Received 5 July 2023, accepted 26 July 2023, date of publication 31 July 2023, date of current version 7 August 2023.

Digital Object Identifier 10.1109/ACCESS.2023.3300036

Applications of Artificial Intelligence in the

Economy, Including Applications in Stock
Trading, Market Analysis, and Risk Management
AMIR MASOUD RAHMANI1 , BAHAREH REZAZADEH 2 ,
MAJID HAGHPARAST 3 , (Senior Member, IEEE), WEI-CHE CHANG1 , AND SHEN GUAN TING1
1 Future Technology Research Center, National Yunlin University of Science and Technology, Douliou, Yunlin 64002, Taiwan
2 Department of Computer Engineering, Science and Research Branch, Islamic Azad University, Tehran 1477893855, Iran
3 Faculty of Information Technology, University of Jyväskylä, 40014 Jyväskylä, Finland
Corresponding author: Majid Haghparast ([email protected])
This work was supported by the University of Jyväskylä (JYU) for Open Access.

ABSTRACT In an increasingly automated world, Artificial Intelligence (AI) promises to revolutionize how
people work, consume, and develop their societies. Science and technology advancement has led humans to
seek solutions to problems; however, AI-based technology is not novel and has a wide range of economic
applications. This paper examines AI applications in economics, including stock trading, market analysis,
and risk assessment. A comprehensive taxonomy is proposed to investigate AI applications in various scopes
of the proposed categories. Furthermore, we will discuss this area’s most significant AI-based techniques
and evaluation criteria. As a final step, we will identify challenges, open issues, and future work suggestions.

INDEX TERMS Internet of Things, artificial intelligence, economy, machine learning, stock market, neural
network.

I. INTRODUCTION and enhancing overall performance in stock trading, market

Artificial Intelligence (AI) is a rapidly evolving field with analysis, and risk management. RL, a branch of AI, has
broad economic applications. AI technologies have become become a powerful tool in portfolio construction, enabling
increasingly important for both the public and private sectors, automated decision-making processes and dynamic adapta-
providing invaluable insights into the current performance of tion to market conditions [1]. By leveraging RL techniques,
the economy, as well as potential future directions. It can portfolio construction processes can benefit from adap-
be used to develop and refine economic models, automate tive and dynamic decision-making, improving risk-adjusted
processes, and inform decision-making. AI is a broad term returns and enhancing portfolio performance [2]. Using RL
covering various technologies and techniques, but the core for portfolio construction involves several steps. The first step
idea is that machines can be trained to ‘‘think’’ and make in using RL for portfolio construction is defining a state rep-
decisions. AI technologies can automate processes, provide resentation that captures relevant market data, such as asset
insights, and conduct predictive analysis. AI is a rapidly prices, trading volumes, and macroeconomic indicators [3].
developing field, and its economic applications are growing Decision-making and policy learning are based on this state
rapidly. representation. RL algorithms require a defined action space,
Recently, Machine Learning (ML), Deep Learning (DL), which specifies the possible actions each stage can take.
and Reinforcement Learning (RL) techniques have gained An action space represents the allocation of funds across
significant traction in various sectors of the economy. various assets in portfolio construction [4]. As another step,
AI-driven approaches have shown promising results for the reward function plays a crucial role in RL algorithms,
improving decision-making processes, optimizing strategies, quantifying the desirable of different actions. It may be built
on risk-adjusted returns, volatility, or downside risk mea-
The associate editor coordinating the review of this manuscript and sures. Optimal decision-making policies through an iterative
approving it for publication was Jose Saldana . process of exploration and exploitation involve evaluating

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 11, 2023 80769
 A. M. Rahmani et al.: Applications of AI in the Economy

different actions, updating the policy based on observed The paper is organized as follows: Section II describes the
rewards, and gradually converging toward an optimal strat- concepts and research methodology. The analysis of chosen
egy [5]. It is the last step of RL utilization in portfolio papers in three main categories is provided in section III.
construction, called policy learning. Section IV answers the Research Questions (RQs) defined in
AI can be used to identify patterns in complex economic Section II and proposes technical analysis and visual reports
data, allowing for more accurate predictions of future eco- based on Section III findings. Several cases of AI applica-
nomic performance. It can also automate processes like tions in successful real-world companies are introduced in
financial transactions, improving efficiency, reducing costs, Section V, and Section VI concludes the paper.
informing decision-making, improving economic model
accuracy, and reducing decision-making time. Furthermore, II. CONCEPTS AND RESEARCH METHODOLOGY
complex economic changes and potential investment oppor- As a foundational step, reviewing existing knowledge of a
tunities can be identified by posing AI. research scope provides researchers with a comprehensive
AI has the potential to revolutionize the labor force through understanding of the current state of research. Addition-
automation. Reducing labor through automation also reduces ally, it assists them in identifying key theories, significant
costs and increases efficiency since it allows higher-skilled advancements, knowledge gaps, and unresolved questions,
workers to be employed. as well as exploring various perspectives, methodologies,
Inflation has been a serious issue happening worldwide and findings. Moreover, it enables researchers to assess the
in recent years. The average per-capita income has not quality and credibility of previous research, ensuring that
increased. However, people must still pay taxes and bills or their review paper builds upon a robust foundation. Therefore,
buy houses. In these circumstances, people are likely to spend in this section, we analyze several recent review papers.
more than their earnings. As a result, people will seek various Cao et al. [10] reviewed modern economies focused on
methods to increase their bank savings [6]. the smart version of Financial Technology (FinTech), which
In finance, stock commerce provides such a chance to is empowered by Data Science and AI (DSAI). DSAI plays
reach your goals. We probably did not have experience a critical role in transforming the economy, including busi-
or knowledge of trading. We might buy stocks based on nesses and personal finance, to be more intelligent and
guesswork or luck. However, the point of investing is to automated. Although the smart FinTech ecosystem contains
prevent your asset from decreasing while increasing its value technologies in vast various scopes, this study introduced
simultaneously. Fortunately, as time passed, experienced some of them and focused on smart FinTech in financial busi-
traders developed many analytical tools, such as elemen- nesses. This study reviewed DSAI techniques that empow-
tary analysis, technical analysis, and measurement, which ered smart FinTech based on a comprehensive taxonomy,
facilitated trading for investors or increased their investment including complex systems methods, quantitative methods,
success rate [7]. However, the tools developed using ML tech- data analytics, DL, privacy-preserving processing, augmenta-
niques have been used in computational sciences and various tion, optimization, and to name a few provided in a table with
applications. their representative applications in smart FineTech. Accord-
Experienced traders are attempting to implement ML to ing to the study’s authors, all FinTech areas have fundamental
predict stock prices. However, every product has its char- processes, such as design, production, operation, promotion,
acteristics, such as market liquidity and size. To solve this optimization, and safeguarding. The authors identified chal-
problem, modifying feature selection or data modeling might lenges and open issues from research analyzed in their paper
be the solution. Currently, AI assists investors with timely and provided DSAI trends and future opportunities.
and precise suggestions for decision-making. However, the Shah et al. [11] investigated the ability of AI and ML
uncertainty of AI-based trading systems makes investors to estimate stock prices, predict future trends, and manage
unwilling to rely on it. Thus, to reduce the risks, the main portfolios. The authors compared the accuracy and error cal-
goal is to try combining multiple algorithms to gain more trust culation of ARIMA, LSTM, Hybrid LSTM, Convolutional
from the investors [8]. Currently, there are several ML mod- Neural Network (CNN), and Hybrid CNN techniques for
els. However, two algorithms have outstanding performance: stock price prediction to standard accuracy measures such
Long Short-Term Memory (LSTM) and Auto-Regressive as Root Mean Square Error (RMSE), MAPE, and MAE.
Integrated Moving Average (ARIMA) [9]. A more sustain- A comparison of all the methods that have been evaluated
able and reliable model is possible with these two algorithms’ has demonstrated that ARIMA is the worst prediction model.
collaboration. The LSTM and Hybrid LSTM models can be used to forecast
At the time, AI is a rapidly evolving field with broad eco- future stock prices, while the NN and Hybrid CNN mod-
nomic applications. AI technologies can automate processes, els, trained by large datasets, provide excellent stock trend
provide insights, and conduct predictive analysis. This paper forecasts. Therefore, a hybrid CNN/LSTM model is most
will discuss AI’s current and potential economic applications, accurate for predicting stocks’ future trends and prices and
its impacts on the labor force, and the potential for increased can be used to build and manage portfolios. Stock fluctua-
economic growth. tions are also influenced by people’s sentiments, which can

80770 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 1. A summary of relevant work on AI applications in the economy.

be observed on social media and analyzed using Natural • Volatility: It refers to the amount and speed at which
Language Processing (NLP). Feeding sentiment analysis data prices change over time.
into a Deep Neural Network (DNN) to predict stock trends • Closing price: The latest price at which a stock is traded
and prices more accurately is possible. Even though this on a typical trading day.
study introduced a properly integrated model for the stock • Efficiency: The time that the system comes up with the
market among its candidate models, many ML techniques result.
may outperform these models and may be a direction for • Accuracy: It assesses the classifier’s ability to predict all
future studies. classes properly.
Mintarya et al. [12] reviewed thirty research papers on • Absolute error: The discrepancy between a quantity’s
ML methods for stock market prediction in a Systematic measured or perceived value and its true value.
Literature Review (SLR) paper. These research papers used • Autocorrelation: The degree of similarity between a par-
Neural Networks (NNs), Support Vector Machines (SVMs), ticular time series and its lagged version over subsequent
LSTM, and other algorithms to predict stock market changes. periods.
According to the comparison, NN is the most frequently • Training performance: The result with the highest
applied technique. This study is limited by a lack of taxonomy accuracy.
and a brief focus on a few papers. • Cumulative abnormal return: Cumulative returns of
Table 1 summarizes the main features of relevant review portfolio optimization strategies.
studies on AI applications in the economy and identifies their
Based on the systematic review method, searching, refining,
limitations. It facilitates the improvement of our research
and choosing several studies are conducted and classified on
contributions.
AI economic applications. The keywords and search terms
Based on the previous research in this scope, the main
and their alternative synonyms we used to find relevant
contributions of this paper are:
research papers are as follows:
• Introducing a comprehensive taxonomy of AI applica- (‘‘AI’’ OR ‘‘artificial intelligence’’) AND (‘‘economy’’
tions in the economy. or ‘‘finance’’) AND (‘‘prediction’’ OR ‘‘application’’) AND
• Identifying the most significant AI-based techniques in (‘‘stock’’ OR ‘‘market’’ OR ‘‘risk’’).
the economy. After finding numerous papers based on our search terms,
• Discovering the critical evaluation criteria in AI appli- we applied several inclusion and exclusion criteria to effec-
cations in the economy. tively choose the most proper papers to refine our search
• Identifying the important challenges of AI applications results.
in the economy. The inclusion criteria are as follows:
• Providing open issues and future direction related to AI
• Taxonomy categories and subcategories are included in
applications in the economy.
the paper’s title.
Determining key terminology can facilitate the understand- • Papers with high citations.
ing of subsequent content and provide clarity for readers. • Papers with a high scientific value.
Therefore, we have defined and listed key expertise terms
The exclusion criteria are as follows:
related to AI and economics to assist readers in the following:
• Papers published before 2017.
• Investment strategy: It helps individual investors meet
• Irrelevant papers.
their financial and investment objectives.
• Papers without technical analysis.
• Portfolio value: The total monetary value of the assets
held in your investment portfolio. Figure 1 shows the paper selection process.
• Accumulated profit: The remaining profit corporations Finally, the following figure illustrates the distribution of
own after deducting dividend expenses. selected papers based on their publication year and publisher.

