UNIT 7 DETERMINISTIC TIME SERIES AND

FORECASTING
Structure
7.0 Objectives
7.1 Introduction
7.2 Problems and Objects of Study of Time Series Data
7.2.1 Components of Time Series
7.2.2 Construction of Time Series: An Example
7.3 Measurement of Trend
7.3.1 Moving Average Method
7.3.2 Suitability of Moving Averages
7.3.3 Examples of Moving Averages
7.4 Method of Fitting Polynomials
7.4.1 Suitability of Least Squares Method
7.4.2 Examples of Least Squares Method
7.5 Monthly or Quarterly Trend Values from Annual Data
7.6 Measurement of Seasonal Variations
7.6.1 Method of Simple Average
7.6.2 Ratio to Trend Method
7.6.3 Ratio to Moving Average Method
7.7 Let Us Sum Up
7.8 Answers or Hints to Check Your Progress Exercises

7.0 OBJECTIVES
After going through this Unit, you will be able to
 construct a trend line for time series data;
 compute moving averages; and
 calculate various measures of seasonal variations.

7.1 INTRODUCTION
A time series is a set of observations on a variable measured at successive points
of time. Usually the variable values are recorded over equal time intervals-yearly,
quarterly monthly, etc. Generally the term ‘time series’ refers to economic data,
but it equally applies to quantitative data collected in other fields also. The time
series of National Income, Agricultural Income, and Agricultural Production are
based on annual observations.

Adapted from IGNOU study material of EEC 13: Elementary Statistical Methods and Survey Techniques,
Unit 11 written by S Bandopadhyay with modifications by K. Barik
Other examples of time series are yield of a crop in different years, population of Deterministic Time
Series and Forecasting
a country over different points of time, sales of a departmental store during
different seasons of the year, quarterly exports of tea, etc. For these types of data
one of the variables is time, denoted by tj and the other which is dependent on
time (such as yield, population, sale or export) is represented by yt. We will
analyse some of these series with the help of the methodology to be developed in
this Unit.

7.2 PROBLEMS AND OBJECTS OF STUDY OF

TIME SERIES DATA
A study of time series data reveals that in general the observed values of the
dependent variable (yt) change over time. These changes are due to interaction of
several forces such as increase in population, change in production techniques,
change of tastes and habits of people, variations in climate, etc. Part of these
changes is long term while others may be seasonal or cyclical. One of the main
objectives of study of time series data is to isolate and measure the effects of
various components. This analysis helps us in understanding the past behavior
and predicting the future. Such predication is of utmost importance to an
economist or a producer who can plan his production much ahead of sales.
7.2.1 Components of Time Series
A graphical representation of time series data reveals changes over time (in rather
exceptional cases the series exhibits no change during the period of observation).
However, these changes are not totally haphazard or random and at least a part of
it can be explained. Some of the movements are periodic in nature while some
others show persistent growth or decline. Along with these some unpredictable
movements, random in nature, are also found to be mixed up. Again, not every
series shows all the movements. It is assumed that the general series has four
important components.

i) Secular or long-term trend(T)

ii) Seasonal variation (S)

iii) Cyclical fluctuation (C)

iv) Irregular or random movement (I)

In the classical approach, it is assumed that the observed value yt may be

represented either as the product of the above components
i.e., yt = T × S ×C × I (multiplicative model)
or, as the sum of components
yt = T + S +C + I (additive model)

155
Summarisation of Although the additive model facilitates easier calculation, the multiplicative
Bivariate and Multi-
variate Data model has been most widely used in analysis of time series.
a) Secular Trend
By secular trend we mean the smooth, regular, long-term changes in the series
when observed over a period of time. Some series may exhibit an upward trend,
some series a downward trend while some others may remain more or less
constant over time. The upward trend of a series may be caused by factors such
as increase in population and improvement in techniques of production. For
example, the pattern of growth of many industries follows closely that of
population growth of the country. Again the advances in technology may give
rise to upward movement of most of the economics time series. But not all time
series will exhibit growth. Some may show decline while some others may show
fluctuations. The time series of crude death rates of a country is likely to show a
declining trend.
b) Seasonal Variations
The graphs of most of the time series reveal that a large number of fluctuations
are imposed on the trend. By seasonal variation we mean the periodic movement
in a time series where the period is not longer than one year. A periodic
movement is that which repeats at regular intervals or periods of time. For
example, the sales of cold drinks increase during summer and decrease during
winter, sales of garments are maximum during some seasons of the year, say
during May or festivals, the number of passengers carried by buses has a peak
during office hours, the number of books borrowed from a library has a peak
during some days of the week, etc. The factors which contribute to this type of
fluctuations are the climatic changes of different seasons, customs and habits
which people follow at different times.
C) Cyclical Fluctuations
By cyclical fluctuations we mean oscillatory movements of a time series, where
the period of oscillation, called the cycle, is more than a year. It includes those
factors leading to alternating periods of expansion and contraction that
characterize most economic and business series. Sometimes these fluctuations are
highly irregular with respect to their Sometimes these fluctuations are highly
irregular with respect to their shape, amplitude and direction. But the phenomena
they reflect – the periods of depression, recovery, boom and collapse-have been
observed in virtually all time series dealing with business and economics data.
d) Irregular Movement
The irregular movement includes component all factors not classifiable
elsewhere. Thus factors such as work stoppage, elections, wars, fire may affect a
particular time series; this category of movement includes all types of variations
not accounted for by secular trend, seasonal or cyclical fluctuations.
Unfortunately, factors of these kinds are frequently indistinguishable from

156
cyclical factors and as such in some discussions cyclical and irregular Deterministic Time
Series and Forecasting
components are combined together.
7.2.2 Construction of Time Series: An Example
As an illustration we prepare a time series according to the multiplicative model.
Table 7.1 presents trend, seasonal and cyclical-irregular components of a
hypothetical series.
Table 7.1: Hypothetical Time Series and its Components (Quarterly)
Year Components
Quarter Series Trend Seasonal Cyclical- Irregular
(yt) (T) (100S) (100CI)
I 79 80 120 82
1
II 58 85 80 85
III 84 90 92 102
IV 107 95 108 105

I 130 100 120 108

2
II 93 105 80 132
III 121 110 92 120
IV 161 115 108 130

I 216 120 120 150

3
II 132 125 80 132
III 150 130 92 125
IV 163 135 108 112

I 176 140 120 105

4
II 112 145 80 97
III 128 150 92 93
IV 142 155 108 85

I 134 160 120 70

5
II 86 165 80 65
III 94 170 92 60
IV 104 175 108 55

In Table 7.1 the series is represented by a multiplicative model, such that

yt = T × S ×C × I.

