You are on page 1of 24

Overview of Qualitative Methods Forecuxtlng

Forecasting is the estimation oI the value oI a variable (or set oI variables) at some Iuture point
in time. In this note we will consider some methods Ior Iorecasting. A Iorecasting exercise is
usually carried out in order to provide an aid to decision-making and in planning the Iuture.
Typically all such exercises work on the premise that if we can predict what the future will be
like we can modify our behaviour now to be in a better position, than we otherwise would
have been, when the future arrives. Applications Ior Iorecasting include:
O lnvenLory conLrol/producLlon plannlng forecasLlng Lhe demand for a producL enables us Lo
conLrol Lhe sLock of raw maLerlals and flnlshed goods plan Lhe producLlon schedule eLc
O lnvesLmenL pollcy forecasLlng flnanclal lnformaLlon such as lnLeresL raLes exchange raLes
share prlces Lhe prlce of gold eLc 1hls ls an area ln whlch no one has yeL developed a rellable
(conslsLenLly accuraLe) forecasLlng Lechnlque (or aL leasL lf Lhey have Lhey havenL Lold
O economlc pollcy forecasLlng economlc lnformaLlon such as Lhe growLh ln Lhe economy
unemploymenL Lhe lnflaLlon raLe eLc ls vlLal boLh Lo governmenL and buslness ln plannlng for
Lhe fuLure
Think Ior a moment, suppose the good Iairy appeared beIore you and told you that because oI
your kindness, virtue and chastity (well - it is a Iairy tale) they had decided to grant you three
Iorecasts. Which three things in your personal/business liIe would you most like to Iorecast?
Personally I would choose (in decreasing order oI importance):
O Lhe daLe of my deaLh
O Lhe wlnnlng numbers on Lhe nexL uk naLlonal loLLery
O Lhe wlnnlng numbers on Lhe uk naLlonal loLLery afLer LhaL one
As you can see Irom my list some Iorecasts have liIe or death consequences. Also it is clear that
to make certain Iorecasts, e.g. the date oI my death, we could (in the absence oI the good Iairy to
help us) collect some data to enable a more inIormed, and hence hopeIully more accurate,
Iorecast to be made. For example we might look at liIe expectancy Ior middle-aged UK male
academics (non-smoker, drinker, never exercises). We might also conduct medical tests. The
point to emphasise here is that collecting relevant data 2, lead to a better Iorecast. OI course it
may not, I could have been run over by a car the day aIter this written and hence be dead already.
Indeed on a personal note I think (nay Iorecast) that companies oIIering Web (digital)
immortality will be a big business growth area in the early part oI the 21
century. Remember
you saw it here Iirst!

1ypex of forecuxtlng problemx,methoJx
One way oI classiIying Iorecasting problems is to consider the timescale involved in the Iorecast
i.e. how Iar Iorward into the Iuture we are trying to Iorecast. Short, medium and long-term are
the usual categories but the actual meaning oI each will vary according to the situation that is
being studied, e.g. in Iorecasting energy demand in order to construct power stations 5-10 years
would be short-term and 50 years would be long-term, whilst in Iorecasting consumer demand in
many business situations up to 6 months would be short-term and over a couple oI years long-
term. The table below shows the timescale associated with business decisions.
Timescale Type of Examples

Short-term Jperating Inventory control
Up to 3-6 months Production planning, distribution

Medium-term Tactical Leasing of plant and equipment
3-6 months - 2 years Employment changes

