2.1 The Chinese contract sample

Despite its size and rapid growth, China's foreign lending remains opaque. The Chinese government has resisted pressure to reveal the size, scope, and terms of its claims on low- and middle-income countries (Dreher et al. forthcoming). This secrecy has been a focus of public debate for a long time. For example, in 2011, a group of bilateral and multilateral creditors urged China to comply voluntarily with information disclosure standards of the OECD's Development Assistance Committee (DAC). The Chinese authorities rejected the call, arguing that the “principle of transparency should apply to north-south cooperation, but [...] it should not be seen as a standard for south-south cooperation.” [9] Ten years later, China still does not participate in the OECD's Creditor Reporting System, the OECD Export Credit Group, or the Paris Club—although its recent commitment to the G20 Common Framework may indicate an evolving position.

To address the evidence gap, we collaborated with AidData—a research lab at the College of William and Mary—to identify all publicly accessible loan agreements between Chinese government institutions and state-owned banks, and sovereign borrowers from low- and middle-income countries. [10] In preparation for the 2021 update of its Global Chinese Official Finance Dataset, AidData recently revised its Tracking Underreported Financial Flows (TUFF) methodology. It now requires the systematic implementation of search procedures that enable the identification of loan agreements in the debt information management systems, official registers and gazettes, and parliamentary websites of low- and middle-income countries.

The implementation of these search procedures resulted in the retrieval of 100 loan agreements between Chinese government institutions and state-owned banks, and government entities in 24 borrowing countries, with a total commitment value of $36.6 billion. All of these loan agreements were drawn from publicly available sources. The dataset consists of every contract that AidData retrieved during the implementation of its updated TUFF methodology; no contract was excluded. The dataset represents about 5% of total estimated Chinese lending between 2000 and 2017 (Horn et al. 2019 estimate total direct lending commitments of $560 billion). As shown in Table 1, our sample includes concessional and non-concessional debt contracts entered into between 2000 and 2020 with China's two main policy banks (China Eximbank and CDB), state-owned commercial banks (Bank of China, Industrial and Commercial Bank of China), state-owned enterprises (e.g., Sinohydro, China Machinery Engineering Corporation) and the central government.

China Eximbank accounts for 76 of the 100 loan agreements in the sample. Out of these 76 loans, 66 are concessional lending instruments (so called Government Concessional Loans or Preferential Buyer Credits). [11] The sample only includes 8 loan contracts with CDB, two of which were co-financed with Chinese state-owned commercial banks. [12] However, the small number of CDB loan agreements in our sample corresponds to substantially larger financial commitment amounts: 8 contracts represent 44% of the overall lending volume captured in our sample. [13] Compared to China Eximbank and CDB lending, loans issued by China's state-owned commercial banks, state-owned enterprises, and the central government are small. Taken together, these three groups of lenders account for only 16 percent of the contracts and 13 percent of the lending volume in our sample.

The distribution of creditors within our sample is broadly in line with creditor composition in the datasets of Morris et al. (2020) and of Horn et al. (2019). In both of these datasets, China Eximbank and CDB represent by far the two most important sources of China's international financial commitments. Morris et al. (2020: 46) analyze 1,046 Chinese government loans to 130 countries between 2000 and 2014 and find that 80 percent of the loans during this period came from China Eximbank and 14 percent came from CDB. While China Eximbank makes a far larger number of loans than CDB, the average size of its loans is substantially smaller than those made by CDB. As a result, in the Morris et al. (2020) dataset, loan commitments from China Eximbank account for 55 percent of overall lending and loan commitments from CDB lending account for 36 percent of overall lending. Very similar patterns are observed in the dataset constructed by Horn et al. (2019). In their dataset, loans from China Eximbank account for 60 percent of the total by number, and 33 percent of the total by monetary value, while loans from CDB represent 18 percent of the total by number, and 42 percent of the total by monetary value.

Figures 1 and 2 further demonstrate that our sample is broadly dispersed across world regions (also see Table A1 in Appendix I for a detailed country list). 47 percent of the loan agreements in the sample are with government borrowers in Africa, and another 27 percent are with government borrowers in Latin America and the Caribbean. The remaining loans in the sample were made to government borrowers in Eastern Europe (11%), Asia (10%), and Oceania (5%).

