You are on page 1of 31

ed

iew
Forecasting Potential Impact of COVID-19 on India and Indonesia’s

ev
GDP Using ARIMA Model

r
er
pe
Pragya Jindal (pragyajindal@gmail.com)

and
ot

Shubham Modi (shubmodi2000@gmail.com)


tn

July 12, 2021


rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
1

ed
Abstract:

iew
In this paper, the Box-Jenkins approach has been used to build the appropriate

Autoregressive-Integrated Moving-Average (ARIMA) model to understand the short-term

economic impact of coronavirus on India and Indonesia’s economies. The quarterly data for

forecasting India and Indonesia's GDP was gathered from the Federal Reserve Bank of St. Louis

ev
to forecast the GDP from 2021-2025. As a broad measure of overall domestic production, GDP

functions as a comprehensive scorecard of a given country’s economic health. We find that the

r
appropriate statistical model for India’s GDP is ARMA (1, 1) and for Indonesia’s GDP is ARMA

er
(2,1). Forecasts show that India and Indonesia will recover completely from the setback in 2025

q4 and 2023 q3 respectively.


pe
.
ot

Keywords:
tn

Box-Jenkins approach, India, Indonesia, Forecasting, Gross domestic product, ARIMA,

Coronavirus, Economy, Unit Root, AR and MA, Pandemic, Time Series, Autocorrelation, ACF
rin

and PACF
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
2

1. Introduction

ed
The recent outbreak of the novel coronavirus disease, now recognized as a global pandemic, is

caused by droplets of saliva or discharge from the nose when an infected person coughs or

iew
sneezes. Most people infected will have mild respiratory illnesses, whereas older people or

people with health issues will develop serious illnesses. The healthcare authorities have

recommended measures such as social distancing and wearing a mask to prevent the spread of

ev
this disease. However, as of today, 140 million people worldwide have been a victim of this

virus. The pandemic has resulted in significant global economic disruption, the largest recession

r
since the Great Depression, threatening both short-term and long-term global growth sternly. The

er
pandemic is disrupting global supply chains and international trade resulting in an economic

slowdown. With countries closing their national borders, the flow of goods and services and the
pe
movement of people has stopped completely.

This study is an attempt to forecast the Gross Domestic Product (GDP) of India and Indonesia
ot

post the COVID-19 outbreak using quarterly time series data based on Box-Jenkins methodology

known as ARIMA (Auto-Regressive-Integrated-Moving-Average) method. The method is based


tn

on the “World Representation Theorem”, which states that every stationary time-series has an

infinite moving average (MA) representation (Jovanovic & Petrovska, 2010). This means its
rin

evolution can be expressed as a function of its past developments”.

This paper aims to understand the short-term economic impact of coronavirus on India and
ep

Indonesia’s economies. We focus on the effect on GDP as GDP is an excellent indicator for

assessing economic development. It is the total of all goods and services produced in the

economy and includes personal consumption, government expenditure, and foreign trade
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
3

balance. It is an important index to assess economic development and how well the economy is

ed
performing. (Abonazel & Abd-Elftah, 2019)

Since COVID has resulted in widespread deaths and has disrupted the business environment,

iew
there has been a huge economic impact on the economy. On the demand side-impact, tourism,

hospitality, and aviation are the worst affected sectors. Whereas, on the supply side,

manufacturing sectors such as automobiles, electronics are facing a shortage of raw materials.

ev
This is because the outbreak has adversely affected the trade of countries with China and the rest

of the world. A survey conducted by the Confederation of Indian Industries (CII) on 200 CEOs

r
indicated that the revenue may fall by more than 10 percent on most Indian firms by Q2 of 2020

er
suggesting that the majority of the firms expects a significant decline in profit by more than 5

percent by the last quarter of 2020. In India, the agricultural sector has been disrupted because of
pe
the non-availability of migrant laborers. Such impacts on the economy call for forecasting the

growth of economies.

