Will Machines Make Better Traders?
Machine learning’s ability to ‘learn’ with date, make predictions, and continuously improve on the accomplishment of its tasks may have earned its permanent spot on the trading floors
Platinum Analytics is leading the way in finding applications for machine learning in forex trading. The next-generation fintech company customizes localized tech solutions for financial institutions, and counts among its clients Tencent, Xinhua, and DBS Bank. An official statement from the company says it is “seeking to revolutionize the structure and competitive dynamic of the forex trading sector with its cutting-edge AI technologies, while boosting the capabilities of traders in Singapore to be among the top in the industry.” Platinum’s AI trading tool uses sentiment and market data analysis to generate trading signals, which results in valuable insight for the trading community to better tap into market data, manage risk and make money.
The algorithm is self-learning and becomes more intelligent over time. Traditionally used in the equities sector, Platinum is the first to pioneer the development of this groundbreaking tool in forex trading.
We met up with Mr. Qi Hong Bao, Chief Technology Officer at Platinum Analytics, to discuss the company’s solutions and services.
AI cannot replace experience and instinct. An experienced and savvy trader, as it stands today, can almost always do better than an AI.
HOW DOES MACHINE LEARNING WORK IN A FOREX TRADING ENVIRONMENT?
The technology is split into three parts - the data aggregator, the natural language processor, and the AI core. The data aggregator is responsible for collecting news, market price, social media, and trade data.
The following are three examples of how the technology works:
A. Parsing and classifying news: HOW IT WORKS:
The data aggregator collects machine readable news from Bloomberg/Thomson Reuters. The Natural Language Processor parses the news item and identifies key sentence structure and topics. The AI core uses statistics based machine learning algorithms to classify the news into several indices.
WHAT IT DELIVERS:
This allows the trader to see a complete landscape of all news items affecting a single currency or a currency pair.
HOW IT IS SETUP:
The client subscribes to news items he is interested in and sets up his dashboard with currencies he wants to watch.
B. Calculating Market Impact HOW IT WORKS:
The data aggregator collects machine readable news from Bloomberg/Thomson Reuters and runs it through the classifier. The Natural Language Processor parses the news item and identifies key sentence structure and topics, using the classified currency as the reference. The AI core uses algorithms to link entities and topics with a polarity dictionary to determine sentiment. Via market data correlation analysis, the algorithm then determines whether this impact is bullish or bearish.
WHAT IT DELIVERS:
This allows the trader to gauge the impact of a piece of news item on the market.
HOW IT IS SETUP:
The client subscribes to news items he is interested in and sets up his dashboard with currencies he wants to watch. As news comes in, the platform will provide impact analysis on each news item.
C. Generating Trading Signals HOW IT WORKS:
The data aggregator collects machine readable news from Bloomberg/Thomson Reuters and runs it through the classifier and impact analysis methods. The Natural Language Processor parses the data and provides a dissection of the data item, in the case of unstructured data. The platform then looks at the results of the data classification and market impact and alerts the trader of trade opportunities. Once an opportunity has been identified, the trader has the choice of manually executing the order or flowing it down to the execution engine.
WHAT IT DELIVERS:
This allows the platform to execute real time analysis on data to identify trading signals in order to maximize trader profit and minimize loss.
HOW IT IS SETUP:
The trader chooses his signal generation parameters and instructs the algorithm to provide a range of entrance and exit signals.
Has the technology been applied to existing businesses?
Yes, major banks are using our technology to automate trading and analyze the forex market.
Describe the set up and its introduction to the client’s existing system. How long is the period of system integration? What will be required from clients interested in availing of your products and services?
We provide the service as SaaS (software as a service). We host it on a cloud-based service platform where clients can log in through secure VPNs via an https portal to use the service. The integration period varies, but is generally measured in weeks. Clients will have a period of requirement gathering with our product team to determine range of data, currency range, and algorithm design.
Who will generate the data for analysis, and who will be in control of said data? What security assurances are there for the data that are being analyzed?
We take our data via three sources: machine readable API from syndicated news sources such as Bloomberg and TR, crawled data from websites, and market data from exchanges. Data encryption, safety, and transmission is handled by the platform. All data, both at rest and in transmission, are encrypted, and all traffic going in and out are all through https and run either via secured VPN or leased line.
How will you monetize the business? What revenue streams are you looking at for Platinum?
We charge fees for both customization and also via SAAS fees.
What are your plans for the company in the near and medium terms?
In the near term, we look to enhance and improve our algorithm accuracy and increase our AI analytics capabilities. In the medium term, we hope to extend our technology to other financial products such as fixed income, equities, and future.
What role will traders assume as AI helps them make decisions?
Traders will still be resonsible for formulating the strategies, as well as providing high level decision making in executing those strategies. The purpose of our AI platform is to free up the traders’ time and attention so that they may spend more time and effort shaping their strategies rather than rote execution.
In what aspects is AI limited in making decisions?
AI cannot replace experience and instinct. An experienced and savvy trader, as it stands today, can almost always do better than an AI. AI is good at taking in a wide array of information quickly and attempt to piece together pattens and recognize trends, but it lacks the focused decision making of a senior trader. The AI is good at solving some problems particularly well, and in this case, the problem is in analyzing data that is both wide and fast, but it struggles when it comes to depth. The human brain, on the other hand, is capable of making decisions based in information depths. In this manner, the AI and the human brain complement each other.
Who will be culpable, or held liable, when AI fails or makes a bad decision?
So, first, I think culpability will primarily need to be determined by a case by case basis. However, just like humans need to do their due diligence in proactively manage their risk, so do the engineers and analysis responsible for creating the AI. It’s up to the AI engineers to make sure the program is robust and guarded against potential risks and filter out elements which may cause poor decisions making. The same expectations from traders should also be given to the AI. That said, in most current implementations of AI, the trader still has the final say in the decision makng process. However, it’s equally important for the AI to not mislead the human trader by giving poorly quantified decisions with no confidence qualifier. If a human trader makes bad decisions which cause financial loss because, say, they came to work drunk, then the culpability lies with the human. If the AI algorithm is implemented with unfinished quanlity assurance or insufficient backtesting, then for the same reason, the culpability is with the AI, even if the human trader had the final say in making the decision. Often, in these cases, ruling out obvious malice or sabotage, culpability will generally lie with those who failed to do due diligence or ignored proper risk management.