ANALYSIS OF TIME SERIES

2-41
TABLE 2.16: TREND VALUES By sENCER'S 15-POINT FORMULA
t 41215)1-3, 3, 4, 3, -3] y,
14) 1412 14]*151 Trend Values
(1) (2) (3) (4) (5) (6) (7) =
(6)
1,401 320
2 1,437
5,791
3 1,460
5,906
4 1,493 23,962
5 6,055
1,516 24,522
6,210
6 1,586 25,105 1,25,474
7 6,351
1,615 25,631
6,489 1,28,612
8 1,634 26,254 1,32,322 5,21,278
9 1,654
6,581 1,628 99
27,100 1,37,008 5,38,960
10 6,833 1,684-25
1,678 28,232 1,42,816
11 7,197 5,63,488 1,760-90
1,867 29,791 1,49,550
12 7,621 5,93,889 1,855-90
1,998 31,439 1,56,828
13 8,140 6,26,226 1,956-96
2,078 32,988 1,64,073
14 8,481 6,57,603 2055-01
2,197 34,378
8,746 1,71,169
15 2,208 35,477 1,78,010
16 9,011
2,263 36,887
17 9,209
2,343 38,280
18 9,921
2,395
10,139
19 2,920
20 2,481

2-5. MEASUREMENT OF SEASONAL VARIATIONS

It has already been
data is the seasonal
pointed out that one of the types of
component. Many economic and business fluctuations found in time st eries

patterns that are pronounced series

for studying seasonal patterns enough to predict
future behaviour of thehave distinct 5ea
in a time
series are series. The objectives
i) To isolate the necessitated by the following reason
the value of the given scasonal variations,
i.e., to determine the
phenomenon, and effect of seasonal swa on
(ii) To eliminate them, i.e.,
seasonal ups and downs in to determine
the value of the
the series. This known as
is phenomenon if there andi
necessary for the study of cyclic variations.
The determination
the given da de-seasonalising
(i) business efficiency, of seasonal effects is of paramount
or () a
n e n t a l s t o r e w o u l d b e i n t e r e s t e d t o s t u d y t h e v a r i a t i o n s i n t h e d e m ah
ned s
phe
lea
mana
n n
i nog

departme importance indof

production programme. (For example,
articles for different months in th f different

due to seasonalswings. order to plan his future stocks to

data is necessary to Moreover, the isolation and cater to the publl mn
study
the data must be given for the effect elimination
of cycle. Obviously, of seasonal fact ati
parts year, ViZ., monthly or
of for the study of seasonar hou
Different methods for measuring seasonal variations quarterly, weekly, daily
A
2-5-1Method of Simple are
discussed
variations in a time series and Averages. This is the simplest method below. o n a l

involves the following steps: of measurinE

 2.42 FUNDAMENTALS OF APPLED STATISTIC
Arrange the data by years and
months (or quarters
if quarterly data are gven
(a)Compute the average x, I =1,2,
12) for the ith month for all the years.
,

month, i =
1,2, ., 12
represents January, February, ..., December respectavey
(at) Compute the average x
of the monthly averages. ie.,
in) Seasonal indices for different months are
5 F =

obtained by expressing monthly averages as

percentage of x. Thus.

Seasonal Index for ith month =(7,/)x 100; i =

1,2. . 12
Remarks 1. If instead of monthly averages, we use
monthly totals for all the years, the resut
remains the same.

2. Total of seasonal indices is 12 x 100 1,200 for

=
monthly data and 4 x 100 = 400 for quareriy
data.
Merits and Demerits. This method is based the basic
any trend and cyclic components and consists in assumption thar the data do not conta
on