VOLUME 11, 2023 80771

 A. M. Rahmani et al.: Applications of AI in the Economy

presented in Figure 3 details various applications of AI in

each category.
ML algorithms are used in AI-based trading applications
to analyze market data, detect trends, and predict potential
buying and selling opportunities. AI algorithms can identify
and interpret patterns in the market that may be too complex
or subtle for humans to recognize [13]. A further benefit of
AI is that it can predict stock price movements, detect market
entry and exit points, suggest optimal portfolio allocations,
automate the trading process, and allow traders to devote
more time to research and analysis.
Market analysis can be conducted using AI to identify
correlations between economic variables, such as consumer
spending, market prices, and macroeconomic indicators.
Through NLP, these applications analyze news articles, social
media posts, and other sources to determine how people
feel about a particular topic or company. It helps investors
identify trends and determine public opinions [14]. As well
as identifying trends, it can compare news sources.
FIGURE 1. The paper section process.
Risk management entails identifying and mitigating risk
factors through various measures. AI applications analyze
large amounts of data to identify potential risks, predict
outcomes, and provide timely insights for decision-making.
This category includes different information, including finan-
cial information, market trends, and historical data [15].
By detecting anomalies, patterns, and correlations, AI algo-
rithms can detect risks like fraud, cybersecurity threats,
or operational inefficiencies. Through continuous data mon-
itoring and analysis, AI systems facilitate proactive risk
management, optimize portfolio risk exposure, and ensure
regulatory compliance. Moreover, AI-based risk manage-
ment solutions enable organizations to improve their response
capabilities, prevent potential losses, and increase their
resilience to rapidly changing risk environments [16].
FIGURE 2. Year of publication and publisher of research papers.
III. SELECTED PAPERS ANALYSIS
According to the proposed taxonomy in Figure 3, we analyze
Figure 2 illustrates the distribution of the selected research the chosen papers in three main categories. First, we examine
papers on AI applications in the economy from various rep- research on AI applications for stock trading, a hot topic in
utable publishers such as IEEE, Springer, Elsevier, and other society today.
miscellaneous publishers.
A. STOCK TRADING APPLICATIONS
This paper uses the systematic review method to address
the following RQs: People today attempt to increase their assets through various
methods, including investing in stocks. Although investing
• RQ1: Which branches of AI are applied in economics? in all products does not guarantee financial success, many
• RQ2: What are the challenges associated with AI appli- people seek ways to predict stock prices. The fluctuating
cations in the economy? historical data due to the fluctuating price. As a result,
• RQ3: Which techniques are most commonly used in AI economists are increasingly involved in developing models
applications in the economy? and methods for improving stock trading and forecasting
• RQ4: What are the most significant evaluation metrics market trends with greater accuracy [17]. The following is a
for AI applications in the economy? review of some of these articles to identify the most common
• RQ5: What are the current open issues and future direc- techniques and solutions, limitations, and challenges in this
tions of AI in the economy? field.
Our analysis of AI applications in the economy is divided Liang et al. [18] developed a hybrid wavelet-DL model
into three main categories: stock trading, market analysis, and that combines Daubechies wavelet (DB) with the bidirec-
risk management. In addition, the comprehensive taxonomy tional LSTM (BLSTM) model called (DB-BLSTM) to cope

80772 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

index futures data in developing markets. Wavelets are uti-

lized for their ability to reduce the interference caused by
high-frequency data and their sensitivity to time-varying
frequencies. Moreover, the researchers constructed trading
strategies based on the proposed hybrid prediction models
and evaluated the results regarding realized returns, Sharpe
ratios (SRs), and maximum drawdowns. After deducting
transaction costs and benchmark returns, the strategies gen-
erate significant excess returns ranging from 15.69% to
45.52%. The strategy also exhibits an impressive SR, indi-
cating favorable risk-adjusted returns. The proposed model
is compared with commonly used models like LSTM and
Artificial Neural Networks (ANN) for predicting stock
index futures intraday trends. The empirical comparison
results demonstrate superior performance in both in-sample
and out-of-sample tests. However, the lack of an inter-
national dataset for evaluating this model’s effectiveness
limits this study which is a potential direction for future
research.
Cohen [19] developed algorithmic trading platforms for
popular cryptocurrencies using ML systems based on the
Relative Strength Index (RSI), Moving Average Convergence
Divergence (MACD), and Keltner Channels trading oscilla-
tors. The LSTM is used to identify optimal trading setups
and generate trading signals for automated intraday trading
of Bitcoin, Ethereum, Binance Coin, Cardano, and XRP
across various timeframes, ranging from 5 to 180 minutes.
The system alternates 5, 15, 30-, 60-, 120-, and 180-minute
timeframes between trading bars to maximize profits. Com-
paring these three strategies, the RSI system outperformed
the Buy and Hold (B&H) strategy for all five cryptocurren-
cies, which supports its effectiveness in improving trading
outcomes under both uptrends and downtrends. In partic-
ular, when trades were split into long and short positions,
the RSI-based system consistently performed better in long
positions, whereas the other systems generated adverse Net
Profit (NP) for Ethereum and Cardano. According to the
results of this study, longer time frames, such as 60 and
120 minutes, offered better trading results than shorter time
frames, such as 5 and 15 minutes. A study concluded that
intraday algorithmic trading systems could be more profitable
for cryptocurrencies and suggested the development of simi-
lar systems in the future based on other AI methods.
Technology has accelerated the development of many
emerging industries in China, resulting in continuous growth
and progress in the nation’s economy. Still, it has also led
to severe turbulence in the stock market, making predicting
stock prices complex and difficult. Therefore, Kan et al. [20]
proposed a dynamic portfolio management model that com-
bines Deep Reinforcement Learning (DRL) and stock trading
to improve stock prediction. Various AI techniques were used
FIGURE 3. The taxonomy of AI applications used in the economy. in their model, and the Shanghai Stock Exchange dataset was
used to simulate it. Based on SR and Accumulated Portfolio
Value (APV) evaluation metrics, this model outperforms its
with complicated periodicity and nonlinearity features in competitors. This model did not consider transaction costs,
the high-frequency data and analyze high-frequency stock an area for future study.

VOLUME 11, 2023 80773

 A. M. Rahmani et al.: Applications of AI in the Economy

Kumar et al. [21] developed an LSTM-based stock price may attest to the feasibility of utilizing a DNN to experiment.
prediction model that improves prediction accuracy and The results show 75% accuracy. Enhancing accuracy and
performance over various timeframes. Mini-batch gradient utilizing more indicators, such as chip analysis, financing,
descent is used as the optimization algorithm in this model and securities lending, in this model is a direction for future
to enhance speed and reduce prediction errors. The authors work.
simulated their proposed model in Python using datasets from Kumar et al. [24] proposed a hybrid CNN-LSTM stock
the Pandas website to compare its accuracy with other DL price prediction model to enhance the efficiency and correct-
models. Results show that this model performs better than ness of stock prediction. They use historical data based on
previous models, speeding up prediction times and reducing the NIFTY50 index. The period of this dataset is between
errors. A lack of measurement of various evaluation metrics 2015 and 2020. They tried various prebuilt and DL models
limited this study, which may form a direction for future to predict the value of the stock in the next ten days. They
research. initially attempted to combine the ARIMA and the Seasonal
Zhao et al. [22] focussed on the complexity of finan- ARIMA model with exogenous components (SARIMAX).
cial markets because of the variety of industries, investors, They discovered this technique had lower accuracy than the
and data types to be considered. In this study, they investi- method combined with CNN and LSTM. As a result, using
gated the development of agent systems for everyday stock the same dataset, the new model, which combines CNN and
trading decision support, and they proposed an improved LSTM, has the best performance. The future work of this
deep Q-learning algorithm model for stock trading prediction research is to enhance the correctness by applying the RL
to deal with divide-and-conquer problems in an integrated method and a larger dataset.
environment. The authors combined Q-learning agents and Selvamuthu et al. [25] predicted the stock price using
Median Absolute Deviation (MAD) algorithms to efficiently three distinct algorithms. The algorithms are Levenberg-
divide and conquer stock trading problems in an integrated Marquardt, Scaled Conjugate Gradient (SCG), and Bayesian
environment. This combination enables stock data processing Regularization. They take data from Indian companies and
from the starting point and analyzing them based on their use the 15-minute data to analyze it with the three algorithms.
characteristics which is an essential foundation for future Finally, they compared the results of the three algorithms.
agent simulation by the Q-learning model. This combined The results indicate that these algorithms have an accuracy
predictive model covers the challenges of single Q-learning of up to 99.9%. However, the accuracy after fifteen min-
methods, including original data preprocessing, reward val- utes decreased to 96.2%, 97.0%, and 98.9%, respectively.
ues, and model testing problems. Under big data, this model Although the accuracy of these three algorithms has reduced,
uses RL methods to forecast changes in stock trading. MAD is as evidenced by this study, the result calculated by the SCG
used to analyze univariate numerical data when high correla- algorithm has the best performance. This research aims to
tions or sample bias are an issue due to its greater adaptability increase the accuracy of the prediction and allow users to
to outliers than the standard deviation. predict specific periods from historical data to get a better
Further, it solves the problem of high correlation between result.
data and optimizes the State, Action, and Reward functions, Ta et al. [26] researched several core components of the
making it suitable for use in financial markets. Using the quantitative trading system. They discovered that ML has
NASDAQ Composite (IXIC) simulator, ‘‘^IXIC,’’ and a real many advantages compared to traditional algorithmic trading.
dataset of the US stock market, the authors demonstrated Using ML, you may synchronously apply certain trading
that their proposed approach outperformed the separate strategies and adapt them based on the real-time market.
Q-learning framework. Additionally, the Q-learning conver- Besides, combining numerous optimized algorithms can con-
gence analysis is evaluated in this study. As a result of trol risks during every transaction. It might be an ideal
the evaluation, this model outperformed a single Q-learning situation in a real-time market, but their model can execute
approach and is highly effective for simulating and predicting effectively after backtesting. The two models both have the
stock trading in 2020. This study examines only the theoret- characteristics of high accuracy and return. Also, the models
ical aspects of stock market prediction, while on the practice have better revenue compared to the S&P 500 ETF-SPY.
side, traditional econometric models are combined with RL Farahani and Hajiagha [27] used ANN to predict stock
models that decrease prediction accuracy. The present model prices and trained their model by Social Spider Optimization
can be further developed by incorporating other potential ML (SSO) and Bat Algorithm (BA), which are called metaheuris-
models to inform investment decisions, forecast stock prices, tic algorithms. Then, they used the Genetic Algorithm (GA)
and create trading strategies in the real market. to select features and choose the best and most relevant indi-
Lee et al. [23] proposed a stock price prediction model cations based on the input’s technical characteristics. They
using LSTM and explored TWSE 0050, the most traded ETF chose the loss function as the evaluation standard to measure
on Taiwan’s stock exchange, as a case study. This research’s the model’s accuracy. In addition, they applied a time series
dataset is the period between Jan 2019 and Oct 2019. Using method to forecast the stock price and compared it to the
LSTM and available data from the Taiwan Stock Exchange model that included ANN-metaheuristic techniques. Finally,

