POLYTECHNIC UNIVERSITY

SALESIAN
---------------------------------------------ECUADOR

PR2DUCsales
Forecast-

Eng. Paul A. Pérez Gosende, MSc.

[email protected]
Forecast:
Process of estimating a future event by projecting past
-
T data into the future.

Prediction:
Process of estimating a future event based on
subjective considerations other than simple data from
the past.

3
Influence of forecasts in other areas
of the Directorate of
Operations

• Design of products

• Plant design

• Capacity management

• Inventory management

4
 DEMAND FORECAST

They are subjective, based on the opinions of

Models managers and experts, the results of consumer
surveys and the estimates of sales personnel.
qualitative When historical information is not available in
this regard, they are the only usable methods.

They are based on historical information

collected on the variables that influence the
Models process, using statistical techniques.
quantitative
DEMAND FORECAST

Qualitative models
 Delphi method
It is one of the best known and consists of selecting a
series of experts (who are unaware of the fact that they
have been selected), led by a coordinator who acts as an
intermediary between them. This asks them for opinions
and forecasts, and from among those received, it selects
those that fall into the 25% of the most pessimistic and
25% of the most optimistic, asking those people for
explanations of how they justify this fact. All information
received, including justifications, are sent to the experts,
repeating the process until a consensus is reached.
 Consult the sales force
It consists of requesting the opinions of sellers
or commercial agents about what will happen in
their area or region. The following problems
may arise: there are usually prejudices and
tendencies according to particular interests,
they can be affected by an excessively optimistic
or pessimistic character and many people must
be involved, which is costly due to the time
required to answer the questionnaires.
 consensus groups
It is based on free exchange in meetings or
meetings between a small group of people. The
idea is that group discussion produces better
forecasts than any individual alone. Participants
can be executives, salespeople or consumers.
Historical analogies
If two things have at least one common
characteristic, an analogy appears. The interest
of this technological forecasting method is to
investigate whether both things also present
other similar characteristics.
It allows the comparative analysis of new
products with other existing and similar ones on
the market.

Market research
It is established to collect data in various ways
(surveys, interviews, etc.) in order to test
hypotheses about the market. It is usually used
to forecast long-term sales and new products.
DEMAND FORECAST

Quantitative models
 Quantitative models

Time series
Causal models
models

1
 TIME SERIES ANALYSIS
It is a statistical method that depends heavily on
historical demand data, with which it projects its
future magnitude and recognizes seasonal trends
and patterns.
Methods
• Empirical forecast
• Simple Moving Average
• Weighted moving average
• Simple exponential smoothing
• Double exponential smoothing
• Simple linear regression
1
 Demand patterns
Repeated observations of the demand for a product or service in the
order they are made form a pattern known as a time series . The five
basic patterns of most time series applicable to demand are:
• HORIZONTAL
• TREND
• SEASONAL
• CYCLIC
• RANDOM

A time series can contain any combination of these patterns.

1
 Horizontal component: data fluctuates around a
horizontal line (a constant mean)
HEPLLE

1
Trend: data systematically increases or decreases over
time

Dantidam

1
Seasonal component: the data show increases and
decreases consistently, that is, they repeat depending on
the time of day, week, month or year.

1
Cyclical component: the data reveal gradual and less
predictable increases and decreases in demand, over long
periods (years, decades).

Random component: Indicates an unpredictable variation

in demand.

1
It is the result of fortuitous causes and cannot be predicted.
It's the main reason why all sales forecasts turn out to be
wrong.

2
 Forecast errors
Forecasts almost always contain errors. These are classified in two ways:
either as bias errors or as random errors.
• Bias errors are the result of systematic mistakes, which is why it is
observed that the forecast is always too high or too low. These errors
are often the result of overlooking or failing to correctly estimate
demand patterns, such as trend, seasonal, or cyclical patterns.
• Random error is the result of unpredictable factors that cause the
forecast to deviate from actual demand.
Forecast analysts try to minimize the effects of bias errors and random
errors by selecting appropriate forecast models, but it is impossible to
suppress errors in all their forms.

2
 Forecast error
In any forecasting method, it is important to
measure its accuracy. The forecast error is
simply the difference obtained by subtracting
the forecast from the actual demand in any
given period:
E t =D t -F t
E t : Forecast error in period t
D t : Real demand in period t
F t : Forecast for period t

2
 Forecast errors
• RSFE : Running sum of forecast errors . It is a measure of total
forecast error, which assesses the bias in a forecast.

RSFE = ∑E t

• BIAS: It is the average forecast error and is also a measure of

bias:
B IA S= 5S
In

2
 Forecast errors
The following errors measure the dispersion of the forecast:

• MSE: Mean Squared Error

M SE =LE í- n
• MAD: Mean Absolute Deviation
MAD = EE E 1 n
• σ: Standard deviation.
2(E
t - B IA S ) -
n- 1

2
 Forecast errors
• MAPE: Mean Absolute percent error. It measures the average
percentage that represents the absolute error of the forecast
with respect to the actual demand for each period
considered. It is the best error measure that can be used to
make comparisons between time series for different
products.