According to the United Nations, the global economy could shrink by up to 1 percent in 2021
ot

due to the coronavirus pandemic, a reversal from the previous forecast of 2.5 percent growth.
tn

Since no one would have forecasted the pandemic and the sluggish economic growth with it, the

previous forecasts of economic growth are irrelevant now. There is a need for new forecasts so

that accurate projections can be made of the economy, taking the coronavirus into account. The
rin

severity of the economic impact will largely depend on the actual size and effectiveness of fiscal

responses to the crisis. (Jamir,2020) The fiscal responses can be optimal only when the
ep

authorities have adequate information about what the future holds. For instance, if the economy

is expected to recover, then inflation may increase and hence banks will increase their interest
Pr

rates, which will have a direct effect on businesses producing goods and services. (Pettinger)

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
4

Private sector companies also use forecasts as a guide to plan their economic activities. Forecasts

ed
allow them to make important decisions related to hiring, spending, and investment that impact

aggregate economic activity. Hence, this study will be a solid foundation for the government and

iew
other policymakers to take into consideration while formulating policies for kick-starting the

economy.

Our motivation to forecast the economic growth of India and Indonesia stems from the fact that

ev
both economies are very similar. Indonesia has a trillion-dollar economy, which is comparatively

similar to that of India. Both countries have large public sectors and have opened up to more

r
foreign investment. The population of both countries is also large with a similar population

er
growth rate (1.2 percent). The available literature usually compares India and China in terms of

the economy, however recent trends suggest that Indonesia is a better fit than China. Even
pe
though India and China are geographically large and populous, they have been on different

economic growth trajectories, such that the Chinese economy is now five times India’s. (Ninan,

2018) Comparing the impact on India and Indonesia will allow us to understand the main areas
ot

of disruptions and how different countries are impacted differently. It will also act as a guide for
tn

authorities to understand the measures that need to be taken to tackle the issue at hand.
rin

The paper is organized as follows. Section 2 discusses a review of the literature, while section 3

presents the data source. In section 4, we discuss the statistical background of the ARIMA
ep

model, which is followed by the establishment of our ARIMA model based on the Box-Jenkins

approach in Section 5. Section 6 offers the summary and the concluding remarks for our study.
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
5

2. Literature Review

ed
While generous exploration on the effect of a pandemic on the worldwide economy is accessible,

writing on investigating the effects of the COVID-19 on GDP is extremely scant. Besides, this is

iew
one of the principal papers that dissects the differential effect of COVID on India and Indonesia's

GDP. This is a chance to create a significant contribution in this particular field. The literature

review is divided into two sections- 1. Covid-19 and Gross Domestic Product, and 2. Forecasting

ev
GDP using the ARIMA model.

2.1 COVID-19 and Gross Domestic Product

r
The evolution of COVID and its economic impact is complex and uncertain which makes it
er
difficult for appropriate decision-makers to formulate macroeconomic policy response.
pe
(McKibbin & Fernando, 2021) conduct empirical research and conclude that even the global

collective containment measures would significantly impact the global economic condition

negatively. Further, (Fezzi & Fanghella, 2020) estimate the short-run impact of COVID-19 on
ot

the economy by using the electricity market data. The results estimate that the three weeks of the

most severe lockdown in Italy reduced the Italian GDP by roughly 30%. The negative impact
tn

was weaker during the later time but at the end of June 2020, the Italian GDP was about 8.5%

lower than it would have been without the outbreak.


rin

2.2 Forecasting GDP using ARIMA Model

The ARIMA model is the most common model used by researchers to forecast GDP. The paper
ep

‘Forecasting Egyptian GDP Using ARIMA Models’ followed the Box-Jenkins approach to build

the appropriate ARIMA model to forecast the GDP of Egypt for 10 years. (Abonazel &
Pr

Abd-Elftah, 2019) expected the GDP to rise because of the various advancements in the

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
6

economy, technological advancements in particular. Ning et al., (2010) used Shaanxi GDP from

ed
1952-2007 to forecast six years GDP (2008-13) using time series data and established ARIMA

(1,2,1) model, while Nyoni (2018) used the ARIMA (1,1,1) model approach to forecasting net

iew
FDI flow of Zimbabwe for two decades using annual time series data from 1980-2017. Zhang

(2013) analyzed yearly information from 1993-2009 to estimate the GDP of Sweden using

ARIMA, VAR, and AR(1) models. He concluded that even though all models can be fitted to

ev
forecast short-run GDP, the ARIMA model was best for forecasting per capita GDP.