80774 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

they selected indicators like the S&P 500, DAX, and FTSE Python to develop their framework and applied a decision
100 as their targets. tree classifier for the model training. They compared this
Sanboon et al. [28] proposed a long-term stock price pre- model to the existing models and proved its high accuracy and
diction model using LSTM. They observed that stock price outperformance. The readability of the results of this model
forecasting is one of the unachievable tasks in the current era. by humans, integrating semantic features to it, and using NN
Also, current forecasting methods cannot predict a stock’s for its development are directions for future work.
long-term price. Therefore, they cannot consider the relation- Ghosh et al. [32] provided a framework to compare the
ship between the prior price and the future price; hence this stock price before and after COVID-19. This model can
research has created a model capable of capturing long-term extract the characteristics of stocks, such as historical volatil-
stock price correlations. The accuracy of the suggested model ity, sectoral outlook, and market sentiment. The explanatory
was evaluated on five equities between January 2015 and features are screened using the suggested Ensemble Fea-
December 2017, and the results were compared. The experi- ture Selection (EFS) approach, which combines the Boruta
mental results show that the suggested model outperforms all and Regularized Random Forest (RRF) algorithms. For pre-
competing models. dicting, Regularized Greedy Forest (RGF) and DNN, two
Safari and Ghavifekr [29] proposed an AI-NN price pre- cutting-edge AI methods, are combined with Kernel Prin-
diction model tunned by Python with a security shell that cipal Component Analysis (KPCA) and Autoencoder (AE).
supports cryptocurrencies. This system can retain a copy of The results indicate that the proportionate importance of the
the final data in the user’s configuration email. Additionally, explanatory elements in futures price forecasting changes
it can predict the price of cryptocurrencies like Bitcoin (BTC) depending on the company and period under examination.
and Ethereum (ETH). They evaluated their idea and proved its Liu et al. [33] proposed a technique for stock ranking
outperformance. The advantage of this system is that it can forecasting by combining the Temporal Convolutional Net-
be extended to other applications, like education, as a future work (TCN) with a Channel-Time dual Attention Module
work. (CTAM) to improve the capability to manage dependencies
Chacón et al. [30] suggested a method for enhancing the within series. Stock industry features were also considered
predictability of financial time series by employing a full by creating an industry-stock Pearson correlation formula and
ensemble empirical modal decomposition with adaptive noise producing a vector that completely describes stock industry
and intrinsic sample entropy. They utilized their methodol- qualities using a matrix factorization algorithm. The effec-
ogy on the S&P 500 index stocks from January 2018 to tiveness of the proposed technique was demonstrated through
April 2020 to evaluate the integrated model. They trained experiments on three datasets. The Investment Return Ratio
his LSTM model for each stock’s closing price in the time (IRR) and SR were calculated based on the technique’s pre-
series to predict the next closing price. The results reveal a diction on the datasets.
relationship between the entropy of the decomposed signal Jang and Seong [3] created a deep RL portfolio opti-
and the prediction accuracy performance, which is seen in mization approach incorporating DL and modern portfolio
the decomposed signal when the short-term complexity of theory. They addressed the multimodal problem using Tucker
the financial time series is minimal compared to the series decomposition and technical analysis as inputs. Their method
energy. It suggests a higher predictive ability after removing surpassed cutting-edge algorithms regarding the SR, annual-
the signal entropy. The highest frequency is greatly improved. ized return, and maximum drawdown.
Moreover, the results show a 31% enhancement in stock price Lee et al. [34] believed that the strong foundation of the
direction prediction using the traditional LSTM architecture. stock market could positively affect the whole economy. As a
Carta et al. [31] proposed an ML method to address a result, they utilized DL and Technical Index to predict the
binary classification problem that aims to foresee the mag- movement of the stock in an abbreviated period. In this
nitude of the future stock price of businesses in the S&P paper, they select TWSE 0050 as their subject, the largest
500 index. A lexicon set is generated from articles published ETF trading volume in Taiwan’s stock exchange. By using
worldwide to identify the most influential words in the market opening price, closing price, and other technical indexes,
within a specific time interval and economic sector. A feature including KD, RSI, and BIAS, it can construct LSTM mod-
engineering process is then performed from the generated els and train them. As the outcomes show, the model’s
lexicon, and the resulting features are fed to a decision tree accuracy is 83.6%.
classifier. A forecasted label (high or low) indicates whether Peng et al. [35] examined the relationship between
the target company’s price will go over or below a specified AI-based Financial Management (FM) and Mineral Resource
threshold the next day. Evaluations of performance utiliz- Management (MRM) in the US economy from 1980-2020.
ing a walk-forward technique and a strong baseline indicate The findings suggested an asymmetric correlation between
that their approach greatly exceeds that of their opponents. FM and Mineral Resource Rent (MRR). Positive shocks of
In addition, the established AI technique is explicable in that FM were negatively related to MRR, while negative shocks
it analyzes the ‘‘white box’’ behind the classifier and gives were positively correlated. This work aims to implement
a set of reasons for the findings achieved. The authors used AI-based financial systems in the mining industry to improve

VOLUME 11, 2023 80775

 A. M. Rahmani et al.: Applications of AI in the Economy

MRM and assure worker performance and comfort, leading other important resources use regular linguistic processing
to longer growth. and assembly learning methods called random forest models
Tang et al. [36] proposed News Distilling Network (NDN) and SVM. The problem of predictability of storage costs
prediction model using NL and collaborative filtering to facil- is known as the problem of characterization to make better
itate similarity measurements. They combined news selection decisions.
mechanisms with NNL architecture to identify the linkage This paper presents a comparison table summarizing the
between stocks and news. They used real-world data to prove main features of research conducted in this area, including the
their framework’s efficiency. main context, applied techniques, evaluation methods, evalu-
Vicente et al. [37] proposed an RL market agents applica- ation metrics, benefits, limitations, and future work. We aim
tion to simulate an intelligent stock market. They examined to reach valuable results for future studies using the technical
the market maker’s approach from an agent-based viewpoint. and statistical interpretations resulting from comparing these
This study analyzes the proposed application’s performance factors.
in non-competitive and competitive situations. They also A summary of the most significant features covered in the
examined their strategy through various experiments and analyzed papers in this category is presented in Table 2.
described the impact of competitive environments on RL The following results are based on the interpretation from
agents’ performance. They demonstrated that RL and DRL comparing the analyzed papers in this section.
techniques are profitable market-making methods, allowing Overarching themes
people to understand their stock market behavior better. Pre- • Application of DL and NN
vious studies on financial news mainly refer to news coverage Designing trading strategies and predicting stock prices based
of the target financial instruments, which may be affected by on DL and NN architectures such as LSTM, CNN, and hybrid
sparse data. models.
Day and Lin [38] designed a robo-advisor using several
• Integration of RL
ML algorithms and DL prediction techniques. The advisor
can assist the user in making decisions by optimizing the Incorporating RL algorithms, such as Q-learning, with port-
investment portfolio. This research integrates various tech- folio management and stock trading strategies.
niques, including ML, data analysis, and investment portfolio • Predictive Models and Hybrid Approaches
optimization. In that case, they hoped they could predict the Enhancing the accuracy of stock price prediction by incor-
trend without using historical data or the investors’ perspec- porating AI techniques, such as wavelet analysis, sentiment
tives. Finally, they chose the algorithms with the highest analysis, technical indicators, and social media data.
accuracy, and the result shows that the Return on Investment Emerging trends
(RoI) is 12% annually. As for future work, they can try adding • Integrating alternative data sources
more technical indicators to train their model to extract the Utilizing non-traditional data sources to supplement histori-
most decisive characteristics and enhance their model. cal data, such as news sentiment, social media, and domain-
Xu and Tan [39] suggested that a risk-diversified, market- specific terminology.
neutral, and dynamic portfolio management model deal with
• Short-term and long-term predictions
systematic risk in portfolio management. They used matching
trading methods, deep RL, and standard portfolio manage- Long-term and short-term stock price predictions using
ment models to make this approach work. Their experiments LSTM.
on 32 stocks in the Chinese A-share market show that the • Explainable AI in stock trading
pairing-based deep portfolio model has the advantage of Understanding the reasoning behind trading decisions and
weighing investment returns and risks in dynamic portfolio developing interpretable models.
management problems. Critical gap
Chhajer et al. [40] presented a study that provided an • Real-time trading strategies
overview of using ML and AI as predictive analytics tools Real-time trading strategies by considering market condi-
for stock market forecasting. They reviewed the advantages tions’ rapid changes.
and disadvantages of ML in this subject and the benefits and
• Generalization of models
drawbacks of using modern technologies for this goal. The
study also focuses on the applicability of three distinct ML Developing AI models to generalize across a variety of finan-
technologies for stock market forecasting, namely ANNs, cial markets.
SVMs, and LSTMs. The study emphasized the importance • Model evaluation and robustness
of utilizing these technologies to make calculated predictions Robustness and reliability of AI-based trading models.
and invest safely in the turbulent stock market. Based on the analysis of the future work column content
Illa et al. [41] applied AI techniques to evaluate the infor- in Table 2, improving the accuracy of forecasting stock trad-
mation submitted to stock exchanges and developed new ing is one of the main concerns in using AI in this area.
techniques to predict equity costs and limit exposure to In most of these papers, developing the proposed predic-
interest on financial exchanges. Materials from reports and tion models is suggested in different ways, including using

80776 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 2. Selected articles of stock trading applications category.

diverse AI methodologies, utilizing a variety of big data, be relevant and important depending on the specific context
and applying additional evaluation criteria. A strong corre- and objectives of the AI-based stock trading system in the
lation exists between these future study directions, which described papers. The following table presents the evaluation
implies improving market predictions’ accuracy and validity. frequency of various criteria in several extensive groups.
A further indication that accuracy is a critical aspect of stock Based on their repetition, we can find the most critical mea-
trading is the frequent assessment of this metric in more than sures in AI-based stock trading.
half of the analyzed papers in this category, as shown in • Accuracy
Table 3. Consequently, it can be concluded that extending AI It refers to the correctness or precision of predictions made
applications to financial applications requires an increasing by an AI-based stock trading model. Precision, recall, and
level of correctness and validity of market volatility predic- F1-score are commonly used in classification tasks and
tions. The second key point worth mentioning is the lack of assessed for binary prediction accuracy. They can be useful
various evaluation criteria for assessing the proposed novel in classifying stock market movements or identifying trading
findings in this area of research. This comparison and anal- signals.
ysis demonstrate that the lack of consideration of different • Performance
evaluation criteria is one of the significant limitations of this While ‘‘performance’’ is a broad term, it encompasses various
study on AI applications in stock training. metrics that evaluate trading strategy effectiveness. Met-
According to the evaluation metrics in the previous table, rics such as R2, DA, PF, APV, SR, SAM, RSI, BIAS,
several metrics measure the same factors. These metrics may Williams%R, MACD, R, MAE, and MAPE can be considered

VOLUME 11, 2023 80777

 A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 2. (Continued.) Selected articles of stock trading applications category.

