ISSN 1410 - 8046 (Print)

ISSN 2460 - 9196 (Online)

Volume 25, Number 3, 2022

Bulletin of Monetary Economics and Banking - Volume 25, Number 3, 2022

THE IMPACT OF DOMESTIC INVESTORS’ PARTICIPATION IN
GOVERNMENT DEBT ON BANK LOANS TO THE PRIVATE SECTOR: A
CROSS-COUNTRY STUDY
Justina Adamanti, Sugiharso Safuan, and Zaäfri Ananto Husodo

NOWCASTING REGIONAL ECONOMIC GROWTH IN INDONESIA

Jesica Nauli Br. Siringo Ringo and Anugerah Karta Monika

INDONESIA’S FINANCIAL STRESS EVENTS AND MACROECONOMIC

DYNAMICS
Sugiharso Safuan, Eric Alexander Sugandi, Okta Qomaruddin Aziz,
and Risna Triandhari

ASYMMETRIC IMPACTS OF MONETARY POLICY SHOCK ON OUTPUT

GAP: EVIDENCE FROM REGIONS IN INDONESIA
Sugeng Triwibowo and Defy Oktaviani

PANDEMIC SHOCKS AND MACRO-FINANCIAL POLICY RESPONSES:

AN ESTIMATED DSGE-VAR MODEL FOR INDONESIA
Advis Budiman, Sugiharso Safuan, Solikin M. Juhro, and
Febrio N. Kacaribu

THE SYSTEMIC RISK IN THE GULF COOPERATION COUNCIL

COUNTRIES’ EQUITY MARKETS AND BANKING SECTORS: A
DYNAMIC COVAR APPROACH
Aktham Maghyereh, Nader Virk, Basel Awartani, and
Mohammad Al Shboul

FROM SCHOOL TO WORK: DOES VOCATIONAL EDUCATION

IMPROVE LABOUR MARKET OUTCOMES? AN EMPIRICAL ANALYSIS
OF INDONESIA
Dyah S. Pritadrajati


Internationally Indexed by
Bulletin of Monetary Economics and Banking
Bank Indonesia

Patron
Board of Governors, Bank Indonesia

Editor-in-Chief
Dr. Perry Warjiyo, Bank Indonesia, Indonesia

Managing Editor
Prof. Paresh Kumar Narayan, Monash University, Australia

Co-Managing Editors
Sahminan, Ph.D., Bank Indonesia, Indonesia

Board of Editors
Prof. Hal Hill, Australian National University, Australia
Prof. Iwan Jaya Azis, Cornell University, USA
Prof. Anwar Nasution, University of Indonesia, Indonesia
Prof. Iftekhar Hasan, Fordham University, USA
Prof. Miranda S. Goeltom, University of Indonesia, Indonesia
Prof. Chun-Ping Chang, Shih Chien University, Taiwan
Prof. Insukindro, Gadjah Mada University, Indonesia
Prof. Ahmet Faruk Aysan, Hamad Bin Khalifa University, Qatar
Prof. Jonathan Batten, RMIT University, Australia
Prof. Niklas Wagner, University of Passau, Germany
Prof. Takahiro Akita, International University of Japan
Prof. Naoyuki Yoshino, Keio University, Tokyo, Japan
Dr. Iskandar Simorangkir, Bank Indonesia, Indonesia
Dr. Solikin M. Juhro, Bank Indonesia, Indonesia
Dr. Haris Munandar, Bank Indonesia, Indonesia

This bulletin is published by Bank Indonesia Institute, Bank Indonesia. Contents and research
outcome in the articles of this bulletin are entirely the responsibility of the authors and do not
represent Bank Indonesia’s views.

We invite all authors to write in this journal. The manuscript is submitted through the Bulletin of
Monetary Economics and Banking's (BMEB) Open Journal System on www.bmeb-bi.org.

The bulletin is published quarterly. Any person who wishes to obtain this publication shall contact
the Research Center of Bank Indonesia, D Tower, 10th floor, Jl. M.H. Thamrin No. 2, Jakarta, Phone
(62-21) 2981-4617. For subscription, please contact: email: [email protected].

Accredited - SK: 36A / E / KPT / 2016

 AUSTRALIAN
BUSINESS DEANS
COUNCIL
Bulletin of Monetary Economics and Banking

Volume 25, Number 3, 2022

CONTENT

The Impact of Domestic Investors’ Participation in Government Debt

on Bank Loans to the Private Sector: A Cross-country Study
Justina Adamanti, Sugiharso Safuan, and Zaäfri Ananto Husodo______ 275 - 290

Nowcasting Regional Economic Growth in Indonesia

Jesica Nauli Br. Siringo Ringo and Anugerah Karta Monika__________ 291 - 322

Indonesia’s Financial Stress Events and Macroeconomic Dynamics

Sugiharso Safuan, Eric Alexander Sugandi, Okta Qomaruddin Aziz,
and Risna Triandhari____________________________________________ 323 - 370

Asymmetric Impacts of Monetary Policy Shock on Output Gap:

Evidence from Regions in Indonesia
Sugeng Triwibowo and Defy Oktaviani____________________________ 371 - 398

Pandemic Shocks and Macro-Financial Policy Responses:

An Estimated DSGE-VAR Model for Indonesia
Advis Budiman, Sugiharso Safuan, Solikin M. Juhro, and
Febrio N. Kacaribu______________________________________________ 399 - 438

The Systemic Risk in the Gulf Cooperation Council Countries’ Equity

Markets and Banking Sectors: A Dynamic CoVaR Approach
Aktham Maghyereh, Nader Virk, Basel Awartani, and
Mohammad Al Shboul__________________________________________ 439 - 470

From School to Work: Does Vocational Education Improve Labour

Market Outcomes? An Empirical Analysis of Indonesia
Dyah S. Pritadrajati______________________________________________ 471 - 492
ii Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

This page is intentionally left blank

 Bulletin of Monetary Economics and Banking, Vol. 25 No. 3, 2022, pp. 291 - 322
p-ISSN: 1410 8046, e-ISSN: 2460 9196

NOWCASTING REGIONAL ECONOMIC GROWTH IN

INDONESIA

Jesica Nauli Br. Siringo Ringo* and Anugerah Karta Monika**

*Corresponding author. BPS Statistics, Indonesia. Email: [email protected]

** Politeknik Statistika STIS, Jakarta, Indonesia

ABSTRACT
This study aims to nowcast gross regional domestic product at the provincial level for
Indonesia. The dynamic factor model and mixed data sampling were applied to three
sets of variables; namely, macroeconomic, financial, and Google Trends. We find that
both methods captured several economic expansions and contractions, including the
recent downturn during the COVID-19 pandemic. By including the pandemic period,
accuracy across the same set of variables and provinces was slightly reduced.

Keywords: Nowcasting DFM; MIDAS; GRDP.

JEL Classifications: C38; C53; E27.

