The full tables behind the talk vs trade page: every lasting shift on both sides, every lag, the month-by-month correlations, graphite on its own, and the price-band effect behind the March 2025 drop.
Every lasting shift we found from Feb 2022, on both sides. Trade can be read as China’s share of EU import value or as tonnes from China.
Each dot is a lasting shift; the line behind it is the 95% range for its date. Triangles point up or down.
Shifts after about Jan 2026 cannot be detected yet, because a new level must last six months (shaded). Trade data end in Jul 2026; talk data in Sep 2026.
Trade: Eurostat Comext, EU27 imports from outside the EU · Talk: 14,577 paragraphs, adjusted for coding errors
Each talk shift is matched to the nearest trade shift in its pairing. A negative lag means trade moved first.
Lag = trade shift month minus talk shift month. Graphite-only pairs use quarters.
Only the two “no stage named” shifts are confirmed by both methods, and both come after the March 2025 trade shift. Components talk had no lasting shift to match.
Rule fixed on 31 Oct 2026, before any trade shift was computed
| Pairing | Trade measure | Talk shift | Nearest trade shift | Lag (months) | Reading |
|---|---|---|---|---|---|
| No stage named · all texts ↔ Anode-grade graphite | China share | 2025-06 (confirmed) | 2025-03 (confirmed) | -3 | trade leads |
| No stage named · all texts ↔ Anode-grade graphite | Volume | 2025-06 (confirmed) | 2025-03 (confirmed) | -3 | trade leads |
| No stage named · the press ↔ Anode-grade graphite | China share | 2025-10 (confirmed) | 2025-03 (confirmed) | -7 | trade leads |
| No stage named · the press ↔ Anode-grade graphite | China share | 2026-04 (PELT only) | 2025-03 (confirmed) | -13 | unrelated (>12 m) |
| No stage named · the press ↔ Anode-grade graphite | Volume | 2025-10 (confirmed) | 2025-03 (confirmed) | -7 | trade leads |
| No stage named · the press ↔ Anode-grade graphite | Volume | 2026-04 (PELT only) | 2025-03 (confirmed) | -13 | unrelated (>12 m) |
| Refining talk ↔ Artificial graphite (3801.10) | China share | 2022-09 (PELT only) | 2022-06 (PELT only) | -3 | trade leads |
| Refining talk ↔ Artificial graphite (3801.10) | China share | 2024-01 (PELT only) | 2024-03 (confirmed) | +2 | simultaneous |
| Refining talk ↔ Artificial graphite (3801.10) | China share | 2024-11 (PELT only) | 2024-03 (confirmed) | -8 | trade leads |
| Refining talk ↔ Artificial graphite (3801.10) | China share | 2026-01 (PELT only) | 2024-03 (confirmed) | -22 | unrelated (>12 m) |
| Refining talk ↔ Artificial graphite (3801.10) | Volume | 2022-09 (PELT only) | 2022-03 (confirmed) | -6 | trade leads |
| Refining talk ↔ Artificial graphite (3801.10) | Volume | 2024-01 (PELT only) | 2024-07 (confirmed) | +6 | talk leads |
| Refining talk ↔ Artificial graphite (3801.10) | Volume | 2024-11 (PELT only) | 2024-07 (confirmed) | -4 | trade leads |
| Refining talk ↔ Artificial graphite (3801.10) | Volume | 2026-01 (PELT only) | 2024-07 (confirmed) | -18 | unrelated (>12 m) |
| Refining talk ↔ Refined natural graphite | China share | 2022-09 (PELT only) | 2023-05 (confirmed) | +8 | talk leads |
| Refining talk ↔ Refined natural graphite | China share | 2024-01 (PELT only) | 2023-05 (confirmed) | -8 | trade leads |
| Refining talk ↔ Refined natural graphite | China share | 2024-11 (PELT only) | 2025-03 (confirmed) | +4 | talk leads |
| Refining talk ↔ Refined natural graphite | China share | 2026-01 (PELT only) | 2025-03 (confirmed) | -10 | trade leads |
| Refining talk ↔ Refined natural graphite | Volume | 2022-09 (PELT only) | 2023-05 (PELT only) | +8 | talk leads |
