Demand Forecasting - II

4104- Chavda Trusha

4117-Gohel Vishal
4128-Makwana Umesh
4139-Parmar Kuldip
4148-Patel Ishita
4151-Patel Keny
4161- Prajapati Tanisha
4182- Solanki Urvesh
4184-Uttekar Diksha

Division - B
MECHANICAL EXTRAPOLATION (OR, TREND PROJECTION) TECHNIQUE

• These methods are based on the past sales pattern.

• It is also known as Time series analysis or Trend Projection method.
• It is based on the assumption that future events are a continuation of the past and historical data can be
used to predict the future.
• It is dispenses with the need for costly market research because the necessary data is already available.
• Simple, inexpensive and quick methods of forecasting.
• These methods yield reasonably accurate results so long as the trend of the data has a persistent tendency to
move in the same direction. But if the data shows significant turns, then the time series analysis may not give
acceptable results. Thus, this analysis is more reliable for forecasting over a short period.
• With the help of time series analysis, we can get the following kinds of predictions :
(a) Finding a trend value for a specific year (or, years)
(b) Finding seasonal fluctuations in the variable, and
(c) Predicting turning points in future movement of the variable.
 Reasons for Fluctuations in Time Series Data
• If the time series data is plotted on a graph, it will show fluctuations and wave–like patterns over time.
• These are usually caused by :Trend (T), Cyclical changes (C), Seasonal changes (S), and Irregular
movements (I).
• In other words, the event of variations in time series is composed of four components : T ; C, S and I.

(1) Trend in the movement of time series is the overall direction of the movement of the variable over a
long period.
• For example , the trend in operating revenues of India Railways from 1977 to 1998.
• Long-term trends generally exist because some of the limited factors, such as population, national
income, competitive conditions, etc. Move steadily and therefore produce only a gradual change over
time.
• The minimum period of a cycle is greater than one year.

(2) The cycle component of a time series is a wavelike, repetitive movement fluctuating about the trend of
the time series.
• For example, depression, recovery, prosperity and recession in general business activity repeat themselves
over time.
• cyclical effects in a time series can be caused by changes in the stockpiling or shortage of commodities,
business expectations, weather, government policy, etc.
• Identification of cycle helps in :
(i) locating turning point of the movement of the variable; and
(ii) making intermediate-range forecasts

(3) The Seasonal movement of time series are those repetitive, fluctuating movements that always occur at a
particular time of the year.
• For example ,sales of woollen clothes are higher during October to January because of the winter season.
• These enable management in making short-term forecasts.

(4) The Random variations or Irregular variations.: These are those variations that are left over after isolating
the other components.
• For example, sudden onset of an epidemic may push up the sales of drug manufacturers.
 Decomposition of Times Series
• The four components of time series related to each other in an additive or multiplicative form :
O = T + S + C + I (Additive Model)
O = T × S × C × I (Multiplicative Model)
where O = Original data
T = Trend
S = Seasonal component
C = Cyclical component and
I = Irregular fluctuations.
• Most of the series relating to business and economics are of multiplicative nature.
Multiplicative model can be converted into additive model, by taking logarithms as :
log O = log T + log S + log C + log I.
 Method of finding Trends
There are 5 trend projection methods

Trend Projection
Methods

Graphic (Fitting
Algebraic (Least Smoothing ARIMA
trend line by Statistical
squares) Techniques Technique
observation

Logarithmic or Moving Average Exponential

Semi averages Straight Line Parabolic
Exponential method Smoothing
1. Fitting Trend line by observation ( Visual time Series Projection)

• This method of estimating trend is elementary, easy and quick.

• It involves merely the plotting of annual sales on a graph, and then estimating just by observation where the
trend line lies.
• The line can then be simply extended to a future period
• Corresponding sales forecast read against that year.
Types of graphs:-
(1) The familiar graph paper with arithmetic scale on both axes.
(2) The graph paper with semi-logarithmic scale
• i.e., having arithmetic scale along the horizontal axis, but having logarithmic conversion of the numbers along
the vertical axis.
• These trends, known respectively as linear and exponential, can be tried and the one that fits more closely to the
data be used for forecasting.
2. Time Series Analysis employing Least Squares Method:-
• This technique is most widely used in practice.
• It is a statistical method With its help a trend line is fitted to the data.
• This line is known as ‘The Line of Best Fit.’
• By extrapolating the trend line for future we can get the corresponding figures of forecasted sales.
• Only one independent variable: time.
• This system of forecasting is considered “naive”.
Linear Trend:-
• The method of least squares may be used either to fit a straight line trend or a non-linear trend.
• The straight line trend represented by the equation:
Sales = a+b (year number)
Or
S = a + b •T
• where a = the constants representing the intercept and b= slope of estimated straight line.
• Equations to determine the values of a & b:
∑S = Na + b∑T
∑ST = a∑T + b∑T²
Where N = no. Of Years.
Illustration 1 :- The following data refer to sales , in thousands of rupees (x), of a certain product during five
years :
Assuming the present trend continues , in which year will you except 1994 sales to be doubled ?