Previous studies demonstrate that Africa, Asia and Latin America are the primary destinations for Chinese government loans (Horn et al. 2019; Dreher et al. 2021). Therefore, our sample likely under-represents Chinese lending to Asia. If Chinese contracts varied systematically by region, this would be of concern for the external validity of our sample. We do not find evidence, however, that contracts differ significantly by geographic region. In fact, our analysis of Chinese contracts reveals that the lending terms are highly standardized, and largely predetermined by the identity of the creditor and the type of lending instrument.

The distribution of our sample is broadly consistent with the global distribution of Chinese government loans by borrowing country income level. Borrowers in middle-income countries account for 90 percent of the loan agreements in the sample, while borrowers in low-income (8%) and high-income countries (2%) account for the remainder. By comparison, the analysis of Horn et al. (2019) demonstrates that, since the turn of the century, Chinese state-owned creditors have made approximately 75 percent of their loans to middle-income countries, 19 percent to low-income countries, and 6 percent to high-income countries.

We conclude from these summary statistics that our sample of 100 contracts is generally in line with the composition of China's global portfolio of loans to government borrowers. While certain subgroups may be over- or under-represented in the data, there is no indication of systematic bias in the composition of the sample. More importantly, our analysis of contracts shows that Chinese lending instruments are highly standardized and do not exhibit significant variation by borrower country, region, or income bracket. Nonetheless, it bears emphasis that this dataset of Chinese debt contracts does not constitute a random sample; the contracts included in our analysis were selected because they were the only ones publicly available at the time of this study.

2.2 The benchmark debt contract sample

China is a state-led economy, and its approach to government-to-government lending often differs from those of OECD governments. We observe a greater array of lenders, terms, and policy mandates in Chinese debt contracts than we do with other governments. To account for the fact that China does not have an obvious peer group within the sovereign lending ecosystem, we established four separate peer groups for benchmarking purposes: (1) bilateral creditors from the OECD, which include government agencies and instrumentalities that coordinate through the OECD's Development Assistance Committee (DAC); (2) non-OECD bilateral creditors (i.e., government creditors from countries that do not participate in DAC, such as the Gulf states or India); (3) multilateral creditors, including regional development banks; and (4) commercial banks. We refer to creditors in the first three categories collectively as official creditors.

China is not unique for failing to publish detailed information about its lending terms. There is no uniform public disclosure standard or practice for bilateral official lenders, although many governments and most multilateral institutions publish information about their lending at varying levels of detail, and many share such information with a subset of other creditors. To address this information gap, particularly when it comes to comparability at the contract level, we constructed a benchmark sample for a single sovereign borrower, Cameroon, which to our knowledge is the only developing country that has maintained a publicly accessible database (via http://dad.minepat.gov.cm/) of its project-related loan contracts with all external creditors. [14]

This database, in principle, should cover all of the Government of Cameroon's project-related loan contracts with external creditors. However, some of the contracts that are stored in the database are incomplete or in an unreadable condition. Also, for some of the loans in the database, no contractual documentation was available. [15] In total, we were able to retrieve 142 debt contracts with 28 different creditors—8 commercial banks, 10 bilateral creditor agencies from 10 different countries (including 3 official export credit agencies), and 11 inter-governmental organizations—that are listed in detail in Table A2 in Appendix I. The composition of the sample is summarized in Figure 4. The International Development Association (IDA), the Islamic Development Bank (ISDB), the African Development Bank (AfDB) and Agence Française de Développement (AFD) are heavily represented in the benchmark set, which likely reflects their institutional mandates and Cameroon's colonial history.

In order to gauge whether the loans in the benchmark sample and the China sample are reasonably comparable, we explore whether they were designed to achieve similar purposes. Figure 5 summarizes the sectoral composition of loans in the China and the benchmark sample. We find considerable overlap between the samples: in both the benchmark and the China sample, most loans financed projects and programs in the transportation, energy, and water supply sectors. These three areas account for roughly 60 percent of the contracts in the Chinese sample and for 50 percent of the contracts in the benchmark sample.

