Demand Estimation and

Forecasting
Demand Estimation and Forecasting
The first question which arises is, what is the
difference between demand estimation and demand
forecasting?
Estimation attempts to quantify the links between
the level of demand and the variables which
determine it.
Forecasting attempts to predict the overall level of
future demand rather than looking at specific
linkages.
For this reason the set of techniques used may differ,
although there will be some overlap between the
two.
Demand Estimation and Forecasting
In general, an estimation technique can be used
to forecast demand but a forecasting technique
cannot be used to estimate demand.

A manager who wishes to know how high
demand is likely to be in two years time might
use a forecasting technique.
A manager who wishes to know how the firms
pricing policy could be used to generate a given
increase in demand would use an estimation
technique.
Marketing Research Approaches to
Demand Estimation
1. Consumer Surveys
2. Consumer Clinics
3. Market Experiments
4. Virtual Shopping & Virtual Management
5. Regression Analysis
Marketing Research Approaches to
Demand Estimation.
Consumer Surveys: The consumer survey
requires questioning customers or potential
customers in an attempt to estimate the
relation between demand for a firms
products and a variety of factors thought to
be important for marketing and profit-
planning purposes.
Marketing Research Approaches to
Demand Estimation.
The technique can be simply applied by
asking question to the consumers about the
quantity of the product they would purchase
at different prices.
In more sophisticated approaches, trained
interviewers develop questionnaire, and
collect information from selected sample.
Sound information can be collected from
consumer survey.
Marketing Research Approaches to
Demand Estimation.
A firm might ask its consumers about
projected purchases under a variety of
different conditions relating to price,
advertising expenditures, prices of
substitutes, and complements, income, and
other variables in the demand function.
Then, firm could forecast total demand by
estimating important parameters in the
demand function.
Marketing Research Approaches to
Demand Estimation.
Limitations of the Consumer Survey Method:
It is not applicable in all cases.
The Consumer survey method cannot collect
quality and quantity information in all the cases.
Usually, consumers could not/ are unwilling to
answer the hypothetical questions. If you ask:
How will you react if 10% price of our product
will be increased?, consumers could not predict
the amount they would purchase exactly.
.

Marketing Research Approaches to
Demand Estimation.
Consumer Clinics/ Laboratory Experiment:
It is controlled experiment.
Participants are given a sum of money and asked
to spend it in a simulated store to see how they
react to changes in the price, packaging,
displays, price of competing goods, and other
factors of demand.
Advocates of this method argue that it is
superior to the consumer survey method as
participants are selected to spend the money
on the good they like most.
Marketing Research Approaches to
Demand Estimation.
Limitations of the Consumer Clinics Method:
Participants are not likely to act really
(compared to actual market) as they know that
they are in artificial situation.
Participants may act to influence firms
decision/strategy.
Generally, small sample size is taken as huge
money is required to conduct the experiment on
large consumers. Appropriate information
cannot be collected from small sample size.

Marketing Research Approaches to
Demand Estimation.
Market Experiments:
Market experiments are conducted under actual
marketplace.
There are many ways of performing market
experiments.
Example: A firm can select three markets A, B,
and C. The firm can change price in market A, it
can change advertisement expenditure in
market B, and it can change packing in market C.
Marketing Research Approaches to
Demand Estimation.
Then, the firm can compare and analyze the
impact of changes in price, advertising
expenditures and packaging.
Study Yourself:
Case Study 4-1: Page 126 (Salvatore)

Marketing Research Approaches to
Demand Estimation.
Virtual shopping and Virtual Management:
It has been recently developed approach.
It is based on modern technology. A represented
sample of consumers shop in a virtual store is
simulated in the computer screen. Consumers
can see shelves of various goods in the screen.
He can see the label of the good he likes by
touching the image of the good.
Consumer can buy the good by touching the
shopping cart.
Marketing Research Approaches to
Demand Estimation.
Virtual shopping and Virtual Management can
be clarified by the Case Study (Case Study 4-3:
Reaching Consumers in the Vanishing Mass
Market: Dominick Salvatore, page No. 129). The
case study adds that traditional way of
advertisement could not capture the attention
of new generation. New generation is so busy,
they feel that they dont have time to watch on
the traditional advertisement. They are
watching two or more TV channels
simultaneously. Virtual management is so
important nowadays.

Marketing Research Approaches to
Demand Estimation.
Firms need to follow virtual management to
deal with the intention of new generation.
..


