China Subsidy Estimates Diverge as OECD and IMF Measure Different Forms of Support

Two major estimates, two different pictures

China's industrial support amounted to 4.4% of its gross domestic product in 2023, according to an International Monetary Fund working paper published on August 8, 2025. Its authors found that support had remained broadly stable as a share of the economy. An OECD announcement on June 1, 2026, presented a different picture: industrial subsidies across 15 industries worldwide reached $108 billion in 2024, with Chinese companies receiving an average of three to eight times the support received by companies in OECD countries over 2005 to 2024.

The OECD also linked subsidies to 60% of the global market share gains recorded by expanding Chinese companies, compared with 22% worldwide. Those findings put government assistance at the center of a debate about manufacturing competition, excess production capacity and trade restrictions.

Yet the headline figures do not measure the same thing. The IMF estimates an economy wide fiscal equivalent relative to GDP. The OECD tracks assistance received by 525 large industrial companies and measures it against revenue. Their samples, periods and definitions differ, so a stable IMF estimate and a rising OECD measure need not contradict each other.

A separate discrepancy concerns the $108 billion figure itself. Some reporting attributes that entire amount to Chinese companies. The OECD's announcement explicitly identifies it as the total across the 15 industries covered globally. Its 1.3% subsidy to revenue ratio also describes that broader sample, rather than OECD countries alone.

The evidence supports a conclusion that Chinese industry receives substantial state assistance. It does not establish a single, universally comparable price tag, nor does it settle whether disputed estimates provide a sound basis for particular trade restrictions.

What the IMF counted

The IMF's industrial policy working paper, written by Daniel Garcia-Macia, Siddharth Kothari and Yifan Tao, estimates four instruments: cash subsidies, tax benefits, subsidized land and subsidized credit. Their combined equivalent fiscal cost was 4.4% of GDP in 2023.

• Cash subsidies: 2.0% of GDP.
• Tax benefits: 1.5% of GDP.
• Land subsidies: 0.5% of GDP.
• Subsidized credit: 0.4% of GDP.

Cash subsidies and tax benefits together account for 3.5 percentage points of the 4.4% total, or about 80%. Credit accounts for roughly 9%. That breakdown matters because uncertainty about interest rate subsidies affects only one part of the IMF estimate, although it can carry more weight in other assessments.

The researchers used WIND financial statements covering more than 5,000 listed companies from 2010 through 2023. They also examined approximately 1.6 million government land sale transactions recorded by the Ministry of Natural Resources over the same period.

The phrase equivalent fiscal cost requires care. A cash grant is a recorded payment. A tax concession represents revenue not collected, while a land or credit subsidy is an estimated benefit against a comparison price or interest rate. Combining them produces a measure of support, not a tally of checks written by government.

What the OECD measured

The OECD's industrial subsidies announcement describes its MAnufacturing Groups and Industrial Corporations database, known as MAGIC. It follows 525 of the world's largest industrial firms from 2005 through 2024, tracking government grants, tax concessions and finance provided below market terms.

The $108 billion recorded for 2024 equaled 1.3% of covered company revenue. That was the second highest ratio in the series, behind 2009. The OECD explains that falling sales during the global financial crisis pushed up the 2009 ratio, while financial rescues of large OECD carmakers largely explain that year's spike in subsidized borrowing.

This illustrates why the denominator matters. A subsidy to revenue ratio can rise because assistance increases, sales fall, or both. A subsidy to GDP ratio responds to a different set of changes. Neither measure can be substituted for the other without reconciling their coverage.

Why borrowing costs are the disputed calculation

Interest rate subsidies are harder to measure than direct grants because researchers must estimate what a company would have paid without state assistance. That hypothetical rate is the benchmark. The difference between it and the actual borrowing cost determines the estimated benefit.

The IMF identifies credit subsidies by comparing effective interest rates in manufacturing with those in other sectors, controlling for financial variables and company type. Those controls seek to separate preferential treatment from differences in company finances and ownership.

Even so, such comparisons depend on whether the benchmark captures the relevant differences between borrowers. They may also miss assistance shared across both the favored sector and its comparison group. Conversely, treating ordinary differences in financing conditions as government support could overstate the benefit.

The OECD announcement identifies finance below market terms as a subsidy category but does not set out the detailed borrowing benchmarks needed to evaluate this criticism. A methodological objection is therefore grounds for closer examination, not proof that a particular estimate is inflated.

The issue is consequential. A March 2025 assessment cites earlier OECD research in which more than half of measured support for covered Chinese companies came from borrowing below market rates. Where credit dominates the total, the choice of benchmark can substantially change the result.

Stable support can still change shape

The IMF finds that the total size of industrial support remained broadly stable over time, while tax benefits became more prominent after the pandemic and other instruments declined slightly. Stability in the total does not mean every policy stayed unchanged.

The cash subsidy rate examined in the paper fell from 2.4% in 2013 to 2.0% in 2023. Its measure of tax benefits relative to profits moved in the opposite direction, rising from 4.4% to 6.3%. These are different measures with different denominators and should not be added together.

Tax benefits are calculated against China's top statutory corporate income tax rate of 25%. The gap between that rate and effective sector tax rates captures several kinds of concessions, including investment incentives and discounts associated with special economic zones. It is a broad estimate rather than an inventory of individual programs.

The land calculation uses a more local comparison. The IMF compares government land prices paid by manufacturing firms with prices paid by firms in other sectors within one kilometer in the same year. This helps account for location and timing, though interpreting the remaining price gap still requires assumptions about comparability.

