Time series analysis and

Forecasting
[email protected]
Outline

The components of a time series

Time series components
Time series models

Finding the trend

Moving averages
Finding the seasonal variations
Forecasting
Forecasting by extrapolation
Deseasonalization
Introduction to time series
A time series is a sequence of observations Example
on a variable measured at successive points in
time or over successive periods of time

A time series depicts the relationship between

two variables. Time is one of those variables
and the second is any quantitative variable.

The components of a time series

▪ Trend
▪ Season variation
▪ Cyclical variation
▪ Random variation or Irregular variation
The components of a time series
Trend Output/
labor hour
35
▪ A trend is the continuous long-term movement
over time in the values of the data recorded. 30

25

Example
20

There are three types of trend in the following 15

example.
Output/ Cost/ Number of 10
Year Downward Trend
labor hour unit ($) employees 5
2004 30 1 100
2005 24 1.08 103
0
2003 2004 2005 2006 2007 2008 2009 2010
2006 26 1.2 96
2007 22 1.15 102 The time series rises or falls continuously. The
2008 21 1.18 103
trend, represented by the red line
2009 17 1.25 98
(A) (B) (C)
The components of a time series
Cost/
Example unit ($)
There are three types of trend in the following 1.30

example. 1.25

Output/ Cost/ Number of 1.20

Year
labor hour unit ($) employees 1.15

2004 30 1 100 1.10

2005 24 1.08 103 1.05

2006 26 1.2 96 1.00

2007 22 1.15 102 0.95

2008 21 1.18 103 0.90

2009 17 1.25 98 0.85

(A) (B) (C) 0.80

2003 2004 2005 2006 2007 2008 2009 2010

Upward trend
The components of a time series
Seasonal variations are movements in the time
series that reoccur each year about the same
time due to the change in the seasons.
These are the rhythmic forces which operate in
a regular and periodic manner over a span of
less than a year. They have the same or almost
the same pattern during a period of 12 months.
This variation will be present in a time series if
the data are recorded hourly, daily, weekly,
quarterly, or monthly.

Example:
• Each year the unemployment rate tends to
goes up in May when students enter the An annual repeating seasonal variation, with low
timber production in January, increasing to a high
summer job market and goes down in
production in April. There is then a slight decrease in
November when retail stores hire temporary July followed by a slight increase in October. There is
help to handle the Christmas rush. also a basic upward trend.

• Sales of ice cream will be higher in summer

than in winter, and sales of overcoats will be
higher in fall than in spring.
The components of a time series
Cyclical variations
Many variables often exhibit a tendency to
fluctuate above and below the long-term trend
over a long period of time. These fluctuations are
called cyclical variations. They cover much longer
time periods than do seasonal variations.

The variations in a time series which operate

themselves over a span of more than one year are
the cyclic variations. This oscillatory movement
has a period of oscillation of more than a year
The components of a time series
There is another factor which causes the variation in the variable under study. They are not regular
variations and are purely random or irregular. These fluctuations are unforeseen, uncontrollable,
unpredictable, and are erratic. These forces are earthquakes, wars, flood, famines, and any other
disasters.
The components of a time series
Mathematical Model for Time Series Analysis

Mathematically, a time series is given as Additive Model for Time Series Analysis
If 𝑦𝑡 is the time series value at time 𝑡. 𝑇𝑡 , 𝑆𝑡 , 𝐶𝑡 , and
𝑌𝑡 = 𝑓(𝑡)
𝑅𝑡 are the trend value, seasonal, cyclic and random
Here, 𝑦𝑡 is the value of the variable under fluctuations at time t respectively. According to the
study at time 𝑡. If the population is the Additive Model, a time series can be expressed as
variable under study at the various time
𝑦𝑡 = 𝑇𝑡 + 𝑆𝑡 + 𝐶𝑡 + 𝑅𝑡 .
period 𝑡1 , 𝑡2 , 𝑡3 , … , 𝑡𝑛 . Then the time series is
t: t1 , t 2 , t 3 , … , t n This model assumes that all four components of
Yt : Yt1 , Yt2 , Yt3 , … , Ytn the time series act independently of each other.

