POLITEKNIK STATISTIKA STIS

For Better Official Statistics

TIME SERIES DECOMPOSITION

Time Series Analysis – Pertemuan 3

 KONSEP
DEKOMPOSISI
TIME SERIES
 For Better Official Statistics

✓Analisis data time series memerlukan identifikasi komponen yang

memengaruhi setiap nilai dalam series
✓Prosedur untuk mengidentifikasi komponen-komponen dalam data
time series → dekomposisi
✓Komponen data time series
▪ Komponen jangka panjang: trend dan siklus (cyclical)
▪ Komponen jangka pendek: musiman (seasonal)
▪ Komponen random: irregular, tidak tertangkap oleh pola
✓Prinsip dasar dekomposisi time series adalah menguraikan data time
series ke dalam komponen-komponen pembentuknya, agar dapat
dipahami lebih baik serta dapat digunakan untuk peramalan

3
 For Better Official Statistics

Time Series

Trend Cyclical Seasonal Calendar Irregular

Component Component Component Component Component

✓Trend → komponen yang mewakili pertumbuhan (atau penurunan) dalam deret

waktu
✓Cyclical → serangkaian fluktuasi atau siklus seperti gelombang dengan durasi
lebih dari satu tahun
✓Seasonal → pola perubahan yang relatif stabil yang muncul setiap periode dan
berulang dari periode ke periode
✓Calendar → efek yang terkait dengan kalender yang tidak berulang dalam
periode yang sama setiap tahun
✓Irregular → fluktuasi yang tidak terduga atau acak 4
 For Better Official Statistics

Manfaat Dekomposisi Time Series

✓Mempertimbangkan deret waktu dalam hal komponennya
(mendekomposisi time series) dapat menekankan (highlight) fitur
penting dari data
▪ Membantu dalam peramalan khususnya jangka pendek
▪ Membantu dalam analisis time series antar individu atau wilayah
▪ Memperbolehkan perbandingan data time series dari bulan ke bulan
▪ Membantu dalam pengambilan kebijakan

✓Pendekatan sederhana adalah dengan mengasumsikan sebuah time

series dapat didekomposisi menjadi tiga komponen
▪ Trend (Trend-Cycle), Seasonal, Irregular

5
 For Better Official Statistics

The Data Puzzle

Trend Trading Day Moving Holiday

Original Data
Seasonal Irregular Other Systematics

6
 For Better Official Statistics

Trend Component
Trend
Underlying
behaviour of
the series

Will contain
long term
business cycles

7
 For Better Official Statistics

Trend Component (2)

✓Komponen trend adalah pergerakan yang mendasari dalam data time
series
✓Mencerminkan level umum yang mendasari time series tersebut
✓Trend bisa naik atau turun dan berpola linier atau nonlinier (eksponensial)
✓Trend dapat disebabkan faktor-faktor seperti:
▪ Pertumbuhan penduduk
▪ Inflasi harga
▪ Pembangunan ekonomi secara umum
✓Pada tahap ini, kita berasumsi bahwa trend mengandung siklus bisnis
jangka panjang (cyclical component)

8
 For Better Official Statistics

Trend Component (3)

Australian Private Gross Fixed Capital Expenditure

9
 For Better Official Statistics

Seasonal and Calendar Component

Trading Day Moving Holiday

Calendar Related
Effects
Seasonal Other Systematics

10
 For Better Official Statistics

Seasonal and Calendar Component (2)

✓Komponen musiman adalah setiap efek yang cukup stabil sehubungan dengan
▪ Waktu tahunan (diobservasi dalam 1 tahun)
▪ Arah yang konsisten (berulang pada periode yang sama)
▪ Besaran yang dapat diprediksi
✓Biasanya bulanan, kuartalan, atau bisa juga mingguan atau harian
✓Komponen musiman dapat disebabkan oleh faktor-faktor seperti:
▪ Waktu libur nasional
▪ Acara kalender
▪ Cuaca
✓Efek yang terkait dengan kalender yang tidak berulang dalam periode yang sama
setiap tahun seperti trading day dan moving holiday termasuk di dalam calendar
component (tidak dicakup di mata kuliah ini). Biasanya ketika efek musiman
dihilangkan dari suatu series, efek calendar juga dihilangkan secara bersamaan
(misalnya, seasonal adjustment oleh Australian Bureau of Statistics)
11
 For Better Official Statistics

Length of Seasonal
✓Data triwulanan, berfluktuasi triwulanan Period of “Season” Number of
dan berulang setiap tahun Pattern Length “Seasons”
→ 1 tahun = 4 triwulan → s = 4 in Pattern
✓Data bulanan, berfluktuasi bulanan dan Week Day 7
berulang setiap tahun
Month Week 4–4½
→1 tahun = 12 bulan → s = 12
Month Day 28 – 31
✓Data mingguan, berfluktuasi mingguan dan
berulang setiap bulan Year Quarter 4
→1 bulan = 4 minggu → s = 4 Year Month 12
✓Data harian, berfluktuasi harian dan Year Week 52
berulang setiap pekan
→1 minggu = 7 hari → s = 7
12
 For Better Official Statistics

