Sector radar: method and current readings

Generated 2026-10-26 by manage.py radar_report from waitan_ludi_app/radar.py; the page is /radar/.

What it is for

One question asked of every strategic sector at once: where do Chinese state support, local preference and export push coincide? The radar puts sixteen sectors on one scorecard so that a trade official, a compliance team, an investor and a researcher read the same figures. Each audience then has its own route in: the risk brief (/radar/brief/), the exposure check for one operation (/exposure/), the listed groups and province matrix (/radar/provinces/), and the robustness checks and test on the radar page.

Method

• Sector document set. Documents whose title names the sector (a Chinese pattern per sector, printed on each brief) plus the members of the sector's topic trees. Same population as the Documents page.
• Policy indicators. Momentum (share of the corpus in the last 24 months over the 24 before, so it is relative to collection); provincial spread (36 months); the shares of documents with a fiscal measure, with a favouring provision or localisation condition, and with import-substitution language.
• Trade indicators. UN Comtrade, China-reported, 2019 to the latest year: compound export growth, growth of exports to the EU-27, the EU share and the export/import ratio. Growth starts in 2022 for HS 2022 codes. From the EU side (Eurostat Comext, EU-reported, euro): China's share of the EU-27's imports from outside the EU.
• Composite. z-scores across sectors (ratios on a log scale, clipped at ±3), weighted mean of those a sector has. The rank range is the 10th-90th percentile over 500 Dirichlet(1) weightings (seed 20261026).
• Not shown. That support caused exports, that money was paid, or the size of support in yuan.
• Sector-set precision. Ten titles per sector drawn at random and read against the sector by the AI assistant: 135/160 on topic in round 1; after tightening the ship and aviation patterns and excluding event notices, general entity lists and a place name, fresh samples of the changed sectors scored 52/60. Labels: reports/radar_sector_validation_2026-10-26.csv.

Current readings

Every sector has a trade series.

Supported and pushing out (policy and trade scores above the mean): New-energy vehicles (exports to the EU +78% a year), Drones and low-altitude (exports to the EU +21% a year). Pushing out, policy now below the mean: Batteries and storage (China 90% of EU imports; policy pressure highest in 2024); EV charging (China 59% of EU imports; policy pressure highest in 2022); Critical minerals (China 83% of EU imports; policy pressure highest in 2018); Shipbuilding (China 25% of EU imports; policy pressure highest in 2023). Read with the history table on /radar/: an export push that follows support with a lag, or outlives it. The one test below does not separate the two. Next wave? (policy above the mean, exports not yet): Pharmaceuticals, AI and computing, Robotics, Semiconductors, Machine tools, Seeds.

Robustness

Movers, policy score now against 2026-04-27: AI and computing +0.22; Seeds +0.15; Pharmaceuticals +0.15; Shipbuilding +0.11; Robotics -0.11; Batteries and storage -0.14; Civil aviation -0.17; Drones and low-altitude -0.36.

One test

Did sectors under more policy pressure at the end of 2021 see faster export growth in 2021–2025? Specified in radar.lead_test before it was first run: policy score from documents dated up to 2021-12-31, compound export growth 2021–2025, Spearman's rho, two-sided permutation p (20,000 shuffles).

rho = -0.24, p = 0.39, n = 14. No detectable association. With this few sectors only a strong association could show; sector shocks dominate (pharmaceutical exports fall with the end of COVID vaccines). The proper design is product-level: HS6 lines targeted by dated policy against untargeted lines in the same chapter (reports/plan_china_shock_eu.md).

Second test, EU side

Did China gain more of the EU's imports in the headings its policy targets than in their neighbours? Specified in eu_lines.py before it was first run: HS4 headings of chapters 84, 85, 87–90 with EU extra-EU imports of €100m+ in 2019 and 2025; treated = headings holding a radar sector's goods; outcome = change in China's share of EU imports; within-chapter difference in means; labels permuted within chapter (20,000 draws).

Difference +0.0 points, p = 0.99; 22 targeted headings, 143 neighbours. China's share rose in the median heading by 5.0 points where policy targets the goods and 5.9 where it does not (2019–2025). The gain is broad, not concentrated in the targeted headings.

Largest gains among untargeted headings: 8707 Bodies, incl. cabs, for tractors, motor vehicles for the tra +65; 8426 Ships' derricks; cranes, incl. cable cranes (excl. wheel-mou +36; 8476 Automatic goods-vending machines, e.g. postage stamp, cigare +32; 8433 Harvesting or threshing machinery, incl. straw or fodder bal +30; 8450 Household or laundry-type washing machines, incl. machines w +30.

Where each sector is pushed

1056 local documents over the last 36 months, by province and sector, with location quotients (the sector's share of the province's documents over its national share). Provinces that lean clearly towards a sector (quotient 1.5 or more, at least 3 documents):

• Guangdong (245): EV charging, Hydrogen
• Shanghai (132): Shipbuilding, Pharmaceuticals
• Sichuan (73): Seeds, Batteries and storage
• Beijing (63): Robotics, Machine tools
• Inner Mongolia (60): Solar PV, Batteries and storage
• Zhejiang (54): Machine tools, Semiconductors
• Hubei (50): Seeds, Medical devices
• Fujian (42): Medical devices
• Shaanxi (31): Drones and low-altitude, Batteries and storage
• Anhui (29): Semiconductors, Batteries and storage
• Chongqing (28): New-energy vehicles, Pharmaceuticals
• Shandong (26): Pharmaceuticals, AI and computing
• Jiangsu (25): Medical devices
• Guizhou (23): Batteries and storage, Solar PV
• Jilin (21): Batteries and storage, Hydrogen

The 15th Five-Year Plan outlines (plan watch)

Mentions per 100,000 characters in the 15 provincial and city outlines of the 15th plan held in full against the 7 outlines of the 14th plan held; within-place change for the 4 places with both (Ordos, Anhui, Chongqing, Kunming). Sector terms are the radar's title patterns applied to the text.

Exploratory lead check (specified after the 15th-plan counts were seen): across sectors, density in the 14th-plan outlines against local documents 2021-25, rho = 0.58 (p = 0.019, n = 16); against China's export growth 2021-25, rho = -0.23 (p = 0.430, n = 14). Density and documents share the corpus's coverage. A mention is not a commitment.

Caveats

• Documents held, not issued: coverage differs by province (/provinces/coverage/). Momentum is relative to the corpus, which removes uniform growth in collection but not collection that favoured one sector.
• Measure, preference and substitution figures are keyword and model screens. Hand-checked precision is published for some classes (local registration 56/60); others are leads.
• HS headings are wider than the policies (8542 is every integrated circuit; 3004 every dosed medicine).
• The composite is a summary for comparison, not an index of harm; every component is shown beside it.