Forecasting Chinese Exports

This is the first edition of Plain Sight, a research letter about the most consequential economic story hiding in public data. Take government targets, customs data, and corporate filings at face value, and model the world without the ideological discount that makes every consensus forecast wrong in the same direction.

Fifteen years of being wrong the same way

Between 2010 and 2025, the International Energy Agency published an annual forecast for global solar deployment. Every single year, for 15 consecutive years, actual deployment exceeded the forecast, systematically and in one direction, often by 50-100%

The 2016 World Energy Outlook projected annual global solar additions of roughly 75 GW a year – every year, flat through 2030 and beyond. Actual additions reached 599 GW in 2024: eight times the projection, eight years out. Earlier vintages drew the same flat line at lower levels; each new edition raised the plateau to meet reality, then held it flat again. Each was published with the institutional authority of the world's most respected energy body, and each was wrong by a factor that grew with every passing year.

China's mid-2025 switch to market pricing pulled installations forward and has cooled the pipeline into 2026; this may be the first year in two decades in which global additions do not grow. A policy pause is not a cost reversal; the learning curve that broke a decade and a half of forecasts is intact.

The IEA was not staffed by fools. The analysts understood solar technology. They tracked installation data. They published rigorous methodology notes. And yet they missed – relentlessly, in one direction – because they were modeling the wrong phenomenon.

The IEA modeled solar as a policy technology. In their framework, deployment was a function of government subsidies, feed-in tariffs, renewable portfolio standards and political will, so each year's forecast assumed the current deployment rate was near a ceiling set by policy.

The IEA's defense is on the record: the WEO scenarios are conditional projections – if stated policies, then this – not forecasts. But the conditions were revised upward every year as policies strengthened, and the misses stayed one-directional anyway. A scenario machine with unbiased errors misses in both directions; 15 consecutive same-direction misses is not a property of conditionality. It is a property of the model.

Solar was on a learning curve: each doubling of cumulative production reduced module costs by roughly 24%, making the next unit cheaper regardless of what any government decided. A technology that cost $4.00 per watt in 2008 cost under $0.10 per watt by 2025. The IEA was modeling a political phenomenon. Reality was an industrial one.

Three years of being wrong the same way

China's export story is the IEA pattern in real time, without the 15 years of hindsight: three consecutive years of the consensus missing in the same direction, for the same reason, with the miss growing each time.⁴

2024: the consensus expected export stagnation. China's 2023 exports were $3.38 trillion¹, down 4.6% in dollar terms after the post-Covid normalization though roughly flat in yuan. Sell-side forecasts expected low single-digit growth at best, citing weak global demand and early tariff signals. The actual result: exports hit $3.58 trillion – 5.9% growth, well above the consensus range.

2025: the consensus expected tariff-driven contraction. Trump's return to office in January 2025 dominated the outlook. Liberation Day tariffs hit in April. US tariffs on Chinese goods peaked at 145%. The consensus narrative was unambiguous: exports would shrink. The actual result: exports grew 5.5% to $3.77 trillion.¹ Exports to the US collapsed 20%, but exports to ASEAN (+13.4%), Africa (+25.8%), and the EU (+8.4%) more than compensated. Brad Setser at the Council on Foreign Relations noted that the IMF's current account forecast was off by roughly double.

2026: the consensus expected slowing growth. Citi's December 2025 outlook forecast export growth slowing to around 3%. Then the January–February data landed. Exports surged 21.8% – against a consensus forecast of 7.1%.² A miss by a factor of three.

The pattern is structural, identical to the IEA solar pattern: the consensus models a political phenomenon – tariffs constrain, rebalancing moderates – where the reality is industrial, China's manufacturing cost advantage compounding with scale, learning, and energy cost reduction. Each unit of capacity built makes the next unit cheaper. The surplus is on a learning curve, not a business cycle.

Where the growth is coming from

The geographic pattern inside the January–February 2026 data is hard to square with the tariff narrative.

Africa up 49.9%. ASEAN up 29.4%. The EU up 27.8%. Latin America up 16.4%. The US actively putting up trade barriers and erecting the Trump tariff wall is the only destination in decline.