The types of borrowers represented in the benchmark sample and the China sample are also remarkably similar. In both samples, the borrower is almost always the central government (99% in the benchmark sample and 94% in the China sample). The remaining borrowers are state-owned enterprises, a government agency, and two special purpose vehicles (project companies) with explicit guarantees from the central government (see Figure 6 below).

In our main analysis, we compare the lending terms of the 100 Chinese contract sample with the 142 contracts of the 28 benchmark creditors. Our analysis therefore entails comparisons of Chinese contracts with borrowers worldwide (including Cameroon) to benchmark contracts with Cameroon as a single sovereign debtor country. This comparison introduces scope for bias if Chinese lending contracts with sovereign borrowers outside Cameroon differ substantially from Chinese loan contracts with Cameroon. Fortunately, this does not appear to be the case, since the terms of Chinese lending contracts in our sample are highly standardized across countries. As we show in Appendix III, all of our main findings hold when we limit our comparison to Chinese and benchmark creditor contracts with Cameroon.

A related concern is that specific characteristics of Cameroon as a borrower could make contracting practices there difficult to compare to contracting practices elsewhere. In other words, our benchmark sample could differ for other developing and emerging market countries. All the evidence we have seen reinforces our impression of standardization by creditor, with banks typically following the LMA template, and bilateral and multilateral creditors relying heavily on their respective general terms and conditions. In order to reflect the high degree of standardization, we have created a typology of contract characteristics by creditor in Appendix II. Nonetheless, we cannot rule out that Cameroon's contracts differ systematically in some way, given the dearth of systematic data and the lack of publicly available sovereign loan contracts. With more data, we could expand our analysis to a broader range of benchmark countries and their contracts with bilateral, multilateral and private creditors.

2.3 Methodology and coding approach

In order to facilitate comparisons between the terms and conditions in the sample of Chinese loan contracts and the benchmark loan contracts, we developed a set of variables that allow for systematic categorization. The variables that we selected follow the structure of the Loan Market Association (LMA) template for single currency term facility agreements in developing markets. We took this approach because, while there is no “international standard” for bilateral official sovereign debt contracts, a variety of private and some official lenders—inside and outside of China—use the LMA template as a basis for their contract design. In total, we code 100 variables, which we group into eight analytical categories:

1. Principal Payment Terms: These variables capture the loan facility, its debt maturity, grace period, repayment schedule, and currency of denomination, as well as bilateral cancellation and debtor prepayment rights.

2. Interest and Fees: The variables in this category identify the interest rate, timing and currency of interest payment, as well as the commitment fee and the arrangement or management fee.

3. Additional Payment Obligations: This is a catch-all qualitative variable created to capture any payment obligations of the borrower not included in (1) and (2), such as currency conversion costs, indemnification costs, or charges related to contract renegotiation or enforcement. This category also captures stabilization or increased cost clauses that require the borrower to compensate the lender for increased costs resulting from policy changes in the borrower country or the creditor country.

4. Credit Enhancement: These variables capture information about third-party credit enhancements and security interests. They cover guarantees (including guarantor identity and guarantee terms and conditions), formal and informal security interests, and escrow and special accounts [16] (including account funding and management arrangements).

5. Conditions, Covenants, and Modification Terms: These variables identify the debtor's commitments apart from the promise to repay the debt with interest. They include commitments addressing status (subordination and pari passu clauses, if any), information disclosure, negative pledge, collective and bilateral restructuring procedures, if any, and linkages to any other contracts, including commodity sales and project operation.

6. Events of Default: These variables identify events of default and their consequences, including acceleration of repayment, suspension of disbursement, and contract termination. Varieties of the cross-default clause feature prominently in the sample and the benchmark contract set.

7. Assignment and Delegation: These variables capture whether and under what conditions the sovereign debtor or the creditor may assign its rights or delegate its obligations to a non-party.

8. Governing Law and Enforcement: These variables identify the law that governs the contract and the agreed dispute settlement forum and procedure (including arbitration and any applicable procedural rules). They cover sovereign immunity waivers, if any, and separately describe waivers of the debtor's immunity from lawsuits and of the immunity of its assets from attachment before and after a court judgment, where applicable.