Long-term Strategic Research and development
Above 2 years Acquisitions and mergers
Product changes
The basic reason Ior the above classiIication is that diIIerent Iorecasting methods apply in each
situation, e.g. a Iorecasting method that is appropriate Ior Iorecasting sales next month (a short-
term Iorecast) would probably be an inappropriate method Ior Iorecasting sales in Iive years time
(a long-term Iorecast). In particular note here that the use oI numbers (data) to which quantitative
techniques are applied typically varies Irom very high Ior short-term Iorecasting to very low Ior
long-term Iorecasting when we are dealing with business situations.
Forecasting methods can be classiIied into several diIIerent categories:
O quallLaLlve meLhods where Lhere ls no formal maLhemaLlcal model ofLen because Lhe daLa
avallable ls noL LhoughL Lo be represenLaLlve of Lhe fuLure (longLerm forecasLlng)
O regresslon meLhods an exLenslon of llnear regresslon where a varlable ls LhoughL Lo be llnearly
relaLed Lo a number of oLher lndependenL varlables
O mulLlple equaLlon meLhods where Lhere are a number of dependenL varlables LhaL lnLeracL
wlLh each oLher Lhrough a serles of equaLlons (as ln economlc models)
O Llme serles meLhods where we have a slngle varlable LhaL changes wlLh Llme and whose fuLure
values are relaLed ln some way Lo lLs pasL values
We shall consider each oI these methods in turn.

uulltutlve methoJx
ethods oI this type are primarily used in situations where there is judged to be no relevant past
data (numbers) on which a Iorecast can be based and typically concern long-term Iorecasting.
One approach oI this kind is the Delphi technique.
The ancient Greeks had a very logical approach to Iorecasting and thought that the best people to
ask about the Iuture were supernatural beings, gods. At the oracle at Delphi in ancient Greece
questions to the gods were answered through the medium oI a woman over IiIty who lived apart
Irom her husband and dressed in a maiden's clothes. II you wanted your question answered you
had to:
O provlde some cake
O provlde an anlmal for sacrlflce and
O baLhe wlLh Lhe medlum ln a sprlng
AIter this the medium would sit on a tripod in a basement room in the temple, chew laurel leaves
and answer your question (oIten in ambiguous verse).
It is thereIore legitimate to ask whether, in the depths oI a basement room somewhere, there is a
laurel leaI chewing government servant who is employed to Iorecast economic growth, election
success, etc. Perhaps there is!
ReIlect Ior a moment, do you believe that making Iorecasts in the manner used at Delphi leads to
accurate Iorecasts or not?
Recent scientiIic investigation (New Scientist, 1st September 2001) indicates that the medium
may have been "high" as a result oI inhaling hydrocarbon Iumes, speciIically ethylene,
emanating Irom a geological Iault underneath the temple.
Nowadays the Delphi technique has a diIIerent meaning. It involves asking a body oI experts to
arrive at a consensus opinion as to what the Iuture holds. Underlying the idea oI using experts is
the -elief that their view oI the Iuture will be better than that oI non-experts (such as people
chosen at random in the street). Consider - what types oI experts would you choose iI you were
trying to Iorecast what the world will be like in 50 years time?
In a Delphi study the experts are all consulted separately to avoid some oI the bias that might
result were they all brought together, e.g. domination by a strong willed individual, divergent
(but valid) views not being expressed Ior Iear oI humiliation.
A typical question might be "In what year (iI ever) do you expect automated rapid transit to have
become common in major cities in Europe?". The answers are assembled in the Iorm oI a
distribution oI years, with comments attached, and recirculated to provide revised estimates. This
process is repeated until a consensus view emerges. Plainly such a method has many deIiciencies
but on the other hand is there a better way oI getting a view oI the Iuture iI we lack the relevant
data (numbers) which would be needed iI we were to apply some oI the more quantitative
As an example oI this there was a Delphi study published in $cience Journ,l in October 1967
which tried to look Iorward into the Iuture (now, oI course, we are many years past 1967 so we
can see how well they Iorecast). any questions were asked as to when something might happen
and a selection oI these questions are given below. For each question we give the upper quartile
answer, the time by which 75 oI the experts believed something would have happened.
O AuLomaLed rapld LranslL upper quarLlle answer 1983 le 73 of Lhe experLs asked ln 1967
LhoughL LhaL by 1983 Lhere would be wldespread auLomaLed rapld LranslL ln mosL urban areas
Lell LhaL Lo anyone who llves ln London!
O Wldespread use of sophlsLlcaLed Leachlng machlnes upper quarLlle answer 1990 le 73 of Lhe
experLs asked ln 1967 LhoughL LhaL by 1990 Lhere would be wldespread use of sophlsLlcaLed
Leachlng machlnes Lell LhaL Lo anyone who works ln a uk school/unlverslLy
O Wldespread use of roboL servlces upper quarLlle answer 1993 le 73 of Lhe experLs asked ln
1967 LhoughL LhaL by 1993 Lhere would be wldespread use of roboL servlces
It is clear that these Iorecasts, at least, were very inaccurate. Indeed looking over the Iull set oI
Iorecasts many oI the 25 Iorecasts made (about all aspects oI liIe/society in the Iuture aIter 1967)
were wildly inaccurate.
%his brings us to our first key point, we are interested in the difference between the
original forecast and the final outcome, i.e. in forecast error.
However, back in 1967 when this Delphi study was done, what other alternative approach did we
have iI we wished to answer these questions?
In many respects the issue we need address with regard to Iorecasting is not whether a particular
method gives good (accurate) Iorecasts but whether it is the best available method - iI it is then
what choice do we have about using it?
%his brings us to our second key point, we need to use the most appropriate (best)
forecasting method, even if we know that (historically) it does not give accurate forecasts.