2-
Fel. 100
MAPE = —t--------
n

2
 Empirical forecast:
Time series method in which the demand
forecast for the next period is equal to the
demand observed in the current period, that is:
Ft+1 = Dt

Where:
F t+1 : Forecast
D t : Demand of the current period

2
 Example:
Forecast the annual demand for certain household
appliances. Calculate the MAD and BIAS of the made
forecast .
Real demand ( Forecast (
Years.
Dt ) Ft ) Error ( Et )
2007 845
2008 920
2009 988
2010 1172
2011 1145
2012 1174
2013 1475
2014

2
 n
XI E t I
t=1
Solution: We do not know the data of
real demand for the year
actual Forecast Mist 2006

Years therefore it is impossible

demand( Dt ( Ft ake
(Et
make a forecast for

.
2007 based on model
empirical, for this reason

2007 8) 45 ) the
model runs from year
2008.
2008 920 845 75
2009 988 920 68
2010 1172 988 184
2011 1145 1172 -27 Since it has not yet passed
year 2014 we do not know
the
2012 1174 1145 29 actual quantity of products
that our

2013 1475 1174 301 customers. For this reason

not
we can determine the
2014 1475 absolute deviation of
prognosis regarding
RSF 6 real values

BIAS RSFE E
2
With historical data
630 =L 3
6 those that are counted, the
■= = 105
0
absolute deviations under
Et E 8 this model has been possible
determine r for 6 periods

MAD = 684 = 114 t 4 (years). For this reason

we consider n=6
6 27
i
 Simple moving average:
The forecast is made by averaging the
demand of the n most recent periods
(generally an odd number), that is:

Ft41 = DttDt-11Dt-2tDt-
n+1
n

2
 The value of t indicates
that

Example
average of actual
demands
corresponding to the
three years prior to the
year that is forecast, that

:
Forecast the annual demand for certain is, a triennium.

household appliances (using t=3).

Calculate the
MAD and the BIAS of the forecast made.
Real demand ( Forecast (
Years.
Dt ) Ft ) Error ( Et )
2007 845
2008 920
2009 988
2010 1172
2011 1145
2012 1174
2013 1475

3
 2014
Solution
We do not know the data of
real demand for years
Real demand ( Forecast ( prior to 2007, therefore

Years. the forecasting model

Dt ) Ft ) Error (Et) this case (moving average

of the three periods
above) can only
2007 845 run from 2010.

2008 920
2009 988
2010 1172 917,67 254,33
2011 1145 1026,67 118,33 Since it has not yet passed
year 2014 we do not know

2012 1174 1101,67 72,33 the

actual quantity of products
that our
2013 1475 1163,67 311,33 customers. For this reason
not

2014 1264,67 we can determine the

absolute deviation of
prognosis regarding
RSFE =LE t = 756,32 real values
RSFE 756,33
BIAS
\Et = 756,32
= 189,08 E
n4 With the historical data
available, the absolute
deviations under this model
have been determined for 4
756,33 periods (years). For this
MAD = 189,08 reason we consider n=4 30

3
3
 Weighted moving average:
Each of the historical demands involved in the
average can have its own weighting. The sum of
such weights is equal to 1.

F t+1 = w,D + W2D_ +"' + w n^t-n+1

Example:
Forecast the annual demand for certain household appliances
(consider t =3 and weights of 0.5, 0.25 and 0.25). Calculate the

3
 MAD and BIAS of the made forecast.
Solution We do not know the data of
real demand for years

Real demand Forecast (

( Forecast (
prior to 2007, therefore

Years.
the forecasting model
Years. this case (moving average
Real demand ( DtDt)) Ft ) Ft Error
) (Et)
Error ( )
weighted of the three
Et
previous periods) only

2007 2007 845 845

can run from
2010.

2008 2008 920 920

2009 2009 988 988
2010 2010 1172 1172 935,25 236,75
2011 2011 1145 1145 1063,00 82
2012 2012 1174 1174 1112,50 61,50
2013 2013 1475 1475 1166,25 308,75
2014 2014 1317,25
R SFE =LE t = 689
BIAS RSFE
n = 689 = 172.25 4
E
\Et = 689 With the historical data
available, the absolute
n deviations under this model
2EtIt have been determined for 4
=1 689 periods (years). For this
MAD = 172.25 reason we consider n=4 33
4

3
3
Simple exponential smoothing:
It is a weighted moving average method that allows
calculating the average of a time series, assigning
recent demands greater weight than previous
demands.

F +1 = Ft +« (D t - F t)

Where:
α: smoothing parameter

3
 Example:
Forecast the annual demand for certain household

a
appliances (take = 0.2). Calculate the MAD and BIAS of
the made forecast.
Real demand ( Forecast (
Years.
Dt ) Ft ) Error ( Et )
2007 845
2008 920
2009 988
2010 1172
2011 1145
2012 1174
2013 1475

3
 2014
Solution To run the model we require previous
forecasts. F
F 2008 = F 2007 + 0.2(^ 2007
Since the exercise does not
2008 =845 + 0.2(845-845)
give us the forecast dataF 2008
used=845
for the
year 2007 (which is the first period in the
data series), the literature suggests that
Forecast ( the actual demand for that first period be
Years. taken as the value of forecast. In this

Real demand ( Dt ) Ft ) Error (Et) way, F 2007 =845 and as a consequence:

2007 845 F 2007

2008 920 845,00 75,00
)

2009 988 860,00 128,00

2010 1172 885,60 286,40
2011 1145 942,88 202,12
2012 1174 983,30 190,70
2013 1475 1021,44 453,56
2014 1112,15
1335.77 RSFE = ^Et = 1335,7
BIAS RSFE — — 000 4
n -- - - ZZz . O3
E
\Et = 1335,7 With the historical data
6 available, the absolute
deviations under this model
have been determined for 6
1335.77 periods (years). For this
MAD --------------= 222.63 reason we consider n=6 36

3
3
Smoothed (exponential) forecasts versus actual demand for a
product over time