Dealing with missing information: Some loan contracts in our data are incomplete or reference additional agreements that are not available to us. In particular, 18 percent of contracts in our sample are missing one or more pages. In these cases, the table of contents can usually be used to infer which parts of the contract are missing. [17] If a contract is incomplete, we cannot rule out that a certain clause exists in the contract. We therefore assign “missing values” rather than “zeros” in such cases. We do so to ensure that contracts with missing information do not enter the sample statistics for the incidence of a specific clause.

Another related problem emerges if a contract references additional legal documents that are not available to us. By way of example, a creditor's “general conditions” can form an integral part of the contract, but they were not consistently available to us. While none of the Chinese contracts refer to separate general conditions, most multilateral and some bilateral creditors use them. In 42 cases (17% of contracts), general conditions are not available to us. In the Chinese contract sample, seven contracts represent only technical and economic cooperation agreements that leave most contractual details to the final (undisclosed) loan agreement. All of these transactions are flagged in our dataset. When coding the information from these contracts, we again assign “missing values” if we cannot find a clause in the contract, since we cannot rule out that the clause is included in the creditor's general conditions or in the final version of the loan agreement.

Similarly, contracts in our samples often reference separate confidentiality agreements, account agreements, or security documents that form part of the transaction and define important terms. In these cases, we know that a certain clause or arrangement exists, but have only limited insights into the details. We discuss these limitations of our study in the presentation of our findings below.

Coding approach: We employed two independent research teams—one at the Georgetown University Law Center and another at the Kiel Institute for the World Economy—to apply a consistent set of variable definitions and coding rules and procedures to the 100 contracts in the Chinese sample and the 142 contracts in the benchmark set. When the coding determinations of these teams were identical, we accepted their assigned values as final. When the two teams reached different coding determinations, we enlisted the support of a senior researcher to identify the underlying source of the discrepancy and apply expert judgment to assign a final value.

We provide more information on the definitions of our variables and the coding rules and procedures that we used to construct the variables in Appendix VI. Our dataset can be accessed at https://www.aiddata.org/how-china-lends. Digitized and scanned PDF copies of the loan agreements that we coded can also be accessed at https://www.aiddata.org/how-china-lends.

3.1 Confidentiality: Chinese contracts contain unusual confidentiality clauses

Sovereign debt contracts with Chinese lenders are more likely to include confidentiality clauses than similar contracts with most other creditors. All CDB contracts and 43% of China Eximbank contracts include such clauses. Some form of confidentiality clause is also common in the benchmark sample: 39 percent of contracts by multilateral creditors, one third of contracts by bilateral creditors and one third of commercial bank contracts include confidentiality undertakings. While benchmark contracts impose confidentiality obligations primarily on the lenders, contracts in our Chinese sample impose them on the borrowers. Confidentiality clauses in Chinese lenders’ contracts are also far broader in scope than those in the benchmark set, covering all the terms, and even the existence of the debt itself.

Figure 7 shows a pronounced shift towards greater secrecy in Chinese lending contracts that is driven by the widespread introduction of confidentiality clauses in China Eximbank contracts around 2014. Whereas only one of 37 China Eximbank contracts prior to 2014 contains a confidentiality clause, all China Eximbank contracts after 2014 include confidentiality clauses.

All China Eximbank contracts beginning in 2014 use substantially the same confidentiality clause, reproduced in Table 2 below. The CDB contracts in our sample follow the LMA template, and also reference separate confidentiality letters. The only publicly available letter of this kind is designed to protect the confidentiality of contract negotiations: it covers all aspects of the transaction and related negotiations, applies to both parties, and expires six months after the contract is signed, or one year after negotiations break up. [18] In our benchmark sample, one-third of the commercial loan contracts use formulations that are nearly identical to or slightly narrower than the confidentiality clause in the LMA template reproduced in Table 2 below.

