Forecasting

UNIT 6 QUANTITATIVE METHOD OF

FORECASTING
Objectives
Upon completion of this unit, you will be able to:

learn the importance of forecasting for decision making

use forecasting techniques in operations management

understand different quantitative techniques of forecasting

know trend analysis, exponential smoothing, decomposition methods, and causal

method of forecasting

find out the suitability of forecasting models

calculate the errors in forecasting

Structures
6.1

Introduction

6.2

Forecasting

6.3

Application to Different Functional Areas

6.3.1

Forecasting in Operations Management

6.4

Specific Forecasting Methods

6.5

Main Classes of Quantitative Models

6.5.1

Time Series Models

6.5.2

Causal Models

6.6

Forecast Error

6.7

Selecting a Suitable Forecasting Method

6.8

Summary

6.9

Self-Assessment Exercises

6.10

Further Readings

6.1

INTRODUCTION

Forecasting is the art and science of predicting the future events. Forecasting was
largely an art, but it has now become a science as well. While managerial judgement
is still required for forecasting, the manager is added today by sophisticated
mathematical tools and methods. While all elements of operations management are
important, forecasting is viewed as the key elements in the operations structure. This
unit is an excellent overview of quantitative forecasting techniques and models and
help recognizing the different models. Also it will help to know their use according to
one's needs. It can be highlighted that the qualitative forecasting is discussed in unit 5
of MS53 . The reader has to read this quantitative forecasting in combination with the
qualitative forecasting. Then only he or she can have a complete understanding of the
subject of forecasting. The needs of the market are changing for us, and we have to
respond more quickly than before. To do so, we have placed a higher emphasis on
forecasting. Students are required to refer MS"8 for different types forecasting
techniques.

6.2

36

FORCASTING

Forecasting, in general, presents an unresolved philosophical dilemma. `You can

never plan the future by the past', said Edmund Burke. But Patrick Henry disagreed: I
know of no way of judging the future but by the past'. Operations managers try to
forecast a wide range of future events that potentially affect success. Main concerned
here is that of

forecasting customer demand for product and services. Forecasting may be shortterm or long-term by nature.
Forecasting is an essential tool in any decision-making process. It's uses vary from
determining inventory requirements for a local shoe store to estimating the annual
sales of video games. The quality of the forecast strongly related to the information
that can be extracted from past data.
Defining Forecasting
A forecast is an estimate of a future event achieved by systematically combining and
casting forward in a predetermined way data about the past. It is simply a statement
about the future. It is clear that we must distinguish between forecast per se and good
forecasts. Good forecast can be quite valuable and would be worth a great deal.
Long-run planning decisions require consideration of many factors: general economic
conditions, industry trends, probable competitors actions, overall political climate,
and so on.
Prediction, on the other hand, is an estimate of a future event achieved through
subjective considerations of managers. This subjective consideration need not occur
in any predetermined way.
Forecasts are possible only when a history of data exists. An established TV
manufacturer can use past date to forecast the number of picture screens required for
next week's TV assembly schedule. A fast-food restaurant can use past data to
forecast the number of hamburger buns required for this weekend's operations. But
suppose a manufacturer offers a new refrigerator or a new car, he cannot depend on
past data. He cannot forecast, but has to predict. For prediction, a good subjective
estimates can be based on the manager's skill, experience, and judgement. One has to
remember that a forecasting technique requires statistical and management science
techniques.
In general, when business people speak of forecasts, they usually mean some
combination of both forecasting and prediction. Commonly, forecasting is substituted
freely for economic forecasting. It implies for some combination of subjective
calculations and subjective judgement. We caution students and operations managers
to avoid misunderstanding.
Forecasts are often classified according to time period and use. In general, short-term
(up to one year) forecasts guide current operations. Medium term (one to three years)
and long-term (over five years) forecasts support decisions on plant location and
capacity Forecast are never perfect. Because it deals with past data, our forecasts will
be less reliable the further into the future we predict. That means forecast accuracy
decreases as time horizon increases. The accuracy of the forecast and the its costs are
interrelated. In general, the higher the need for accuracy translates to higher costs of
developing forecasting models. So how much money and manpower is budgeted for
forecasting? What possible benefits are accrued from accurate forecasting? What are
possible cost of inaccurate forecasting? The best forecast are not necessarily the most
accurate or the least costly. Factors such as purpose and data availability play
important role in determining the desired accuracy of forecast.