Marketing Research Approaches to
Demand Estimation.
Regression Analysis:
Regression analysis is the most commonly
used approach to demand estimation.
British Scholar Francis Galton introduced
the term REGRESSION.







Types of
Regression Models
17
Regression
Models
Linear
Non-
Linear

2+ Explanatory
Variables
Simple Multiple
Linear
1 Explanatory
Variable
Non-
Linear
Marketing Research Approaches to
Demand Estimation.
Regression Analysis (contd)
Regression analysis is concerned with the
study of the dependence of one variable, the
dependent variable, on one or more other
variables, the explanatory variables, with a
view to estimating and/or predicting the
population mean or average value of the
former in terms of the known or fixed value
of the latter.
Marketing Research Approaches to
Demand Estimation.
Difference between actual value and
estimated value is called residual. Residual is
denoted by e or u.
Here, e = Y ^ Y
According to OLS, sum of squares of residuals
must be minimum to estimate the Best fit
for the regression line.

Demand Estimation By Regression
Analysis
Though various other methods can be used
for demand estimation, regression analysis is
most commonly used method.
Regression analysis:
provides more information,
is generally less expensive.

Demand Estimation By Regression
Analysis
Following steps are followed for demand
estimation by regression analysis:
Step 1 Model Specification: First of all, the we
have to specify the model to be specified. This
involves the most important variables (such as
price of the commodity P
X
, consumer income I,
the number of consumers in the market N, the
price of related goods P
Y
, consumers taste T
etc) that are believed to affect the demand for
the commodity.


Demand Estimation By Regression
Analysis
We may specify the following general
demand function:
Q
X
= f(P
X
, I, N, P
Y
, T,)

Specifying the model is so challenging.
We should include five/six variables that
are believed to affect the demand
highly. Taking too many variables
creates econometric difficulties such as
multicollinearity.



Demand Estimation By Regression
Analysis
Step 2 Collecting Data on the Variables:
Data for required variables should be
collected. We may collect either TIME-SERIES
DATA or CROSS-SECTIONAL DATA or,
POOLED DATA.


Demand Estimation By Regression
Analysis
A time series data is obtained by
observing response variable at regular
time periods (yearly, quarterly, monthly.
etc.).
Example
Year: 1995 1996 1997 1998 1999
Sales: 78.7 63.5 89.7 93.2 92.1
Demand Estimation By Regression
Analysis
In a cross-sectional data, all
observations in the sample are taken
from the same point in time and
represent different individual entities
(such as households, houses, etc.)

Student ID Sex Age Height Weight
777672431 M 21 61 178 lbs.
231098765 M 28 511 205 lbs.
111000111 F 19 58 121 lbs.
898069845 F 22 54 98 lbs.
000341234 M 20 62 183 lbs
Example of Cross-sectional
Data


Demand Estimation By Regression
Analysis
The of data to be collected depends on the
availability.
Sometimes, we may use proxy data. For
example, a proxy for consumers price
expectation in each period might be the actual
price changes from the previous period.
Sometimes, we could drop the variable such as
taste if we could not quantify it.
For secondary data, the source should be
reliable.

Demand Estimation By Regression
Analysis
Step3 Specifying the Form of Demand Function:
Third step is to specify the functional form of
the model to be estimated. The model can
be specified either in the linear or non-linear
form.
The simplest form, probably also the most
realistic, is the linear form.
Demand Estimation By Regression
Analysis
A linear form of the generalized demand
function is:
Q
X
= a
0
+ a
1
P
X
+ a
2
I, N + a
3
N + a
4
P
Y
+ ..+ e
Where as are the parameters (coefficients) to be
estimated, and e is the error term.
Interpretation of parameters is crucial.
a
1
is the marginal change in Q
X
due to one unit
change in P
X.

Similarly, we can interpret other parameters.
Demand Estimation By Regression
Analysis
In some cases, non-linear relationship fit the
data better than the linear form.
Suppose, we specify the model in non-linear
form as
Q
X
= a (P
X
)
b
1
(I)
b
2

To make it linear, we take log on both sides,
log Q
X
= log a + b
1
log P
X
+ b
2
log I

(i)

It is called Log-linear form.