A time series model can be expressed as Multiplicative Model for Time Series Analysis
some combination of these four The multiplicative model assumes that the various
components. Two types of models are components in a time series operate
commonly associated with time series: proportionately to each other. According to this
additive model and multiplicative model. model
The additve model is expressed as 𝑦𝑡 = 𝑇𝑡 × 𝑆𝑡 × 𝐶𝑡 × 𝑅𝑡
Measuring Forecast accuracy

Percentage error (PE):

PE𝑡
= 100 ∗ 𝑌𝑡 − 𝐹𝑡 /𝑌𝑡

Mean absolute
Mean error : Mean percentage error percentage error :
Mean squared error :
1 n 1 𝑛 1 n 1 n
𝑀𝐸 = ∑t=1 et 𝑀𝑆𝐸 = ∑ 𝑒𝑡2 MPE = ∑t=1 PEt 𝑀𝐴𝑃𝐸 = ∑t=1 PEt
𝑛 𝑛 𝑡=1 n 𝑛
Finding the trend

Three principal methods to find a trend. Moving averages (MA) is a series of

arithmetic averages over a given number of time
Inspection: The trend line can be drawn by periods. It is the estimate of the long run average
eye on a graph in such a way that it appears to of the variable.
lie evenly between the recorded points, that is,
a line of best fit drawn by eye.
MOVING AVERAGE OF ORDER 𝑘
Regression analysis: This method makes
the assumption that the trend line is a straight
line. Periods of time are numbered, and the ∑ ( most recent k data values )
𝐹𝑡+1 =
regression line of the data on those periods is 𝑘
found. That line is taken to be the trend.
𝑌𝑡 + 𝑌𝑡−1 + ⋯ + 𝑌𝑡−𝑘+1
Moving average: This method attempts to =
remove seasonal variations by a process of 𝑘
averaging.
where
Ft+1 = forecast of the times series for period t + 1
Yt = actual value of the time series in period t
Calculating trend by MA, Seasonal variation by additive model

Finding the seasonal variations Three-

Sales month Seasonal variation
Month
The additive model of time series analysis is ($) moving ($)
average ($)
Y = T + S + R. January 125
February 145 152 (145-152) =-7
If we assume that the random component R is
negligible, the seasonal component can be found March 186 154 (186-154) = 32
as April 131 156 (131 -156) = -25
S=Y—T May 151 158 (151 -158) =-7
June 192 160 (192-160) = 32
called the de-trended series. July 137 162 (137-162) = -25
August 157 164 (157-164) =-7
A negative variation means that the actual figure September 198 166 (198-166) = 32
in that period is less than the trend and a October 143 168 (143-168) = -25
positive figure means that the actual is more November 163 170 (163-170) =-7
than the trend. December 204
Seasonal variation by multiplicative model

The multiplicative model of time series

analysis can be expressed as Three-month
Seasonal
Month Sales ($) moving
Y = T × S × R. variation ($)
average ($)

In this model, only T is expressed in the original February 145 152 (145/152) = 0.95
units, S and R are stated in terms of percentages.
March 186 154 (186/154)= 1.21
If the random component R is negligible, which April 131 156 (131/156) = 0.84
means R ≈ 1 in this case, the seasonal
percentage is calculated by
This suggests that month 1 is usually 95% of the
S = Y/T trend, month 2 is 121% and month 3 is 84%. The
multiplicative model is a better method to use when
the trend is increasing or decreasing over time, as
the seasonal variation is also likely to be increasing
or decreasing.
Example seasonal variation
Example: Output at a factory appears to vary
with the day of the week. Output (in units) over
the last three weeks has been as follows

Week 1 Week 2 Week 3

Mon 92 94 96
Tue 115 121 127
Wed 104 107 110
Thu 131 135 140
Fri 74 76 78

Find the seasonal variation for each of the fifteen

days, and the average seasonal variation for each
day of the week using the moving average method
by using

a. additive model.

b. multiplicative model.
Forecasting
Forecasting by extrapolation Forecasts of future values should be made as
follows.
Extrapolation is the process of estimating,
beyond the original observation range, the • Calculate a trend line using moving
value of a variable on the basis of its averages or regression.
relationship with another variable.
• Use the trend line to forecast future trend
line values.

• Adjust these values by the average seasonal

variation. With the additive model, add the
variation. With the multiplicative model,
multiply the trend value by the variation
proportion.
Multiplicative model vs additive model in forecast

The multiplicative model is better than the additive model for forecasting when the trend is
increasing or decreasing over time. The additive model suffers from the somewhat unrealistic
assumption that the components are independent of each other. In most instances, movements
in one component will have an impact on other components.

The multiplicative model is often preferred. It assumes that the components interact with each
other and do not move independently.

Since extrapolation is based on historical data alone, and does not include effects of
developments, it can be used for short-term forecasts only for specific areas, where no
untypical developments are expected.