Before subsidy figures are used to justify new trade measures, the benchmark behind below-market borrowing should be made comparable across countries. The central issue is that the OECD/MAGIC treatment makes China's BMB estimate LPR-based, and this can make measured borrowing support several times larger than estimates based on more comparable interbank or government-bond benchmarks.

Chinese electric vehicles, solar panels, steel and other manufactured goods are again at the centre of an international trade argument. In Washington and Brussels, the language is familiar: subsidies, overcapacity and market distortion. What is newer is the growing reliance on firm-level subsidy databases to give the debate more quantitative grounding.

Precision, however, depends on measurement. A subsidy figure that looks like a hard fact may rest on a modelling choice that most readers never see. That is why the interest-rate benchmark behind one component of the OECD's Manufacturing Groups and Industrial Corporations (MAGIC) database deserves more scrutiny. Put more directly, if the MAGIC calculation treats China's below-market borrowing as LPR-based, the reported China subsidy gap can be magnified by the choice of benchmark itself.

The OECD's 2026 MAGIC report separates industrial support into three categories: government grants, corporate income-tax concessions and below-market borrowings (BMB). The OECD reports that firms based in China received, on average, three to eight times more government support than OECD-based firms between 2005 and 2024, and that almost 60 per cent of Chinese firms' global market-share gains can be explained by subsidies. The OECD also states that its results do not prejudge WTO or countervailing-duty assessments. That headline comparison depends heavily on BMB, the component most exposed to the interest-rate yardstick used for China.

The decomposition matters. A large part of the reported China-other-region gap, especially in the 2005-24 average, comes from BMB. This means the debate is not only about how much support China receives, but also about whether the OECD/MAGIC BMB yardstick makes that support look several times larger.

That distinction matters. Grants and tax concessions are closer to observable fiscal support. BMB is a constructed estimate. It compares a firm's actual borrowing cost with a counterfactual market benchmark, and then applies the gap to a measure of debt. If the benchmark used for China is LPR-based while comparator benchmarks are closer to interbank funding or low-risk sovereign rates, measured Chinese BMB can be mechanically inflated.

The issue is not whether China uses industrial policy. So do the United States, the European Union, Japan, South Korea and many others. The narrower issue is whether countries are being compared with the same yardstick.

A comparability question can be seen in related OECD materials. In its 2024 work on industrial subsidies, the OECD compares banks' average corporate lending rates with applicable base rates: base rates for OECD banks are described as currency-specific interbank rates, while China's benchmark lending rates refer to the Loan Prime Rate (LPR) after 2019 and earlier PBOC benchmark lending rates before then. China is therefore assessed against a lending-rate benchmark, while the comparator for OECD banks is closer to wholesale interbank funding.

This distinction is not semantic. The LPR is calculated by the National Interbank Funding Center from bank quotations over open-market operation rates and serves as a pricing reference for bank lending. By contrast, an interbank rate reflects banks' wholesale funding quotations, while a government-bond yield is closer to a low-risk sovereign reference. Treating these different financial-market layers as interchangeable can turn an LPR-based China estimate into a larger measured BMB number, even before any substantive policy judgement is made.

A simple example shows why this matters. The People's Bank of China's 2026 first-quarter monetary policy report states that, in March 2026, the one-year LPR was 3.0 per cent. The same report shows that 48.81 per cent of new renminbi loans that month were priced below the LPR, 9.31 per cent at the LPR and 41.88 per cent above it. If the LPR is used as a stand-alone cutoff, nearly half of new lending would begin on the "below benchmark" side of the calculation. That is too broad a signal for industrial-policy support, because below-LPR lending may reflect borrower quality, collateral, bank competition or ordinary loan-market pricing. This example shows why an LPR-based BMB measure can overstate policy support if below-LPR pricing is treated too broadly as below-market borrowing.

The point is not that below-LPR loans can never involve policy support. Some may. The point is that the LPR is not equivalent to an interbank funding cost or a low-risk sovereign benchmark. Using LPR as China's base rate before adding firm-level credit-risk and guarantee adjustments therefore risks making ordinary commercial lending appear as LPR-based BMB.

Because the firm-level MAGIC database, loan-level inputs, credit-spread assumptions, debt-currency structure and full calculation details are not public, the OECD's headline numbers cannot be independently replicated from public materials. But a narrower test is possible: hold the firms, debt proxy and simplified formula constant, and change only the base rate. That test asks a direct question: how much larger does the China BMB estimate become when the calculation is LPR-based rather than government-bond- or SHIBOR-based?

Using Chinese A-share listed firms from 2019 to 2024, mapped broadly to MAGIC-related industries, such a benchmark-sensitivity exercise gives large differences. In the Top-147 large-firm sample, the LPR-based BMB measure used to approximate the OECD/MAGIC treatment is 16.68 times the China government-bond-based estimate and 5.65 times the SHIBOR-based estimate; in 2024 alone, it is 17.89 times and 7.38 times larger, respectively.

The pattern is not confined to one sector. In the broad MAGIC-industry mapping, the LPR benchmark produces higher measured BMB than both alternative base rates across all covered sectors. The gap also widens in more recent years, which is precisely when subsidy and overcapacity claims have become more politically salient. This is therefore not a sector-specific outlier; it is a systematic feature of using an LPR-based benchmark for China.

These numbers do not mean that the OECD's full estimate would fall by exactly the same multiples. Its full method includes credit-risk spreads, guarantee adjustments, maturities, sector definitions and non-listed firms. But they do show that the base-rate choice alone can magnify the measured China BMB by several-fold, enough to change how the headline subsidy gap should be interpreted.

There is also a scope limitation. MAGIC focuses on three firm-level instruments: grants, corporate tax concessions and BMB. Industrial policy is broader. It can include procurement rules, tariffs, non-tariff barriers, local-content requirements, investment screening, consumer subsidies and regulatory preferences. Some of these tools are used by many economies, but may not appear in the same way in a firm-level subsidy database.

The same caution applies to the OECD's claim that subsidies explain almost 60 per cent of Chinese firms' market-share gains. That figure is not directly observed. It is a model-based attribution that uses a subsidy-intensity variable including BMB and an instrumental-variables estimate. If China's BMB component is magnified by an LPR-based benchmark, then the almost-60-per-cent attribution is also benchmark-sensitive and should not be read as a standalone fact about China's manufacturing performance.

Industrial-policy debates need measurement. But measurement should discipline policy conclusions, not pre-judge them. Before subsidy estimates play a larger role in trade measures or claims about global imbalance, benchmark rates, sample boundaries and sensitivity tests should be made transparent. A database can inform debate only if its yardsticks are comparable; otherwise an LPR-based BMB approach may magnify China's measured subsidy gap before the policy debate even begins.

Yuankun Li, Assistant Research Fellow, Institute of World Economics and Politics, Chinese Academy of Social Sciences