6.3

Quantitative Methods of Forecasting

APPLICATION TO DIFFERENT FUNCTIONAL

AREAS

Forecasting is one input to all types of business planning and control, both inside and
outside the operations function. Marketing uses forecasts to plan products,
promotion, and pricing. Finance uses forecasting for managing cash flows and as an
input to financial planning. Accountants rely on forecasts of costs and revenues for
tax planning. Human resource personnel need forecasts for recruiting.
The main focus of this unit is on forecasting on operations function. It serves as an
input for decision on process design, capacity planning, and inventory control. For
process design purposes, forecasting is needed to decide on the type of process and
the degree of automation to be used. For example, a low forecast of future sales
might indicate that little automation is needed and the process should be kept as
simple as possible. If greater volume is forecast, more automation and more elaborate
process including line flow might be justified. Since process decisions are long-range
in nature, they can require forecasts for many years into future. Forecast can measure
the variability in demand during lead time

37

Forecasting

that in turn can help carry proper safety stock levels. Appropriate safety stock
inventory levels could minimise overall carrying and stockout costs associated with
these items.

6.3.1 Forecasting in operations management

In studying forecasting, we must be careful not to be emotional in immersing
ourselves in techniques and loose track of the reasons for forecasting. Forecasting is
an important component of operations planning. It is absolutely necessary for
planning, scheduling, and controlling the system to facilitate effective and efficient
output of goods and services.
Forecasting is helpful in operations management as regard to the aggregate demand
forecast. It is obtained by estimating expected volumes of sales, expressed in dollars,
and then converting the sales dollars into homogeneous production units. Production
unit can be subdivided into component parts and converted into labor or material
requirements. These resource forecasts are used to plan and control operation
subsystems as shown Figure 6.1

Refer Figure 6.1. There are three types of operations sub-functions which need
forecasting. These operations sub-functions are planning the system, scheduling the
system, and controlling the system. Each one will be discussed below in detail.

38

Planning the system: Managers need to forecast demands so that they can design or
redesign processes necessary to meet demand. Automated, continuous flows facilitate
high production volumes; manual or semi-automated, intermittent flows are generally
more economical for smaller production volumes. The demand forecast is critical to
this design. We have discussed a bit of this at the beginning of the unit. Wide
variation between anticipated demand and actual demand can result in excessive
operations"costs. Capacity planning utilities forecasting at different levels. A long
range forecasting is needed for planning the total capacity of facilities. For medium
range capacity decisions, a detailed forecasting will be needed to determine the
subcontracting, hiring plans, and equipment utilisation. Shoat-range capacity
decisions, including assignment of available people and machines to jobs or activities
in the near future, should be detailed in terms of individual products and they should
be highly accurate. If capacity is not expanded fast enough, both individual firms and
the national economy suffer. On the other hand, too much capacity is burdensome.
For example, Jet aircraft, at $20 million each, cannot be purchased and stocked for
occasional demand, since the cost of excess capacity is considerable. Boeing,
McDonnell Douglas, and airbus- the world's largest commercial aircraft producerstry

very hard to have manufacturing plants size a to meet exactly the number of aircraft
demanded. If the plants are too large, it will be costly to the firm.

Quantitative Methods of Forecasting

Scheduling the system: Job scheduling in intermittent and continuous operations is

more stable if demand forecasts are accurate. Accurate demand forecasting is needed
for best utilisation of the existing conversion system. Managers need intermediate run
demand forecasts for three months, six months, and a year into the future. Both
current and future workforce levels and production rates must be established from
these forecasts.
Controlling the system: In regards to controlling inventory, production, labor, and
overall costs, managers need accurate demand forecast. Accurate forecasts are
needed for the immediate future- hours, days, and weeks ahead. Thus a computerised
forecasting system may be needed for these decisions.
In general, there are different types of decisions in operations and different associated
forecasting requirement as shown in Table 6.1. A peep into the table indicates that
there are two types of forecasting methods in operations management: qualitative
methods of forecasting and quantitative methods of forecasting. It is to be reminded
that qualitative forecasting has been discussed in unit S. This unit will deal with the
quantitative methods of forecasting only.
Table 6.1 Forecasting uses and methods
Uses of
forecasting
For operations
Process design

Time Accuracy
horizon required

Number of
products

Long

Single or few Top

Medium

Capacity planning Long

Medium
Facilities
Aggregate
Medium High
planning
Scheduling
Inventory
management

Short
Short

Highest
Highest

Management Forecasting
level
methods

Single or few

Top

Few

Middle

Qualitative
and causal
Qualitative
and causal
Causal and

Many
Many

Lower
Lower

Time series
Time series
Time series

Source: Operations Management by R.G Schroeder, McGraw-Hill.

Qualitative forecasting depends on managerial experience, because they do not use
any h specific quantitative models. This method is helpful when past data are not
available to the managers or not reliable. Thus different individuals can apply the
same qualitative methods hilt can arrive at different forecasting results. Because of
non-availability of data, the managers can utilise the forecasting by using the
qualitative methods. Some of the well-known qualitative methods are:

Management judgment

Consensus

Writing `scenarios' of the possible events that might occur

Judgmental methods(Delphi technique)

Based on the individual's feeling and expert opinion

Some of these methods are discussed in unit
Quantitative forecasting is used when:

Past data is available, and

Past data can be fitted into a pattern that can be expected to continue into the
future.