The China Eximbank confidentiality clause excerpted here binds the sovereign debtor. It applies to the entire agreement and, potentially, to a broader set of dealings between the debtor and China Eximbank “in connection with” the contract. On the other hand, the clause contains a carve-out that would allow the sovereign debtor to make disclosure required by law. It is unlikely, however, that this carve-out would be broad enough to allow a debtor to disclose China Eximbank contract terms to its Paris Club creditors, since the Paris Club process and output are at best “soft law.” The LMA template, in contrast, imposes more robust non-disclosure obligations on the lenders (“finance parties”), which presumably reflects the fact that banks obtain confidential business information in the course of their credit assessment before they issue a loan. The debtor nondisclosure obligations are narrowly drawn, limited to banks’ funding costs, and expressly justified by reference to securities regulations. [19]

The Arab Bank for Economic Development in Africa, the Islamic Development Bank, the OPEC Fund for International Development and the Kuwait Fund for Arab Economic Development all have a version of the clause, reproduced in Table 2, that requires the borrower to keep the lender's documents and correspondence confidential. It is much narrower than the China Eximbank clause. Two official bilateral debt contracts in our benchmark sample, both from Agence Française de Développement (AFD), have confidentiality terms resembling China Eximbank’s, committing the debtor not to disclose any of the agreement.

Expansive debtor confidentiality undertakings that extend beyond contract negotiation present multiple political and debt management problems. First, they try to hide government borrowing from the people whose taxes are bound to repay it. Second, they impede budget transparency and sound fiscal management. Third, they hide the sovereign's true financial condition from its other creditors. Creditors may charge the government higher interest rates to reflect the uncertainty and potential for subordination. Fourth, potential for hidden debt can impede debt restructuring. At this writing, Zambia's bondholders are refusing to proceed with debt renegotiation citing insufficient information about China's claims on the country (Bavier and Strohecker 2021). More broadly, a lack of trust in the debtor's financial reporting can derail crisis response and recovery.

We have not found any evidence of judicial enforcement of the confidentiality clauses, but we have identified at least one instance of CDB invoking them in response to a video obtained and released by investigative journalists that revealed the terms of Ecuador's multi-billion dollar oil-backed debt to CDB. The release of the video shortly after the deal was signed prompted public debate about the new borrowing (Zurita et al. 2020). In response, the head of CDB's Resident Mission in Ecuador wrote to his counterpart in Ecuador's Ministry of Finance, complaining about the borrower's apparent breach of the confidentiality letter, called on the Ecuadorian government to launch a leak investigation, and demanded that it take measures to mitigate the reputational damage to CDB caused by the video. [20] The CDB letter also implicitly threatened to withhold future financing if the borrower did not adequately address the incident.

3.2 Seniority and Security: Chinese lenders use formal and informal collateral arrangements to maximize their repayment prospects

Chinese state-owned banks use liens, escrow and special accounts much more extensively than either the official or the commercial lenders in the benchmark set. Whereas 29% of the debt contracts in the Chinese sample use one or more of these devices, only 7% of OECD bilateral creditors and 1% of the multilateral creditors in the benchmark set do so. No contract with non-OECD bilateral creditors in the benchmark set uses any of these security devices.

Figure 8 further shows that collateralization practices vary across Chinese lending institutions: 6 out of 8 CDB loans in our sample benefit from some form of security interest, but only 22% of the China Eximbank loans do. [21] The distinct mandates of these institutions may help explain the difference: CDB operates without formal subsidies from the central government, and probably has stronger incentives than China Eximbank to write contracts to minimize repayment risk. Because CDB makes larger loans than China Eximbank, it must manage additional credit and liquidity risks. In our sample, the average face value of a China Eximbank loan is $200 million, while the average CDB loan has a face value of $1.5 billion. Any and all of these features would lead CDB to use credit enhancements when lending to risky borrowers.

The most common way of securing repayment in the Chinese contract sample is the use of escrow or special accounts. Sovereign borrowers commit to maintain and fund bank accounts either at the lending institution or at a bank “acceptable to the lender” throughout the life of the loan, and to route through these accounts project revenues and/or cash flows that are unrelated to the project funded by the loan. Debt contracts describe the accounts as part of the debt repayment process; however, they function above all as a security device.