Article history:
Received : October 03, 2021
Revised : February 02, 2022
Accepted : June 24, 2022
Available Online : November 30, 2022
https://doi.org/10.21098/bemp.v25i3.1815
292 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

I. INTRODUCTION
The purpose of this study is to establish a nowcasting model of regional economic
growth in Indonesia. Economic growth is a widely used indicator for monitoring
and evaluating the economy, both at sectoral and regional levels. However,
regional economic growth or Gross Regional Domestic Product (GRDP) growth
is released with substantial delays, although up-to-date data on GRDP are
required for regional planning. This is released five weeks after the end of the
reference quarter. As a result, assessing regional economic activities becomes
difficult for policymakers, especially during an unusual event like COVID-19
pandemic; for a survey of the COVID-19 literature, see Narayan (2021). To obtain
an overview of current economic conditions, data nowcasting is performed by
economists. According to Agostino et al. (2015), nowcasting is an act of monitoring
economic conditions in real-time through high-frequency indicators, indicators
with different frequencies, and those with various release times. Gil et al. (2019)
used the terminology of very short-term forecasts to define nowcasting. Unlike
national data, regional data typically have a narrower data range and lower
quality due to the smaller number of samples and data sources. This issue made
regional nowcasting difficult but there is still a need to offer timely information.
In Indonesia, there was no nowcasting studies or components at the regional level
until recently. Therefore, we attempt to answer the following research question: by
using limited sets of variables at the regional level, can we perform nowcasting to
obtain accurate estimation of regional economic growth?
Another challenge in nowcasting is the timeliness of high-frequency data from
the official statistics. Google Trends is a database of Google search data that are
available on a daily, weekly, and monthly basis. Another question is whether the
quality by Google Trends data will create a more accurate estimation of regional
economic growth? The reason behind this was the assumption that the internet
search about economics could also capture the economic activity itself. Along
with the Google Trend data, the accuracy with data from official statistics which
was available on a monthly basis, that was openly accessible by anyone, and were
relevant to economic activity will be compared. Thus, this study addresses these
two research questions using monthly macroeconomic, financial, and Google
Trends data at the regional level.
To build projections that can describe conditions in the present quarter,
the nowcasting process requires monthly indicators and, as a result, we need
econometric methodologies that can mix data from different frequencies. By
disaggregating GDP into high-frequency series, Luthfiana and Nasrudin (2018)
came up with a solution. Rather than disaggregating GDP, two nowcasting
methodologies are used, namely the Dynamic Factor Model (DFM) with
expectation-maximization algorithms by Bańbura and Reichlin (2010) and the
unrestricted mixed data sampling (U-MIDAS) by Foroni et al. (2011). These
nowcasting models are motivated by the fact that, unlike other econometric
methods (e.g. Distributed Lag Model, Bridge Model), DFM and U-MIDAS can link
the monthly data to quarterly GRDP growth. DFM’s method can combine a large
number of variables into a single factor. This ensures that variables can be used
without being restricted by multicollinearity assumption. We also use U-MIDAS
that has been recognized for nowcasting variables by mixing data frequencies.
Nowcasting Regional Economic Growth in Indonesia 293

According to results, both models can capture the direction and magnitude
of current economic conditions using a limited set of variables. Also, the use of
Google Trends data generated the best nowcasting accuracy compared to those
from official statistics.
The contribution of this study are as follows: First, nowcasting studies based
on Indonesia are still focused at the national level. At that level, Tarsidin et al. (2018)
and Luciani et al. (2018) performed nowcasting studies, while Dewati et al. (2018)
developed a nowcasting model in Sumatra, Java, and Eastern Indonesia. This
study focuses on nowcasting at the regional level to provide the foundation for
regional economic assessment. Second, this is the first study to map Google Trends
topics and categories to the GRDP sector. The approach by Heikkinen (2019) and
Woloszko (2020) was integrated to leverage Google Trends topics and categories
to minimize the subjectivity of selecting the keywords in order to obtain Google
Trends index. Furthermore, the final contribution is to provide useful nowcasting
results due to the recent economic downturn. An appropriate approach is made
to the data range by distinguishing the period prior to the COVID-19 pandemic
from the period following the pandemic. By doing this, the models could still be
beneficial to estimate the current economic condition despite economic contraction
due to the pandemic.

II. DATA AND METHODOLOGY

A. Data
The dataset includes monthly variables and quarterly GRDP growth over the period
2012Q2 to 2020Q4 for 13 Indonesian provinces. The monthly variables consist
of farmer exchange rate, exports value, imports value, demand deposit, saving
deposit, time deposit, loans for working capital, loans for investment, loans for
consumption, credit to micro, small, and medium enterprise, and Google Trends
variables. Due to the availability of data, our study employed seven provinces out
of 34 provinces in Indonesia. According to the National Standardization Agency,
the provinces’ name abbreviations are North Sumatra (SU), West Sumatra (SB),
Riau (RI), Jambi (JA), Lampung (LA), Bangka Belitung Islands (BB), Riau Islands
(KR), Central Java (JT), Banten (BT), Bali (BA), East Java (JI), West Kalimantan
(KB), and South Kalimantan (KS). There are several provinces with zero search
volume indices (SVI) during the first few years of the data range considered in this
study and this is a limitation of the Google Trends data because of lower search
intensities. Due to this limitation, the nowcasting process can only be done in 7
provinces. An appropriate approach was made to the recent economic downturn
by dividing the data range into two sub-samples, namely a pre-pandemic (2012Q2-
2019Q4) and a period that includes pandemic (2012Q2-2020Q4). To establish an
appropriate approach to the COVID-19 pandemic where economic growth is
decreased drastically, the two types of periods are used. This situation will affect
the nowcasting performance, therefore, evaluating the benefit of the nowcasting
model in that period is necessary.
Moreover, our datset can be categorized into three categories namely,
macroeconomic, financial, and Google Trend. More specifically, there are three
macroeconomic variables, nine financial variables, and twelve Google Trends
294 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

variables. Google Trends data is collected by the application of a scraping data

method using the RStudio package called gtrendsR. This transformation is needed
to induce stationarity and to adjust the type of data. Table 1 provides the summary
of variables used in this study.

Table 1.
Summary of Variables Used in the Model
This table describes the variables used in this study. These variables are divided into three categories. The first
set includes macroeconomic variables from Badan Pusat Statistik (BPS), the official statistic agency Indonesia. The
second set is financial variables from Bank Indonesia, the central bank in Indonesia. The third set is Google Trends.
DFM block also refers to the name of sets of variables. Variable block abbreviation: FIN for Financial, MACRO for
Macroeconomics, GT for Google Trends. The transformation that we apply is also reported.
Transformation (1) is the difference q-to-q to the difference year-on-year: (Yt-Yt-4)-(Yt-1-Yt-5)
Transformation (2) is the difference month-to-month: Yt-Yt-1
Transformation (3) is growth year-on-year:
Transformation (4) is the year-on-year difference of the natural logarithm: ln Yt - lnYt-12

DFM Data Lag

Variable Name Unit Transformation
Block Source release
GRDP Growth PDRB - BPS 6 weeks % 1
Farmer Exchange
FER MACRO BPS 3 days Index 2
Rate
Exports EKS MACRO BPS 5 weeks Million US$ 3
Imports IMP MACRO BPS 5 weeks Million US$ 3
Demand Deposit GIR FIN BI 5 weeks Million Rp 4
Saving Deposit TAB FIN BI 5 weeks Million Rp 4
Time Deposit SIM FIN BI 5 weeks Million Rp 4
Working Capital MOD FIN BI 5 weeks Million Rp 4
Investment INV FIN BI 5 weeks Million Rp 4
Consumption KON FIN BI 5 weeks Million Rp 4
Micro MIK FIN BI 5 weeks Million Rp 4
Small KEC FIN BI 5 weeks Million Rp 4
Medium MEN FIN BI 5 weeks Million Rp 4
GT Google
Google Trends GT Real time Index 3
(A-I) Trends

