You are on page 1of 20

J. Inst. Eng. India Ser.

B (October 2022) 103(5):1459–1478


https://doi.org/10.1007/s40031-022-00762-2

ORIGINAL CONTRIBUTION

Deep Learning‑Based Trend Analysis on Indian Stock Market


in COVID‑19 Pandemic Scenario and Forecasting Future
Financial Drift
Janmenjoy Nayak1 · Pandit Byomakesha Dash2 ·
Bighnaraj Naik3 · Subhashree Mohapatra4 ·
A. R. Routray5

Received: 30 August 2021 / Accepted: 24 May 2022 / Published online: 4 July 2022
© The Institution of Engineers (India) 2022

Abstract This present study has used the long-short-term has been studied. Here, the analysis is based on the impact of
memory (LSTM) network-based deep learning architecture COVID-19 including the effect of lockdown. (3) A predic-
to analyze the influence of the current widespread COVID- tion model has been proposed for the study of the behavior
19 on the Indian stock market. The major contribution of of the Indian stock index (Nifty 50) during the COVID-19
this work is as follows: (1) Designing LSTM-based deep pandemic. (4) Comparison of the efficacy of the suggested
neural network is used to study the effect of the COVID- approach with other existing baseline regression models.
19 outbreak and Lockdown on the Indian stock exchange
(Nifty 50), and (2) designing a prediction model to capture Keywords COVID-19 · Deep learning · LSTM ·
the effect of various COVID-19 waves in India on Indian Pandemic · Stock market
Stock exchange. The outcomes of the analysis show that the
increase in daily new confirmed cases, recovered cases, and
death cases have a significant adverse impact on the trend of Introduction
the stock market. Moreover, the results of the work have also
analyzed the impact of government policy such as ‘lockdown In the last week of December 2019, many residents in the
city’ with a reaction to increased Pandemic cases. This work Hubei Province of Wuhan city were infected by unidentified
is briefly summarized as follow: (1) LSTM-based deep neu- pneumonia [1]. Within the first few weeks of the epidemics,
ral network is used for this study to analyze the effect of the thousands of individuals have been infected and more than
COVID-19 outbreak on the Indian stock exchange. (2) The hundreds of people have been killed. After the analysis of
Indian Stock exchange affected by the COVID-19 pandemic throat swabs, the CDC researchers named the novel virus
NCP in January [2]. As its genome structure is similar to
SARS-CoV, the novel coronavirus is named SARS-CoV-2
* Bighnaraj Naik by the ICTV [3]. The disease caused by SARS-CoV-2 was
bighnaraj_naik@ieee.org
named COVID-19 by the WHO in February 2020 [4]. In a
1
Department of Computer Science, Maharaja Sriram short period, the novel coronavirus spread quickly to other
Chandra BhanjaDeo University, Baripada, Mayurbhanj,
regions of the world and has changed the life of many people
Odisha 757003, India
2
around the world.
Department of Information Technology, Aditya Institute
The major issue in the detection of the COVID-
of Technology and Management (AITAM), Tekkali 532201,
India 19 virus is the diagnosis of the disease. Although the
3 RT-PCR method was used as the standard approach for
Department of Computer Application, Veer Surendra Sai
University of Technology, Burla, Odisha 768018, India identifying the COVID-19 infection, they require more
4 time, presents low sensitivity, and limited stock because
Department of Computer Science & Engineering,
Institute of Technical Education and Research, Siksha ‘O’ of the pandemic. The drawbacks of RT-PCR may be
Anusandhan University, Bhubaneswar 751030, India overwhelmed by the usage of medical radiographs like
5
Department of Information & Communication Technology, X-rays, scanners, and CT. The ML and DL methods assist
Fakir Mohan University, Balasore, Odisha, India the medical practitioners in the precise calculation and

13
Vol.:(0123456789)
1460 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

speedily recognition of pneumonia [5–8]. Therefore,


the present world health situation is also demanding the
utilization of these techniques for the identification of
COVID-19 infection efficiently. Despite the expensive
computational simulations on basic physical models and
limited understanding of the underlying mechanism, the
ML approaches are capable of developing predictive
models which are considered the advantage of the ML
approach [9–12]. The other advantage is the extraction
of obscure patterns from vast, noisy, and composite data-
sets through the execution of distinct statistical, proba-
Fig. 1  India’s growth trajectory since 2011
bilistic, and optimization techniques. Because of these
advantages, ML techniques can be expeditiously utilized
in the various applications of health management such bacterial and viral pneumonia from thoracic imaging data
as prognosis of disease, manufacturing, and discovery [24]. Further, several attempts have been made to identify
of drugs, diagnosis of medical radiographs, research and several chest CT images [25]. As an important compara-
clinical trials, maintenance of smart health records, col- tive analysis, a comparison among seven distinct notable
lection of crowdsourced data, prediction of the outbreak, deep learning neural networks architectures was given by
better radiotherapy, etc. [13–16]. An ML approach such as Hemdan et al. [26]. They have used the dataset consist-
the FFT-Gabor scheme has been proposed by Al-Karawi ing of 50 images in the experiments consisting of equal
et al. [17] to investigate COVID-19 infection from Chest halves of both normal and infected patients. Further, the
CT radiographs. Further, it is evident from the simula- ImageNet dataset was used to pretrain the models [27]
tion results that the developed model predicts results with and radiography was used to train the classifier. From the
an average accuracy of 95.37%. To predict the spread of results, it is concluded that the best performing architec-
COVID-19, an ARIMA prototype has been suggested by tures were the VGG19 and the DenseNET201. For dis-
[18]. For the next two days, the authors have forecasted tinguishing several images such as X-ray, normal, and
various parameters based on the study about the predomi- pneumonia, a novel architecture of convolutional neural
nance of COVID-19. Further, this paper also indicates the network named COVID-net was proposed by Wang et al.
ARIMA forecast graph for the epidemic occurrence and [28]. Compared with previous work, in their study authors
predominance. Though ML tools have been efficiently have used a much larger dataset comprising 16,756 chest
applied in the prophecy and forecasting of COVID-19, radiography images from over ten thousand patients. Fur-
there remain some drawbacks such as full appliance of ther, the empirical outcome depicts that the suggested
biomedical data, temporal dependency, failure in main- model outperforms with overall accuracy and sensitiv-
tenance of high-dimensionality, diversity, sparsity, incon- ity of 92.4%, and 80%, respectively, for COVID-19. Due
sistency, and new patterns that cannot be recognized with to numerous advantages like superior performance, and
supervised approaches. Due to the incremental architec- the use of automatic feature extraction without human
tural structure, the deep learning techniques gain the intervention, DL approaches are widely applied in the
ability to resolve most of the complex machine learning prophecy and forecasting of COVID-19 infection instead
tasks. The features such as dynamic updated learning, of ML approaches.
better performance, and potential in handling multi-modal In the second week of March 2020, due to a larger
and composite data and so on made the DL prototypes death rate of over 10,000 with huge affected cases across
the best precise techniques for supervising healthcare the globe, the WHO declared the epidemic a pandemic.
datasets such as categorization of brain abnormalities, Businesses throughout the globe are functioning in the
categorization of pathogenic bacteria, detection of dis- panic of a looming crumple of international financial
tinct types of carcinoma, and partitioning of biomedical markets as approximately 162 countries are firmly going
radiographs [17–23]. Formerly, several DL approaches into lockdown. This plight was hit hard with a stagnant
have been developed for determining and distinguishing economic rise in the last year, particularly in an advanced