Activity A
Think of any organisation of manufacturing or service sector. Evaluate its short-term,
medium-term and long-term forecasting techniques. Which of the forecasting
technique are in use? '

39

Forecasting

6.4

SPECIFIC FORECASTING METHODS

We have been highlighting that forecasting is important in operations and strategic

planning. In stead of getting too deeply into specific types of forecasts for varying
situations, we summarize that the long-range strategic planning and facilities
decisions use less analytical qualitative methods of forecasting, and operational
planning in production and inventory control uses more analytical, time series
analysis. Causal forecasting techniques are used for a variety of planning situations
but are specially in intermediate-term planning.
Table 6.2 summarizes modern forecasting techniques being used in industries. There
are three types of forecasting techniques: qualitative forecasting, naive (time series/
forecasting, and causal forecasting. Qualitative and naive models are the most
frequently used forecasting techniques in operations management. The causal models
are very costly models to implement and are not suitable for short-term forecasting
typically needed h operations managers. Even though qualitative models are very
popular, it has been discussed in unit 5, and this unit will limit its discussion to
quantitative forecasting.

6.5

MAIN CLASSES OF QUANTITATIVE MODELS

1) Time series models

i)

Trend projection (long-term)

ii)

Decomposition methods (intermediate-term)

iii) Smoothing methods (short-term)

a)

exponential smoothing

b)

Naive forecasting

2) Causal models

40

i)

Regression and correlation analysis

ii)

Econometric model

Quantitative Methods of Forecasting

iii) Simulation method

iv) Input-Output model
A causal model attempt to relate some quantity to other factors. Time series models
attempt to relate some quantity strictly to time. A time-series is a time ordered
sequence of observations which have been taken at regular intervals over a period-of
time (hourly, daily, weekly, monthly, annually etc.).
Some of the important forecasting techniques are explained below.
6.5.1

Time Series Models

i)

Trend Projection

Reasons for studying trends are:

1)

The study of secular trends allows us to describe a historical pattern.

2)

Studying secular trends permits us to project past patterns, or trends, into the future.

3)

In many, situations, studying the secular trend of a time series allows us to

eliminate the trend component from the series.

Trend can be a straight line or a curvilinear. Example of straight line trend is increase
of pollutants in the environment as shown in Figure 6.1(a) . Common example of
curvilinear relationship is the life cycle of a new product. When a new product is
introduced, its sales volume is low (i), as the product gains recognition and success,
unit sales grow at an increasing rate (ii). After the product is firmly established, the
unit sales grow at a stable rate (iii). Finally, as the product reaches the end of its life
cycle, unit sales begin to decrease (iv) in Figure 6.1(b).

Trend reflects the effect of global movements in the time series. Consider the linear
trend. Yt= bo +blx
Where T, = estimated value of the dependent variable
bo = Y- intercept
bl = slope of the trend line
x = independent variable (time in trend analysis)
Objective is to find out the value of slope(bl) and intercept(b) from the given set of
independent values of x.
We can describe the general trend of many time series using a straight line. But we
are faced with the problem of finding the best-fitting line. We can use the least
squares methods to calculate the best-fitting line, or equation.
Least-squares Method
Find b,, and b1 such that we get the "best" line which fits the data.
Given (Y1, Y2, Y3, ..., Yn), find b, and bl such that the demand

41

Forecasting

Example: Sales data of a company are given for 14 time periods (t) with respective
sales data (yt). Forecast the sales value for the period 15.
(Adapted from course material, Department of Management Studies, IIT, Delhi).

42

Quantitative Methods of Forecasting

Tt is measured in units of Yt but St , Ct and Rt are all measured in relative terms.

Values of St, Ct, and Rt >1 indicate effects above the trend (above the average level).

43

Forecasting

Steps in decomposition method

Step 1: Identify seasonal factors: For the actual series D, compute a moving average,
whose length is equal to the seasonality (e.g. 12 in monthly data, 4 in quarterly data ,
7 in daily data). Before calculation of moving average, let's define it.
What is a moving average?
Moving average is an average that is repeatedly updated. As we move into the future
and. as new observations become available, the oldest values in the series are
discarded, thereby keeping the average current The message given by the moving
averages technique is that while history helps to plan the future, only a small
fragment of the past is relevant. Thus, if we have the monthly demand data, we may
calculate average of only six month's values in the data to get the forecast for the next
month and such a moving average is called a six month moving average. The
selection of the number of periods in the moving average what is being forecasted
and (ii) the characteristics of the demand. Simple n -period moving average forecast
for the period (t+l) = (aDk) x 1/nwhere.k = (t - n+l) tot.
The moving average technique can be made simple by an example. For 4- period
moving average, we need four observations to compute.
Example: (Adapted from course material, Management Department, IIT, Delhi)
The following table gives the number of patients in a family practice for last four
years. Each year is divided into four quarters and the respective patients being treated
are included for forecasting.