| Refining talk ↔ Refined natural graphite | Volume | 2024-01 (PELT only) | 2023-05 (PELT only) | -8 | trade leads |
| Refining talk ↔ Refined natural graphite | Volume | 2024-11 (PELT only) | 2024-10 (PELT only) | -1 | simultaneous |
| Refining talk ↔ Refined natural graphite | Volume | 2026-01 (PELT only) | 2024-10 (PELT only) | -15 | unrelated (>12 m) |
| Mining talk ↔ Natural graphite, all | China share | 2022-09 (PELT only) | 2024-06 (confirmed) | +21 | unrelated (>12 m) |
| Mining talk ↔ Natural graphite, all | China share | 2024-01 (PELT only) | 2024-06 (confirmed) | +5 | talk leads |
| Mining talk ↔ Natural graphite, all | Volume | 2022-09 (PELT only) | 2022-12 (confirmed) | +3 | talk leads |
| Mining talk ↔ Natural graphite, all | Volume | 2024-01 (PELT only) | 2024-02 (PELT only) | +1 | simultaneous |
| Graphite · no stage named (quarterly) ↔ Anode-grade graphite (quarterly) | China share | 2023-Q4 (PELT only) | 2023-Q4 (BP only) | +0 | simultaneous |
| Graphite · no stage named (quarterly) ↔ Anode-grade graphite (quarterly) | China share | 2025-Q4 (PELT only) | 2025-Q2 (confirmed) | -6 | trade leads |
| Graphite · no stage named (quarterly) ↔ Anode-grade graphite (quarterly) | Volume | 2023-Q4 (PELT only) | 2025-Q2 (confirmed) | +18 | unrelated (>12 m) |
| Graphite · no stage named (quarterly) ↔ Anode-grade graphite (quarterly) | Volume | 2025-Q4 (PELT only) | 2025-Q2 (confirmed) | -6 | trade leads |
We shuffled the shift dates 5,000 times, keeping how many shifts each series has, and counted how often a talk shift lands within a year of a trade shift.
Number of talk shifts with a trade shift within a year, against the average over 5,000 random histories
With two to five trade shifts in 54 months, nearly any talk shift has a trade shift within a year. The p-value is the share of random histories with at least as many matches; none is below 0.30.
5,000 random draws per pairing · at least 6 months between shifts
| Pairing | Trade measure | Shifts talk / trade | Matched | Expected | p |
|---|---|---|---|---|---|
| No stage named · all texts ↔ Anode-grade graphite | China share | 1 / 5 | 1 | 1.00 | 1.00 |
| No stage named · all texts ↔ Anode-grade graphite | Volume | 1 / 3 | 1 | 0.93 | 0.93 |
| No stage named · the press ↔ Anode-grade graphite | China share | 2 / 5 | 1 | 2.00 | 1.00 |
| No stage named · the press ↔ Anode-grade graphite | Volume | 2 / 3 | 1 | 1.86 | 1.00 |
| Refining talk ↔ Artificial graphite (3801.10) | China share | 4 / 2 | 3 | 3.09 | 0.80 |
| Refining talk ↔ Artificial graphite (3801.10) | Volume | 4 / 2 | 3 | 3.10 | 0.80 |
| Refining talk ↔ Refined natural graphite | China share | 4 / 2 | 4 | 3.08 | 0.30 |
| Refining talk ↔ Refined natural graphite | Volume | 4 / 2 | 3 | 3.10 | 0.80 |
| Components talk ↔ Anode-grade graphite | China share | 0 / 5 | 0 | — | — |
| Components talk ↔ Anode-grade graphite | Volume | 0 / 3 | 0 | — | — |
| Mining talk ↔ Natural graphite, all | China share | 2 / 2 | 1 | 1.55 | 0.97 |
| Mining talk ↔ Natural graphite, all | Volume | 2 / 5 | 2 | 2.00 | 1.00 |
| Graphite · no stage named (quarterly) ↔ Anode-grade graphite (quarterly) | China share | 2 / 5 | 2 | 1.99 | 0.99 |
| Graphite · no stage named (quarterly) ↔ Anode-grade graphite (quarterly) | Volume | 2 / 2 | 1 | 1.55 | 0.96 |
| Graphite · refining talk (quarterly) ↔ Artificial graphite (quarterly) | China share | 0 / 2 | 0 | — | — |
| Graphite · refining talk (quarterly) ↔ Artificial graphite (quarterly) | Volume | 0 / 4 | 0 | — | — |
Shifts are one view; whole series are another. We correlated the month-to-month changes in talk and trade at every lag from −12 to +12 months.
