BAMS2923 STATISTICAL ANALYSIS FOR BUSINESS

CHAPTER 3: TIME SERIES ANALYSIS

Time series is a set of observations of a variable taken at specific

times, usually at equal time intervals.

Examples of time series data

1. Daily output of a factory for the past three weeks
2. Monthly sales of a company over the last two years
3. Annual total operating costs of a warehouse for the last 10
years

Time series analysis is concerned with the movement of the

variable’s values from one time period to the next, it enables us
to understand the past and use the past information to forecast
what will happen in the future and thus help with production,
planning and other future decision making.

Historigram
 A graph of a time series
 X axis represents time e.g. day, week, month, quarter, year
 Y axis represents the values of the variable

Historigram: Quarterly fuel consumption (in liter)

545
540
Fuel consumption (liter)

535
530
525
520
515
510
505
500
495
1 2 3 4 1 2 3 4 1 2 3 4
Quarter

Chapter 7 – Page 1
 Historigram: Monthly sales of CD (in $'000)

6
5
Sales ($'000)

4
3
2
1
0
Feb

June

Oct
July

Sept

Nov

Dec
Mar

Apr

May

Aug
Jan

Month

Components / movements/ variations of a time series

A time series can be decomposed into 4 components
1. Trend (T)
2. Seasonal variation (S)
3. Cyclical variation (C)
4. Random variation (R)

Trend
 is the underlying long-term movement of the variable over
time
 Three types of trend:
1. Downward trend
2. Upward trend
3. Constant trend/ No clear movement

Example 1
Historigram: Quarterly sales Air cond. (000s)

6
5.5
5
Sales (000s)

4.5
4
3.5
3
2.5
2
1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
Quarter

Chapter 7 – Page 2
 Overall there appears to be an increase in the number of
sales although the sales figures are fluctuating quarter by
quarter. The general /overall movement indicates an upward
trend

Seasonal variations
 are short-term regular periodic movement in the value of the
variable
 are due to different circumstances which prevail and affect
results at different times of the year, days of the week, times
of the day etc

For example:
1. Sales of ice cream are higher in summer than in winter, this
recurring pattern repeats every year
2. Cash sales in departmental store are higher on weekend than
on weekday each week

Cyclical variations
 are long-term oscillations or swings about the trend
 may or may not be periodic
 are commonly associated with economic / trade cycle,
representing intervals of prosperity, recession, depression
and recovery
Trade cycle
Y prosperity

recovery
recession

depression
X

Chapter 7 – Page 3
Random variations
 Purely random and irregular once-and-for all events which are
completely unpredictable
 Accidental fluctuations for e.g. the weather may be
particularly hot resulting in above normal ice cream sales for
1 year. Catastrophic fluctuations for e.g. fire, earthquake,
flood etc
 Since they are unpredictable, we assume that in the long run
they will tend to cancel each other out, and that in our
analysis, we may initially ignore their impact

Example 2
A company is investigating the number of order (thousands) of
three of its products, A, B and C, and has quarterly data available
for the past three years. Draw a historigram of the data for each
of the three products and comment on the trend and seasonal
components.

Product
Year Quarter A B C
2020 1 84 5 2.0
2 90 8 2.4
3 112 9 1.8
4 107 10 1.6
2021 1 88 14 2.1
2 95 17 2.3
3 118 18 1.7
4 116 22 1.6
2022 1 98 24 2.0
2 106 24 2.3
3 128 27 1.8
4 124 28 1.7

Chapter 7 – Page 4
Solution:

Historigram: Number of order for product A

140
Number of order ('000s)

130

120

110

100

90

80
1 2 3 4 1 2 3 4 1 2 3 4
1990 1991 1992 Year
2020 2021 2022
2020

Product A shows upward trend and seasonal component.

Historigram: Number of order for product B

30
Number of order ('000s)

25

20

15

10

0
1 2 3 4 1 2 3 4 1 2 3 4
1990 1991 1992
Year
2020 2021 2022

Product B shows an upward trend but no seasonal component.

Historigram: Number of order for product C

2.5
2.4
Number of order ('000s)

2.3
2.2
2.1
2
1.9
1.8
1.7
1.6
1.5
1 2 3 4 1 2 3 4 1 2 3 4
1990 1991 1992
Year
2020 2021 2022

Product C shows constant trend (approximately) and seasonal

component.