13
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478 1461

Table 1  Few major steps taken by the Indian Government to tackle the pandemic and its impact on the economy
Date Measure taken

27th March 2020 374,000 crore has been announced by the Reserve Bank of India to the country’s financial system
1st April 2020 To handle the COVID-19 pandemic in India, $1 bn has been approved by World Bank approved
3rd April 2020 28,379 crore funds were released to states for tackling the COVID-19 pandemic
17th April 2020 50,000 crore special finance has been announced by RBI Governor to NABARD, SIDBI, and NHB
18th April 2020 To protect Indian companies during the pandemic, India’s foreign direct investment policy has
been changed by the government
17th May 2020 20 lakh-crore package has been announced by the government to manage this pandemic and
upsurge the economic growth

country like India as the economy was already on a down- researches available in the literature that shows the impact
ward trajectory since the turn of FY 2018–2019 as shown of epidemic/pandemic diseases on SMRs.
in Fig. 1. According to credit rating agencies and the The present study highlights the impact of the most
World Bank, India’s rise for FY 2021 has declined with recent pandemic on stock market indices, particularly the
the poor figures in the last three decades [29, 30]. consequences of the COVID-19 transmissible disease on
To combat COVID-19, the Indian Government has the Indian stock market. In this work, the authors have
announced a lockdown up to 14th April 2020 and then used LSTM as a deep learning tool to analyze the trend
extended up to 3rd May 2020. In the period of the first of the COVID-19 outbreak by predicting and forecasting
21 days of complete lockdown, the Indian economy is it. Moreover, the impact of this contagious disease on the
likely to lose over 32,000 crores every day [31, 32]. The Indian stock exchange trend is also studied. In this regard,
following are the different sectors affected by the lock- the following are major points that highlight the major
down such as 53% of the businesses [33], supply chains contributions of the work.
have been kept under stress as there was a lack of clar-
ity in streamlining, and a large number of farmers fac- a. LSTM-based deep neural network is used for this
ing uncertainty [34], daily wage groups are at risk [35], study to analyze the effect of the COVID-19 outbreak
live event industries have witnessed a loss of 3000 crores on the Indian stock exchange.
[36]. Within a short span of one month, unemployment b. The Indian Stock exchange affected by the COVID-
surged from 6.7% to a drastic of 26% (from March–April) 19 pandemic has been studied. Here, the analysis is
[37]. Major firms in India such as Tata Motors, UltraTech based on the impact of COVID-19 including the effect
Cement, Aditya Birla Group, Grasim Industries, Bharat of lockdown.
Forge, Larsen and Toubro, and Thermax have temporarily c. A prediction model has been proposed for the study
delayed or minimized their operations. of the behavior of the Indian stock index (Nifty 50)
To handle the plight of the pandemic for security and during the COVID-19 pandemic.
safety of food and additional funds for healthcare, the d. Comparison of the efficacy of the suggested approach
Government of India is released several measures as with other existing baseline regression models.
shown in Table 1.
SMRs of a particular country have a major impact
on their respective economic growth and vice versa. Related Work
The impact of SMRs due to several external sources has
already been studied in many articles, such as sport- The immediate actions by the Country/State governments,
activities [38], disasters [39], environmental [40], and manufacturing companies, media, and the public in a reac-
political events [41, 42]. In past, the SMRs have also been tion to the rise in COVID-19 cases, have created a demand
affected by the many epidemic infections as per the fol- and supply blows, and therefore the COVID-19 disease has
lowing studies: SARs outbreak [43, 44], and Ebola Virus been recognized as a most impactful cause than other similar
Disease (EVD) outbreak [45]. But, there are only a few events [46]. This economic crisis as a result of falling sup-
ply and demand is now affecting the employability status of