Bars right of zero: talk moves first. Left: trade moves first. Dashed lines: the 95% band from 2,000 shifted series.
A bar has to clear the dashed band to beat chance, and after correcting for testing six pairings none does. A Granger test (does one series help predict the other?) gives a few small p-values, mainly mining talk and natural graphite. With 32 tests on 53 months they are exploratory, not findings.
Feb 2022 – Jul 2026 overlap (54 months) · first differences · Holm correction across six pairings
| Pairing | Best lag | r | 95% band | p | p (corrected) | Granger talk → trade | Granger trade → talk |
|---|---|---|---|---|---|---|---|
| No stage named · all texts ↔ Anode-grade graphite · China share | -2 | 0.36 | ±0.42 | 0.52 | 1.00 | 0.222 | 0.067 |
| No stage named · all texts ↔ Anode-grade graphite · Volume | +5 | 0.35 | ±0.42 | 0.97 | 1.00 | 0.464 | 0.091 |
| No stage named · the press ↔ Anode-grade graphite · China share | -10 | 0.30 | ±0.42 | 0.97 | 1.00 | 0.404 | 0.729 |
| No stage named · the press ↔ Anode-grade graphite · Volume | -10 | 0.35 | ±0.47 | 1.00 | 1.00 | 0.631 | 0.857 |
| Refining talk ↔ Artificial graphite (3801.10) · China share | -6 | 0.41 | ±0.56 | 0.59 | 1.00 | 0.106 | 0.395 |
| Refining talk ↔ Artificial graphite (3801.10) · Volume | +3 | -0.45 | ±0.46 | 0.08 | 0.50 | 0.005 | 0.179 |
| Refining talk ↔ Refined natural graphite · China share | +2 | -0.32 | ±0.45 | 0.80 | 1.00 | 0.107 | 0.378 |
| Refining talk ↔ Refined natural graphite · Volume | -11 | -0.22 | ±0.47 | 1.00 | 1.00 | 0.354 | 0.572 |
| Components talk ↔ Anode-grade graphite · China share | +9 | 0.25 | ±0.55 | 1.00 | 1.00 | 0.456 | 0.670 |
| Components talk ↔ Anode-grade graphite · Volume | +6 | 0.23 | ±0.54 | 1.00 | 1.00 | 0.269 | 0.396 |
| Mining talk ↔ Natural graphite, all · China share | -4 | 0.50 | ±0.53 | 0.18 | 1.00 | 0.090 | 0.001 |
| Mining talk ↔ Natural graphite, all · Volume | +1 | 0.39 | ±0.47 | 0.70 | 1.00 | 0.014 | 0.357 |
So we used quarters. Graphite claims naming no stage rose in late 2023 with the licences, and fell in late 2025, after the trade drop.
Top: share of graphite paragraphs claiming a dependency with no stage named. Bottom: China’s share of EU anode-grade graphite imports.
Talk shifts: up in Q4 2023 (same quarter as a trade rise: together), down in Q4 2025 (two quarters after the trade drop: trade first). Random dates match as often (p = 0.99). Graphite refining talk had no lasting shift.