Example: Strong Seasonality

Job advertisements (ACT)

13
 For Better Official Statistics

Irregular Component

Short term Not systematic

fluctuations or predictable

Irregular

14
 For Better Official Statistics

Irregular Component (2)

✓Time series memiliki tingkat volatilitas, menyebabkan nilai berosilasi di sekitar
level trend umum
✓Komponen irregular (sering juga disebut komponen remainder) merupakan
fluktuasi jangka pendek yang tidak sistematis (tidak berulang) dan tidak dapat
diprediksi
▪ Termasuk ketidakpastian yang mendasari ("noise") dari time series
✓Perubahan kenaikan atau penurunan tidak beraturan baik dari sisi waktu maupun
durasinya
✓Kadang-kadang tingkat ketidakteraturan sangat besar yang dapat menghasilkan
nilai yang ekstrim
✓Variasi irregular bisa disebabkan oleh:
▪ Faktor alam: bencana alam, wabah
▪ Kejadian tak terduga: perang, krisis 15
 For Better Official Statistics

Irregular Component (3)

Example: Very Irregular Series Value of Building Approvals (ACT)

16
 For Better Official Statistics

Examples of Real-World Events

Trend Seasonal Irregular
Business cycle Seasonal: Strikes/industrial action
Population growth/decline Christmas Natural catastrophes (floods)
Economic growth/decline End of financial year War/assassinations
Inflation School holidays Elections
Globalisation Climatic season/agricultural Major events (e.g. Olympics)
Policy changes (e.g. First home cycle Change of government
owners grant) School leavers Terrorism
Analysis or definitional Public holidays Tax cuts/bonuses
changes – survey design, use of Weekends
admin data, etc. Academic year
Exchange rates Tourism
Interest rates Melbourne cup
Social change Calendar:
Climate change Ramadhan
Labour growth Easter
Ageing population Days per month
Technological change Trading days
Leap years
17
 For Better Official Statistics

Model Dekomposisi
Dengan mengasumsikan time series hanya terdiri dari tiga komponen, 𝒀𝒕 dapat
dituliskan sbb:
𝑌𝑡 = 𝑓 𝑇𝑡 , 𝑆𝑡 , 𝐼𝑡
di mana 𝑌𝑡 = data pada periode 𝑡
𝑇𝑡 = trend-cycle component pada periode 𝑡
𝑆𝑡 = seasonal component pada periode 𝑡
𝐼𝑡 = irregular component pada periode 𝑡

Terdapat dua jenis model dekomposisi, yaitu

1. Model Aditif → 𝑌𝑡 = 𝑇𝑡 + 𝑆𝑡 + 𝐼𝑡
2. Model Multiplikatif → 𝑌𝑡 = 𝑇𝑡 × 𝑆𝑡 × 𝐼𝑡
18
 For Better Official Statistics

Model Dekomposisi (2)

✓Model aditif cocok jika besaran fluktuasi musiman (seasonal) tidak
berubah (konstan) terhadap level data
✓Jika fluktuasi musiman proporsional terhadap level dari series, maka
model multiplikatif lebih tepat
✓Model multiplikatif lebih umum digunakan pada variabel ekonomi
✓Sebagai alternatif, gunakan transformasi Box-Cox untuk mengubah
hubungan multiplikatif menjadi aditif:
𝑌𝑡 = 𝑇𝑡 × 𝑆𝑡 × 𝐼𝑡 → log 𝑌𝑡 = log 𝑇𝑡 + log 𝑆𝑡 + log 𝐼𝑡

19
 For Better Official Statistics

→ Dekomposisi Adititif
(Trend & Seasonal)
→ Constant Variability

→ Dekomposisi Multiplikatif
(Trend & Seasonal)
→ Variability Increasing

20
 For Better Official Statistics

Permintaan listrik
minggguan di England dan
Wales pada Bulan April
2005 dan April 2013

21
 For Better Official Statistics

Seasonal Adjustment
✓Useful by-product of decomposition: an easy way to calculate seasonally
adjusted data.
✓Seasonally adjusted data is useful to highlight variation due to the
underlying state of the economy rather than the seasonal variation
✓Additive decomposition: seasonally adjusted data given by
𝑌𝑡 − 𝑆𝑡 = 𝑇𝑡 + 𝐼𝑡
✓Multiplicative decomposition: seasonally adjusted data given by
𝑌𝑡
= 𝑇𝑡 × 𝐼𝑡
𝑆𝑡
22
 For Better Official Statistics