Demand Estimation By Regression
Analysis
Interpretation in log-linear form is different
from linear form.
Parameters b
1
and b
2
in (i), indicates the
percentage changes (or average elastic ties).
b
1
= price elasticity of demand ,i.e., b
1
is the
percentage change in due to one unit
change in price.

b
2
= income elasticity of demand , i.e., b
2
is
the percentage change in due to one unit .

Demand Estimation By Regression
Analysis
Model (i) is called log-linear form
because it can be written as linear form.
From (i),
log Q
X
= log a + b
1
log P
X
+ b
2
log I


Putting log Q
X
= Y,
log a = A,
log P
X
= X
1,
and
log I

= X
2,
we get

Y

= A+ b
1
X
1
+ b
2
X
2 ,
which takes the
linear form.





Demand Estimation By Regression
Analysis
Now, question arises: which form is better?
In real world, generally, both linear and non-
linear models are estimated. And, one that
gives the better results is reported.
..
Demand Estimation By Regression
Analysis
Step 4 Testing the Econometric Results:
Fourth and last step in estimation of demand
function by regression analysis is to
test/evaluate the regression results. Mainly,
we test/examine:
(i) Sign of the estimated slope coefficient (s)
must be checked to see if it conforms to
what is postulated in theoretical grounds.




Demand Estimation By Regression
Analysis
(ii) t-test must be conducted on the statistical significance
of the estimated parameters to determine the degree of
confidence that we can have in each of the estimated
slope coefficients..
(iii) Finally, we see whether the estimated demand
equation pass other econometric tests or not, i.e.,
econometric problems such as multicollinearity,
heteroscedasticity, and autocorrelation should not be
there in the model. If any of these problems are detected
from the tests, measures must be applied to overcome
these problems. Generally, heteroscedasticty is more
likely to present if cross-sectional data are used, whereas
autocorrelation is more likely to present when time-series
data are used. Multicollinearity arises when..
Demand Estimation By Regression
Analysis
Case Study 4-4:Estimation of the Demand for
Air Travel
page 153, Salvatore
East-west airlines estimated the demand for air travel
between India and Singapore from 1994 to 2000.
Regression equation is

ln Q
t
= 3.342 1.452 In P
t
+ 1.92 In GNP
t

(-5.071) (7.286)
R

2
= 0.95

Demand Estimation By Regression
AnalysisCase Study
Case Study (contd)
Where, Q
t
= number of passengers per year
travelling (in thousands)


P
t
= average yearly airfare
GNP
t
= Indian gross product
Numbers in parentheses refer to the
estimated t-statistics.
Estimated coefficients give elasticities
because demand equation is transformed by
using natural log.





Demand Estimation By Regression
Analysis
Case Study(contd)
Here, price elasticity = -1.452, which
indicates that a 10% increase in average
airfares would reduce the number of airline
passengers by 14.52%.
Here, income elasticity = 1.92, which
indicates that a 10% rise in GNP in India
would increase the number of passenger by
19.2%.

Demand Estimation By Regression
Analysis
Case Study(contd)
Since absolute value of price elasticity is more
than 1, demand for air travel between India and
Singapore is price elastic.
Since of value of income elasticity is more than
1, demand for air travel between India and
Singapore is luxury.
We can find very nice result as (i) signs of the
estimated slope coefficients are as postulated




Demand Estimation By Regression
Analysis
Case Study(contd)
by demand theory (ii) R

2
= 0.95, i.e., airfare and
GNP explain 95% of the variation in
percentage of number of passengers flying
between India and Singapore.

Question
Explain concept of the demand estimation by
regression analysis. (10 marks)
or,
Explain the steps of demand estimation by
regression analysis. Explain with one
example. (10 marks/20 marks)
Demand Estimation and
Forecasting
The firm needs to have information about
likely future demand in order to pursue
optimal pricing strategy. It can only charge a
price that the market will bear if it is to sell
the product.
Most business decisions are made in the face
of risk or uncertainty. The aim of economic
forecasting is to minimize risk or uncertainty.

Demand Estimation
and Forecasting
Forecasting the demand and sales of the firms
product usually begins with a macroeconomic
forecast of the general level of economic activity
for the economy as a whole, or gross national
product (GNP) because demand and sales of the
products varies with the of economy.
The data for macroeconomic forecast are
provided by government and non-government
agencies.

