Specialist artificial intelligence models tailored to finance are outperforming mainstream tools from technology giants such as Google and Amazon, according to new research.
A project led by Dr Eghbal Rahimikia of Alliance Manchester Business School, working with researchers from University College London and Shanghai University, has produced more than 600 AI models designed specifically for time series forecasting. This is the analysis of historical, time‑stamped data to predict future trends.
Their work has resulted in the development and rollout of more than 600 AI models programmed specifically for time series forecasting, which is when historical and time-stamped data is analysed for predicting the future.
Known as Time Series Foundation Models, these AI tools have existed for several years. But existing models have been broader in scope, making predictions for not just financial forecasting but also areas like customer demand and energy consumption.
Rahimikia and his research team have, instead, focused on developing AI models dedicated to financial forecasting and trained them using only financial data to ensure predictions wouldn’t be generalist.
Using data from 94 countries and the Isambard-AI supercomputer, the biggest of its kind in the UK, Rahimikia and his colleagues spent 18 months and a total of 50,000 GPU hours developing, training, testing and improving the models.
The next stage of this project saw Rahimikia’s research team compare the specialist AI models with commercially available versions developed by tech giants to see which fared better when it came to financial forecasting.
From this comparison, they concluded that models programmed solely for financial forecasting were capable of outperforming the bigger, more well-known models.
The researchers have since published their models – and the methodology used to develop them – online, allowing financial global professionals and organisations to improve their forecasting and portfolios.
Since the release of the models a few months ago, they’ve amassed more than 8,000 downloads and are currently being used by experts in academia and the financial services industry.
This isn’t the first time that Rahimikia has explored the impact of AI on finance in his research work. In a previous study titled “Re(Visiting) Large Language Models in Finance”, he looked at how custom-made, specialist AI tools could take on more well-known models being used by financial professionals.
“Prediction in financial markets is crucial for both the day-to-day management of firms, as well as analysing wider macro-economic trends. Through this study, we wanted to explore ways to advance the science of prediction and translate our academic research into real-world applications to accelerate the UK’s fintech AI capabilities,” he said.
“By showing the feasibility of developing specialised AI models like this, we can not only reduce costs for financial institutions and businesses but also improve their forecasting abilities. Better predictions mean better resource allocation, smarter investments, more resilient institutions and a stronger economy.”