monthly quarterly)
(or values
eliminating irregular components by averagEe
over
different years. Since most of the economic time senes iave irenE
theseassumptions are not in general true and as such this method,
practical utility. though simple. is not of Darn
Example 2.13. Use the method of monthly averages to determine the monthly indices for
the following datd of production of a commodity for the
2003, years 2002, 2004
MMonth (Production in lakhs
of tonnes) Production in lakhs of tonnes
Month
2002 2003
2004 2002 2003 2004
January 12 15 16 July 16 17 16
February 11 14 15 August 13 12 13
March 10 13 14 September 11 13 10
April 14 16 16 October 10 12 10
May 15 16 15 November 12 13 11
June 15 15 17 December 15 14 15
TABLE2-17: COMPUTATION OF SEASONAL IN:DICES
(Production in lakhs of tonnes) Monthly Seasonal
Month 2002 2003 2004 Total
Average Index
(1) (2) (3) (4 (5) (6) (5) +3 (7)
= =
(6) +I Average of averages.
Jan. 12 15 16 43 14 33 a 104 886 163-95
Feb. 11 14 15 40 13-33 97-566 12
March 10 13 14 37 12-33 90-247 136625
April 14 16 16 46 33 112-205easonal Index for Jan
May 15 16 15 46 15 33
June 15 15 17
112-205 14 33
13-6625 x 100
47 15-66 114 620
July 16 17 16 49 16-33 119-524 104886
Aug. 13 12 13 38 12-66 92-662 Seasonal Index for Feb.
Sept. 11 13 10 34 11-33 82-928
Oct. 10 12 10 32 10-66 78-024
13-33
13-6625 X 100
Nov. 12 13 11 36 12-00 87-832 = 97-566
Dec. 15 14 15 44 14-66 and so on.
Total
107-301
492 163-95 1,200
Averages 41 13-6625 100
tolw
ANALYSIS OF TIME SERIES
2-43
2-5:2. Ratio to Trend
Method. This method is an improvement over the
averages method and is based on the Simple
sin

assumption that seasonal variation tor

15 constant factor of the trend. The measurement of seasonal variation any by
givens
this ariation onth
consists of the following method
steps
( ) Compute the trend values by the principle of least squares by fitting an
mathematical curve (straight line, 2nd appropriad
degree parabolic curve or exponential curve, etc 1ate
(uExpress the original data as the percentage of the trend
multiplicative model, these percentages will, therefore contain thevalues. Assuming
seasonal, cyclicthe
t

irregular components. and

and
(Lu) The cyclic and
irregular components are then wiped out by averagng the
percentages for different months (quarters) if the data are monthly (quarterly), thus
us
with indices of seasonal variations. Either arithmetic mean or median can be leavine
averaging, but median is preferred to arithmetic mean since the latter used for
gives undue
weightage to extreme values which are not primarily due to seasonal
abnormal values, modified mean (which swings. If there are few
consists of calculating arithmetic mean
dropping out the extreme or abnormal values) may be used with after
advantage.
(v) Finally, these indices, obtained in
step (ii), are adjusted to a total of 1200 for
monthly data or 400 for quarterly data by multiplying them throughout by a constant k given
by
1,200 4,00
Total of the indices andk =
Total of the indices
monthly and quarterly data
for
respectively.
Merits and Demerits. Since this method attempts at ironing out the
components by the process of averaging, the purpose will be cyclical irregular or

variationsare known to be absent or they

accomplished only if the cyclical
are not
hand, if the series exhibits pronounced cyclical pronounced even if present. On the other
so

square method can never follow the actual data as

swings, the trend values obtained by the
lea
as such the seasonal indices obtained
closely as 12-month moving average 31
by *ratio to trend' method are
than those obtained by 'ratio to moving liable to be more biaseu
average' method discussed in § 2-5:3
The obvious advantage of this method over
the moving
that 'ratio to trend' can be average method lies in the
obtained for each
Such, unlike the 'ratio to moving average method,month for which the data are avallable
there is no loss of data. a
Remark. The calculations are
vearly totals averages) and then
(or
simplified to a
great extent by first
fitting a trend equation
modification of the trend equation, as obtaining the monthly
illustrated in the
(or
quarterly) trend values by a su

following example:
Year 1Qrt. I1 Qrt. III Qrt
1995 30
Exampé 2-16. Using Ratio to Trend 40 36
methoddetermine the quarterly seasonal 1996 34 44
52 50
indices for the adjoining data:
1997 40 58 54
1998 54 76 68
1999 80 92 86
 2.44 FUNDAMENTALS OF APPLIED STATISTICS
TABLE 218: COMPUTATION OF LINEAR TREND
Total of Average of| Trend First of all, we will determine
quarterlyquarterl values the trend values for the
Year y
values values (y) =t- 1997 quarterly by fitting
(Million averages
(MillionRs) Rs.)
a linear trend by the method
140 35 -2 (c.f. Remark
1995 4 70 32 of least squares
above, $ 2-5-2).
1996 180 45 -1 1 5 Let the straight 1line trend
44
200 50
equation be:
1997 0 0 56 y =a + bx ..(*)
260 65 Origin 1997, x unit: 1 year)
1998 65 68 y Quarterly averages
1999 340 85 2 (Million Rs.)
4 170 80