Stock and Watson (1989) constructed a GDP index coincident based on

characteristics that follow the core concept of ​​calculating GDP. These include the
supply, demand, and income side of the equation. By considering the availability
of regional variables in Indonesia, three macroeconomic variables, namely export,
import, and Farmer Exchange Rate are used as the proxies for supply, demand,
and income component of GDP. The supply-side is represented by export, while
the demand-side is represented by import. Based on expenditure approaches,
export and import are the components of GRDP. According to Badan Pusat
Statistik (2020), Farmer Exchange Rate is used as a measure of purchasing power
of farmers in rural areas. Although Farmer Exchange Rate has never been utilized
in Indonesian nowcasting studies, this study uses it to portray the income side
because purchasing power can describe income, especially in rural areas.
Nowcasting Regional Economic Growth in Indonesia 295

In several nowcasting studies, financial variables have proven efficient in

capturing economic movements. According to Glocker and Wegmueller (2020),
banking indicators contain information about liquidity in the financial sector.
Bhadury et al. (2020) also formed a financial block in DFM due to its ability to
increase accuracy in nowcasting GDP growth. However, the financial variable is
more useful in capturing future conditions (leading indicator) and the recovery
period in economic activity. According to Narayan (2019), financial sectors,
specifically financial technology, have a significant and positive impact on
economic growth in Indonesia. Financial variables from Bank Indonesia is also
used in this analysis, which includes private deposits, loans, and credit to micro,
small, and medium enterprise. Private deposit is divided into demand, saving,
and time deposit. Furthermore, loan is divided by the purpose: working capital,
investment, and consumption, while credit to the enterprise is divided by the size
of the enterprise: micro, small, and medium.
One of the most notable big data sources provided by Google is Google Trends
data. This data is defined as an index based on a search ratio and Woloszko (2020)
explained the index using the following formula:

(1)

SVI refers to search volume indices, SV represents search volume, SVT

indicatestotal search volume, c represents the category or topic, and t is time.
Therefore, SVI is the ratio of searches for a particular category and time to the total
searches for that time. To ensure the index result is at a maximum value of 100, the
ratio is multiplied by the constant . The index value of 100 indicates
that the search was most popular at that time and location.
As proposed by Heikkinen (2019) and Woloszko (2020), a combination of topics
and categories was fitted to GRDP sectors based on the production approach. To
combine the subsectors into the GRDP sectors, the first component of the principal
component analysis is used. Table 2 reports the concordance between Google
Trends topics and categories and GRDP sectors.

Table 2.
Google Trends Topics and Categories
This table shows the concordance result of Google Trends topics and categories to GRDP sectors based on the
production approach. The first column is the industry or sector of GRDP based on the production approach. The
second column is the Google Trends topic/category that is fitted to GRDP sectors. The first component of principal
component analysis (PCA) is used to compress subsectors into a sector, e.g. food production, agriculture, animal
product & services, forestry are compressed to sector A (agriculture, forestry, and fishing).

Industry Google Trends Topic/Category

(1) (2)
A. AGRICULTURE, FORESTRY, AND FISHING
Food Crops Food Production
Plantation Crops Agriculture
Livestock Animal Product & Services
Forestry Forestry
296 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Table 2.
Google Trends Topics and Categories

Industry Google Trends Topic/Category

(1) (2)
B. MINING AND QUARRYING
Crude Petroleum, Natural Gas Oil & Gas
Coal, Lignite, Iron Ore, and Other Mining Metals & Mining
C. MANUFACTURING INDUSTRY
Manufacture of Food and Products and Beverages Food & Drink
Manufacture of Textiles, and Wearing Apparel Textiles & Nonwovens
Manufacture of Wood Wood
Manufacture of Paper and Paper Products Paper
Manufacture of Chemicals Chemical Industry
Manufacture of Cement Cement
Manufacture of Transport Equipment, Machinery, and
Automotive Industry
Equipment
D. ELECTRICITY AND GAS
Electricity Electricity
Manufacture of Gas Gasoline
Manufacture of Clean Water Water Filters & Purifiers
E. CONSTRUCTION
F. WHOLESALE AND RETAIL TRADE
Wholesale and Retail Trade Retail Trade
Hotel Hotel & Accommodation
Restaurants Restaurants
G. ACCOMMODATION AND COMMUNICATION
Railways Transport Rail Transportation
Land Transport Urban Transportation
Sea Transport Maritime Transportation
Air Transport Aviation, Airport Parking &
Communication Internet & Telecom
H. FINANCIAL, REAL ESTATE, AND BUSINESS ACTIVITIES
Bank Banking
Financial Institutions Non-Bank Financial Markets
Financial Supporting Service Business Finance
Real Estate Real Estate
Business Activities Business Service
I. SERVICES
Public Administration and Defence State & Local Government
Government Contracting &
Other Public Administration Service
Procurement
Social Society Social Services
Entertainment and Recreation Entertainment & Industry
Individual and Household Housing & Development
Nowcasting Regional Economic Growth in Indonesia 297

B. Methodology
In this study, we follow the DFM approach proposed by Bańbura and Reichlin
(2010) using maximum likelihood estimation in an expectation-maximization
algorithm. They state the model by the following specification:

(2)

xt=(x1t,x2t,…,xnt)’ is a set of stationary and standardized monthly variables for

t=1,…,T. The vector of common factors with order r×1 is expressed as ft where
we set a fixed number of factors, r=2. The vector of idiosyncratic components is
expressed as εt. The loading matrix of the factors that explained the contribution
of the variable to the common factor is expressed as Λ. The common factors, ft, are
assumed to follow the VAR process with order 2 as follows:

(3)

A1,A2 is the coefficient matrix of AR(p). Idiosyncratic components, εi,t, are

assumed to follow the AR(2) process as follows:

(4)

for i≠j. The index s represents the lag in the model where s=1,…,q.
Bańbura and Reichlin (2010) handle the mixed frequency issue by transforming
the quarterly growth of GRDP using Mariano and Murasawa (2003) approximation.
We notated ytQ as quarterly growth of GRDP and yt as an unobserved monthly
growth rate of GRDP where yt=ΔYtM. Mariano and Murasawa (2003) linked yt
with ytQ by assigning the value of the quarterly variable to the third month of the
respective quarter, which is denoted by YtQ,t=3,6,9,…. The quarterly level of GRDP
can be expressed as the sum of its unobserved monthly level, YtM:

(5)

yt is assumed to have the same factor model as the monthly variables as follows:

(6)

(7)

To link yt with the observed GRDP data, Mariano and Murasawa (2003)
constructed a partially observed monthly series and use the following
approximation:
298 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

(8)

The MIDAS variant used in this study is unrestricted MIDAS (U-MIDAS).

In this context, unrestricted means that the polynomial lag function of the
high frequency variable is not included. U-MIDAS is built with a simple linear
polynomial lag and can be estimated using Ordinary Least Squares (OLS) method.
Foroni et al. (2015) stated that U-MIDAS can be well applied to data with small
frequency differences, such as monthly and quarterly. Leboeuf and Morel (2014)
specified the U-MIDAS model as follows:

(9)

where Yt(Q) is a reference variable with quarterly frequency, Xt(Mj) is a monthly

variable.
The stationary test is done using Augmented Dickey-Fuller (ADF) test. ADF test
is also used to check the robustness of the model by testing the residual of the
fitted model. Gujarati (2004) states that the white noise assumption has an average
of zero, constant variance, and is not autocorrelated. Residuals that meet the
stationarity condition have also satisfied the white noise assumption.