How do you make forecasts for the period 17 i.e. for the first quarter of 5'' year,
considering the trend component, seasonal component, cyclical component and
random component?

44

Solution: The following table calculates the 4-period moving averages (MAs) and
centered MAs. The sample calculations are shown below. You can calculate the
values very fast and accurately with the help of MS-Excel.

Quantitative Methods of Forecasting

Here the time period t =1, 2...16

4-period MA

(920+916+895+905) / 4 = 909 (at period 4)

(916+895+905+947) / 4 = 915.75 (at period 5)

and

(1201+1142+1106+1163 ) / 4 = 1153 (at period 16)

from period 6 to 15, you can calculate the values accordingly and cross check with
tabulated vales.
Note that the first moving average (= 909) is for the whole first year i.e. it is
associated with all the four quarters of the whole year. Therefore it is reasonable to
center it. This is the (Tt Ct) component. Now question is where do we center it2 It can
be between 2nd and the 3rd quarter.
We can handle this by taking the midpoints of successive moving averages. For
example: (909+915.75) / 2 = 912.375 (can be centered at 3' quarter) that is the
average of the MA of 4th quarter and MA of 5th quarter.
This indicates that you are using 50 observations and centering at the middle.
Similarly, (915.75+919.25) / 2 =917.50 (center at 4)
Other values are calculated and tabulated for your reference.
Step 2: From Tt Ct component (centered MA), identify seasonal-random component.

St R t =

Yt
Tt C t

For Year 1
Q3=> 895 /912.375= 0.981
Q4=> 905 / 912.5 = 0.986

45

Forecasting

Year 1
Year 2

Year 3

Year 4

Q3
Q4
Q1
Q2
Q3
Q4
Q1
Q2
Q3
Q4
Q1
Q2

0.981
0.986
1.029
1.005
0.964
0,995
L048
0.997
0.939
0.930
1.113
1.013

Note: We loose 4 points, 2 at the beginning (i.e. Ql and Q2) and 2 at the end (Q3 and
Q4).
Step 3: Seasonal factor (St) of each quarter can be obtained by taking averages from
each year.
1st quarter effect = (1.029+1.048+1.113) /3 =1.063
2nd quarter effect = (1.005+0.997+1.013) / 3 =1.005
3rd quarter effect = (0.981+0.964+0.939) / 3 = 0.961
4th quarter effect = (0.986+0.995+0.93) / 3 =0.970
Step 4: Deseasonalize the series now
Tt Ct Rt= Yt/ St
For year 1
QI
Q2
Q3
Q4

920/ 1.063 =865.475

916/1.005 = 911.442
895/0.971 = 931.32
905/0.970 = 932.99 and so on.

Deseasonalized series
865.475

Year 1

911.442
931.320

Step 5: Next important point is to identify the trend

in the deseasonalized series.

932.990
890.874

Year 2

925.373
969.072
Year.3

960.458
989.691
1136.318
1150.884

46

1198.969

and calculating the values of bo, b1 and get the value

as Tt = 825.94 + 19.12 t
Step 6: Obtain cyclical component C1Rt = Yt / (StTt)
Where (Yt / S) is deseasonalised series.' Before
calculating CtRt, you must calculate the values of Tt
i.e. the trend in the

947.264

1129.821

From trend projection, we know that Eq. 6.1 is

Tt = bo+b1t;t= 1,2,..., 16

938.605
936.971

Now we have calculated the deseasonalized value.

Year 4

deseasonalised series for all the 16 periods.

The values are given in the table. Students are required to check the values. Then
calculate the values of Ct Rt

Quantitative Methods of Forecasting

It is required to forecast the values for the

period 17 i.e. for the first quarter of 5th year.
T17=825.94 + 19.12(17) =1150.98
Considering cyclical factor which is usuall
forecasted by using smoothing techniques
(explained in this unit) C17= 1.02
Now T17C17 = 1150.98 (1.02) = 1174.
Note: Use Ms-Excel to calculate the values.