13
1462 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

the country, for example, seventeen million un-employabil- They presented a technique to assess the systematic threat
ity cases have been added to their un-employability status in the stock markets. The considered models were based
within three weeks of the crisis [47]. Therefore, the financial on the projected deaths of COVID-19 that are released on
markets or stock markets throughout the world have seen a 2nd April 2020 by Washington University. The authors
downgrade movement [48]. Some Stock market indices have have estimated recovery as well as the decline in the index
resisted the worst one-day fall in the last two decades. In this for the next 3 months of the study. Q-Gaussian distribu-
regard, predicting this uncertainty is challenging work for tions were considered, as the forecast is an anticipation
the researcher. However, several studies were already carried of a prediction. The authors have supposed that COVID-
out in the last couple of months which are presented in the 19 has an exogenous as well as a deterministic impact
subsequent sub-section. on the stock market and applied the projected technique
Al-Awadhi [49] has examined the influence of infec- to address the systematic risk with a Q-Gaussian distri-
tious diseases on the stock exchange and used the analysis bution. The consequences of COVID-19 on well-known
of panel regression in the experiment to determine the stock markets in main concerning countries such as Japan,
COVID-19 effect. From the experimentation, the authors Korea, Indonesia, Malaysia, the USA, Singapore, etc., are
have concluded that the findings signified that the raise discussed by Liu et.al [53]. The authors have used an
in both COVID-19 confirmed and death has major nega- event study technique and indicated that the main affected
tive consequences on stock markets. Ahmar and Val [50] areas, as well as countries of stock markets, have rap-
have introduced a method to predict the confirmed cases idly fallen after the spread of the pandemic. They have
of COVID-19 and IBEX that are obtained from the Spain confirmed that the result of regression also maintained
stock market. The authors have used the SutteARIMA and the difficult consequences of COVID-19 confirmed
ARIMA method. They have considered the data (COVID- cases on stock markets. Based on the investigations, the
19 detected cases in Spain) from 12th February 2020 to authors have concluded that there was a major negative
9th April 2020. To evaluate the prediction techniques, consequence of the COVID-19 pandemic on stock mar-
mean absolute percentage error and predicting accuracy kets. It was found to have more impact on COVID-19 in
measures are applied. From the experimentation, the Asian county stock markets. The researchers [54] have
authors have inferred that the SutteARIMA method is predicted by analyzing the previous scenarios of stock
more pertinent than the ARIMA method in computing markets along with the current scenarios. As the stock
IBEX and COVID-19 cases in Spain. It is found to have a market presents the investor’s view of expectations for
MAPE value of 0.1905 which is smaller than the ARIMA the upcoming days, the author has mentioned how the
and IBEX value of 0, 0202 for defined cases in Spain. spread of COVID-19 has kept the stock markets exactly
Pavlyshenko [51] has analyzed the impacts and regres- in difficult situations. The author has mentioned the lat-
sion approach of COVID-19 on the stock exchange. To est research that is reported by Ramelli and Wagner [55]
forecast the COVID-19 spread, the logistic curve along which included the behavior of investors in three phases.
with Bayesian regression was applied. The authors have The author has represented the effects of the COVID-19
compared the analysis of COVID-19 with regression to scenario and the risks to the investors. As per the author’s
the influence of other disasters. The authors have mainly point of view, wide actions with fiscal policy interfer-
considered the significance of finding a maximum num- ences have needed to evade further negative effects of
ber of COVID-19 cases per day for estimating half an the COVID-19 alarm. Kizys et. al. [56] have presented
instance of its spread. From the analysis, the authors have the primary study of investors’ assessments of the results
deduced that several disasters for different reasons have of companies due to the COVID-19 effect. The authors
shown their impacts on the stock markets. Due to this have focused on the results of company debt and cash
reason, Bayesian inference showed its significance to assets as well as international contact of US firms. They
diagnose the indecision of the impacts as independently. organize the initial study into three phases that include
The two forecast techniques of the S&P500 that relays incubation, outbreak, fever and described the actions fea-
on COVID-19 have been introduced by Karina et.al [52]. turing each phase. Majorly, it was found to focus on the

13
Table 2  Studies on previous literature
Author and year Contribution Pros Cons References

Fernandes Nuno (2020) Discussed about the impact of COVID-19 out- Presented a rough estimate of global economic The model does not take into account the direct [46]
break on the economic growth of industries and cost of COVID-19 outbreak in different sectors and indirect health cost and the probable spillo-
countries across the world and GDP growth of 30 different countries in vers to the financial sector
different scenarios during the outbreak
Al-Awadhi et al. (2020) Analyzed the impact of COVID-19 on the For analyzing the COVID-19 impact on stock Considered only daily growth in the confirmed [49]
outcome of the Hang Seng Index and Shanghai market, panel regression approach used which cases and deaths for analyzing the impact of
stock market minimizes estimation bias and multicollinear- COVID-19 on stock market returns
ity. Also used to determine the time-varying
relationship between dependent and independent
variables
Ahmar et al. (2020) Short-term forecasting of COVID-19 cases and The daily forecasting of COVID-19 confirmed Impact of lockdown restrictions on daily forecast- [50]
IBEX stock in Spain cases and IBEX stock in Spain has been done ing of COVID-19 cases and stock market has
effectively using Sutte ARIMA model when not been considered
compared to ARIMA model
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

Pavlyshenko (2020) Analyzed the impact of COVID-19 on stock Analyzed the uncertainty of crises impacts using An expert is required for setting up the informa- [51]
market and modelling the spread of COVID-19 Bayesian inferences tive prior distributions in Bayesian inference
outbreak in different countries
Karina et al. (2020) Explored the impact of COVID-19 outbreak on Assessed the stochastic and systematic risk in Systematic risk has not been estimated using [52]
S&P500 stock market in USA the stock market using two forecasting models enhanced tools
based on COVID-19 deaths and the 2-month-
sconsideration since the first confirmed case
recorded in USA
Liu et al. (2020) Assessed the effect of COVID-19 crisis on 21 Analyzed the immediate impact of COVID-19 Not considered the investor sentiment and fears of [53]
leading stock market indices in Japan, Korea, outbreak on the returns of stock market using uncertainties
Singapore, the USA, Germany, Italy, UK and OLS regression approach which considered Demographic variables have not been considered
so on panel data for 21 market indices in a 35-day in the analysis of the model
window
Khan et al. (2020) Explored the impact of COVID-19 outbreak on Based on efficient market theory and intertempo- Considered only weekly new cases of COVID-19 [57]
sixteen countries stock markets ral asset price theory, investigated the impact of pandemic
COVID-19 outbreak on daily and weekly stock
prices using pooled OLS, t-test and Mann–
Whitney test approaches
Lee et al. (2020) Investigated the effect of COVID-19 sentiment on In addition to twitter data, the proposed approach The proposed approach has been evaluated [58]
the US stock market using big data has used DNSI or Google Trends data to explore by considering data from the initial stage of
the initial impact of COVID-19 sentiment on COVID-19 pandemic
US stock market by industry
Kizys et al. (2021) Explored whether Government responses regard- The results have been analyzed by considering 72 Not considered rational or irrational herding [56]
ing COVID-19 crisis can alleviate herding countries daily stock market data. and govern- behaviour of Investor in international stock
behaviour of Investor in international stock ment response to COVID-19 are measured market
market using Oxford COVID-19 Government Response
Tracker