157
Summarisation of
Bivariate and Multi-
Thus the observation 79 (of I quarter of 1st year) = 80 ×
120 82
variate Data × .
100 100

Similarly, 112 (of II quarter of 4th year) = 145 ×

80 97
× .
100 100

Thus, each quarterly figure (yt) is the product of the secular trend (T), the
seasonal index (S), cyclical and the irregular component (CI). Such a synthetic
composition looks very much like an actual time series and has encouraged use
of the model as the basis for the analysis of time series data.

7.3 MEASUREMENT OF TREND

At times we are interested to know the trend movement in a time series. In such
circumstances, we have to eliminate the effects of other components (seasonal,
cyclical and irregular) from the series.
Two important methods of measuring trend are the ‘Moving Averages Method’,
and the ‘Method of Fitting Polynomials’. In moving average method, secular
trend is obtained by smoothing out fluctuations by the process of averaging. In
the latter, a polynomial of suitable degree is chosen either for the original
variables or for its transformed variable and its constants are determined by the
method of least squares. The choice of the degree of the polynomial can be made
by plotting the data on a graph paper where different scales, arithmetic, semi-
logarithmic or double-logarithmic scales may be used. Measurement of trend is
necessary for studying the behavior of the time series and for forecasting the
future.
7.3.1 Moving Averages Method
This is a simple method of smoothing out fluctuations of a series by calculating a
number of averages covering overlapping periods of the series. The first step
consists in selecting proper period of the moving average. If the period chosen
is 3 years, the moving averages are obtained by calculating a series of mean
values of three consecutive values covering overlapping periods of the series.
Denoting the original series by y1, y2, y3,….., the mean of the first three values,
given by (y1+ y2+ y3)/3, is placed at the midpoint of the period covering first
three years. This is the first moving average value. The second moving average
value is obtained by calculating the mean of the values covering the period from
second to fourth year. This is given by (y2+ y3+ y4)/3 and is placed at the mid-
p0int of the period covering second to fourth year. This process is repeated. It is
clear that some of the values for the years at the beginning as well as at the end
cannot be obtained by this method.
Two cases may be distinguished, viz., when the period of moving average is odd
and when it is even. If the period is odd (for example, if the period is three years),
the first moving average is placed at the second year, the second moving average
is placed at the third year and so on.
158
 If, however, the period is even (for example four years), the moving average Deterministic Time
Series and Forecasting
value fall between two consecutive years and ‘centering’ is necessary for getting
trend values for various years.
As an illustration, let us consider a schematic representation for the calculation of
centered 4-year moving averages. Here we will present two methods- the direct
method (Table 7.2), as well as the short-cut method (Table 7.3).
Table 7.2 Calculation of centered 4-year moving averages (Direct Method)

Year y1 4-year Moving Total 4-year Moving Centered Centered 4-

Average Moving year moving
Total

(1) (2) (3) (4) (5) (6)

1 y1 - -

2 y2 - -

y1 + y1 + y1 + y1 = T1 T1 / 4

3 y3 (T1 + T2)/4 (T1 + T2)/8

y2 + y3+ y4 + y5 = T2 T2 / 4

4 y4 (T2 + T3)/4 (T2 + T3)/8

y3 + y4+ y5 + y6 = T3
T3 / 4

5 y5 (T3 + T4)/4 (T3+ T4)/8

y4 + y5+ y6 + y7 = T4 T4 / 4

6 y6 - -

7 y7 - -

In the above illustration, the period of moving averages is 4 years. Both in the
direct and in the short-cut method col. 3 shows the 4-year moving totals. The first
value (T1) is placed between the second and the third year, the second moving
total (T2) is placed between the third and the fourth year and so on. The centered
4-year moving averages are placed at the third year, fourth year, by taking a
further 2 item moving average in the direct method (Table 7.2.) In the short cut
method (Table 7.3), the calculation of the 4-year moving average is omitted (as
shown in col. 4 of Table 7.2 in the direct method). Instead, the 2-item moving
totals of the 4-year moving averages are obtained (col. 4 and 5).
159
Summarisation of You should note that for a 4-year moving average, the procedure for centering
Bivariate and Multi-
variate Data leaves out 4/2 = 2 years at the end of the series each.
Table 7.3 Calculation of centered 4-year moving averages (Short Method)

Year y1 4-year Moving Total 2-item moving Centered 4-year

(M.T.) total (M.T.) moving average
(M.A.)
(1) (2) (3) (4) (5)
1 y1 - -
2 y2 - -
y1 + y2 + y3 + y4 = T1
3 y3 (T1 + T2) (T1 + T2)/8
y2 + y3+ y4 + y5 = T2
4 y4 (T2 + T3) (T2 + T3)/8
y3 + y4+ y5 + y6 = T3
5 y5 (T3 + T4) (T3+ T4)/8
y4 + y5+ y6 + y7 = T4
6 y6 - -
7 y7 - -

7.3.2 Suitability of Moving Averages

Moving average method is simple to apply but the success of this method
depends on the proper choice of the period. Moving average with a period
exactly equal to or a multiple of the period exactly equal to or a multiple of the
period of the cycle present in the series will completely eliminate the cyclical
component and give an estimate of the trend. This method is flexible but some
trend values at the beginning and at the end of the series have to be left out and
their number increases with increase in the period of the moving average. Again
as moving average assumes no law of change; the method cannot be used for
forecasting future trend.
7.3.3 Examples of Moving Averages
Example 7.3.1: Calculate the three and five year moving averages of the
following data:

Year 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

Production
18 19 20 22 20 19 22 24 25 24 25 26
(‘000 tons)

160
Steps of Calculation Deterministic Time
Series and Forecasting
1) In Table 7.3.1 the figures in col. 3 are obtained as the sum of three
consecutive values of col. 2. Thus the first moving total (M.T.) is 57 = 18
+ 19 + 20 and is placed against 2001. The second moving total 61 = 19 +
20 + 22 is placed against 2002.

2) The three-year moving average (M.A.) in col. 4 is obtained by dividing

the corresponding three-year moving total in col. 3 by 3, the period of the
moving average. Thus 57 ÷ 3 = 19, 61 ÷ 3 = 20.3, etc.

