GEMs and the 600-800 Billion Dollar Data Dividend

For years now, stakeholders from the private, public, and official sectors have campaigned for the release of data from the Global Emerging Markets Database (GEMs) consortium, which tracks the default and recovery rates of borrowers from 29 development finance institutions (DFIs) including the major multilateral development banks (MDBs). This persistence paid off: the GEMs consortium just released its most comprehensive dataset, disclosing 40 years of data for lending to governments and 30 years of data for lending to state-owned enterprises and the private sector.

One motivation behind the public data push was so other stakeholders, including investors and credit rating agencies (CRAs), could see that lending to emerging markets and developing economies (EMDEs) is less risky than often perceived for transactions involving MDBs and DFIs. Indeed, based on the latest data set, a key observation of the GEMs consortium is that EMDEs' sovereign credit ratings significantly overstate the default risk of private projects structured by MDBs and DFIs. Data on state-owned enterprises show more volatility but are still low, with average default and recovery rates of 2.6 percent and 88 percent respectively between 1994 and 2024. For the private sector, default rates averaged 3.5 percent and recovery rates averaged 73 percent, although median recovery rates were above 90 percent.

There is some correlation between income levels and default rates, with Europe and Central Asia usually on the low end, and higher rates prevailing in sub-Saharan Africa (SSA): as an example, average defaults for lending to the private sector in Europe and Central Asia were 2.7 percent compared to 6 percent for SSA. But recovery rates do not appear associated with country income levels: median recovery rates for SSA are 97.8 percent and 86.5 percent for Europe and Central Asia.

This data disclosure has already proven hugely consequential because Standard & Poor’s has announced that factoring newly available GEMs data into their assessments of MDB portfolio risk had a big impact on MDB capital requirements under their methodology: they now estimate that the major MDBs are sitting on an extra $600-$800 billion in lending capacity that could be deployed over the next ten years without putting their ratings at risk. This extraordinary result underscores why data transparency is so important. Just by providing more information about their own credit track records, MDBs can drive much more accurate assessments of risk—to their benefit and to the benefit of private investors.

The campaign for GEMs

The first major public push for GEMs data came in the 2022 report, “Boosting MDBs’ Investing Capacity,” (also known as the Capital Adequacy Report, or CAF). The report was prepared by an independent experts’ group commissioned by the G20 to identify ways to use MDB capital more efficiently while maintaining high institutional ratings. Included among its recommendations was a call to provide public access to GEMs data, with the authors arguing that: “Wider access to GEMs would build investor understanding and strengthen risk assessment, expanding investor interest and better risk transfer opportunities for MDB assets, improving capital efficiency and scaling capacity for MDBs” (italics ours).

GEMs data are privileged: only the member DFIs have access. Stakeholders argued that because the data are collected by donor-funded institutions, they should be in the public domain. This proved easier said than done. Securing access to GEMs data at a level granular enough to be of value to CRAs and investors requires a highly disaggregated and fully scrubbed database, harmonized across institutions, while still maintaining business confidentiality for individual projects.

Initial data disclosures were too aggregated to be of much use. But after considerable internal deliberations and external urging, the GEMs consortium released data by region, sector and year for sovereign and sovereign-guaranteed loans and state-owned enterprises (i.e., companies with at least 50 percent public ownership) and by country for private lending. To preserve anonymity in line with non-disclosure agreements required by underlying contracts, there are some gaps- especially at the sector level- but overall, these data offer considerable insights into the risk profiles of MDB and DFI borrowers. To enable even more fulsome datasets going forward, we believe these contracts should explicitly allow for more data disclosure.

Key findings—sovereign and sovereign guaranteed lending

What GEMs data show across the board is that MDB lending risks are relatively low, including in most low and lower-middle-income countries (LICs and LMICs), for all categories (sovereign, SOE, and private). Starting with sovereign lending, data over four decades show that loan repayments have been consistently near or at 100 percent, with average default rates of only 0.77 percent. High repayment rates are to be expected- the MDBs benefit from preferred creditor status and once a country defaults to an MDB, access to capital markets is usually cut off. Restoring credibility as a borrower requires making good on these payments, which is why recovery rates are also very high, at 95 percent on average, regardless of region or lending volumes. (See Figures 1 and 2.)