80778 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 2. (Continued.) Selected articles of stock trading applications category.

performance metrics as they assess different aspects of a • Investment-related metrics

trading model’s effectiveness.
Metrics like NP, PP, ARV, TheilsU, DPA, and IRR focus on
• Risk Management measuring financial gains or returns generated by a trading
Metrics such as robustness, MDD, IA, TI, and NSE are strategy.
indicators of risk management and a trading strategy’s ability According to the table above, accuracy and performance
to withstand adverse market conditions. are equally important and more significant than the other

VOLUME 11, 2023 80779

 A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 3. The evaluation factors for stock trading applications category.

dimensions of stock trading. In light of AI-based stock trad- Intermediate (WTI) dataset of crude oil prices spanning
ing, which predicts and optimizes stock price fluctuations 24 years provides the NN with new critical economic and
in an automated manner, the accuracy and performance of seasonal factors as input. The model has a five-dollar-per-
intelligent models are crucial evaluation factors. barrel accuracy limit for its weekly oil price predictions. The
experimental findings show that the achieved right prediction
B. MARKET ANALYSIS APPLICATIONS rate of 88% is greater than rates reported in other relevant
In this subsection, we analyze the application of AI in market research. The authors suggested that NNs may be better at
analysis in three subcategories, including technical analysis, predicting oil prices.
sentiment analysis, and market segmentation; each has sev- Al-Fattah [44] introduced a novel model for predicting the
eral subcategories. price of gasoline. This model uses AI, GA, ANN, and data
As a paper that focused on oscillators in the technical anal- mining techniques to predict the gasoline price. It monitors
ysis subcategory of market analysis applications, Lee [42] and documents the price fluctuations of gasoline. The result
proposed an innovative Chaotic Type-2 Transient-Fuzzy shows that the prediction matches the historical data. In addi-
Deep Neuro-Oscillatory Network (CT2TFDNN) system for tion to explaining the behavior and capturing the dynamics
long-term financial prediction in worldwide investment. The of oil-price volatility, the model demonstrated the ability to
CT2TFDNN extends the author’s original work on the chaotic predict the direction of changes in oil market volatility with
discrete-time neural oscillator with deep transient-chaotic an accuracy of 88%.
features, the Lee oscillator. An effective representation of an Vekaria et al. [8] proposed an economy prediction scheme
IT2FLS with a chaotic transient-fuzzy club function, effective that combines AI, Big Data Analytics (BDA), and the Inter-
time-collection community schooling, and forecasting of the net of Things (IoT) to promote economy-boosting activities
usage of a chaotic deep neuro-oscillatory community with called ξ boost. The scheme uses LSTM for early economic
retrograde signaling. CT2TFDNN no longer offers a rapid, prediction, particularly in the case of the COVID-19 pan-
chaotic fuzzy-neuro deep studying and predicting result but demic in India. The authors evaluated the prediction accuracy
also effectively determines the huge facts overtraining and of their scheme using Python and different economic datasets.
impasse problems, which might be normally imposed using They also compared it to other existing models based on
conventional Recurrent Neural Networks (RNN) with the MAPE and RMSE. Results showed that their scheme per-
use of classical sigmoid-primarily based activation functions. formed exceptionally well in predicting the economy and
CT2TFDNN comprises 2048 trading-day time-collection could be applied to other countries for future research.
monetary data and the top-10 fundamental monetary indica- Alonso de Armiño et al. [45] proposed a hybrid AI sys-
tors as fuzzy monetary indicators for the real-time prediction tem to examine how economic factors and transportation
of 129 global monetary products, including nine fundamental patterns interact. The temporal patterns of road traffic and
cryptocurrencies, 84 global currencies, 19 fundamental com- macroeconomic developments are studied using supervised
modities, and 17 global monetary indices. and unsupervised NNs and clustering algorithms. The sug-
Khashman and Carstea [43] described a powerful, super- gested approach is validated by connecting Spanish road
vised NN-based oil price prediction system. The West Texas transportation data and macroeconomic changes over six

80780 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 4. Selected articles of market analysis applications category.

VOLUME 11, 2023 80781

 A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 4. (Continued.) Selected articles of market analysis applications category.

years (2011–2017). Using data visualizations of the inter- This research shows a relationship between the news and the
connected relationships between road transportation patterns price of Bitcoin. Based on the news content, it has been dis-
and macroeconomic indicators, the results clarify the data’s tributed into 18 categories, including crime content related to
fascinating underlying structure. The results of the clustering cryptocurrencies, the economy, and the market. The analyzed
approaches similarly showed the same data structure. Several result shows some specific terms or words that will cause the
accurate predictions were generated by analyzing the road price movement of the market, mostly the Bitcoin market.
traffic data as a time series and forecasting the future values Moreover, the effect mostly happened within 24 hours after
of the primary series. These findings supported the predicted the news had been released.
connection between data on road transportation and macroe- Koch et al. [47] analyzed 34,209 news articles to inspect
conomic indices. the connection between stock market fluctuation and news
Coulter [46] deployed Natural Language Process to ana- sentiment. This study categorizes news sentiment into three
lyze the incidents of international news and Bitcoin price. groups: positive, negative, and neutral. Given the analyzed

80782 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 5. The evaluation factors for market analysis applications category.

result, when the whole economy is not in a stable scenario, the Smith and O’Hare [51] investigated the link between news
spillover effect of the news sentiment may assist the investors sentiment and tweet affection. Some of the CEOs will post a
from the institution in making decisions. Besides, the result tweet on social media. This research proved it does not have
also shows that although the directional spillover has a strong high relation to the stock price of its companies. However,
effect on the development of BREXIT, not in every scenario. news unrelated to the finical will not affect the stock price.
In the end, the impact of the news sentiment does not have a In contrast, the news sentiment of business is associated with
strong connection to the outcome of BREXIT. stock movements. Nonetheless, the strength of the association
Caporale et al. [48] investigated the connections between suggests that price movement drives sentiment, apart from
press indexes and portfolio flows. Meanwhile, this research the tumultuous economic times caused by the SARS-COV-2
sorts out a comprehensive factor based on all positive and pandemic in 2020.
negative news titles. The period is from 2007 to 2017. Yashina et al. [52] developed a finical asset analysis tool.
As a result, news indexes are the crucial factor influencing The point of this tool is to decrease and improve invest-
cross-border portfolio flows. ment strategies when crises come. Based on the mathematical
Barbaglia et al. [49] produced a method of extracting statistics method, the system can boost investment strategies
emotional information from the news based on the current by indexing fluctuation and economic crises. This system has
economic scenario. This method can analyze the emotion been evaluated by the dataset of the financial crises of Russia
from two aspects: the meaning of the word or based on the in 2008-2009, 2014-2015, and 2019-2020. With the algorithm
score corresponding with the specific terms. For the experi- of this system, you will be able to monitor the finical market
ment, they collected news data from six publishers in the US situation in real-time.
and a total of 6.6 million articles and 4.2 billion words. As the Shahzad et al. [53] investigated the relationship between
results show, multiple economy emotion indexes are highly AI enterprises and basic materials/oil & gas companies in
related to the fluctuation of the business cycle. In addition, Islamic markets. They discovered that AI was a net recip-
they also found that including the economic factors and emo- ient of shocks and that oil and gas-related companies were
tional reasons simultaneously can increase the accuracy of the the cause-in-quantiles. The authors proposed that COVID-19
prediction significantly. provided an opportunity to strengthen the involvement of AI
Yang et al. [50] built an undirected weight for news topics. innovations with basic materials and oil and gas companies.
The reason is that they found out that sentiment analysis for The findings have significance for AI application developers,
business has some unsolved issues. Such as, the DL system resource policymakers and managers, and investors interested
cannot identify jokes, stories, or words with two meanings. in advanced technologies. AI may be regarded as a crucial
To address this issue, they normalized the sentences in the link in the supply chain of basic materials and oil and gas
paper as nodes. As a result, when the system identifies the businesses.
articles, it will not be limited by a single word. Further- Wang and Zhao [54] proposed a prediction model based
more, to increase the accuracy of emotion analysis, they on AI and combining big data analytics. They believe that,
mentioned a sentiment analysis named BuSeD. The finding due to the limitations of human comprehension, the con-
indicated that the sentiment analysis tool has great economic ventional economic model is less accurate than the modern
importance. one. Now, the digital economy accomplishes the goal of

