In FX trading, artificial intelligence (AI) is the most potentially disruptive technology for predictive analysis. However, when creating an AI application, there is always the problem of gaps in the data and technology. How has a joint venture between Refinitiv and the Bank of China addressed these challenges, and moved an AI innovation from idea to reality in a short space of time?
- Refinitiv and the Bank of China have collaborated to launch an innovative AI application through Eikon for FX trading signal prediction that uses the power of machines and human knowledge to achieve its objectives.
- Refinitiv data is fundamental in the field of AI innovation. This helps in the creation of AI applications, where the starting point is choosing the correct datasets and combing them to maximize alpha.
- Eikon’s open platform enables customers to build their own apps and rapidly share FinTech innovations to the global community. Refinitiv’s APIs also enable a seamless content integration with the customer’s system, which supports their innovations.
For more data-driven insights in your Inbox, subscribe to the Refinitiv Perspectives weekly newsletter.
Using AI innovation in FX trading has been on people’s minds for quite some time. However, it has now become a more practical proposition because of advances in big data and machine learning (ML).
FX traders are increasingly using these advances as the basis for predictive analysis.
The Bank of China has run FX trading for more than 70 years. Despite using deep-learning algorithms to predict the FX price movement for only a couple of years, the Bank’s Digital Asset Management Department has made significant progress.
The challenges of AI innovation and FX trading
There are four major challenges when using AI innovation with FX trading:
Data: It’s very challenging to figure out what kind of data and which data combinations would be the most appropriate when building an FX trading AI model, and also how to acquire a high quality data source at right time. In the machine learning (ML) world, this is known as ‘feature engineering’.
Algorithms: There are many algorithms under each ML framework that are designed for different purposes. New algorithms are invented from time to time, while existing algorithms are being improved. For example, ‘reinforce learning’ and ‘supervise learning’ are two typical ML frameworks for FX analytics, but there are also many other options in terms of algorithm selection. This makes it a challenge to pick effective ML frameworks and algorithms for different business cases.
Platforms: An innovation team needs a platform that can easily integrate different datasets used in the ML training process, and to also provide enough computation power (GPUs/CPUs) to handle big data and to accelerate the training process.
Domain knowledge: A smart machine can learn, but business objectives are defined by people. Smart FX traders are needed who have deep domain knowledge so that they can provide market insight and help the machine to use their market experiences in a programmatic way. The data scientist takes the critical role of bridging the gap between technology and business.
Helping FX traders with app collaboration
The Bank of China has teamed up with Refinitiv to launch a new FX price prediction app: DeepFX.
DeepFX is an Eikon app developed by the Bank of China’s Digital Asset Management Department. Using Refinitiv’s trusted, high-quality data, the department’s FX AI model has improved significantly, proving to be more accurate and stable than other sources used by the team in the past.
The ‘Lite’ version is free with a subscription to Eikon. The trading signals it generates can be used by traders or FX researchers as a reference of short-term trading direction. Besides improved prediction accuracy, DeepFX also predicts the strength of a trading signal to help manage a position.
The model-building process has considered many factors influencing FX rates from a data perspective, but has also integrated traders’ experiences and insights. Furthermore, the model has been trained with a large amount of historical data, which aims to cover as many abnormal conditions as possible so that it can similarly handle extreme market behavior in the future.
For example, a model might not behave correctly during the COVID-19 pandemic unless it has been trained with the data of 2008 financial crisis, or another crisis before that. DeepFX has performed steadily during the COVID-19 pandemic by using a broad enough range of historical data for mode training.
As the Solution Business Director in the Refinitiv Key Strategic Account team, I worked closely with the Bank of China team on platform innovation. Together, we narrowed down the requirements and finalized the solution of integrating Refinitiv data/API, Bank of China’s AI/ML, cloud deployment and the Eikon App as a presentation layer together, and led production release successfully.
In the DeepFX screenshot below, two AI models have been pre-trained by advance deep learning algorithms and Refinitiv data in the back-end.
In the table on the left-hand side, Signal 1 shows the output of one model that uses an aggressive strategy, while Signal 2 shows the output of a different model that uses a relatively conservative strategy.
Both of them are able to predict the short-term FX price movement of six major currency pairs and generate trading signals every five minutes. The output value of Signal 1 or Signal 2 ranges from -1 to 1 and indicates a recommended position by model; the “+/-” symbol stands for suggested trade direction (“+” for long and “-” for short).
The top-right-hand side of the chart shows up to 10 days of back-testing results of the two models for the selected currency pair. The buy and hold (BAH) line on the chart is the baseline for performance metrics of the two models.
How Refinitiv can help
Refinitiv data helps to build customers’ confidence in innovation, as the future of trading becomes more data driven. Our market data gathers real-time and historical insights from hundreds of sources and expert partners worldwide, using 25 years of historical tick history data and covering 500 global venues and third-party contributors.
The content quality, coverage and the span of Refinitiv’s historical data are fundamental when building an AI application.
Eikon’s open platform is a great place to facilitate customer innovation. It enables them to build and plug into a wide array of APIs and innovative apps to get the information they need and to build solutions. And, unlike ‘closed’ models, Eikon is a catalyst for innovation in the global financial services industry.
Our global Eikon community, with more than 300K professionals, can also help customers to collaborate more effectively on a global scale.
Chinese banks have benefited significantly from globalization. For one, their FinTech capability will be visible to more than 300K global professionals via the Eikon App. Meanwhile, the Bank of China team can also learn from other global banks on how they run FinTech innovations. Through the global community, brilliant ideas and solutions can be shared and implemented.
Our product’s modern API with native Python support provides consistent access to rich content. Also, having seamless API integration with customers’ ML infrastructure helps them to empower their platforms and accelerate work on mining data and running analytics.
Refinitiv innovation ecosystem
Predictive analytics is just a part of the FX trading workflow. There are many complex FX business challenges that require not only AI innovation but also the leveraging of as many resources as possible.
An innovation ecosystem is an important concept to help solve those complex business challenges. It achieves its aims by combining all the available resources and efficient collaboration.
Refinitiv Labs has been working on building a Refinitiv innovation ecosystem with our customers, partners and colleagues in order to drive business opportunities via innovations.
Meanwhile, I have acted as a bridge between our key strategy accounts and our lab, and will continue working with our lab team to identify customers’ pain points and make full use of our innovation ecosystem to help customers solve their business challenges.