SERIES

613
ANALYSIS OF TIME
612 STATISTICAL METHODS

The Method of Least Squares

Merits and Limitations practice. It is a mathematical method
This method is most widely used in to the data in suCn a manner that
and with its help a trend line is fitted
Merits. The method of moving
averages has the followving advantages : the following two conditions are satisfied:
squares.
compared to the method of least
" This method js cimnle as (1) E (Y- Y) = 0, computed
the reason that if a few actual values of Y and the
" It is a flexible method of measuring trend tor i.e., the sum of deviations of the
the entire calculations are not
more figures are added to thhe data, values.
values of Y is zero.
Changed-we onlyget some more trend (2) E (Y- Y is least.
happens to coincide with the period deviations of the actual and computed
" If the period of movingin average i.e., the sum of the squares of the hence the name method of least
of cvclical fluctuations the ata. such fluctuations are automaticaily values is least from this line and
eliminated. squares. The line obtained by this method is known 'as the line of best
" The moving average has the advantage that it follows the general fit. a straight line trend
movements of the data and that its shape is determined by
the data The method of least squares may be used either to fit
rather than the statistician's choice of a mathematical function. of a parabolic trend.
. It is particularly effective if the trend of a series is very iregular. The straight line trend is represented by the equation
Y= a+ bX
are, however, Some limitations of this mnethod too. from
Limitations. There where Yc is used to destgnate the trent values to distinguish them
These are : the actual values, a is the Y intercept or the computed trend figure of
. Trend values cannot be computed for all the years. The longer the the Y variable when X=0. b represents the slope of the trend line or
period of moving average. the greater the number of years for which amount of change in Y variable that is associated with a change of one
trend values cannot be obtained. For example, in a three-yearly unit in'X variable. The Xvariable in time series analysis represents time.
moving average, trend values cannot be obtained for the first year Whenever we fit any straight line trend by the least squares method, three
and last year. in a five-yearly moving average for the first two years things should be specified:
and the last two years. and so on.
(1) Which year was selected as the origin ?
" Great care has to be exercised in selecting the period of moving (2) What is the unit of time represented by X ? Is it half year, one year or
average. No hard and fast rules are available for the choice of the five years ?
period and one has to use his own judgment. (3) In what kind of units is Y being measured? Is it production in tonnes,
Since the moving average is not represented by a mathematical function sales in rupees, price in rupees,employment in thousands of workers ?
this method cannot be used in forecasting which is one of the main In order to determine the values of the constants a and b the following two
objectives of trend analysis. normal equations are to be solved
" Although theoretically we say that if the period of moving average 2Y= Na +bX ...()
happens to coincide with the period of cycle, the cyclical fluctuations EXY= aEX+ b X
are completely eliminated. but in practice since the cycles are by no
means perfectly periodic, the lengths of the various cycles in any where N represerts Tmumber of years (months or any other period) for
given series will· usually vary considerably and, therefore, no moving which data are given.
average can completely remove the cycle. The best result would be It should be noted that the first equation is merely the summation of the
obtained by a moving average whose period was equal to the average given function. the second is the summation of X multiplied by the given
function.
length of all the cycles in the given series. However, it is diffcult to
determine the average length of the cycle until the cycles are isolated We can measure the variable Xfrom any point of time in origin such as the
from the series.
first year. But the calculations are very much simplified when the mid-point in
Finally. when the trend situation is not linear (a straight line) the time is taken as the origin because in that case the negative values in the first
moving average lies either above or below the true sweep of the data. half of the series balance out the positive values in the second half so that
Consequently, the moving average is appropriate for trend computations £X=0. In other words, the time variable is measured as a deviation from its
only when : mean. Since 2 X= 0the above two normal equations would take the form.
(a) the purpose of investigation does not call for current EY= Na ...()
or forecasting. analysis
EXY= b X ...()
(b) the trend is linear, and
(c) the cyclical variations are regular both in period The values of a and b can now be determined easily.
and amplitudes. Since 2Y= Na
Unfortunately, these conditions are encountered very
infrequently.
614
STATISTICAL METHODS ANALYSIS OF TIME SERIES 615