13
1463
1464 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

Fig. 2  LSTM network structure

primary impact of the COVID-19 pandemic on the stock Indian data for predicting the future stock market index. The
exchange. The authors have demonstrated how the crisis LSTM-based DL approach is used for designing this predict-
of COVID-19 has affected the stock market investors and ing model. The prediction performance is validated through
morphed into a broader financial as well as economic four major evaluation metrics such as MAPE, ­R2 score,
crisis. Onali [59] has made a first-ever investigation to RMSE, and MAE. Moreover, the outcome of the suggested
know the COVID-19 impact on the US stock exchange. approach has been compared with other baseline methods
The author has used the GARCH method with powerful (RR, LASSO, LR, Elastic net, etc.) to check the effective-
standard errors that permits changes in detailed cases as ness of the work. The comparative result indicates that the
well as deaths due to COVID-19. Based on the author’s LSTM-based approach outperforms others by a significant
analysis, it was suggested that the alterations in the num- margin.
ber of cases and fatalities in the US and other six highly Furthermore, the outcome of this study concludes that the
distressed countries by the COVID-19 disaster has no increasing rate of Pandemic cases and lockdown announce-
impact on the US stock exchange, apart from the amount ment has a significant impact on stock indices.
of accounted cases for China. Markov-Switching tech-
niques implied that the negative influence of the volatility
index (VIX) on the stock exchange has risen threefold at Methodology
the February 2020 end. The major advantages and limi-
tations of some prestudied models are given in Table 2. This proposed system has may be seen as a process of
The prevailing pandemic ‘COVID-19’ has not only designing an LSTM-based [60] model for modeling time
changed the lifestyle of individual countrymen but also has series prediction of Indian stock closing price in the pres-
shown a major role in transforming the decision policies ence/absence of COVID-19 spread in India. The proposed
of companies and governments. These instantaneous policy model makes use of (1) past COVID-19 spread time series
transformations have supplied the uncertainty in the stock consisting of ‘Confirm case’, ‘Recover case’ and ‘Death
exchange and enhanced the volatility. In this regard, this case’, and (2) past Indian stock closing price as input and
work analyzes the effect of rising COVID-19 cases (con- forecasts the future trends. Figure 4 represents the abstract
firmed, recovered, and death) and government lockdown view of the proposed approach, and the details of the pro-
policy on the trend of the Indian stock exchange. The data posed model have been presented in Algorithm 1 and Fig. 5.
of the Nifty 50 index under the national stock exchange
(NSE) have been chosen for the experiment. This work has Proposed LSTM‑Dropout Model for Trend Analysis
used the historical data of Nifty 50 in addition to COVID-19 of Indian Stock Market and impact of COVID‑19
Spread

The proposed prediction model (Algorithm 1) based on


LSTM-Dropout Model has three major computation steps:

13
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478 1465

i) Computation of Gate F t known as forget gate, S t known


as input modulation gate, I t known as input gate and Ot


known as output gate(Fig. 2), ii) Computation of gradient
using backpropagation through time (Fig. 3) (Backpropaga-
tion phase), iii) Computation of new network parameters
and Updating them, and iv) Computation and imposing of
dropout on F t , S t , I t and Ot is done to avoid overfitting.

To improve model performance and reduce overfitting,


the inputs and recurrent connections ton LSTM nodes are
excluded using a probabilistic method. This is a regulariza-
tion method, called Dropout, in which these many connec-
tions (Fig. 2) are excluded from the activation and weight
updation, thereby resulting in less complex networks with
less chance of overfitting (Fig. 4).
Fig. 3  Back propagation through time (back propagation phase)

Fig. 4  The overall framework


of the suggested LSTM-drop-
out-based approach for forecast-
ing Stock market trend

13
1466 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

Fig. 5  Working schema of the proposed approach

13
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478 1467

13
1468 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

13
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478 1469

The semantic representation of the proposed work can be duration of data corresponding to both stock exchange and
realized in Fig. 5. It has been considered five parameters COVID-19 cases indices is taken from 22nd Jan 2020 to
of LSTM as many hidden layers and nodes in each layer 29th May 2021. The initial date of the confirmed case in
(controls the number of weights to be adjusted. If it is more India was reported on ­30th Jan 2020 in the state of Kerala.
then, training time is higher), learning rate (controls the Therefore, to make a fair analysis the training data cor-
degree of change of network weights), decay (responsible responding to the stock market has been from seven days
for the reduction of learning rate over the epoch), activa- back from the initial reporting. Depending on the spread
tion function (controls the activation of nodes), and dropout of COVID-19 positive cases and stock index, the proposed
rate (minimizes the overfitting by removing some fraction of work has been executed with the following results: (1) Pre-
the network that contributes less towards final output). The diction of Nifty50 index, (2) Advanced one-month and two-
value of these parameters is set as follows: several hidden month forecasting of Nifty50. Therefore, the model sepa-
layers and nodes in each layer—2, 200 respectively, Learn- rates the historical data into two groups; training (22nd Jan
ing rate—0.001, Activation Function—Relu, and Dropout 2020 to 29th May 2021) and testing (30th May 2021 to 20th
rate—0.1. August 2021).

Error Index
Critical Outcomes
To validate the proposed COVID-19 prediction approach,
The short-term day-to-day prediction is very important for the MAPE, R2 score (coefficient of determination), RMSE,
supporting the government’s strategic decisions. In this and MAE are considered evaluation indices in this work.
regard, this work concentrate on the prediction of COVID- These indices are mathematically described in Eqs. 12 to
19 confirmed cases with its impact on the Indian stock 15, respectively. Here, D (n) represents the data in the nth
exchange. sample.
M
Data 1 ∑ || Da (n) − Df (n) ||
MAPE = | | (12)
N n=1 || Da (n) |
|
The COVID-19 Indian confirmed, recovered and death data
are obtained from the website1 mentioned in Dong et al. M
(2020). Similarly, the Indian stock exchange data (Nifty50) 1∑
MAE = D (n) − Df (n) (13)
was collected from the official website of the NSE.2 The N n=1 a

1
https://​syste​ms.​jhu.​edu/​resea​rch/​public-​health/​ncov/.
2
https://​www1.​nsein​dia.​com/.