Year Sales (S)

1993 605

Solution :-
1994 715
a=594.5
b = 53.5 1995 830
t= 15.617 (Year 2009 which is 16th year from 1993)
1996 790

1997 835
Non-linear Trends:

The trend equation given above assumes that there is a linear (or, proportional) change is sales over time. In
fact, the trend equation can take a linear or a non-linear form.
The non-linear trends can be parabolic, exponential, geometric.
(1) Second and higher degree polynomial trends:
Second degree or quadratic trend : S = a-bT+cT²
Third degree or cubic trend : S = a-bT+cT²-dT³
• Similarly, build up trend equations for polynomials of degree higher than three, but they are seldom used
in business forecasting in practice.
• A second-degree polynomial gives us a quadratic trend while with Third degree polynomial we get cubic
trend.
(2) Exponential or semi-log trend: S= a e bt
log S = log a + bt
This trend equation assumes a constant growth rate in variable S over time.
(3) Double-log trend : S = aTb
log S = log a + b log T
Double-log function assumes that the value of elasticity of S remains
constant over time.
 The empirical appropriateness of linear and non-linear trend can be assessed with the help of the
method of first and second differences.

•First differences of successive observations are

nearly constant then the straight line trend curve is
most appropriate.
• Alternatively, a second-degree of quadratic trend
curve may be considered.
•The appropriateness of fitting a quadratic trend
curve to the given data can be tested by the
method of second
differences .
•If second differences are almost constant, a
second-degree or quadratic trend curve is most
appropriate.
• When a given time series is increasing or
decreasing at a constant rate (not constant amount)
it could be more
appropriate to fit exponential trend curve.
3) Forecasting Through Decomposing a Time Series.
• A time series is composed of trend, seasonal fluctuations, cyclical movements and irregular
• variations.
• If the available data is quarter-wise/month-wise, it is possible to identify the seasonal effect.
• If this data is available for a sufficiently long period of time, the trend and cyclical effects can also be found
out.
4) Smoothing Methods:
• Smoothing methods attempt to cancel out the effect of random variations on the values of the series.
• Once this effect is removed, a clearer indication of the underlying movement in the series is revealed.
• The two main smoothing methods are:
• (i) Moving average
• (ii)Exponential Smoothing
(i) Method of Moving Averages (M.A.)
• Moving averages consist of a series of arithmetic means calculated from overlapping groups of successive
values of a time series. Each moving average is based on values covering a fixed time interval, called ‘period
of moving average and is shown against the centre of the period. The composition of items is adjusted
successively by replacing the first value of the previously averaged group by the next value below that group.
Thus, for the time series values Y1, Y2, Y3 ,..., for different time periods, the moving average of period n is
given by:
• 1st value of moving average ( Y1 )= 1/n (Y1 +Y2 +...+Y n).
• 2nd value of moving average ( Y2 )= 1/n (Y2+Y3 +...+Y n+1 ).
• 3rd value of moving average and so on. ( Y3 )= 1/n (Y3 +Y4 +...+Y n+2 ), and so on