Exports and surpluses unlikely to retrench

The consensus view is that the surplus will moderate from its record $1.19 trillion¹ due to flat or declining exports is implying that several of four things must occur:

1. China's cost advantage must erode. It is not eroding: energy costs are falling (solar is now the cheapest electricity source in history), labor productivity is rising and automation is accelerating, so the learning curve compounds.

2. Category expansion must stop. It is not stopping. China is capturing new industrial categories: lower-end semiconductors, industrial robots, CNC equipment, adding new layers of exportable dominance on top of the existing base.

3. The Level 2-to-3 transition must stall. It is not stalling. Three billion people live at $2–8/day, Rosling's Level 2³; as electricity costs fall and urbanization continues, 1.5–2 billion will cross into the $8–32/day range – Level 3, where the consumption basket is almost entirely Chinese-manufactured.

4. The 2026 energy crisis must not accelerate the timeline. It is accelerating it. Every country that watched Hormuz close is rethinking energy sovereignty. The buyer for Chinese solar panels, batteries, and EVs is not a climate activist – it is a finance minister who just watched their diesel import bill triple.

For the consensus to be right, all four must hold. For us to be right, any one can fail.

There is no overcapacity. There is capacity on a cost curve the consensus cannot model, meeting demand from populations whose purchasing power it has not yet learned to count.

1. The export series currency basis. China's General Administration of Customs publishes in both yuan and US dollars, and the two disagree by enough to matter. Every figure in this post is on the dollar basis.

   • 2025: exports US$3.77 trillion, up 5.5%; imports US$2.58 trillion, roughly flat; the trade surplus US$1.19 trillion, a record. December exports ran 6.6% above the previous December. GACC annual release, January 2026.
   • The same 2024 year grew 7.1% in yuan – 25.45 trillion yuan – against 5.9% in dollars. The gap is the currency, not the cargo. A reader checking the yuan series will get a different number for every growth rate quoted here, which is why the basis is named.
   • 2024 back-solves to US$3.57–3.58 trillion from the 2025 level and the 2025 dollar growth rate. 2023's US$3.38 trillion, and the 4.6% dollar decline that year against a roughly flat yuan outturn, are carried from the same dollar series.
   ↩

2. The January–February 2026 print. Exports of US$656.58 billion, up 21.8% on the same two months of 2025 – the largest gain in four years, against 6.6% in December and 5.5% across 2025. Exports to the United States fell 11% over the same window.

   • The consensus was a Reuters poll of economists at 7.1%.
   • So the two-month 2025 base is US$539.1 billion, and the consensus rate implies US$577.3 billion. The outturn cleared it by US$79.3 billion, 13.7% above where the consensus had the period. That is the third pair in Exhibit 2.
   • Customs reports January and February together, because the timing of the lunar new year makes either month alone uninterpretable.
   ↩

3. Rosling's levels. Hans Rosling, Factfulness (2018): four levels divided at roughly $2, $8 and $32 a day in 2011 purchasing power, holding populations of about 1, 3, 2 and 1 billion. Those are the book's figures, and the three billion at Level 2 is the second of them.

   • Gapminder has since moved the underlying data to 2017 and then 2021 purchasing power while holding the boundaries, so the current level populations differ from the book's. The claim here needs only the order of magnitude and the direction of travel.
   • These are income levels. They are not the energy tiers on the workings page above, and the two are not interchangeable.
   ↩

4. Figures on Exhibit 2. The chart is drawn in growth rates rather than dollar levels to accommodate the year-to-date 2026 bars.

   • Actual growth, dollar basis: 2024 up 5.9%, 2025 up 5.5%, January–February 2026 up 21.8%.
   • Consensus: the 2026 figure is the published Reuters poll. The 2024 and 2025 figures are a Plain Sight compilation of the sell-side ranges current at the start of each year, expressed as one rate and rounded to the nearest whole percent so that nothing turns on a decimal – about +1% for 2024, about −8% for 2025.
   • The dollar gaps above each pair come from the same compiled levels: US$3.42 trillion against US$3.58 trillion in 2024, US$3.30 trillion against US$3.77 trillion in 2025, and US$577 billion against US$657 billion over January and February 2026.
   ↩