When the product has a stable demand it is advisable to use a small value of ^ (0.05 or
0.1)
If demand is experiencing growth it would be better to consider a higher value of ^
(perhaps 0.15 or 0.30). The faster the growth, the higher the value of ^ should be. 37
 Double exponential smoothing:
(Also called trend-adjusted exponential smoothing)

An upward or downward trend in data collected over a sequence of periods

causes the exponential forecast to always fall short or lag behind the actual
occurrence.
Exponentially smoothed forecasts can be corrected by adding an adjustment to
the trends, for which two smoothing constants are used. In addition to α, the
trend equation uses a smoothing constant β, which reduces the impact of the
error that occurs between reality and forecast.
…………… Yo F I Tt = Ft +Tt
F
= F I7 t +c (D - F IT t)
t+1
Tt+1 = Tt- ^ (F t+1 - FI T t)

Where:
α and β: smoothing constants
FIT t :Forecast including trend for period t
F t+1 : Exponentially smoothed forecast for period t+1
T t+1 : Exponentially smoothed trend for period t+1 D t : Real demand for period t.
 Example:
Forecast the annual demand for certain household appliances.
Assume an initial smoothed forecast of 840 units, a bias of 10 ( a
= 0.3 and β=0.8). Calculate the MAD and BIAS of the made
forecast.
Real Smoothed Smoothed Forecast
including Error (E t =
Years. demand ( Dt forecast (F trend (T t
trend (FIT D t -FIT t )
) t ) )
t )
2007 845
2008 920
2009 988
2010 1172
2011 1145
2012 1174
2013 1475
2014
4
 Solution:
Real Smoothed
Smoothed Forecast
Years. demand ( forecast (F t Error ( Et )
trend (T t ) including
)
Dt ) trend (FIT t )
2007 845 840 10 850 -5
2008 920 848,50 8,80 857,30 62,70
2009 988 876,11 23,85 899,96 88,04
2010 1172 926,37 44,98 971,35 200,65
2011 1145 1031,54 93,13 1124,68 20,32
2012 1174 1130,77 98,01 1228,79 -54,79
2013 1475 1212,35 84,86 1297,21 177,79
2014 1350,55 127,53 1478,08
R SFE = ^Et = 489,71
BIAS RSFE 489.71
= 69,96 \Et = 609,29
E
n7
n

XE t I
=1
MAD = 2
7 4
3
609.29 = 87,04
 Linear regression method:
Causal method in which a variable (known as the dependent
variable) is related to an independent variable through a linear
equation of the form:
Y = a + bx
When the independent variable is time periods, it is considered a
time series method, in any other case it is considered a causal
method.

Where:
Y: dependent variable
X: independent variable
a: intersection of the line with the Y axis
b: slope of the line

4
The objective is to find the values of a and b
that minimize the sum of the deviations
quadratics of the actual data points that
are represented in the time series

4
 Linear regression method:
Y=a+bx
where

. = Intersection of the line with the Y axis b =

Slope of the line
a = y — bx
y = Average of all y
x = Average of all x n2XY-(2X)(2Y)
x = x value of each data point y = y value of
n}x2—(x)2
each data point

n = Data point number

Y = Value of the dependent variable calculated with the regression equation
Example:
Forecast the annual demand for certain household
appliances. Calculate the MAD and BIAS of the made

4
forecast.
Real demand ( Forecast (
Years.
Dt ) Ft ) Error ( Et )
2007 845
2008 920
2009 988
2010 1172
2011 1145
2012 1174
2013 1475
2014
First, the linear regression equation must be obtained
Y = a + b^
This equation will allow us to determine the demand forecast Y
when the independent variable X (the number of year, month or

4
 other causal variable) takes values in the domain for which it is
Solution 1
In this case the variable
independent X is the
defined.
period number (year)
but work with
numbers 2007…2013 turns
out
in cumbersome
Years. x Demand (Y) XY 2 x

To determine the2007
calculations. By
that is assigned
equation-3we must find 845the parameters
-2535 a and9b
through the following
conveniently values
2008 mathematical
-2 formulas:
920 -1840 4
smaller than
simplify the calculation
2009 -1 988 -988 1
2010
n 2 XY -( 2 ^)( 2Y0) 1172 0 0
n22011
X 2 -(^) 2 1 1145 1145 1
We replace the
To table
carrysumsout these 2012
calculations,
2 it is convenient
1174 to 2348 4
use a table like
the one presented
in the formulas and
we get a and b
below: 3
2013 1475 4425 9
ADDITI 0 7719 2555 28
Then we substitute the values of
b= - =91.25 a and b in the equation Y=a + bx y
we get the regression equation
= 91,25
nx2 - linear

(2x)2
a = y-bx = 1102.71 Y = 1102.71+ 91.25X

4
 Solution 1
Real demand ( Forecast (
Years.
Dt ) Ft ) Error (Et)
2007 845 828,96 16,04
2008 920 920,21 -0,21
2009 988 1011,46 -23,46
2010 1172 1102,71 69,29
2011 1145 1193,96 -48,96
2012 1174 1285,21 -111,21
2013 1475 1376,46 98,54
2014 1467,71
Once the linear regression equation is obtained, we can obtain the forecasts for the different months by giving X the value of each period.
R SFE = YEt = 0,03
Thus, for the year 2007 the forecast according to the regression equation would be:
Y=a+b·X
\Et = 367,71
F = 1102.71+ 91.25·(-3) F
2007 2007 = 828.96 E
BIAS
n RSFE = 0.03 = 0.004
XI E t I
n 7
t =1

