Reduces processing time from 48 hours to two hours for trades
Improves hedging model performance by 1-5%
Outperforms the buy-and-hold strategy by 28.75%
Vertex AI platform allows for transparent development
UOB Asset Management (UOBAM) sought to create new products to further enhance investment outcomes for customers, leveraging Vertex AI to build a reliable hedging model for better results.
UOB Asset Management (UOBAM) sought to create new products to further enhance investment outcomes for customers, leveraging Vertex AI to build a reliable hedging model for better results.
Every advantage counts in the financial markets, and for United Overseas Bank Asset Management (UOBAM), utilizing data and algorithms to deliver better investment outcomes for its customers has always been a key part of the process.
Starting in 2019, UOBAM began its journey of incorporating artificial intelligence (AI) into its projects, led by the new Invest Tech team. This group is tasked with running a series of proof-of-concepts that would lead to new product launches from which investors can benefit.
Given the platform to learn more about generative AI use cases with Google Cloud, UOBAM was provided the opportunity to explore the deep learning functionalities and capabilities of Vertex AI. Having previously developed a proprietary hedging AI model that had not yet been deployed, the team decided to put the technology to the test when it comes to expanding AI capabilities in hedging processes. This test bed would reveal how AI could further enhance returns.
“We were particularly keen to understand whether AI techniques could be applied to reliably predict financial market movements, and thereby generate above-benchmark returns” says Paul Ho, Senior Director of Asia Equities at UOBAM.
A framework was developed to define the boundaries, mechanisms, and objectives of the AI model and process, separating them from the other performance enhancement tools.
We were particularly keen to understand whether AI techniques could be applied to reliably predict financial market movements, and thereby generate above-benchmark returns.
Paul Ho
Senior Director of Asia Equities, UOBAM
An AI model that delivers promising results
The team started with providing 1,000 features for Vertex AI to work with, down from 2,000 compared to the original model. This allowed the platform to improve model efficiency by analyzing data and suggesting investment strategies faster.
Historical financial planning data and other market indicators, like interest rates and economic trends, were also taken into consideration. Approximately one trade was also generated every fortnight.
For this model, success was defined as the ability to determine the most impactful set of data, improve performance by generating better hedging strategies, quicker processing time, and improved explainability with increased transparency. The results were favorable in all areas.
Processing time was improved significantly, dropping from 48 hours per trade to two hours. At the same time, the performance of the hedging model registered a 1.86% improvement on the F1 score, plus a 4.77% improvement in its recall score, bringing the total recall score to 92.11%.
Google Cloud demonstrated that it could improve our hedging model further and enable advanced hedging capabilities. This has set us on a path to potentially continuing our work as we explore the features of its AI tools and learn how they can be applied to wider performance enhancement needs.
Paul Ho
Senior Director of Asia Equities, UOBAM
Incorporating the technology into the existing AI and machine learning model development lifecycle was also beneficial, producing results that demonstrated a clear improvement in the model based on all the significant metrics. The returns of the improved model successfully outperformed the buy-and-hold strategy by 28.7%, with the model adding an extra 9.9% increase over the original which is an improvement of 52.55%, a significant advancement over the existing model’s returns.
When translated to a live portfolio, these results can potentially deliver close to an extra $10 million for every $100 million deployed. It also suggests that, with the necessary commitment and the right tools, it is possible to continue to refine and upgrade existing models.
“Google Cloud demonstrated that it could improve our hedging model further and enable advanced hedging capabilities. This has set us on a path to potentially continuing our work as we explore the features of its AI tools and learn how they can be applied to wider performance enhancement needs,” says Ho.
The performance and most importantly, the results, have proven useful in helping UOBAM continue on its journey of AI augmentation. The team continues to explore the possibility of broadening and deepening the returns outperformance of the model, provided it stays ahead of the curve while being compliant with internal and external governance policies.
Established in 1986, UOB Asset Management (UOBAM) offers individual and institutional investors in Asia a comprehensive suite of products from retail unit trusts and exchange-traded funds to customized portfolio management services for institutional clients.
Industry: Financial Services
Location: Singapore
Products: Vertex AI