The debt contracts in our sample and benchmark set reference separate account agreements that appear to contain most of the detailed provisions governing the accounts. We have access to only one such agreement in our sample, and therefore cannot provide a systematic assessment of how these accounts work. However, many of the debt contracts contain enough detail to convey a general sense of account operation.

• All account arrangements with available information require the debtor to maintain a minimum account balance; in most cases, the minimum is the annual principal, interest, and fees due under the debt contract.

• In 70% of the Chinese transactions with a special account, all revenues from the associated projects must be deposited in the account.

• In 38% of the Chinese account arrangements, the account is financed from unrelated sources, either instead of or in addition to project revenues. In our sample, these sources include the export revenues from oil (Ecuador and Venezuela), bauxite (Ghana), and revenue from financial assets (Costa Rica). Other contracts require the borrower to provide sufficient funding from sources that are not limited to project revenues, but they do not specify the source(s).

• In 5 CDB contracts (with Argentina, Ecuador and Venezuela), the lender also has the ability to block the debtor from withdrawing the funds. These contracts expressly limit the debtor's withdrawal rights to those specified in the account agreement. We have only one such account agreement: between CDB and a state-owned development bank (BANDES) in Venezuela. Under this agreement, BANDES is not permitted to make any withdrawals from the Collection Account during the 35-day period prior to any repayment date or if withdrawals would violate the minimum debt service coverage ratio. CDB, on the other hand, is “entitled at any time and without notice to BANDES, to […] appropriate, set-off or debit all or parts of the balances in the Collection Account to pay and discharge all or part of BANDES’ liabilities to CDB.” [22] In the 4.7 billion USD loan to Argentina's Ministry of Finance by CDB, ICBC and BOC, all project revenue is collected in a Project Trust Account and withdrawals are limited to pay for fees, loan repayments, and specified project expenses, in the prescribed order of priority. [23]

Box 1 illustrates the use of special accounts in a 2010 loan from CDB to the government of Ecuador. The loan agreement is linked to an oil purchase agreement between PetroEcuador and PetroChina. Our sample includes two additional oil-backed CDB loans to Ecuador and a similarly structured lending arrangement between CDB and BANDES in Venezuela.

Only 3 of the 142 contracts in our benchmark set have comparable account arrangements.

• Cameroon's contract with Commerzbank Paris requires the government to deposit payments from the UN into an escrow account and to maintain a minimum account balance equal to a year's principal and interest payments due.

• A contract between Cameroon and the French government development agency (AFD) requires the borrower to deposit royalty payments equal to 150% of annual payments due from Cameroon in an account formally pledged to AFD, and to maintain a minimum account balance equal to two loan payments due.

• Provisions relating to a reserve account in a 2003 African Development Bank loan with Cameroon are illegible in the version of the contract that was published by the government of Cameroon.

Special accounts of the sort described here are standard in limited-recourse project finance.[26] Their function is to help lenders manage credit, operational, transfer, and legal risk, among others. Such accounts appear to be rare in bilateral official and multilateral lending practice. A handful of high-profile exceptions prove the rule:

• U.S. emergency loans to Mexico beginning in 1982, and again in 1994, required Mexico to route proceeds from state oil sales through Mexico's account at the Federal Reserve Bank of New York; however, even under the more restrictive 1994 agreement, Mexico could withdraw the funds so long as it was not in default (General Accounting Office 1996). In addition, the United States committed to buy oil from Mexico at a discount price. The 1994 arrangement addressed a mix of financial and political imperatives, notably U.S. congressional opposition to the extraordinary assistance package.

• Budget transparency, fiscal management, and related governance concerns led to the establishment of a London-based escrow account, which featured prominently in the World Bank's ill-fated financing for the Chad-Cameroon pipeline. Chad's petroleum revenues from the new pipeline flowed through the accounts between 2004 and 2006. Withdrawals had to be approved by a public oversight board before funds could be transferred to Chad's treasury. The World Bank suspended most disbursements to Chad and froze the account in 2006, after Chad changed its law and, according to the Bank, took fiscal measures in contravention of its agreement. Settlement later the same year allowed Chad to make a partial withdrawal; a subsequent review concluded that the arrangement was fragile and ultimately ineffective as a policy tool (World Bank 2006, 2009).

