TIME SERIES ANALYSIS

Basic Components of a Time Series;

Time Series Analysis;
Estimation of Trend; and
Estimation of Seasoned Variation.

OBJECTIVES
At the end of the topic, readers should be able to
a) know the meaning of a time series;
b) know the basic components of a time series;
c) estimate the trend by the use of moving averages and least squares methods;
d) carry out exponential smoothening; and
e) estimate seasonal index.

Introduction
A time series is a set of data that are successively collected at regular intervals of time. The regular
interval of time can be daily, weekly, monthly, quarterly or yearly. Examples of some time series data
include:
a) Monthly production of a company;
b) Daily sales at a medical store;
c) Amount of annual rainfall over a period of time; and d) Money deposited in a bank on various
working days.
Furthermore, it is essential to know that when a time series is analysed, it has the following benefits:
i. Understanding the past behaviour of a variable and be able to:
• determine direction of periodic fluctuations; and
• predict future tendencies of the variable.
ii. Determining the impact of the various forces influencing different variables which then facilitate
their comparison, such as:
• the differences that may have to do with price of commodities;
• the physical quantity of goods produced, marketed or consumed in order to make a comparison
between periods of time, schools, places and etc.
iii. Knowing the behaviour of the variables in order to iron out intra-year variations as control events.

1
BASIC COMPONENTS OF A TIME SERIES
In time series, the existence of fluctuations from time to time is caused by composite forces which are
at work constantly.
These factors have four components viz:
a) Secular Trend or Secular Variation (T);
b) Seasonal Fluctuations or Variation (S);
c) Cyclical Variation (C); and
d) Irregular or Random Variations (I).

Secular Trend or Secular Variation (T)

The pattern of time series trend may be linear or non-linear. It is linear when the series values are
concentrated along a straight line on the time –plot. Time – plot is known as the graph of time series
values against different time points. The sketches of time plot showing typical examples of a secular
trend are given below.

2
Cyclical Variation (C)
This is a long-term oscillation or wavelike fluctuation about the trend line of a time series. It is similar
to seasonal variation with a difference of reoccurring in more than one year period. Cyclical variations
are called business cycles in the sense that periods of prosperity followed by recession or depression
and then recovery are caused by aggregate economic conditions rather than seasonal effects. The
length of the cycle varies between four and seven years.
The cyclical variation is less predictable. The figure depicted below shows the
pattern of cyclical variation.

Irregular or Random Variation (I)

This is a variation caused by sporadic events (unpredictable events) such as floods, strikes, disasters,
wars.
Random variation represents the residual variation in time series which cannot be accounted for by the
three other components (i.e Trend, Seasonal, Cyclical variations).
See the figure below for a typical random variation.

3
TIME SERIES ANALYSIS
This is an act of analysing and interpreting time series data. It can be said to be an investigative
method into the time series components; which a times is also referred to as decomposition of a time
series.
The first step in the time series analysis is the drawing of the “time-plot”. This will
clearly show the pattern of the series movement.
Let Y represent the series by convention and T, S, C, I as components earlier indicated in paragraph
5.2. The two models of time series are additive and multiplicative models. For the additive, it is
written as
Y = T + S + C + I ……5.1
while in multiplicative model, it is written as;
Y = TSCI ……5.2
Most of the time the data available can be used to estimate only the trend and seasonal variation.

ESTIMATION OF TREND
There are various methods of estimating trend. However, this study text will consider the following
methods:
a) Moving Average method,
b) Least Squares method; and
c) Experimental Smoothening method
Moving Average Method
This is a popular method of trend estimation. The trend is obtained in this method by smoothing out
the fluctuation. The procedure of computing the trend by the Moving Average (M.A) method depends
on whether the period of moving average desired, atimes called the order, is even or odd:

where the moving averages are expected to be written against the middle items considered for each
average and the numerators of equation 5.4.1.1 are called the moving totals.
The following are procedural steps of computing trend by moving average (M.A):
a. Determine the order to be used whether odd or even.
b. When it is odd, directly apply equation 5.4.1.1 to get the desired moving average (M.A); and
c. If it is even, first obtain the moving totals of order n from the given series and then obtain a 2
combined n moving totals of earlier obtained moving totals. The moving totals lastly obtained will be
divided by 2n to give the desired M.A. The purpose of using two moving totals is to overcome the

4
problem of placing the M. A against middle items considered. By convention, M. A. must be written
against middle item of the items considered. Hence, moving totals make this possible.
The following examples illustrate the trend estimation byMoving Average (M.A) method:

5
Merits and Demerits of Moving Average (M.A) Method
Merits:
(i) The method is simple if compared with least squares method;
(ii) The effect of cyclical fluctuations is completely removed if the period of Moving Average (M.A)
is equal to the average period of cycles; and
(iii) It is good for a time series that reveals linear trends.

Demerits
(i) The extreme values are always lost by Moving Average (M.A) method;
(ii) The method is not suitable for forecasting; and
(iii) It is not good for non-linear trend.

6
Least Squares Method (L.S.M)
Recall that by least squares method, fitting a linear regression of variable (y) on variable (x) gives y =
a + bx, where a and b are constants and are the intercept and regression coefficient respectively. This
method can equally be extended to a time series data by taking the time period t as variable x and the
time series value as variable y. Hence, the trend line by L.S.M is given as :

The following examples give illustrations of Trend estimation by L.S.M.

7
a) Merits of LSM

8
i. No extreme values are lost in this method as in the case of the M. A.
ii. The method is free from subjective error.
iii. The method can be used for forecasting.

b) Demerits
i. It requires more time for computation.