(Bloomberg) — It’s been a source of intrigue and confusion among currency traders for years. Just how does China decide each day where it’s going to set its fixing rate for the yuan — a reference point that determines its permitted trading range for the next session?
Most Read from Bloomberg
Now, a small group of traders are using artificial intelligence models in an attempt to decode the nation’s opaque currency policy.
Among them is Ding Di, who works for a Shanghai-based bank’s investment department. For the past year, Di has been working with the Gemini-powered AI model Xiao Long, also known as Little Dragon, to see if it could help him predict where the People’s Bank of China would set the daily yuan reference rate.
Di got an early taste of success on March 4. Then three days into the US war with Iran, Xiao Long computed the day’s dollar-yuan fixing at 6.9122, just two pips below the official number of 6.9124.
The model was nearly spot on because it had identified a change in how the fixing was set the day before, said Di, who declined to identify his employer as he’s not authorized to speak publicly and the AI experiment is for his personal use.
While the central bank focused on factors such as the US-China interest rate differential and the Federal Reserve’s easing mode in the months before the war, it seemed to have begun attaching more weight to the yuan’s overnight moves, Xiao Long’s findings showed. That suggested the PBOC no longer intended to let the yuan extend its appreciation since last year, at least around that time, Di said.
“The fixing is official pricing, not something arbitrary or casually disclosed,” said Di in an interview. “AI approaches things from policymakers’ perspective. It may interpret signals as a particular way of thinking by the central bank.”
Di is not alone in using AI to penetrate the thinking of the PBOC. A small group of like-minded peers say they are training AI tools in a bid to detect patterns from the minuscule changes made daily to the fixing. A tentative conclusion is that the central bank has been reining in the managed currency since the war started, after letting it appreciate against the dollar in the months before that.
The PBOC didn’t respond to a faxed request for comment.
China’s central bank doesn’t hold frequent briefings, unlike some of its major peers, often leaving investors and businesses guessing at its intent from broad statements. The reference rate is the only other visible clue for its foreign exchange policy, with the yuan confined to a 2% trading range around it in the onshore market. Beijing’s policy stance has come under renewed scrutiny as its exports surged, with critics saying an undervalued currency played a part.
Across 225 episodes over the past four quarters, Xiao Long has correctly called the direction more than two-thirds of the time, according to Di. In over 40% of the cases, its forecasts were within 20 pips away from the official levels.
A Singapore-based hedge fund trader using a similar AI tool to Di’s has reached comparable conclusions. His model shows that the fixing had reflected an appreciation bias, but since March, the data suggest that the PBOC has become more tolerant of slower yuan gains.
Early Days
The yuan has strengthened over 8% against the dollar since President Donald Trump launched his sweeping global tariffs in early April last year, surprising some in the market. Analysts have rushed to upgrade their estimates, with Goldman Sachs Group Inc. saying the yuan is more than 20% undervalued.
Nominally speaking, the PBOC sets the yuan fixing based on market makers’ price quotations that reference the previous day’s close, supply and demand, and major currencies’ moves, with unspecified weightings attached.
In practice, the fixing deviates from basket-implied levels “quite regularly,” said Stephen Jen, chief investment officer at Eurizon SLJ Capital. It’s a structurally “semi-transparent” system, one that provides signals while retaining discretion, he added.
To Jen and some other observers, the fixing reflects the PBOC’s preference for currency stability while it seeks to address needs ranging from economic fundamentals to capital flows and geopolitical risks.
To be sure, it’s still early days and many remain skeptical that the fixing’s opaque mechanism lends itself to efforts to draw any meaningful conclusions.
From time to time, the Little Dragon also suffers from hallucinations, a phenomenon where AI generates false or fabricated information, according to Di. But with continued tracking and constant upgrading, Di now has a better sense of the AI model’s limitations.
“It’s now a human–machine symbiosis — with human input, AI can form its own observations,” he said. “That’s what makes it so interesting.”
Most Read from Bloomberg Businessweek
©2026 Bloomberg L.P.