Centred Moving Average

• When the period of M.A. Is odd , the moving averages are placed against the mid-value of the period.
• When the period of M.A. Is even , the moving averages are placed between the two central values of the
period.
• First centred moving average = ½ (Y₁′ + Y₂′)
• Second centred moving average = ½ (Y₂′ + Y₃′) , and so on.
(ii) Exponential Smoothing
 Another smoothing technique, called exponential smoothing, is a very popular approach for short-term
forecasting. This method determines values by computing exponentially weighted stem.
 The weights assigned to each value reflect the degree of importance of that value. More recent values being
more relevant for forecasting, these are assigned greater weight than previous period values. It may be noted
that weights (w) are so assigned that w lies between zero and unity (0 <= w <= 1).
 The smoothing scheme begins by setting smoothened value equals to observed value for the period (t = 1). F1
= A1 .
 The smoothed average value for any subsequent time period is found with help of the equation:
 Smoothing Formula:
 Fₜ = wAₜ + (1 – w) Ft-1
Where:
Fₜ = Forecast for next period (new smoothed value)
Aₜ = current observed value
Ft-1 = Forecast for current period (previous smoothed value
(5) ARIMA METHOD
• The ARIMA method (Auto-Regressive Integrated Moving Average) was
developed by Box and Jenkin.
• This method is highly suitable to situations where the inherent pattern in the underlying series is highly
complex and difficult to understand.
• This method combines smoothing method with auto-regressive method.
• It can be used primarily for short-term forecasting.
• It combines the smoothing method and the auto-regressive method.
Five Stages of the Analysis in this method:-
1. Removal of the Trend
2. Model Identification.
3. Parameter Estimation
4. Verification
5. Forecasting
Barometric (Or Leading Indicator) Technique
Barometric technique is based on the presumption that a relationship can exist among various time
series.
For e.g, Industrial Production overtime and industrial loans by commercial banks over time may move
in the same direction.
There are three kinds of relationships among economic time series:
1) Leading Series: A leading series consists of the data that move ahead of the series being compared.
e.g., Application for the amount of housing loan over time is a leading series for the demand of
construction material; birth rate of children is the leading series for demand of seats in schools.
2) Coincident Series: When data in series moves up and down along with some other series, it is
known as a coincident series. For e.g, a series of data on national income is often coincident with the
series of employment in an economy (over a short period).
3) Lagging Series: Where data moves up and down behind the series being compared. For e.g, Data on
Industrial wages overtime is a lagging series when compared with series of price index of Industrial
Workers.
Comparison (Barometric Technique vs. Mechanical Extrapolation)
Point Barometric Technique Mechanical Extrapolation
Basis of Forecasting Uses current economic events Assumes the future will simply
and indicators to predict future continue the past patterns.
trends.
Data Used Statistical indicators (leading, Past data and historical trends only.
coincident, lagging).
Nature of Prediction More realistic because it uses Less reliable when conditions
present signals. change.
Purpose Shows direction of change in Extends past values into the future.
economy or industry.
Limitations Should only be used as a Does not account for current
complement to other methods. economic changes.
1. Coincident Indicators and Lagging Indicators:
• These are those variables whose movement coincides with or falls behind general economic
activity or market trends.
• These are generally used to confirm or refute the validity of the forecasts arrived at through
the use of leading indicators.
• E.g., Gross National Product, Index of Industrial Production, Retail Sales, Number of
industrial employees, etc.

2. Leading Indicators:
• Those variables whose movement precedes the movement of some other related variable are
known as leading indicators.
• E.g., Loan Application with financial institutions, birth rate, enrolment in school and prices of
food grains.

Limitations:
1. To Locate the leading indicator for the variable whose forecast is being undertaken;
2. To estimate a mathematical/statistical relationship of leading indicator with the variable
under forecast;
3. To Find out of the forecasted values of the variable; and
4. If possible, to verify the validity of the forecast with the help of coincident indicators
and/or lagging indicators.
Example:
• The example explains how leading indicators are used for forecasting.
Here, X = number of children born in a town on a particular day, and Y = demand for school
seats (when those children turn 5 years old).
• Since children need school seats 5 years after birth, the birth rate (X) becomes a leading indicator
for the future demand for school seats (Y).
• The relationship is written as:
𝑌 =𝑓 𝑋
• where s = time lag (5 years).
• To find the best prediction, we test different lags (1 year, 2 years, 3 years, etc.) and choose the one that
fits best using statistical methods like least squares.
• Finally, if the best equation found is:
𝑌 = 121 + 6𝑋
• Then the forecast for the coming years can be made by plugging in values of the leading indicator. For
example:
𝑌 = 121 + 6𝑋
𝑌 = 121 + 6𝑋
and so on.
Limitations of using lead indicator method for forecasting:

1. It is not easy to identify suitable leading, lagging, or coincident indicators. In many

cases, a proper leading indicator may not even exist (e.g., demand for mangoes). Without
a good indicator, forecasting becomes impossible.

2. Even if an appropriate indicator is found out, changes in fashions and tastes may render
this indicator redundant over time.

3. The time gap (lead time) between the indicator and the actual event may be too small to
be useful. For example, if births lead to an increase in demand for tinned milk only
slightly over a month, it gives producers too little time to adjust production.

4. Leading indicators mainly tell us whether the variable will increase or decrease, but not
how much it will change. So, they cannot accurately predict the exact size of future
demand
3. Index Numbers:
• To Overcome some of the problems of leading indicator method, the following
methods were developed:
i) Diffusion indices
ii) Composite indicators

i) Diffusion indices:
• In this method, a group of leading indicators is selected. We check how many of these indicators
have risen or fallen during a certain period. For example, if 9 indicators are used and 6 show a rise,
the diffusion index is 66.7%.
• If more than 50% indicators rise → future increase is expected.
• If more than 50% fall → future decrease is expected.