Regrexxlon methoJx
ou have probably already met linear regression where a straight line oI the Iorm a bX is
Iitted to data. It is possible to extend the method to deal with more than one independent variable
X. Suppose we have k independent variables X
, X
, ..., X
then we can Iit the regression line
a b
... b

This extension to the basic linear regression technique is known as 2ultiple regression. Plainly
knowing the regression line enables us to Iorecast given values Ior the X

ultlple equutlon methoJx
ethods oI this type are Irequently used in economic modelling (econo2etrics) where there are
many dependent variables that interact with each other via a series oI equations, the for2 oI
which is given by economic theory. This is an important point. Economic theory gives us some
insight into the basic structural relationships between variables. The precise numeric relationship
between variables must oIten be deduced by examining data.
As an example consider the Iollowing simple model, let:
O x personal lncome
O ? personal spendlng
O l personal lnvesLmenL
O r lnLeresL raLe
From economic theory suppose that we have
= a
+ b
) (spending a linear function of disposable income)
I = a
+ b
r (investment linearly related to the interest rate)
and the balancing equation
X = + I (income = spending + investment)
where a
are constants.
Here we have 3 equations in 4 variables (X,,I,r) and so to solve these equations one oI the
variables must be given a value. The variable so chosen is known as an exogenous variable
because its value is determined outside the system oI equations whilst the remaining variables
are called endogenous variables as their values are determined within the system oI equations,
e.g. in our model we might regard the interest rate r as the exogenous variable and be interested
in how X, and I change as we alter r.
Usually the constants a
are not known exactly and must be estimated Irom data (a
complex procedure). Note too that these constants will probably be diIIerent Ior diIIerent groups
oI people, e.g. urban/rural, men/women, single/married, etc.
An example oI an econometric model oI this type is the UK Treasury model oI the economy
which contains many variables (each with a time subscript), complicated equations, and is used
to look at the eIIect oI interest rate changes, tax changes, oil price movements, etc.
For example the UK Treasury equation |New Scientist, 31st October 1993| to predict consumer
spending looks like:
-0.018 0.0623DDlog
- 0.00448log



)| etc
O L Llme perlod (quarLer) ln quesLlon
O u change ln varlable beLween Lhls quarLer and lasL quarLer
O C consumer nondurable spendlng for Lhe quarLer ln quesLlon
O u unemploymenL raLe
O ? real dlsposable lncome ad[usLed for lnflaLlon loss on flnanclal asseLs
O lnflaLlon lndex for LoLal consumer spendlng
O nlW neL flnanclal asseLs of Lhe personal secLor
O W gross physlcal wealLh of Lhe personal secLor
II you click here you will Iind a model that enables you to play with the UK economy.
Historically econometric techniques/methods tend to have large Iorecast errors when Iorecasting
national economies in the medium-term. However recall one oI our key points above: we need to
use the most appropriate (best) forecasting method, even if we know that (historically) it
does not give accurate forecasts. It can be argued that such techniques are the most
appropriate/best way oI making economic Iorecasts.