• Offshore accounts and revenue pledges were popular in 19th and early 20th century sovereign lending, before the advent of restrictive sovereign immunity. Such arrangements had mixed success in improving creditor repayment prospects (e.g., Borchard and Hotchkiss 1951, Wynne 1951, Maurer 2013).

Account arrangements of the sort we identify, when used in full-recourse sovereign lending, can pose multiple policy challenges. First, they encumber scarce foreign exchange and fiscal revenues. Second, the encumbrance can be easy to hide. This follows from the fact that banks’ set-off rights against their account holders are usually found in background laws and regulations, and do not require a formal pledge, registration, or disclosure on the part of the debtor. Any additional contractual undertakings may also be kept confidential. In contrast, a lender that wishes to take effective security interest in a physical asset must enter into a separate agreement and make a public filing to get a priority claim against the asset. Third, undisclosed routing of revenue flows to special accounts impedes the accuracy of a debt sustainability analysis and multilateral surveillance work. If a substantial portion of a country's revenue streams is earmarked for the benefit of a single creditor, conventional measures of debt sustainability are likely to overestimate the country's true debt servicing capacity to all creditors. In balance of payments crises, this can undermine IMF programs, adding to the effective adjustment burden of the country and deepening haircuts for other creditors in the event of a debt restructuring. Fourth, control over revenue flows can give the lender considerable bargaining power vis-à-vis the debtor and other creditors, [27] which can translate into political leverage in the context of government-to-government lending.

Historical and contemporary experience with special accounts suggests that they may offer lenders limited, if any, protection. Cash-strapped debtors usually do not hesitate to redirect payment flows. However, it may be more difficult to do so if the creditor is also the source of those payment flows, as in the case with the oil-backed loan contracts discussed earlier. Special accounts may also help creditors deflect political pressure at home, reassuring shareholders and voters that risky debt would be repaid.

In contrast to the prevalence of special accounts, only 5 of the Chinese loans in our sample explicitly reference a formal security interest or pledge. In these cases, pledged assets include financial instruments (in Costa Rica and Honduras), mining rights (in the DRC), and project output and equipment, as well as shares in a project company (in Sierra Leone). We find little evidence in our contract sample that China's state-owned banks routinely use physical infrastructure—like a seaport or a power plant—as collateral. This finding stands in contrast to the prominent media and political narrative, which holds that China's lending practices are designed to appropriate strategic physical assets in poor countries (for a critique of this narrative, see Bräutigam and Kidane 2020).

The only debt contract in our sample that appears to entail a pledge of physical assets is for a syndicated loan from China Eximbank and ICBC to upgrade and expand a seaport in Sierra Leone. The contract contains several references to pledged collateral in the form of physical or financial assets that could be transferred to the lender and liquidated in the event of default. However, we were not able to obtain any of the security agreements referenced in this or any other contract in the Chinese sample, and do not have enough information to define the pledged assets or the operation of the collateral scheme with specificity.

In summary, Chinese lenders in our sample appear to prefer collateral in the form of bank accounts, with contractual minimum balance requirements to ensure that the lender would have cash to seize in the event of default. By comparison, collateral in the form of illiquid physical assets is more burdensome to secure and sell, harder to keep confidential, and more likely to draw unfavorable media coverage and political controversy.

3.3 Seniority and “No Paris Club”: Chinese contracts enable lenders to seek preferential repayment without saying so

Only two debt contracts with China's state-owned banks formally claim senior status: an ICBC loan to Argentina and a collateralized loan from ICBC and China Eximbank loan to Sierra Leone, discussed earlier. On the other hand, all contracts in our Chinese sample commit the borrower to exclude the debt from any multilateral restructuring process, such as the Paris Club of official bilateral creditors, and from “comparable treatment” that the Paris Club requires the debtor to seek from its other creditors. Such a promise is unlikely to be enforceable in the court of any major financial jurisdiction; however, combined with other contract terms, it could give the lender more bargaining power in a crisis.