3) The five-year moving totals in col.5 are obtained as the sum of five
consecutive values in col.2. Thus the first moving total against the year
2002 is 99 = 18 + 19 + 20 + 22 + 20.

4) The five-year moving total in col. 5 by 5. Thus moving average for 2005
is 107 ÷ 5 = 21.4.

Table 7.3.1: Calculation of (I) 3-year Average (II) 5-year Moving Average

Year Production 3-year 3-year 5-year 5-year

M.T. M.A. M.T. M.A.

(1) (2) (3) (4) (5) (6)

2000 18 - - - -

2001 19 57 19.0 - -

2002 20 61 20.3 99 19.8

2003 22 62 20.77 100 20.6

2004 20 61 20.3 103 20.6

2005 19 61 20.3 107 21.4

2006 22 65 21.7 110 22.0

2007 24 71 23.7 114 22.8

2008 25 73 24.3 120 24.0

2009 22 74 24.7 124 24.8

2010 25 75 25.0 - -

2011 26 - - - -

161
Summarisation of
Bivariate and Multi- Note that for 3-year centered moving averages = 1 year, and for 5-year
variate Data
centered moving averages = 2 years, respectively, are left out both at the
beginning and the end of the series.

Example 7.3.2: Compute trend values for the following time series using 4-
yearly moving averages.

Year 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018

Yield 52 54 55 57 58 61 63 66 67 70
(qntls.)

Solution:
Table 7.3.2(a): Calculation of 4-year moving average (Direct Method)

Year Yield 4-year 3-year 2 item M.T. of (Centered)

M.T. M.A. col.4(centered) 4-year
M.A.

(1) (2) (3) (4) (5) (6)

2009 52 - - - -

2010 54 218 54.50 - -

2011 55 224 56.00 110.50 55.250

2012 57 231 57.75 113.75 56.875

2013 58 239 59.75 1177.50 58.750

2014 61 248 62.00 121.75 60.875

2015 63 257 64.25 126.25 63.125

2016 66 266 66.50 130.75 65.375

2017 67 - - - -

2018 70 - - - -

162
 Table 7.3.2(b): Calculation of 4-year Moving Average (Shortcut Method) Deterministic Time
Series and Forecasting
Year Yield 4-year 2-item (Centered)
M.T. M.A 4-year M.A

(1) (2) (3) (4) (5)

2009 52 - - -
2010 54 218 - -
2011 55 224 442 55.250
2012 57 231 455 56.875
2013 58 239 470 58.750
2014 61 248 487 60.875
2015 63 257 505 63.125
2016 66 266 523 65.375
2017 67 - - -
2018 70 - - -

Steps of Calculation (Direct Method) – see Table 7.3.2(a)

1) Col. 3 is the sum of four consecutive values in col.2

Thus, 52 + 54 + 55 + 57 =218, 54 + 55 + 57 + 58 = 224.

2) Col. 4 = col. 3÷4. Thus 218 ÷ 4 = 54.5, 224 ÷ 4 = 56.

3) Col. 5 = Sum of two consecutive values in col.4.

Thus, 54.5 + 56.0 = 110.5, 5.00 + 57.75 = 113.75.

4) Col. 6 = col. 5÷2. Thus 110.5÷2 = 55.25.

Steps of calculation (shortcut method) – see Table 7.3.2 (b)

1) Col. 4 is the sum of two consecutive values in col.3.

Thus, 218 + 224 = 442, 224 + 231 = 455.

2) Col.5 = col.4 ÷ 8. Thus 442 ÷ 8 = 55.25

Example 7.3.3
Find trend values for the following series using a 3-year weighted moving
average with weights 1, 2, 1.

Year 1 2 3 4 5 6

Value 2 3 5 6 8 11

163
Summarisation of Solution:
Bivariate and Multi-
variate Data Table 7.3.3: Calculation of 3-year weighted moving average

Year Value 3-year weighted 3-year weighted

moving total moving average
(M.T.) (M.A)

(1) (2) (3) (4)

1 2 - -

2 3 13 3.25

3 5 19 4.75

4 6 25 6.25

5 8 33 8.25

6 11 - -

Step of calculation
1) Col. 3 figures are the weighted moving totals of col.2 figures with weights 1,
2, 1.
Thus 1 × 2 + 2 ×3+ 1×5 =13
1 × 3 + 2 × 5 + 1×6 =19
2) Col.4 = col.3 ÷ (sum of weights, i.e., 4)
Thus 13 ÷ 4 = 3.25, 19 ÷ 4 = 4.75

Example 7.3.4: Calculate the 4-quarter moving average for the following time
series data

Year
Quarter
2015 2016 2017 2018

1 62 66 72 79

2 58 60 67 74

3 72 74 80 88

4 60 64 69 77

164
Solution: Deterministic Time
Series and Forecasting
Year Quarter Value 4-quarter Centered 4-quarter
(M.T.) (M.T.) (M.A.)

(1) (2) (3) (4) (5) (6)

2015 1 62 - - -
2 58 - - -
259 - -
3 72 508 63.50
256
4 60 514 64.25
258

2016 1 66 518 64.75

260
2 60 524 65.50
264
3 74 534 66.75
270
4 64 547 68.38
277
2017 1 72 560 70.00
283
2 67 571 71.38
288
3 80 583 72.88
295
4 69 597 74.63
302
2018 1 79 612 76.50
310
2 74 628 78.50
318 - -
3 88 - - -
4 77 - - -

165
Summarisation of Check Your Progress 1
Bivariate and Multi-
variate Data 1) Given below is data on index of production for the period 2011 to 2020.

Year 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020

Index of 109.2 119.8 129.7 140.8 153.8 152.2 152.6 163.0 175.3 184.3
Production

1) Fit the trend line and predict the index of production for the year 2012 by
3-year moving averages method.
…………………………………………………………………………...…
………....………………………………………………………………..….
……..…………………………………………………………………….…
……………………...………………………………………………………
………………………………...……………………………………………

2) Bring out the advantages and disadvantages of moving averages method.