Graphite paragraphs, adjusted for coding errors · Comext quarterly to Q2 2026
| Quarter | No stage named | Graphite paragraphs |
|---|---|---|
| 2022-Q1 | 0.0% | 3 (thin) |
| 2022-Q2 | 0.0% | 3 (thin) |
| 2022-Q3 | 10.6% | 11 (thin) |
| 2022-Q4 | 10.7% | 11 (thin) |
| 2023-Q1 | 12.8% | 40 |
| 2023-Q2 | 4.6% | 31 |
| 2023-Q3 | 0.0% | 51 |
| 2023-Q4 | 45.1% | 80 |
| 2024-Q1 | 7.4% | 13 (thin) |
| 2024-Q2 | 0.7% | 31 |
| 2024-Q3 | 27.1% | 12 (thin) |
| 2024-Q4 | 35.3% | 23 (thin) |
| 2025-Q1 | 10.5% | 33 |
| 2025-Q2 | 24.5% | 20 (thin) |
| 2025-Q3 | 28.5% | 15 (thin) |
| 2025-Q4 | 5.6% | 29 (thin) |
| 2026-Q1 | 0.0% | 10 (thin) |
| 2026-Q2 | 1.6% | 19 (thin) |
Anode-grade and refined graphite share one customs code. We split it by price per tonne, and Chinese prices sit close to the line between the two bands.
China’s share of EU import value for CN 2504.10, by price band and in total, by quarter
In Q2 2025 the anode band falls about 70 points while the refined band rises. The whole code falls about 15 points: that is the safer size of the drop, and the one the main page uses.
Eurostat Comext via Pillar 1 · bands set by the Pillar 1 team
| Quarter | Anode band | Refined band | Whole code | Anode t from China | Refined t from China |
|---|---|---|---|---|---|
| 2022-Q1 | 34.8% | 40.7% | 34.9% | 635 | 2,439 |
| 2022-Q2 | 56.5% | 46.4% | 38.7% | 1,306 | 3,170 |
| 2022-Q3 | 49.2% | 48.2% | 38.0% | 1,324 | 4,578 |
| 2022-Q4 | 28.3% | 58.2% | 36.8% | 813 | 5,798 |
| 2023-Q1 | 25.7% | 36.0% | 26.8% | 833 | 1,565 |
| 2023-Q2 | 32.6% | 45.5% | 28.7% | 832 | 1,803 |
| 2023-Q3 | 37.2% | 41.1% | 36.0% | 926 | 1,045 |
| 2023-Q4 | 46.7% | 14.4% | 41.6% | 1,486 | 410 |
| 2024-Q1 | 58.9% | 49.8% | 47.7% | 753 | 1,591 |
| 2024-Q2 | 48.5% | 62.3% | 40.8% | 678 | 3,448 |
| 2024-Q3 | 75.5% | 1.7% | 53.2% | 2,021 | 23 |
| 2024-Q4 | 61.4% | 56.9% | 48.0% | 590 | 1,996 |
| 2025-Q1 | 82.7% | 29.1% | 55.1% | 2,284 | 676 |
| 2025-Q2 | 14.0% | 62.7% | 40.1% | 122 | 2,608 |
| 2025-Q3 | 21.5% | 48.8% | 35.9% | 138 | 1,307 |
| 2025-Q4 | 36.9% | 55.4% | 35.1% | 462 | 873 |
| 2026-Q1 | 27.8% | 53.7% | 41.5% | 344 | 1,302 |
| 2026-Q2 | 14.3% | 61.6% | 41.4% | 150 | 2,315 |
EU imports of anode-band natural graphite from China, tonnes per quarter
Chinese anode-band tonnes fell from over 2,000 a quarter to between 120 and 460 from Q2 2025. The fall in volume is real; its size in shares is inflated by the band line.
Eurostat Comext via Pillar 1