Trends and Methods in Stock Price Forecasting

Xiayang Sun

International Business School Suzhoug, Xi'an Jiaotong-Liverpool University, Jiangsu Province, China

Keywords: Stock Predictions, Machine Learning, Deep Learning, Technical Analysis, Financial Market Predictions.

Abstract: Accurate stock market forecasting is increasingly vital in today’s fast-paced financial environment, as it

directly impacts economic stability and investment decisions. This paper provides an overview of various

stock prediction methods, addressing both traditional techniques and modern advancements in artificial

intelligence. It explores the shortcomings of fundamental and technical analyses, which rely heavily on

historical data, and contrasts these with innovative machine learning and deep learning approaches that better

capture complex market patterns. The review covers key models such as long short-term memory (LSTM)

and convolutional neural networks (CNN), as well as hybrid methods that enhance prediction accuracy.

Challenges such as market unpredictability, data quality issues, and the interpretability of AI-driven models

are examined. By analyzing these methods from multiple perspectives, this paper identifies future

opportunities for improving prediction effectiveness, suggesting advancements in computational efficiency

and the inclusion of alternative data sources. This research underscores the importance of continually evolving

prediction techniques to meet the demands of dynamic financial markets.

1 INTRODUCTION

The stock market originated from the Dutch East

India Company in the 17th century, which was the

first company in the world to publicly issue stocks

and set up an exchange. As the Industrial Revolution

progressed, stock exchanges in London and New

York were gradually established, and the stock

market began to become globalized. Since the 20th

century, technological progress has stimulated the

rapid growth of electronic transactions in global asset

markets. However, the 2008 financial crisis exposed

the fragility of the market. Nowadays, more

convenient methods such as big data analysis and

artificial intelligence applications are profoundly

changing stock predictions and trading strategies. For

example, Wall Street uses quantitative trading models

to significantly improve the accuracy of investment

decisions.

The importance of stock forecasting in financial

markets is reflected at both the macro and micro

levels. At the macro level, stock forecasts provide an

important basis for the formulation of economic

policies and macroeconomic regulation. Through

market forecasts, governments and financial

https://orcid.org/0009-0003-0317-9042

institutions can gain insight into economic trends in

advance and formulate effective monetary and fiscal

policies to avoid economic overheating or recession.

Accurate market forecasts help ensure the stability of

financial markets and maintain the healthy

development of the economy. For example, during a

financial crisis, accurate forecasts can help

policymakers take quick action to reduce economic

losses. Companies can predict market trends and

reasonably arrange financing and investment plans,

thereby optimizing resource allocation and improving

corporate competitiveness. For investors, stock

forecasts directly affect their investment returns and

risk management. Research shows that machine

learning-based stock prediction models greatly

increase the efficiency of marketing forecasts. For

example, research conducted by Fischer and Krauss

(2018) using long short-term memory (LSTM)

networks showed that this method performed well in

stock price prediction and was able to provide more

accurate market expectations. In addition, research by

Sezer et al. (2020) also pointed out that deep learning

architecture has significant advantages when dealing

with the problem of predicting the accuracy of stock

Sun, X.

Trends and Methods in Stock Price Forecasting.

DOI: 10.5220/0013214600004568

In Proceedings of the 1st International Conference on E-commerce and Artiﬁcial Intelligence (ECAI 2024), pages 269-273

ISBN: 978-989-758-726-9

269

return, bringing higher returns and better risk

management to investors.

This paper will review stock prediction from the

following perspectives: First, it will explore the basic

analysis and technical analysis methods of the stock

market, including the application of fundamental

analysis and technical indicators. This section will

explain the principles and limitations of traditional

prediction methods. Secondly, it will focus on the

channels of machine learning and deep learning

techniques over stock forecasting, and evaluate their

effects and advantages in market prediction by

comparing various algorithm models. Thirdly, it will

discuss the main challenges faced by stock prediction,

such as data uncertainty, market volatility, and model

overfitting. Finally, it will proactively plan the

direction of stock forecasts and explore how to further

optimize the effectiveness and practicality of

anticipations through technological innovation. It

aims at providing a dialectical understanding of the

ongoing status of stock prediction and points out the

direction for future research. Although machine

learning models have shown a good performance in

stock forecasting, their application also has some

limitations. First, current forecasting models tend to

rely too much on historical data, which makes them

perform poorly in the face of sudden market changes.

In addition, at the macro level, financial market

participants are not yet deeply aware of the use of new

technologies and tools, which has led to a certain lag

in the popularization and application of these

technologies. On the other hand, although big data

and artificial intelligence technologies can provide

more market insights, there are still challenges in the

precision, completeness and timeliness of the data. In

the future, how to improve the interpretability and

applicability of these technologies remains a pressing

issue.