Forecast =1174 (1.063) =1247.96 (as 1.063 is the seasonal factor for the 1 quarter).
=> Forecast = 1274.96 for 1st quarter of the 5th year.
Activity B
In the Atlanta area, the number of daily calls for repair of speedy copy machines has
been recorded as follows:

a)

Prepare a three-period moving-average forecast for the data. What is the error on
each day?

b)

Prepare a three period weighted-moving-average forecast using weights of WI

=0.5, w2 =0.3 ans w3= 0.2.

c)

Which of these two forecasts is the best.

47

Forecasting

iii)

Smoothing Technique

It is suitable for short-term forecasting. It is generally averaged by weighting the past

data. The extreme case is nave forecasting.
a)

Exponential Smoothing

Exponential smoothing is an averaging method that exponentially decreases the

weighting of old demands. The models are well known and often used in operations
management. They are popular because of the fact that (i) they are readily available
in standard computer software packages, and (ii) they require relatively little data
storage and computation.
Exponential smoothing is distinguished by the special way it weights each past
demand. The pattern of weights is exponential in form. Demand for the most recent
period is weighted most heavily and the weights decreases exponentially for the old
periods. In other words, the weights decreases in magnitude the further back in time
the data are weighted, the decrease is nonlinear.
Let Ft = forecast for next period t
As each time period expires, a new forecast is made.
Forecast for next period (Ft) = a* (actual demand for most recent period)
+ (1-a) *(demand forecast for most recent period)
= > Ft= a Y t-1 + (1- ) F t_1

Eq. (6.8)

Where Y t.1= the most recent observation

F t.1= the old forecast

= smoothing parameter, 0 < < 1, which indicates that one has to specify
the value of .
Why is this model called exponential smoothing?
The expansion of the above equation (6.8) shows that
F1= Y1+ (1- ) F
Also Ft-1= Y t.2 + (1- ) F t z, putting the value of Ft-1 in eq.6.8, we get
Ft= Y t_1 + (1- ) [ Y t-2 + (1- ) F t-2)
or Ft=a Y .,+(1- ) Y t_2+(1- )2 F t.2 Eq. (6.9)
Similarly F t-2 = Y + (1- ) F t-3 and substituting the value of F t-2

Eq. (6.9)

We get Ft= a Y t-1+ (1- ) a Y t-2+(1- ) [ Y t-3+ (1- ) F t-3]

=> Ft = [Yt-1+ (1- ) Y t-2+(1- )2Y as ] + (1-a)3 F t-"s
=> Ft = a [(I- ) Y t-1+(1- )1 Y t-2+(1- )2Y t-3. + (1- )3 F t-3

Eq. (6.10)

The expansion of Eq. (6.10) can be continued further, but it will not help us. Since 0
< a < 1, the term a(l-a), a (I-a)', a (1-a)2 and so forth are successively smaller in the
equation. More specifically these weights decrease exponentially. Weights decrease
as we consider past observations. The most recent observation has the highest weight
and the weights are decreasing exponentially(or geometrically) as we go to past
observations. Thus exponential smoothing is also called the exponentially weighted
moving average.
Rearranging and rewriting Eq. (6.8)
Ft=Ft-l + (Yt-1-Ft-1)
Where the term (Y t-1- F t-1) is called the error term: (e t-1)
The form of the forecasting equation above is known as the `error correction form' of
the ES model.
The higher the value of a, the faster the system responds to current data: As a
increases, there is more correction.
If =1 => Ft = Y t-1(Naive forecast)

48

= 0 => Ft= F t-1 (Forecast never changes and data has no role)

Quantitative Methods of Forecasting

These are the two extreme cases.

The limitation of ES is that (i) it is not suitable for data that include long-term
upward or downward movements(trend) in data.(ii) the use of ES would produce
forecast that were too low for upward trend and too high for downward trend(i.e.
either overestimated or underestimated).
Example: Sales data for 10 periods are given. Forecast the data for the next period
i.e. for the 11' period. Assume a =0.4 and F1= 700 (initial value)
t

Yt

700

724

720

728

740

742

758

8'

750

770

10

775

Solution: We can use the equation (6.8) Ft= F + a (Y t-1- F t-l) for solving the problem.
t

Yt

Ft

700

700

724

700

720

709.6

728

713.76

5
6

740
742

719.46
727.67

758

733.4

750

743.24

770

745.95.

10

775

755.57

11

763.34

For calculating Ft (i.e. forecasting for the period t) we need the value of Ft-1.,. For
simplicity we can assume the first value of Ft equal to 700 i.e. the value equal to Yt.
Next value can be forecasted as: Ft = Ft-1 + a (Y t.4 - F t-1.)
Where Ft-1 = 700 (just previous value), Y t-1 = 700 (just previous -value) and a = 0.4
=>

700 + 0.4 (700-700) =700

For next forecasting: F t-1=700, Y t-1 =724 and a = 0.4

=>

700 + 0.4(724-700) = 709.6

The last forecasting is F11 = 755.57 + 0.4(775-755.57) =763, 34

Similarly rest of the calculation can be carried out. The calculated values are
tabulated in the table above.
Activity C
Using the data in activity B, prepare exponentially smoothed forecasts for the
following cases:

49

Forecasting

a)

a =0.land Ft =130

b)

a = 0.3 mid ft =130

.
Activity D
Compute the errors of bias and absolute deviation for the forecasts in activity c.
Which of the forecasting models is the best?