1lme xerlex methoJx,unulyxlx
ethods oI this type are concerned with a variable that changes with time and which can be said
to depend only upon the current time and the previous values that it took (i.e. not dependent on
any other variables or external Iactors). II
is the value oI the variable at time t then the
equation Ior

, ...,
, t)
i.e. the value oI the variable at time t is purely some Iunction oI its previous values and time, no
other variables/Iactors are oI relevance. The purpose oI time series analysis is to discover the
nature oI the Iunction I and hence allow us to Iorecast values Ior
Time series methods are especially good Ior short-term Iorecasting where, within reason, the past
behaviour oI a particular variable is a good indicator oI its Iuture behaviour, at least in the short-
term. The typical example here is short-term demand Iorecasting. Note the diIIerence between
demand and sales - demand is what customers want - sales is what we sell, and the two may be
In graphical terms the plot oI
against t is as shown below.

The purpose oI the analysis is to discern some relationship between the
values observed so Iar
in order to enable us to Iorecast Iuture
values. We shall deal with two techniques Ior time
series analysis in detail and brieIly mention a more sophisticated method.

ovlng uveruge
One, very simple, method Ior time series Iorecasting is to take a 2oving ,ver,ge (also known as
weighted moving average).
The moving average (m
) over the last L periods ending in period t is calculated by taking the
average oI the values Ior the periods t-L1, t-L2, t-L3, ..., t-1, t so that



To Iorecast using the moving average we say that the Iorecast Ior all periods beyond t is just m

(although we usually only Iorecast Ior one period ahead, updating the moving average as the
actual observation Ior that period becomes available).
Consider the Iollowing example: the demand Ior a product Ior 6 months is shown below -
calculate the three month moving average Ior each month and Iorecast the demand Ior month 7.
Month 1 2 3 4 5 6
Demand (100's) 42 41 43 38 35 37
Now we cannot calculate a three month moving average until we have at least 3 observations -
i.e. it is only possible to calculate such an average Irom month 3 onward. The moving average
Ior month 3 is given by:
(42 41 43)/3 42
and the moving average Ior the other months is given by:
(41 43 38)/3 40.7
(43 38 35)/3 38.7
(38 35 37)/3 36.7
We use m
as the Iorecast Ior month 7. Hence the demand Iorecast Ior month 7 is 3670 units.
The package input Ior this problem is shown below.

The output Irom the package Ior a three month moving average is shown below.