VOLUME 11, 2023 80783

 A. M. Rahmani et al.: Applications of AI in the Economy

economic connectivity and meticulous data sharing so that In the market analysis applications papers, several gaps are
the statistics of the economy and mathematical analytics will identified, which are proposed in the following:
be more accurate. Besides, with the help of AI, it can analyze Critical gaps
it more objectively and comprehensively. In their model, they • Industry-specific applications:
integrate a variety of variables, including potential political
Although many general market analysis papers are available,
issues, human activity elements, and social environment fac-
there is a lack of research on AI applications in specific
tors. The outcome shows that their developed model can be
industries, such as healthcare, energy, or transportation.
the basic economic statistics, analysis, and decision-making
• Real-time market monitoring:
model.
Xiangyan [55] examined how big data and AI technology Further investigation is required in developing real-time mar-
have affected economic forecasting and analysis. According ket analysis systems capable of adapting to dynamic market
to the typical point data paradigm, this study proposes a conditions and providing timely insights.
macroeconomic interval forecasting approach based on the Most applications of AI in market analysis are related to
stock market, fund market, and futures market data. After forecasting market trends and prices based on sentiments
hands-on results demonstrate that the macroeconomic inter- and other factors. Therefore, accuracy is the most significant
val prediction model can be reasonably fitted to the Shanghai metric for evaluating these models’ efficiency. The frequency
fund index, futures market transaction amount, Shenzhen of utilized evaluation metrics supports this conclusion.
component index, and narrow money supply in the interval
financial data. The findings of the experimental investiga- C. RISK MANAGEMENT APPLICATIONS
tion demonstrate that the suggested model performs well El Qadi et al. [56] focused on credit assessments and exe-
in predicting patterns of economic growth and that it may cuted benchmarks based on different ML models. The objec-
be applied to projecting upcoming economic development tive of this model is to predict whether the company will have
projections. a financial problem during the given time horizon. This model
This paper presents a comparison table summarizing the will help credit insurance companies decrease risks when
main features of research conducted in this area, including the providing company loans. The model can evaluate the possi-
main context, applied techniques, evaluation methods, evalu- bility that the company will break the contract. They give an
ation metrics, benefits, limitations, and future work. We aim expert-aligned feature relevance score that identifies the dis-
to reach valuable results for future studies using the technical parity between a credit risk expert and a model feature attribu-
and statistical interpretations resulting from comparing these tion explanation to quantify the convergence more effectively
factors. toward enhanced human-aligned decision-making.
A summary of the most significant features covered in Koo and Kim [57] found that hybrid models integrating
the analyzed papers in this category is presented in Table 4. ANNs and GARCH-type models have been developed. Still,
Furthermore, Table 5 shows the evaluation factors for this the fluctuating time series distribution is mainly concen-
category. trated at zero. The above reasons make the predictive per-
The following results are based on the interpretation from formance of the overall probability density function domain
comparing the analyzed papers in this section. low because the weights in the network are only trained to the
Overarching themes high-frequency region. To address this issue, they suggested a
• AI in economic predictions: novel hybrid model based on a nonlinear filtering method and
Predicting economic trends and market movements using AI a GARCH-type model to lower the volatility concentration
techniques like NNs, clustering, and big data analysis. feature. They applied a root-type function for the filtering,
and according to the study, the suggested hybrid model
• Sentiment Analysis:
(VU-GARCH-LSTM) obtained an efficiency gain of 21.03%
Using sentiment analysis in social media and events to assess in RMSE after comparing the data with the average effi-
the impact of the news on the market and make informed ciency of the current hybrid model combining LSTM and
investment decisions. GARCH-type models. In addition, the model enhances the
• AI Algorithm Integration: forecast performance in areas where the label density is likely
Enhancing the accuracy of market analysis by combining correct by predicting the distribution to be comparable to the
different AI algorithms. label distribution.
Emerging trends Gonzales and Hargreaves [58] aimed to address the chal-
• COVID-19 impact analysis:
lenge of decision-making in the stock market. They created
three distinct approaches to developing a stock recommender
Using AI to analyze the effects of COVID-19 on various system that focuses on the needs and interests of the investors.
sectors of the economy. They used hierarchical clustering to understand groups of
• Linking traditional and social media: traders with similar preferences, which improved compu-
The correlation between traditional media, social media, and tational efficiency. The K-Nearest Neighbor (kNN), Sin-
stock price. gular Value Decomposition (SVD), and Association Rule

80784 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 6. Selected articles of risk management applications category.

Mining (ARM) algorithms were investigated and tested. The and market durability. They discovered that ESG reputational
expected returns and value-at-risks were assessed to ensure risk negatively impacted company growth opportunities, lim-
profitability and support risk-informed decisions. The aver- iting market durability. The authors confirmed their empirical
age rate of return for the short-term, medium-term, and findings through several robustness checks, providing useful
long-term portfolios was 4.15 %, 10.24 %, and 23.17 %, insights for stakeholders and market participants.
respectively. Gangwar et al. [60] developed a hybrid simulation-
Fafaliou et al. [59] investigated the impact of Environ- optimization approach that combines scenario and risk
mental, Social, and Governance (ESG) reputational risk on analysis to optimize production plant scheduling under
the market durability of a sample of US-traded corporations volatile energy conditions. Despite traditional methods’
from 2007 to 2019. They used dynamic empirical analysis to mathematical complexity, the authors attempted to over-
evaluate the correlation between firms’ ESG reputational risk come their slowness and inapplicability. This approach was

VOLUME 11, 2023 80785

 A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 6. (Continued.) Selected articles of risk management applications category.

used to optimize the operation of a cryogenic air separation Companies’ credit risk analysis and banking risk manage-
plant under the uncertainty of electricity prices for short- ment optimization using AI and ML.
term scheduling. Based on MATLAB and a dataset collected • Forecasting Market Volatility: Utilizing AI-based
from a cryogenic air separation plant in Tarragona, Spain, volatility forecasting models for risk assessment and
they evaluated their model. This approach can be easily portfolio management.
adapted to scheduling facilities in several energy-intensive • Integration of AI in Manufacturing and Supply Chain:
industries, such as metallurgy, cement, or pulp and paper AI in risk management beyond the financial sector to
manufacturing. the manufacturing industries.
Ahmed et al. [61] measured bank risk using the Maha- Emerging trends
lanobis Distance (MD) and an Adaptive Neuro-Fuzzy Infer-
• ESG reputational risk:
ence System within five categories: Capital Adequacy,
Credit, Liquidity, Earnings Quality, and Operational Risk. Using AI to assess and manage reputational risk related to
Based on datasets and simulations with MATLAB, they ESG factors.
examined different risk indexes and the functional model’s • Digital Manufacturing risk analysis:
sensitivity. As a result, the Net Interest Margin Ratio (NIMR) Digital manufacturing processes risk analysis using inte-
and the Credit Adequacy Ratio (CAR) are the most significant grated AI, including supply chain management and
factors in determining bank risk. In contrast, the Provisional optimization.
Loan Ratio (PLR) is disappointing. Using ANNs and opti- Critical gaps
mization techniques is also beneficial in evaluating credit • Explainability and transparency:
risk.
A transparent and explainable risk management system is
Rodríguez-Espíndola et al. [62] developed a risk man-
needed for gaining trust and mitigating risks.
agement behavioral model combining AI, big data, cloud
• Real-time risk assessment:
computing, and blockchain technologies with institutional
theory. There is a need to develop real-time risk assessment
Table 6 shows a summary analysis of selected articles systems that can adapt to provide timely risk alerts
in this category. Furthermore, Table 7 shows the evaluation We present a grouping of metrics that measure similar factors
factors for this category. based on evaluation metrics used in AI-based risk manage-
This paper presents a comparison table summarizing the ment applications. The relevance and importance of these
main features of research conducted in this area, including the metrics will vary depending on the specific context and objec-
main context, applied techniques, evaluation methods, evalu- tives of the AI-based risk management application.
ation metrics, benefits, limitations, and future work. We aim • Accuracy: It generally refers to the correctness or preci-
to reach valuable results for future studies using the technical sion of predictions made by an AI-based risk manage-
and statistical interpretations resulting from comparing these ment model.
factors. • Performance: Metrics like RoR, Tobin’s Q, and perfor-
The following results are based on the interpretation from mance are associated with evaluating a risk management
comparing the analyzed papers in this section. strategy’s financial performance or profitability.
Overarching themes • Risk assessment: Metrics such as VaR, CVaR, Extreme
• Credit risk assessment: (extreme value analysis), Risk, RRI, SR, NIMR,

80786 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

TABLE 7. The evaluation factors for the risk management applications category.

and Kaiser-Meyer-Olkin (KMO) are indicators of risk economic forecasting, it also presents several challenges.
assessment and the ability of a risk management model Some challenges are:
to measure and quantify potential risks.
• Model evaluation metrics: Metrics such as RMSE, A. ACCURACY
MAE, AP@k (Average Precision at k), RMSEA, Good- AI models are often used to predict economic trends and
ness of Fit Index (GFI), Tucker-Lewis Index (TLI), make financial decisions. However, due to the complexity
Comparative Fit Index (CFI), and Normed X2 are used of the global economy, these models can be prone to errors
to assess the performance and accuracy of the risk man- and inaccuracies. AI models are based on algorithms and
agement model itself. often rely on incomplete or outdated data, which can lead
• Data analysis and ML metrics: Metrics like Correlation to inaccurate predictions. Additionally, AI models are often
Coefficient (CC), Area Under the Curve (AUC), kNN, unable to learn and adapt to changes in the economy as
SVD, ARM, Kolmogorov-Zurbenko Index (KZI), and quickly as needed, meaning they may not accurately predict
Worst-Case Weighted Importance (WWI) are associated future outcomes. Also, the accuracy of AI-based economic
with data analysis techniques, ML algorithms, or feature predictions depends on the data quality used to train the AI
importance. system. For the training data to be reliable, it needs to be up-
Based on repeated assessment criteria, accuracy is the to-date and reflect the current state of the economy. The AI
most significant criterion in this category. In addition, as the system will produce inaccurate results if the data is outdated
name of this category indicates, the criteria related to risk or incomplete [8].
measurement are more critical.
B. EXTERNAL IMPACTS
IV. DISCUSSION To make a prediction model in an AI application, we usually
The previous section reviewed the AI applications in three need historical data to train and test the model. We can also
categories of the economy, including stock trading, market modify or combine different algorithms to improve the accu-
analysis, and risk management. In this section, we answer the racy of the prediction. Still, external factors, such as financial
predefined RQs based on technical and statistical analysis and storms, corporate reform, negative news, and major disasters,
visually report the results. will affect the stock price. Therefore, we hope to discover
the cycles or patterns of these external factors that will affect
• RQ1: Which branches of AI are applied in economics?
stock prices [35].
According to the techniques that are involved in the analyzed
research papers, the AI branches applied in the economy can C. BIAS
be summarized as follows:
Another challenge of AI in the economy’s prediction is bias.
Notably, time series analysis is frequently used in the
AI models often rely on historical data to make predictions,
economy, not considered a subfield of AI, but a statistical
which can be biased. If a dataset contains a certain bias, this
analysis technique used to analyze and model time-dependent
can be reflected in the AI model’s predictions and lead to
data. Furthermore, ARIMA and SARIMAX can be applied to
inaccurate or unfair results. Additionally, AI models can be
AI applications, are not associated with a specific branch of
vulnerable to malicious actors who may manipulate the data
AI, and are widely used in statistical analysis and forecasting.
to manipulate the results [63].
Figure 4 shows the applied AI branches in the economy.
• RQ2: What are the challenges associated with AI appli- D. COMPLEXITY
cations in the economy? AI models often require a large amount of data to make accu-
AI is becoming more and more important to the economy. rate predictions. This data often needs to be collected from
This is because AI technologies make businesses more effi- multiple sources, which can be difficult and time-consuming.
cient and productive. While AI offers immense potential for Additionally, the AI models themselves can be complex and

VOLUME 11, 2023 80787

 A. M. Rahmani et al.: Applications of AI in the Economy

FIGURE 5. The AI techniques applied in the economy.