…………………………………………………………………………...…
………....………………………………………………………………..….
……..…………………………………………………………………….…
……………………...………………………………………………………
………………………………...……………………………………………

7.4 METHOD OF FITTING POLYNOMIALS

Method of fitting polynomials is perhaps the best and the most objective method
of determining trend. Here an appropriate type of polynomial is selected for trend
and the constants appearing in the trend equation are determined on the basis of
the given time series data.
The choice of an appropriate polynomial is facilitated by a graphical
representation of the data for which, apart from the usual arithmetic scales, semi-
logarithmic or doubly-logarithmic scales may be used. If the plotted data show
approximately a straight line tendency on an ordinary graph paper, the used is
Y = a + bx (straight line or first degree polynomial).
If they show a straight line on a semi-logarithmic graph paper, the equation is log
Y = a + bx, which is obtained by taking logarithm of Y = A.Bx (exponential
function). Note that a = log A and b = log b.
Some times a second or a third degree polynomial may also be fitted.
Y = a +bx +cx2 (second degree polynomial or parabola)

166
Y = a +bx +cx2 +dx3 (third degree polynomial) Deterministic Time
Series and Forecasting
The constants appearing in the above equations (such as a,b,c…) are obtained by
applying the principle of “least squares”, as in regressions (see unit 5). This
states that the values of the constants will be such as to make the sum of squares
of the deviations
∑(𝑦 − 𝑌) minimum,

where y = observed value,

Y = expected value obtained from the trend equation of the type

Y = a +bx

Or,

Y = a +bx +cx2 etc., and the summation is taken over all the observations.

In the case of straight line fitted by the method of “least squares”, the constants a
and b are determined from the following normal equations:
∑ 𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥 and

∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥

where n is the number of years covered.

Similarly in the case of parabola or second degree polynomial the constants a, b

and c are determined from the three normal equations
∑ 𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥 + 𝑐 ∑ 𝑥

∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥 + 𝑐 ∑ 𝑥

∑𝑥 𝑦 = 𝑎∑𝑥 + 𝑏∑𝑥 + 𝑐∑𝑥

Rule for writing down the normal equations

To get the first normal equation, multiply each observation by coefficient of a in
that equation and take sum over all the n observations.

Thus for straight line y = a + bx, as the coefficient of a is 1, the first normal
equation is ∑ 𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥.

For the second normal equation, multiply each observation by the coefficient of b
in that equation and take sum over all the n observations. In the case of straight
line, coefficient of b is x. So, the second normal equation is ∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 +
𝑏∑𝑥 .
Now we will consider trend fitting for periods covering odd (Table 7.4) and even
(Table 7.5) number of years taking the first degree polynomial.

167
Summarisation of
Bivariate and Multi-
Case I: Odd number of years (n = 5)
variate Data Table 7.4

Year y x x2 xy

(1) (2) (3) (4) (5)

1 y1 -2 4
2 y2 -1 1
3 y3 0 0
4 y4 1 1
5 y5 2 4
total Sy 0 10 Sxy

The normal equations are:

∑ 𝑦 = 5𝑎 + 𝑏 ∑ 𝑥 = 5𝑎 (Since ∑ 𝑥 = 0)
∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥 = 10𝑏
∑ ∑
∴𝑎 = , 𝑏=

where the origin (x = 0) is taken at the midpoint of the interval covered by 5

years, i.e., at the third year and unit of time = 1 year. In real life situations the
actual values of y’1 s are considered. Hence ∑ 𝑥 and ∑ 𝑥𝑦 will be known.

Case II: Even number of years (n = 6)

Table 7.5
Year y x x2 xy
(1) (2) (3) (4) (5)
1 y1 -5 25
2 y2 -3 9
3 y3 -1 1
4 y4 1 1
5 y5 3 9
6 Y6 5 25
total ∑𝑦 0 70 𝑥𝑦

The constants a and b will be obtained from the following equations

∑ 𝑥 = 6𝑎

168
∑ 𝑥𝑦 = 70𝑏 Deterministic Time
Series and Forecasting
Here the origin (x = 0) will be in the middle of 3rd and 4th year and the unit of x
=6 months.
7.4.1 Suitability of Least Squares Method
Trend lines are used for description of the growth or decline of the time series
and as an aid to the study of the long-term trend of the economy. The method of
fitting polynomials completely eliminates personal bias and trend values for all
the given periods can be obtained. This is, however, not possible with moving
average method. But the choice of the type of the polynomial curve is arbitrary
and one cannot be sure whether a linear or parabolic curve will represent the
trend best. The choice of the trend equation may itself lead to a bias. It is,
however, possible to get some idea of the pattern of trend from the scatter
diagram of the data.
7.4.2 Examples of Least Squares Method
Example 7.4.1
Fit a straight line trend by the method of least squares to the following data:
1) The data given below give the index of industrial production from 1961 to
1970

years 2005 2006 2007 2008 2009 2010 2011

Production 81 92 100 105 112 120 120

Estimate the production for 2012.

Solution:

Here the number of years is odd (n = 7) Let y = a+bx be the equation of the
straight line trend with origin (x = 0) at 2008 and one unit of x = 1 year. The least
squares normal equations are (see unit 5)

∑ 𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥

∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥

Thus, substituting the values of ∑ 𝑦, ∑ 𝑥 y, ∑ 𝑥 and ∑ 𝑥 from the above table in

the normal equations, we get

7a = 736, so a = 105.1

28b = 203, so b = 7.21

The trend equation is
Y = 105.1+7.21x, with origin at 2008 and unit of x =1 year. The value of x for
2012 would be 4.
169
Summarisation of Table 7.4.1: Fitting straight Line Trend
Bivariate and Multi-
variate Data
Year Production (y) x x2 xy

(1) (2) (3) (4) (5)

2005 81 -3 9 -243

2006 92 -2 4 -184

2007 100 -1 1 -100

2008 105 0 0 0

2009 112 1 1 112

2010 120 2 4 240

2011 126 3 9 378

Total 736 0 28 203

Hence, using the trend equation the estimate for 2012 is Y = 105.1 + 4 × 7.21 =
133.94.

Example: 7.4.2
Fit a straight line trend to the following time series data:

Year 2010 2011 2012 2013 2014 2015

Profits: 3.1 3.3 3.6 3.2 3.7 3.9

Estimate the profit for 2016.

Solution:

Here the number of years is even (n = 6). Let y = a+bx be the trend equation
with origin.

(n = 0) mid-way between 2012 and 2013 and unit of x = 6 months.