.
Advantages: Simple exponential smoothing and other exponential smoothing models
share the advantage of requiring very few data points be stored. To update the
forecast from period to period, you need only a, last period demand, and last period
forecast. Remember this model incorporates in the new forecast all past demands.
The model is easy to understand and easily computerised for thousands of part
numbers, supply items and inventory items. The model is helpful in both the
manufacturing and service sectors.
Difference Between the Decomposition Method and Smoothing Method
1) The decomposition methods assume that the seasonality for a particular period
of the seasonal cycle can be treated as constant, whereas smoothing method
constantly update the seasonal ratios.
2) The decomposition methods determine a particular linear or non-lenear trend in
the data and use this relationship in all future forecasting, the smoothing
methods constantly track and update the trends (up and down) in every period.
3) The decomposition methods attempt to determine the cyclical factor and use it
across the board.
4) The basic difference between the decomposition methods and the smoothing
methods is that, the later (i) continually tracks and modifies the demand
components and (ii) does not isolate the demand components but gives the
correctness in the procedure.
If the demand pattern is more stable over a long period of time, the decomposition
method is suitable, whereas if the demand pattern is fluctuating more frequently, the
smoothing techniques may be more useful.
Decomposition method, as explained earlier, is the forecasting required for
intermediate range forecasting. This type of forecasting is helpful in' production
planning. Decomposition methods may also use smoothing in their analysis.

6.5.2 Causal Models

i)

Regression Analysis

It is a causal forecasting model in which, from historical data, a functional

relationship is established between variables and then use to forecast dependent
variable values. We consider a simple regression equation here for two variables and
their linear relationship.
Ft=a+bXt
Where is the forecast for period t, given the value of the variable X in period t. The
coefficient a and b are the intercept and slope of the line. The coefficient a and b are
computed by the following two equations: 41?
b =

a=

50

n ( X t D t ) ( X t )( D t )
( X

2
t

) ( X t )2

Dt b X t
n

Where D= a + bX and n is the number of periods.

Example: (Adapted from Production/Operations Management by Adam and Ebert).

A paper box company makes carryout pizza boxes. The operations planning
department knows that the pizza sales of a major client are a function of the
advertising dollars the client spends, an account of which they can receive in advance
of the expenditure. Operation planning is interested in determining this relationship
between the client's advertising and sales. The amount of pizza boxes the client will
order, in dollar volume, is known to be a fixed percent of sales.

Quantitative Methods of Forecasting

Quarterly advertising and sales

Forecast for the next period i.e. for the period 11.
Solution: Computing b and then a, where advertising is Xt
For quarter t, sales are Dt for quarter t , sales are Dt for quarter t, and forcast is Ft for
future period t

b= [10(328) -96 (30) ] / [(10(1060) - 96* 96] = 0.29

a= [30-0.29(96) ] / (10) = 0.22
Thus, the estimated regression line, the relationship between future sales Ft and
advertising Xt is Ft= 0.22 + 0.29Xt
The operations planner can now ask for planned advertising expenditures , and from
that sales can be forecasted.
Now substitute 11 for Xt into the equation above gives
Ft = 0.22+0.29(11) = 3.41
So sales are forecast as $3,410,000.

51

Forecasting

Note: A variety of regression models are discussed in block 5 of MS-8. Students have
to read this unit in combination with block 5 of MS-8 so as to get a clear
understanding of the regression models for forecasting.
Activity E
Describe the uses of qualitative, time series and causal forecasting in your
organisation.
.........................................................................................................................................
.........................................................................................................................................
.........................................................................................................................................

6.6

FORECAST ERROR

Whether it is a simple smoothing or more advanced smoothing, an estimate of

forecast error should be computed along with the smoothed average. This error
estimate might be used for several purposes: .
i)

To set safety stocks or safety capacity and thereby ensure a desired level of
protection against stockout.

ii)

To monitor erratic demand observations.

iii) To determine when the forecasting method is no Ionger tracking actual demand
and needs to be reset.
In forecasting, a commonly used forecasting error is mean absolute deviation or
MAD. It is a forecast error measure that is the average forecast error without regard
to direction; calculated as the sum of the absolute value of forecast error for all
periods divided by the total number of periods evaluated. It is mathematically defined
as:

MAD=

Sum of the absolute value of the forecast error for all periods
Number of periods