K. REAL-TIME RISK ASSESSMENT

Establishing systems for assessing risks continuously in
real-time and providing timely alerts.
FIGURE 4. Applied AI branches in the economy. It is essential to adjust to market changes, create effective
models across multiple markets, evaluate the performance
require a lot of computing power, which can be expensive and and reliability of these models, explore AI applications
difficult to obtain [10]. within specific industries, monitor markets in real-time,
Human Interaction: Another challenge of AI in the econ- ensure transparency in decision-making, and assess risks
omy’s prediction is that AI models cannot accurately predict accordingly.
human behavior. AI models often cannot accurately predict • RQ3: Which techniques are most commonly used in AI
how people will respond to certain events or react to cer- applications in the economy?
tain economic conditions. This means that AI models may We integrated AI techniques belonging to the same AI sub-
not accurately predict the effects of certain events on the fields to estimate the percentage of applications of each AI
economy [64]. branch in the economy. Figure 5 illustrates the portion of
applications in each category:
E. REAL-TIME TRADING STRATEGIES
The graph shows that NNs, with 35%, are the most com-
Development of strategies that can quickly adjust to market
monly used AI techniques in the economy, followed by ML
changes.
techniques, with 27%.
F. GENERALIZATION OF MODELS • RQ4: What are the most significant evaluation metrics
The development of AI models can work across various for AI applications in the economy?
financial markets. After investigating and analyzing selected papers, we found
the objectives that all articles focus on, among which accu-
G. MODEL EVALUATION AND ROBUSTNESS racy and correct prediction rate are the most concerned. The
Assuring that AI-based trading models are reliable and results are shown in Figure 6.
resilient to various scenarios. RQ5: What are the current open issues and future direc-
tions of AI in the economy?
H. INDUSTRY-SPECIFIC APPLICATIONS
Expanding research on AI applications in specific industries The most important open issues of AI for the economy can
like healthcare, energy, or transportation. be:
1. Increasing automation and job losses: AI is rapidly
I. REAL-TIME MARKET MONITORING becoming more common in the economy, leading to increased
Enhancing real-time market analysis systems and providing automation and job losses. AI technology is being used to
timely market insights. replace human labor in many industries, from manufacturing
to financial services. As AI technology continues to improve
J. EXPLAINABILITY AND TRANSPARENCY and become more widespread, more and more jobs will
Development of transparent risk management systems to likely be replaced by machines and robots. This will lead to
build trust and enable AI-driven decision-making to be increased inequality, as those with the skills to work with AI
understood. technology will reap the benefits, while those without those

80788 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

4. Ethical implications: AI can potentially revolution-

ize the economy but also presents important ethical issues.
A major concern is the impact on employment. Automation
could create economic displacement as machines increas-
ingly replace human labor, especially in the service sector.
This could lead to a significant rise in unemployment and
could exacerbate existing income inequality. Another ethical
issue is privacy. AI systems are increasingly used to collect
and analyze data, making it important to consider the impli-
cations of this data collection and what rights citizens have
to their data. Additionally, there is a need to ensure equitable
access to AI to share its benefits widely and not concentrated
in certain areas or groups. These ethical considerations must
be addressed to ensure that AI is used responsibly and for the
benefit of all [55].
FIGURE 6. The evaluation metrics applied AI in the economy.
5. Economic impact: AI has become a transformational
economic force. AI has had a major impact on the health-
skills may struggle to find work. Additionally, using AI can care, finance, logistics, manufacturing, and retail indus-
decrease wages for those employed, as machines can work tries. AI automates mundane tasks, optimizes processes, and
for less than humans and do not require the same benefits. predicts customer behavior. AI is also helping to improve
This can decrease overall wages for those employed in the customer service, increase efficiency, and reduce costs. AI is
economy, decreasing consumer purchasing power [52]. also helping to increase the accuracy of data analysis, which
2. Data privacy and security: AI has recently been can provide valuable insights that can lead to better decision-
increasingly implemented into the economy, and its poten- making. In addition, AI is being used to drive innovation and
tial for revolutionizing our business is immense. However, create new products and services. AI is changing the way
as with any rapidly evolving technology, certain risks are businesses operate, leading to increased customer satisfaction
associated with using AI. One of the most pressing con- and improved customer loyalty. AI also drives increased
cerns is data privacy and security. AI relies on large data productivity and efficiency, helping businesses remain com-
sets to make decisions, meaning that any data collected by petitive globally. AI is also helping to create new job opportu-
companies using AI must be carefully safeguarded from nities, as it opens up new fields and roles for human workers.
potentially malicious actors. Companies need robust data The overall impact of AI on the economy is profound, and its
security processes to ensure that any collected data is not potential is yet to be fully realized [56].
exposed to outside parties while allowing AI to function The following are some possible future directions for AI in
properly. Furthermore, companies should also consider how the economy:
they will protect their customers’ data and how they could 1. Cryptocurrency: The increasing popularity of cryp-
be held accountable for any potential misuse of this data. tocurrencies allows researchers to study AI’s impact on
By addressing these issues, companies can ensure that their digital currencies and the global economy. AI plays an impor-
use of AI is as secure and ethical as possible, allowing them tant role in predicting cryptocurrency prices through data
to benefit from this technology without compromising their analysis, pattern recognition, and ML [65]. It identifies cryp-
customers’ privacy [53]. tocurrency trading trends, predicts future prices, and detects
3. Regulatory challenges: Regulatory challenges, in par- fraud and money laundering. Here is an overview of how AI
ticular, are a major obstacle to the growth of AI in the is applied to cryptocurrency price predictions:
economy. AI-based systems are complex and highly dynamic, • Data analysis: AI algorithms, including price move-
making it difficult to create effective laws that govern their ments, trading volumes, market sentiment, and other
use. Furthermore, due to the global nature of AI, it is not factors, can analyze a wide range of historical cryptocur-
easy to create laws that are both effective and applicable rency data. AI models can identify patterns, correlations,
across multiple jurisdictions. Privacy and data security are and anomalies in cryptocurrency prices.
also major concerns when using AI, as the large amounts • Pattern Recognition: Identifying patterns in complex
of data collected can be used to identify individuals, track data sets is one of the most valuable abilities of AI.
their movements, and influence their behavior. Finally, the Several AI algorithms can identify patterns in historical
potential for AI to create or displace jobs is a key area of price charts, such as support and resistance levels, chart
concern, as introducing AI technologies could drastically patterns, and other technical indicators. These patterns
reduce the number of jobs available in certain fields. Gov- can provide insights into future price movements.
ernments must create regulations that protect workers from • ML: Predictive models can be developed using AI
displacement and ensure that the benefits of AI are distributed models that learn from historical cryptocurrency data.
fairly [54]. It is possible to establish relationships between various

VOLUME 11, 2023 80789

 A. M. Rahmani et al.: Applications of AI in the Economy

factors and the movement of cryptocurrency prices 5. ML algorithms for trading: ML algorithms can be
through training ML algorithms on vast datasets. These used to develop trading strategies that automatically adjust
models can predict based on new data inputs and chang- to changing market conditions. This can potentially increase
ing market conditions [66]. profits and reduce risk [68].
• Sentiment analysis: AI-powered sentiment analysis 6. Risk management: AI can be used to analyze data
techniques analyze social media posts, news articles, and related to financial risks, such as credit risk, market risk, and
online cryptocurrency discussions. AI models can iden- operational risk. This can help financial institutions manage
tify potential price influences by gauging the sentiment risk more effectively and make better decisions about lending
of market participants. As a result of this analysis, short- and investing [69].
term market sentiment can be predicted, and its potential
impact on price trends can be evaluated. V. REAL CASES
• Hybrid approaches: Many cryptocurrency price predic- Here, we provide several real cases to support the AI appli-
tion models combine multiple AI techniques. In some cation scenarios discussed in this paper:
models, data is incorporated from various sources, • Renaissance Technologies,1 or RenTech, is an Ameri-
including market data, news sentiment, social media can hedge fund and investment management firm based
sentiment, and macroeconomic indicators. Combin- in East Setauket, New York, specializing in systematic
ing multiple AI approaches, these hybrid models trading using quantitative models derived from mathe-
are intended to provide more accurate and robust matical and statistical analysis in the design and execu-
predictions. tion of its investment programs.
2. Exchange rate fluctuations: AI plays a significant • Virtu Financial,2 a leading high-frequency trading firm,
role in analyzing and predicting exchange rate fluctuations. is a notable case study in high-frequency trading which
The proposed taxonomy shows that AI can predict exchange is an American company that provides financial ser-
rate fluctuations through data analysis, predictive modeling, vices, trading products, and market-making services.
sentiment analysis, news and event impact analysis, and algo- Virtu employs sophisticated AI algorithms to analyze
rithmic trading. It is important to note that various complex vast market data and execute trades within milliseconds.
factors affect exchange rate fluctuations, including economic • Trade Ideas3 uses AI-based pattern recognition algo-
indicators, political events, market sentiment, and global rithms to identify trading opportunities. As a result of
economic trends. Exchange rates remain subject to many their algorithms, users receive real-time trading alerts
unpredictable factors, despite AI’s ability to provide valu- based on technical patterns, such as moving averages,
able insights. Therefore, AI supports decision-making and support and resistance levels, and chart patterns.
improves forecasting accuracy but does not guarantee exact • Wealthfront4 is a robo-advisor platform that utilizes AI
predictions of exchange rate movements. This challenge is a algorithms to optimize portfolio composition and sug-
potential direction for in-depth research. gest optimal asset allocations and rebalancing strategies
3. Accounting fraud detection: AI can detect account- to help clients reach their financial objectives. It con-
ing fraud using advanced algorithms and data analysis siders risk tolerance, investment goals, and historical
techniques. The following are some ways in which AI can performance.
detect accounting fraud: • AlphaSense5 provides financial forecasting and predic-
tive analytics using AI. Using a wide range of data,
• Anomaly Detection
including financial statements, news articles, and indus-
• Pattern Recognition
try reports, AlphaSense assists users in making accurate
• NLP
predictions about the earnings of companies, the move-
• Data Integration and Analysis
ment of stock prices, and the direction of the market.
• Predictive Analytics
• MarketPsych6 specializes in market sentiment analysis.
• Network Analysis
They use AI algorithms to process vast amounts of news
• Continuous Monitoring
articles, social media posts, and other textual data to
These AI-driven techniques can be combined with human gauge market sentiment. Their sentiment indicators give
expertise in forensic accounting and fraud investigation to traders and investors insights into the market’s mood.
enhance the ability to detect accounting fraud, mitigate risks, • MSCI Barra7 is a prominent provider of market segmen-
and protect financial integrity. Accounting fraud detection tation solutions that uses AI techniques to classify stocks
and prevention efforts are significantly strengthened by AI’s
1 https://www.rentec.com/Home.action?index=true
ability to process vast amounts of data, detect subtle patterns,
2 ://www.virtu.com/
and provide timely alerts. 3 https://www.trade-ideas.com/
4. NLP: With NLP, AI can analyze news articles, financial 4 https://www.wealthfront.com/
reports, and other documents related to the economy. This can 5 https://www.alpha-sense.com/
help researchers identify emerging trends and make informed 6 https://www.marketpsych.com/home
decisions about investment opportunities [67]. 7 https://www.msci.com/