3. Data

r
The data for forecasting India and Indonesia's GDP was gathered from the Federal Reserve Bank

er
of St. Louis. The time-series data of quarterly current price Gross Domestic Product for twenty

years was gathered from the first quarter (q1) of 2000 to the last quarter (q4) of 2020 to
pe
anticipate short-term GDP up to the final quarter (q4) of 2025. The data was first adjusted for

any discrepancies and was then converted into US Dollar, $, to have a uniform denomination.
ot

4. Time Series and Econometric Model


tn

Auto-Regressive Integrated Moving Average (ARIMA) is a class of models that explains a given

time series based on its past values: its lags, and the lagged forecast errors equation can be used
rin

to forecast future values. ARMA is composed of two distinct models- autoregressive (AR)

models and moving-average (MA) models.


ep

A pure Auto-Regressive (AR only) model is one where Yt depends only on its lags, that is, yt is

a function of the ‘lags of yt ’:


Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
7

yt = a0 + a1yt-1 + a 2yt-2 + … + apyt-p, where yt-1 is the lag1 of the series, a1 is the coefficient of lag1

ed
that the model estimates and a0 is the intercept term.

On the other hand, a pure Moving Average (MA only) model is one where yt depends only on the

iew
lagged forecast errors:

yt = a0 + εt + β1εt-1 + β 2εt-2 + … + β qεt-q ,where the error terms are the errors of the autoregressive

ev
models of the respective lags.

r
An ARIMA model is one where the time series was differenced at least once to make it

er
stationary and the AR and the MA terms were combined. So the equation becomes:
pe
yt = a0 + a1yt-1 + a2yt-2 + … + apyt-p + εt + β1εt-1 + β 2εt-2 + … + β qεt-q

An ARIMA model is characterized by three components- the order of the AR component (p), the

order of MA (q), and the minimum number of differencing required to make the series stationary
ot

(d). In this paper, we chose the ARIMA model to forecast GDP because ARIMA models
tn

frequently outperform more sophisticated structural models in terms of short-run forecasting

ability. (Stockton and Glassman (1987) and Litterman (1986)). This is because it can provide a

short-run forecast for sufficiently large amounts of data on the concerned variables very
rin

precisely. Moreover, the ARIMA model helps find the best fit of a time series model to past

values of a time series. Therefore, the ARIMA forecasting technique outlined in this paper will
ep

not only provide a benchmark by which other forecasting techniques may be appraised, but will

also provide input into forecasting in its own right. Besides, ARIMA is a parametric method and
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
8

it should work better for relatively short series when the number of observations is not sufficient

ed
to apply more flexible methods.

iew
Choosing Auto Regressive Moving Average (ARMA) models for forecasting the impact of

COVID on GDP requires an understanding of Autocorrelation Function (ACF), and Partial

Autocorrelation Function (PACF) since they are necessary to determine the order of AR and/ or

ev
MA terms. Though ACF and PACF do not directly dictate the order of the ARMA model, the

plots facilitate understanding the order and provide an idea of which model can be a good fit for

r
the time-series data. The ACF plot is a bar chart of coefficients of correlation between a time

er
series and its lagged values. It explains how the present value of a given time series is correlated

with the past values. On the other hand, PACF explains the partial correlation between the series
pe
and the lags of itself. This function plays an important role in data analysis aimed at identifying

the extent of the lag in the model.


ot

The transformed Autocorrelation Function (ACF) and Partial Autocorrelation Function (PACF)

helps in identifying the parameters of p and q of ARIMA models. The ACF and PACF plots are
tn

considered together to obtain the most accurate model. We expect the ACF plot to gradually

decrease for the AR process and simultaneously the PACF to have a sharp drop after p
rin

significant lags.
ep

5. Results and Interpretations

The ARIMA approach to forecasting the GDP consists of four steps- 1. Identification of the
Pr

model, 2. Estimation of the model, 3. Diagnostic Checking of the model and 4. Forecasting.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
9

5.1 Identification of the model

ed
Before identifying the best model, it is imperative to ensure that the series is stationary. A

iew
stationary time series is one whose properties do not depend on the time at which the series is

observed. Thus, time series with trends, or with seasonality, are not stationary and will affect the

value of the time series at different times (Rob J Hyndman & George Athanasopoulos). A series

ev
needs to be stationary because it makes the analysis easier to understand and detect and model

necessary tools and procedures in time series analysis (Palachy, 2019). We plotted the GDP of

r
both India and Indonesia over time to check for stationarity (see Figures 1 and 2). We see that

there is an upward trend for both countries and hence the series are not stationary. We also

er
confirmed this by plotting the ACF. The values of ACF tend to degrade to zero quickly for
pe
stationary time series, while for non-stationary data the degradation will happen more slowly. In

