AI innovation on FX trading with Bank of China
As our CEO quoted at smart person powered by smart machine ?- “AI and ML play in the transformation of financial services”, using AI/ML for predictive analysis is one of the important innovation areas of forex trading workflow, there are always gaps between data, technologies and business when building an applicable AI application, a joint innovation between Refinitiv and Bank of China provides a practice how to fill the gap and move an AI innovation from an idea to real in short time.??????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????????
1.???? AI innovation on FX trading will not only need the power of machine, but also human experiences and knowledge, the deep collaboration between Refinitiv and Bank of China makes full use of the power of both human and machine to achieve innovation objectives.
2.???? Refinitiv high quality market data and historical data are fundamental of the AI innovation. Anyone working in AI knows the importance of avoiding bad data garbage in and garbage out, choosing proper datasets and combine them to maximize alpha is the starting point of the innovation.
3.???? Eikon’s innovative open platform enables customers to build own apps and share Fintech innovations to global community rapidly, our APIs also enables a seamless content integration to customer’s system to support their innovations.
?
Artificial Intelligence on FX trading has entered people's vision for years. However, it started to become more practical these days with advances in big data and machine learning, nowadays FX traders are becoming increasingly rely on predictive analytics based on big data and AI algorithms. As one of the top 4 banks in China who runs forex business over 70 years, Bank of China has seen the trend, its digital asset management department started researches of using deep learning algorithms to predict FX price movement a couple of years ago and made significant progress on it.
THE CHALLENGE:
There are 4 major challenges when doing AI innovation on Forex:? data, algorithm, platform and domain knowledge. Firstly, figuring out what kind of data and data combinations will be best fit for building a Forex AI model and how to get high quality data source at right time are very challenging. The process is also called ‘feature engineering’ in Machine Learning world. Secondly, there are so many algorithms under each ML framework designed for different purposes and they are developing quickly, engineers might propose reinforce learning framework to treat a learning process as game playing e.g. go or chess, or they can choose supervise learning framework to treat a learning process as classification problem(long/short) or a regression problem and each ML frameworks also includes many subtle algorithms underneath like MDP, DRQN, ANN, LR etc., so setup right learning goal and choose proper algorithm for specified FX analytics use cases are very challenging. Thirdly, an innovation team needs a platform which can easily integrate different datasets used by ML training process and provide enough computation power (GPUs/CPUs) to handle big data and accelerate training process. Finally, smart machine could learn, but the business cases and objectives are defined by human, it needs smart FX traders who have deep domain knowledge to provide market insight and help machine to use their market experiences? in programmatic way, datascientist is taking critical role to fill the gap between technology and business.
THE SOLUTION:
Bank of China teams up with Refinitiv to launch new foreign exchange price prediction app - DeepFX to solve the challenges, DeepFX is an Eikon App independently developed by Digital Asset Management Department of Bank of China. By using trustable high quality Refinitiv data, the performance of its FX AI model has been improved significantly on both accuracy and stability comparing with other sources they have ever tested. ?The innovative adoption of Refinitiv historical datasets assists the AI model to be able to consider extreme market situations through learning the data of 2008 finical crisis and show a stable performance during covid19, the model design is not only focus on prediction accuracy like most of AI applications are focusing, but also maintain a good shape of the performance curve, it means even we could achieve 90% of all trades are profitable, one crisis in the 10% situation like covid19 can absorb all your profits and generate irreparable losses, the innovation tried to avoid this by using large amount of historical data and datascientist from Bank of China also enhanced generic ML algorithm to make it more suitable for FX use cases and also be able to consider trader’s experiences and insights as one of decision factors in program.
?
Solution Business Director in Refinitiv KSA team, who is responsible for defining and driving Fintech Innovations with our global key strategic accounts and using cutting-edge technologies to deliver tangible solutions through a win-win way, had been worked closely with the expert team of Bank of China on platform innovation, we narrowed down the requirements together and combined with Refinitiv data/API, AI/ML, Cloud Deployment and Eikon App as presentation to form a final proposal and led the Eikon App release successfully.
领英推荐
?
Below is the DeepFX App UI in Eikon, two AI models behind the Eikon App have been pre-trained by advance deep learning algorithm and Refinitiv data in Bank of China’s ML computation farm, Singal1 on left panel shows the output of one model being trained with aggressive strategy and Singal2 shows the output of another model being trained with relative conservative strategy. Both of them can predict short-term FX price movement of six major currency pairs and generate trading signals every next 5 minutes, the range of the signals is [-1,1], +/- indicates the trade direction (“+” for long and “-“ for short) and the value of a signal indicates a recommended position. The right panel shows up to 10 days back-testing result of the two models for selected currency pair, the BAH (Buy and Hold) line is the base line for performance metrics. ?
CUSTOMER BENEFITS:
·?????? Refinitiv data helps to build customer’s innovation confidence as future trading will become more data driven (see the latest Refinitiv data catalog). Our market data gathers real-time and historical insights from hundreds of sources and expert partners worldwide, 25 years of historical tick history data covering 500 global venues and 3rd party contributors, customers are very impressive about our content quality, coverage and the span of Refinitiv historical data which are the fundamental of building an AI application
·?????? Eikon’s open platform is a good place for combination innovation, it allows customer to build and plug into a wide array of APIs and apps to get the information needed and build solutions in it. And, unlike ‘closed’ models, Eikon is a catalyst for innovation in the global financial services industry, this is well demoed in the case.
·?????? 300K+ professionals in Eikon open platform is attractive, our global Eikon community is not only facilitating the connection between human to machine, machine to machine but also human to human, no matter for human or machine, they can learn from each other by leveraging Eikon platform’s capabilities, brilliant ideas and solutions can be quickly implemented and spread out in the world.
·?????? Our product’s modern API with native Python support provides consistent access approach to rich content and seamless content integration with customers’ ML computation farm and helps to empower customer’s platform, improve its platform efficiencies and seamless platform integration makes datascientist’s works be much easier to manipulate, interpolate data and run analytics.
?
FUTURE:
·?????? With the development of AI innovation on Forex, the first question might come about is will AI replace trader’s work finally if it looks so powerful? From a recent blog posted at Future of Trading: The best of people and machines, by Michael Chin, Managing Director and Global Head of Trading Proposition at Refinitiv, ?the answer is people will still be in charge, AI innovation will offer more opportunities on analytics and automations and the fusion of technology and finance will help to move trader’s career path up to next level, more traders will start to learn how to use python for analytics, automation or even building AI model by themselves.
·?????? Besides the AI innovation on pre-trade analytics to support decision making, innovations have been happening in other key components of the trading workflow, for example, how to optimize trade routes, achieve best execution and reduce transaction costs etc., on one hand automation could remove the human’s emotion in trading process while speeding up trading execution which is good, on the other hand traders should never let technology limit trading opportunities, automation allows you to become more productive and focus on alpha generation or risk mitigation, compliance etc., you should never miss opportunities when you see it.
·?????? As Refinitiv continues enhancing our content, technology, product and platform, a health innovation ecosystem based on that can facilitate AI innovation and maximized the value of both Refinitiv and customers, our innovation lab is open to customer’s feedback where KSA, SBD team is a bridge of that for Key Strategic Account, see the all projects from Refinitiv Innovation Lab at? https://labs.refinitiv.com/ . At last, no matter it is machine learning or human learning, learning itself is a continuous cycle, so does innovation.