80790 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

and assets into various indices and segments based on REFERENCES

factors such as industry classifications, geographical [1] T. Cui, S. Ding, H. Jin, and Y. Zhang, ‘‘Portfolio constructions in cryp-
regions, and market capitalizations. By using these seg- tocurrency market: A CVaR-based deep reinforcement learning approach,’’
Econ. Model., vol. 119, Feb. 2023, Art. no. 106078.
mented indices, investors can monitor the performance
[2] Q. Y. E. Lim, Q. Cao, and C. Quek, ‘‘Dynamic portfolio rebalancing
of specific market segments. through reinforcement learning,’’ Neural Comput. Appl., vol. 34, no. 9,
• PayPal,8 an online payment platform, utilizes AI algo- pp. 7125–7139, May 2022.
rithms for fraud detection. These algorithms analyze [3] J. Jang and N. Seong, ‘‘Deep reinforcement learning for stock portfolio
optimization by connecting with modern portfolio theory,’’ Expert Syst.
transaction data, user behavior patterns, and other fac- Appl., vol. 218, May 2023, Art. no. 119556.
tors to identify and prevent fraud, ensuring financial [4] S. A. Saeidi, F. Fallah, S. Barmaki, and H. Farbeh, ‘‘A novel neuromorphic
transaction security. processors realization of spiking deep reinforcement learning for portfolio
management,’’ in Proc. Design, Autom. Test Eur. Conf. Exhib. (DATE),
• LendingClub,9 a peer-to-peer lending platform, uses AI Mar. 2022, pp. 68–71.
algorithms to assess loan applicants’ creditworthiness. [5] V. M. Ngo, H. H. Nguyen, and P. Van Nguyen, ‘‘Does reinforcement learn-
Using AI to analyze credit histories, income data, and ing outperform deep learning and traditional portfolio optimization models
other factors, LendingClub’s system determines the risk in frontier and developed financial markets?’’ Res. Int. Bus. Finance,
vol. 65, Apr. 2023, Art. no. 101936.
associated with each loan request, allowing them to [6] H. Li, ‘‘Application of artificial intelligence control system in economic
make informed lending decisions. management,’’ in Proc. Int. Conf. Comput. Technol., Electron. Commun.
• BlackRock10 is a leading investment management firm (ICCTEC), Dec. 2017, pp. 933–938.
that uses AI-based risk management tools to assess [7] Y. Chen, ‘‘Based on neural network prediction and decision-making meth-
ods in environmental economy application research in the system,’’ in Proc.
market risks. Their system analyzes market indicators, IEEE Int. Conf. Artif. Intell. Ind. Design (AIID), May 2021, pp. 100–103.
economic data, and portfolio positions to identify poten- [8] D. Vekaria, A. Kumari, S. Tanwar, and N. Kumar, ‘‘ξ boost: An AI-based
tial risks. It also conducts stress tests on investment data analytics scheme for COVID-19 prediction and economy boosting,’’
IEEE Internet Things J., vol. 8, no. 21, pp. 15977–15989, Nov. 2021.
portfolios, enabling proactive risk management. [9] M. Zhang, ‘‘Artificial intelligence and application in finance,’’ in Proc.
11th Int. Conf. E-Educ., E-Bus., E-Manag., E-Learn., 2020, pp. 317–322.
VI. CONCLUSION [10] L. Cao, Q. Yang, and P. S. Yu, ‘‘Data science and AI in FinTech: An
overview,’’ Int. J. Data Sci. Anal., vol. 12, no. 2, pp. 81–99, Aug. 2021.
In conclusion, AI has the potential to transform the economy
[11] J. Shah, D. Vaidya, and M. Shah, ‘‘A comprehensive review on multiple
by increasing efficiency, reducing costs, and generating new hybrid deep learning approaches for stock prediction,’’ Intell. Syst. Appl.,
sources of value. It has the potential to create new oppor- vol. 16, Nov. 2022, Art. no. 200111.
tunities for those who are willing to embrace it and have [12] L. N. Mintarya, J. N. M. Halim, C. Angie, S. Achmad, and A. Kurniawan,
‘‘Machine learning approaches in stock market prediction: A systematic
a significant impact on the way businesses operate. As AI literature review,’’ Proc. Comput. Sci., vol. 216, pp. 96–102, Jan. 2023.
technology continues to evolve and become more accessible, [13] D. Sheth and M. Shah, ‘‘Predicting stock market using machine learning:
it is essential to recognize the potential consequences of Best and accurate way to know future stock prices,’’ Int. J. Syst. Assurance
Eng. Manage., vol. 14, no. 1, pp. 1–18, Feb. 2023.
its widespread adoption. Governments and businesses must
[14] Y.-T. Chang and N.-H. Fan, ‘‘A novel approach to market segmentation
consider optimal ways to manage and regulate AI to ensure selection using artificial intelligence techniques,’’ J. Supercomput., vol. 79,
it is utilized responsibly and ethically. At the same time, no. 2, pp. 1235–1262, Feb. 2023.
it is essential to recognize the potential for AI to create new [15] Y. Zhang, S. Ye, J. Liu, and L. Du, ‘‘Impact of the development of FinTech
by commercial banks on bank credit risk,’’ Finance Res. Lett., vol. 55,
economic opportunities and open up new collaboration and Jul. 2023, Art. no. 103857.
innovation avenues. By understanding the implications of [16] H. Pallathadka, E. H. Ramirez-Asis, T. P. Loli-Poma, K. Kaliyaperumal,
AI and developing the necessary policies and regulations, R. J. M. Ventayen, and M. Naved, ‘‘Applications of artificial intelligence
the economy can reap the benefits of AI without sacrificing in business management, e-commerce and finance,’’ Mater. Today: Proc.,
vol. 80, pp. 2610–2613, 2023.
human dignity and social justice. This paper examines several [17] B. J. Rothschild, J. S. Ault, and S. G. Smith, ‘‘A systems science approach
application scenarios, such as stock market predictions, oil to fisheries stock assessment and management,’’ in Stock Assessment. Boca
price analysis and forecasting, and research on how AI may Raton, FL, USA: CRC Press, 2023, pp. 473–492.
[18] D. Liang, Y. Xu, Y. Hu, and Q. Du, ‘‘Intraday return forecasts and high-
impact the global economy. Using DL, numerous research frequency trading of stock index futures: A hybrid wavelet-deep learning
articles predict the direction of the economy. High accu- approach,’’ Emerg. Markets Finance Trade, vol. 59, no. 7, pp. 2118–2128,
racy rates are achieved in all employed models; some are 2023.
even efficient or offer other benefits. Furthermore, the article [19] G. Cohen, ‘‘Intraday algorithmic trading strategies for cryptocurrencies,’’
Rev. Quant. Finance Accounting, vol. 61, pp. 395–409, May 2023.
demonstrates that numerous techniques are applied to the [20] X. Kan, M. Miao, L. Cao, T. Xu, Y. Li, and J. Jiang, ‘‘Stock price prediction
data, with DL carrying the most weight. Finally, we hope to be based on artificial neural network,’’ in Proc. 2nd Int. Conf. Mach. Learn.,
focused on applications of cryptocurrency price predictions, Big Data Bus. Intell. (MLBDBI), Oct. 2020, pp. 182–185.
[21] S. S. Kuma, D. Chitradevi, and S. Rajan, ‘‘Stock price prediction using
exchange rate fluctuations in macro and microeconomics, and deep learning LSTM (long short-term memory),’’ in Proc. 2nd Int. Conf.
detecting accounting fraud illegally performed by companies Adv. Comput. Innov. Technol. Eng. (ICACITE), Apr. 2022, pp. 1787–1791.
in the future. [22] P. Zhao, S. Yin, B. V. D. Kumar, and T. J. Yew, ‘‘Improving quantitative
stock trading prediction based on MAD using Q-learning technology,’’ in
Proc. Int. Conf. Artif. Intell. Smart Commun. (AISC), Jan. 2023, pp. 57–60.
8 https://www.paypal.com/
[23] M.-C. Lee, J.-S. Liao, S.-C. Yeh, and J.-W. Chang, ‘‘Forecasting the short-
9 https://www.lendingclub.com/
term price trend of Taiwan stocks with deep neural network,’’ in Proc. 11th
10 https://www.blackrock.com/ca Int. Conf. E-Educ., E-Bus., E-Manag., E-Learn., 2020, pp. 296–299.