III. MAIN FINDINGS

The discussion on main findings of this study is divided into five subsections. In the
first subsection, the GRDP growth rate movement is discussed, and in the second
subsection, the correlation variables are examined. The third subsection discusses
nowcasting GDP at the national level, while the fourth and fifth subsections
examine nowcasting GRDP using DFM and MIDAS, respectively.

A. GRDP Growth Movement

Every region’s economic growth follows a different pattern. However, the recent
economic downturn due to the COVID-19 pandemic affected all provinces. Figure 1
shows that in the second quarter of 2020, all provinces saw a significant contraction.
To prevent the spread of the virus, the government imposed social restrictions
in various provinces at the beginning of the pandemic. This resulted in a major
decline in business performance and income levels. Firms were restraining their
spendings, while households were limiting their consumption. Bali experienced
the deepest contraction in that period with a contraction rate of -11.06%. In the
third and fourth quarters of 2020, this value fell to -12%.
Nowcasting Regional Economic Growth in Indonesia 299

Figure 1.
Quarterly Growth of GRDP in 13 Provinces in Indonesia
This figure depicts the quarterly year-on-year growth of GRDP over the period 2019Q1-2020Q3 to show the period
before the COVID-19 pandemic the beginning of the COVID-19 pandemic period that directly affects economic
grwoth.

10

-5

-10

-15
Q1 Q2 Q3 Q4 Q1 Q2 Q3
2019 2020
North Sumatra West Sumatra Riau Jambi
Lampung Bangka Belitung Islands Riau Islands Central Java
East Java Banten Bali

B. Correlation of Economic Growth with Variables Used

Based on the correlation analysis of economic growth and 3-month-average of
all variables, Farmer Exchange Rate in Riau Islands had the strongest correlation
(0.76) with GRDP growth. In general, the correlation coefficients are relatively low.
The highest correlation value in West Kalimantan is 0.23. This means that none of
the variables in West Kalimantan has the correlation value that is exceeding 0.23.
Although the correlation between variables is small, nowcasting is still feasible.
According to Giannone et al. (2008), any information from variables that are
deemed unnecessary should not be discarded. Each variable may potentially affect
the current-quarter estimates and precision.
Farmer Exchange Rate in Bali and Riau Islands has the strongest link with
GRDP increase. This indicates that these variables can accurately describe
economic activity, especially in terms of income. The export variable has the
highest correlation among other monthly variables in West Kalimantan and South
Kalimantan. However, the financial variables have the highest correlation in
Riau, East Java, West Sumatra, Jambi, and Bangka Belitung Islands. In Lampung,
Banten, and Central Java, the Google Trends variable from the concordance sector
of electricity, gas, and clean water sector indicates a high correlation.
300 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Table 3.
Correlation Coefficient
This table reports the Pearson correlation coefficient between quarterly GRDP growth and all monthly variables. The
monthly variables used are transformed into 3-months-average.

Variable SU RI LA BT JT JI BA SB JA KR BB KB KS
FER 0.198 0.025 0.463 0.293 0.185 0.375 0.741* 0.587 0.509 0.760* 0.520 0.171 0.610
EKS 0.379 0.366 0.486 0.041 0.255 0.062 0.575 0.499 0.499 0.607 0.441 0.234* 0.735*
IMP 0.457 0.287 0.628 0.475 0.441 0.127 0.378 0.555 0.239 0.576 0.250 0.037 0.634
GIR 0.378 0.223 0.455 0.469 0.415 0.550 0.110 0.595 0.624 0.402 0.275 0.169 0.633
TAB 0.621* 0.263 0.513 0.558 0.447 0.563 0.500 0.627 0.673 0.675 0.497 0.145 0.489
SIM 0.530 0.356 0.481 0.540 0.323 0.509 0.490 0.648* 0.690 0.702 0.562 0.179 0.597
MOD 0.454 0.311 0.560 0.523 0.367 0.432 0.409 0.597 0.700* 0.510 0.564* 0.183 0.554
INV 0.602 0.310 0.675 0.436 0.506 0.517 0.490 0.629 0.659 0.673 0.447 0.169 0.686
KON 0.578 0.259 0.468 0.511 0.337 0.454 0.450 0.633 0.698 0.719 0.540 0.142 0.599
MIK 0.528 0.217 0.572 0.527 0.389 0.570* 0.494 0.616 0.646 0.690 0.520 0.170 0.451
KEC 0.479 0.273 0.510 0.510 0.442 0.564 0.476 0.575 0.673 0.710 0.536 0.216 0.589
MEN 0.365 0.511* 0.521 0.464 0.343 0.420 0.396 0.214 0.628 0.419 0.493 0.085 0.645
GT_A 0.223 0.311 0.566 0.636 0.539 0.299 0.450
GT_B 0.264 0.297 0.626 0.587 0.573 0.335 0.514
GT_C 0.456 0.326 0.628 0.574 0.555 0.478 0.477
GT_D 0.220 0.348 0.681* 0.662* 0.617* 0.422 0.485
GT_E 0.119 0.252 0.484 0.456 0.434 0.200 0.394
GT_F 0.187 0.167 0.484 0.420 0.343 0.236 0.139
GT_G 0.048 0.202 0.434 0.293 0.304 0.013 0.136
GT_H 0.358 0.308 0.618 0.550 0.539 0.394 0.407
GT_I 0.078 0.279 0.497 0.525 0.497 0.185 0.317
*variable with the highest correlation amongst a province

C. Nowcasting Indonesia’s GDP Growth

To analyze the consistency of variables used for nowcasting at the regional and
national levels, the nowcasting model is compared using different sets of variables.
At the national level, we use the variables that are commonly used in the national
level nowcasting studies: export, import, narrow money, broad money, index of the
manufacturing industry, index of general wholesale prices, and foreign tourists.
Table 4 shows that the Root Mean Squared Error (RMSE) values are similar for
each set of variables. The national level variables in nowcasting studies produce
the lowest accuracy. This shows that while nowcasting at the national level can use
the same variables as nowcasting at the regional level, the outcomes will be worse. 
Gil et al. (2019) mention that using the same variables in nowcasting models
at the regional and national levels will produce similar accuracy. However, the
results at the national level are smoother and more accurate and this is because of
the various data problems at the regional level. By taking into account the result
at the national level, the use of these variables at the regional level is possible once
the results at the national level are accurate. 
Nowcasting Regional Economic Growth in Indonesia 301

Table 4.
RMSE of Nowcasting at National Level-by Sets of Variables
Regional level variables are the sets of variables that we will be used at our study at the regional level. National
level variables are the commonly used variables at the national level nowcasting studies. These variables are only
available at the national level, except export and import. M1 is narrow money, M2 is broad money, IBS is index of
the manufacturing industry, IHPB is index of general wholesale prices, foreign tourists are all foreign visitor directly
arrived in Indonesia. FIN+MACRO covers all the variables from macroeconomics and financial categories. FIN refers
to financial variables, MACRO represnets Macroeconomics variables, and GT represents Google Trend variables.

Set of
Regional Level Variables National Level Variables
Variables
Estimation
DFM MIDAS DFM
Method
Export, Import, M1, M2, IBS, IHPB,
FIN+
Sets Variable GT FIN MACRO GT Foreign Tourists, Rice Price, Coal
MACRO
Price
RMSE 1.0586 1.5049 1.1977 0.1016 0.8647 1.0737

D. Nowcasting Regional Economic Growth using DFM

Figure 2 presents results of nowcasting with financial, macroeconomic, and Google
Trend variables. This figure shows that the nowcasting model can capture the
economic downturn in the COVID-19 pandemic. In the first quarter of 2020, a low
magnitude along with the economic shock was demonstrated by the nowcasting
results. However, in the fourth quarter of 2020, economic movements in the seven
provinces began to experience expansion and a rebound economic pattern was
also shown in the nowcasting models.