Key findings—lending to state-owned enterprises

For state-owned enterprises (i.e., municipalities and companies that are at least 50 percent state-owned), GEMs reported an overall default rate of 2.6 percent between 1994 and 2024. No defaults were recorded for South Asia, but the number and volume of contracts is low relative to other regions: 221 deals worth 2.9 billion euro. Default rates were highest in sub-Saharan Africa, at 5.2 percent, but this is for 3,265 contracts valued at nearly 86 billion euro. (See Figures 3 and 4.)

The large number of contracts in sub-Saharan Africa enabled GEMs to provide a sectoral breakdown for SOE default rates. The sectors are too broadly defined to offer major insights but the data we have indicate that default risks are not highly variable: financial institutions (5.2 percent), non-financial institutions (5.2 percent), banking (6.4 percent), non-bank financials (3.5 percent), infrastructure (5 percent) and services (6.1 percent).

Average SOE loan recovery rates between 1994-2022 were 86 percent, exceeding 80 percent in four of the five regions of operation. (Recovery rates for 2023 and 2024 are not disclosed, as defaults during this time remain unresolved.) Sub-Saharan Africa accounts for nearly 58 percent of the recoveries, with an overall average rate of 87 percent. Interestingly, the lowest average rate of recovery was in East Asia and the Pacific, at 73.5 percent. (See Figure 3.)

Key findings—lending to the private sector

For the private sector, we have data by region and country—a first. GEMs looked at 15,500 contracts between 1994-2024 from 169 countries and found that the average default rate was 3.5 percent, which GEMs reports is on par with B-rated companies from S&P and B3 rated companies from Moody’s, where default rates are 3.3 and 4 percent respectively. These are similar to many corporate borrowers from advanced economies.

As one might expect, default rates on MDB credits to private companies are negatively correlated with country income levels: the average default rate in UMICs was 2.7 percent, compared to 3.8 percent in LMICs, and 7.1 percent in LICs. Europe and Central Asia had the lowest default rates (2.7 percent) and sub-Saharan Africa the highest (at 6 percent). The average SSA default rate is comparable to that of S&P B to B- rated companies in advanced countries. Corporate borrowing rates for such companies have averaged between 6.5 and 8 percent since the start of 2024, as compared to SSA borrowing rates of ~8 to 8.5 percent over the same period.

Recovery rates averaged 73 percent, although median recovery rates were above 90 percent, suggesting that a few large transactions may have dominated. Interestingly, and in contrast to default rates, recovery rates do not necessarily rise with country income levels: the lowest level—68 percent—was observed in Latin America and the Caribbean, while the highest average rate of recovery was in sub-Saharan Africa, at 78 percent.

There are some data by sector and region for credits to the private sector, although significant gaps make cross-regional comparisons challenging. Average default and recovery rates and median recovery rates for all regions are provided for three sectors from 1994-2024: (1) consumer discretionary (e.g., autos, apparel, home furnishings, electronics and leisure); (2) financials (e.g., capital markets, real estate, insurance); and (3) materials (e.g., construction, chemicals, metals, mining). Financials accounted for 36 percent of the overall loan portfolio and had the lowest default rate at 2.3 percent. (See Tables 1 and 2.)

Default rates were lowest in Latin America and the Caribbean (LAC) for consumer discretionary and materials (3.3 and 3.6 percent respectively), and in Europe and Central Asia for financials (1.1 percent).

Average default rates were highest in sub-Saharan Africa for all three sectors: consumer goods,11 percent, financials, 4 percent and materials, 6.8 percent. Notably, sub-Saharan Africa was the only region where average default rates exceeded 10 percent in any sector. The highest rate of default was in health care, at 17 percent, followed by consumer discretionary and consumer staples, with average rates of 11 and 10.3 percent respectively. Default rates in the region were relatively low (e.g., below 5 percent) for utilities, industries, and financials. (See Table 1.)

Median recovery rates for consumer goods were highest in South Asia (99 percent) followed by sub-Saharan Africa (97 percent). For financials, the highest median recovery rates were in sub-Saharan Africa for financials (99.5 percent) and South Asia (98.5 percent). Two regions registered 100 percent recovery rates for materials: South Asia and LAC.