(Forecast demand - actual demand)

n
Where n is the number of periods
The above expression is just the average error observed, without regard to sign, over
all the past periods of forecasting. The MAD is similar to standard deviation, except
we have not squared the errors for each period and then taken the square root of
them. This is called Mean square error (MSE). Mathematically it can be written as:
MSE = (1/n) x S (Yt -Ft) 2
MAD = (1/n) x S Yt Ft 1/2
MAD is used to determine whether the forecast is tracking with the actual time-series
values. To determine this, a tracking signal is computed, as follows:

Tracking signal =

Running sum of forecast error RSFE

=
MAD
MAD

Tracking signal is thus a computation of bias in the numerator divided by the most
recent estimate of MAD. Bias is a less commonly used error measure.
Bias

= (sum of forecast error for all periods) / number of periods

= (forecasted demand - actual demand) / n

In computerised forecasting systems it is extremely important to incorporate error

controls of the type discussed above. This will ensure that the system does not run
out of control.

52

Example: Read the table below Forecast values and the actual values are given.
Calculate RSFE, Forecast error, cumulative error, MAD. Calculate also the tracking
signal (TS).

Quantitative Methods of Forecasting

Solution: The tracking signal is calculated in the table below which is self explanatory.

For forecast to be in control 89% of the errors are expected to fall within. 2
MADs , 98% within 3MADs, or 99% within 4MADs.

6.7

SELECTING A SUITABLE FORECASTING

METHOD

In this section we will present a framework for selecting a suitable qualitative, timeseries, and casual methods of forecasting. The framework is based in large part on the
survey conducted by Wheelwright and Clarke (1976), who identified factors that
companies consider important when they select a forecasting method. The most
important factors are indicated as follows:
i) User and System Sophistication: How sophisticated are the managers who are
expected to use the forecasting results? It has been found that the forecasting method
must be matched to the knowledge and sophistication of the user. Generally
speaking, managers are reluctant to use result's from techniques they do not
understand. Another factor is the status of forecasting systems currently in use.
Wheelwright and Clarke found that forecasting systems tend to evolve toward more
mathematically sophisticated methods; they do not change in one grand step. So the
method chosen must not be too advanced or sophisticated for its users or too far
advanced beyond the current forecasting system
ii) Time and Resource Available: The selection of a forecasting method will
depend on the time available in which to collect the data and prepare the forecast.
This may involve the time of .users, forecasters and data collectors. The time
required is also closely related to the necessary resources and the costs of the
forecasting method. The preparation of a complicated forecast for which most of the
data must be collected may take several months and cost thousands of money Value.
For routine forecasts made by computerized systems, both the cost and the amount of
time required may be very modest
iii) Use or Decision Characteristics: The forecasting method must be related to the
use or decisions required. The use is closely related to such characteristics as
accuracy required, time horizon of the forecast, and number of items to be forecast.
For example, inventory and scheduling decisions require highly accurate short range
forecasts for a large number of items. Time-series methods are ideally suited to these
requirements. On the other hand, decisions involving process and facility planning
are long range in nature; they require less accuracy for, perhaps, a single estimate of
total demand. Qualitative or casual methods

53

Forecasting

tend to be more appropriate for those decisions. In the middle time range are
aggregate planning and budgeting decisions which often utilize time-series or casual
methods.
iv) Data Availability: The choice of forecasting method is often constrained by
available data. An econometric model might require data which are simply not
available in the short run; therefore another method must be selected. The BoxJenkins time-series method requires about 60 data points (5 years of monthly data).
v) Data Pattern: The pattern in the data will affect the type of forecasting method
selected. If the time-series is flat, a first order method can be used. However, if the
data show trends or seasonal patterns, more advanced methods will be needed. The
pattern in the data will also determine whether a time-series method will suffice or
whether casuals model are needed. If the data pattern is unstable over time, a
qualitative method maybe selected. Thus the data pattern is one of the most important
factors affecting the selection of a forecasting method.
Another issue concerning the selection of forecasting methods is the difference
between fit and prediction. When different models are tested, it is often thought that
the model with the best fit to historical data (least error) is also the best predictive
model. This is not true. For example, suppose demand observations are obtained over
the last eight time periods and we want to fit the best time-series model to these data.
A polynomial model of degree seven can be made to fit exactly through each of the
past eight data points. But this model is not necessarily the best predictor of the
future.
Finally, an interesting question concerning model selection is the accuracy of
qualitative human forecasting versus quantitative model based forecasting. Ebert
(1976) compared humans to model for a variety of underlying time-series demand
patterns. He found that, when the data included a great deal of random noise or nonlinear seasonal patterns, models did better than humans provided that care was taken
in fitting the models. However, when simple (first order) exponential models with an
arbitrary value of a were used, the humans often performed better than the models.
This research indicates that quantitative models do not always provide better
forecasts than humans.