Seasonal Adjustment (2)

Total employment in US retail

23
 For Better Official Statistics

Seasonal Adjustment (3)

✓We use estimates of 𝑆 based on past values to seasonally adjust a current
value.
✓Seasonally adjusted series reflect remainders as well as trend. Therefore
they are not “smooth” and “downturns” or “upturns” can be misleading.
✓It is better to use the trend-cycle component to look for turning points.
✓Country practices:
▪ Australia: https://www.abs.gov.au/statistics/economy/national-accounts/australian-
national-accounts-national-income-expenditure-and-product/mar-2022

▪ Singapore: https://tablebuilder.singstat.gov.sg/table/TS/M015651
24
TREND-CYCLE
ESTIMATION
 For Better Official Statistics

Pemodelan Komponen Trend

✓Jika komponen trend diestimasi secara terpisah, terdapat 2 metode
sederhana yang dapat dilakukan:
▪ Moving average smoothing (Pertemuan 2)
▪ Linear, quadratic, and exponential trend model

26
 For Better Official Statistics

Trend Linier
✓ The time series is modelled
with a straight line:
𝑇෠𝑡 = 𝑏0 + 𝑏1 𝑡
✓ It is the simplest model and
may suffice for short-run
forecasting or as a baseline
model
✓ Linear trend is fitted by using
ordinary least squares
formulas
✓ Note: instead of using the
actual time values (e.g., years),
use an index 𝑡 = 1, 2, …

27
 For Better Official Statistics

Trend Kuadratik
✓ The time series is modelled
with a curve:
𝑇෠𝑡 = 𝑏0 + 𝑏1 𝑡 + 𝑏2 𝑡 2
✓ Fitting a quadratic model is a
way of checking for
nonlinearity
✓ If 𝑏0 does not differ
significantly from zero, then
the linear trend would suffice

28
 For Better Official Statistics

Trend Eksponensial
✓ The time series is modelled
with an exponential function:
𝑇෠𝑡 = 𝑏0 𝑏1𝑡
✓ There may not be much
difference between a linear
and exponential model when
the growth rate is small and
the data set covers only a few
time periods

29
 For Better Official Statistics

Pros and Cons

30
 CLASSICAL
DECOMPOSITION
 For Better Official Statistics

Classical Decomposition
✓ The classical decomposition method originating in the 1920s is a
relatively simple procedure, and forms the starting point for most
other methods of time series decomposition
✓ There are two forms of classical decomposition: an additive
decomposition and a multiplicative decomposition
✓ In classical decomposition, we assume that the seasonal component
is constant from year to year
✓ For multiplicative seasonality, the 𝑠 values that form the seasonal
component are sometimes called the “seasonal indices”
32
 For Better Official Statistics

Additive Decomposition
Step 1
Compute the trend-cycle component 𝑇෠𝑡 using a (centered) moving
average: 𝑠 is even, use 2×𝑠−𝑀𝐴; 𝑠 is odd use, 𝑠−𝑀𝐴
Step 2
Calculate the detrended series: 𝑌𝑡 − 𝑇෠𝑡
Step 3
To estimate the seasonal component 𝑆መ𝑡 for each season, simply average
the detrended values for that season. These seasonal component values
are then adjusted to ensure that they add to zero
Step 4
The remainder component is calculated by subtracting the estimated
seasonal and trend-cycle components: 𝐼መ𝑡 = 𝑌𝑡 − 𝑇෠𝑡 − 𝑆መ𝑡
33
For Better Official Statistics

34
 For Better Official Statistics

Multiplicative Decomposition
Step 1
Compute the trend-cycle component 𝑇෠𝑡 using a (centered) moving
average: 𝑠 is even, use 2×𝑠−𝑀𝐴; 𝑠 is odd use, 𝑠−𝑀𝐴
Step 2
Calculate the detrended series: 𝑌𝑡 Τ𝑇෠𝑡
Step 3
To estimate the seasonal component 𝑆መ𝑡 for each season, simply average
the detrended values for that season. These seasonal indexes are then
adjusted to ensure that they add to 𝑠
Step 4
The remainder component is calculated by dividing out the estimated
seasonal and trend-cycle components: 𝐼መ𝑡 = 𝑌𝑡 Τ𝑇෠𝑡 Τ𝑆መ𝑡
35
For Better Official Statistics

36
 VARIOUS
DECOMPOSITION
METHODS
 For Better Official Statistics

History of Time Series Decomposition

✓Classical method originated in 1920s.
✓Census II method introduced in 1957. Basis for X-11 method and variants
(including X-12-ARIMA, X-13-ARIMA)
✓STL method introduced in 1983
✓TRAMO/SEATS introduced in 1990s
Methods Used by National Statistics Offices
ABS uses X-12-ARIMA
US Census Bureau uses X-13ARIMA-SEATS
Statistics Canada uses X-12-ARIMA
ONS (UK) uses X-12-ARIMA
EuroStat use X-13ARIMA-SEATS
38
For Better Official Statistics