 Limitations of Diffusion Indices

• The time gap (lead time) between indicator changes and actual business cycle changes is very
small, so forecasting becomes difficult.
• If more indicators are added to increase accuracy, interpretation becomes more confusing.
• This method can show only the direction (upward or downward movement), not the exact
amount of change.
• Diffusion indices should be used only as a supporting method, not the main forecasting tool.
2. Composite Indicators
Composite indicators are formed by combining several leading indicators. This method reduces the risk
of wrong predictions because random errors in different indicators balance each other.

Steps to Prepare Composite Indicators

Multiply each indicator

value by its weight, add
Assign weights to each
them, and divide by the
indicator based on: total weights to get the
1. its economic importance, composite index.
Remove the trend from 2. how strongly it moves the
each indicator series and main variable,
smooth the data using 3. consistency in timing,
Set of proper indicators the moving average
method. 4. smoothness of movement,
is identified for the
forecast. 5. how current or updated
the indicator is.
 Factor like competitors' advertising, one's own
STATISTICAL

advertising, own price, competitor's price, etc.
might have undergone a change and, therefore,
METHODS time will not be able to explain the movement in
sales.
 In such cases we may use statistical methods. One
may use statistical tools to
construct estimating equations, and tests can be
carried out to see whether or
not any observed association between past sales
and another variable is statistically significant.
These methods are also called economic methods.
1) Naive Models
 Naive models are often as effective as more sophisticated models, and are cheaper and
easier to use.
 They are generally useful where situations are stable or are experiencing only a
gradual change, so long as their results are fairly accurate they will be quite helpful.
 For example, in a stable situation one may use the ratio of advertising outlay and
sales of the past to forecast how future sales will react to future advertising outlays.
2. Correlation and Regression Method
 This is perhaps the most popular method of forecasting. Unlike time series analysis, the correlation
and regression analysis does not limit itself to “time” as the independent variable.
 It recognises the fact that sales depend upon factory other than time.
 Correlation Analysis:
 Correlation means covariance, or the two series varying together, e.g. Income and expenditure of
families. When the two series vary in the same direction, it is known ad positive Correlation (r>0),
while when these vary in the opposite direction, it is known as negative Correlation (r<0).
 Correlation Analysis can be of two types:
1. Simple or Partial Correlation
2. Multiple Correlation
Coefficient of Correlation
 To know the closeness of variables ( say X and Y ), we find correlation coefficient (r)
by using the following formula:
Regression Equation Method
 In the case the trend of the dependent variable is approximately linear we fit in a Linear equation like ,
Export of good X = a ( national income) + b ( domestic prices of X ) + c ( international
prices of X ) + d ( weather).
 It may be noted that X can be taken as independent variable while Y as dependent, and vice versa. In
the first case we will get regression of Y on X, while in the second case, it is regression of X on Y .
 Let us discuss the various regression equations which can be used for forecasting exercise.
1. Fitting Simple Linear Regression
 In this case a straight line is fitted to the data containing one dependent variable and only one
independent variable, e.g.,
 Sales = a+b .( price )
(i) Graphical Method
 In this Method, we plot the sets of data of the two variable ( dependent and independent variables ) on
a graph below. The regression line is then approximately by sketching it freehand in such a manner
that the line passes through the middle of the scatter of points.
 (ii)Least Square Method

 In the least squares method of estimating regression line S = a +bP,we need to find the value of the
constans a and b with the help of the normal equations:
𝛴𝑆 = 𝑛𝑎 + 𝑏𝛴P
𝛴𝑆𝑃 = 𝑎𝛴𝑃 + 𝑏𝛴P2
• By simplifying (1) and(2), we get the regression coefficient of S on P and the intercept of the
regression equation (a) :

bSP = n𝛴SP-𝛴S𝛴P/n𝛴P2-(𝛴p)2 and a= 𝛴S-b𝛴p/n

• By substituting the calculated values of constants a and b in the regression equation
• S= a + bp, we get the required regression equation to forcast S. We can also find regression equation
by either of the following two formula :
1. (S -𝑠̅̇ ) = bSP (P-𝑝̅ )
2. (S -𝑠̅̇ ) = r * 𝜎 / 𝜎 (P−𝑝̅ )
Illustartion :
bSP = n𝛴SP-𝛴S𝛴P/n𝛴P2-(𝛴p)2

12(7,098)-(240)(384)/12(5,708)-(240)2

-6984/10,896
= - 0.641

a= 𝛴S-b𝛴p/n
= 384 –(240) (-0.641)/12
= 44.82

So the equation of the regression line would be,

S = 44.82 – 0.641 P.