Figures 3 and 4, we can see that the degradation has happened very slowly, thus indicating that

our series is non-stationary.


ot

Figures 1 and 2: GDP at Level from 2000-2010


tn
rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
10

Figures 3 and 4: ACF of GDP for India and Indonesia

ed
iew
r ev
er
We also conducted the Dickey-Fuller test to confirm our analysis. The test evaluates whether a

time series variable is stationary or non-stationary and whether it possesses a unit root. In the
pe
Dickey-Fuller Test, the null hypothesis is that the series possesses a unit root and is not

stationary, and the alternative hypothesis is that the series is stationary. The p-value for India is

0.2870, whereas for Indonesia it is 0.2859 (see Table 1 and Table 2). Since the p-values for both
ot

are more than 0.05, we can not reject the null hypothesis and therefore conclude that the series is
tn

non-stationary. As a result, we transform the series by taking the first difference of the natural

logarithms of the values in the series to attain stationary in the first moment. Transformations

such as logarithms help to stabilize the variance of a time series, whereas differencing helps
rin

stabilize the mean of a time series by removing changes in the level of a time series and therefore

eliminating (or reducing) trend and seasonality.


ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
11

ed
Table 1: Dickey-Fuller Test for Unit Root- India

iew
ev
Table 2: Dickey-Fuller Test for Unit Root- Indonesia

r
er
pe
ot
tn

The equation representing the transformation is given by:

Xt= lnGDPt - lnGDPt-1, where lnGDPt= loge(GDPt)


rin

After the transformation, we can see from Figures 5 and 6 that there is no trend. In the

Dickey-Fuller test (see Table 3 and 4), the p-values for both India and Indonesia are 0, which is
ep

less than 0.05. Hence, we now reject the null hypothesis and conclude that the transformed series

is stationary.
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
12

Figures 5 and 6: GDP at First Difference

ed
iew
r ev
Table 3: Dickey-Fuller Test for Unit Root at First Difference- India

er
pe
ot

Table 4: Dickey-Fuller Test for Unit Root at First Difference- Indonesia


tn
rin
ep

If the autocorrelations are positive for many lags (10 or more), then the series needs further

differencing. In our case, the first difference was enough to satisfy the condition. Since we take a

transformed equation into consideration, which is already differenced and does not need any
Pr

more differentiation, the value of (d) is 0.

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
13

ACF and PACF functions are essential tools required for the identification of the model. The lags

ed
at which the PACF cuts off is the indicated number of AR terms (p). In Figures 6 and 7, we

observe that for India the AR term is 3, whereas for Indonesia it is 1. We similarly look at the

iew
ACF function for the MA terms (q). The MA component for India is 3, whereas it is 2 for

Indonesia.

ev
Figure 6 and 7: PACF of India and Indonesia

r
er
pe
ot
tn

Figure 8 and 9: ACF of India and Indonesia


rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
14

In Figures 1 and 2, we observe a break (known as structural breaks) in the trend. A structural

ed
break is an unexpected change over time in the parameters of regression models. We understand

and tackle the issue of the structural break because an ARIMA model controlled for a break is

iew
better than an ARIMA model that is not controlled for a break(s) to forecast short-term GDP.

This is because structural breaks can lead to huge forecasting errors and invalid conclusions. A

fixed-parameter model cannot be expected to forecast well if the true parameters of the model

ev
change over time. In their 2011 paper, Pettenuzzo and Timmermann show that including

structural breaks in asset allocation models can improve long-horizon forecasts and that ignoring

r
breaks can lead to large welfare losses. Inoue and Rossi (2011) show the importance of

er
identifying parameter instabilities for improving the performance of the models. We recognize

the breaks in our model and control them by generating a Fourier series. A Fourier series
pe
represents a periodic function of sine and cosine functions that control for smooth breaks. It

automatically captures the structural break by nonlinearity that will reduce in the model. We

tested the Fourier series with multiple frequencies (up to 3), however, the coefficients were
ot

significant only for frequency one.


tn

5.2 Estimation of the model


rin

We consider multiple tentative models based on the AR and MA components found to find the

best model. The statistical significance of the lags will define the decision for the combination of
ep

the ARIMA model. We select the most parsimonious model as Box-Jenkins methodology

emphasized that a parsimonious model gives a better forecast than an over-parameterized model

by picking the model that gives the smallest number of parameters to be estimated. (Jamir,2020)
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
15