Data are also provided for communication services, consumer staples, energy, health care, industrials, IT, real estate, and utilities, but it is more sporadic.

The maps below show default and recovery rates by country, where available. It correlates largely with the regional data but there are some interesting outliers, notably Myanmar (15.26 percent) and Turkmenistan (13.64 percent).

Default and recovery rates by creditor—a closer look

Average default rates across lending type over three decades show a broad downward trend, though with spikes during periods of stress (see Figure 8). Private and SOE default rates peaked around the turn-of-the-century financial crises in Latin America and East Asia, surpassing 8 percent for public and private borrowers, and subsequently remained at or below 4 percent, except for a jump in 2020 when private defaults reached an average of 5 percent.

Recovery rates show a similar pattern, with exceptionally high rates for sovereign lending and lower rates for SOEs and private lending. Aside from a dip in the mid-1990s for private lending, average recovery rates for all categories never fell below 60 percent. For sovereign data, rates hovered at or near 100 percent.

While the sovereign data results should not come as a surprise, the data for private and SOE lending may help shift perceptions about the risk of doing business in emerging and developing economies. To recap, key findings are:

• For state-owned enterprises (i.e., publicly guaranteed debt), default and recovery rates averaged 2.6 percent and 88 percent respectively between 1994 and 2024.
• For the private sector, default rates averaged 3.5 percent and recovery rates averaged 73 percent, although median recovery rates were above 90 percent. The average SSA default rate is comparable to that of S&P B to B- rated companies in advanced countries.
• There is some correlation between income levels and default rates, with higher rates prevailing in sub-Saharan Africa. But recoveries in sub-Saharan Africa are also higher than averages for other regions.
• Gaps in the data on sector-level default rates make it difficult to come to any definitive conclusions around risk. There are three sector categories where data are available for all regions: consumer goods, financials and materials. The data suggest that risks in financials are especially low, with average default rates ranging from 2 to 4 percent, followed by materials (where rates range from 4-6.8 percent) and consumer spending (3.3 to 11 percent).

Private investors can benefit from the elements that underpin strong default and recovery rates (e.g., preferred creditor status and well collateralized deals) by co-investing with MDBs and DFIs.

Moving forward

In addition to shifting private sector risk perceptions, GEMs data are critical for determining how much capital (equity) DFIs and MDBs need to hold against possible credit losses for all lending categories. Until recently, CRAs have not had access to the detailed default and recovery data that would allow them to properly access the risk of MDB loans and therefore accurately calculate MDB capital requirements. CRAs have been relying on proxy data in many instances, especially for private lending and recovery rates. Using the GEMs data, S&P’s model now suggests that the MDBs can lend over time up to an additional $800 billion backed by the equity MDBs already have on hand. That is a major finding in a world where shareholders are increasingly reluctant to fund further MDB capital increases. Both Moody's and Fitch, with their different methodologies, would do well to re-run their MDB models using these new data to update and improve the accuracy of their risk weights for MDB portfolios.

As rating agencies produce reports showing large volumes of additional lending headroom, MDB shareholders and management will have the final say in the matter. Shareholders have a responsibility to use the capital funded by their taxpayers both prudently and efficiently. The credit rating agencies have assigned high ratings (often AAA) to these institutions, affirming that they do have strong prudential controls. But more progress is needed on efficiency, which requires MDBs to put their additional headroom to use by increasing lending levels. What’s also needed now is continued attention and resolve from the stakeholders who championed the public release of the GEMs data. Specifically, they should continue to press for a stand-alone, searchable database and elimination or revision of non-disclosure agreements that preclude more granular data dissemination. Stakeholders can use G20 and G7 communiques and other joint statements to focus attention and drive more progress.

Using the power of MDBs’ own data to unlock lending capacity costs very little while yielding a great deal, and is in the interest of all shareholders, whether from borrowing countries or creditor countries. With this latest release, the GEMs consortium has effectively displayed the power of data disclosure, underscoring why transparency should remain a priority for all stakeholders in the development finance community.