Chapter 7 – Page 5
Combining The Components
Given the four components (T, S, C, R) of any time series, they
are usually combined in one of the two alternative models.
1. The additive model
2. The multiplicative model

The additive model

 Assume that the observed value Y is the sum of the 4
components: 𝑌 =𝑇+𝑆+𝐶+𝑅
 Units of measurement of T, S, C and R are the same as those
of the observed value of Y
 Since there will not be sufficient data to identify the cyclical
component and the random variation is unpredictable, we
assume that C and R are 0. The model is reduced to:
𝑌 =𝑇+𝑆
 The additive model will be most appropriate where the
variations about the trend are of similar magnitude in the
same period of each section.
150
140
130
120
110
100
Evening

Evening

Evening

Evening

Evening
Afternoon
Morning
Afternoon

Morning
Afternoon

Morning

Morning
Afternoon

Morning
Afternoon

Monday Tuesday Wednesday Thursday Friday

The multiplicative model

 Assume that the observed value Y is the product of the 4
components: 𝑌 =𝑇×𝑆×𝐶×𝑅
 T has the same unit of measurement as the observed values
of Y, whereas the values of S, C and R are pure ratios or
percentages
 Again the lack of data will often mean that the cyclical element
cannot be identified and the random variation is
unpredictable, we assume that C and R are 1 (or 100%). The
model is reduced to:
Chapter 7 – Page 6
 𝑌 = 𝑇×𝑆
 The multiplicative model will be most appropriate for
situations where the variations about the trend are of the
same proportionate size (or percentage) of the trend in the
same period of each section.
300

250

200

150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Measuring The Components

The analysis of a set of time series data consists of breaking
down the data into the components (decomposition) and
trying to find an estimate for each of them.

(a) Identifying the trend

 There are several methods of identifying a trend
 If the broad underlying movement appears to be linear, then
a regression trend (least squares method) would be
appropriate
 If there appears to be a curvilinear trend, then the moving
average method might provide better answers

The Least squares method

 A statistical technique used to calculate the ‘line of best fit’
 Assumption: the trend line, whether up or down, is a straight
line
 The variable concerned as Y and the time variable as X
 The X values will be the rankings of time periods
 The trend is given by Yˆ = a + bX
nXY − (X )(Y ) Y X
where b = nX 2 − (X ) 2 and a = −b
n n

Chapter 7 – Page 7
Example 3
Find the least squares trend line for the time series. Hence
forecast the sales for the year 2023.

Year 2018 2019 2020 2021 2022

Sales (thousand units) 27 30 33 37 41

Solution:
Let Y = sales (thousand units)
X = ranking by year

Year X Y X2 XY
2018 1 27 1 27
2019 2 30 4 60
2020 3 33 9 99
2021 4 37 16 148
2022 5 41 25 205
Total 15 168 55 539

n = number of time points = 5

Chapter 7 – Page 8
Moving averages method
 Most commonly used method to identify the trend
 Involves the calculation of a set of averages
 A moving average is obtained by replacing each value in a
series by the mean of itself, some values directly preceding
and directly following it
 By using moving averages of appropriate orders, we can
eliminate seasonal, cyclic, and random variations. Thus,
leaving only the trend movement

Note:
 The period of the moving average must coincide with the
length of the natural cycle of the series. Examples:
(i) Moving average trend of numbers unemployed
for the quarters for several years must have a
period of 4.
(ii) Moving average trend of total monthly sales of
a business for a number of years must have a
period of 12.
(iii) Moving average trend for the daily takings of a
supermarket (open six days per week) over a
number of weeks must have a period of 6.

 Every moving average trend value calculated must

correspond to the median of the time points for the values
being averaged
(i) For odd number of periods such as 3, 5, 7; the
moving averages will correspond to one
particular time point. Hence the moving
averages give the trend. (Refer to Example 3)
(ii) For even number of periods, the moving
averages are lying between two time points; thus
the “centred” technique is used. The centered
moving averages give the trend. (Refer to
Example 4)

Chapter 7 – Page 9
Example 4
Absentees due to sickness
Week Monday Tuesday Wednesday Thursday Friday
1 4 7 8 11 18
2 3 8 10 13 21
3 6 9 13 17 28

Find the trend for the time series using moving averages method.