For India, we consider six different models (see Table 5)- AR(2) , MA (3), AR (1), ARMA (1,1),

ed
ARMA (2,1), ARMA (1,2). For Indonesia, we consider the following 5 models- AR (1), ARMA

(1,1), ARMA (1,2), ARMA (1,3), ARMA (2,1).

iew
Table 5: Results of Tentative Model Estimation for India (See Appendix A1-A6 for Results)

ev
Model AR(2) AR(1) MA(3) ARMA ARMA ARMA

(1,1) (1,2) (2,1)

r
Coefficient Insignificant Insignificant Significant Significant Significant Significant

AIC -285.420 -269.298 er


-292.149 -293.389 -292.149 -291.969
pe
BIC -270.980 -257.264 -277.709 -281.355 -277.709 -277.529
ot

Based on the tentative model estimation result, four models, MA(3), ARMA(1,1), ARMA(1,2),

and ARMA(2,1), are significant at a 5% level of significance. Out of them, ARMA (1,1) has the
tn

lowest AIC and BIC. After that, two models, MA(3) and ARMA (1,2) have the second-lowest

AIC and BIC values. We thus carried out the Debold Mariano test (see Appendix A7) to find the
rin

best model between MA (3) and ARIMA (1,2) and then performed the same test to compare it to

ARMA (1,1). From the results, we found that the ARMA (1,2) model is a better forecast than
ep

MA (3). We then compared ARMA (1,2) with ARMA (1,1) (see Table 6) and found that ARMA

(1,1) is the most parsimonious model for forecasting.


Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
16

Table 6: Diebold Mariano Test for ARMA (1,2) and ARMA (1,1)- India

ed
iew
r ev
(Remark- dlyhat3 is ARMA (1,1) and dlyhat4 is ARMA (1,2))
er
pe
Similarly, we performed the Diebold Mariano Test for Indonesia (see Table 7) with five different

models- AR (1), ARMA (1,1), ARMA (1,2), ARMA (1,3), ARMA (2,1). We compared the two

models with the lowest AIC and BIC (AR (1) and ARMA (2,1)) (see Table 8). We found that
ot

ARMA (2,1) is the better forecasting model.


tn

Table 7: Results of Tentative Model Estimation for Indonesia (See Appendix A8-A12 for Results)
rin

Model AR(1) ARMA (1,1) ARMA (1,2) ARMA (1,3) ARMA (2,1)

Coefficient Insignificant Insignificant Significant Significant Significant


ep

AIC -403.672 -402.909 -405.658 -405.523 -406.273


Pr

BIC -391.639 -388.469 -391.218 -388.676 -391.833

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
17

Table 8: Diebold Mariano Test for AR (1) and ARMA (2,1) - Indonesia

ed
iew
r ev
er
(Remark-dlyhat1 is ARMA (2,1) and dlyhat2 is AR(1))
pe
5.3 Diagnostic Checking of the model

The best models were further taken to perform the Ljung-Box Test of Squared Residuals (see
ot

Table 9 and 10) to check for autocorrelation. The results show that all the probability values from
tn

lag 1 to lag 20 are higher than 0.05 for both India and Indonesia, which indicates that the

residuals are white noise and there is no autocorrelation in the model. Appendix A13 and A14

show that all the lag structures of ACF fall within the 95% confidence interval, indicating that all
rin

information has been captured. This strengthens our arguments for the selected model and thus

we can go ahead with forecasting.


ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
18

Table 9: Ljung-Box Test for ARMA (1,1)- India

ed
iew
r ev
er
pe
Table 10: Ljung-Box Test for ARMA (2,1)- Indonesia
ot
tn
rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
19

ed
5.4 Forecasting Results

Since ARMA (1,1) and ARMA (2,1) are fit to forecast the GDP data for India and Indonesia

iew
respectively, we forecast the Gross Domestic Product values for the next five years (from 2021q1

to q4 2025q4) to understand the short term impact of coronavirus on the economy.

ev
In India, the quarterly GDP grew by roughly 2.5% from 2010-2018 (see figure 10). 2019

onwards, there was a drop in quarterly GDP to about 1.8% because of various internal and

r
external reasons such as global slowdown, demonetization, and poor GST implementation. We

er
forecast that the GDP will fall to a quarterly growth rate of about 1.3% from 2021 because of the

economic slowdown due to COVID. We predict that from 2023 q3 onwards, the economy will
pe
kickstart and grow at an increasing rate. In 2025 q4, the economy will return to the quarterly

growth rate of 1.8%, which was observed before coronavirus.