The normal equations are

∑ 𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥

∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥

170
 Table 7.4.2: Fitting straight Line Trend Deterministic Time
Series and Forecasting
Year Profit (y) x x2 xy
(Rs. lakhs)

(1) (2) (3) (4) (5)

2010 3.1 -5 25 -15.5
2011 3.3 -3 9 -9.9
2012 3.6 -1 1 -3.6
2013 3.2 1 1 0.0
2014 3.7 3 9 11.1
2015 3.9 5 25 19.5
Total 20.8 0 70 4.8

So, substituting the values of ∑ 𝑦 , ∑ 𝑥𝑦, ∑ 𝑥, and ∑ 𝑥 from the above table in
the normal equations, we get
6a = 20.8, or a = 3.47
70b = 4.8, or b = 0.07
The trend equation is
Y = 3.47 + 0.07x, with origin at the middle of 2012 and 2013 and unit of x = 6
months.
For 2016, x would be 7.
So, estimate for 2016 is
Y = 3.47 + 0.07 × 7 = 3.47 + 0.49 =3.96
Hence the estimated profit for 2016 is Rs. 3.96 lakhs.
Example: 7.4.3
The sales of a company (in thousands of rupees) for the year 2010 to 2016 are
given in the following table. Fit an exponential trend (Y = A,Bx) and estimate the
sales for 2017.

Year 2010 2011 2012 2013 2014 2015 2016

Sales 32 47 65 92 132 190 275

Solution:
Here the number of years is odd (n = 7). Taking log of both sides of the given
equation, we can write log Y = logA + x logB. Let a = logA and b= logB. Thus
we can write

171
Summarisation of logY = a + bx.
Bivariate and Multi-
variate Data Further, we take origin (x = 0) at 2013 and one unit of x = 1 year. The least
squares normal equations are:
∑ log𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥
∑ 𝑥log𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥
Table 7.4.3: Fitting Straight Line Trend

Year Sales (y) x x2 logy x. logy

2010 32 -3 9 1.5051 4.5153

2011 47 -2 4 1.6721 3.3442

2012 65 -1 1 1.8129 1.8129

2013 92 0 0 1.9638 0

2014 132 1 1 2.1206 2.1206

2015 190 2 4 2.2788 4.5576

2016 275 3 9 2.4398 7.3119

Total 833 0 28 13.7931 4.3237

So, substituting the values of ∑ log𝑦 , ∑ 𝑥. log𝑦, ∑ 𝑥, and ∑ 𝑥 from the above
table in the normal equations, we get
7a = 13.7931, or a = 1.97
28b = 4.3237, or b = 0.154
Thus, the fitted function is logy = 1.97 + 0.154x or Y = antilog(1.97 + 0.154x).
For 2017, x would be 4.
Thus, the estimated value for 2017 is
Y = antilog (1.97+0.154×4) = antilog 2.586 = 385.48.
A case with even number of years can be attempted as in the fitting of a straight
line (see Example 7.4.2).

Example: 7.4.4
The following table shows the production of cement in India during 2002 to
2008.
Fit a second degree polynomial to the data.
172
 Deterministic Time
Series and Forecasting
Year 2012 2013 2014 2015 2016 2017 2018

Production 23.7 27.1 30.2 33.1 36.4 39.3 45.0

Solution:
Here the number of years is odd (n = 7). Let y = a + bx+cx2 be the trend
equation with origin (x = 0) at 2015 and unit of x = 1year. The normal equation
is:
∑ 𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥 + 𝑐 ∑ 𝑥
∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥 + 𝑐 ∑ 𝑥
∑𝑥 𝑦 = 𝑎∑𝑥 + 𝑏∑𝑥 + 𝑐∑𝑥
Table 7.4.4: Fitting Second Degree Polynomial

Year y x x2 x3 x4 xy x2 y

2012 23.7 -3 9 -27 81 -71.1 213.3

2013 27.1 -2 4 -8 16 -54.2 108.4

2014 30.2 -1 1 -1 1 -30.2 30.2

2015 33.1 0 0 0 0 0.0 0.0

2016 36.4 1 1 1 1 36.4 36.4

2017 39.3 2 4 8 16 78.6 157.2

2018 45.0 3 9 27 81 135.0 405.0

Total 234.8 0 28 0 196 94.5 950.5

Substituting the values from the table in the normal equations

7a + 28c = 234.8
28b = 94.5
28a + 196c = 950.5
Solving these three equations, simultaneously, we get
a = 33
b = 3.37
c = 0.134
Hence, the second degree polynomial is

173
Summarisation of Y = 33 + 3.37x + 0.134x2,
Bivariate and Multi-
variate Data with origin (x = 0) at 2015 and unit of x = 1 year.
Example: 7.4.5
Fit a second degree polynomial to the following data. Estimate the trend value for
2012.

Year 2006 2007 2008 2009 2010 2011

Annual Indian 23.7 27.1 30.2 33.1 36.4 39.3

Imports (108 Rs.)

Solution:
Here the number of years is even (n = 6). Let y = a + bx+cx2 be the trend
equation with origin (x = 0) mid-way between 2008 and 2009 and unit of x = 6
months. The normal equations are ∑ 𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥 + 𝑐 ∑ 𝑥
∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥 + 𝑐 ∑ 𝑥
∑𝑥 𝑦 = 𝑎∑𝑥 + 𝑏∑𝑥 + 𝑐∑𝑥
Table 7.4.5: Fitting Second Degree Polynomial

Years y x x2 x3 x4 xy x2y

2006 507 -5 25 -125 625 -2535 12675

2007 602 -3 9 -27 81 -1806 5418

2008 681 -1 1 -1 1 -681 681

2009 914 1 1 1 1 914 914

2010 1255 3 9 27 81 3765 11295

2011 1361 5 25 125 625 6805 34025

Total 5320 0 70 0 1414 6462 65008

Substituting the values from the table in the normal equations

6a + 70c = 5320
70b = 6462
70a + 1414c = 65008
Solving the above three equations simultaneously, we get
a = 829.2, b = 92.31 and c = 4.924.
The second degree polynomials is

174
Y = 829.2 + 92.31x + 4.924x2, Deterministic Time
Series and Forecasting
with origin (x = 0) mid-way between 2008 and 2009 and unit of x = 6 months.
For 2012, x would be 7.
Therefore, estimate for 2012 is
Y = 829.2 + 92.31 × 7 = 3.47 + 4.924 × (7)2
= 829.2+646.17+241.28 = 1716.65.