Solution:
Let Y = absentees due to sickness

5-day 5-day
Week Day Y moving moving
total average
(Trend)
1 Monday 4
Tuesday 7
Wednesday 8 48 9.6
Thursday 11
Friday 18
2 Monday 3
Tuesday 8
Wednesday 10
Thursday 13 58 11.6
Friday 21 59 11.8
3 Monday 6 62 12.4
Tuesday 9 66 13.2
Wednesday 13 73 14.6
Thursday 17
Friday 28

Chapter 7 – Page 10
Example 5
Sales (000’s)
Year Quarter 1 Quarter 2 Quarter 3 Quarter 4
1 - 42 55 36
2 16 50 53 40
3 28 58 69 59
Find the trend using the moving averages method.

Solution: Let Y = sales (000’s)

Year Q Y 4-quarter 4-quarter 4-quarter
moving moving Centred
total average moving
average
(Trend)
1 1 -

2 42

3 55
149 37.2500
4 36

2 1 16

2 50

3 53
171 42.7500
4 40 43.7500
179 44.7500
3 1 28 46.7500
195 48.7500
2 58 51.1250
214 53.5000
3 69

4 59
Chapter 7 – Page 11
Measuring the Components
(b) Identifying the seasonal component
Seasonal variation has a distinctive characteristic of being
periodic with a recurring pattern from year to year, month to
month, week to week or day to day.

Using the additive model ( Y = T + S )

Given the actual data (Y) and the calculated trend values (T):
S=Y–T
The procedure for calculating the seasonal variation is given as
follows:
STEP 1 Calculate, for each time point, the value of Y – T.
STEP 2 For each season in turn, find the average of Y–T
values.
STEP 3 If the total of the averages differs from zero, adjust
the values so that their total is zero.

NOTE:
Using additive model to obtain the average seasonal variation,
the total average seasonal variations should be zero. If it is not,
the following adjustment is necessary:

0 − actual average total

𝐀𝐝𝐣𝐮𝐬𝐭𝐦𝐞𝐧𝐭 =
No. of seasons

Adjusted average seasonal variation

= Average seasonal variation + Adjustment

Chapter 7 – Page 12
Example 6
Using the time series and moving averages trend obtained in
Example 5, calculate the average seasonal variations by using
the additive model and interpret your results.

Solution:
4-quarter
4-quarter 4-quarter Centred
Year Quarter Y moving moving moving (Y–T)
total average average
(Trend)
1 1 -

2 42

3 55
149 37.2500
4 36 38.2500
157 39.2500
2 1 16 39
155 38.7500
2 50 39.2500
159 39.7500
3 53 41.2500 11.7500
171 42.7500
4 40 43.7500 -3.7500
179 44.7500
3 1 28 46.7500 -18.7500
195 48.7500
2 58 51.1250 6.8750
214 53.5000
3 69

4 59

Chapter 7 – Page 13
Average seasonal variations(000’s):
Quarter
Year 1 2 3 4 Total
1
2
3
Total -41.7500

Average -20.8750 -3.3125

*Adjustment 0.8281
Adjusted
average -20.0469 0

0 − (−3.3125)
* Adjustment = = 0.8281
4

Interpretation of the average seasonal variations:

Due to seasonal fluctuation, sales in quarter 1 was
20.0469(‘000’s) units below average sales. Sales in quarter 2
was 9.6406(‘000’s) units above average sales. Sales in quarter
3 was 12.5781(‘000’s) units above average sales. Sales in
quarter 4 was 2.1719(‘000’s) units below average sales.

Chapter 7 – Page 14
Using the multiplicative model ( Y = T  S )
Given the actual data (Y) and the calculated trend values (T):
S=Y÷T (in ratio form)
or S = (Y ÷ T)  100 (in percentage form)

The procedure for calculating the seasonal variations is given

below:
STEP 1 Calculate, for each time point, the value of Y ÷ T
STEP 2 For each season in turn, find the average of Y ÷ T
values.
STEP 3 If the total of the averages differs from the number
of time points or moving average, adjust the values
so that their total is the desired number.