ot

Figure 10: Forecasted GDP for India from 2021q1 to 2025q4


tn
rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
20

ed
In Indonesia, the quarterly GDP grew by roughly 2.3% from 2010-2018 (see figure 11). 2019

onwards, there was a drop in quarterly GDP to about 1.8% because of the global slowdown. We

foresee a negative growth rate (-2%) in the first half of 2020 because of coronavirus. We forecast

iew
that the GDP will gradually increase to a quarterly growth rate of about 1.4% from 2021.

Further, we predict that by 2023 q3, the economy will recover completely from the economic

shock. Post that, the economy will grow more than it has ever had (2.6% growth from 2025 q3 to

ev
2025 q4).

Figure 11: Forecasted GDP for Indonesia from 2021q1 to 2025q4

r
er
pe
ot
tn
rin

In conclusion, we observe that COVID has a short-term effect on the economy of both India and

Indonesia. Indonesia recovers faster than India because Indonesia will be able to better control

COVID than India. The situation in India will be worse because India has a higher population
ep

density and poor health infrastructure. Since the disease is contagious, it will be difficult for

India to contain the virus. In addition, India lacks the appropriate healthcare facilities to cure the
Pr

country’s large population. Since the population in Indonesia is roughly 20% of India’s

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
21

population, the situation will be much better there. Further, Indonesia scores more than India in

ed
terms of growth, development, and infrastructure, which are important indicators in analyzing

how an economy can handle economic shocks.

iew
It is important to note that these are only predicted values, but the economy is a dynamic,

uncertain and complex system. The government’s response to economic aid packages and fiscal

policy will play a big role in determining the country’s GDP over the next few years. We should

ev
focus on the risk of adjustment in the economic activities and prevent the economy from severe

fluctuations and adjust the corresponding target value according to the actual situation. see

r
Wabomba et al (2016).

6. Conclusion er
pe
The study aimed to model and forecast India and Indonesia’s GDP based on the Box-Jenkins

approach based on the quarterly data (from 2000 q1 to 2020 q4). The four stages of the
ot

Box-Jenkins approach were conducted to obtain an appropriate ARIMA model for the GDP. We

used the models to forecast the GDP for the next five years (from 2021 q1 to 2025 q4). The
tn

forecasting helped us in understanding the differential short-term impact of coronavirus on the

economy of both countries. From the results, we see that Indonesia is better equipped to tackle
rin

the impact of the disease. The results in this paper will provide a guide to governments and the

appropriate authorities to implement the necessary measures to minimize the impact of

coronavirus on the economy. We believe that the two economies will leave this crisis stronger.
ep

This is because of various reasons. Firstly, lockdowns have channeled many people online for

learning and working, accelerating the shift to digital services. This will promote technological
Pr

innovation and lead to more advanced economies in the future. Secondly, there will be changes

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
22

to the country’s healthcare system as countries will prepare for possible future pandemics.

ed
Further, companies and governments will seek to address supply-chain flaws revealed by the

crisis, which will drastically improve the business environment in the future.

iew
r ev
er
pe
ot
tn
rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
23

References

ed
1. Global economy could shrink by almost 1% in 2020 due To Covid-19 pandemic: United

Nations. (n.d.). Retrieved May 01, 2021, from

iew
https://economictimes.indiatimes.com/news/international/business/global-economy-could

-shrink-by-almost-1-in-2020-due-to-covid-19-pandemic-united-nations/articleshow/7494

3235.cms?utm_source=contentofinterest&utm_medium=text&utm_campaign=cppst

ev
2. Abonazel, M. R., & Abd-Elftah, A. I. (2019). Forecasting Egyptian GDP using ARIMA

models. Reports on Economics and Finance, 5(1), 35-47. doi:10.12988/ref.2019.81023

r
3. Pettinger, T., & Musgrave, R. (2020, February 04). What is the importance of economic

forecasts? Retrieved May 01, 2021, from


er
https://www.economicshelp.org/blog/1976/economics/importance-of-economic-forecasts/
pe
#:~:text=The%20main%20importance%20of%20economic,help%20policymakers%20m

ake%20better%20decisions.&text=Forecasts%20are%20important%20because%20polic

y,to%20have%20the%20full%20effect.
ot

4. Jamir, I. (2020). Forecasting potential impact of COVID-19 outbreak on India's GDP


tn

Using ARIMA Model. SSRN Electronic Journal. doi:10.2139/ssrn.3613724

5. Jovanovic, B., & Petrovska, M. (2010). Forecasting Macedonian GDP: Evaluation of


rin

different models for short-term forecasting PDF Logo.

