Generated by Codex with GPT 5.6 Sol XHigh

China’s official urban unemployment rate presents an implausibly calm picture of the world’s largest labor market. Since the measure was introduced in 2018, it has landed between 5% and 6% in 99 of 115 months. Even during the covid-19 shock it never rose above 6.2%. The article argues that this stability is not reassuring. It is a warning that the statistic may be shaped by the government’s targets and by the way China’s migrant workforce moves in and out of the survey.

When a measure becomes a target

The problem is captured by Goodhart’s law: once policymakers use a statistic for control, the relationship it measures tends to break down. China sets targets for many parts of its economy, so officials have strong incentives to make the published numbers conform to political goals. Since 2018 the annual target for surveyed urban unemployment has usually been about 5.5%, rising to 6% in 2020. The reported rate’s narrow range around those figures has led Andrew Batson of Gavekal Dragonomics to suggest that it may be managed to suppress volatility, though that possibility cannot be proved.

The survey’s design can also dampen genuine changes. It covers urban workers, including migrants from the countryside. When city jobs are plentiful, more migrants arrive and expand the pool of potential workers, limiting the fall in the unemployment rate. When conditions weaken, unemployed migrants may return to rural areas and disappear from the urban survey, limiting the rise. The movement of people therefore gives the statistic a self-stabilizing quality even without direct manipulation.

That stability leaves economists struggling to answer basic questions. China’s export boom may be generating less manufacturing employment than its output suggests. Artificial intelligence may be reducing entry-level white-collar opportunities. Weak wage growth may also be increasing the risk of renewed deflation. A labor-market measure that barely moves cannot confirm or reject any of these concerns.

An old statistic starts speaking again

Batson proposes revisiting a cruder measure that dates to 1978: the number of people registered as unemployed with local authorities, typically as a first step toward claiming benefits. For decades this figure was itself a government target and was almost as stable as today’s survey rate. It faded from view after the newer measure appeared. Officials stopped targeting it and no longer publish a percentage rate, but the raw annual total still appears in the official database.

That neglected count has now moved sharply. Registered unemployment reached 12.7m at the end of 2025, nearly 16% more than a year earlier. Rule changes introduced in 2020 made registration easier, particularly for migrants living far from their registered home, so the absolute level must be treated carefully. But those changes do not explain the latest jump. The rise is therefore a plausible sign of real labor-market stress.

The reversal suggests a corollary to Goodhart’s law: when a measure stops being a target, it can become useful again. Once officials and markets stopped paying attention to registered unemployment, there was less reason to keep it smooth, allowing it to recover some informational value.

The cost of statistical control

The lesson is not that the older count is a perfect substitute. Both measures have serious limitations, and analysts must assemble a picture from imperfect proxies. The deeper problem is that political pressure damages the feedback policymakers need. A government that treats uncomfortable data as a threat may gain a tidier dashboard while losing sight of the economy it is trying to steer.

China has already shown how this can end. When youth-unemployment figures became a focus of criticism in 2023, publication was suspended. The same fate could await registered unemployment if its recent rise attracts too much attention. In that system, a measure used as a target stops being informative; a measure that becomes embarrassing may stop appearing at all.