7.5 MONTHLY OR QUATRELY TREND VALUES

FROM ANNUAL DATA
In a time series, annual data may be available in different forms such as (1)
monthly or quarterly averages fo02r each year, and (ii) annual totals.
If the trend equation is fitted to the monthly or quarterly data, there will be no
difficulty in obtaining monthly or quarterly values. However, it is not advisable
to fit a trend line by the method of least square to the monthly or quarterly data.
The trend line is usually fitted to the annual data and then the trend values may
be obtained for different months (or quarters). The method of obtaining monthly
(or quarterly) trend equation is discussed below.
Let y = a+bx be an annual trend equation. If we divide both sides of this equation
by 12, we get a monthly average equation. Thus = + 𝑥 is a monthly
average equation. Denoting by = ,𝐴 = 𝑎𝑛𝑑 𝐵 = , we can write the
monthly average equation as Y =A+Bx. Here, Y is monthly and average and B
denotes change in monthly average per unit change in x,i.e., 1 year.
To get a monthly equation, we have to determine the corresponding rate of
change of Y. since B is the average monthly change in Y per year, will denote
average change per month. Thus, the monthly equation can be written as
𝑌 = 𝐴 + 𝑥 or 𝑌 = + 𝑥, where x denotes month rather than year.

Similarly, we can write 𝑌 = + 𝑥 as the quarterly average equation, where a

unit of x denotes one year, and 𝑌 = + 𝑥 as the quarterly average equation,
where a unit of x denotes one quarter.
Shifting of Origin
We define a, in the equation Y = a + bx, as the value of trend in the year of
origin. Thus, with the shifting of origin the value of a changes. Let us assume
that the year of origin. (i.e., x = 0) is 2015 and we want to change it to 2018.
We note that x = 3 for 2018, therefore, trend for 2018 = a +3b. Treating this as
constant term in the trend equation, y = (a + 3b) + bx becomes the new trend
equation with 2018 as origin.

175
Summarisation of Example 7.5.1
Bivariate and Multi-
variate Data The trend equation for certain production data is Y = 150 +24x (y = annual
production in thousand tons and x = time with origin at 2008, unit of x = 1 year).
Estimate the trend value for May 2013.
Solution: The monthly trend equation is
𝑌= + 𝑥 = 12.5 + 0.167𝑥,

where Y = monthly production, unit of x = 1 month and origin at 2008, i.e., 30th
June 2008.
To estimate the trend for May 2013, we substitute x = 58.5 in the above
equation. Thus, we get Y = 12.5 + 0.167 × 58.5 = 22.25 (‘000 tons)
Example 7.5.2
The trend equation fitted to quarterly average sales for 7 years is given by y =
250 + 20x (unit of x = 1 year, origin = 30th June 2010). Estimate the trend value
for the first quarter of 2013 (January-March).
Solution: Here the quarterly average refers to average per quarter for each year.
The quarterly trend equation is given by 𝑌 = 250 + 𝑥, where Y = quarterly
sales, x = 1 quarter and origin at 30th June 2010.
The interval between 30th June 2010 and the 1st quarter of 2013 are 10.5 quarters.
Thus, to obtain the trend for 1st quarter of 2013, we substitute x = 10.5 in the
above equation.
Hence, the required trend is Y = 250 + 5× 10.5 = 302.5.
Check Your Progress 2
1) Fit a straight line trend to the following data and show how to obtain the
monthly trend values from the trend line fitted to the given time series.
Obtain two such monthly values.

Year 2010 2011 2012 2013 2014

Monthly Production: 38 40 41 45 47
(in ‘000 tons)

…………………………………………………………………………...…
………....………………………………………………………………..….
……..…………………………………………………………………….…
……………………...………………………………………………………
……………………...………………………………………………………
…………………...…………………………………………………………
……………………………...………………………………………………

176
2) The trend equation for certain production data is y = 240 + 48x (y = annual Deterministic Time
Series and Forecasting
production in tons, x = time with origin at 2010, unit of x = 1 year).
Estimate the trend for October 2016.
…………………………………………………………………………...…
………....………………………………………………………………..….
……..…………………………………………………………………….…
……………………...………………………………………………………
………………………………...……………………………………………
3) The trend equation fitted to quarterly average sales data is given by y =
60 + 8x (unit of x = 1year, origin = 30th June, 2018). Estimate the trend
value for first quarter (Jan-March.) of 2020.

…………………………………………………………………………...…
………....………………………………………………………………..….
……..…………………………………………………………………….…
……………………...………………………………………………………
……………………...………………………………………………………
………………………………...……………………………………………

7.6 MEASUREMENT OF SEASONAL VARIATIONS

There are a number of methods of measuring seasonal variations in time series,
depending on how the other components such as cyclical, trend and irregular
movements are present in it. For simplicity, we shall consider seasonal variations
in monthly or quarterly data only, but the procedure for weekly or daily data will
be similar. It may be mentioned here that annual data contains no seasonal
variation. The application of the methods, to be discussed below, will give us
4(or 12) numbers for quarterly (or monthly) data. These will be termed as
seasonal indices and are normally expressed as percentages. A figure of a
particular quarter (or month) indicates whether that quarter is above or below the
normal quarter. For example, a value of 80 for particular quarter indicates that the
business for exports or sales (say) during that quarter is slack and it is below the
normal quarter by 20%. We will consider only the multiplicative model for the
measurement of seasonal variations. The main methods for the measurement of
seasonal variation are given below.

1) Method of Simple Average

2) Ratio to Trend Method

3) Ratio to Moving Average Method

177
Summarisation of 7.6.1 Method of Simple Average
Bivariate and Multi-
variate Data This method assumes that the time series variable y is made up of only two
components, viz., seasonal (S) and irregular or random component (I). Thus, we
can write
Y = S.I
When we take average of y values for each month or quarter of all the years, the
irregular component gets eliminated and we are left with seasonal component.
We will illustrate this method in Table 7.6.
Table 7.6 Illustration of the Method of Simple Average

Years Quarters

I II II IV

1 y1 Y2 Y3 Y4

2 Y5 Y6 Y7 Y8

3 Y9 y10 y11 y12

4 y13 y14 y15 y16

5 y17 y18 y19 Y20

Total T1 T2 T3 T4

Average A1 A2 A3 A4

S.I. s1 s2 s3 s4

S.I. (adjusted) S1 S2 S3 S4

Explanatory notes:
a) T1 = y1 + y5+ y9 + y13 + y17 is the total of y values of first quarter of each year.
Similarly, T2, T3 and T4, are the totals of second, third and fourth quarters of
each year respectively.
b) Ai is the ith quarter average = , where i = 1, 2, 3, 4, and n denotes the number
of years.
∑
c) G is defined as the grand average = .

d) 𝑠 = × 100, 𝑖 = 1, 2, 3, 4.

e) s = s1 + s2+ s3 + s4

178
f) S1, S2, S3, and S4, are the seasonal indices for the first, second, third and the Deterministic Time
Series and Forecasting
fourth quarters respectively, where 𝑆 = × 400, 𝑖 = 1, 2, 3, 4. Note that the
sum of these 4 index numbers must be equal to 400. Further, 𝑆 = 𝑠 if s =
400.
g) For a time series with monthly data, the sum of 12 seasonal indices, one for
each month, must be equal to 1200.
Example: 7.6.1
Compute seasonal indices for the following data by the Method of Simple
Average.