A typical “No Paris Club” clause in the Chinese contract sample is reproduced below:

[T]he Borrower hereby represents, warrants and undertakes that its obligations and liabilities under this Agreement are independent and separate from those stated in agreements with other creditors (official creditors, Paris Club creditors, or other creditors), and the Borrower shall not seek from the Lender any kind of comparable terms and conditions which are stated or might be stated in agreements with other creditors. [28]

Three CDB contracts with Argentina's Ministry of Economy contain an even more expansive variation on the theme:

The Borrower shall under no circumstances bring or agree to submit the obligations under the Finance Documents to the Paris Club for restructuring or into any debt reduction plan of the IMF, the World Bank, any other multilateral international financial institution to which the State is a part of, or the Government of the PRC without the prior written consent of the Lender. [29]

Comparability of treatment is one of six core Paris Club principles; it covers both commercial and official bilateral creditors that are not members of the Paris Club. [30] The stated objective of comparability is burden-sharing: governments are loath to grant relief if their taxpayers end up subsidizing other creditors instead of helping countries in distress. Comparability has long been a pillar of the international financial architecture, and has shaped international sovereign debt markets for decades (see Gelpern 2004; Schlegl, Trebesch and Wright 2019). In theory, a sovereign debtor that fails to secure comparable treatment from official or private non-Paris Club creditors risks losing its Paris Club relief, and potentially its IMF and other multilateral financing. In practice, no Paris Club treatment has ever been undone for lack of comparability, in part because it is assessed in the aggregate, treating all non-Paris Club creditors as a group, and defined loosely enough to accommodate a wide variety of creditor concessions.

A debtor that flouts the comparability principle and follows through on its preferential treatment promise to CDB or China Eximbank would be in serious breach of Paris Club norms, and would likely damage its relationships with the IMF, the World Bank, and other official and commercial creditors. As part of the Common Framework agreed in November of 2020, China and other G20 members that are not part of the Paris Club agreed to restructure their claims on the poorest sovereign borrowers in tandem with the Paris Club, implying broadly the same terms, including comparability of treatment for both official and commercial claims. Although the G20 statement is too vague to amount to a definitive commitment, it stands in tension with 74% of the Chinese contracts in our sample, which explicitly reject burden sharing with other creditors.

No contracts with private or non-Paris Club official creditors in our benchmark set include similar clauses. The closest analogue in recent sovereign debt history may be the term forswearing future debt restructuring in the 1990s Brady Bonds (restructured bank claims that represented substantial debt relief):

[The sovereign] will not, directly or indirectly, seek any restructuring or rescheduling of the Bonds or any provisions thereof, nor will it, directly or indirectly, seek or request any loans, advances, extensions of credit or other financial accommodation from any holders of the Bonds or any affiliate thereof based upon such holdings.[31]

Although such clauses became ubiquitous in Brady Bonds, they were also widely understood to be unenforceable: it is hard to see what remedy a court might fashion for breach of such a term. On the other hand, a lender could do real economic damage to the debtor if it has access to a “special account” and threatens to use self-help, seizing the account should the debtor attempt collective restructuring negotiations. Moreover, “No Paris Club” clauses could have expressive and political functions, which may be more potent in the government-to-government context.

3.4 Policy influence: Cancellation, acceleration, and stabilization clauses can give Chinese lenders sway over policies in the borrowing country

Chinese lenders in our contract sample retain the right to cancel the loan and demand immediate repayment under a wide range of circumstances, including political and economic developments not directly connected to the lending relationship. In contrast, the sovereign debtor's exit options are limited once the contracts are signed. For instance, cross-default and cross-cancellation clauses in some of the Chinese contracts trigger if the debtor takes action adverse to “any PRC entity” in the borrowing country. Such terms position Chinese state-owned institutions to act in concert, amplifying their collective bargaining power vis-à-vis the developing country. All CDB contracts include the severing of diplomatic relations with China as an event of default. Events of default in 90% of the contracts in our Chinese sample include broadly defined policy changes in the creditor or in the debtor country. Normally in the event of default, the lender can accelerate principal and interest repayment. Default triggers of the sort we have identified in Chinese debt contracts potentially amplify China's economic and political influence over a sovereign borrower. We elaborate below.