6. Ninan, T., Gupta, S., Chandna, H., & Bhardwaj, A. (2018, May 12). When comparing

economies, PAIR India with Indonesia, not China. Retrieved May 01, 2021, from
ep

https://theprint.in/opinion/when-comparing-economies-pair-india-with-indonesia-not-chi

na/57650/
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
24

7. McKibbin, W., & Fernando, R. (2021). The global macroeconomic impacts of

ed
COVID-19: Seven Scenarios. Asian Economic Papers, 1-30. doi:10.1162/asep_a_00796

8. Fezzi, C., & Fanghella, V. (2020). Real-time estimation of the short-run impact of

iew
covid-19 on economic activity using electricity market data. Environmental and Resource

Economics, 76(4), 885-900. doi:10.1007/s10640-020-00467-4

9. Stockton, D. J., & Glassman, J. E. (1987). An evaluation of the forecast performance of

ev
alternative models of inflation. The Review of Economics and Statistics, 69(1), 108.

doi:10.2307/1937907

r
10. Litterman, R. B. (1986). Forecasting with Bayesian Vector autoregressions: Five years of

er
experience. Journal of Business & Economic Statistics, 4(1), 25. doi:10.2307/1391384

11. Forecasting: Principles and Practice (2nd ed). (n.d.). Retrieved May 01, 2021, from
pe
https://otexts.com/fpp2/stationarity.html

12. Palachy, S. (2019, September 22). Stationarity in time series analysis. Retrieved May 01,

2021, from
ot

https://towardsdatascience.com/stationarity-in-time-series-analysis-90c94f27322

13. Masum, M. (2020, August 13). Identifying AR and ma terms Using ACF AND PACF
tn

plots in time SERIES Forecasting. Retrieved May 01, 2021, from

https://towardsdatascience.com/identifying-ar-and-ma-terms-using-acf-and-pacf-plots-in-
rin

time-series-forecasting-ccb9fd073db8#:~:text=The%20ACF%20and%20PACF%20plots

%20indicate%20that%20an%20MA%20
ep

14. Inoue, A., & Rossi, B. (2011). Identifying the sources of instabilities in macroeconomic

fluctuations. Review of Economics and Statistics, 93(4), 1186-1204.


Pr

doi:10.1162/rest_a_00130

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
25

15. Sammy Wabomba, M. (2016). Modeling and Forecasting Kenyan GDP USING

ed
Autoregressive Integrated moving average (ARIMA) Models. Science Journal of Applied

Mathematics and Statistics, 4(2), 64. doi:10.11648/j.sjams.20160402.18

iew
r ev
er
pe
ot
tn
rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
26

Appendix

ed
A1: AIC and BIC of AR (2)- India

iew
r ev
er
A2: AIC and BIC of ARMA (1,1)- India
pe
ot
tn

A3: AIC and BIC of ARMA (2,1)- India


rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
27

A4: AIC and BIC of ARMA (1,2)- India

ed
iew
ev
A5: AIC and BIC of MA (3)- India

r
er
pe

A6: AIC and BIC of AR (1)- India


ot
tn
rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
28

A7: Diebold Mariano Test for MA (3) and ARMA (1,2)

ed
iew
r ev
(Remark- dlyhat 1 is ARMA (1,2) and dlyhat 2 is MA (3))

er
A8: AIC and BIC of ARMA (1,1)- Indonesia
pe
ot
tn

A9: AIC and BIC of ARMA (2,1)- Indonesia


rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
29

A10: AIC and BIC of ARMA (1,2)- Indonesia

ed
iew
ev
A11: AIC and BIC of AR (1)- Indonesia

r
er
pe

A12: AIC and BIC of ARMA (1,3)- Indonesia


ot
tn
rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823
30

A13: ACF of residuals for ARMA (1,1)- India

ed
iew
r ev
er
A14: ACF of residuals for ARMA (2,1)- Indonesia
pe
ot
tn
rin
ep
Pr

This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=3884823

You might also like