367.71
MAD = = 52.53

4
5
If we had assigned other values to 1 and so on. As can be seen, the regression equation has changed, but despite this the forecast for each
year remains unchanged. This is demonstrated by the calculation of the forecast for the year 2007:
Y=a+b·X
F 2007 = 828.96+91.25·(0) F 2007 = 828.96 Solution 2
x Demand (Y) XY x 2
2007 0 845 0 0
2008 1 920 920 1
2009 2 988 1976 4
2010 3 1172 3516 9
2011 4 1145 4580 16
2012 5 1174 5870 25
2013 6 1475 8850 36
ADDITI 21 7719 25712 91

b= - =91.25 a = y-bx = 828.96

= 91,25
nx2 -
(2x)2
Y = 828.96+ 91.25X
Equation of
linear
7 regression
5
1
 Solution 2
Real demand ( Forecast (
Years.
Dt ) Ft ) Error (Et)
2007 845 828,96 16,04
2008 920 920,21 -0,21
2009 988 1011,46 -23,46
2010 1172 1102,71 69,29
2011 1145 1193,96 -48,96
2012 1174 1285,21 -111,21
2013 1475 1376,46 98,54
2014 1467,71
R SFE =LE t = 0,03
RSFE 0.03
BIAS
n7 0.004 LEL t = 367,71
The BIAS and the
n
IF E t I MAD
t =1 367.71 they also remain
MAD = = 52.53
unalterable

5
 Solution 3:
Years. x Demand (Y) XY x 2

2007 2007 845 1695915 4028049

2008 2008 920 1847360 4032064
2009 2009 988 1984892 4036081
2010 2010 1172 2355720 4040100
2011 2011 1145 2302595 4044121
2012 2012 1174 2362088 4048144
2013 2013 1475 2969175 4052169
ADDITI 14070 7719 15517745 28280728

Yob = = 91,25 a = y-bx = -182309.786

nSx 2 - (2x)2
Equation of
Y = -182309.786+ 91.25X
7
linear regression J
5
3
 Solution 3:
Real demand ( Forecast (
Years.
Dt ) Ft ) Error (Et)
2007 845 828,96 16,04
2008 920 920,21 -0,21
2009 988 1011,46 -23,46
2010 1172 1102,71 69,29
2011 1145 1193,96 -48,96
2012 1174 1285,21 -111,21
2013 1475 1376,46 98,54
2014 1467,71
R SFE =LE t = 0,03
RSFE 0.03
n7 0.004 LEL t = 367,71

MAD =

BIAS n n XI Et I n - 1 t
5
 CONCLUSION:
By analyzing the mean absolute deviations (MAD) for each forecast
we can conclude that the linear regression method provides the MAD
and the BIAS closest to zero , so we can consider that this method for
the practical case under study is the most reliable to make demand
forecasts according to the behavior of historical series.

Forecast Method MAD BIAS

Empirical 114 105
moving average 190 189,08
Weighted moving average 173 172,25
Smoothing exp. simple 223 222,63
Smoothing exp. double 101 81,62
Simple linear regression 52 0,004
 Exercise
The following table shows the actual sales of upholstered armchairs
made by a furniture manufacturer and the corresponding forecasts.
contributions to each of the last eight months.
Calculate the RSFE, MSE, σ, MAD, and MAPE for this product.
Month Demand ( Dt
Forecast ( Ft )
(t) )
1 200 225
2 240 220
3 300 285
4 270 290
5 230 250
6 260 240
7 210 250
8 275 240

5
 Solution:
Quadratic Absolute Absolute
Mont Demand ( Forecast ( Error (
error (E t error |E t percentage
h (t) Dt ) Ft ) Et ) error (|E t
) 2
|
|/Dt)*100

1 200 225 -25 625 25 12,50%

2 240 220 20 400 20 8,3
3 300 285 15 225 15 5
4 270 290 -20 400 20 7,4
5 230 250 -20 400 20 8,7
6 260 240 20 400 20 7,7
7 210 250 -40 1600 40 19
8 275 240 35 1225 35 12,7
TOTAL -15 5275 195 81,3%

5
 Conclusions:
Cumulative forecast error (bias) RSFE = ∑Et=-15
R SFE _________
S
Average forecast error (average bias) BIAS = ^
— = -1.875
2
_ EEt 5275
Mean square error M SE =--------= — = 659,4
n 8 ,
•------------------------
Standard deviation or = 2 [ E t-(- 1 , 875 )]
2
, TO
Absolute mean deviation MAD = LlE
Í1 =195 = 24.4 n 8 ,

Absolute mean percentage error z. 1

E
D^ 00

An RSFE of -15 indicates that the forecast has a slight tendency to overestimate demand. The MSE,
σ, and MAD statistics provide measures of forecast error variability. A MAD of 24.4 means that the
average forecast error was 24.4 units in absolute value. The value of σ=27.4 indicates that the
distribution of forecast errors within the sample has a standard deviation of 27.4 units. A MAPE of
10.2% implies that, on average, the forecast error was plus or minus 10% of the actual observed
demand. These measures become more reliable as the number of data periods increases. 55
 Proposed exercise:
2. The sales made by Dalworth Company in the last
year have been: Month Demand
(millions USD)
a) Forecast sales for the months between May and December
Jan. 20
using the four-month moving average method.
b) Apply a three-month weighted moving average to forecast Feb. 24
sales between April and December (use weights of 3/6, Sea. 27
2/6, and 1/6).
c) Use exponential smoothing with α=0.6 to forecast sales for Apr. 31
the months April through December. Suppose the initial May. 37
forecast for January was 22 million. Jun. 47
d) Compare the performance of the three methods using the Jul. 53
MAD as the performance criterion. Which method would Aug. 62
you recommend? Sep. 54
e) Compare the performance of the three methods using Oct. 36
MAPE as a performance criterion. Which method would
you recommend?
Nov. 32
f) Compare the performance of the three methods using the Dec. 29
MSE as the performance criterion. Which method would you recommend?