Figure 2.
DFM Nowcasting Results in 7 Provinces-by Sets of Variables
This figure shows the year-on-year growth rate of GRDP (black line) compared to DFM nowcast using financial and
macroeconomic variables (blue line), and google trends variables (red line) in 7 Indonesian provinces.

A. North Sumatera
6
4
2
0
-2

PDRB
FIN+MAKRO
GT
-4

2014 2016 2018 2020

302 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Figure 2.
DFM Nowcasting Results in 7 Provinces-by Sets of Variables (Continued)
B. Riau
8
6
4
2
2
-2

PDRB
-4

FIN+MAKRO
GT
-6

2014 2016 2018 2020

C. Lampung
10
8
6
4
2
0

PDRB
-2

FIN+MAKRO
GT
-4

2014 2016 2018 2020

Nowcasting Regional Economic Growth in Indonesia 303

Figure 2.
DFM Nowcasting Results in 7 Provinces-by Sets of Variables (Continued)
D. Banten
5
0
-5

PDRB
FIN+MAKRO
GT
-0

2014 2016 2018 2020

E. Central Java
6
4
2
0
-2
-4

PDRB
FIN+MAKRO
GT
-6

2014 2016 2018 2020

304 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Figure 2.
DFM Nowcasting Results in 7 Provinces-by Sets of Variables (Continued)
F. East Java
5
0

PDRB
-5

FIN+MAKRO
GT

2014 2016 2018 2020

G. Bali
5
0
-5
-10

PDRB
FIN+MAKRO
GT
-15

2014 2016 2018 2020

Figure 3 depicts nowcasting in six provinces without the Google Trend

variables. Overall, the decline in the economic movement since the COVID-19
pandemic can be captured using nowcasting with financial and macroeconomic
variables. This is demonstrated by the low magnitude since the second quarter of
2020. In the second quarter, the negative values are
​​ obtained in Riau Islands and
Bangka Belitung Islands, while in the third quarter they are found negative in all
provinces. 
Nowcasting Regional Economic Growth in Indonesia 305

Figure 3.
DFM Nowcasting Results in 6 Provinces-by Sets of Variables
This figure shows the year-on-year growth rate of GRDP (black line) compared to DFM nowcast using financial and
macroeconomic variables (blue line) in 6 Indonesian provinces.

A. West Sumatra
5
0

PDRB
-5

FIN+MAKRO

2014 2016 2018 2020

B. Jambi
10
5
0

PDRB
FIN+MAKRO

2014 2016 2018 2020

306 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Figure 3.
DFM Nowcasting Results in 6 Provinces-by Sets of Variables (Continued)
C. Riau Islands
5
0

PDRB
-5

FIN+MAKRO

2014 2016 2018 2020

D. Bangka Belitung Islands

6
4
2
0
-2
-4

PDRB
FIN+MAKRO
-6

2014 2016 2018 2020

Nowcasting Regional Economic Growth in Indonesia 307

Figure 3.
DFM Nowcasting Results in 6 Provinces-by Sets of Variables (Continued)
E. West Kalimantan
8
6
4
2
0
-2

PDRB
FIN+MAKRO
-4

2014 2016 2018 2020

F. South Kalimantan
8
6
4
2
0
-2

PDRB
-4

FIN+MAKRO

2014 2016 2018 2020

Table 5 presents the DFM performance for 13 Indonesian provinces. The

nowcasting results are compared with the univariate model, AR(2), and the
forecasting results across provinces. Our findings suggest that DFM’s performance
relative to AR(2) is better in the data range that includes economic shocks.
Therefore, this implies that economic shock is better captured with econometric
models that include more information than using univariate models. Assuming
the comparison is made without considering the ratio of the two models, the
results reveal that both DFM and AR(2), produce smaller RMSE in data before the
pandemic. This indicates that the models perform better without including the
economic shock period.
308 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

The results of GRDP forecasting in the North Sumatra, West Sumatra,

Jambi, Riau Islands, Central Java, Banten, Bali, East Java, and South Kalimantan
indicates that adding more observations for GRDP forecasting in 2019Q4 does not
significantly increase the forecasting accuracy between the forecasting periods.
In this case, the difference between the forecasted periods is merely 0.013 point.
For example, the RMSE of Riau remains at 1.03, in backcasting, nowcasting, and
forecasting. This may indicate that the variables used are not exposed to significant
shocks. Also, this condition occurs in several provinces experiencing economic
shocks, namely Riau Islands and South Kalimantan.

Table 5.
DFM RMSE of Nowcasting Using Financial and Macro Sets of Variables-by
Methodologies and Provinces
This table reports the accuracy measurement using Root Mean Squared Error (RMSE). The notation -2, -1, 0, 1, 2
states the number of months before the release of the quarterly GRDP value. Backcasting is conducted in months (-2)
and (-1). Nowcasting is conducted in the month (0). Forecasting is conducted in months (1) and (2).  2019Q4 refers to
the nowcast of the fourth quarter of 2019, period prior to the COVID-19 pandemic. 2020Q4 refers to the nowcast of
the fourth quarter of 2020, period that includes COVID-19 pandemic. Order 2 for AR is chosen based on the use of
lag order in DFM so that the forecast includes the same amount of lag. FIN+MACRO includes macroeconomics and
financial variables.

FIN+MACRO Relative to AR(2)