Total y 280 x =0
Exy
|=10 120
The normal equations for estimating a and b in (*) are

280
bEx a = 56
y na +
5
xy = a x +b 120
b = 1 10

Hence, the straight line trend is given by the equation

Ye 56 + 12x lorigin 1997, x unit = 1 year] ..(*)

Putting x =
-2, -1, 0, 1, 2,
obtain the average quarterly trend values for the
we
years
1995 to 1999 respectively, which are given in the last column of the Table 2.18.

From the trend equation (*), we observe that:

Yearly increment in trend values = b = 12 Quarterly increment = = 3

The positive value of b implies that have

we an increasing trend. NeKt, we determine the
quarterly trend values as follows
For the year 1995, the average quarterly trend value is
32 which is, in fact, the trend
value for the middle quarter, i.e., half of the 2nd quarter and half of 3rd quarter, of 1995.
ince the quarterly increment is 3, we obtain the trend values for the 2nd and 3rd quarters of
year 1995 as 32 1-5 and 32
-

1-5, i.e., 30-5 and 33.5 respectively and consequently the

rend value for first quarter is 305-3 27-5 and 4th quarter is 33-5 +3 36-5. Similarly we
Can get the trend values for other
years as given in the following table.
 6D o.992y

, u3
22

ANALYSIs OF TIME SERIES

2-45
TABLE 2-19: COMPUTATION OF SEASONAL INDICES
Year Trend Values Trend Eliminated Values
(Given valyes aso of trend values)
1Qrt I1 Qrt. IIl Qrt
1995
Il1 Qrt. IV Qrt. Qrt 1Rrt IV Qrt
27-5 30-5 33-5 36-5 09-1 1311 1075 93.1
1996 39-5 425 45.5 48-5 86 1 122-4 109.9 90-7
1997 51-5 545 57-5 60-5 77-7 106-4 93-9 70-3
1998 63-5 66-5 69-5 85-0 114-3
72-5 97-8 85-5
1999
5 78-5 81-5 84-5 106-0 117-1 105-5 97-0
Total 436-9 591-3 514-6 445-6
Average (A.M.) (Seasonal Indices)
92-38 118-26 102-92 89.12
Adjusted Seasonal Indices -
92-07 117-36 102-14 88-44
The indices obtained above are
adjusted to
+118-26 + 102.92+ 89.12 403-08, which is
a total of 400 (since the sum of
indices 92-78
=

greater than 400) by multiplying each

them of
by constant factor k, called correction
a
400
factor given by:k Sum of 400 =

2-5-3. Ratio to Moving Average Method. As 403-08 0-9924

indices 403-08 0-9924

eliminates periodic movements if the extent (period of pointed out earlier, moving average
of theoscillatory movements sought to be eliminated. moving average) is equal to the period
Thus for a monthly data, a 12-month
moving average should completely eliminate the seasonal
pattern and intensity. The method of getting seasonal movements if they are of constant
indices by moving
followingsteps involves the average
Calculate
values
the centred 12-month moving
average of the data. These moving average
will give estimates of the combined effects
of trend and
(ii) Express the
original data (except for 6 months in the cyclic variations.
end) percentages of the centred moving
as
beginniFg and
eginning and 6 months
months atat the
tne
model, these percentages would then average values.) Using multiplican
components. represent the seasonal and irregula
(i) The preliminary seasonal indices
are
random component by averaging these now obtained by eliminating the
either arithmatic mean or median irregular
percentages. As discussed in $ 2.5-2, Step t
(iv) The sum of (preferably median) be
these indices S (say) will not, in
=
can used for averag1n
(or quarterly) data. general, be 1,200 (or 400) for monthly
Finally, an
1,200 (or 400) by multiplying adjustment is done to make the sum ndiees

