QUALITATIVE FORECASTING METHODS Qualitative forecasting methods are subjective

QUALITATIVE FORECASTING METHODS

Qualitative forecasting methods are subjective, based on the executive opinions, sales staff
opinions, consumer surveys and survey of the experts. They are appropriate when past data are
unavailable. Consumer surveys, sales staff opinions and executive opinion can be used to obtain
quick estimates, while survey of experts, which is also known as Delphi method is used for long
range predictions.

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

Situations where qualitative forecasting method is often used:
• New product idea generation and development.
• Strengthens and weaknesses of products or brands.
• Understanding motive force of purchase decision dynamics.
• Studying emotions and attitudes on societal and public affairs issues.

FORECASTING TECHNIQUES
QUANTITATIVE METHODS
TIME SERIES
ANALYSIS
SMOOTHING
TECHNIQUES
*Moving average method
*Weighted moving average
method
*Exponential smoothing
method
TREND
ANALYSIS
*Graphic method
*Linear regression
*Non -linear regression
CAUSAL
APPROACH
*Regression analysis
QUALITATIVE METHODS
*Consumer Surveys
*Executive Opinions
*Sales Staff Opinions
*Delphi Method

C O N S U M E R S U R V E Y S
Some companies conduct their own market surveys regarding specific consumer purchases.
Surveys may consist of telephone contacts, personal interviews, or questionnaire as a means of
obtaining data. In this technique, either the sampling method or the census method can be used to
survey the customers. The expected sales can be calculated as follows:
ES = (nR/nS) (N.AS)
ES = Expected Sales
nR = No. of households reporting demand for the product
nS = No. of households surveyed
AS = Average sales per demand reporting households
N = T otal number of households

E X E C U T I V E O P I N I O N S
The forecasting is done quickly and easily, without need of elaborate statistics. A team of high
level managers meet and collectively build a forecast. But, there is a drawback in this method.
The opinion of on e member might prevail and because of this there are less likely chances to get
a good forecast.

S A L E S S T A F F O PI N I O N S
In this technique, forecasts are developed by the opinion of the sales force as these people who
have continued contacts with customers. But sometimes the sales force may not understand the
plans of customers exactly which leads to wrong estimation.

D E L PH I M E T H O D
Delphi method was fi rst developed and used by RAND C orporation, California in the 1950s,
during the Cold War. This is a group te chnique in which a panel of experts is questioned
individually about their perceptions of future events. The experts do not meet as a group, in order
to reduce the possibility that unanimity is reached because of most influential personality factors.
Hence the forecasts and other arguments are collected and summarized by another group and
returned to the experts along with further questions and opinions. This until decision is reached.

QUANTITATIVE FORECASTING METHODS

Quantitative forecasting models are used to forecast future data as a function of past data. The
selection of a particular quantitative method depends on whether we are dealing with time series
or cross -sectional data. These techniques are usually applied to short or intermediate range
deci sions.
Forecasting with cross -sectional data
The dependant variable would be forecasted by using the data on its drivers, while dealing with
cross -sectional data. The drivers can be different variables. Firstly, using past data for all
variables, the relationship is estimated. Following that, the estimated model is used to predict the
future values of the dependent variable by putting the future values of the independent variables.
Forecasting with time series data
In time serie s technique, the forecast is based on the past values of the variable under
consideration. It is calculated from known facts for future under some assumption that the
prevailing pattern in the data is going to continue in the future also. Now we can study the details
of time series data and its components.
TIME SERIES DATA
A time series may be defined as a set of data points of a variable collected and recorded in
chronological order, of the time periods at uniform intervals. According to Morris Hamburg “A
time series is a set of statistical observations arranged in chronological order” . The collection of
data points can be done every hour, day, week, month or year or 10 years etc at particular
intervals. Example for time series data:

YEAR

SALES ($m)
2001 15
2002 18
2003 20
2004 16
2005 22
2006 23
2007 25
2008 17

2009 20
2010 22
2011 24
2012 25
2013 27
2014 29
2015 30
Table 1: Sales of LCD TVs
If we observe time series, it is considered to have the following four components.
• Trend effect(T)
• Cyclical variations(C)
• Seasonal variations(S)
• Random or irregular variations(I)
TREND EFFECT
A trend is the changes that have occurred as a result of the upward or downward movement of
time series data over a long period of time.
CYCLICAL V ARIATIONS
These long term and recurrent variations in many economic and business series have taken place
because of the fou r phases in the business cycle, namely (a) prosperity (b) recession (c)
depression and (d) recovery.
SEASONAL VARIATIONS
The changes that have taken place within a year as a result of change in a season which is a
period or part of one year.
RANDOM OR IRREGULAR VARIATIONS
The changes that have taken place as a result of unpredicted forces like floods, earthquakes etc .
CAUSAL FORECAST
The aim of causal forecasting method is to make the best statistical relationship between a
dependent variable and one or more independent variable. The most common model of causal
approach in forecast used in practice is regression analysis. In causal fore casting methods, when
someone tries to predict a dependent variable using a single independent variable it is known as
simple regression method and while using more than one independent variable it is known as
multiple regression method.

FORECAST ACCURACY
The relevance or importance of a technique for forecasting a time series data can be evaluated by
looking at the forecast accuracy of that technique. It can be defined as follows:
Forecast Error = Actual Value – Forecast
The forecast errors are analogous t o the residuals in regression analysis. In regression analysis, a
residual is the difference between the observed values of the dependent variable and the
estimated value of the dependent variable.
The commonly used measures for forecasting errors as follo ws:
• Mean Absolute Deviation ( MAD ), is the average absolute error measure,

MAD =??A t – Ft?/ n

• Mean Squared Error ( MSE ), is the average squared errors,

MSE = ? (A t – Ft)2 / n

• Mean Absolute Percent Error ( MASPE ), is the average absolute percent error measure,
MAPE = ( (?A t – Ft?/ A t) × 100 )/ n
Where, At is the actual value and Ft is the forecast, and value of t varies from 1 to n.
In case of determining which technique is giving better forecasting results these measures can be
considered. A forecasting technique providing lowest values of MAPE, MSE and MAD can be
finally selected for forecasting.
Time series analysis can be classified into Smoothing techniques and Trend analysis.
SSMM OOOOTTHHIINNGG TTEECCHHNNIIQQUUEESS
When time series data do not show real and significant trend , smoothing techniques are used, as
given below.
• Moving Average Method
If the time series data shows a constant trend or horizontal pattern, moving average method may
be used. This method is based on the average of the most recent k data values as the forecast for
next period in the time series data. Moving average forecast of order k is defined by,
Ft = ? A t -1 / k = (A t-k + . . . . . + A t-2 + A t-1) / k

Where, F t = Forecast for time period t
At-1 = Actual value in period (t -i); i= 1,2,3, . . . . ,k.
K = No: of periods in the moving average
In this method, every time , the oldest data point is replaced by a new observation, while
computing the new average and this way the average measure keeps moving over time period.
To illustrate this method, let us take data on yearly sales of LED TVs as shown in Table 1. The
calculations for moving average with k=3 are shown in table 2 and the calculations with k=5 are
presented in Table 3.
Table 2: Calculations for Moving average when k = 3
Year Sales Forecast,
k=3
Forecast
Error
Absolute
Error
Squared
Error
Percentage
Error
1 15
2 18
3 20
4 16 17.67 -1.67 1.67 2.79 .10
5 22 18.00 4.00 4.00 16.00 .18
6 23 19.33 3.67 3.67 13.47 .16
7 25 20.33 4.67 4.67 21.81 .19
8 17 23.33 -6.33 6.33 40.07 .37
9 20 21.67 -1.67 1.67 2.79 .08
10 22 20.67 1.33 1.33 1.77 .07
11 24 19.67 4.33 4.33 18.75 .18
12 25 22.00 3.00 3.00 9.00 .12
13 27 23.67 3.33 3.33 11.09 .12
14 29 25.33 3.67 3.67 13.47 .13
15 30 27.00 3.00 3.00 9.00 .10
40.67 160.00 1.80