5
 a) Avg. Mobile
Month Dt F
t and
t

Jan. 20
Feb. 24
Sea. 27
Apr. 31
May. 37 25,5 11,5
Jun. 47 29,75 17,25
Jul. 53 35,5 17,5
Aug. 62 42 20
Sep. 54 49,75 4,25
Oct. 36 54 -18
Nov. 32 51,25 - 19,25
Dec. 29 46 -17
MAD 15,59

MSE 267,21

MAP 38,71

6
 b) Avg. c) Exponential
Weighted smoothing
F
t Mobile t and F
t and
t
22 -2
20,8 3,2
22,72 4,28
24,83 6,17 25,29 5,71
28,50 8,50 28,72 8,28
33,33 13,67 33,69 13,31
41,00 12,00 41,67 11,33
48,33 13,67 48,47 13,53
56,50 -2,50 56,59 -2,59
56,50 -20,50 55,04 -19,04
46,33 -14,33 43,61 -11,61
37,00 -8,00 36,65 -7,65
11,04 9,13
147,09 98,73
27,84 22,65

6
 Proposed exercise:
1. The monthly demand for units manufactured by ACME Rocket Company
has been the Month Demand Month Demand
following:
May. 100 Sep. 105
Jun. 80 Oct. 110
Jul. 110 Nov. 125
Aug. 115 Dec. 120

a) Apply the exponential smoothing method to forecast the number of units

from June to January. The initial forecast for May was 105 units; α=0.2.
b) Calculate the absolute percentage error corresponding to each of the
months from June to December, and also the MAD and MAPE of the
forecast error at the end of December.

6
 DEMAND FORECAST

Causal Models
Causal models:
They are a type of quantitative method that uses historical
data on independent variables, such as promotional
campaigns, economic conditions, and competitors'
activities to forecast demand.
Simple regression
model..... . .
Multiple regression model
• Econometric model
• Purchase intention survey method
• Input-output model (technical coefficients method)

Causal models: Linear regression

Example 1:
The following data relates the sales figures of a bar in a small Hotel, with
the number of guests over 18 years of age registered during the last

6
 month: 2x
Sales, and Guests,x xy
week Guests bar sales

330 16 256 5.280

1 16 $330
270 12 144 3.240
2 12 270
380 18 324 6.840
3 18 380
300 4 14
14 196 300 4.200

y = 1,280 2X = 60 x2
= 920 z
xy =19,560
The marketing manager J. knows from experience that the bar's daily sales
level is strongly_ related
2X
to the number_ ofzyguests
=
staying at the hotel. If a
total of 20 adultV—guests = 60have already confirmed 1.280
= 15 their= 320
reservation for next
week, how much are
X = n
the sales
4 expected
y =to
n be? 4

xy - n xy
= 19.560 – (4) (15) (320)
2
= 18
920- (4) ( 15 ) 20

a = y - bx = 320 – 18(15) = 50
The estimated regression equation is therefore

y = 50 + 18 x or Sales = 50 + 18 * Qty. of guests 62

6
 If the forecast is 20 guests next week, how much are sales expected to be?

y = 50 + 18 x

Sales = 50 + 18 (20)
= 410

Simple Linear Regression Line

C
0 400
or 350
O
300 Historical lawsuit
(
250
200
150
100
50 •

4 8 12 16 20

6
Guests

6
 Correlation coefficient (r) How good is
the relationship of the historical data of the independent variable
with respect to the behavior of the dependent variable?
In these cases, the unknown is usually solved by determining the
correlation coefficient, which is nothing more than an index that
measures the direction and strength of the relationship between
the independent variable and the dependent variable .

n 2 xy - 2x2y
n2x n 2 and 2 -
2 2 and
The value of r varies between 1 and -1
The value of the correlation index varies in the interval [-1,1]:

• If r approaches 1 , then there is a large direct relationship of the data, in other

6
 words that a large part of the behavior of the dependent variable is explained by
the behavior of the independent variable (If r = 1, there is a perfect positive
correlation )

• If r approaches -1, an inverse relationship occurs (decreases in the independent

variable will be accompanied by increases in the dependent variable and vice versa).

• If r approaches zero on both sides , we can conclude that there is no linear

relationship. But this does not necessarily imply that the variables are independent:
there may be some type of non-linear relationship between the variables.

• The closer r is to ±1, the better the fit of the regression line to the points on the
graph.

7
Correlation coefficient (r)

7
1
 Correlation coefficient (r)
Determination coefficient (r²)
It is equal to the square of the correlation coefficient and
takes values in the interval [0;1]
It measures the amount of variation that the dependent
variable presents with respect to its mean value, which is
explained by the regression line.
It is desirable that r 2 approaches 1, which means that the
variations of the dependent variable and the forecast
generated by the regression equation are closely related.