Demand Estimation
and Forecasting
A firm /industry use macroeconomic
forecasts as inputs.
Firms use various methods for forecasting.
Selection of the forecasting method depends
on :
(i) the cost of preparing the forecast and
the benefit of that results from its use,
(ii) the lead time in decision making,

Demand Estimation
and Forecasting
(iii) the time period of the forecast (short-
term or long-term),
(iv) the level of accuracy desired,
(v) the quality and availability of the data,
and
(vi) the level of complexity of the
relationships to be forecast.
The goal of forecasting
To transform available data into equations
that provide the best possible forecasts of
economic variablese.g., sales revenues and
costs of productionthat are crucial for
management.



Types of Forecasting

Various types of classification
of forecasting.
One way of classification:
1. Short-run forecasting
2. Long-run forecasting
Purpose of Short-run Forecasting
Deciding appropriate price policy
Setting sales targets
Formulating an appropriate production policy
Forecasting the financial requirement
Reducing cost of purchasing raw materials and
controlling inventory.
Evolving a suitable advertising and promotional
programmed

Purposes of Long-run Forecasting
Planning of new unit and expansion of an
existing unit
Planning long-term financial requirements
Planning manpower requirements
Providing guideline for demand forecasts for
related industries
Guiding the government

Need of Demand Forecasting
Predicting of sales
Arrangement of raw materials
Arrangement of labor
Capital expansion plan
Investment and inventory policy
Plant expansion or contraction plant

Steps in Demand Forecasting
The following steps must be considered for
demand forecasting:
Identification of objective
Establish the relation demand and
determinants
Selecting proper method of forecasting
Analysis and interpretation of results
Presenting the findings in readable form

Types of Demand Forecasting on the basis of
techniques
Two types of techniques:
Statistical methods
Survey methods
Statistical methods
Regression Analysis
Trend projection
Barometric methods
Input/output matrix


Survey Methods
1. Consumer Survey
2. Opinion Poll
Forecasting
Another way of classification of forecasting
is:
1. Qualitative Forecasting
2. Quantitative Forecasting
Forecasting Approaches
Used when situation is
vague & little data exist
New products
New technology
Involve intuition,
experience
e.g., forecasting sales on
Internet
Qualitative Methods Quantitative Methods
Forecasting Approaches
Used when situation is
vague & little data exist
New products
New technology
Involve intuition,
experience
Used when situation is
stable & historical data
exist
Existing products
Current technology
Involve mathematical
techniques
Quantitative Methods Qualitative Methods
Quantitative Forecasting
Select several forecasting methods
Forecast the past
Evaluate forecasts
Select best method
Forecast the future
Monitor continuously forecast accuracy
Causal
Models
Quantitative Forecasting Methods
Quantitative
Forecasting
Time Series
Models
Regression
Exponential
Smoothing
Trend
Models
Moving
Average
Criteria of good forecasting
Economy
Simplicity
Durability
Flexibility
Accuracy
Reliability


Limitations of demand
forecasting
Error
o Economic error
o Statistical error
o Instrumental error
o Human error
o Situational error
Durability of goods

Limitations ..
Biasness
o Investigator
o Informant
o Correspondence

Demand Forecasting By Time-Series
Analysis
Time-series data refers to the values of a
variable arranged chronologically by days,
weeks, months, quarters, or years.
Demand forecasting by time-series analysis
can be explained under the following steps:
Step 1: Plotting in graph: First of all, past values
of variable that we seek to forecast (say, sales
of a firm) on Y-axis, and time on the X-axis.
Demand Forecasting By Time-Series
Analysis
We may observe the movement of the time-
series over time.
We forecast/predict the values for future on
the basis of past values. Assumption is that
time series will continue to move as in the
past. This is why time-series analysis is often
referred as naive forecasting.
Components of Time-Series Data/
Reasons of Fluctuations in Time-series
Data:
1. Secular trend
2. Cyclical fluctuations
3. Seasonal variation
4. Irregular/ random Influences
Components of Time Series
Trend
Seasonal
Cyclical
Irregular
Components of
1. Secular Trend: Secular trend refers to a long-
run increases or decrease in the data. For
example, sales of a good may be affected by
(i) increase in population, and (ii) increase in
income.