Chooxlng between forecuxtx
One problem with this Iorecast is simple - how good is it? For example we could also produce a
demand Iorecast Ior month 7 using a two month moving average. This would give the Iollowing:
(42 41)/2 41.5
(41 43)/2 42
(43 38)/2 40.5
(38 35)/2 36.5
(35 37)/2 36
Would this Iorecast (m
3600 units) be better than our current demand Iorecast oI 3670 units?
Rather than attempt to guess which Iorecast is better we can approach the problem logically. In
Iact, as will become apparent below, we already have sufficient information to make a logical
choice between forecasts if we look at that information appropriately.
In an attempt to decide how good a Iorecast is we have the Iollowing logic. Consider the three
month moving average given above and pretend Ior a moment that we had only demand data Ior
the Iirst three months, then we would calculate the moving average Ior month 3 (m
) as 42 (see
above). This would be our forec,st Ior month 4. But in month 4 the outco2e is actually 38, so we
have a diIIerence (error) deIined by:
O ettot fotecostootcome 4238 4
Note here that we could equally well deIine error as outco2e-forec,st. That would just change
the sign oI the errors, not their absolute values. Indeed note here that iI you inspect the package
output you will see that it does just that.
In month 4 we have a Iorecast Ior month 5 oI m
40.7 but an outcome Ior month 5 oI 35
leading to an error oI 40.7-35 5.7.
In month 5 we have a Iorecast Ior month 6 oI m
38.7 but an outcome Ior month 6 oI 37
leading to an error oI 38.7-37 1.7.
Hence we can construct the table below:
Month 1 2 3 4 5 6 7
Demand (100's) 42 41 43 38 35 37 .
Forecast - - - m
- - - 42 40.7 38.7 36.7
Error - - - 4 5.7 1.7 .
Constructing the same table Ior the two month moving average we have:
Month 1 2 3 4 5 6 7
Demand (100's) 42 41 43 38 35 37 .
Forecast - - m
- - 41.5 42 40.5 36.5 36
Error - - -1.5 4 5.5 -0.5 .
Comparing these two tables we can see that the error terms give us a measure oI how good the
Iorecasting methods (two or three month moving average) would have been had we used them to
Iorecast one period (month) ahead on the historical data that we have.
In an ideal world we would like a Iorecasting method Ior which all the errors are zero, this would
give us conIidence (probably a lot oI conIidence) that our Iorecast Ior month 7 is likely to be
correct. Plainly, in the real world, we are hardly likely to get a situation where all the errors are
zero. It is genuinely diIIicult to look at (as in this case) two series oI error terms and compare
them. It is much easier iI we take some Iunction oI the error terms, i.e. reduce each series to a
single (easily grasped) number. One suitable Iunction Ior deciding how accurate a Iorecasting
method has been is:
O ovetoqe spooteJ ettot
The logic here is that by squaring errors we remove the sign ( or -) and discriminate against
large errors (being resigned to small errors but being adverse to large errors). Ideally average
squared error should be zero (i.e. a perIect Iorecast). In any event we preIer the Iorecasting
method that gives the lowest average squared error.
We have that Ior the three month moving average:
O average squared error 4 + 37 + 17/3 1713
and Ior the two month moving average:
O average squared error (13) + 4 + 33 + (03)/4 1219
The lower oI these two Iigures is associated with the two month moving average and so we
preIer that Iorecasting method (and hence preIer the Iorecast oI 3600 Ior month 7 produced by
the two month moving average).
Average squared error is known technically as the 2e,n squ,red devi,tion (M$D) or 2e,n
squ,red error (M$).
ote here that we have actually done more than distinguish between two different forecasts
(i.e. between two month and three month moving average). We now have a criteria for
distinguishing between forecasts, however they are generated - namely we prefer the
forecast generated by the technique with the lowest MSD (historically the most accurate
forecasting technique on the data had we applied it consistently across time).
This is important as we know that even our simple package contains many diIIerent methods Ior
time series Iorecasting - as below.

"uestion - do you think that one of the above forecasting methods ALWAYS gives better
results than the others or not?

lngle exponentlul xmoothlng
One disadvantage oI using moving averages Ior Iorecasting is that in calculating the average all
the observations are given equal weight (namely 1/L), whereas we would expect the more recent
observations to be a better indicator oI the Iuture (and accordingly ought to be given greater
weight). Also in moving averages we only use recent observations, perhaps we should take into
account all previous observations.
One technique known as exponential smoothing (or, more accurately, single exponential
smoothing) gives greater weight to more recent observations ,nd takes into account all previous
DeIine a constant where 0 1 then the (single) exponentially smoothed moving average
Ior period t (
say) is given by


(1- )
(1- )
(1- )
So you can see here that the exponentially smoothed moving average takes into account all oI the
previous observations, compare the moving average above where only a Iew oI the previous
observations were taken into account.
The above equation is diIIicult to use numerically but note that:


(1- )|
(1- )
(1- )