i.e., by expressing the throughout by a constant factor of the

preliminary seasonal indices as the
apithmetic mean. The resultant gives the 1,200/S (or i =

ther

Merits and Demerits. Of all desired indices of percentage o

the moving average the methods of measuring seasonal variations
'method is the most seasonal atiot
estimating the seasonal
cyclical components from 1luctuations
satistactory, variations.tnd
in a time series flexible and widely used me
the indices because it irons out both tre
However, an obvious drawback ofofthisseasonal variations.
the beginning and at the
end method is that there Valuesi n

quarters) of the first and accordingly

year and last six
is loss
seasonal indices for of some trend (or

determined. months (or 2 first six mon Cannot be

quarters) of the last

ast year
2.46 FUNDAMENTALS OF APPLIED STATISTICS
amark.
Rem Additive Model. If we
additive model of the time
use
series. then the method of moving
a v e r
computing
a g e s
for seasonal indices involves the
following steps. [We shall state the steps 1Or
d these can be modified accordingly
onthly data and
for quarterly and other
data.)
Obtain 12-month moving average values. These will contain trend
represent (7T+ C).
and cyclic components, 1.e,
theywill
i) Trend eliminated values are obtained on
s e r i e s values to give:
subtracting these moving average values from the
giventime
y M.A. values (T +S + C + =
)-(T + C) S+I =

ii) Irregular component is eliminated on

averaging these (S + D values for each month over
different years and we get the preliminary indices for each month.
(iv) Sum of the indices should be zero. In case it is not
the preliminary indices in step (iii)
adiusted to a total of zero by subtracting from each of them a so,
constant factor,
are

k Sum of monthly seasonal

indices
Exampe 2.17. Apply ratio to
moving average method to ascertain -seasonal indices from
the following tat :
Year and No. of persons Year and No. of persons Year and No. of persons
Month visiting a place Month visiting a place Month visiting a place
2002 of interest 2003 of interest 2004 of interest
Jan. 90 Jan. 100
Feb. Jan. 110
85 Feb. 89
March Feb. 93
70 March 74 March 78
April 60 April 62
May 55 April 66
May 55 May 58
June 45 June 47 June 40
July 30
July 30
Aug. 40
July 35
Aug. 43 Aug. 45
Sept. 70 Sept. 65
Oct. 120
Sept. 72
Oct. 127 Oct. 130
Nov. 115 Nov. 118
Dec. 118
Nov. 118
Dec. 120 Dec. 124
Solution.
TABLE 2-20 COMPUTATION OF MOVING AVERAGES im
Year & Month No. of persons visiting 12-point 12-point M. 12-point MA.
a place of interest moving totals centered Hatio to M1A.
(1) (2) (3) (4) (3)
=
+2 (5) A2) + (5)] x 100
2002 Jan. 90
Feb. 85
March 70
April 60
May 55
June 45
898 748
July 30
908 75-67 75-3 39-85 S
Aug 40
75-8
Sept. 70
912 76-00 528 0
916 76-2 91-9
Oct. 120 76-33
76-4 157-1
O
2-48

FUNDAMENTALS OF APPLIED STATISTICS

Example 2.18. Calculatë seasonal indices
following daka: by the ratio to moving average method from
Year Prices (in 10
the
Rupees
2001 per kg.)
Quarter 2002
2003 2004
75
60 86
65 90 100
54 72
Q 59 63
80 66
78
72
Solution. Assuming multiplicative model of time series, the
85 93
values eliminated by expressing
are
trend (Moving
are
the Average
given in the last column (6) of the given values (y,) as a percentage of trend values and
following Table 2-22.
TABLE 2-22:
COMPUTATION OF
MOVING
Year Price
4-Quarterly 2-Period of Col (3) AVEBAGES
Quarter Moving Totals 4-Quarter
Moving Totals Moving Average Ratio to Moving y -M.A.
(1) (2) Average
(3) (4) (6) =
(4) +
( (6) [(2)
2001 75 =
+ (6)]x 100 (7)
II 60
III 54 248
63-376 SA;63:3
IV 59 259 507 85-2071 - 9-375