13
1470 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

Fig. 6  COVID-19 in India a total confirmed cases daily basis, b new confirmed cases daily basis

13
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478 1471

Fig. 7  Nifty50 tend during


COVID-19 pandemic a Nifty
50 tend from 22/01/2020 to
16/07/2021, b Nifty 50 tend
from 25/03/2020 to 15/04/2020,
c Nifty 50 tend from 05/04/2021
to 30/04/2021

13
1472 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

Table 3  Prediction result of confirmed cases for India Table 5  Prediction for Nifty50 Closing Price in India from 6 to 20th
2
Aug 2021 (using Proposed Method)
Model studied RMSE MAE MAPE R Score
Date Proposed method
Proposed LSTM-dropout 0.0073 0.0065 0.0030 0.9640
Forecast value APE
LSTM 0.0124 0.0129 0.0084 0.9145
Linear regression 0.0358 0.0365 0.0245 0.8269 8/6/2021 16,287.9 0.003061
Ridge regression 0.0349 0.0314 0.0236 0.7827 8/7/2021 16,287.9 0.003061
Lasso regression 0.0937 0.0799 0.0335 0.6482 8/8/2021 16,287.9 0.003061
Elastic net regression 0.0617 0.0489 0.0134 0.5982 8/9/2021 16,307.4 0.003023
8/10/2021 16,328.6 0.002979
8/11/2021 16,330.7 0.002976
8/12/2021 16,410.5 0.002817
Table 4  Prediction of COVID-19 confirmed cases in India from May
30,2021 to August 20, 2021 8/13/2021 16,570.3 0.002493
8/14/2021 16,570.3 0.002493
Model studied RMSE MAE MAPE R2 Score
8/15/2021 16,570.3 0.002493
Proposed LSTM-dropout 0.0081 0.0026 0.0021 0.9961 8/16/2021 16,603.3 0.00243
LSTM 0.0154 0.0143 0.0103 0.9015 8/17/2021 16,653.2 0.002323
Linear regression 0.0198 0.0717 0.0763 0.6872 8/18/2021 16,608.9 0.002417
Ridge regression 0.0424 0.0964 0.068 0.5839 8/19/2021 16,608.9 0.002417
Lasso regression 0.0251 0.0732 0.0745 0.4841 8/20/2021 16,494.1 0.00265
Elastic net regression 0.0325 0.0237 0.0226 0.6047 Mean 0.002713

Fig. 8  The efficiency of the suggested LSTM method vs. other baseline methods (train sample)

13
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478 1473

Fig. 9  The efficiency of the suggested LSTM method vs. other baseline methods (test sample)

Table 6  Prediction for Nifty50 Closing Price in India from 6 to 20th Aug 2021 (Using Baseline model)
Date Actual LR RR LASSO ENET
Forecast value APE Forecast value APE Forecast value APE Forecast value APE

8/6/2021 16,238.2 16,538.4 0.018487 16,638.4 0.024646 15,548.16 0.042495 15,298.1 0.057894
8/7/2021 16,238.2 16,538.4 0.018487 16,638.4 0.024646 15,578.29 0.040639 15,328.2 0.056041
8/8/2021 16,238.2 16,538.4 0.018487 16,638.4 0.024646 15,608.43 0.038783 15,358.4 0.054181
8/9/2021 16,258.25 16,557.9 0.018431 16,657.9 0.024581 15,638.56 0.038115 15,388.5 0.053496
8/10/2021 16,280.1 16,579.1 0.018366 16,679.1 0.024508 15,668.7 0.037555 15,418.7 0.052911
8/11/2021 16,282.25 16,581.2 0.01836 16,681.2 0.024502 15,698.83 0.035832 15,448.83 0.051186
8/12/2021 16,364.4 16,661 0.018125 16,761 0.024236 15,457.75 0.055404 15,207.7 0.070684
8/13/2021 16,529.1 16,820.8 0.017648 16,920.8 0.023698 15,299.9 0.074366 15,049.9 0.089491
8/14/2021 16,529.1 16,820.8 0.017648 16,920.8 0.023698 15,299.9 0.074366 15,049.9 0.089491
8/15/2021 16,529.1 16,820.8 0.017648 16,920.8 0.023698 15,299.9 0.074366 15,049.9 0.089491
8/16/2021 16,563.05 16,853.8 0.017554 16,953.8 0.023592 15,371.65 0.071931 15,121.65 0.087025
8/17/2021 16,614.6 16,903.8 0.017406 17,003.8 0.023425 15,414.7 0.07222 15,164.7 0.087267
8/18/2021 16,568.85 16,859.4 0.017536 16,959.4 0.023571 15,299.9 0.076586 15,049.9 0.091675
8/19/2021 16,568.85 16,859.4 0.017536 16,959.4 0.023571 15,371.65 0.072256 15,121.65 0.087345
8/20/2021 16,450.5 16,744.6 0.017878 16,844.6 0.023957 15,414.7 0.062965 15,164.7 0.078162
Mean 0.017973 Mean 0.024065 Mean 0.057859 Mean 0.073089