NOTE:
Using the multiplicative model to obtain the seasonal
variations, the total of the average seasonal variations would be
predicted to be n where n is the number of seasons/time periods
in a section. If it is not, an adjustment is necessary as below:

No. of seasons
𝐀𝐝𝐣𝐮𝐬𝐭𝐦𝐞𝐧𝐭 =
Actual average total

Adjusted average seasonal variation

= Average seasonal variation  Adjustment

Chapter 7 – Page 15
Example 7
Using the time series and moving averages trend obtained in
Example 5, calculate the average seasonal variation using the
multiplicative model and interpret your results.

Solution:
4-quarter
4-quarter 4-quarter Centred
Year Quarter Y moving moving moving (Y÷T)
total average average
(Trend)
1 1 -

2 42

3 55
149 37.2500
4 36 38.2500
157 39.2500
2 1 16 39
155 38.7500
2 50 39.2500
159 39.7500
3 53 41.2500 1.2848
171 42.7500
4 40 43.7500 0.9143
179 44.7500
3 1 28 46.7500 0.5989
195 48.7500
2 58 51.1250 1.1345
214 53.5000
3 69

4 59

Chapter 7 – Page 16
Average seasonal variations:
Quarter
Year 1 2 3 4 Total
1
2 -
3
Total -
Average 3.9214
*Adjustment -
Adjusted
0.5147 1.2283 1.3105 0.9464 4
average
4
*Adjustment = = 1.0200
3.9214

Interpretation of the average seasonal variations:

The seasonal variations obtained in multiplicative model are not
true variations but percentage increases or decreases. Without
seasonal fluctuation, the seasonal variation should be 100%.
Due to seasonal fluctuation, sales in quarter 1 was 48.53% below
average sales. Sales in quarter 2 was 22.83% above average
sales. Sales in quarter 3 was 31.05% above average sales.
Sales in quarter 4 was 5.36% below average sales.

Chapter 7 – Page 17
Forecasting Using Trend and Seasonal Factor
 A particular use of time series analysis is in forecasting
 Assume that patterns of past data will be broadly continued,
at least into the short-term future
 Using the past data to predict the future values

Forecasting a value for a future time point involves the following

steps.
STEP 1 Estimate a trend value for the time point.
STEP 2 Identify the (average) seasonal variation value
appropriate to the time point.
STEP 3 Combining these two values together giving the
required forecast.
Additive Model: 𝑌𝑒𝑠𝑡 = 𝑇𝑒𝑠𝑡 + 𝑆
Multiplicative Model: 𝑌𝑒𝑠𝑡 = 𝑇𝑒𝑠𝑡 × 𝑆

where 𝑌𝑒𝑠𝑡 = Estimated data value

𝑇𝑒𝑠𝑡 = Estimated trend value
𝑆 = appropriate seasonal variations

𝑇𝑒𝑠𝑡 = Last trend value + Average rate of change in trend × 𝑑

Last trend − first trend

Average rate of change in trend =
n−1

𝑑 = No. of period deviated from last trend

where
𝑛 = No. of trend calculated

Alternatively; the trend can be obtained graphically by joining

the first and the last of the calculated trend values to obtain a
linear projection of the trend. This method can be employed
with fairly ‘steady linear’ trend.

Chapter 7 – Page 18
Example 8
Using seasonal variations (additive model) obtained in Example
6, forecast the sales for the first two quarters of year 4.

Solution:

Average rate of change in trend =

Seasonal
Estimated Trend Forecast sales
variation
Year Quarter d (000’s), 𝑇𝑒𝑠𝑡 (000’s), 𝑌𝑒𝑠𝑡
(000’s)
4 1

Example 9
Using the seasonal variations (multiplicative model) obtained in
Example 7, forecast the sales for the first two quarters of year 4.