(1- )

Hence the exponentially smoothed moving average Ior period t is a linear combination oI the
current value (
) and the previous exponentially smoothed moving average (
The constant is called the s2oothing const,nt and the value oI reIlects the weight given to
the current observation (
) in calculating the exponentially smoothed moving average
period t (which is the Iorecast Ior period t1). For example iI 0.2 then this indicates that 20
oI the weight in generating Iorecasts is assigned to the most recent observation and the remaining
80 to previous observations.
Note here that

(1- )
can also be written

- (

) or current
Iorecast previous Iorecast - (error in previous Iorecast) so exponential smoothing can be
viewed as a Iorecast continually updated by the Iorecast error just made.
Consider the Iollowing example: Ior the demand data given in the previous section calculate the
exponentially smoothed moving average Ior values oI the smoothing constant 0.2 and 0.9.
We have the Iollowing Ior 0.2.


42 (we always start with


0.2(41) 0.8(42) 41.80

0.2(43) 0.8(41.80) 42.04

0.2(38) 0.8(42.04) 41.23

0.2(35) 0.8(41.23) 39.98

0.2(37) 0.8(39.98) 39.38
Note here that it is usually suIIicient to just work to two or three decimal places when doing
exponential smoothing. We use
as the Iorecast Ior month 7, i.e. the Iorecast Ior month 7 is
3938 units.
We have the Iollowing Ior 0.9.



0.9(41) 0.1(42) 41.10

0.9(43) 0.1(41.10) 42.81

0.9(38) 0.1(42.81) 38.48

0.9(35) 0.1(38.48) 35.35

0.9(37) 0.1(35.35) 36.84
As beIore
is the Iorecast Ior month 7, i.e. 3684 units.
The package output Ior 0.2 is shown below.

The package output Ior 0.9 is shown below.

In order to decide the best value oI (Irom the two values oI 0.2 and 0.9 considered) we choose
the value associated with the lowest SD (as above Ior moving averages).
For 0.2 we have that
O ,u (4241)+(418043)+(420438)+(412333)+(3998 37)/3 1329
For 0.9 we have that
O ,u (4241)+(411043)+(428138)+(384833)+(3333 37)/3 832
Note here that these SD values agree (to within rounding errors) with the SD values given in
the package output above.
Hence, in this case, 0.9 appears to give better Iorecasts than 0.2 as it has a smaller value oI
Above we used SD to reduce a series oI error terms to an easily grasped single number. In Iact
Iunctions other than SD such as:
O ,Au (mean absoluLe devlaLlon) average | error |
O blas (mean error) average error also know as CumulaLlve lorecasL Lrror
exist which can also be used to reduce a series oI error terms to a single number so as to judge
how good a Iorecast is.
For example, as can be seen in the package outputs above, the package gives a number oI such
Iunctions, deIined as:
In Iact methods are available which enable the optimal value oI the smoothing constant (i.e. the
value oI which minimises the chosen criteria oI Iorecast accuracy, such as mean squared
deviation (SD)) to be easily determined. This can be seen below where the package has
calculated that the value oI which minimises SD is 0.86 (approximately).

Note here that the package can be used to plot both the data and the Iorecasts as generated by the
method chosen. Below we show this Ior the output above (associated with the value oI which
minimises SD oI

Note here that the choice oI criterion can have a large eIIect on the value oI e.g. Ior our
example the value oI which minimises AD is 0.59 (approximately) and the value oI
which minimises bias is 1.0 (approximately).
To illustrate the change in AD, bias and SD as changes we graph below AD and bias
against the smoothing constant ,

and below SD against .

Below we graph the value oI the Iorecast against . One particular point to note is that, Ior this
example, Ior a relatively wide range oI values Ior the Iorecast is stable (e.g. Ior 0.60
1.00 the Iorecast lies between 36.75 and 37.00). This can be seen below - the curve is "Ilat" Ior
high values.