2 TRADITIONAL METHODS

There exist two common types of stock market

analysis methods: fundamental analysis and technical

analysis. These two methods have their own focuses

and can help investors understand and forecast future

stock trends from different perspectives.

2.1 Fundamental Analysis

Fundamental analysis seeks to forecast a company's

future stock performance by assessing its underlying

value. This method focuses on the company's

financial status, industry prospects, management

capabilities and macroeconomic factors. Specifically,

fundamental analysis includes the following key

aspects. First is financial analysis, which evaluates

the profitability, liquidity and debt-paying ability of a

business with the assistance of having a deep vision

of the entity's balance sheet, income and cash flow

statement. These financial indicators provide

investors with the company's current financial health

and future profit potential. The second is industry

analysis, which evaluates the development trend of

the industry, the market competition situation and the

impact of technological innovation when it is

integrated with company activities. By understanding

the firm's positioning or competitive advantages in

the industry, investors can more accurately predict its

long-term performance. The third is macroeconomic

analysis, which analyzes the effect of macroeconomic

elements such as financial cycles, monetary policies,

and fiscal policies on corporate operations. Changes

in the macroeconomic environment often directly or

indirectly affect the company's operating conditions

and stock prices. The advantage of fundamental

analysis is that it can provide value judgments for

long-term investments. However, this method has

certain limitations when dealing with short-term

market swings. Apart from that, due to the asymmetry

of information acquisition, the analysis results may be

biased.

2.2 Technical Analysis

Technical analysis is based on historical stock prices

and capacity data, using graphs and technical metrics

to forecast approaching stock price flows. Technical

analysis accepts that almost every market insight is

reflected in stock values, so by analyzing the pattern

of rate changes, investors can predict future price

trends. Common technical indicators are shown

below: The first one is the Moving Average (MA) by

calculating the average value of stock prices during a

certain duration of time, the processing average can

ease price volatility and help identify trend directions.

Commonly used include Simple Moving Average

(SMA) and Exponential Moving Average (EMA).

The second one is the Relative Strength Index (RSI),

which measures the rise and fall of stock prices and

helps identify overbought or oversold situations in the

trade. RSI values are usually between 0 and 100.

When RSI is above 70, the market has the possibility

of overbought, and when it is below 30, the market

may be oversold. The third one is Bollinger Bands,

Bollinger Bands contain three lines, including a

center line and two upper and lower track lines, which

are used to measure market volatility. When the price

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

270

touches the upper or lower track, it usually means that

the market may reverse. The fourth one is MACD

(Moving Average Convergence Divergence), a trend-

tracking momentum indicator used to reflect the

relationship between short-term and long-term price

changes, helping investors identify buy or sell signals.

The advantage of technical analysis is that it can

quickly respond to market changes and cooperate

well with short-term traders. However, technical

analysis counts too much on previous data and may

not accurately predict sudden events or abnormal

movements in the market.

3 DEEP LEARNING METHODS

In modern financial markets, predicting stock prices

has always been the focus of investors and

researchers. With the advancement of artificial

intelligence applied science, deep learning has

behaved excellent potential in stock market

prediction. Compared with traditional prediction

methods, deep learning can effectively extract

features from a large amount of complex financial

data and capture the nonlinear patterns of market

behavior.

3.1 Introduction to Deep Learning

Methods

Deep learning has gradually come to be one of the

dominant methods for stock market prediction due to

its powerful modeling capabilities. Typical deep

learning architectures, such as LSTM and

convolutional neural networks (CNN), have been

frequently used in time series prediction missions.

Hochreiter and Schmidhuber (1997) first proposed

the LSTM network. This model solves the gradient

vanishing problem faced by traditional neural

networks when operating long time series data by

introducing memory units, and has been widely used

in stock value forecasting. Subsequently, Bao et al.

(2017) proposed a new financial time series

prediction framework by combining LSTM with

stacked autoencoders (SAE), which effectively

improved the prediction accuracy.

At the same time, K. Kim and H. Kim (2019)

proposed a model that combines LSTM and CNN,

which significantly improved the prediction

performance by fusing features from different data

representations. Similarly, Eapen et al. (2019) studied

a novel CNN and bidirectional LSTM hybrid model,

and its excellent performance in stock market index

prediction proved the efficiency of the model.

3.2 Performance of Deep Learning on

Some Problems

LSTM is one of the most commonly used deep

learning methods, and it is especially good for

operating time series data. According to the study by

Goenka et al., the prediction error of the LSTM

network was reduced by about 18% compared with

the random forest-based method in processing time

series prediction of stock values (Goenka, 2024).