Solution:

Average rate of change in trend =

Seasonal
Estimated Trend Forecast sales
variation
Year Quarter d (000’s), 𝑇𝑒𝑠𝑡 (000’s), 𝑌𝑒𝑠𝑡
(000’s)
4 1

Chapter 7 – Page 19
Deseasonalising a time series
 The effect is to smooth away seasonal fluctuations, leaving
a clear view of what might be expected ‘had seasons not
existed’
 Also known as seasonally adjusted data
 Seasonally adjusted data for additive model: Y − S
 Seasonally adjusted data for Multiplicative model: Y  S

Example 10
Deseasonalise the time series data in Example 6 (Additive
model).

Solution:
Sales Seasonal Deseasonalised
(000’s) variation (000’s) sales (000’s)
Year Q Y S Y-S
1 1 - - 20.0469 -
2 42 9.6406 32.3594
3 55 12.5781 42.4219
4 36 - 2.1719 38.1719
2 1 16 - 20.0469 36.0469
2 50 9.6406
3 53 12.5781
4 40 - 2.1719
3 1 28 - 20.0469
2 58 9.6406
3 69 12.5781
4 59 - 2.1719

Chapter 7 – Page 20
Example 11
Seasonally adjust the following time series data in Example 7
using the multiplicative model.

Solution:
Seasonally
Sales Seasonal adjusted sales
(000’s) variation (000’s)
Year Q Y S Y÷S
1 1 - 0.5147 -
2 42 1.2283 34.1936
3 55 1.3105 41.9687
4 36 0.9464 38.0389
2 1 16 0.5147 31.0861
2 50 1.2283
3 53 1.3105
4 40 0.9464
3 1 28 0.5147
2 58 1.2283
3 69 1.3105
4 59 0.9464

Chapter 7 – Page 21
Example 12
The following table shows the daily income (in RM) of a
stallholder who sells shoes from Wednesday to Sunday over a
3-week period:

Week Wednesday Thursday Friday Saturday Sunday

1 180 250 500 600 640
2 150 280 450 550 650
3 210 230 390 520 580

(a) Obtain a suitable moving average trend for the data.

(b) Using the additive model, compute the average seasonal
variations.
(c) Forecast the daily income for Thursday of week 4.
(d) Find the de-seasonalise data for Friday of week 3.
(Answer: (c) 212.0667; (d) 358.0667)

Solution:
a) b)
Week Day Y 5-Day MT 5-Day MA (T) Y - T
1 Wed 180
Thu 250
Fri 500
Sat 600
Sun 640 2170 434 206
2 Wed 150 2120 424 -274
Thu 280 2070 414 -134
Fri 450 2080 416 34
Sat 550 2140 428 122
Sun 650 2090 418 232
3 Wed 210 2030 406 -196
Thu 230 2000 400 -170
Fri 390 1930 386 4
Sat 520
Sun 580

Chapter 7 – Page 22
b)
Week Wed Thu Fri Sat Sun Total
1 66 172 206
2 -274 -134 34 122 232
3 -196 -170 4
Total 104 294 438
Average 34.6667 147 219 13.6667
Adjustment -2.7333 -2.7333 -2.7333
Adjusted
31.9333 144.2667 216.2667 0
average

Adjustment =

c) Average rate of change in trend =

Week Day d 𝑇𝑒𝑠𝑡 S 𝑌𝑒𝑠𝑡

4 Thu

d)
Week Day Y S Y-S
3 Fri

Chapter 7 – Page 23
BAMS2923 STATISTICAL ANALYSIS FOR BUSINESS
TUTORIAL 3 (TIME SERIES ANALYSIS)

1. The table below shows the number of components produced

by a factory during the morning, afternoon and evening shifts
last week.

No. of components produced

Day
Morning Afternoon Evening
Monday 127 114 134
Tuesday 130 115 138
Wednesday 128 117 142
Thursday 131 116 141
Friday 132 120 144

(a) Plot the time series data.

(b) Establish the trend using a 3-point moving averages.
(c) Estimate the average shift variations using additive
model.
(d) Forecast the production for the three shifts on Monday
of this week.

2. The following table shows the quarterly sales (in RM million)

of a company over a period of 3 years:

Quarter
Year
1 2 3 4
1 32 48 40 16
2 24 56 40 8
3 24 48 32 16

(a) Calculate the four-quarter centred moving averages.

(b) Assume an additive model, calculate the average
seasonal variations.
(c) Estimate the sales for the first two quarters of year 4.
(d) De-seasonalise the series.