Years Quarters

I II III IV

1992 72 68 80 70

1993 76 70 82 74

1994 74 66 84 80

1995 76 74 84 78

1996 78 74 86 82

Solution:
Table 7.6.1: Calculation of Seasonal Indices

Years Quarters

I II III IV

1992 72 68 80 70

1993 76 70 82 74

1994 74 66 84 80

1995 76 74 84 78

1996 78 74 86 82

Total 3776 352 416 384

Average 75.2 70.4 83.2 76.8

S.I. 43 92.15 108.90 100.52

179
Summarisation of Explanatory Notes:
Bivariate and Multi-
variate Data . . . .
Grand Average 𝐺 = = = 76.4

Seasonal Index for Quarter I, i.e., S1 = 98.43

Seasonal Index for Quarter II, i.e., S2 = 92.15

Seasonal Index for Quarter III, i.e., S3 = 108.90

Seasonal Index for Quarter IV, i.e., S4 = 100.52

Since the sum of these indices = 400, no adjustment is needed.

7.6.2 Ratio to Trend Method
If the data contain trend to an appreciable extent, we first find an appropriate
trend equation to determine the trend for various quarters or months. Usually the
monthly or quarterly trend values are obtained from the quarterly (or monthly)
average trend equation. The trend is then eliminated by expressing the original y
values as percentages of the corresponding trend values. This method is based
upon the assumption that cyclical variations are either not market or completely
absent.
Symbolically, we can write
× 100 = × 100 = 𝑆𝐼 × 100

From this, the irregular component can be eliminated by the use of Simple
Average Method.
Example: 7.6.2
The following table shows the sales (9n’000 Rs.) in a departmental store for five
different years. Obtain the seasonal indices by Ratio to Trend Method.

Years Quarters

I II III IV

2000 502 1632 605 362

2001 526 1700 680 390

2002 556 1820 780 422

2003 590 1955 888 464

2004 632 2110 1002 515

180
Solution: Deterministic Time
Series and Forecasting
Let us fit a straight line trend to the data on quarterly averages. The trend
equation fitted to quarterly averages y = a + bx, where y denotes quarterly
average of the year and the unit of x = 1 year. The table below has been
constructed from the given data by computing the averages of 4 quarters of each
year.
Table 7.6.2(a): Fitting Linear Trend

Years Quarters

y x x2 xy

2000 775 -2 4 -1550

2001 824 -1 4 -824

2002 894 0 0 0

2003 974 1 1 9774

2004 1065 2 4 2130

Total 4532 0 10 730

The normal equations are

∑ 𝑦 = 𝑛𝑎 + 𝑏 ∑ 𝑥
∑ 𝑥𝑦 = 𝑎 ∑ 𝑥 + 𝑏 ∑ 𝑥
Substituting values from the above table into the normal equations for linear
trend, we get
5a = 4532 or a = 906.4
10b = 730 or b = 73.
Thus, the quarterly average trend equation is T = 906.4 + 73x (origin: 2002, unit
of x = 1 year)
Note that we are using T instead of Y (used earlier in fitting of trends) From this,
we can write the quarterly trend equation as
𝑇 = 906.4 + . 𝑥 = 906.4 + 18.25𝑥

(origin: 30th June, 2002 unit of x = 1 quarter)

Shifting the origin to 3rd quarter (mid-point of third quarter) of 2002, the
quarterly trend equation becomes
T = 906.4 + 18.25 (x + 0.) = 906.4 + 9.125 + 18.25x
= 9.125 + 18.25x.
181
Summarisation of Putting appropriate values of x, we can get the trend values (T) for various
Bivariate and Multi-
variate Data quarters. The next step is to express the original values (y) as percentage of trend
i.e., (y ÷ T)×100, giving “trend ratios”. The trend values along with the trend
ratios are shown in Table 7.6.2(b).
Table 7.6.2(b): Calculation of Trend Ratios

Year Quarter x T = y (y ÷
915.525 + T)×100
18.25x

2000 I -10 733.0 502 68

II -9 7751.3 1632 217

III -8 769.5 605 779

IV -7 787.8 362 46

2001 I -6 806.0 526 65

II -5 824.3 1700 206

III -4 842.5 680 81

IV -3 860.8 390 45

2002 I -2 879.0 556 63

II -1 897.3 1820 203

III 0 915.5 780 85

IV 1 933.8 422 45

2003 I 2 952.0 590 62

II 3 970.3 1955 201

III 4 988.5 888 90

IV 5 1006.8 464 46

2004 I 6 1025.0 632 62

II 7 1943.3 2110 202

III 8 1061.5 1002 94

IV 9 10779.8 515 48

182
The trend ratios are now arranged by quarters and the seasonal indices are Deterministic Time
Series and Forecasting
calculated by the method of simple averages.

Table 7.6.2(c) : Calculation of Seasonal Indices

Years Quarters
I II III IV
2000 68 217 79 46
2001 65 206 81 45
2002 63 203 85 45
2003 62 201 90 46
2004 62 202 94 48
Total 320 1029 429 230
Average 64.0 205.8 85.8 46.0
S.I. 63.74 209.98 85.46 45.82

7.6.3 Ratio to Moving Average Method

This method is used when the time series data is composed of all the four
components. We know that moving averages, with period equal to the periodic
variations, and completely eliminates those variations. Thus, if we take 4 period
moving average (M) of quarterly data (or 12 period moving averages of monthly
data), the seasonal variations (and some irregular movements) are eliminated
from the original (y) values. Further, by expressing y as a percentage to moving
average, i.e., (y ÷ M)× 100, we get values consisting of seasonal and irregular
components. The seasonal component is, then, isolated from irregular component
by the use of the Method of Simple Average.

Symbolically, we can write

× 100 = = 𝑆𝐼"
′

Example: 7.6.3
Use the Ratio to Moving Average Method to calculate seasonal indices for the
following data.