7
2
Continuing with the hotel example, calculate the correlation coefficient:

2x
Sales (and) Guests (x) xy 2 and

108.900
330 16 256 5.280
270 12 144 3.240 72.900
380 18 324 6.840 144.400
90.000
300 14 196 4.200

I y = 1.280 2x = 60 2 x2
2 and 2 =
= 920 2xy =19,560 416,200

(4) (19.560) - (60) (1.280)

(4) (920)- (60)2 (4) (416.200)- (1.280)2

1.440 1.440
0,993619798

2.112.000 1453,27217
Positive correlation r= 0< r <1
7
 Example 2:
An operations manager needs to determine the sales forecast for a type of
bronze door hinge. To do this, ask the Marketing department for information
on the advertising budget for this product during the previous five months.
Knowing that in June $1750 will be spent on advertising for this product:
a) Apply linear regression to develop a sales forecast for June.
b) Do you think it is correct to select advertising expenses as a causal variable
for demand?
Sales Advertising
Month
(Thousands of units) (thousands of USD)
Jan. 264 2,5
Feb. 116 1,3
Sea. 165 1,4
Apr. 101 1,0
May. 209 2,0

7
 Solution:
The regression equation is: Y= - 8,135 + 109,229X r=0.98
r2
=0.96
Since the value of the correlation coefficient is very close to 1 (r=0.98), it is concluded that
there is a strong positive relationship between sales and advertising expenses, and that the
choice was correct.
A value of r2=0.96 implies that 96% of the variation observed in sales is explained by
advertising expenses. In practice, most relationships between advertising and sales are not
that strong because often other variables, such as the general economic situation and
competitors' strategies, combine to affect sales.

Since the advertising expense will be $1750, the forecast for June is:

Y= - 8.135 + 109.229*(1.75)
Y= 183,016 units
Software download OM explorer: Excel plug-in
http://wps.pearsoned.co.uk/
ema_ge_krajewski_opsmgmt_9/140/35881/9185615.cw/i ndex.html

7
 Example 3:
Chicken Palace periodically offers meals that include five pieces of
chicken to take home, at special prices. Based on the historical
observations and calculations represented in the following table,
determine the regression equation, the correlation coefficient, and
the coefficient of determination. How many meals does Chicken
Palace think it will sell for $3.00 each?

Observation Price Meals sold

1 $2,70 760
2 $3,50 510
3 $2,00 980
4 $4,20 250
5 $3,10 320
6 $4,05 480

7
 Solution:
The regression equation is: Y= 1454.60 – 277.63

r = - 0.84
r2
= 0.71

The correlation coefficient shows a negative correlation between

the variables. The coefficient of determination is relatively small,
indicating that other variables (besides price) significantly affect
sales.
If the manager finds the regression equation satisfactory, the
estimated sales at a price of $3.00 per meal would be:

Y= 1454.60 – 277.63 (3.00)

Y= 621.71 ≈ 622 meals

7
 Example 4:
Ohio Milk Products manufactures and distributes ice cream in Ohio, Kentucky and
West Virginia. The company wants to expand its operations by opening another
plant in northern Ohio. The size of the new plant will be calculated based on the
expected demand for ice cream in the area served by said plant. A market study is
currently being conducted to determine what that demand will be.
Ohio Swiss wants to estimate the relationship between the manufacturing cost per
gallon and the number of gallons sold in a year to determine the demand for ice
cream and therefore the size of the new plant. The following data has been
collected:
a) Develop a regression equation for Plant Cost per Thousands of
thousand
of gallons gallons sold
decide the size of the plant, depending on the
cost per gallon for any location. 1 1015 416,9
b) Calculate the correlation coefficient and the 2 973 472,5
determination coefficient. Comment on your 3 1046 250
regression equation in light of these 4 1006 372,1
c) Suppose that the financial study estimated a 5 1058 238,1
cost of $1005 per thousand gallons of ice cream 6 1068 258,6
produced in the new plant. Under this criterion 7 967 597
How much should the new 8 997 414
9 1044 263,2
plant?
10 1008 372 73
 Solution:
The regression equation is: Y= 3573 – 3.15

r = - 0.94
r2
= 0.89
The correlation coefficient shows a strong negative correlation
between the variables.
The coefficient of determination indicates that 89% of the variation
in demand is explained by the total cost of production, although it
appears that said variation depends to a lesser extent on other
variables not identified in the study. According to the regression
line, the plant should produce 407,000 gallons of ice cream at a
cost of $1.005 per gallon.

Y= 3573 – 3.15 (1005)

Y= 4 07.02 ≈ 407 thousand gallons
74
 Example 5:
EXAMPLE 15.5: Forecasting Using a Causal Relationship
Carpet City Store in Carpenteria keeps annual records of its sales (in square yards), as well as the number of new home
licenses in this area.

NUMBER of CONSTRUCTION SALES

YEAR LICENSES (SQUARE YARDS)

1999 18 13 000
2000 15 12 000
2001 12 11000
2002 10 10 000
2003 20 14 000
2004 28 16 000

2005 35 19 000
2006 30 17 000
2007 20 13 000

Carpet City's operations manager believes that it is possible to forecast sales if the start of housing projects for the year
is known.
Determine the forecast for 2008 based on the linear regression method. Consider that 25 construction
licenses were granted this year.