Specifically, in the prediction task of day trading data,

the mean absolute percentage error (MAPE) of the

LSTM model dropped from 14% to 11.5%. This

shows that LSTM can effectively capture the long-

term dependence on stock prices and improve

prediction accuracy.

A hybrid deep recurrent neural network (RNN) is

also a widely used model. In their research,

Karahasan et al. combined a simple exponential

smoothing mechanism with RNN, and the results

showed that its prediction stability under extreme

market conditions increased by about 25%

(Karahasan, 2024). Specific data shows that the

hybrid model ’ s forecast error during extreme

market fluctuations is reduced by 2.5%, while the

error of the traditional RNN model is 3.3%. This

shows that the hybrid model can provide more stable

prediction results when dealing with high-volatility

market data.

In the study of Islamic financial markets, the

optimized dynamic dense graph convolutional

network model proposed by Dey et al. performed well

in a specific market environment, and the prediction

accuracy increased by about 13% (Dey, 2024).

Research data shows that the model ’ s accuracy

increased from 72% to 81% when predicting stock

price trends in the Islamic financial market,

demonstrating its adaptability and predictive

capabilities in different market environments.

In addition, Gil et al. evaluated the performance

of multiple deep learning models in detail and

proposed an improved data preprocessing method,

which increased the accuracy of model predictions by

an average of 8% (Gil, 2024). Their research showed

that by improving the data preprocessing strategy, the

prediction error of the deep learning model was

reduced from the original 0.021 to 0.019, thus

effectively improving the overall performance of the

model.

3.3 Advantages and Limitations of

Deep Learning

Although deep learning has many advantages in stock

Trends and Methods in Stock Price Forecasting

271

price forecasting, the method still has certain

limitations. Gu et al. (2020) pointed out that although

deep learning has advantages in capturing nonlinear

relationships in the market, the impact of its model

complexity and data noise on the prediction results

cannot be ignored. In addition, Fischer and Krauss

(2018) showed that deep learning models may show

instability when dealing with extreme market

fluctuations, which poses a challenge to the accuracy

of predictions.

These studies show that although deep learning

has broad application prospects in stock market

prediction, its limitations also remind us to be

cautious when applying these methods and combine

them with traditional analysis ways to boost the

trustworthiness and consistency of predictions.

4 MACHINE LEARNING

METHODS

Machine learning methods have sparked a revolution

in stock market predictions. Traditional models such

as linear regression and decision trees have been

replaced or supplemented by more advanced machine

learning algorithms, including random forests,

support vector machines (SVM), and ensemble

methods. These practices significantly enhance the

effectiveness and dependability of stock forecasts by

learning complex patterns from historical data.

In one study, Luo et al. used linear regression and

random forest models to forecast the closing value of

Google stock. The results show that the prediction

accuracy of the random forest is about 15% bigger

than the linear regression model (Luo, 2024).

Specifically, the random forest model got an accuracy

of 85% within a one-month prediction window, while

the linear regression model showed an accuracy of

70%. This shows that when processing complex stock

market data, the random forest model can capture

market trends more effectively and provide higher

prediction accuracy.

Another machine learning method that performs

well is extreme gradient boosting (XGBoost). In the

study by Goenka et al., the prediction accuracy

increased by about 20% by combining the integration

method of XGBoost and LSTM network models

(Goenka, 2024). Research shows that in standard time

series prediction tasks, the aggregation of XGBoost

and LSTM significantly reduces the mean square

error (MSE) from 0.023 for a single model to 0.018.

This combined method can capture more subtle

market changes and improve the durability of the

model.

In addition, the Transformer model also performs

well in stock price prediction. Zhang proposed an

improved Transformer model in his research, which

reduces the prediction error by about 12% compared

to the traditional LSTM model when dealing with

complex market dynamics (Zhang, 2024). Specific

data shows that after using the improved Transformer

model, the prediction accuracy increases to 88%,

while the traditional LSTM model is 76%. This shows

that the Transformer model can better handle data in

non-linear and dynamic market environments and

improve the accuracy of predictions.

5 CHALLENGES AND FUTURE

DIRECTIONS

Despite the great potential of stock market prediction,

there are still many challenges in practical

application. These challenges include data

complexity, market volatility, model overfitting, and

interpretive issues. In addition, with the advancement

of technology, the field of stock prediction has also

shown new development directions.

5.1 Data Complexity and Uncertainty

Stock market data are usually high-dimensional,

noisy, and heterogeneous, which brings challenges to

the construction of predictive models. The complex

and diverse behaviors of market participants result in

data full of noise and outliers, which may increase

forecast errors by up to 15% (Gil, 2024). In addition,

the heterogeneity of data makes general models

perform poorly when applied across markets or

industries, and improving the robustness of the model

is the key to solving this problem (Karahasan, 2024).

