RMIT student leads breakthrough research in deep learning for stock price forecasting

RMIT student leads breakthrough research in deep learning for stock price forecasting

Nguyen Quoc Anh, an RMIT Economics and Finance student, has tackled stock price forecasting by using advanced deep learning models.

His research has been accepted for publication and presented at prestigious international conferences, ranked by the Computing Research and Education Association of Australasia (CORE).  

By late May, Vietnam saw a record-breaking 7.94 million in stock trading accounts. 

The stock market is influenced by various factors, such as human psychology, macroeconomic and microeconomic conditions, and monetary policies. As a result, the financial sector has become unpredictable for traders seeking stable profits at low-risk levels. 

Is there any indicator that can mitigate risks in financial trading? This question sparked Quoc Anh's interest in algorithmic trading research. It is still nascent in Southeast Asia due to less adaptable trading environments and lower liquidity compared to North America or Europe. Therefore, machine learning and deep learning models have not advanced significantly in this region. 

“Although my major is in Economics and Finance, I found myself drawn to the practical applications of computer science. This could be a potential breakthrough in economic science, addressing long-standing research questions. 

“In finance, while time series data is abundant, the approaches remain traditional,” Quoc Anh said.  

Under the supervision of RMIT lecturers in the Blockchain enabled business program Dr Ha Xuan Son and Dr Thai Trung Hieu, Quoc Anh has led a paper titled “Phase space reconstructed neural ordinary differential equations model for stock price forecasting”. This research was accepted for publication in the Association for Information Systems (AIS) eLibrary. It was also presented at the 31st Pacific Asia Conference on Information Systems (PACIS) 2024, ranked A, on 4 July at The Reverie Saigon, Ho Chi Minh City. 

Quoc Anh presented at the PACIS 2024. Quoc Anh presented at the PACIS 2024.

In this paper, Quoc Anh proposed a novel application of Neural ordinary differential equations (neural ODE or NODE) in stock price forecasting. NODE could also be understood as a sophisticated deep learning model for predictive modelling. 

“Overall, I aim to test the model’s forecasting ability based on the daily stock prices of six US companies from the technology, finance, and pharmaceutical sectors from 2003 to 2023,” Quoc Anh said. 

The model followed a 70:20:10 split, with 70 per cent of the data allocated to training, 20 per cent for finetuning/validation, and the remaining 10 per cent for testing.  

“To leverage the model’s forecasting ability, I incorporated a data pre-processing technique called ‘Phase space reconstruction’ (PSR) from the Chaos theory to transform the original ‘Open, High, Low, Close, Volume’ (OHLCV) stock data into a multi-dimensional space.”  

By doing this, NODE could learn more meaningful underlying patterns that exist in traditional OHLCV sets. Quoc Anh fine-tuned NODE across ten different parameters, testing over 40 values to find the optimal settings.  

“Next, NODE is compared against six different state-of-the-art baselines, including RNN, Transformer, SVR, LSTM, CNN, and CNN-LSTM.  

Quoc Anh’s stock price forecasting model Quoc Anh’s stock price forecasting model

“Finally, NODE demonstrates superior forecasting accuracy compared to its closest competitor, LSTM, in predicting long-term values with minimal deviations over 1000-time steps, reducing errors by more than 70 per cent for each stock,” said Quoc Anh. 

Dr Ha Xuan Son said the model addresses the limitations of traditional deep learning methods in capturing complex, non-linear stock market dynamics. 

“It demonstrates superior long-term forecasting accuracy and effectively captures sudden market shifts like flash crashes.  

“Beyond stock forecasting, the model shows promise in predicting other chaotic systems, as evidenced by tests on Lorenz and Mackey-Glass datasets,” said Dr Son. 

Quoc Anh encountered various challenges during his research, with technology proficiency being the most demanding. 

"I had to self-learn languages like Python, SQL, LaTeX, utilise tools such as GitHub, MongoDB, Lightning AI, retrieve data from APIs, and master the skill of searching for and comprehending research, primarily through technology forums or YouTube tutorials,” said Quoc Anh. 

Dr Son said Quoc Anh possesses a natural aptitude for academic inquiry, showing keen analytical skills and a talent for innovative problem-solving. 

“His ability to grasp complex concepts quickly and apply them creatively sets him apart as a promising young scholar,” said Dr Son. 

In addition to “Phase space reconstructed neural ordinary differential equations model for stock price forecasting”, Quoc Anh is the lead author of a paper titled “Transforming stock price forecasting: Deep learning architectures and strategic feature engineering”. This paper will be published in Springer’s Lecture Note of Artificial Intelligence (LNAI) and presented at the 21st International Conference on Modelling Decisions for Artificial Intelligence (MDAI) 2024, hosted at Meiji University, Japan. 

Quoc Anh was a presenter at the Digital3 conference at RMIT Vietnam in 2023. Quoc Anh was a presenter at the Digital3 conference at RMIT Vietnam in 2023.

In the future, Quoc Anh plans to explore innovative applications in data science, delve deeper into the significance of time series data, and develop automation models that address diverse societal needs, ranging from pattern recognition to early cancer detection. 

The third-year RMIT student emphasised “I want to encourage my peers to learn hand-in-hand with action. There's always more to learn, especially in the digital transformation era.”

Story: June Pham

26 July 2024

Share

  • Research and Innovation
Engage

Related news

sugarcane-bagasse-pickled-mustard-greens-and-salted-whole-anchovies-helped-rmit-researchers-uncover-native-yeasts-with-the-potential-to-produce-sustainable-edible-oils.jpg

Vietnam's tiny yeasts show big promise for sustainable edible oils
Science and technology
Icon / Small / CalendarCreated with Sketch. 20 Oct 2026

What do sugarcane bagasse, pickled mustard greens and salted anchovies have in common? They have helped RMIT researchers discover promising oil-producing yeasts.

ethnic-minority-students-took-part-in-practical-training-designed-to-build-their-ai-skills.jpg

MOET and RMIT Vietnam launch AI skills program for ethnic minority students
Campus and community
Icon / Small / CalendarCreated with Sketch. 19 Oct 2026

Vietnam’s Ministry of Education and Training (MOET) and RMIT Vietnam launched an AI skills development program for more than 5,400 ethnic minority students across eight higher education institutions.

Woman looking at a background depicting a digital brain and AI

Can AI truly forget, or does it just pretend to?
Science and technology
Icon / Small / CalendarCreated with Sketch. 16 Oct 2026

Amidst rising AI adoption in Vietnam, research from RMIT University reveals that systems may appear to forget sensitive data but can still recover it under pressure, raising urgent questions for data privacy.

Two older women dancing on a paved path in a park surrounded by trees and sculptures.

Vietnam is ageing fast. Are we ready?
Communication and design
Icon / Small / CalendarCreated with Sketch. 14 Oct 2026

RMIT Vietnam researcher Catherine Earl found that many ASEAN countries, including Vietnam, are ageing at a pace that could see them "grow old before they grow rich”.

Load more +