13
1474 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478



√1 ∑ M Prediction Result of Testing Data
( )2
RMSE = √ D (n) − Df (n) (14)
M n=1 a The proposed model was trained and tested by consider-
ing data as mentioned in Sect. 4.1. The obtained results are
∑ ∑ ∑
M Da (n) ∗ Df (n) − Da (n) Df (n) represented in Table 3. The result shows that the proposed
R2 = ��
∑� �2 �∑ �2 � � ∑ � �2 � ∑ �2 � LSTM model effectively predicts the confirmed case of India
M Da (n) − Da (n) − M Df (n) − Df (n) with a mean error indices value of MAPE: 0.0028, MAE:
(15) 0.0058, RMSE: 0.0069, and ­R2 Score: 0.9740. The com-
parative performance analysis with other baseline regression
models as illustrated in Table 3 shows the effectiveness of
Discussion the developed approach. The suggested approach outper-
forms other baseline approaches concerning each evaluation
Figure 6a depicts the total COVID-19 cases in India and indices.
Fig. 6b shows the daily added new confirmed cases. From As the error indices obtained by this analysis are fortu-
these figures, it can be analyzed that the total case will keep nately respectable, the proposed approach is trained with
rising until the curve is sloped. Moreover, the status of daily the historical Nifty50 trend data combined with COVID-19
new cases is stabilized by an average value of 3500 cases cases in India. The testing stock market data which are from
over the last 15 days. As the doubling rate and uncertainty 30th May 2021 to 20th August 2021 are used to validate the
of Indian data is not too high compared to other countries, proposed method. Table 4 shows the result obtained from
the suitability of predicting COVID-19 cases with its impact this analysis. The result shows that the proposed LSTM
on the stock market has not been harshly affected. model effectively predicts the Nifty50 tend with a mean
Meanwhile, the announcement of COVID-19 as a pan- error indices value of MAPE: 0.0030, MAE: 0.0065, RMSE:
demic by WHO, continuing lockdown, and maintaining 0.0073, and R­ 2 Score: 0.9640. The training result regarding
social distancing are a few major issues that have a serious Nifty50 tends data is presented pictorially in Fig. 8. The
impact on the country’s economy. In addition to this, the testing result in the trend of stock market data is depicted
influence of COVID-19 on the share market is so high that in Fig. 9. The results reveal that the LSTM-based proposed
the investors have stop buying; as a result, the selling stock approach outperforms other baseline approaches.
results in a sharp drop in the share index. In this regard, Moreover, to provide a day-to-day error analysis of testing
Fig. 7 shows the trend of Nifty50 in graphical form. Figure 7 data, Tables 5 and 6 provide the actual and forecasted value
shows that the curve of Nifty50 shows a negative trend or of the Nifty 50 index with an average percentage error value.
downfall from 30/03/2020 (first-phase lockdown), stabilized
after (19/05/2020), and from 10/04/2021 (second-phase Advanced Forecasting
partial lockdown), stabilized after (16/07/2021). Moreover,
during each lockdown or its extensions’ announcement, the Advance forecasting of COVID-19 spread is useful for the
trend of Nifty50 shows a transient movement. government’s future planning to stop spreading. Also, the
Based on these analyses, the forecasting of time-series advanced forecasting of the stock index is helpful for the
data of COVID-19 is processed by the developed model. stockbroker and individual company’s preventive action.
The outcome of the proposed work is analyzed by different In this regard, this section has presented the result of the
case studies as follows: The first COVID-19 case in India is advanced 60 days prediction result. Figure 10 shows the
reported on 31st Jan 2020. After this, the central government forecasting result of COVID-19 cases for India for one
of India has decided to implement a full lockdown in the month ahead of 21st August 2021 and similarly, the fore-
country from 25 March 2020 (in the first phase) and a partial casting results of Nifty50 two months ahead of 21st August
lockdown on 05th April 2021 (in the second phase). In this 2021 are presented in Fig. 11. The dotted straight red lines
regard, the proposed model has taken the Nifty 50 data from indicate the lockdown/unlock imposed to date. As the Indian
before lockdown, i.e. up to 29th May 2021 for training and government has implemented the lockdown with different
forecast the Nifty 50 stock index for advanced two months. criteria, the impact of the lockdown on the stock market has
It can easily observe from the figure that the implementa- shown a different trend.
tion of lockdown has a positive effect on the stock exchange
of India.

13
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478 1475

Fig. 10  Forecasted tend of Nifty 50 for one month

Fig. 11  Forecasted tend of Nifty 50 for two month

Conclusion transformations have supplied the uncertainty in the stock


exchange and enhanced the volatility. In this regard, this
The current Pandemic ‘COVID-19’ has not only changed work analyzes the influence of increasing COVID-19 cases
the lifestyle of individual countrymen but also has shown with various restrictions (lockdown and partial lockdown)
a major role in transforming the decision policies of imposed by Govt. and its impact on the trend of the Indian
companies and government. These instantaneous policy stock exchange. The data of the Nifty 50 index under the

13
1476 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

NSE have been chosen for the experiment. This work has Declaration
used the historical data of Nifty 50 in addition to COVID-
19 Indian data and lockdown durations for predicting the Conflict of interest The authors declare that this manuscript has no
future stock market index. The LSTM-based DL approach conflict of interest with any other published source and has not been
published previously (partly or in full). No data have been fabricated
is used for designing this prediction model. The prediction or manipulated to support the conclusions.
performance is validated through four major evaluation met-
rics such as MAPE, R ­ 2 score RMSE, and MAE. Moreover,
the outcome of the suggested approach has been compared Appendix
with other baseline methods (RR, LR, LASSO, Elastic net,
etc.) to check the effectiveness of the work. The comparative
result indicates that the LSTM-based approach outperforms
others by a significant margin. Furthermore, the outcome Abb Actual meaning
of this study concludes that the increasing rate of Pandemic
LSTM long-short-term-memory
cases and various restrictions on lockdown announcement
CDC Center for disease control
has a significant impact on stock indices. However, a few
NCP novel coronavirus pneumonia
limitations of the proposed work are also found as it takes
SARS-CoV-2 Severe Acute Respiratory Syndrome Corona Virus-2
more space and time for training the network. Also, the
ICTV International Committee on Taxonomy of Viruses
model is sensitive to the randomly generated initial weights
WHO World Health Organization
and always it has been a challenging issue to decide the
RT-PCR Reverse Transcription Polymerase Chain Reaction
optimal values of dropout rate.
CT Computer tomography
ML Machine Learning
DL Deep Learning
Future Scope
ARIMA Autoregressive Integrated Moving Average
SMR Stock market returns
Though several machine learning and deep learning mod-
GRU​ Gated Recurrent Unit
els are well developed to predict the behavior of the stock
NSE National Stock Exchange
market, still few issues are open. The unusual or stochas-
MAPE Mean Absolute Percentage Error
tic nature of the stock market makes the models impera-
RMSE Root Mean Square Error
tive for solving the issue. Especially, extraction of valuable
MAE Mean Absolute Error
features from the acquired data and analysis of their impact
on the behavior of the stock market is a challenging task.
Mainly, the intelligent models should be capable of real-
izing the diversity of prediction variables which are helpful
in forecasting. As a future insight, the proposed model may
References
be trained to efficiently work with a few interesting factors 1. WMHCI, Wuhan Municipal Health Commission Infection Data
involved with live testing such as large differentiation in (2020). http://​wjw.​wuhan.​gov.​cn/​front/​web/​list2​nd/​no/​710
price, unpredicted/unusual behavior of events, etc. Moreo- 2. C. Huang, Y. Wang, X. Li, L. Ren, J. Zhao, Y. Hu, L. Zhang, G.
ver, due to the popularity of online trading, it will be a major Fan, J. Xu, X. Gu, Z. Cheng, Clinical features of patients infected
with 2019 novel coronavirus in Wuhan, China. The Lancet
requirement to develop intelligent models to simulate the 395(10223), 497–506 (2020)
behavior of stock users addicted to social media sentiments. 3. J. Cui, F. Li, Z.L. Shi, Origin and evolution of pathogenic coro-
Also, working on the models dealing with market volatility naviruses. Nat. Rev. Microbiol. 17(3), 181–192 (2019)
and analysis of risk factors will be of greater future interest. 4. WHO, Laboratory Testing for Coronavirus Disease (COVID-19)
in Suspected Human Cases: Interim Guidance, 19 March 2020
The proposed method can be useful in some other applica- (No. WHO/COVID-19/laboratory/2020.5) (World Health Organi-
tions like predictive analysis on viral disease growth rates, zation, 2020)
weather forecasting, time series forecasting, supply chain, 5. H. Aly, S.M. Youssef, Bio-signal based motion control system
demand forecasting, etc. using deep learning models: a deep learning approach for motion
classification using EEG and EMG signal fusion. J. Ambient
Intell. Human. Comput. 593, 1–12 (2021)
6. J.L. Casteleiro-Roca, M. Gomes, J.A. Méndez-Pérez, H. Alaiz-
Funding This submission does not support by any funding from any Moretón, M.D.C. Meizoso-López, B.A. Rodríguez-Gómez, J.L.
sources. Calvo-Rolle, Electromyogram prediction during anesthesia by
using a hybrid intelligent model. J. Ambient. Intell. Humaniz.
Comput. 11(11), 4467–4476 (2020)