A broader assessment finds support beyond grants

Rhodium Group's state support assessment, published on March 17, 2025, by Camille Boullenois, Agatha Kratz and Daniel H. Rosen, argues that conventional subsidy measures capture only part of China's assistance to industry. Its account includes implicit government guarantees, discounted industrial park facilities, selective regulation and state influence over commercial decisions.

The report also provides a useful check on claims of rapidly rising direct grants. It says grants to listed Chinese companies grew 67% between 2016 and 2023, while GDP grew 70%. On those figures, grants rose in absolute terms but did not rise as a share of GDP, consistent with the IMF's broader finding of stability.

Other estimates collected in the assessment illustrate the range of measurements. Earlier OECD research covering 13 sectors from 2005 to 2019 put support at 4.5% of covered Chinese company revenue, against 0.69% for OECD companies, roughly a 6.5 times difference. A CSIS estimate placed Chinese industrial policy spending at $248 billion, or 1.73% of GDP, in 2019.

These are not interchangeable estimates. They concern different periods, instruments and populations. A medical technology study cited in the assessment estimated support averaging at least 5.3% of company revenue during 2017 to 2022, a sector specific result rather than a national total.

Rhodium also cites tax concessions equivalent to 0.75% of revenue for China based firms, compared with 0.32% for OECD firms, 0.28% for Indian firms and 0.47% elsewhere. Its finding that 98.58% of Chinese A share companies received subsidies in 2020 shows how widespread grants were, but says nothing by itself about the amount each recipient received.

Market share is not the same as productivity

The OECD links subsidies to approximately 22% of market share gains among expanding firms worldwide and 60% among expanding Chinese firms. The Chinese figure is about 2.7 times the global figure. It does not mean subsidies explain 60% of China's total exports or manufacturing output.

The OECD says assistance increased market shares without producing significant gains in productivity or profitability. Secretary-General Mathias Cormann described the concern in its June 1 announcement:

Large and persistent industrial subsidies can distort global markets, creating unfair competitive advantages and contributing to excess supply capacity,

The IMF examines a different outcome: domestic resource allocation. Its model estimates that industrial policy reduced aggregate total factor productivity by about 1.2% relative to a baseline without industrial policy, potentially lowering GDP by up to 2%. Total factor productivity measures how efficiently labor and capital are combined to produce output.

Those results require stronger modeling assumptions than the IMF's subsidy accounting. They use policies implemented from 2009 to 2018 and allocation outcomes in 2018, rather than measuring the effects of the 2023 subsidy total. The paper also leaves out possible benefits from correcting market failures, such as knowledge spreading between firms, and does not estimate international spillovers. Its productivity result is therefore not a complete balance sheet of industrial policy.

Ownership and trade rules complicate the picture

State ownership is associated with greater assistance in both studies, but neither confines industrial policy to state enterprises. The OECD finds that companies with state ownership above 25% received substantially more support on average than private competitors, particularly through grants and preferential borrowing.

The IMF finds that state enterprises tend to receive lower interest rates and higher cash subsidy rates after accounting for their sectors. Its estimated cash subsidy advantage averages 0.7 percentage points. Private companies, however, receive greater tax benefits and dominate sectors favored by industrial policy.

Rhodium's assessment argues that assistance can also operate through private actors responding to government priorities. That broader concept of state support does not automatically match the legal definition of a subsidy that can be challenged under trade rules.

The World Trade Organization's Agreement on Subsidies and Countervailing Measures requires evidence of a financial contribution by a government or qualifying body that confers a benefit. Actions against subsidies also depend on requirements concerning specificity and adverse effects. An economic estimate of national support cannot substitute for evidence about the program, recipient and harm involved in a particular case.

This distinction cuts both ways. Some policies may affect competition without fitting neatly into a conventional subsidy total. Some amounts counted in an economic study may not establish a legal basis for duties. Broad figures should guide investigation, not be treated as automatic authorization for trade barriers.

The dates and evidence that matter next

The research sequence helps explain the debate. Rhodium's assessment predates the IMF paper, while the OECD release adds company observations through 2024:

• March 17, 2025: Rhodium Group publishes its assessment of conventional subsidies and wider state support.
• August 8, 2025: The IMF working paper estimates support at 4.4% of GDP in 2023 and finds broad stability over time.
• June 1, 2026: The OECD announces its latest findings and public availability of the MAGIC database.
• June 3, 2026: The OECD Ministerial Council Meeting is scheduled to begin, with industrial policy among its stated themes.

The debate is also feeding into European trade discussions. German Economy Minister Katherina Reiche visited China in the week before the OECD announcement to call for fair competition. China's role in global markets and its relationship with the European Union were expected to feature at the G7 summit in mid June, though no exact summit date is specified here.

The most useful next step is a reconciliation of methods: which firms are covered, which instruments are included, how borrowing benchmarks are selected and how company observations are expanded into national estimates. Public access to MAGIC is a concrete improvement, but access must be accompanied by enough methodological detail to reproduce and compare the calculations.

No reconciled estimate or deadline for producing one has been announced in these accounts. The evidence supports scrutiny of substantial Chinese state assistance, while leaving its precise scale and trend unsettled. Policymakers should distinguish documented support and demonstrated harm from uncertain totals before using those totals to justify protectionism.

Key Points

• The IMF estimates four forms of Chinese industrial support at 4.4% of GDP in 2023, with the total broadly stable over time.
• The OECD's $108 billion figure covers its global sample across 15 industries, not Chinese firms alone.
• Chinese firms received three to eight times the support received by OECD country firms over 2005 to 2024, according to the OECD.
• GDP based and revenue based estimates cannot be compared directly without reconciling coverage and methods.
• Interest rate benchmarks are a central source of uncertainty, especially where subsidized borrowing accounts for much of measured support.
• Economic subsidy estimates do not by themselves establish the legal grounds for trade duties.