Province AR(2)
-2 -1 0 1 2 -2 -1 0 1 2
Panel A: 2019Q4
SU 0.4545 0.3809 0.3208 0.3247 0.3255 0.3248 0.8382 0.7058 0.7144 0.7163 0.7147
SB 2.7956 1.0419 1.0319 1.0279 1.0234 1.0258 0.9700 0.9607 0.9570 0.9528 0.9550
RI 1.2916 2.3490 2.3720 2.3739 2.3570 2.3706 0.8403 0.8485 0.8491 0.8431 0.8480
JA 1.2577 2.2381 2.2353 2.2349 2.2341 2.2440 0.8567 0.8556 0.8555 0.8552 0.8589
LA 1.4237 1.4391 1.4450 1.4362 1.4307 1.4408 1.1142 1.1188 1.1120 1.1077 1.1155
BB 1.3052 1.2404 1.4403 1.4497 1.4539 1.4078 0.6920 0.8035 0.8088 0.8111 0.7854
KR 0.5317 1.1680 1.1851 1.1831 1.1866 1.1765 1.1324 1.1490 1.1471 1.1505 1.1407
JT 1.0741 1.0412 1.0351 1.0358 1.0346 1.0378 0.8278 0.8230 0.8236 0.8226 0.8252
BT 2.6125 1.1986 1.2103 1.2035 1.1972 1.1974 0.8419 0.8501 0.8453 0.8409 0.8410
BA 1.7924 1.1509 1.1613 1.1795 1.1773 1.1857 0.8818 0.8898 0.9037 0.9020 0.9084
JI 1.0314 0.3998 0.3995 0.3994 0.3994 0.4078 0.7520 0.7513 0.7511 0.7512 0.7669
KB 2.7152 2.4343 2.4324 2.4625 2.4626 2.4654 0.8965 0.8958 0.9069 0.9070 0.9080
KS 1.2595 1.3295 1.3309 1.3318 1.3293 1.3413 1.0555 1.0567 1.0574 1.0554 1.0649
Panel B: 2020Q4
SU 1.26 0.6676 0.6849 0.6892 0.6901 0.6932 0.5299 0.5436 0.5470 0.5477 0.5502
SB 1.892 1.3118 1.3261 1.3011 1.3063 1.3050 0.6933 0.7009 0.6876 0.6904 0.6897
RI 2.836 2.3277 2.3229 2.3179 2.3174 2.3501 0.8207 0.8190 0.8173 0.8171 0.8286
JA 2.629 2.1275 2.1731 2.1737 2.1766 2.2270 0.8093 0.8267 0.8269 0.8280 0.8471
LA 1.715 1.0363 1.0383 1.0422 1.0423 0.9978 0.6044 0.6056 0.6079 0.6079 0.5820
BB 1.941 1.4029 1.4949 1.4867 1.4871 1.5540 0.7229 0.7703 0.7661 0.7663 0.8008
KR 2.352 1.4893 1.4877 1.4824 1.4759 1.5427 0.6332 0.6325 0.6303 0.6275 0.6559
JT 2.033 1.3819 1.3749 1.3854 1.3869 1.3555 0.6798 0.6763 0.6815 0.6822 0.6668
BT 2.299 1.5795 1.5767 1.5738 1.5692 1.7140 0.6872 0.6859 0.6847 0.6827 0.7457
BA 2.255 1.2523 1.2653 1.2521 1.2115 1.3184 0.5553 0.5611 0.5552 0.5373 0.5847
JI 1.711 1.1675 1.1966 1.1906 1.1845 1.5350 0.6823 0.6993 0.6958 0.6923 0.8971
KB 2.94 2.4322 2.4724 2.4586 2.4742 2.4791 0.8272 0.8409 0.8362 0.8415 0.8431
KS 1.798 1.3654 1.3666 1.3675 1.3805 1.4014 0.7595 0.7602 0.7606 0.7679 0.7795
Nowcasting Regional Economic Growth in Indonesia 309

For further examination, we compare the nowcasting results of Google Trend

variables andfinancial and macroeconomic variables by the econometric methods:
the DFM and AR(2). Table 6 shows that the RMSE obtained using the Google Trend
variables differs for each province. In general, all estimates before the pandemic
show a smaller RMSE across provinces. This means that the model works better
with the absence of economic shock. One of the reasons behind this is that this
model does not distinguish between the treatment of non-linearity in the data.
Woloszko’s (2020) revealed similar findings for a period that include economic
shock due to the global economic crisis in 2008 indicating that forecasting results
have worsened.
In the sample period up to 2019Q4 which excludes economic shock due to
the COVID-19 pandemic, forecasting with the Google Trend variables produces
better projections in all provinces, except in the case of Riau, Lampung, and East
Java than those financial and macroeconomic variables. While comparing with
the univariate model, our results suggest that North Sumatra, Banten, and Bali
outperformed the AR model, while Riau, Lampung, Central Java, and East Java
have higher accuracy using the AR model. This means that the Google Trend
variables has been able to provide meaningful information in cases where the
RMSE of those provinces is only slightly different from others, except in Riau. This
demonstrates that, in the presence of stable economic conditions, the Google Trend
variables can be effectively used to explain the direction of economic movement. 

Table 6.
DFM RMSE of Nowcasting Using Google Trends Set of Variabels-by
Methodologies and Provinces
This table reports the accuracy measurement of DFM using root mean squared error (RMSE). The result based on
Google Trend variables is compared to FIN+MACRO variables. FIN+MACRO includes all macroeconomics and
financial variables. FIN refers to financial variables, MACRO represnets Macroeconomics variables, and GT represents
Google Trend variables.

2019Q4 2020Q4
Province Relative to Relative to Relative to Relative to
GT GT
FIN+MACRO AR(2) FIN+MACRO AR(2)
SU 0.1339 0.2945 0.3517 1.0688 1.1846 0.8484
RI 4.7698 1.7062 2.0181 5.1350 2.2108 1.8106
LA 1.3311 1.0306 1.2821 1.7085 1.4280 0.9965
JT 1.2496 0.9936 1.2101 3.2775 2.3719 1.6122
BT 0.5517 0.3875 0.4763 2.1523 1.4409 0.9364
BA 0.1957 0.1499 0.1650 2.0710 1.5540 0.9184
JI 1.0699 2.0123 2.6991 2.8971 2.3911 1.6931

E. Nowcasting Regional Economic Growth Using MIDAS

In general, MIDAS nowcasting models based on financial, macroeconomic,
Google Trend variables, shown in Figure 4 reveal the direction of economic
decline since the pandemic. Some models in several provinces can well depict the
fall in economic growth that reached negative values in all provinces in the second
quarter of 2020. Furthermore, economic rebound movements can be captured
by nowcasting in the fourth quarter of 2020 as expansion begins. Lampung has
310 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

shown the best nowcasting result in the fourth quarter of 2020. This captured
economic rebound movements due to expanding economic conditions. Visually,
the nowcasting movement using DFM was smoother than the MIDAS. 

Figure 4.
MIDAS Nowcasting Results in 7 Provinces-by Sets of Variables
This figure shows the year-on-year growth rate of GRDP (black line) compared to MIDAS nowcast using financial
variables (blue line), macroeconomic (green line), and google trends variables (red line) in 7 Indonesian provinces.

A. North Sumatra
5
0
-5

PDRB MAKRO
FIN GT
-10

2014 2016 2018 2020

B. Riau
5
0

PDRB MAKRO
-5

FIN GT

2014 2016 2018 2020

Nowcasting Regional Economic Growth in Indonesia 311

Figure 4.
MIDAS Nowcasting Results in 7 Provinces-by Sets of Variables (Continued)
C. Lampung
10
8
6
4
2
0

PDRB MAKRO
-2

FIN GT
-4

2014 2016 2018 2020

D. Banten
10
5
0
-5

PDRB MAKRO
FIN GT

2014 2016 2018 2020

312 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Figure 4.
MIDAS Nowcasting Results in 7 Provinces-by Sets of Variables (Continued)

E. Central Java
5
0

PDRB MAKRO
-5

FIN GT

2014 2016 2018 2020

F. East Java
10
5
0
-5

PDRB MAKRO
FIN GT
-10

2014 2016 2018 2020

Nowcasting Regional Economic Growth in Indonesia 313

Figure 4.
MIDAS Nowcasting Results in 7 Provinces-by Sets of Variables (Continued)
G. Bali
10
5
0
-5
-10

PDRB MAKRO
-15

FIN GT

2014 2016 2018 2020

In general, the nowcasting results in Figure 5 show that the MIDAS model
accurately captures the fall in economic activity that happened in those six
provinces during the COVID-19 pandemic. Except in West Sumatra, almost
all models with macroeconomic variables reports negative magnitudes in the
second quarter of 2020. Only the Riau Islands can capture negative values with
the financial variables during that period. Several regions, like West Sumatra and
the Bangka Belitung Islands, witnessed an increase in nowcasting direction in the
fourth quarter of 2020, while Jambi continued to fall. 

Figure 5.
MIDAS Nowcasting Results in 6 Provinces-by Sets of Variables
This figure shows the year-on-year growth rate of GRDP (black line) compared to MIDAS nowcast using financial
variables (blue line), and macroeconomic variables (green line) in 6 Indonesian provinces.

