156 economies, ~1,200 products, thirty years of bilateral trade, distilled into one map. Copper marks what Indonesia exports competitively.
US$259.5B EXPORTED TO 216 PARTNERS · UN COMTRADE 2023Export shares by product chapter. Scrub the years: watch oil, gas, and wood give way to palm and coal — then the nickel complex bloom.
Indonesia’s top export chapters, racing from 1995 to 2024. Copper bars are the nickel chain. Watch coal and palm overtake oil and gas — then processed nickel climb from nowhere after the 2020 ban.
Indonesia banned raw nickel-ore exports in 2014, relaxed the ban, then closed it for good in 2020. Ore exports collapsed; processed nickel — ferroalloys, mattes, stainless — surged. That part of the story is true, and here it is in the data.
But the same file records tonnes. Between 2013–2015 and 2021–2024, the dollars rose 14.6× — and the tonnage rose 34.6×.
The average price of a tonne of Indonesia's processed nickel fell 58%, from US$5,454 to US$2,307 a tonne. That is not a price shock; it is a change in what is being shipped. The basket moved from 75% concentrated nickel mattes to 43% ferro-nickel and 36% stainless steel — products that are mostly iron by weight. Downstreaming is real, and it moved Indonesia into a much bulkier, much cheaper-per-tonne product. Every dollar earned now carries 2.4 times the material, and therefore that much more furnace, freight and fuel. Chapter 10 finds the fuel.
Values in current US$; HS92 codes 2604 (ore) vs 7202, 7501–7503, 7218–7220 (processed); quantities in metric tonnes. The code list survives a check: at HS6, ferro-alloys containing no nickel are 0.17% of the basket, and narrowing it to ferro-nickel plus HS 75 alone still gives 10.0× in value. Source: BACI.
Indonesia ranks #78 in economic complexity in 2024 — versus #76 a decade earlier. Fifteen times the processed-nickel exports; a rank that barely moved.
Complexity rewards the diversity and sophistication of the whole export basket. Downstreaming multiplied the value of one chapter, but a single-commodity climb does not diversify a basket — so Indonesia held its place in the mid-seventies while its neighbours moved up: over the same decade Vietnam +12, the Philippines +3, India +7 places. Every line here is computed from the same corpus.
Read the rank as a band, not a seat. An independent review of this chapter found that the index is densely packed where Indonesia sits: in 2024, 11 economies lie inside 0.095 of the complexity score — about 9% of one standard deviation — so ten places of rank separate almost nothing. This series moves a mean of 4.3 places a year with a largest single-year swing of 14. The decade difference above is real in direction and not resolvable in size. What survives is the shape: downstreaming did not diversify the basket, and that is true on our numbers and on Harvard's.
Rank check, restated. The like-for-like test is a rank correlation across all economies, not a comparison of one country's seat. Re-running our complexity ordering against the Harvard Atlas's published HS92 index gives Spearman ρ between 0.946 and 0.987 in every year from 1995 to 2024 — a check written into this repository before any result was read and, until the review, never run. On Indonesia's own seat the two pipelines differ by more: ranked over the Atlas's own country list we place Indonesia #73 in 2024 against the Atlas's #69, and a mean of 11.1 places lower across the thirty years. The review section below has the full comparison.
Products Indonesia does not yet export competitively, placed by how close they sit to what it already makes (density) and how sophisticated they are (PCI). The upper-right corner is the strategic frontier.
Chapters 01–06 are export-only by definition, not by oversight: economic complexity is built on revealed comparative advantage, which asks what a country can sell the world better than the world sells it. Imports cannot enter that calculation — an ECI computed on imports would measure purchasing power, not capability. So that analysis stays exactly as it is.
But the case is named for a network, and a network has counterparties. The four chapters that follow are the half the map was missing: who actually buys Indonesia, what Indonesia cannot make for itself, who it runs surpluses and deficits against — and whether the nickel story above survives contact with the import data.
China took 22.5% of Indonesia’s goods exports in 2024, against 3.9% in 1995. But Japan took 26.0% in 1995 — more than China takes now.
We expected to find a customer base narrowing onto Beijing. The data says something more interesting. Measured across the whole customer list, Indonesia is less concentrated today than it was thirty years ago: the Herfindahl–Hirschman index of export partners fell from 1097 in 1995 to 588 by 2015 as Japan’s post-war dominance dissolved and India, Korea, Vietnam and the Philippines each took a slice.
Then it reversed. Since 2015 the index has climbed back to 844, and essentially all of that climb is one country: China roughly doubled its share in a decade. So the honest reading is a U — two decades of diversification, one decade of re-concentration, with the second half of that curve still well short of where Indonesia started. Below, each band is a partner’s share; the strip beneath is the concentration itself.
Bands are the partners that ever reach a year’s top twelve; the rest are pooled as “all other partners”. Hong Kong, Macao and Taiwan are separate BACI reporting territories and are not folded into China — see the footer. The concentration strip is deliberately not zero-based: the whole signal is a swing between roughly 48.5% and 58.4%, which a zero baseline would flatten into a straight line. Read it for shape, not height.
Indonesia imported US$231.5B of goods in 2024 — 32.4% of it from China alone. Its supplier concentration is now 1307 against 844 for its customers.
This is the finding chapter 07 sets up, and the review corrected its shape. Both sides of the network are U-shaped — the supplier side simply turned earlier and went further. Import concentration fell from 846 in 1995 to 594 in 1999, stayed below its 1995 level for 19 of 30 years, crossed back above it in 2015, and has since climbed to 1307 — now 54% above where it started, while the customer side is still 23% below. Indonesia sells to a world it widened and has been re-narrowing for a decade; it buys from one it is re-narrowing faster. China’s share of what it buys (32.4%) runs well ahead of China’s share of what it sells (22.5%).