Solution. (0 FITTING THE STRAIGHT LINE TREND

Q 0r

XY
N Year Production X x? Trend values
(m. tonnes) Yc
Since SXY= b X
240 9 84
2001 80
The constant 'a is simply equal to the mean of Yvalues and the constant 'b 2002 180 86*
90 -2
gives the rate of change. 2003 92 1 92 1 88
It should be noted that in case of odd number of years, when the deviations 90
are taken from themiddle year Xwould always be zero provided there is no 2004 83
+ 94 1 92
gap in the data given. However, in case of even years also EX will be zero if 2005 94 + 1
+ 198 4 94
the X origin is placed midway between the two middle years. For example, if 2006 99 +2
the vears are 2002, 2003. 2004. 2005, 2006 and 2007, we can take deviations 2007 92 +3 +276 96
from the middle year 2004.5. Thus the deviations would be N=7 EXY=56 x'=28 E Yo= 630
£ Y= 630 EX=0
-2.5. - 1.5, - 0.5, + 0.5, + 1.5, + 2.5 for the varivus years and the total X
would be zero. Hence both in odd as well as in even number of years we can The equation of the straight line is Yc= a+bX.
use the simple procedure of determining the values of the constants a and b. Since E X=0 ;a=
EXY
The arithmeticstraight line trend fitted by the method of least squares is
by far the most widely applied trend curve. This particular trend curve is Y-630, N=7, EXY=56, E*= 28,
applicable for those series in which period-to-period changes are constant in a =
630 = 90;and b= 56 =2
absolute amount. 7 28
A quick method for assessing the appropriateness of the straight line mnodel Hence the equation of the straight line trend is Ye= 90 +2 X;
is the method of frst differences. If the differences between successive Origin, 2004; Xunits, one year; Yunits, production in milion tonnes.
observations of a series are constant (or nearly so), the arithmetic straight line For X=-3, Yo= 90 +2(-3) = 84
should be taken to be an appropriate representation of the trend component. For X= -2, yc =90 +2(-2) =86
The method of first differences is explained below : For X= -1,yc= 90 +2 (- 1)= 88.
METHOD OF FIRST DIFFERENCES Similarly, by putting X= 0, 1,2, 3, we can obtain other trend values. However, since the value
of b is constant, first trend value need be obtained and then if the value b is positive we may
Year Sales ('000 units) First differences Continue adding the value of bto every preceding value. For 2002 it will be 84 +2= 86, for 2003
2000 120 20 it willbe 86 +2 =88, and so on. If bis negative, then instead of adding we will deduct.
2001 140 22 (i) The graph of the above data is given below:
2002 162 18
2003 180 23 Linear Trend by the Method of Least Squares
100
2004 203 22
2005 225 20
2006
2007
245
268
20
23
(tPomrndiulcetosn) 95 Trend Line

It is clear from the above series that the differences in successive

observations are nearly constant and hence the arithmetic straight line is an
appropriate model for assessing the trend component of the series. 85 Actual Data
It may be pointed out that very few time series exhibit this type of constant
change over a period of time, say, over a period of several business cycles. It is
often necessary to fit other types of lines or curves. 80

liistration10. Below are given the figures of production (milion tonnes) of a sugar factory :
Year 2001 2002 2003 2004 2005 2006 2007 79
Production (m. tonnes) 80 90 92 83 94 99 92
() Fit a straight line trend to these figures. 0
(i) Plot these figures on a graph and show the trend line. 2001 2002 2003 2004 2005 2006 2007
([Link]., HPU: [Link],. Bangalore Uniw.) Years
616
STATISTICAL METHODs ANALYSIS OF TIME SERIES
617