Trend Component
Overall Upward or Downward Movement
Data Taken Over a Period of Years

Sales
Time
Components of
2. Cyclical Fluctuations: Cyclical fluctuations
are refer to economic expansions and
contractions. Business cycles affect the data.
Cyclical Component
Repeating up & down movements
Due to interactions of factors influencing
economy
Usually 2-10 years duration
Mo., Qtr., Yr.
Response
Cycle
Components of .
3. Seasonal Variation: Seasonal variation refers
to the regularly recurring fluctuation in
economic activity during a certain time
period, generally during each year. Sale of
woolen clothes hikes in summer.
Seasonal Component
Regular pattern of up & down fluctuations
Due to weather, customs etc.
Occurs within one year
Mo., Qtr.
Response
Summer
1984-1994 T/Maker Co.
Seasonal Component
Upward or Downward Swings
Regular Patterns
Observed Within One Year

Sales
Time (Monthly or Quarterly)
Components of
4.Irregular or Random Influences: Random
influences are the variations resulting from
natural disasters, strikes, earthquake or other
unusual events.

Irregular Component
Erratic, unsystematic, residual fluctuations
Due to random variation or unforeseen
events
Union strike
War
Short duration &
non-repeating
1984-1994 T/Maker Co.
Random or Irregular
Component
Due to Random Variations of
Nature
Accidents
Time Series Forecasting
Linear
Time
Series
Trend?
Smoothing
Methods
Trend
Models
Yes No
Exponential
Smoothing
Quadratic Exponential
Auto-
Regressive
Moving
Average
Trend Projection
This the simplest form of time-series analysis.
It deals with long-run effects.
Regression analysis used to take the form of
S
t
= S
0
+ b t
Where, S
t
= the value of sales to be forecasted
for year t

S
0
= estimated value of the sales in the base
year
b = absolute amount of growth of sales per year
Trend Projection
Suppose, fitting a regression line to the data
of computer sales from 2001 to 2011, we get
the estimated regression equation as
S
t
= 10 + 0.50 t
Suppose the estimated model is statistically
as well as econometrically significant.
Based on the past experience, we can
forecast the amount of computer sales in
2012, 2012, 2013, and so on.


Trend Projection
Remember that (in our example)
t = 1 for 2001,
t = 2 for 2002,
t = 3 for 2003 and so on.
We can forecast the sales for 2013, 2014, 2015,
and so on.
Sales for 2013 is S
13
= 10 + 0.5 (13) = 16.5 units
Sales for 2014 is S
14
= 10 + 0.5 (14) = 17 units
and so on.
Note: We can depict the trend line in the graph.
Trend Projection
Question 1: Using the method of least squares,
fit a straight trend line and find the trend
values for the following data. Also predict the
likely demand for 2013.

Year 2001 2002 2003 2004 2005 2006 2007
D(thous
ands)
59 62 61 65 67 65 69
Trend Projection
Solution: Let the trend line be Y = a + b x
where, x = X middle year (because n = 7,odd)
x = X - 2004
(Note: if n = even number, we put
x= 2(X A.M. of two middle years)
Then, we compete the table.

Trend Projection
Year (X) Demand (y) x = X- 2004 x
2
xy Trend
values y
c

2001 59 -3 9 -177 59.5
2002 62 -2 4 -124 61
2003 61 -1 1 -61 62.5
2004 65 0 0 0 64
2005 67 1 1 67 65.5
2006 65 2 4 130 67
2007 69 3 9 207 68.5
y = 448 x = 0

x
2
= 28

xy = 42

y
c
= 448

Trend Projection
Since, x = 0, a = x/n = 448/7 = 64
and, b = xy/ x
2
= 42/28 = 1.5
Puting these values of a and b in y = a + b x,
we get
Y = 64 + 1.5 x
Or, y
c
= 64 + 1.5 x which is the required trend
line.

Trend Projection
Putting x= -3, -2, -1, 0, 1, 2 and 3, we get the trend
values for 2001, 2002, 2003, 2004, 2005, 2006
and 2007 respectively as follows:
For x =-3, y
c
= 64 + 1.5 (-3) = 59.5
For x =-2, y
c
= 64 + 1.5 (-2) = 61
For x =-1, y
c
= 64 + 1.5 (-1) = 62.5
For x =0, y
c
= 64 + 1.5 (0) = 64
For x =1, y
c
= 64 + 1.5 (1) = 65.5
For x =2, y
c
= 64 + 1.5 (2) = 67
For x =3, y
c
= 64 + 1.5 (3) = 68.5

Trend Projection
Again, to forecast the trend value for 2013, put
X = 2013 in x = X -2004, we get
x = 2013-2004=9
Putting x = 9 in y
c
= 64 + 1.5 x, we get
y
c
= 64 + 1.5 (9) = 77.5
Hence, predicted demand for 2013 is 74.5
thousands.