The bill itself is an industrial shopping list, not a consumer one: machinery, fuels, chemicals, food. Note the top line — the largest single chapter Indonesia imports is mineral fuels, bought by one of the world’s largest energy exporters, which is the refining gap in one number. Below the area chart, each chapter split by who supplies it, because import dependence is only politically live when it has an address.
Indonesia’s trade balance is not one number but a structure: surpluses with the economies that consume its commodities — India, the United States, the Philippines, Japan — and deficits with the economies that sell it the machines. The sharpest case is China, which buys more from Indonesia than anyone else and still sells it more than it buys, leaving Indonesia in deficit with its single largest customer. In 1995 that relationship was roughly balanced. Scrub the years to watch the structure form.
Top 18 partners by total trade in the selected year. Bars are exports minus imports, both FOB. Hover for the two sides separately.
Chapter 03 showed processed-nickel exports growing roughly fifteen-fold while ore exports went to zero. That is an export-side fact. It implies an import-side one: you cannot build dozens of smelters without buying the furnaces, the electrodes and the inputs to feed them. If the story is true, it should be visible in what Indonesia bought.
So we pre-registered the test before looking, and committed the thresholds to the repository before computing a single import number. Base window 2013–2015, peak window 2021–2024. H1: imports of machinery and electrical equipment (HS 84 + 85) rise at least 50%. H2: that rise beats total import growth by at least 10 points, so a general import boom cannot pass on its own. H3: at least one smelter input — coke, graphite electrodes, chrome ore, stainless scrap — at least doubles. All three is a pass; two is partial; fewer is a failure, and it publishes as one.
It failed. Capital-goods imports grew +29% while all imports grew +27% — the smelter build-out is essentially invisible in the machinery bill.
That is a real result and it stands. Indonesia’s furnace purchases are either too small against a US$231.5B import bill to register, spread across too many years to show up in a window comparison, or booked as investment flows the goods data never sees. On the capital-goods hypothesis, the answer is no.
But look at what the test found on its way to failing. The feedstock did not grow — it detonated. Chromium ores imports rose 124× between the same two windows, and imported coal — in the country that exports more coal than any other on earth — rose 14× to US$3.2B a year. Coke, quicklime and limestone flux all more than tripled. Ferrous scrap, which we expected to rise, fell.
So the smelters are visible in the import data after all; we simply pointed the instrument at the wrong shelf. Downstreaming did not show up as a one-off equipment purchase. It showed up as a permanent new import dependency — US$3.6B a year of coal, coke, chrome and flux that Indonesia did not previously buy, set against US$26.5B a year of additional processed-nickel exports. 13.6% of the headline gain is spent buying the inputs that make it possible — and this bill scales with the 34.6× rise in tonnage from chapter 03, not with the 14.6× rise in receipts. Two things it does not count: the US$0.8B a year of ore exports given up, and the domestic coal burned in the smelters' captive power plants, which never appears in an import statistic. The figure is a floor.
Corrected after review. This paragraph previously set the input bill against "US$28.5B a year of additional processed-nickel exports" and called the ratio "around an eighth". US$28.4B is the peak-window level; the increment over the base window is US$26.5B, which makes the ratio nearer a seventh. The direction of the argument is unchanged; the arithmetic was wrong and is now computed rather than typed.
A prior we got wrong, on the record. Coal and limestone were entered as controls — Indonesia has both domestically, so we predicted they would not move. They moved the most. The likely reason is geographic and metallurgical rather than statistical: the smelters sit in Sulawesi, far from the Kalimantan and Sumatra coal basins, and much Indonesian coal is thermal grade rather than the metallurgical input a furnace wants. That explanation is an inference, not a measurement, and is flagged as such.
Two totals, two published benchmarks, tolerances fixed before either was computed. The import side reconciles almost exactly. The export side does not: BACI puts Indonesia’s goods exports roughly a sixth above what Indonesia itself reports, every year we can check. The export-reconciliation check fails, and rather than quietly widening the tolerance we are printing the gap.
The gap is not an artefact of this extension. It is systematic across years, it is spread across chapters rather than sitting in one product, and it is not caused by aggregate or “nes” partner codes — every destination in the file is a real economy. The leading explanation is that BACI is a mirror-reconciled reconstruction: it harmonises what Indonesia says it shipped against what its 213 partners say they received, and for Indonesia the partners consistently report receiving more. That gap is a well-documented feature of Indonesian commodity trade, not a bug in this pipeline — and our totals reproduce the case’s existing export figures to nine significant figures, so the discrepancy is between BACI and Jakarta, not between us and BACI.
Does this undermine chapters 01–06? Largely no, and for a structural reason: economic complexity is computed from revealed comparative advantage, which is a ratio of shares. A level difference that is roughly proportional across products cancels in the numerator and denominator. It does matter for any statement about levels — which is why the hero quotes UN Comtrade’s US$259.5B for 2023 and not BACI’s figure, and why every dollar total in chapters 07–10 is labelled as BACI’s.
An independent read of the same data against the published literature. It runs the rank correlation against the Harvard Atlas that this repository registered and never computed — it passes at ρ 0.975 — and then shows why the rank this page leads with still cannot bear the weight: 11 economies sit inside 9% of one standard deviation of Indonesia. It reads the tonnes the case never opened, and finds 34.6× the material for 14.6× the money. And it corrects three claims this page used to make — about the neighbours' decade, about the supplier side, and about the arithmetic of the nickel gain.
Read the review article →