
July ?? ?? equals 0.5
3a. Exponential Smoothing
How to choose α
depends on the emphasis you want to place
on the most recent data

Increasing α makes forecast more

sensitive to recent data
Forecast Effects of
Smoothing Constant 
Ft+1 = Ft +  (At - Ft)
or Ft+1 =  At + (1- ) At - 1 + (1- )2At - 2 + ...
w1 w2 w3

Weights
= Prior Period 2 periods ago 3 periods ago
 (1 - ) (1 - )2

= 0.10
10% 9% 8.1%
= 0.90 90% 9% 0.9%
To Use a Forecasting Method

 Collect historical data

 Select a model
 Moving average methods
 Select n (number of periods)
 For weighted moving average: select weights
 Exponential smoothing
 Select 

 Selections should produce a good forecast

…but what is a good forecast?
A Good Forecast

 Has a small error

 Error = Demand - Forecast
Measures of Forecast Error
et
nn

a. MAD = Mean Absolute Deviation 

 AA --FF
t=1
tt tt
MAD
MAD== t=1
nn

nn

b. MSE = Mean Squared Error  A - F 

 At t - Ft t 
22

t=
t =11
MSE =
MSE =
nn

c. RMSE = Root Mean Squared Error RMSE

RMSE == MSE
MSE

 Ideal values =0 (i.e., no forecasting error)

 nn

MAD Example 
 AA --FF
t=1
tt tt = 40 =10
MAD
MAD== t=1 4
nn

What
What isis the
the MAD
MAD value
value given
given the
the
forecast
forecast values
values in
in the
the table
table below?
below?
At Ft
Month Sales Forecast |At – Ft|
1 220 n/a
2 250 255 5
3 210 205 5
4 300 320 20
5 325 315 10
= 40
 nn

 A - F 
 At t - Ft t 
22

= 550 =137.5
MSE/RMSE Example MSE =
MSE =
t =t =11
nn 4

What
What isis the
the MSE
MSE value?
value? RMSE = √137.5
=11.73
At Ft
Month Sales Forecast |At – Ft| (At – Ft)2
1 220 n/a
2 250 255 5 25
3 210 205 5 25
4 300 320 20 400
5 325 315 10 100
= 550
 Measures of Error
1. Mean Absolute Deviation
(MAD)
n
t At Ft et |et| et 2
e
MAD  1
t
84 = 14
Jan 120 100 20 20 400 n
6
-16 16
Feb 90 106 256 2a. Mean Squared Error
-1 1 1 (MSE)
Mar 101 102 n

-10 10 100 
 te  2

April 91 101 MSE  1 1,446

17 17 289 n = 241
May 115 98 6
-20 20 400
2b. Root Mean Squared Error
June 83 103 (RMSE)
-10 84 1,446
An accurate forecasting system will have small MAD, RMSE  MSE
MSE and RMSE; ideally equal to zero. A large error may
indicate that either the forecasting method used or the = SQRT(241)
parameters such as α used in the method are wrong.
Note: In the above, n is the number of periods, which is
=15.52
Forecast Bias

 How can we tell if a forecast has a positive or

negative bias?

 TS = Tracking Signal
Good tracking signal has low values

RSFE  (actual t  forecast t )

TS = = t
MAD Mean absolute
MAD deviation
30
Quantitative Forecasting Methods

Quantitative
Forecasting

Time Series Regression

Models Models

2. Moving 3. Exponential
1. Naive
Average Smoothing
a) simple a) level
b) weighted b) trend
c) seasonality
Exponential Smoothing (continued)

 We looked at using exponential

smoothing to forecast demand with
only random variations
Ft+1 = Ft +  (At - Ft)
Ft+1 = Ft +  At –  Ft
Ft+1 =  At + (1-) Ft
Exponential Smoothing (continued)

 We looked at using exponential

smoothing to forecast demand with
only random variations
 What if demand varies due to
randomness and trend?

 What if we have trend and seasonality

in the data?
 Regression Analysis as a Method for
Forecasting
Regression analysis takes advantage
of the relationship between two
variables. Demand is then
forecasted based on the
knowledge of this relationship and
for the given value of the related
variable.

Ex: Sale of Tires (Y), Sale of Autos (X)

are obviously related

If we analyze the past data of these

two variables and establish a
relationship between them, we may
use that relationship to forecast the
sales of tires given the sales of
automobiles.

The simplest form of the relationship

is, of course, linear, hence it is Sales of Autos (100,000)
referred to as a regression line.
 Formulas
yy == aa ++ bb xx

where,
where,


x

xy  n x y
y
b
 x  nx2 2

x
y
a  y  bx
Regression – Example
yy == aa++ bb X
X b
 xy  n x y a  y  bx
 x  nx
2 2

MonthAdvertising Sales X 2 XY
January 3 1 9.00 3.00
February 4 2 16.00 8.00
March 2 1 4.00 2.00
April 5 3 25.00 15.00
May 4 2 16.00 8.00
June 2 1 4.00 2.00
July

TOTAL 20 10 74 38
 General Guiding Principles for
Forecasting

1. Forecasts are more accurate for larger groups of items.

2. Forecasts are more accurate for shorter periods of time.
3. Every forecast should include an estimate of error.
4. Before applying any forecasting method, the total system
should be understood.
5. Before applying any forecasting method, the method should
be tested and evaluated.
6. Be aware of people; they can prove you wrong very easily
in forecasting
 READ THE CHAPTERS ON
 Forecasting
 Product and service design