VOLUME 11, 2023 80791

 A. M. Rahmani et al.: Applications of AI in the Economy

[24] P. Kumar, S. Adhikari, P. Agarwal, and A. Sahoo, ‘‘Stock prices predic- [45] C. A. D. Armiño, M. Á. Manzanedo, and Á. Herrero, ‘‘Analysing the
tion based on social influence & historic data,’’ in Proc. 14th Int. Conf. intermeshed patterns of road transportation and macroeconomic indicators
Contemp. Comput., Aug. 2022, pp. 276–283. through neural and clustering techniques,’’ Pattern Anal. Appl., vol. 23,
[25] D. Selvamuthu, V. Kumar, and A. Mishra, ‘‘Indian stock market prediction no. 3, pp. 1059–1070, Aug. 2020.
using artificial neural networks on tick data,’’ Financial Innov., vol. 5, no. 1, [46] K. A. Coulter, ‘‘The impact of news media on Bitcoin prices: Modelling
pp. 1–12, Dec. 2019. data driven discourses in the crypto-economy with natural language pro-
[26] V.-D. Ta, C.-M. Liu, and D. Addis, ‘‘Prediction and portfolio optimization cessing,’’ Roy. Soc. Open Sci., vol. 9, no. 4, Apr. 2022, Art. no. 220276.
in quantitative trading using machine learning techniques,’’ in Proc. 9th [47] A. Koch, T. L. D. Huynh, and M. Wang, ‘‘News sentiment and international
Int. Symp. Inf. Commun. Technol., 2018, pp. 98–105. equity markets during BREXIT period: A textual and connectedness anal-
[27] M. S. Farahani and S. H. R. Hajiagha, ‘‘Forecasting stock price using ysis,’’ Int. J. Finance Econ., early access, May 2022.
integrated artificial neural network and metaheuristic algorithms compared [48] G. M. Caporale, F. Menla Ali, F. Spagnolo, and N. Spagnolo, ‘‘Cross-
to time series models,’’ Soft Comput., vol. 25, no. 13, pp. 8483–8513, border portfolio flows and news media coverage,’’ J. Int. Money Finance,
Jul. 2021. vol. 126, Sep. 2022, Art. no. 102638.
[28] T. Sanboon, K. Keatruangkamala, and S. Jaiyen, ‘‘A deep learning model [49] L. Barbaglia, S. Consoli, and S. Manzan, ‘‘Forecasting with economic
for predicting buy and sell recommendations in stock exchange of Thailand news,’’ J. Bus. Econ. Statist., vol. 41, no. 3, pp. 708–719, 2022.
using long short-term memory,’’ in Proc. IEEE 4th Int. Conf. Comput. [50] M. Yang, B. Jiang, Y. Wang, T. Hao, and Y. Liu, ‘‘News text mining-based
Commun. Syst. (ICCCS), Feb. 2019, pp. 757–760. business sentiment analysis and its significance in economy,’’ Frontiers
[29] A. Safari and A. A. Ghavifekr, ‘‘International stock index prediction using Psychol., vol. 13, Jul. 2022, Art. no. 918447.
artificial neural network (ANN) and Python programming,’’ in Proc. 7th [51] S. Smith and A. O’Hare, ‘‘Comparing traditional news and social media
Int. Conf. Control, Instrum. Autom. (ICCIA), Feb. 2021, pp. 1–7. with stock price movements; which comes first, the news or the price
[30] H. D. Chacón, E. Kesici, and P. Najafirad, ‘‘Improving financial time change?’’ J. Big Data, vol. 9, no. 1, pp. 1–20, Dec. 2022.
series prediction accuracy using ensemble empirical mode decomposition [52] N. Yashina, O. Kashina, S. Yashin, N. Pronchatova-Rubtsova, and
and recurrent neural networks,’’ IEEE Access, vol. 8, pp. 117133–117145, I. Khorosheva, ‘‘Digital methods of technical analysis for diagnosis of
2020. crisis phenomena in the financial market,’’ Int. J. Technol., vol. 13, no. 7,
[31] S. M. Carta, S. Consoli, L. Piras, A. S. Podda, and D. R. Recupero, pp. 1403–1411, 2022.
‘‘Explainable machine learning exploiting news and domain-specific lexi- [53] U. Shahzad, M. G. Asl, M. Panait, T. Sarker, and S. A. Apostu, ‘‘Emerging
con for stock market forecasting,’’ IEEE Access, vol. 9, pp. 30193–30205, interaction of artificial intelligence with basic materials and oil & gas
2021. companies: A comparative look at the Islamic vs. conventional markets,’’
[32] I. Ghosh, T. D. Chaudhuri, E. Alfaro-Cortés, M. Gámez, and N. García, Resour. Policy, vol. 80, Jan. 2023, Art. no. 103197.
‘‘A hybrid approach to forecasting futures prices with simultaneous consid- [54] L. Wang and L. Zhao, ‘‘Digital economy meets artificial intelligence:
eration of optimality in ensemble feature selection and advanced artificial Forecasting economic conditions based on big data analytics,’’ Mobile Inf.
intelligence,’’ Technol. Forecasting Social Change, vol. 181, Aug. 2022, Syst., vol. 2022, Oct. 2022, Art. no. 7014874.
Art. no. 121757. [55] P. Xiangyan, ‘‘Prediction algorithm of digital economy development trend
[33] H. Liu, T. Zhao, S. Wang, and X. Li, ‘‘A stock rank prediction method based on big data,’’ Math. Problems Eng., vol. 2022, pp. 1–10, Apr. 2022.
combining industry attributes and price data of stocks,’’ Inf. Process. [56] A. E. Qadi, M. Trocan, N. Díaz-Rodríguez, and T. Frossard, ‘‘Feature
Manage., vol. 60, no. 4, Jul. 2023, Art. no. 103358. contribution alignment with expert knowledge for artificial intelligence
[34] M.-C. Lee, J.-W. Chang, J. Hung, and B.-L. Chen, ‘‘Exploring the effec- credit scoring,’’ Signal, Image Video Process., vol. 17, no. 2, pp. 427–434,
tiveness of deep neural networks with technical analysis applied to stock Mar. 2023.
market prediction,’’ Comput. Sci. Inf. Syst., vol. 18, no. 2, pp. 401–418, [57] E. Koo and G. Kim, ‘‘A hybrid prediction model integrating GARCH
2021. models with a distribution manipulation strategy based on LSTM networks
[35] X. Peng, S. Mousa, M. Sarfraz, N. Abdelmohsen A, and M. Haffar, for stock market volatility,’’ IEEE Access, vol. 10, pp. 34743–34754, 2022.
‘‘Improving mineral resource management by accurate financial man- [58] R. M. D. Gonzales and C. A. Hargreaves, ‘‘How can we use artificial
agement: Studying through artificial intelligence tools,’’ Resour. Policy, intelligence for stock recommendation and risk management? A proposed
vol. 81, Mar. 2023, Art. no. 103323. decision support system,’’ Int. J. Inf. Manage. Data Insights, vol. 2, no. 2,
[36] T.-H. Tang, C.-C. Chen, H.-H. Huang, and H.-H. Chen, ‘‘Retrieving Nov. 2022, Art. no. 100130.
implicit information for stock movement prediction,’’ in Proc. 44th Int. [59] I. Fafaliou, M. Giaka, D. Konstantios, and M. Polemis, ‘‘Firms’ ESG
ACM SIGIR Conf. Res. Develop. Inf. Retr., Jul. 2021, pp. 2010–2014. reputational risk and market longevity: A firm-level analysis for the United
[37] Ó. F. Vicente, F. F. Rebollo, and F. J. G. Polo, ‘‘Deep Q-learning market States,’’ J. Bus. Res., vol. 149, pp. 161–177, Oct. 2022.
makers in a multi-agent simulated stock market,’’ in Proc. 2nd ACM Int. [60] S. Gangwar, D. Fernández, C. Pozo, R. Folgado, L. Jiménez, and D. Boer,
Conf. AI Finance, Nov. 2021, pp. 1–9. ‘‘Scheduling optimization and risk analysis for energy-intensive industries
[38] M.-Y. Day and J.-T. Lin, ‘‘Artificial intelligence for ETF market prediction under uncertain electricity market to facilitate financial planning,’’ Com-
and portfolio optimization,’’ in Proc. IEEE/ACM Int. Conf. Adv. Social put. Chem. Eng., vol. 174, Jun. 2023, Art. no. 108234.
Netw. Anal. Mining (ASONAM), Aug. 2019, pp. 1026–1033. [61] I. E. Ahmed, R. Mehdi, and E. A. Mohamed, ‘‘The role of artificial
[39] F. Xu and S. Tan, ‘‘Dynamic portfolio management based on pair trading intelligence in developing a banking risk index: An application of adaptive
and deep reinforcement learning,’’ in Proc. 3rd Int. Conf. Comput. Intell. neural network-based fuzzy inference system (ANFIS),’’ Artif. Intell. Rev.,
Intell. Syst., Nov. 2020, pp. 50–55. early access, pp. 1–23, Apr. 2023.
[40] P. Chhajer, M. Shah, and A. Kshirsagar, ‘‘The applications of artificial [62] O. Rodríguez-Espíndola, S. Chowdhury, P. K. Dey, P. Albores, and
neural networks, support vector machines, and long–short term mem- A. Emrouznejad, ‘‘Analysis of the adoption of emergent technologies for
ory for stock market prediction,’’ Decis. Anal. J., vol. 2, Mar. 2022, risk management in the era of digital manufacturing,’’ Technol. Forecasting
Art. no. 100015. Social Change, vol. 178, May 2022, Art. no. 121562.
[41] P. K. Illa, B. Parvathala, and A. K. Sharma, ‘‘Stock price prediction [63] A. Taeihagh, ‘‘Governance of artificial intelligence,’’ Policy Soc., vol. 40,
methodology using random forest algorithm and support vector machine,’’ no. 2, pp. 137–157, 2021.
Mater. Today, Proc., vol. 56, pp. 1776–1782, Jan. 2022. [64] A. Arslan, C. Cooper, Z. Khan, I. Golgeci, and I. Ali, ‘‘Artificial intelli-
[42] R. S. T. Lee, ‘‘Chaotic type-2 transient-fuzzy deep neuro-oscillatory net- gence and human workers interaction at team level: A conceptual assess-
work (CT2TFDNN) for worldwide financial prediction,’’ IEEE Trans. ment of the challenges and potential HRM strategies,’’ Int. J. Manpower,
Fuzzy Syst., vol. 28, no. 4, pp. 731–745, Apr. 2020. vol. 43, no. 1, pp. 75–88, Apr. 2022.
[43] A. Khashman and C.-G. Carstea, ‘‘Oil price prediction using a supervised [65] P. Giudici, ‘‘FinTech risk management: A research challenge for artificial
neural network,’’ Int. J. Oil, Gas Coal Technol., vol. 20, no. 3, pp. 360–371, intelligence in finance,’’ Frontiers Artif. Intell., vol. 1, p. 1, Nov. 2018.
2019. [66] B. Rezazadeh, P. Asghari, and A. M. Rahmani, ‘‘Computer-aided meth-
[44] S. M. Al-Fattah, ‘‘Artificial intelligence approach for modeling and fore- ods for combating COVID-19 in prevention, detection, and service pro-
casting oil-price volatility,’’ SPE Reservoir Eval. Eng., vol. 22, no. 3, vision approaches,’’ Neural Comput. Appl., vol. 35, pp. 14739–14778,
pp. 817–826, Aug. 2019. May 2023.

80792 VOLUME 11, 2023

A. M. Rahmani et al.: Applications of AI in the Economy

[67] T.-I. Theodorou, A. Zamichos, M. Skoumperdis, A. Kougioumtzidou, MAJID HAGHPARAST (Senior Member, IEEE)
K. Tsolaki, D. Papadopoulos, T. Patsios, G. Papanikolaou, is currently a Researcher with the Faculty of
A. Konstantinidis, A. Drosou, and D. Tzovaras, ‘‘An AI-enabled stock Information Technology, University of Jyväskylä,
prediction platform combining news and social sensing with financial Finland. He is an Associate Editor of the Clus-
statements,’’ Future Internet, vol. 13, no. 6, p. 138, May 2021. ter Computing (Springer) and Journal of Com-
[68] D. Lv, S. Yuan, M. Li, and Y. Xiang, ‘‘An empirical study of machine putational Electronics (Springer). He is also an
learning algorithms for stock daily trading strategy,’’ Math. Problems Eng., Editorial Board Member of Optical and Quantum
vol. 2019, pp. 1–30, Apr. 2019.
Electronics journal (Springer).
[69] D. Mhlanga, ‘‘Financial inclusion in emerging economies: The application
of machine learning and artificial intelligence in credit risk assessment,’’
Int. J. Financial Stud., vol. 9, no. 3, p. 39, Jul. 2021.

AMIR MASOUD RAHMANI received the B.S. WEI-CHE CHANG received the bachelor’s degree
degree in computer engineering from Amirkabir in information management from the National
University, Tehran, in 1996, the M.S. degree in Yunlin University of Science and Technology,
computer engineering from the Sharif University in 2022. He is currently pursuing the master’s
of Technology, Tehran, in 1998, and the Ph.D. degree in artificial intelligence with the Interna-
degree in computer engineering from Islamic Azad tional Graduate Institute. His research interests
University (IAU), Tehran, in 2005. His research include big data and the Internet of Vehicles.
interests include the Internet of Things, cloud/fog
computing, and evolutionary computing.

BAHAREH REZAZADEH received the B.S.

degree in information technology engineering SHEN GUAN TING received the associate degree
from the Urmia University of Technology, in 2013, in information management from the National
the M.S. degree in MBA from the University Taipei University of Business, in 2020, and the
of Tehran, in 2015, and the second master’s bachelor’s degree in information management
degree in computer engineering from the Sci- from the National Yunlin University of Science
ence and Research Branch, Islamic Azad Univer- and Technology, in 2022, where he is currently
sity (IAU), Tehran, in 2022. She is currently a pursuing the degree with the International Grad-
Senior Researcher with IAU. Her research inter- uate Institute of Artificial Intelligence.
ests include the Internet of Things, distributed
computing, and artificial intelligence.

VOLUME 11, 2023 80793