8
 BIBLIOGRAPHY:
1. CHASE, JACOBS, AQUILANO. Operative administration.
McGraw Hill. Mexico, 2014. pp. 486512

8
 Proposed exercise
16, Assume that your inventory of merchandise for sale is maintained on the basis of forecast demand. If the
distributor's sales staff calls on the first day of each month, calculate your sales forecast using the three
methods requested here.
REAL

DEMAND FORECAST 5
Delphi method 6
Consult the sales force 7
consensus groups 8
Historical analogies 9
Market research 10
TIME SERIES ANALYSIS 13
Demand patterns 14
Forecast errors 20
Forecast error 21
Forecast errors 22
Forecast errors 23
Forecast errors 24
Empirical forecast: 25

8
 Example: 26
Solution: 27
Simple moving average: 28
Example: 29
Solution 31
Weighted moving average: 32
Example: 32
Solution 34
Simple exponential smoothing: 35
Example: 36
Solution 38
Linear regression method: 44
Linear regression method: 46
Example: 46
Solution 1 49
Solution 1 50
Solution 2 51
Solution 2 52
Solution 3: 53
Solution 3: 55
Exercise 58
Solution: 59
Proposed exercise: 61
Proposed exercise: 63
Causal models: 64

8
 Causal models: Linear regression 65
Determination coefficient (r²) 72
Example 2: 74
Solution: 74
Example 3: 75
Solution: 76
Example 4: 78
Solution: 79
Example 5: 80
BIBLIOGRAPHY: 81
Proposed exercise 82
Self-study exercises: 85
Self-study exercises: 88
Self-study exercises: pg 157s 89
Self-study exercises: 90
Self-study exercises: 91
Self-study exercises: 91
Self-study exercises: 92
Self-study exercises: 93
Self-study exercises: 95
Self-study exercises: 95
Self-study exercises: 96

a) Using a three-month simple moving average, what is the forecast for September?

8
 b) Using a weighted moving average, what is the forecast for September with relative values of 0.20, 0.30,
and 0.50 for June, July, and August, respectively?
c) Using simple exponential smoothing, and assuming that the June forecast was 130, forecast September
sales with a smoothing constant a of 0.30.

Self-study exercises:
17. The historical demand for a product is as follows:
DEMAND

DEMAND FORECAST...................................................................................................................................................................................5
Delphi method..................................................................................................................................................................................................6
Consult the sales force......................................................................................................................................................................................7
consensus groups..............................................................................................................................................................................................8
Historical analogies..........................................................................................................................................................................................9
Market research..............................................................................................................................................................................................10
TIME SERIES ANALYSIS...........................................................................................................................................................................13
Demand patterns.............................................................................................................................................................................................14
Forecast errors................................................................................................................................................................................................20
Forecast error..................................................................................................................................................................................................21
Forecast errors................................................................................................................................................................................................22
Forecast errors................................................................................................................................................................................................23

8
 Forecast errors................................................................................................................................................................................................24
Empirical forecast:.........................................................................................................................................................................................25
Example:.........................................................................................................................................................................................................26
Solution:.........................................................................................................................................................................................................27
Simple moving average:.................................................................................................................................................................................28
Example:.........................................................................................................................................................................................................29
Solution..........................................................................................................................................................................................................31
Weighted moving average:.............................................................................................................................................................................32
Example:.........................................................................................................................................................................................................32
Solution..........................................................................................................................................................................................................34
Simple exponential smoothing:......................................................................................................................................................................35
Example:.........................................................................................................................................................................................................36
Solution..........................................................................................................................................................................................................38
Linear regression method:..............................................................................................................................................................................44
Linear regression method:..............................................................................................................................................................................46
Example:.........................................................................................................................................................................................................46
Solution 1.......................................................................................................................................................................................................49
Solution 1.......................................................................................................................................................................................................50
Solution 2.......................................................................................................................................................................................................51
Solution 2.......................................................................................................................................................................................................52
Solution 3:......................................................................................................................................................................................................53
Solution 3:......................................................................................................................................................................................................55
Exercise..................................................................................................................................................................................................................58
Solution:.................................................................................................................................................................................................................59
Proposed exercise:..........................................................................................................................................................................................61

8
 Proposed exercise:..........................................................................................................................................................................................63
Causal models:...............................................................................................................................................................................................64
Causal models: Linear regression...................................................................................................................................................................65
Determination coefficient (r²).........................................................................................................................................................................72
Example 2:.........................................................................................................................................................................................................74
Solution:.............................................................................................................................................................................................................74
Example 3:.........................................................................................................................................................................................................75
Solution:.............................................................................................................................................................................................................76
Example 4:.........................................................................................................................................................................................................78
Solution:.............................................................................................................................................................................................................79
Example 5:......................................................................................................................................................................................................80
BIBLIOGRAPHY:.........................................................................................................................................................................................81
Proposed exercise...........................................................................................................................................................................................82
Self-study exercises:.......................................................................................................................................................................................85
Self-study exercises:.......................................................................................................................................................................................88
Self-study exercises: pg 157s.........................................................................................................................................................................89
Self-study exercises:.......................................................................................................................................................................................90
Self-study exercises:.......................................................................................................................................................................................91
Self-study exercises:.......................................................................................................................................................................................91
Self-study exercises:.......................................................................................................................................................................................92
Self-study exercises:.......................................................................................................................................................................................93
Self-study exercises:.......................................................................................................................................................................................95
Self-study exercises:.......................................................................................................................................................................................95
Self-study exercises:.......................................................................................................................................................................................96

8
 a) Using a four-month simple moving average, calculate a forecast for October.
b) Using simple exponential smoothing with a = 0.2 and a forecast for September = 65, compute a forecast
for October.
c) Using simple linear regression, calculate the trend line for the historical data. On the x-axis, let April = L
May = 2, etc., while on the y-axis is demand.
d) Calculate a forecast for October.

Self-study exercises:
20, Your manager is trying to determine which forecasting method to use. Based on the following historical
data, calculate the following forecast and specify which procedure you would use.

