Chinese online brokerage tests first AI chatbot for stock trading as it talks to regulators about compliance

Tiger Brokers, an online brokerage, has created a trade bot powered by generative artificial intelligence technology, similar to what powers OpenAI’s ChatGPT, to test how smart machines can replace humans in bonds and stock trading.

“We’re keeping a close eye on AI and industry advancement since November last year,” Jacques Li, head of global communications at the Xiaomi-backed brokerage, told the Post in an interview on Wednesday.

TigerGPT is currently only available as an invite-only beta for a small cluster of users. Like similar services that have emerged over the past several months, the AI chatbot can generate responses to prompts on a wide range of topics. But this bot specialises in financial information, with the goal of providing timely data to help users of Tiger’s online trading platform with their investment decisions.

China to embrace AI advances but also control risks as ChatGPT wave spreads

The Beijing-based brokerage, owned by Nasdaq-listed Up Fintech Holdings, took three months to “think thoroughly” about whether it wanted to provide its own AI to address some of the pain points of the platform’s 2 million account holders and 9 million users, Li said.

The project was initiated in January, and TigerGPT finally launched this month, making Tiger Brokers the first company to incorporate ChatGPT-like services into an online brokerage platform.

One user pain point, according to Li, was the difficulty of gathering information that is critical to assessing investment decisions. Li said TigerGPT can help users save time on market research and get more up-to-date information.

TigerGPT was trained on a vast amount of premium content that the company has access to, according to Li. The result is a chatbot capable of analysing current affairs and macroeconomic trends – such as how many times the US Federal Reserve has raised interest rates this year.

As the name would suggest, TigerGPT uses OpenAI’s large language models (LLMs) known as Generative Pre-Trained Transformer (GPT) models, according to the company. However, Tiger Brokers is using the older GPT-3 model. ChatGPT launched with GPT-3.5, while the more sophisticated GPT-4 was launched earlier this year.

Users can ask about a wide range of information about companies in their portfolio, but currently in either English or Mandarin Chinese only. TigerGPT can give company fundamentals based on earnings reports, as well as third-party analysis for a fuller picture.

These capabilities make comparing company performance as simple as a quick prompt. TigerGPT can, for example, pull up Tesla’s price-to-earnings ratio on command and compare it against rival electric vehicle brands. This could give users more immediate access to information that might otherwise require multiple searches for each company and self-tabulated data.

“Sometimes the market moves really fast, up and down, and investors would want to know what is really happening,” Li said. While other financial information services can address this need, Li said that Tiger Brokers “can provide [users] with timely answers of why this is happening, why this is relevant and how this will impact their investment strategy.”

However, ChatGPT and other generative AI services, such as Google’s Bard, are known for giving inaccurate answers. Li acknowledged that the same issue applies to TigerGPT in its early days, but he said the project’s 50-person team has been fine-tuning it on a daily basis and feeding it the latest market information to improve accuracy.

Hong Kong stocks suffer amid US$26 billion sell-off in Tencent and BYD

Li said Tiger Brokers has been in touch with regulators around the world from the beginning of the project in order to address their concerns, including in Hong Kong.

“AI, among other technologies, should be subject to rigorous regulation,” he said. “In Hong Kong, we’re trying to make sure this feature is under their guidance and in full compliance.”

In the next iteration of TigerGPT, the company intends to introduce an audio function so users can speak their queries and listen to the response.

When Microsoft-backed OpenAI launched ChatGPT last November, rival Big Tech firms were caught off guard by the product’s instant global popularity. Tech firms with their own AI models – from Google and Meta Platforms in the US to Baidu and Alibaba in China – raced to launch their own chatbots. However, LLMs’ propensity for giving unexpected and fallacious responses has limited how they can be used in certain contexts.

In the financial sector, the prevalence of robo-advisers and algorithmic investment suggests that the latest evolution of AI may pose no challenges to existing regulatory frameworks, according to You Chuanman, director of the Institute of Internal Auditors Centre for Regulation and Global Governance with the Chinese University of Hong Kong, Shenzhen Campus.

“The regulation of robo-advisers has been in practice for the last decade,” You said. “I don’t see a paradigm shift of the regulatory structure.”