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.

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