A. West Sumatra
8
6
4
2
0
-2

PDRB
FIN
-4

MAKRO

2014 2016 2018 2020

314 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Figure 5.
MIDAS Nowcasting Results in 6 Provinces-by Sets of Variables (Continued)
B. Jambi
12
10
8
6
4
2

PDRB
0

FIN
-2

MAKRO

2014 2016 2018 2020

C. Riau Islands
10
5
0

PDRB
FIN
-5

MAKRO

2014 2016 2018 2020

Nowcasting Regional Economic Growth in Indonesia 315

Figure 5.
MIDAS Nowcasting Results in 6 Provinces-by Sets of Variables (Continued)
D. Bangka Belitung Islands
6
4
2
0
-2

PDRB
FIN
-4

MAKRO

2014 2016 2018 2020

E. West Kalimantan
10
5
0

PDRB
FIN
-5

MAKRO

2014 2016 2018 2020

316 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Figure 5.
MIDAS Nowcasting Results in 6 Provinces-by Sets of Variables (Continued)
F. South Kalimantan
6
4
2
0
-2

PDRB
-4

FIN
MAKRO
-6

2014 2016 2018 2020

Based on the comparison results between data ranges, the nowcasting before
the COVID-19 pandemic provides better results (see Table 7). This indicates that
the MIDAS model works better by removing the period of economic shock, except
for the Google Trend variables in East Java. The results in East Java shows that the
use of the Google Trend variables in the period that includes COVID-19 pandemic
also significantly outperformed the other set of variables.
According to the result of nowcasting using the MIDAS method for two data
ranges, the forecast with the Google Trend variables produced the smallest RMSE
compared to financial and macroeconomic variables. Since this model can include
information from all the months contributed to the current quarter, the availability
of Google Trend variables in real-time is an additional advantage. Meanwhile,
financial and macroeconomic variables only include information from the first
month of the current quarter and its lags. 

Table 7.
MIDAS RMSE of Nowcasting Using Financial, Macro, and Google Trends Sets of
Variables-by Methodologies and Provinces
This table reports the accuracy measurement of the MIDAS model using Root Mean Squared Error (RMSE). FIN is not
divided into nine variables as in Table 1 due to the number of variables in parsimony MIDAS model. FIN is divided
into three variables of private deposit, loans, and credit to micro, small, and medium enteprise by aggregating the
nine variables.

MIDAS Relative to AR(2)

Province AR(2)
FIN MACRO GT FIN MACRO GT
Panel A: 2019Q4
SU 0.4545 0.1152 0.1872 0.0024 0.2534 0.4119 0.0054
RI 2.7956 6.6534 5.0195 1.3032 2.3800 1.7955 0.4662
LA 1.2916 1.4660 0.9418 0.3810 1.1350 0.7292 0.2949
JT 1.2577 0.8151 0.5906 0.4045 0.6481 0.4696 0.3216
BT 1.4237 1.2523 0.9694 0.5371 0.8796 0.6809 0.3773
Nowcasting Regional Economic Growth in Indonesia 317

Table 7.
MIDAS RMSE of Nowcasting Using Financial, Macro, and Google Trends Sets of
Variables-by Methodologies and Provinces (Continued)
MIDAS Relative to AR(2)
Province AR(2)
FIN MACRO GT FIN MACRO GT
BA 1.3052 1.3881 0.8532 0.1086 1.0635 0.6537 0.0832
JI 0.5317 0.1908 0.2490 1.6820 0.3588 0.4683 3.1634
SB 1.0741 0.5397 0.6595 0.5025 0.6140
JA 2.6125 6.1407 4.1465 2.3505 1.5872
BB 1.7924 2.4549 1.0979 1.3696 0.6125
KR 1.0314 0.8766 1.3440 0.8499 1.3030
KB 2.7152 5.2611 3.4639 1.9377 1.2757
KS 1.2595 0.6759 1.0195 0.5366 0.8095
Panel B: 2020Q4
SU 1.26 1.2259 0.3187 0.0070 0.9730 0.2530 0.0056
RI 1.892 6.8862 4.8062 1.6820 2.4281 1.6946 0.5931
LA 2.836 1.9058 1.2122 0.9718 1.1116 0.7070 0.5668
JT 2.629 1.5882 2.0516 0.4159 0.7813 1.0092 0.2046
BT 1.715 2.5014 2.6631 0.7080 1.0882 1.1586 0.3080
BA 1.941 3.1114 1.1810 1.0907 1.3798 0.5237 0.4837
JI 2.352 1.8717 2.0334 0.0525 1.0938 1.1884 0.0307
SB 2.033 1.5054 1.5673 0.7956 0.8284
JA 2.299 6.3408 4.7507 2.4120 1.8072
BB 2.255 4.0595 1.5255 2.0919 0.7861
KR 1.711 1.6655 4.1195 0.7081 1.7515
KB 2.94 6.8001 5.5033 2.3127 1.8717
KS 1.798 1.7151 1.4742 0.9540 0.8200

F. Robustness Check
The robustness test is established by testing the residual of the models to confirm
the model adequacy. Table 8 reports results of the ADF unit root test. Our results
indicates that the residual of the fitted models is white noise. The t-test value is
compared to Mackinnon’s critical value and once the t-test value is less than the
critical value, the test results are considered in the rejected region. The critical value
of Mackinnon is -1,95 and according to the results, all the t-test values are less than
the critical value. Therefore, the residuals of all the fitted DFMs are concluded to
be statistically white noise.
318 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

Table 8.
DFM Robustness Check-ADF Test
This table reports the results of the ADF test of DFM’s residuals with the null hypothesis of a unit root.

2019Q4 2020Q4
Province FIN+MACRO GT FIN+MACRO GT
t-Stat p-value t-Stat p-value t-Stat p-value t-Stat p-value
SU -7.6407 0.01 -6.1859 0.01 -5.4886 0.01 -6.2133 0.01
RI -8.6342 0.01 -5.9486 0.01 -8.8206 0.01 -6.4598 0.01
LA -6.1122 0.01 -6.2876 0.01 -5.9483 0.01 -6.0108 0.01
BT -9.543 0.01 -7.1830 0.01 -7.8082 0.01 -5.4060 0.01
JT -12.0953 0.01 -6.8009 0.01 -8.3887 0.01 -6.1365 0.01
JI -7.7764 0.01 -6.8003 0.01 -6.0545 0.01 -6.6142 0.01
BA -10.2676 0.01 -6.6241 0.01 -8.3873 0.01 -5.5662 0.01
SB -9.8484 0.01 -8.909 0.01
JA -8.5525 0.01 -8.9187 0.01
KR -5.2809 0.01 -6.5007 0.01
BB -6.6305 0.01 -6.3656 0.01
KB -10.8239 0.01 -9.9954 0.01
KS -10.3879 0.01 -8.0919 0.01

Table 9 reports results of an ADF unit root test for the residual obtained from
MIDAS modelsOur results inicate that all the -test values are less than the critical
value. Therefore, the residuals of all the fitted MIDAS models are also statistically
white noise.

Table 9.
MIDAS Robustness Check-ADF Test
This table reports the results of the ADF test of MIDAS’ residuals with the null hypothesis of a unit root.