This section must be in two columns.

5.2 Market Volatility and Model

Overfitting

Market volatility is another significant challenge, and

the nonlinearity and unpredictability of market prices

make models susceptible to overfitting. By

introducing regularization technology and integration

methods, such as XGBoost and random forest, the

overfitting phenomenon can be effectively reduced,

and the prediction error is reduced by about 10%

(Dey, 2024).

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

272

5.3 Interpretive Issues and

Transparency

Despite their excellent performance in predictions,

the black box attribute of deep learning models

remains a major issue. Market participants need to

know the model’s decision-making procedure to

make informed decisions. Through visualization

technology and attention mechanism, the

interpretability of the model can be improved and

investors' trust in the prediction results can be

enhanced (Gil, 2024).

5.4 Sub Subsection Titles

In the future, stock forecasts will move in the

following directions. Comprehensive forecasts across

markets and asset classes focus on developing

universal models capable of handling multiple

markets and asset classes to provide more

comprehensive market forecasts. Higher computing

efficiency and real-time prediction try to apply

efficient algorithms to provide real-time prediction

results. Also, enhanced interpretability and

transparency contribute crucially to improving the

interpretability of the model and increasing the

practical application of the technology. Last is the

utilization of emerging data sources, it effectively

utilizes social media and big data for stock prediction

and develops new data processing technologies. The

domain of stock market forecasting is in a stage of

rapid development, and although there are many

challenges, it is also full of opportunities. Future

research will continue to explore the robustness,

interpretability, and computational efficiency of the

model to achieve more accurate and interpretable

market forecasts.

6 CONCLUSIONS

In conclusion, this paper has explored the importance

and methodologies of stock market forecasting,

examining both traditional and modern approaches.

Stock prediction holds a critical role in financial

markets, aiding in policy formulation, corporate

decision-making, and investment strategies.

Predicting stock movements helps in minimizing

risks and optimizing returns, making it essential for

financial stability. However, there are limitations to

current models, especially when faced with sudden

market fluctuations, and their dependence on

historical data. Additionally, the complexity of AI-

driven models poses challenges in interpretation and

real-time applicability. Future developments should

prioritize computational efficiency, merging

alternative data sources, and enhancing the

transparency of AI models. This will ensure better

adaptability and more accurate stock forecasts,

aligning with the changing nature of economic

markets.

REFERENCES

Bao, W., Yue, J., Rao, Y., 2017. A deep learning framework

for financial time series using stacked autoencoders and

long short-term memory. PLoS ONE, 12(7), e0180944.

Dey, R., Kassim, S., Bhagwat, S., Mahajan, RA., 2024.

Stock Market Prediction Using Optimized Dynamic

Dense Graph Convolutional Network for Islamic

Financial Market. SSRN.

Eapen, P., Rajan, S., Chandy, T., 2019. A hybrid model for

stock price prediction using Long Short Term Memory

and Convolutional Neural Network. International

Journal of Advanced Science and Technology, 28(15),

1-10.

Fischer, T., Krauss, C., 2018. Deep learning with long

short-term memory networks for financial market

predictions. European Journal of Operational Research.

Gil, GL., Duhamel-Sebline, P., McCarren, A., 2024. An

Evaluation of Deep Learning Models for Stock Market

Trend Prediction. arXiv preprint arXiv:2408.12408.

Goenka, A., 2024. Developing an Integrated Smart Model

to Enhance the Efficacy of Stock Market Prediction by

Leveraging XGBoost and Long Short-Term Memory

Networks. International Journal of Technology and

Business Management.

Gu, S., Kelly, B., Xiu, D., 2020. Empirical Asset Pricing

via Machine Learning. Review of Financial Studies.

Hochreiter, S., Schmidhuber, J., 1997. Long Short-Term

Memory. Neural Computation, 9(8), 1735-1780.

Karahasan, O., Bas, E., Egrioglu, E., 2024. A Hybrid Deep

Recurrent Artificial Neural Network with a Simple

Exponential Smoothing Feedback Mechanism.

Information Sciences. Elsevier.

Kim, K., Kim, H., 2019., Forecasting stock prices with a

feature fusion LSTM-CNN model using different data

representations. Expert Systems with Applications,

103, 25-37.

Luo, Z., 2024. The Prediction of Google Stock Closing

Price Based on Linear Regression Model and Random

Forest Model. Scitepress.

Sezer, O. B., Gudelek, M. U., Ozbayoglu, A. M., 2020.

Financial time series forecasting with deep learning: A

systematic literature review: 2005–2019. Applied Soft

Computing, 90, 106181.

Zhang, H., 2024. Stock Price Forecast Based on Improved

Transformer. Google Books.

Trends and Methods in Stock Price Forecasting

273