f instead of middle year as origin, we take fst year as ongin the soluton would be as folowe : EXY 136
Production XY x2 Y and b= = 4.857
Year E28
(m. tonnes)
Hence Y= 76 + 4.857 X
80 84 For 2012. X will be + 8 Y012 = 76 + 4.857 (8) = 114.86.
2001 90 1 Thus the likely sales for the year 2012 is 114.86 m. tonnes.
90 86
2002 184 4
folowing
IlJástration 12. Fit a straight line trend by the method of least squares to the predicted
92 2 88
2003
3 249 9 |data. Assuming that the same rate of change continues, what would be the
83 90
2004 earning for the year 2009? 2004 2005 2006 2007
94 4 376 16 92 Year 2000 2001. 2002 2003
2005 40 65 72 69 60 87 95
2006
5 495 25 94 Earnings (As. lakhs) 38
6 552 36 Do not plot the trend values on the graph.
2007 92 96 ([Link]., M.D. Uniu.; [Link].. M.K. Uniu.; [Link]., Osmania Unw., 2007)
EY= 630 )X- 21 XY =1,946 x²=91 E Ye= 630
N=7 Solution. FITTING OF STRAIGHT LINE TREND BY
Ye=a+ bX; Y= Na+ b X; X Y= aX+ b X? THE METHOD OF LEAST SQUARES
Substituting the value, 630 = 7a+21 b ..() Deviations Deviations
1946 =21 a +91b
Year Earnings
...(0) (As. lakhs) from 2003.5 multiplied by 2
Multiplying Eqn. () by 3, 1890 =21 a+ 63b X XY
1946 = 21 a+91 b
2000 38 3.5 -7 266 49
- 28 b=-56 or b= 2 2001 40 2.5 -5 200 25
2002 65 -1.5 -3 - 195
Substituting the value of b in Egn. (), -1 -72 1
630 = 7 a+21 (2) 2003 72 -0.5
7a=630 42 =588 or a=84 2004 69 +0.5 +1 + 69 1
+ 1.5 +3 + 180
Thus the equation is Yo= 84 +2 X 2005 60
tonnes. + 435 25
Origin 2001 : X units, one year; Y units, production in million 2006 87 +2.5 +5
2007 95 +3.5 +7 + 665 49
For X=0, Y=84
the
Note : The difference in the two equations is because of the difference in origin. Inoriain N=8 £ Y=526 E X=0 EXY=616 X'=168
first case 2004 was taken as origin whereas in the second case 2001 was taken as
However, trend values are the same. The equation of the straight line trend is
Illustration 11. Fit a straight line trend for the following series. Estimate the value for 2012: a=
EY 526 = 65.75
N8
Year 2001 2002 2003 2004 2005 2006 2007 EXY 616
Production of Steel 60 72 75 65 80 85 95 b= 3.667; Y= 65.75+3.667 X.
168
(m. tonnes) For 2009, X will be + 11.
(B.A., (H) Econ., Delhi Unw. 2007) When Xis 11, Ywill be
Y=65.75+3.667 (11) =65.75 +40.337 = 106.087.
Solution. FITTING STRÄIGHT LINE TREND Thus the estimated earnings for the year 2009 are Rs. 106.087 lakhs.
Year Production of Steel Deviations bo) XY The same result willbe obtained if we do not multiply the deviations by 2. But in that case our
from 2004 computations would be more difficult as could be seen below:
(m. tonnes)
Year Sales in Deviations from
thousands of 2003.5 XY
2001 : 60 -3 - 180
rupees
2002 72 -2 - 144 4
Y
2003 75 -1 -75 1
2000 38 -3.5 133.00 12.25
2004 65 2001 40 -2.5 -100.00 6.25
2005 80 +1 +80 1 2002 65 -1.5 97.50 2.25
4 2003 72 - 0.5 36.00 0.25
2006 85 +2 + 170
2004 69 +0.5 + 34.50 0.25
2007 95 +3 + 285
2005 60 + 1.5 + 90.00 2.25
N=7 EY=532 EX=0 EXY=136 x'=28 6.25
2006 87 +2.5 +217.50
The equation of the straight line trend is 2007 95 +3.5 + 332.50 12.25
Y= a+bX N=8 2Y=526 2 X=0 £XY=308 Ex'=42.00
Since EX=0;a EY_ 532
N=76
 618 STATISTICAL METHODS
ANALYSIS OF TIME SERIES 619