Trend Projection
Question 2: Using the method of least squares,
fit a straight trend line and find the trend
values for the following data. Also predict the
likely demand for 2013.

Ye
ar
200
1
200
2
200
3
2004 200
5
200
6
2007 2008
D(th
ousan
ds)
59 62 61 65 67 65 69 68
Trend Projection
Solution: Let the trend line be Y = a + b x
where, x= 2(X A.M. of two middle years)
Here middle years are 2004 and 2005.
A.M. of middle years = (2004 + 2005) /2
= 2004.5
Then, we proceed as in Question No. 1.

Seasonal variation
It doesnt deal with long-run changes, it deals
with short-run changes.
Various methods are used for demand
forecasting by Seasonal variation.
Here, we illustrate method of Simple
averages.
Example: Suppose we have the monthly data
of computer sales (in hundreds) for three
years as given in the table ( in the next slide).


Seasonal Variation
Month Computer sales ( in hundreds)
2001 2002 2003
Jan 12 15 16
Feb 11 14 15
March 10 13 14
Apr 14 16 16
May 15 16 15
June 15 15 17
July 16 17 16
Aug 13 12 13
Sept 11 13 10
Oct 10 12 10
Nov 12 13 11
Dec 15 14 15
Seasonal Variation
We have to follow certain steps for forecasting.
Step 1: Arrange the data by years, months or
quarters or weeks
Step 2: compute the arithmetic average for each
period
Average for January = (12+15+16) /3 =14.33
Average for February = (11+14+15)/3=13.33
And so on.

Seasonal Variation
Step 3: Compute average of monthly averages
Average of monthly averages
= (14.33 + 13.33 + 12.33 + )/12 = 13.6625
Step 4: Calculate Seasonal index for each month
Seasonal index for January = (monthly average for
Jan/average of monthly averages) 100
=(14.33/13.6625) 100 = 104.88
Similarly, we can compute seasonal indices for
other months.
Month computer sales (in hundreds) Total Average Seasonal
Index
2001 2002 2003
Jan 12 15 16 43 14.33 104.88
Feb 11 14 15 40 13.33 97.56
March 10 13 14 37 12.33 90.24
Apr 14 16 16 46 15.33 112.20
May 15 16 15 46 15.33 112.20
June 15 15 17 47 15.66 114.62
July 16 17 16 49 16.33 119.52
Aug 13 12 13 38 12.66 92.66
Sept 11 13 10 34 11.33 82.92
Oct 10 12 10 32 10.66 78.02
Nov 12 13 11 36 12.00 87.83
Dec 15 14 15 44 14.66 107.30
Total 492 163.95 1200
Average 41 13.6625 100
Moving Average Method
Series of arithmetic means
Used only for smoothing
Provides overall impression of data over time
Moving Average Method
Series of arithmetic means
Used only for smoothing
Provides overall impression of data over time

Used for elementary forecasting
Moving Average Graph
0
2
4
6
8
93 94 95 96 97 98
Year
Sales
Actual
Moving Average
[An Example]
You work for Firestone Tire. You
want to smooth random
fluctuations using a 3-period
moving average.
1995 20,000
1996 24,000
1997 22,000
1998 26,000
1999 25,000
Moving Average
[Solution]
YearSales MA(3) in 1,000
1995 20,000 NA
1996 24,000 (20+24+22)/3 = 22
1997 22,000 (24+22+26)/3 = 24
1998 26,000 (22+26+25)/3 = 24
1999 25,000 NA