FIN MACRO GT
Province
t-Stat p-value t-Stat p-value t-Stat p-value
Panel A: 2019Q4
SU -5.03 0.01 -6.0839 0.01 -4.9854 0.01
RI -6.0553 0.01 -6.3129 0.01 -5.2437 0.01
LA -6.3996 0.01 -6.7547 0.01 -5.2949 0.01
BT -4.212 0.015 -4.5101 0.01 -7.0843 0.01
JT -6.1316 0.01 -5.4985 0.01 -4.6146 0.01
JI -5.2226 0.01 -4.5264 0.01 -5.5617 0.01
BA -5.971 0.01 -5.274 0.01 -4.6859 0.01
SB -5.8778 0.01 -5.0079 0.01 -5.8962 0.01
JA -5.5877 0.01 -5.8066 0.01 -6.0575 0.01
KR -5.0118 0.01 -6.0602 0.01 -4.631 0.01
BB -4.6231 0.01 -5.9785 0.01 -4.9594 0.01
KB -5.6407 0.01 -6.176 0.01 -6.9523 0.01
KS -5.6889 0.01 -4.6871 0.01 -5.7938 0.01
Nowcasting Regional Economic Growth in Indonesia 319

Table 9.
MIDAS Robustness Check-ADF Test (Continued)
FIN MACRO GT
Province
t-Stat p-value t-Stat p-value t-Stat p-value
Panel B: 2020Q4
SU -4.6183 0.01 -10.0129 0.01 -5.9585 0.01
RI -4.926 0.01 -5.6935 0.01 -6.3188 0.01
LA -4.8337 0.01 -5.8354 0.01 -5.4395 0.01
BT -6.4209 0.01 -4.5637 0.01 -5.6331 0.01
JT -5.32 0.01 -9.2904 0.01 -8.7746 0.01
JI -5.001 0.01 -4.7999 0.01 -8.0197 0.01
BA -5.9125 0.01 -5.434 0.01 -8.2113 0.01
SB -5.3242 0.01
JA -6.3516 0.01
KR -5.7286 0.01
BB -5.5299 0.01
KB -5.8105 0.01
KS -5.5471 0.01

IV. CONCLUDING REMARKS

The urge to learn about the current economic conditions based on data is strong;
however, it is limited due to substantial delays in the release of GRDP growth
data. Based on the nowcasting results using a selected set of variables, this
study concludes that nowcasting quarterly GRDP growth can be achieved at the
provincial level in Indonesia using macroeconomics, financial, and Google Trends
data. As the COVID-19 pandemic began in the first quarter of 2020, the DFM and
MIDAS models are able to capture the reduction in economic activity. Apart from
the pandemic, the two models accurately reflect several economic expansions and
contractions that occurred in several provinces. The results of nowcasting using
the DFM method were adapted to economic growth in the fourth quarter of 2020.
Except for the macroeconomic variables in the Jambi and Bangka Belitung Islands,
the MIDAS model also produces negative magnitudes.
The DFM and MIDAS model performance shows better accuracy in a data
range that does not include periods of economic shocks. Model accuracy across
the same set of variables and provinces is slightly reduced once the pandemic
period is included. Furthermore, the use of Google Trends variable produces the
best accuracy when compared to other categories of variable using the MIDAS
model, both before and after the COVID-19 pandemic. Using the DFM model, the
use of Google Trends variables also produces the best accuracy, but only in the
period leading up to the pandemic.
The nowcasting procedure can be adopted by researchers and policy makers,
including the central bank and local government. Every time new monthly data is
provided, the nowcasting model can be updated. Therefore, the local government
can adapt the Federal Reserve Bank of New York’s method, which publishes the
most recent nowcasting results on their official website. This can be useful for
provincial development plans, as well as information for policy makers.
320 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

REFERENCES
Agostino, A. D., Giannone, D., Lenza, M., & Modugno, M. (2015). Nowcasting
Business Cylce: a Bayesian Approach to Dynamic Heterogeneous Factor
Models. Finance and Economics Discussion Series, 066.
Badan Pusat Statistik. (2020). Statistik Nilai Tukar Petani. BPS.
Bańbura, M., Giannone, D., & Reichlin, L. (2010). Nowcasting. ECB Working Paper
No. 1275.
Bhadury, S., Ghosh, S., & Kumar, P. (2020). Nowcasting Indian GDP Growth Using
a Dynamic Factor Model. RBI Working Paper Series, 3.
Dewati, W., Prakasa, R. R., Fitrama, R., Ariyanti, D., Pratama, D. H., Sendrata,
D., Warsono, & Syafii, E. (2018). Nowcasting Konsumsi Rumah Tangga dan
Investasi Regional Sumatra, Jawa, dan Kawasan Timur Indonesia (KTI).
Working Paper Bank Indonesia, 10.
Foroni, C., Marcellino, M., & Schumacher, C. (2011). U-MIDAS : MIDAS Regressions
with Unrestricted Lag Polynomials, 178, 57–82.
Foroni, C., Marcellino, M., & Schumacher, C. (2015). Unrestricted Mixed Data
Sampling (MIDAS): MIDAS Regressions with unrestricted Lag Polynomials.
Journal of the Royal Statistical Society. Series A: Statistics in Society.
Giannone, D., Reichlin, L., & Small, D. (2008). Nowcasting: The Real-time
Informational Content of Macroeconomic Data. Journal of Monetary Economics,
55, 665–676.
Gil, M., Leiva‐Leon, D., Perez, J. J., & Urtasun, A. (2019). An Application of
Dynamic Factor Models to Nowcast Regional Economic Activity in Spain.
Banco de Espana Occasional Paper, 1904.
Glocker, C., & Wegmueller, P. (2020). Business Cycle Dating and Forecasting with
Real-time Swiss GDP Data. Empirical Economics, 58, 73–105.
Gujarati, D. N. (2004). Basic Econometrics. Tata McGraw-Hill.
Heikkinen, J. (2019). Nowcasting GDP Growth Using Google Trends. Jyväskylä
University.
Leboeuf, M., & Morel, L. (2014). Forecasting Short-term Real GDP Growth in the
Euro Area and Japan Using Unrestricted MIDAS Regressions. Bank of Canada
Discussion Paper, 3.
Luciani, M., Pundit, M., Ramayandi, A., & Veronese, G. (2018). Nowcasting
Indonesia. Empirical Economics, 55, 597–619.
Luthfiana, P. S., & Nasrudin, N. (2018). Disaggregation and Forecasting of the
Monthly Indonesian Gross Domestic Product (GDP). Bulletin of Monetary
Economics and Banking, 20, 529–556.
Mariano, R. S., & Murasawa, Y. (2003). A New Coincident Index of Business Cycles
Based on Monthly and Quarterly Series. Journal of Applied Econometrics, 18,
427–443.
Narayan, S. W. (2019). Does Fintech Matter for Indonesia’s Economic Growth?
Bulletin of Monetary Economics and Banking, 22, 437–456.
Narayan, P. K. (2021). COVID-19 Research Outcomes: An Agenda for Future
Research. Economic Analysis and Policy, 71. doi: 10.1016/j.eap.2021.06.006
Stock, J. H., & Watson, M. W. (1989). New Indexes of Coincident and Leading
Economic Indicators. NBER Macroeconomics Annual, 4, 351–409.
Nowcasting Regional Economic Growth in Indonesia 321

Tarsidin, Idham, & Rakhman, R. N. (2018). Nowcasting Household Consumption

and Investment in Indonesia. Bulletin of Monetary Economics and Banking, 20,
375–404.
Woloszko, N. (2020). Tracking Activity in Real Time with Google Trends. OECD
Economics Departmen Working Papers, 1634.
322 Bulletin of Monetary Economics and Banking, Volume 25, Number 3, 2022

This page is intentionally left blank