2002 I 86 264 523 65-375 s9LCS 90-2485 - 6-375

II 65 273 587 67 125 128.1192 18-875

III 63 294 567 70-875 91-7108 - 5-875

IV 80 298 592 74-000 85-1351 - 11-000

2003 I 90 305 603 75-375 106 1360 4-625
II 72 308 613 76-625 117-4551 13-375
I 66 313 621 77-625 92-7536 - 5-625

323 636 79500 83-0189

IV 85 -13.500
2004 I 100 329 652 81-500 104-2945 3-500
II 335 664 83-000 120-4819 17-000
78 678
343 84 750
III 72 92-0354 - 6-750

IV 93
TABLE 2-23 COMPUTATION OF SEASONAL INDICES
(MULTIPLICATIVE MODEL)
Year Trend Eliminated Values
1Qrt. I1 Qrt. IIlQrt. IVQrt.
2001 85-2071 90-2485
2002 128-1192 91-7108 85-1351 106.1360
2003 117-4551 92-7536 83-0189 104-2945
2004 120-4819 92.0354
Total 366-0562 276-4998 253.3611 300-6790
Average (A.M.) (S.I 122-0187 92-1666 84-4537 100-226 Total 398-865
Adjusted Seasonal Indices 122 3603 92-4246 84 6902 100-5066 399.985 - 400
= (S. I)x k
 ANALYSIS OF TIME SERIES
2.49
The seasonal indices obtained as averages (A.M.) above are adjusted to a total of 400. hw
multiplying each of them by a constant factor, by
R 400
S u m of Seasonal Indices 400
398-865 IT0028

If
we assume additive model of the time series, then the trend eliminated values, (short
term and irregular fluctuations), are obtained on subtracting the trend (M.A.) values frorm
the given time series values
(y,) i.e., by the formulae:
Short-term fluctuations = y, - (M.A. values) = S +I, and are given in the last column (7
of Table 2.22. These values are then used to obtain the seasonal indices as given in
the last
row of following Table 2.24.

TABLE 2.24: COMPUTATION OF SEASONAL INDICES (ADDITIVE MODEL)

Trend Eliminated Values
Year
1Qrt. II Qrt. III Qrt. IVQrt.
2001 - 9.375 - 6-375

2002 18-875 - 5-875 - 11-000 4-625

2003 13.375 - 5-625 - 13.500
3-500
2004 17-000 - 6-750

Total 49-250 - 18-250 - 33-8755 1-750

Average (A.M.) (S.I.) 16-417 - 6-083 - 11-292 0-583 Total =-0-375
Adjusted Seasonal Indices
16-511 - 5-989 - 11.1988
0-677 Total = 0-001 0
= (S. I. - k)

The seasonal indices obtained as averages (A. M.) above are adjusted to a total zero by
subtracting from each of them a constant factor (k):

13/202
2 0 1 k Sum of seasonal indices 0-375 -0-094.
254.Link Relative Method. This method, also known as Pearson's method, is base
on averaging the link relatives. Link relative is the value of one season expressed as
percentage of the value of the preceding season. Here, the word 'season' refers to time periad
it would mean month for monthly data, quarter for quarterly data, etc. Thus, for monthy
data

Link relative for any month Current

= months value 100
Previous month's yalue 100
(2-40)

The steps involved in this nmethod may be summed up as follows

Convert the original data into link relatives (L.R.) by the formula in (2-40).
e a c h month
i i As in the case of 'ratio to trend' method/ average the link relatives for eacn for
(quarter) of the data are monthly (quarterly). Mean or median may be
averaging.\ b a s e

(iii) Convert the average (Mean or Medan) link relatives into chain relat1ves on
of the first season. Chain relative (C.R.) for any season is
inlyi
obtained on
m
the link relative of that season by the chain relative of the
dividing by 100, i.e.,
preceding sea
2-50
FUNDAMENTALS OF APPLIED STATISTICS
C.R. for any season LR.of that season) x (C. R of preceding season
100 (240a)
CR. for first season is taken as 100.
Thus for monthly
T data, the chain relative for first
(month), for
be 100.
season i.e.. January, 1s
taken to
LR. of Feb. x CR. of Jan
CR for February LR. of Feb. (C.R. of Jan = 100)
100

C.R. for March LR. of March x C.R. for Feb.