MONTH ACTUAL DEMAND MONTH ACTUAL DEMAND

1 62 7 76
2 65 8 78
3 67 9 78
4 68 10 80
5 71 11 84
6 73 12 85

a) Calculate a three-month simple moving average forecast for periods 4 through 12.
b) Calculate the three-month weighted moving average with weights of 0.50, 0.30, and 0.20 for periods 4

8
 to 12.
c) Compute a simple exponential smoothing forecast for periods 2 through 12 using a pro initial forecast
(F,) of 61 and an a of 0.30.
d) Calculate the mean absolute deviation (MAD) of the forecasts made with each technique in periods 4 to
12. Which forecasting method do you prefer?

Self-study exercises: pg 157s

I. Demand for stereo headphones and CD players for joggers has led Nina Industries to grow almost 50% in
the past year. The number of joggers continues to increase, so Nina expects demand to increase as well,
because, so far, there have been no safety laws that prevent joggers from using headphones. The demand
for this inmates from last year was the following:

DEMAND (UNITS) DEMAND (UNITS)

MONTH MONTH

January 4200 July 5 300

February 4 300 August 4900
March 4 000 September 5400
April 4400 October 5700
May 5 000 November 6300
June 4700 December 6 000

8
 a) Using a least squares regression analysis, what would you estimate the demand to be for each month of
the coming year?

Self-study exercises:
II. The current tabulated demand for an item over a period of nine months (January to September) is given
below. Your supervisor wants to try two testing methods to see which one worked best over the period.

MONTH REAL MONTH REAL

January 110 June 180

February 130 July 140

March 150 August 130

April 170 September 140

May 160

a) Forecast April to September with a three-month moving average.

b) Using simple exponential smoothing with an alias of 0.3, calculate from April to September, b) Use the
M AD to decide which method produced the best forecast in the six-month period

9
 Self-study exercises:
3, The following tabulations are actual unit sales for six months and an initial forecast for January.
a) Compute the forecasts for the remaining five months with simple exponential smoothing with a = 0.2.
b) Calculate the MAD of the forecasts.

REAL FORECAST
100 80
January 94
February 106
March 80
April May June 68
94

Self-study exercises:
9. Not all items in your stationery store are evenly distributed as far as demand is concerned, so you decide to

9
 forecast demand to plan your assortment. The past data from usual account books, for the month of August,
are as follows:

Weekly 300 Week 3 600

Week 2 400 Week 4 700

a) Using a three-week moving average, what would be your forecast for the coming week?
b) With exponential smoothing with a = 0.20. If the exponential forecast for week 3 was calculated as the
average of the first two weeks [(300 + 400)/2 = 350], what would be your forecast for week 5?

Self-study exercises:

EXAMPLE 15.2: Least squares method

The sales of a product line in a company during the 12 quarters of the last 3 years are as follows:

QUARTER SALES QUARTER SALES

1 600 7 2 600

2 1550 8 2 900

9
 3 1 500 9 3 800

4 1500 10 4 500

5 2 400 11 4 000

either 3100 12 4 900

The company wants to forecast each quarter of the fourth year; that is, the 13th, 14th, 15th and 16th quarters

Self-study exercises:

21. Perform a regression analysis on demand without seasonal factors to forecast demand in the summer of 2008, given
the following historical demand data.

YEAR SEASON REAL DEMAND

2006 Spring 205
Summer 140
Autumn 375
Winter 575
2007 Spring 475
Summer 275

9
Autumn 685
Winter 965

9
 Self-study exercises:

23, The actual demand for a product in the previous three months was:

Three months ago 400 units

Two months ago 350 units
Last month 325 units

t Using a three-month simple moving average, make a forecast for this month.
o If this month the actual demand was 300 units, what would be your forecast for next month?
) Using simple exponential smoothing, what would be your forecast for this month if the uniform forecast me
b exponential three months ago was 450 units and the uniformity constant was 0.20?

Self-study exercises:

26. Below are the sales revenues for a large utility company from 1997 to 2007. Forecast revenue from 2008 to 2011.

9
 Use your best judgment, intuition, or common sense as to which model or method to use, as well as what period of
data to include.

REVENUE (MILLIONS) REVENUE (MILLIONS)

1997 S4 865.9 2003 $5 094.4

1998 5 067.4 2004 5108.8

1999 5515.6 2005 5550.6

2000 5728.8 2006 5 738.9

2001 5 497.7 2007 5860.0

2002 5197.7

Self-study exercises:
27. Mark Price, new production manager for Speakers and Company, has to figure out which variable most affects
demand for his line of stereo speakers. You are not sure whether the unit price of the product or the effects of further
marketing are the main drivers of sales and you want to apply a regression analysis to find out which factor drives
the demand for your market the most. do. The relevant information was collected in an extensive marketing project
spanning the last 10 years and resulted in the following data:

SALES/UNIT ADVERTISING (THOUSANDS

(THOUSANDS) OF DOLLARS)
YEAR UNIT PRICE

1996 400 280 600

1997 700 215 835

9
 1998 900 211 1100
1999 1300 210 1400
2000 1150 215 1200
2001 1200 200 1300
2002 900 225 900
2003 1100 207 1100
2004 980 220 700
2005 1234 211 900
2006 925 227 700
2007 800 245 690

a) Perform a regression analysis in Excel® based on this data. Using your results, answer the following questions.
b) Which variable, price or advertising, has a greater effect on sales and how do you know?
c) Predict Speakers and Company's average annual speaker sales based on the regression results, if the price was
$300 per unit and the amount spent on advertising. city (in thousands) was $900.