Chapter 7 – Page 24
3. The table below shows the number of tourists who visit a town
in October 2022.

Number of tourists
Week
Mon Tue Wed Thu Fri Sat Sun
1 172 392 432 427 405 655 689
2 192 402 442 438 416 665 701
3 202 414 454 450 428 677 718
4 212 424 467 460 440 686 725

(a) Obtain a suitable moving average trend for the data.

(b) Using the additive model, compute the daily seasonal
variations.
(c) Forecast the number of tourists for Tuesday for the first
week of November 2022.
(d) Find the de-seasonalise data for Sunday of week 4.

4. The table below shows the number of students studying

Statistics in university from year 2018 to year 2022.

Number of students
Year Semester 1 Semester 2 Semester 3
2018 359 262 135
2019 330 271 151
2020 299 279 173
2021 280 250 190
2022 272 221 252

(a) Obtain a suitable moving average trend for the data.

(b) Using the multiplicative model, compute average
seasonal variations.
(c) Forecast the number of students for the second
semester of year 2023.
(d) Find the de-seasonalise data for third semester of year
2022.

Chapter 7 – Page 25
5. The daily output (in units) from a factory is as shown in the
following table for a four-week period.

Daily output (units)

Week Monday Tuesday Wednesday Thursday Friday
1 360 400 480 600 660
2 350 430 490 580 680
3 380 440 490 590 690
4 390 450 400 600 690

(a) Use the method of moving averages to find a set of trend

values.
(b) Use the multiplicative model, find the daily variations
pertaining to each day of the week.
(c) Forecast the output for Monday and Tuesday of week 5.

6. The sales of a product (RM’000) for May year 1 to April year

4 are given in the following table:

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

1 - - 103 128 203 324
2 68 90 109 144 231 354
3 70 95 116 153 245 374
4 73 99

(a) Obtain a suitable moving average trend for the sales.

(b) Using the multiplicative model, compute the average
seasonal variations.
(c) Estimate the expected sales for Jul/Aug of year 4.
(d) Find the seasonally adjusted sales for Jan/Feb of year 4.

Answers

1. (b) 125, 126, 126.3333, 127.6667, 127, 127.6667, 129,

130, 129.6667, 129.3333, 129.6667, 131, 132
(c) 1.5944, -12.1889, 10.5944

Chapter 7 – Page 26
 (d) 0.5833; 134.7610, 121.5610, 144.9276

2. (a) 33, 33, 34, 33, 32, 31, 29, 29

(b) –7.75, 20.75, 7.25, -20.25
(c) 19.5358, 47.4644
(d) 39.75, 27.25, 32.75, 36.25, 31.75, 35.25, 32.75,
28.25, 31.75, 27.25, 24.75, 36.25

3. (a) 453.1429, 456, 457.4286, 458.8571, 460.4286, 462,

463.4286, 465.1429, 466.5714, 468.2857, 470,
471.7143, 473.4286, 475.1429, 477.5714, 479,
480.4286, 482.2857, 483.7143, 485.4286,
486.7143, 487.7143
(b) -269.9864, -60.3197, -20.7959, -27.1769, -50.8911,
196.9184, 232.2517
(c) 435.6261
(d) 492.7483

4. (a) 252, 242.3333, 245.3333, 250.6667, 240.3333, 243,

250.3333, 244, 234.3333, 240, 237.3333, 227.6667,
248.3333
(b) 1.2630, 1.0515, 0.6856
(c) 260.1584
(d) 367.5613

5. (a) 500, 498, 504, 506, 502, 506, 512, 514, 514, 516,
518, 520, 522, 504, 506, 506
(b) 0.7359, 0.8674, 0.9175, 1.1590, 1.3199
(c) 373.2485, 440.2922

6. (a) 153.1667, 155, 158.6667, 163.5, 166.1667, 166.75,

167.75, 169.0833, 171, 173.8334, 175.75, 176.3334
(b) 0.4247, 0.5592, 0.6644,0.8652, 1.3820, 2.1046
(c) 161.6770
(d) 171.8860

Chapter 7 – Page 27
Past Year Questions

Chapter 7 – Page 28