100

CR. for December =Lk. ot Dec. x C.R. for No.

100
Now, by taking this December value as a base, a new chain relative for January can be
obtained as
New C.R. for January= LR. of January xCR for December
(2406)
100
Usually, this will not be 100 due to trend and so we have to adjust the chain relatives for
trend.
(iw) This adjustment is done by substracting a 'correction factor from each chain relative.
If we write

d Second (New) C.R for January -100 241)

then, assuming linear trend, the correction factor for February, March,.. December is d,
2d,, 11d respectively.
() Finally, adjust the corrected chain relatives to total 1,200 by expressing the corrected
chain relatives as percentages of their arithmetic mean. The resultant gives the adjusted
monthly indices of seasonal variatians.
Merits apd Demerits. (i) The link relatives averaged together contain both the trend
and cycliccómponents. Although thetrendis subsequently eliminated by applving correction.
themethod is effect+ve only if the growth is of constant amount or constant rate, i.e., if the
assumption of linear trend is valid.
( Thouglh not so simple as the moving average method. or so readily adaptable as
others to the construction of some or more complex types of seasonal movements, the actual
computations of the link relative method are much less extensive.
(iii) This method utilises data more completely than moving average method as there is
loss of only one link relative, i.e., for the tirst season. while in case ot moVing average methvd
we are deprived of some of the values (trend and seasonal) in the beginning and at the end

Year 1996 1997 1998 1999 2000

Example/24. Compute the seasonalQuar
31 34
indices by the Link Relatives' method for
the adjoining data relating to the average
I 30 35 31
II 26 28 29 31 36
quarterly prices (Rs. per kg.) of a III 22 22 28 25 26
commodity for five years: 36 36 32 35 33
ANALYSIS OF TIME SERIES
25
Solution.
TABLE 2.25: COMPUTATION OF SEASONAL INDICES BY LINK RELATIVES METH
OD
Link Relatives
Year IQrt. II Qrt. IIi Qrt.
1996 86-7 84-6
TVQrt.
1997
1409
112-9 80-0 78-6
1998
163-6
86-1 935 96.6
1999
1143
96:9 100-0 80-7
2000
140-0
97-1 105-9 72:2
Total of LR:s
1269
393 4461 442-7
685-7
Average L.R. (A.M.) 393
98-25
4
4461 93:22
593-22 442:
5
82-54 685-7
5 137 14
93-22x 100 93-22 x 82-54 76-95 x 137-14
Chain Relatives 100 100 100 100
93-22 = 76:95 = 105-4

Adjusted C.R:'s 100

93-22-0-875 76-95-1-75 105-4 -2-625
92.345 75-2 = 102 775
Seasonal Indices 108-02 99-75 81-23 111-0
Explanation of Steps:
The new (second) C.R. for 1st quarter is
L.R. of1st Qrt. x C.R.of last (4th) Qrt.105-4 x 98-25
100 100 = 103-5

Correction factor, d 3.5

= (103-5 100-0) ="= 0.875
Adjusted chain relatives are obtained by
2-625 from the chain relatives of
second, third
subtracting 0-875, 2 x 0-875 =
1-75,3 x
and fourth
quarters respectively.
Average of Adjusted Chain Relatives= 100+92-345
+75-200102775 99.58
Seasonal Variation Index for any quarter Adjusted C.R.
x 100
Seasonal indices have been obtained in the 92-58
last ro of
2-5-5.
De-seasonalisation of Data. As already Table 225.
seasonal variations is (i) to measure
them and (ii) to pointed out, the objective
Elimination of the seasonal effects from eliminate them from the s i
the data. It helps the given
to
us adjust the given time values is termed as
with trend
component, cyclic and irregular series for de-seaso
seasonal variations, thu
time series, the movements.
de-seasonalised (seasonality
given values by the eliminated)Assuming
corresponding indices of seasonal multiplicativeyi
values are obtained o
variations.
Deasonalised Data == S TC
Deasonalisation is
specially needed for the
businessmen and management
TCI I ta
. l s a"

study of cyclic
forecasting, and managerial executives
contro.
for
planning
It also
componenafuture production pro
helps in
proper interpretation o
e da