Note here that the above graphs imply that in Iinding a good value Ior the smoothing constant it
is not usually necessary to calculate to a very high degree oI accuracy (e.g. not to within 0.001
Ior example).

ore uJvunceJ tlme xerlex forecuxtlng
Time series Iorecasting methods more advanced than those considered in our simple package do
exist. These are based on AutoRegressive Integrated Moving Average (ARIMA) models.
Essentially these assume that the time series has been generated by a probability process with
Iuture values related to past values, as well as to past Iorecast errors. To apply ARIA models
the time series needs to be stationary. A stationary time series is one whose statistical properties
such as mean, variance and autocorrelation are constant over time. II the initial time series is not
stationary it may be that some Iunction oI the time series, e.g. taking the diIIerences between
successive values, is stationary.
In Iitting an ARIA model to time series data the Iramework usually used is a Box-Jenkins
approach. It does however have the disadvantage that whereas a number oI time series
techniques are Iully automatic, in the sense that the Iorecaster has to exercise no judgement other
than in choosing the technique to use, the Box-Jenkins technique requires the Iorecaster to make
judgements and consequently its use requires experience and "expert judgement" on the part oI
the Iorecaster. Some Iorecasting packages do exist that make these "expert choices" Ior
you.Himalaya Publishing House
Demand Forecasting
Demand Forecasting using QuaIitative & Quantitative methods
After gathering data from the primary and secondary sources, the analysts then attempt to
forecast the demand levels in the future. The tools that are available for forecasting can be
divided into three broad categories, which are explained in detail below:
Qualitative Methods
These methods rely on experts who try to quantify the level of demand from the available qualitative data.
The two most widely followed methods are:
O Jury of execution opinion method: Opinions of a group of experts is called for and these are then
combined to arrive at the estimated demand.
O Delphi Method: n this method a group of experts are sent questionnaires through mail. The
responses received are summarised without disclosing the identities. Further mails are sent for
clarification in cases of extreme views. The process is repeated till the group reaches to a
reasonable agreement.
Quantitative Methods
These methods forecast demand levels based on analysis of historical time series. The important
methods in this category are:
%7end p7ojection methods
These methods involve determining the trend of consumption based on past consumption and project
future consumption by extrapolating this trend. The trend relations may be represented in one of the
following ways:
O Linear relationship Y
= a + b t
O Exponential relationship Y
= a e

O Polynomial relationship Y
= a
+ a
t + a
+.+ a

O Cobb Douglas relationship Y
= a t

n the above relationships Y
represents the demand for the year t, a and b are constants.
ponential smoothening method n this method, forecasts are modified whenever errors are
observed. For example, if the forecast value for the year t, F
, is less than the actual value for the year S
the forecast value for the year F
is set more than F
. n general, F
- e

a Smoothening parameter (value lies between 0 and 1) ; F
Forecast for the year t+1

error in the forecast for the year t = F
- S

Moving Ave7age Method
According to this method, the forecast for the next period represents a simple or weighted
arithmetic average of the last few observations.
What is a demand forecast?
A demand Iorecast is the prediction oI what will happen to your company's existing product
sales. It would be best to determine the demand Iorecast using a multi-Iunctional approach. The
inputs Irom sales and marketing, Iinance, and production should be considered. The Iinal
demand Iorecast is the consensus oI all participating managers. ou may also want to put up a
Sales and Operations Planning group composed oI representatives Irom the diIIerent departments
that will be tasked to prepare the demand Iorecast.
Determination oI the demand Iorecasts is done through the Iollowing steps:
Determine the use oI the Iorecast
Select the items to be Iorecast
Determine the time horizon oI the Iorecast
Select the Iorecasting model(s)
Gather the data
ake the Iorecast
Validate and implement results
The time horizon oI the Iorecast is classiIied as Iollows:
Description Forecast Horizon
Short-range Medium-range Long-range