13
J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478 1477

7. S. Saravanan, R. Karthigaivel, V. Magudeeswaran, A brain tumor Conference on Computer Vision and Pattern Recognition (IEEE,
image segmentation technique in image processing using ICA- 2009), pp. 248–255
LDA algorithm with ARHE model. J. Ambient. Intell. Humaniz. 28. L. Wang, A. Wong, COVID-Net: a tailored deep convolutional
Comput. 12(5), 4727–4735 (2021) neural network design for detection of COVID-19 cases from
8. R. Preetha, S.V. Jinny, Early diagnose breast cancer with PCA- chest radiography images (2020). arXiv:​2003.​09871
LDA based FER and neuro-fuzzy classification system. J. Ambi- 29. The Hindu. PTI. World Bank sees FY21 India growth at 1.5–2.8%,
ent. Intell. Humaniz. Comput. 12(7), 7195–7204 (2021) slowest since economic reforms 30 years ago. 12 April 2020.
9. S. Sahoo, P. Das, P. Biswal, S. Sabut, Classification of heart Retrieved 27th Apr 2020
rhythm disorders using instructive features and artificial neural 30. The Times of India, World Bank Sees FY21 India Growth at 1.5–
networks. Int. J. Med. Eng. Inform. 10(4), 359–381 (2018) 2.8%—Slowest Since Economic Reforms Three Decades Ago. 12
10. V. Gupta, M. Mittal, R-peak detection in ECG signal using Yule– Apr 2020. Retrieved 27th Apr 2020
Walker and principal component analysis. IETE J. Res. 67(6), 31. The Hindu Business Line, Covid-19 Lockdown Estimated to Cost
921–934 (2021) India $4.5 Billion a Day: Acuité Ratings. 2 Apr 2020. Retrieved
11. X. Li, D. Li, Y. Deng, J. Xing, Intelligent mining algorithm for 27th April 2022
complex medical data based on deep learning. J. Ambient. Intell. 32. The Hindu @businessline PTI, Experts Peg India’s Cost of Cor-
Humaniz. Comput. 12(2), 1667–1678 (2021) onavirus Lockdown at USD 120 bn. Retrieved 27th Apr 2022
12. A. Al-Yousef, H. Eloqayli, M. Obiedat, A. Almoustafa, Predicting (2020)
treatment outcome of spinal musculoskeletal pain using artificial 33. M. Mukherji, Coronavirus Impact: Indian Industry Seeks Relief
neural networks: a pilot study. Int. J. Med. Eng. Inform. 13(3), Measures to Aid Economy. Livemint. Retrieved 27th Apr 2020
237–253 (2021) (2020)
13. Y. Ma, H. Zhang, Deep mining of communication information 34. The Hindu @businessline. Jump up to:a b Chaudhry, Siraj A.
association based on discrete Fourier transform. J. Ambient Intell. "Covid-19 puts India’s Food Supply Chain to a Stress-Test.
Human. Comput. 53, 1–12 (2021) Retrieved 27th Apr 2020 (2020)
14. V. Gupta, M. Mittal, V. Mittal, A. Gupta, ECG signal analysis 35. G. Das, 136 million jobs at risk in post-corona India. Livemint.
using CWT, spectrogram and autoregressive technique. Iran J. Retrieved 27th Apr 2020 (2020)
Comput. Sci. 4(4), 265–280 (2021) 36. The Economic Times.Jump up to:a b c d Goyal, Malini, Covid-
15. V. Gupta, M. Mittal, A novel method of cardiac arrhythmia detec- 19: How the Deadly Virus Hints at a Looming Financial Crisis
tion in electrocardiogram signal. Int. J. Med. Eng. Inform. 12(5), (2020). Retrieved 27th Apr 2020
489–499 (2020) 37. Y. Mahesh, Unemployment Rate Touches 26%. (Centre for Moni-
16. V. Gupta, M. Mittal, V. Mittal, Chaos theory: an emerging tool for toring Indian Economy (CMIE), 2020). Retrieved 27th Apr 2022
arrhythmia detection. Sens. Imaging 21(1), 1–22 (2020) 38. O. Kowalewski, P. Śpiewanowski, Stock market response to potash
17. D. Al-Karawi, S. Al-Zaidi, N. Polus, S. Jassim, Machine Learning mine disasters. J. Commod. Mark. 43, 100124 (2020)
Analysis of Chest CT Scan Images As A Complementary Digital 39. R. Buhagiar, D. Cortis, P.W. Newall, Why do some soccer bettors
Test of Coronavirus (COVID-19) Patients (2020). medRxiv lose more money than others? J Behav Exp Finance 18, 85–93
18. D. Benvenuto, M. Giovanetti, L. Vassallo, S. Angeletti, M. Cic- (2018)
cozzi, Application of the ARIMA model on the COVID-2019 40. M. Guo, Y. Kuai, X. Liu, Stock market response to environmental
epidemic dataset. Data Brief 29, 105340 (2020) policies: Evidence from heavily polluting firms in China. Econ.
19. M. Talo, U.B. Baloglu, Ö. Yıldırım, U.R. Acharya, Application of Model. 86, 306–316 (2020)
deep transfer learning for automated brain abnormality classifica- 41. A. Bash, K. Alsaifi, Fear from uncertainty: An event study of
tion using MR images. Cogn. Syst. Res. 54, 176–188 (2019) Khashoggi and stock market returns. J. Behav. Exp. Financ. 23,
20. S.A.A. Ismael, A. Mohammed, H. Hefny, An enhanced deep 54–58 (2019)
learning approach for brain cancer MRI images classification 42. S. Shanaev, B. Ghimire, Is all politics local? Regional politi-
using residual networks. Artif. Intell. Med. 102, 101779 (2020) cal risk in Russia and the panel of stock returns. J. Behav. Exp.
21. N.E.M. Khalifa, M.H.N. Taha, D.E. Ali, A. Slowik, A.E. Has- Financ. 21, 70–82 (2019)
sanien, Artificial intelligence technique for gene expression by 43. M.H. Chen, S.S. Jang, W.G. Kim, The impact of the SARS out-
tumor RNA-Seq data: a novel optimized deep learning approach. break on Taiwanese hotel stock performance: an event-study
IEEE Access 8, 22874–22883 (2020) approach. Int. J. Hosp. Manag. 26(1), 200–212 (2007)
22. C.S. Ho, N. Jean, C.A. Hogan, L. Blackmon, S.S. Jeffrey, M. 44. C.D. Chen, C.C. Chen, W.W. Tang, B.Y. Huang, The positive
Holodniy, N. Banaei, A.A. Saleh, S. Ermon, J. Dionne, Rapid and negative impacts of the SARS outbreak: a case of the Taiwan
identification of pathogenic bacteria using Raman spectroscopy industries. J. Dev. Areas 43, 281–293 (2009)
and deep learning. Nat. Commun. 10(1), 1–8 (2019) 45. R. Ichev, M. Marinč, Stock prices and geographic proximity of
23. I.R.I. Haque, J. Neubert, Deep learning approaches to biomedical information: evidence from the Ebola outbreak. Int. Rev. Financ.
image segmentation. Inform. Med. Unlock. 18, 100297 (2020) Anal. 56, 153–166 (2018)
24. S. Rajaraman, S. Candemir, I. Kim, G. Thoma, S. Antani, Visuali- 46. N. Fernandes, in Economic Effects of Coronavirus Outbreak
zation and interpretation of convolutional neural network predic- (COVID-19) on the World Economy (2020). Available at SSRN
tions in detecting pneumonia in pediatric chest radiographs. Appl. 3557504
Sci. 8(10), 1715 (2018) 47. S. Chaney, D. Harrison, U.S. Jobless, Claims Soar for Third
25. M. Anthimopoulos, S. Christodoulidis, L. Ebner, A. Christe, S. Straight Week, (2020). https://​www.​wsj.​com/​artic​les/u-​s-​surge-​
Mougiakakou, Lung pattern classification for interstitial lung dis- in-​unemp​loyme​nt-​claims-​expec​ted-​to-​conti​nue-​11586​424605
eases using a deep convolutional neural network. IEEE Trans. 48. W. Watts, Coronavirus Stock-Market Volatility is Creating the
Med. Imaging 35(5), 1207–1216 (2016) Largest Daily Price Swings Since 1929 Crash (2020). https://​
26. E.E.D. Hemdan, M.A. Shouman, M.E. Karar, Covidx-net: a www.​marke​twatch.​com/​story/​stock-​market-​inves​tors-​have-​to-​
framework of deep learning classifiers to diagnose covid-19 in go-​back-​to-​1929-​to-​find-​daily-​swings-​this-​wild-​2020-​04-​07
x-ray images (2020). arXiv:​2003.​11055. 49. A.M. Al-Awadhi, K. Al-Saifi, A. Al-Awadhi, S. Alhamadi, Death
27. J. Deng, W. Dong, R. Socher, L.J. Li, K. Li, L. Fei-Fei, Ima- and contagious infectious diseases: impact of the COVID-19 virus
genet: A large-scale hierarchical image database, in 2009 IEEE on stock market returns. J. Behav. Exp. Finance 27, 100326 (2020)