6.8

SUMMARY

The unit illustrated the concept of business forecasting in operations management and
define forecasting as the use of the past data to future events. Prediction, on the other
hand, refers to subjective estimates of the future. Skill, experience, and sound
judgement of a manager are required for good predictions.
We studied the tradeoffs between cost and accuracy in forecasting approaches.
Generally, less expensive forecasting approach leads to less accurate results. We
discussed two basic groupings of forecasting techniques: naive (time-series) models
i.e. trend analysis, decomposition method and exponential technique, and casual
models i.e. regression technique. Research results illustrated that forecasting model
depends upon the length of the forecast period, the level of noise, the measure of
forecast error, and, the demand pattern. Often, forecasts are not made with statistical
models but individuals can and do intuitively use past data to forecast future events.

6.9

54

SELF-ASSESSMENT EXERCISES

1)

How does a quantitative forecasting differ from a qualitative forecasting?

2)

What do you understand by time series analysis? How would you go about
conducting such an analysis for forecasting the sales of a product in your firm?

3)

Compare time series analysis with other methods of forecasting? Briefly

summarize the strength and weakness of various methods.

4)

A test was conducted on a given process for the purpose of determining the
effect of an independent variable X (such as process temperature) on certain
characteristics of a finished product Y (such as density). Twenty observation
were taken and the following results were obtained:

Quantitative Methods of Forecasting

X = 5.0 ( X 1 X ) = 160, ( X 1 X )(Y1 Y ) = 80

Y = 3.0 (Y Y ) 2 = 83.2
Assume a model of the type Y = bo +bl X +I
a)

Calculate the fitted regression equation

b)

Prepare the analysis of the variable table

c)

Determine 95% confidence limits for the true mean value of y when
1)

X=5.0

2)

X=9.0

5) A survey of used car sales in a city for the 10 year period 1976-85 has been
made. A linear trend was fitted to the sales for month for each year and the
equation was found to be Y= 400 + 18t, where t=0 on January 1, 1981 and t is
measured in 1/2 year (6 monthly) units.
a)

use this trend to predict sales for June, 1990.

b)

If the actual sales in June, 1987 are 600 and the seasonal index for June
sales is 1.20, what would be the relative cyclical, irregular index for June,
1987?

6) What would be the considerations in the choice of a forecasting methods?

7) Find the 4-quarter moving average of the following time series representing the
quarterly production of coffee in an Indian state.
Production (in tones)
Year

Quarter I

Quarter II

Quarter III

Quarter IV

1983

10

17

1984

10

16

1985

18

1986

15

19

1987

14

21

8) Given below is the data of production of a certain company in lakhs of units.

Year
Production

1991

1992

1993

1994

1995

1996

1997

15

14

18

20

17

24

27

a)

Compute the linear trend by the method of least spa re.

b)

Compute the trend values of each of the year.

9) Contrast forecasting and prediction and give an example of each.

10) Which would use in evaluating a forecast, MAD or Bias? Why?
11) Present evidence that suggests forecasting is an important tool in the service
sector.
12) Describe the cost \ accuracy tradeoffs associated with both sophisticated
statistical models and intuitive forecasting.
13) An ice creme parlor experiepced the following demand for ice cream last month.
The current forecasting procedure is to use last year's corresponding weekly sales
as this year's forecast.

June 1.

Forecasted
demand
(in gallons)
210

Actual
demand
(in galloons)
200

June 8

235

225

June 15

225

200

Week

June 22
270
Calculate MAD and bias and interpret each

260

55

Forecasting

14 A manufacturer of mole and gopher poison has experienced the following

monthly demand for an environmentally sound pesticide poison.
Month

Actual demand
(in cases)

February

620

March

840

April.

770

May

950

June

1000

a)

Using a simple average, what would the forecast have been for May and June.

b)

What would be the three-month simple moving average have been for May and June?

c)

Which forecasting method would you recommend? Why?

6.10 FURTHER READINGS

Adam,E.E., and Ebert, Ronald J., 1997. Production-and Operations Management,
Fifth Edition, Prentice-Hall of India, New Delhi.
Chary, S.N., 1997. Theory and Problems in Production and Operations Management.
TMH Outline Series, New Delhi.
Levin, R.I., and Rubin, D.S., 1997. Statistics for Management. Prentice Hail of India,
New Delhi.
Narasimhan, L., McLeavey, D.W., and Billinton, P.J., 1997. Production Planning and
Inventory Control, Prentice Hall of India, New Delhi.
Schroeder, R.G., 1985. Operations Management. Second Edition, McGraw-Hill, New
York.

56