Duration Usually less than 3
months, maximum oI
1 year
3 months to 3 years ore than 3 years
Applicability Job scheduling,
worker assignments
Sales and production
planning, budgeting
New product
Iacilities planning
How is demand forecast determined?
There are two approaches to determine demand Iorecast (1) the qualitative approach, (2) the
quantitative approach. The comparison oI these two approaches is shown below:
Description "ualitative Approach "uantitative Approach
Applicability Used when situation is vague &
little data exist (e.g., new products
and technologies)
Used when situation is stable &
historical data exist
(e.g. existing products, current
Considerations Involves intuition and experience Involves mathematical techniques
Techniques Jury oI executive opinion
Sales Iorce composite
Delphi method
Consumer market survey
Time series models
Causal models

"ualitative Forecasting Methods
our company may wish to try any oI the qualitative Iorecasting methods below iI you do not
have historical data on your products' sales.
"ualitative Method Description
1ury of executive
The opinions oI a small group oI high-level managers are
pooled and together they estimate demand. The group uses
their managerial experience, and in some cases, combines
the results oI statistical models.
Sales force composite Each salesperson (Ior example Ior a territorial coverage) is
asked to project their sales. Since the salesperson is the one
closest to the marketplace, he has the capacity to know what
the customer wants. These projections are then combined at
the municipal, provincial and regional levels.
Delphi method A panel oI experts is identiIied where an expert could be a
decision maker, an ordinary employee, or an industry expert.
Each oI them will be asked individually Ior their estimate oI
the demand. An iterative process is conducted until the
experts have reached a consensus.
Consumer market
The customers are asked about their purchasing plans and
their projected buying behavior. A large number oI
respondents is needed here to be able to generalize certain
Quantitative Forecasting Methods
There are two Iorecasting models here (1) the time series model and (2) the causal model. A
time series is a s et oI evenly spaced numerical data and is o btained by observing responses at
regular time periods. In the time series model , the Iorecast is based only on past values and
assumes that Iactors that inIluence the past, the present and the Iuture sales oI your products will
On the other hand, t he causal model uses a mathematical technique known as the regression
analysis that relates a dependent variable (Ior example, demand) to an independent variable (Ior
example, price, advertisement, etc.) in the Iorm oI a linear equation. The time series Iorecasting
methods are described below:

%ime Series
ave Approach Assumes that demand in the next period is the same as demand in
most recent period; demand pattern may not always be that stable
For example:
1 JuIy saIes were 50, then Augusts saIes wiII aIso be 50

%ime Series
Moving Averages
A is a series oI arithmetic means and is used iI little or no trend is
present in the data; provides an overall impression oI data over time
A simple moving average uses average demand Ior a Iixed
sequence oI periods and is good Ior stable demand with no
pronounced behavioral patterns.
F 4 D 1 + D2 + D3] / 4
F Iorecast, D Demand, No. Period
(see illustrative example - simple moving average)
A weighted moving average adjusts the moving average method to
reIlect Iluctuations more closely by assigning weights to the most
recent data, meaning, that the older data is usually less important.
The weights are based on intuition and lie between 0 and 1 Ior a
total oI 1.0
WMA 4 (W) (D3) + (W) (D2) + (W) (D1)
WA Weighted moving average, W Weight, D Demand, No.
(see illustrative example - weighted moving average)
The exponential smoothing is an averaging method that reacts
more strongly to recent changes in demand by assigning a
smoothing constant to the most recent data more strongly; useIul iI
recent changes in data are the results oI actual change (e.g., seasonal
pattern) instead oI just random Iluctuations
F t + 1 a D t + (1 - a ) F t
F t 1 the Iorecast Ior the next period
D t actual demand in the present period
F t the previously determined Iorecast Ior the present period
a weighting Iactor reIerred to as the smoothing constant
(see illustrative example - exponential smoothing)
%ime Series
The time series decomposition adjusts the seasonality by
multiplying the normal Iorecast by a seasonal Iactor
(see illustrative example - time series decomposition)