Y 526 = 65.75: b e X Y 308

=7.333
42 when X=0, Y- 88,803
The advantage of this method is that the value of b gives annual increment of change rahee when X= +1, Y= 88.803 + 1.38 - 90.183
than 6-monthiy increment, as in the first method discussed above. Hence we will not ho when X= +2, Y= 88.803 + (1.38 x 2) = 91.563
double the value of b to obtain yearly increment. It is clear from the above illustration thati o when X= +5, Y=88.803+ (1.38 x5) =95.703
first case the value of b is half of what we obtained from the second method (b was 367 in (i)After eliminating the trend we are left with cyclical and irreqular variations.
first case and 7.333 in the second case). (üin The monthly increase in the production of sugar is
llustration 13. Below are given the figures of production (in million tonnes) ot a sugar b/12, i.e., 1.38/12 = 0.115 m. tonnes.
factory : Merits and Limitations.
Year Production Year Production
(m. tonnes) (m. tonnes) Merits. The merits of this method can be enumerated here:
2001 77 2005 91
2002 88 2006 This is a mathematical method of measuring trend and as such there
98
2003 2007 is no possibility of subjectiveness.
90
2004 85
The line obtained by this method is called the line of best fit because
(Ò Fit a straight line by the 'least squares' method and tabulate the trend values it is this line from where the sun of the positive and negative
(in Eliminate the trend. What components of the series are thus left over ? deviations is zero and the sum of the squares of the deviations least,
(i) What is monthly increase in the production of sugar ?
M Com Jamnu Uni.: M. Com.. Osimania Uniu.: B. Com. Kerala(M.A., Kurukshetra Uniu.: i.e.. E (Y- Y = 0and E (Y- Y least.
Uniu.: AIMA uno:
M.A. Econ.. GND Univ.) " Trend values can be obtained for all the given time periods in the
Solution. () The equation of the straight line trend is Yo=a+bX series. This is not possible in some other methods like method of
Since E Xis not zero, the value of aand b will be obtained directly by solving moving averages.
normal equations: the following two
Limitations. The limitations of the method are given below:
EY= Na+bEX ..()
E XY= aEX+ bE X " Great care has to be exercised in selecting the type of trend curve to
Production Taking 2004
Year (m. tonnes) Trend value be fitted, ie., linear, parabolic or some other type. Carelessness in
as ongin XY x? Yo (Y- Y
this respect may lead to fallacious results.
X
" This method is more tedious and time-consuming compared to other
2001 77 4 308 16 83.2831. methods.
-6.283
2002 88 -2 -176 4 86.043. +1.957 " Predictions are based only on long-term variations, ie., trend and the
2003 94 - 1/ 94 87.423 +6.577 impact of cyclical, seasonal and iregular variations is ignored.
2004 85 0 88.803 Being a mathematical method it is not flexible-the addition of even
3.803
2005 91 +1 + 91 1 90.183 +0.817 one more observation makes it necessary to do all the computations
2006 98 +2 + 196 91.563 +6.437 again.
2007 90 +5. + 450 25 95.703 - 5.703 The method cannot be used to fit the growth curves like Gompertz
N=7 Y=623 EX=‘ :XY =+ 159 x?=51 E Yo=623 E(Y- Yo) =0 curve, logistic curve, etc.
Hence, mathematical methods of fitting trend are not foolproof-in fact, they
623=73+ b
159¬ a+51 b ...) can be the source of some of the most serious errors that are made in
Multiplying the second equation by 7, we get ...i) statistical work. They should never be used unless rigidly controlled by a
1113=7 a+357 b separate logical analysis. Trend fitting depends upon the judgment of the
Deductingequation (i) from {) ...(in statistician, and a skilfully made freehand sketch is oftern more practical than
- 490 =-356 b 490 a refined mathematical formula. *
b= 356 1.38
Substituting the value of bin equation (i Second Degree Parabola. The simplest example of the non-linear trend
623 =7 a+ 1.38
7a= 623 - 1.38= 621.62
is the second degree parabola, the equation of which is written in the
So the equation of straight line trend is or a= 88.803 form :
Y= 88.803 + 1.38 X Y= a+bX+ cX
when X=-4, Y= 88.803 + 1.38
when X=-2. Y= 88.803 + 1.38 (-4) =88.803 - 5.52 = 83.283
(-2) =
where c is still the Y intercept, b is the slope of the curve at the origin
when X=-1, Y= 88.803 -1.38 = 87.42388.803 -2.76 = 86.043 Riggleman and Frisbee : Business Statistics, p. 312.
 620
STATISTICAL METHODS
621
ANALYSIS OF TIME SERIES
and cis the rate of change in the slope. When numerical values for o
and chave been derived, the trend value for any year may be computed we solve the following normal equations:
by substituting in the equation the value of X for that year. The values of Solution: To determine the values of a, b and
ab and c can be determined by solving the following three normal egua. Y- Na +bE X+ cEX
tions simultaneously 2XY= a L X+ bEx+c X
£Y= Na +bX+ cEX ...() zxY=aLX+b+cEX
EXY= aX+bX+cEß Year Prices Trend
.(i) XY Values
Y=ax+bE+cEx ..(ü)
(As.)
(Yc)
Note that the first equation is merely the summation of the given function, 2002 100 2 16 -200 400 97.717
the second is the summation of X multiplied into the given function, and the 2003 107 1 -1 1 - 107 107 110.401
third is the summation of X multiplied into the given function. 2004 128 0 0 126.657
When time origin is taken between two middle years, 2 Xwould be zero, In 2005 140 +1 1 + 1 1 + 140 140 146.485
that case the above equations are reduced to 2006 181 +2 4 + 8 16 + 362 724 169.885
1728 196.857
£Y= Na+cE X ..()
2007 192 +3 9 +27 81 +576

