DEEPSEEK’S success is intensifying pressure on China’s quantitative hedge funds to embrace artificial intelligence (AI) or risk becoming obsolete, according to one of the most aggressive users of AI in the industry.
“In three years, quants that do not use AI will inevitably be eliminated,” said Feng Ji, chief executive officer of Baiont Quant, which uses machine learning to do trading with no human intervention. “DeepSeek is a final call to those who still do not believe in the power of AI.”
Baiont manages more than six billion yuan (S$1.1 billion), mostly from institutional clients such as securities firms. Its long-only product has beaten China’s CSI 1000 Index of small-cap stocks by 66 percentage points since the fund’s inception two years ago.
China’s financial and asset management industries are already rushing to leverage DeepSeek’s R1, with dozens having integrated the AI reasoning model into their own systems to reduce costs and boost efficiency. For quants, the ability to better tap AI is critical for survival, as the 1.3 trillion yuan industry faces cut-throat competition, the rapid decay of factors that trading models rely on to generate returns, and a regulatory clampdown aimed at protecting retail investors.
“The breakthrough progress by DeepSeek is undoubtedly exciting for the entire quantitative investment industry,” another quant firm, Hangzhou Longqi Scientific Investment, said. “DeepSeek has used its technological strength to legitimise quantitative investing and make people realise the power of technology in investing.”
DeepSeek was spawned in 2023 by Zhejiang High-Flyer Asset Management, a top quant co-founded by Liang Wenfeng in 2015. After machine learning helped push High-Flyer’s assets under management to more than 90 billion yuan at its peak in 2021, Liang shifted his focus to Artificial General Intelligence.
BT in your inbox
Start and end each day with the latest news stories and analyses delivered straight to your inbox.
Many other Chinese quants have been applying AI in various ways. They mostly use machine learning in different parts of the investment process to varying extents, typically combining researcher-developed factors – attributes of securities that are believed to influence future performance – with machine-found ones to build strategies. While chatbots including OpenAI’s ChatGPT are not yet powerful enough to directly produce complete quant strategies, firms such as Longqi have also been using them to accelerate research.
As their training costs decline and capabilities grow, chatbots’ use in investment research is expanding rapidly, with DeepSeek R1 having become one of the best choices for daily work for performance and pricing, according to China Merchants Securities. Potential scenarios include developing alternative factors and monitoring transactions for compliance risks, analysts led by Ren Tong wrote in a report this month.
The ability to leverage AI made a difference last year when quants tackled wild gyrations in Chinese stocks. Shanghai Manfeng Asset Management used deep learning to cull more stocks from its multi-factor model, leading to a tendency for contrarian bets that helped it avert a collapse in the meltdown. Hainan Zhengren Quant Private Fund Management said reinforcement learning allowed it to develop models that skipped the use of factors and helped it adapt to market swings more swiftly.
Yet none of those approaches would suffice in the next phase of competition, according to Baiont, which is part-owned by technology pioneer Kai-Fu Lee, founder of 01.AI and Sinovation Ventures.
New blueprint
“Most firms still see AI just as an alternative tool while their production line remains the same, but that’s absolutely not enough,” said Feng, who holds a PhD in machine learning from Nanjing University. “The entire factory blueprint needs to be reconstructed.”
Feng sees a “super transformation period” where firms will need to shift to a system of AI-driven investment from start to finish. Quants are still mostly “human-centred”, relying on “brilliant ideas” of researchers to develop and upgrade strategies while computer algorithms are used to automate trading, he said.
Baiont, set up in 2020, built its entire production line with AI, which performs everything from factor exploration to strategy development and execution without humans, according to Feng. Its 30-strong team, with no researcher developing factors such as most quants do, maintains the “foundation model to do all tasks end-to-end” by monitoring key indicators, he said.
“If you were previously an automobile manufacturer and you wanted to switch to electric vehicles, you cannot just replace the engine production section with a new-energy battery section and assume that your factory doesn’t need any other changes,” he said. “That simply will not help you truly achieve the ultimate improvement in performance.”
Baiont’s fund that beat the CSI 1000 index is an enhanced index fund, which uses AI to help select better-performing stocks from the benchmark’s members. Its market-neutral fund made a combined 37 per cent return since June 2022, including a 7 per cent gain last year when half of such products tracked by CSC Financial lost money.
The firm’s AI-based production line also means upgrades to its strategies can be planned as technology evolves, defying limits of researchers’ human inspirations. “The cool thing about AI systems is that you can actually schedule the upgrade, and that’s quite remarkable for quant firms,” Feng said.
Quants with AI advantages are gradually developing a new generation of methods that combine customised features, which can improve algorithm efficiency, curb model risk and reduce strategy crowding, according to a Guotai Junan Securities report in 2022. Such algorithms are hard to replicate, “making them a core competitive advantage”, the analysts wrote.
“We have already moved from the age of cold weapons to that of firearms,” Feng said. “If you are still wielding a spear, you will be in great danger when it comes to close combat.” BLOOMBERG