MAD = (? ?A t – Ft?) / n
=40.67/12
=3.39
MSE= (?(A t – Ft)2) / n
=160/12
=13.33

MAPE = ( (?A t – Ft?/ A t) × 100 )/ n
=(1.80/12)×100
=15
Table 3: Calculations for Moving Average when k=5
Year Sales Forecast,
k=5
Forecast
Error
Absolute
Error
Squared
Error
Percentage
Error
1 15
2 18
3 20
4 16
5 22
6 23 18.20 4.80 4.80 23.04 .21
7 25 19.80 5.20 5.20 27.04 .21
8 17 21.20 -4.20 4.20 17.64 .25
9 20 20.60 -0.60 0.60 0.36 .03
10 22 21.40 0.60 0.60 0.36 .03
11 24 21.40 2.60 2.60 6.76 .11
12 25 21.60 3.40 3.40 11.56 .14
13 27 21.60 5.40 5.40 29.16 .20
14 29 23.60 5.40 5.40 29.16 .19
15 30 25.40 4.60 4.60 21.16 .15
36.80 166.24 1.52

MAD = (? ?A t – Ft?) / n
= 36.80/10
= 3.68
MSE = (?(A t – Ft)2) / n
=166.24 / 10
=16.62
MAPE = ( (?A t – Ft?/ A t) × 100 )/ n
= (1.52/10)×100
=15.20

If we observe our data we can see that our error measures in case of k =3 are less than the three
error measures in case of k=5, so we can say that 3 -year moving average is giving better
forecasts in case of the data on LED TV sales. Th e graphical plot of the actual sales and the
estimated sales with two kinds of forecasts is given in Figure 1. MA(3) and MA(5) represents
the moving average forecast with k=3 and moving average forecast with k=5 respectively.

Figure 1: Moving Average Forecasts for Three and Five Years.
Now the forecast for the 16 th year can be calculated as given below:
F16 = (27+29+30)/3
=28.67
Therefore, the forecast for LED TV sales for the 16 th year is $28.67m.
• Weighted Moving Average Method
Weighted moving average method is naturally similar to a moving average, except that it allows
more weight to the most recent data in a time series. So this method is more meditative of the
recent observations in the time series. The sum of weighting shoul d up to 1 (or 100%). For
example, if five weights are used then five most recent observations are used to make the
forecast. The most recent value may be assigned a weight of 5/15, the next most recent value a
weight of 4/15, the next one a weight of 3/15, the next one a weight of 2/15, and the next after
that a weight of 1/15. Similarly, if four latest observations are used for forecasting then they may
0
5
10
15
20
25
30
35
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Sales
MA(3)
MA(5)

be assigned weights as 4/10, 3/10, 2/10, and 1/10 respectively where the largest weight is given
to the most latest observation. However, the selection of weights is subjective and somewhat
arbitrary. Generally suitable weighing scheme is arrived at by using trial and error technique.
Ft = w t-1(A t-1) + w t-2(A t-2) + . . . + w t-k(A t-k)
Where
wt-1 = weight for period t -1, so on
At-1 = actual value for period t -1, so on.
Let us calculate the weighted moving average forecast for the 7 th year using the data
given below:

Year Sales
1 15
2 18
3 20
4 16
5 22
6 23
Weighted moving average forecast using five observations:
F7 = .33 (23)+.27(22)+.20(16)+.13(20 )+ .07(18)
= 20.59
F7 = .4(23)+.3(22)+.2(16)+.1(14 )
= 21

• Exponential Smoothing Method
As in case of the moving average method, smoothing exponential method for forecasting is also
suitable, for short term forecasts. Therefore, these methods are appropriate to use, in situations
where the long term trend of the data cannot be established. The forecast for period t is defined
on the basis of forecast for the period t and a fraction of the difference between the forecast and
the actual value for period t.
Ft =Ft-1 + ?(A t-1 – Ft-1)
Where, F t = Forecast for period
Ft-1 = Forecast for the period (t -1)
? = Smoothing co -efficient (where 0