..()
N=6 2Y=848 EX=3 z=19 -27 X- 115 EXY
=771 =3,099 848.002
EXY=aX+cEX*
..(ü) 848 =6 a+3 b+19 c ...)
The value of b can now directly be obtained fromequation (i) and that of a ...(i)
and c solving () and (ü) simultaneously. Thus 771 =3 a+ 19 b+27 c
3,099 = 19 a+27 b+ 115 c ..(ü)
E()-c b=
EXY C=
NXn-2(x z(Y Multiplying the second equation by 2 and keeping the first as it is, we get
a=
N NE (X-(2X 848 =6 a+3b+ 19 c
1,542 = 6 a+38 b+54 c
Asecond degree trend equation is considered to be an appropriate model for
the secular trend component of a time series when the data appear not to fall
in a straight line. A test of the appropriateness of a second degree equation in 694 =-35 b-35c ...(iv)
modelling a given time series is based on the method of second differences., or 35 b+35 C= 694
When the differences are taken of the first differences and the result is a Multiplying Eqn. () by 19 and Eqn. (i) by 3, we get
constant (or nearly so) between successive observations of the given time 14.649 = 57 a+ 361 b+513c
series. a second degree equation may be an appropriate model for measuring 9.297 =57 a+ 81 b+ 345 c
the secular trend.
5,352 = 280 b+ 168 c
The method of second differences is explained below : Multiplying equation (iv) by 8, we have
METHOD OF SECOND DIFFERENCES 280 b + 280 c= 5.552
Solving equations (iv) and (v)
X Y= 6+2 X+ 1.5 ² First differences Second differences 280 b+280 c=5,552
9.5 280 b+ 168 c= 5,352
16.0 6.5
25.5 9.5 112 c=200 or C=1.786
4 38.0 12.5 3
Substituting the value of c in Eqn. (iv),
5 53.5 15.5 35 b+ (35 x 1.786) = 694
6 72.0 18.5 35 b= 694 - 62.5 631.5 or b=18.042
7 93.5 21.5 3 848 =6 a+3(18.042) +19 (1.786) =6 a +54.126 +33.934
118.0 24.5 6 a=759.94 or a= 126.657
Thus a= 126.657, b= 18.042 and c= 1.786
It is clear from above that the second differences are constant throughout. Substituting these values in the equation,
llustration 14. The prices of a commodity during 2002-2007 are yven below. Fit a Y= 126.657 + 18.042 X+ 1.786 X*
parabola Y- a+bX+cX* to these data. Estimate the price of the commodity for the year when X -2
2008: Y- 126.657 + 18.042 (-2) + 1.786 (-2)
Year Prices Year Prices = 126.657 - 36.084 + 7.144 - 97.717
2002 100 2005 140 when X=-1
2003 107 2006 181 Y= 126.657 + 18.042(-1) +1.786 (- 1)
2004 128 = 126.657 - 18.042 + 1.786 = 110.401
2007 192
Also plot the actual and trend values on the graph. when X= 1,
([Link]., M.D. Univ., 2002 (adapted): [Link]. (H), Delhi Uni., 2006)
 622 STATISTICAL METHODS
Y= 126.657 +18.042 +1.786 146.485
when X = 2,
Y- 126.657 +18.042 (2) +1.786 (2)- 169.,885
when X= 3.
Y= 126.657 +18.042 (3) +1.786 (3)- 196.857
Price for the year 2008 equation.
For 2008 X would be equal to 4. Putting X= 4 in the
Y= 126.657 + 18.042 (4) + 1.786 (4)
= 126.657 + 72.168 + 28.576-227.401
Thus the likely price of the commodity for the year 2008 is Rs. 227.41 aporox
below:
The graph of the actual and trend values is given

Second Degree Parabola

(Y=a +bx +cx)
200 -

180

(PrRisc.e) 160 Trend Line

L 140

120 Actual Data

100

2002 2003 2004 2005 2006 2007

Years

MEASURING TRENDS BY LOGARITHMS

The trends discussed so far were plotted on arithmetic
also be plotted on a semi-logarithmic (or semi-log) chartscales.
in the
Trends may
form of a
sraight líne or a non-linear curve. A straight ine on the semi-log
shows the increase of Y values of a time chart
series at
straight line on an arithmetic chart indicates the increase a constant rate (a
amount). When a given time series is increasing or at a constant
stant rate and is not approaching some upper limit, decreasing at a Coi
proximated leasSt by an exponential curve. It should be it is usually ap
are talking of Constant rate and not constant noted that here wo
absolute amount.
logarithmic model is suggested whenever a scatter liagram Also Sen
points of the series on semi-logarithmic graph paper of the data
line. When it is a non-linear curve on the indicates a stralgt
semi-log
shows the increase at varying rates, depending on the chart an upward culve
slopes-the steeper the slope, the higher is the rate of shapes of the
The types of trend usually computed by logarithms increase.
are :