Moving Average
Year Response Moving
Ave
1994 2 NA

1995 5 3
1996 2 3
1997 2 3.67
1998 7

1999 6 NA
94 95 96 97 98 99
8
6
4
2
0
Sales
5
Moving Average
Year Response Moving
Ave
1994 2 NA
1995 5 3
1996 2 3
1997 2 3.67
1998 7 5
1999 6 NA
94 95 96 97 98 99
8
6
4
2
0
Sales
Exponential Smoothing
Method
Form of weighted moving average
Weights decline exponentially
Most recent data weighted most
Requires smoothing constant (W)
Ranges from 0 to 1
Subjectively chosen
Involves little record keeping of past data
Youre organizing a Kwanza meeting. You
want to forecast attendance for 1998 using
exponential smoothing
( = .20). Past attendance (00) is:
1995 4
1996 6
1997 5
1998 3
1999 7
Exponential Smoothing
[An Example]
1995 Corel Corp.
Exponential Smoothing
Time Y
i
Smoothed Value, E
i
(W = .2)
Forecast
Y
i + 1
1995 4 4.0 NA
1996 6 (.2)(6) + (1-.2)(4.0) = 4.4 4.0
1997 5 (.2)(5) + (1-.2)(4.4) = 4.5 4.4
1998 3 (.2)(3) + (1-.2)(4.5) = 4.2 4.5
1999 7 (.2)(7) + (1-.2)(4.2) = 4.8 4.2
2000 NA NA 4.8
E
i
= WY
i
+ (1 - W)E
i-1
^
Exponential Smoothing [Graph]
0
2
4
6
8
93 96 97 98 99
Year
Attendance
Actual
Weight
W is...
Prior Period
2 Periods
Ago
3 Periods
Ago
W W(1-W) W(1-W)
2
0.10 10% 9% 8.1%
0.90 90% 9% 0.9%
Forecast Effect of Smoothing
Coefficient (W)
Y
i+1
= WY
i
+ W(1-W)Y
i-1
+ W(1-W)
2
Y
i-2
+...
^
Linear Time-Series Forecasting
Model
Used for forecasting trend
Relationship between response variable Y &
time X is a linear function
Coded X values used often
Year X: 1995 1996 1997 1998 1999
Coded year: 0 1 2 3 4
Sales Y: 78.7 63.5 89.7 93.2 92.1

Linear Time-Series Model
Y
Time, X
1

Y b b X
i i

0 1 1
b
1
> 0
b
1
< 0
Linear Time-Series Model [An
Example]
Youre a marketing analyst for Hasbro Toys. Using
coded years, you find Y
i
= .6 + .7X
i
.
1995 1
1996 1
1997 2
1998 2
1999 4
Forecast 2000 sales.
^
Linear Time-Series [Example]
Year Coded Year Sales (Units)
1995 0 1
1996 1 1
1997 2 2
1998 3 2
1999 4 4
2000 5 ?
2000 forecast sales: Y
i
= .6 + .7(5) = 4.1
The equation would be different if Year used.
^
The Linear Trend Model
i i i
X . . X b b Y

743 143 2
1 0
Year Coded Sales
94 0 2
95 1 5
96 2 2
97 3 2
98 4 7
99 5 6
0
1
2
3
4
5
6
7
8
1993 1994 1995 1996 1997 1998 1999 2000
Projected to year
2000
Coeffi ci ents
I n te r ce p t 2. 14285714
X V a r i a b l e 1 0. 74285714
Excel Output
Time Series Plot
01/93 01/94 01/95 01/96 01/97
Month/Year
16
17
18
19
20
Number of Surgeries
(X 1000)
Surgery Data
(Time Sequence Plot)
Source: General Hospital, Metropolis
Time Series Plot [Revised]
01/93 01/94 01/95 01/96 01/97
Month/Year
183
185
187
189
191
193
Number of Surgeries
(X 100)
Revised Surgery Data
(Time Sequence Plot)
Source: General Hospital, Metropolis
Exponential Time-Series
Forecasting Model
Used for forecasting trend
Relationship is an exponential function
Series increases (decreases) at increasing
(decreasing) rate
Y
Year, X
1
Exponential Time-Series Model
Relationships
b
1
> 1
0 < b
1
< 1
Exponential Weight [Example Graph]
94 95 96 97 98 99
8
6
4
2
0
Sales
Year
Data
Smoothed
C o e f f i c i e n t s
In t e r c e p t 0 . 3 3 5 8 3 7 9 5
X V a r i a b l e 10 . 0 8 0 6 8 5 4 4
Exponential Trend Model
i
X
i
b b Y

1 0

or
1 1 0
b log X b log Y

log
i

Excel Output of Values in logs
i
X
i
) . )( . ( Y

2 1 17 2
Year Coded Sales
94 0 2
95 1 5
96 2 2
97 3 2
98 4 7
99 5 6
a n t i l o g (. 3 3 5 8 3 7 9 5 ) = 2 . 1 7
a n t i l o g (. 0 8 0 6 8 5 4 4 ) = 1 . 2