By assuming value 30 to P in the regression equation , we can get the corresponding estimated sales level :

S ( when P is 30 0= 44.82- 0.641 * 30

= 25.29(‘000 units).
2. Fitting Non-linear Regression Equation
(i) Logarithmic Model:
Let us plot the data given in illustration 6 on (a) the usual arithmetic scale and (b) the semi-logarithmic
scale graph papers. If we find that the freehand trend line in the first graph exhibits a curved trend, while
in the second graph the price series over time exhibits a linear trend, then a log-linear model would be
more appropriate to fit.
The log-linear model would be of the form:
log S = a + bP
Where:
log S = logarithm of sales
P = Price
Note that one property of the logarithmic (or power) function of the form S = aPᵇ is that the exponent of
the independent variable is elasticity. For example, b is the price elasticity here.
(ii) Parabolic Regression Model

• Sometimes we need to fit a curved trend line which, by a change in variable, could not be reduced to a
linear form (unlike the logarithmic model). The curved line can be a second-degree polynomial, third-
degree polynomial, etc.
Let us assume it is a second-degree polynomial (the simplest and most popular), given by the equation:

S = a + bP + cP²

The statistics a, b, and c can be calculated from the set of normal equations:

ΣS = Na + bΣP + cΣP²
ΣSP = aΣP + bΣP² + cΣP³
ΣSP² = aΣP² + bΣP³ + cΣP⁴

For polynomials higher than the second degree, we can similarly write the set of normal equations to
find the values of the constants.
Illustration-7

The sales department of a firm is planning to expand sales to Rs. 10 lakh. The complaint to the sales department
points out that in the past this firm's sale proceeds and advertisement expenditure have a very high correlation
of +0.75.

The past data revealed that:

The firm's average sales per year has been Rs. 4 lakh with a variance of Rs. 30,000.
Its average annual advertisement expense has been Rs. 1 lakh with a variance of Rs. 10,000.
How much advertising expenditure must this firm therefore incur to achieve its sales target?
Solution Using the Regression Line Formula:
Let: The regression equation of Y on X is:
Average annual advertisement expense be Y (Y - Ȳ) = r (σᵧ / σₓ) (X - X̄)
Average annual sales be X Substituting the values:
We are to find the regression of Y on X. (Y - 1,00,000) = 0.75 × (10,000 / 30,000) × (X -
Given Data: 4,00,000)
X̄ = Rs. 4,00,000 (Y - 1,00,000) = 0.25 × (X - 4,00,000)
Ȳ = Rs. 1,00,000 Y = 0.25X - 1,00,000 + 1,00,000
σₓ = Rs. 30,000 Therefore, the regression equation is:
σᵧ = Rs. 10,000 Y = 0.25X
r = +0.75

Conclusion
If the target sales are X = Rs. 10,00,000, then the estimated advertisement expenditure is:
Y = 0.25 × (Rs. 10,00,000) = Rs. 2,50,000
Thus, to achieve the sales target of Rs. 10 lakh, the sales department should spend Rs. 2.5 lakh on
advertisement.
3. Multiple Regression Analysis

When more than one independent variable is used in a regression model, we perform multiple regression analysis.
The Multiple Regression Model
A multiple regression model for sales can be stated as:
Sales = a(Price) + b(Advertising) + c(Salesmen Visits) + d(Rivals' Price) + e(Disposable Income) + u
Where:
a, b, c, d, e are the partial regression coefficients.
Each coefficient shows the effect of its corresponding variable on sales, all other factors held constant.
For example, 'a' represents the percentage change in sales from a 1% change in price.
The constant u represents the effect of all variables not included in the equation but that still affect sales.
Alternative Model Form
• The regression equation can also be written in a multiplicative
form:
• Sales = (Price)ᵃ (Advertising)ᵇ (Salesmen Visits)ᶜ (Rivals'
Price)ᵈ (Disposable Income)ᵉ u
• Limitations of the Regression Method
• This method forecasts based on past data. If the future does not
reflect the past, the forecasts will be inaccurate.
• The accuracy of the forecast depends heavily on the accurate
measurement of the independent variables. Any error in them
reduces the reliability of the forecast.
Engaging the audience

• Make eye contact with your audience to create a sense of intimacy and involvement
• Weave relatable stories into your presentation using narratives that make your
message memorable and impactful
• Encourage questions and provide thoughtful responses to enhance audience
participation
• Use live polls or surveys to gather audience opinions, promoting engagement and
making sure the audience feel involved
Thank you
Brita Tamm
502-555-0152
brita@[Link]
[Link]