39
 For Better Official Statistics

X-11 Decomposition
Advantages
✓Relatively robust to outliers
✓Completely automated choices for trend and seasonal changes
✓Very widely tested on economic data over a long period of time
Disadvantages
✓No prediction/confidence intervals
✓Ad hoc method with no underlying model
✓Only developed for quarterly and monthly data

40
 For Better Official Statistics

Extensions: X-12-ARIMA and X-13-ARIMA

✓The X-11, X-12-ARIMA and X-13-ARIMA methods are based on Census II
decomposition
✓These allow adjustments for trading days and other explanatory variables.
✓Known outliers can be omitted
✓Level shifts and ramp effects can be modelled
✓Missing values estimated and replaced
✓Holiday factors (e.g., Easter, Labour Day) can be estimated

41
 For Better Official Statistics

X-13ARIMA-SEATS
Advantages
✓Model-based
✓Smooth trend estimate
✓Allows estimates at end points
✓Allows changing seasonality
✓Developed for economic data
Disadvantages
✓Only developed for quarterly and monthly data

42
 STL
DECOMPOSITION
 For Better Official Statistics

STL Decomposition
✓ STL: “Seasonal and Trend decomposition using Loess”
✓ Very versatile and robust
✓ Unlike X-12-ARIMA, STL will handle any type of seasonality (daily, weekly, etc.)
✓ Seasonal component allowed to change over time, and rate of change controlled
by user
✓ Smoothness of trend-cycle also controlled by user
✓ Robust to outliers
✓ Not trading day or calendar adjustments
✓ Only additive
✓ Take logs to get multiplicative decomposition
✓ Use Box-Cox transformations to get other decompositions
44
 For Better Official Statistics

STL Decomposition (2)

✓ Algorithm that updates trend and seasonal components iteratively
✓ Starts with 𝑇෠𝑡 = 0
✓ Uses a mixture of loess and moving averages to successively refine the trend and
seasonal estimates
✓ The trend window controls loess bandwidth applied to deasonalised values
✓ The season window controls loess bandwidth applied to detrended subseries
✓ Robustness weights based on remainder

45
 For Better Official Statistics

STL Decomposition (3)

46
 TIME SERIES
DECOMPOSITION
WITH FOUR
COMPONENTS
 For Better Official Statistics

Dekomposisi 4 Komponen
✓ Misalnya time series 𝒀𝒕 didekomposisi menjadi 4 komponen
termasuk komponen siklus 𝑪𝒕 dengan model multiplikatif sebagai
berikut:
𝒀𝒕 = 𝑻𝒕 × 𝑪𝒕 × 𝑺𝒕 × 𝑰𝒕
𝒀𝒕
= 𝑪𝒕 × 𝑰𝒕
𝑻𝒕 × 𝑺𝒕
✓ Rata-rata bergerak dapat digunakan untuk memuluskan
irregularities, sehingga hanya meninggalkan komponen siklus
✓ Siklus adalah suatu perubahan atau gelombang naik atau turun
yang terjadi pada periode yang tidak tetap (durasi biasanya
minimal 2 tahun)

48
 For Better Official Statistics

Variasi Siklus
✓Kondisi perekonomian mengalami gelombang siklus,
contohnya :
▪ Resesi
▪ Pemulihan Mempunyai periode
▪ Ledakan - boom disebut lama siklus
▪ Krisis

✓ Lama siklus sulit ditentukan, namun dapat didekati dengan jarak

puncak-puncak atau lembah-lembah yang paling menonjol atau
sering berulang

49
 For Better Official Statistics

Contoh Penghitungan
Trend Indeks
Musiman 3 − 𝑀𝐴
Linier

Catur- 𝑻∙𝑪∙𝑰
Tahun 𝑌 𝑇 𝑆 𝑻 ∙ 𝑪 ∙ 𝑰 = 𝒀ൗ𝑺 𝑪∙𝑰= 𝐶
wulan 𝑻
1 60 47.56
2005 2 65 53.47 100.00 65.00 121.56
3 70 59.39 100.00 70.00 117.87 117.75
1 75 65.31 100.90 74.33 113.82 113.58
2006 2 78 71.22 100.43 77.67 109.05 107.85
3 80 77.14 103.00 77.67 100.68 99.74
1 75 83.06 100.90 74.33 89.50 89.99
2007 2 68 88.97 95.77 71.00 79.80
3 70 94.89
Total 641

50
 POLITEKNIK STATISTIKA STIS
For Better Official Statistics

Terima Kasih!