Year Summer Monsoon Autumn Winter

2009 30 81 62 119
2010 33 104 86 171
2011 42 153 99 221
2012 56 172 129 235
2013 67 201 136 302

183
Summarisation of Solution:
Bivariate and Multi-
variate Data Table 7.6.3: Calculation of Seasonal Indices by Ratio to Moving Average Method

Year Quarter y 4-period M.T. Centered Total 4-period M.A. (M) ( y ÷ M)×100

2009 Sum 30 - - - -
Mon 81 -- - -- -
292 - - -
Aut 62 597 73.38 84.50
295
Win 119 613 76.63 155.30
318
2010 Sum 33 660 82.50 40.00
342
Mon 104 736 92.00 113.04
394
Aut 86 796 99.63 86.32
403
Win 171 855 106.88 160.00
452
2011 Sum 42 917 114.63 36.64
465
Mon 153 980 122.50 124.90
515
Aut 99 1044 130.50 75.86
529
Win 221 1077 134.63 164.16
548
2012 Sum 56 1126 140.75 39.79
578
Mon 172 1170 146.25 117.61
592
Aut 129 1195 149.38 86.36
603
Win 235 1235 154.38 152.23
632
2013 Sum 67 1271 158.88 42.17
639
Mon 201 1345 168.13 119.55
706 - - -
Aut 136 - - - -
Win 302 - - - -

184
The moving ratios are now arranged by quarters and the seasonal indices are Deterministic Time
Series and Forecasting
calculated by the method of simple averages.

Year Quarters

Summer Monsoon Autumn Winter

2009 - - 84.50 155.30

2010 40.00 113.04 86.32 160.00

2011 36.64 124.90 75.86 164.16

2012 39.79 117.61 86.36 152.23

2013 42.17 119.55 - -

Total 158.60 475.10 333.04 631.69

Average 39.65 118.78 83.26 157.92

S.I. 36.69 118.89 83.34 158.08

Check Your Progress 1

1) The following data represent the production of finished steel tins for the
years 1992 to 1995:
Production of Finished Steel Tins (000 tons)

Year Jan. Feb. Mar Apr. May. Jun. Jul. Aug. Sep. Oct. Nov. Dec.

1992 420 414 502 365 368 332 390 396 429 417 422 496

1993 491 466 516 337 342 360 409 402 372 391 394 446

1994 463 465 478 310 325 406 415 437 438 445 430 416

1995 502 487 536 404 418 429 489 492 475 456 476 476

Compute the seasonal indices by the Method of Simple Averages.

…………………………………………………………………………...…………
....………………………………………………………………..….……..….….…
…………………………………………………………….………………………..
.………………………………………………………………………………..……
…...……………………………………………………………………………....…
………………………………………..................................................................….
185
Summarisation of 2) The following table gives the production of steel in India from 1992 to
Bivariate and Multi-
variate Data 1995 (in ’000) over the different quarters.

Year 1st 2nd 3rd 4th

quarter quarter quarter quarter

1992 1336 1065 1215 1335

1993 1463 1039 1183 1161

1994 1306 1041 1290 1321

1995 1525 1251 1456 1408

Obtain seasonal indices by the method of Ratio to Trend, assuming linear trend.
…………………………………………………………………………...…………
....………………………………………………………………..….……..….….…
…………………………………………………………….………………………..
.………………………………………………………………………………..……
…...……………………………………………………………………………....…
………………………………………..................................................................…

3) Given the following quarterly sales figures in thousands of rupees for the
years 2006 to 2009. Find the specific seasonal by the method of moving
averages.

Years Quarters

I II III IV

2006 290 280 285 310

2007 320 305 310 330

2008 340 321 320 340

2009 370 360 362 380

…………………………………………………………………………...…………
....………………………………………………………………..….……..….….…
…………………………………………………………….………………………..
.………………………………………………………………………………..……
…...……………………………………………………………………………....…
………………………………………..................................................................…

186
4) The seasonal indices for the sales of garments of a particular type in a certain Deterministic Time
Series and Forecasting
shop are given below:
Quarter Seasonal Index
Jan-Mar 97
Apr-Jun 85
Jul-Sep 83
Oct-Dec 135
If the total sales in the first quarter of a year are Rs. 15,000, determine how
much worth of garments of this type should be kept in stocky by the shop
owner to meet the demand for each of the other three quarters of the year?

…………………………………………………………………………...…………
....………………………………………………………………..….……..….….…
…………………………………………………………….………………………..
.………………………………………………………………………………..……
…………………………………………………………………………...…………

7.7 LET US SUM UP

Time series is a set of observation on a variable recorded over time – usually at
equal intervals. The change in the variable concerned can be explained to some
extent on the basis of components of time series. These components are trend,
seasonal variations, cyclical fluctuations and random movements. The observed
value of the variable may be represented either as the product of the aforesaid
components (multiplicative model) or as the sum of the components (additive
model).
Trend can be measured by the method of moving averages and fitting equations
by the method of least squares. Once we estimate the trend, we can predict future
values and also estimate the monthly or quarterly values from the annual trend.
Seasonal variation can be estimated by three methods: Method of Simple
Averages, Ratio to Trend Method and Ratio to Moving Average Method. The
method of simple average is used to average out the irregular component. The
ratio to trend method can be used if cyclical variations are supposed to be absent
while the ratio to moving average method is recommended when the time series
variable is composed of all the four components.

7.10 ANSWERS OR HINTS OT CHECK YOUR

PROGRESS EXERCISES
Check Your Progress 1
1) Read Example 7.3.1 and answer.
2) Read Example 7.3.2 and answer.
187
Summarisation of Check Your Progress 2
Bivariate and Multi-
variate Data 1) y = 42.2 + 2.3x, origin: 2012, unit of x = 1 year (monthly average equation)
y = 42.3 + 0.19x, origin: July, 2012, unit of x = 1 month (monthly equation)
Estimate for March 2011 (x = -16) = 39.23
Estimate for September 2013 (x = 14) = 44.98
2) 45.17 tons
3) 73
Check Your Progress 3

1) 10957, 107.00, 118.69, 82.71, 84.87, 89.19, 99.47, 100.87, 100.11, 99.82,
100.58, 107.12

2) 112.27, 86.62, 100.21, 100.90

3) 104.2, 97.9, 96.5, 101.4

4) Rs. 13144; 12835; 20876.

188