13
1478 J. Inst. Eng. India Ser. B (October 2022) 103(5):1459–1478

50. A.S. Ahmar, E.B. del Val, SutteARIMA: short-term forecasting 56. R. Kizys, P. Tzouvanas, M. Donadelli, From COVID-19 Herd
method, a case: Covid-19 and stock market in Spain. Sci. Total Immunity to Investor Herding in International Stock Markets: The
Environ. 729, 138883 (2020) Role of Government And Regulatory Restrictions (2020). Avail-
51. B.M. Pavlyshenko, Regression Approach for Modeling COVID-19 able at SSRN 3597354
Spread and its Impact on Stock Market (2020) arXiv:​2004.​01489 57. K. Khan et al., The impact of COVID-19 pandemic on stock mar-
52. A.C. Karina, A.M. Fernando, N.N. Morteza, H. Michael, Fore- kets: an empirical analysis of world major stock indices. J. Asian
casting the Effect of COVID-19 on the S&P500 (2020) arXiv:​ Finance Econ. Bus. 7(7), 463–474 (2020)
2005.​03969 58. H.S. Lee, Exploring the initial impact of COVID-19 sentiment
53. H. Liu, A. Manzoor, C. Wang, L. Zhang, Z. Manzoor, The on US stock market using big data. Sustainability 12(16), 6648
COVID-19 outbreak and affected countries stock markets (2020)
response. Int. J. Environ. Res. Public Health 17(8), 2800 (2020) 59. E. Onali, Covid-19 and Stock Market Volatility (2020). Available
54. E. Dong, D. Hongru, G. Lauren An Interactive Web-Based Dash- at SSRN 3571453
board To Track COVID-19 in Real Time. The Lancet Infectious 60. S. Hochreiter, J. Schmidhuber, Long short-term memory. Neural
Diseases (2020). https://​syste​ms.​jhu.​edu/​resea​rch/​public-​health/​ Comput. 9(8), 1735–1780 (1997)
ncov/
55. S. Ramelli, A.F. Wagner, Feverish stock price reactions to covid- Publisher’s Note Springer Nature remains neutral with regard to
19 (2020) jurisdictional claims in published maps and institutional affiliations.

13

You might also like