The World Bank COVID-19 response: Politics as usual?

There are, of course, also potential commercial benefits to firms and contractors supplying relief aid but these are private goods (being both rivalrous and excludable).

This figure includes IFC and MIGA funds; the figure for IBRD and IDA commitments is $104 billion. Actual IBRD/IDA commitments through June 2021 were $98 billion (World Bank, 2021b). Morris et al. (2021) expect 60% of these commitments to disburse by that point, a lower percentage than they believe appropriate given the scale of the pandemic but a substantially higher percentage than is typical in past years. The World Bank (2020b, 36), for its part, states that it aims to disburse “the largest share…within 12 to 18 months,” details a number of streamlined administrative procedures, and notes that IDA lending contains a greater-than-usual share of fast disbursing development policy finance, but also notes that disbursement may be delayed by supply chain problems.

Throughout this paper, we focus exclusively on International Bank for Reconstruction (IBRD) loans and International Development Association (IDA) credits and grants. Our COVID loan designations are based primarily on project titles (including the terms “COVID” and “pandemic”). Among these, we checked cases where preparation may have started before the pandemic (as indicated by low project identification numbers; see Fig. 3 and Kilby (2013b)) to verify that their central focus was COVID-related. We also reviewed other projects with the word “emergency” in their titles to determine if the emergency was the pandemic. All results reported are based on IBRD and IDA lending only, excluding trust funds and other product lines (the sums for these are small) plus any multi-country loans where commitments cannot be attributed to individual countries.

The World Bank has provided various COVID lending figures in press releases and fact sheets. An October 14, 2020 factsheet (World Bank, 2020e) indicates $22 billion in IBRD/IDA commitments for COVID-related activities, 36% for “strengthening policies, institutions and investment for rebuilding better” (the 4th of four thematic pillars the World Bank articulated). Interpreting this category as “business as usual” rather than COVID-related, the figure is $14 billion through early October, in line with our calculations.

Throughout this paper, as is common in the empirical literature on lending by international financial institutions (IFIs), we use the terms project and loan synonymously. In fact, individual projects are often financed with more than one simultaneously issued loan or credit, reflecting different sources of funding, different currencies, and, in some cases, different financing terms. Following World Bank practice, all such loans are bundled together and treated as a single project/loan. Again following the literature and World Bank practice, the only time we treat different loans for the same project separately is when there are supplemental loans (also called additional finance (AF)) that are approved separately at a later point in time and added to an existing project.

In some cases, this follows a formal provision in existing project agreements called a “Contingent Emergency Response Component” (CERC) that allows the borrowing government to convert part of the loan or credit commitment to emergency funds. For example, the World Bank approved a $200 million IDA credit to Ghana in May 2019, for the Greater Accra Resilient and Integrated Development Project (Project ID P164330), the primary goal of which was improved flood and waste management. This loan included a $65 million CERC. On March 25, the World Bank approved activation of the CERC. On April 2, the World Bank also approved a $35 million COVID-19 Emergency Preparedness and Response Project and on November 10, it approved a $130 million AF loan for this project.

Our analysis focuses on new lending during the pandemic (e.g., Ghana’s Emergency Preparedness and Response Project and its AF). We do not analyze redirection of previously committed funds to COVID relief (e.g., Ghana’s CERC) since that depends also on a country’s pre-pandemic situation, i.e., having undisbursed balances, CERCs, etc.

The right-hand side regular loan cluster is disproportionately AF loans (dashed lines) that add funds to existing projects and are known for short preparation periods (Kilby 2013b). For more on AF loans, see Kersting and Kilby (2019). The World Bank (2020c, 18) states that emergency health projects approved to respond to the pandemic had an average preparation time of 27 days as of June 2020.

We define countries as eligible to borrow based on WDI (2020) coding.

Personal correspondence, June 30, 2021.

For each such vote, we record a 1 if the country votes the same as the U.S., 0 if it votes opposite the U.S., and ½ if it abstains or is absent when the U.S. votes. Our voting measure is the average of these scores across all votes on UNGA resolutions designated as important by the U.S. State Department in the calendar year. Since the vast majority of UNGA votes take place in November and December each year, we use the alignment measure from the previous year (2019).

COVID-19 numbers are cumulative as of June 30, 2020. Results are similar using figures from April 1 (just prior to the start of COVID lending) and somewhat less significant (for the COVID coefficients) using end-of-year data; other results do not depend on which measure we choose. Looking at the expected impact of growth follows from the World Bank’s internal justification for lending: “Potential tightening of credit conditions, weaker growth, and the diversion of expenditures to fight the pandemic are likely to cut into government revenues and their ability to invest to meet infrastructure, health, education, and other priority development goals" (World Bank, 2020b, 11).

It is possible that we still face an omitted variable problem, e.g., if other factors related to economic policy that matter to the World Bank are correlated with our geopolitical variables. Vreeland and Dreher (2014) explore this in depth in the case of nonpermanent UNSC membership and conclude that nonpermanent membership appears to be exogenous in this context. To address the endogeneity issue for UNGA voting alignment, one might consider an instrumental variables approach. However, finding a valid instrument that would generate meaningful results is problematic. If we interpret UNGA alignment as vote buying, identifying exogeneous variation in voting (e.g., due to quirks in the process) is appealing. However, if we interpret UNGA alignment as reflecting country (or government) preferences, then the exogenous component of voting could be systematically unrepresentative of preferences and therefore fail to capture the quantity of interest. This argues for including sensible controls and against trying to uncover a “clever” natural experiment. The robustness section below explores a range of other controls. For a more general discussion about the local average treatment effect (LATE) versus the average treatment effect (ATE), see Deaton (2010) and Gibson (2019).

For the Column (4) sample, we also explore using the panel nature of the data. First, we include year dummies to allow for change over time. Second, we estimate a conditional logit with country fixed effects. The average marginal effect for UNGA voting alignment continues to be positive and statistically significant throughout. In contrast, the significance of UNSC non-permanent membership depends on the specification.

In the Table 3 specifications, there is some evidence of multicollinearity between # COVID Cases and # COVID Deaths but nothing that materially impacts our results.

At this point in time, 2020 trade data in the IMF DOTS are available only for two-thirds of our sample. Results available on request.

For example, Tanzania’s late President John Magufuli, a well-known COVID denier, suppressed reporting after April 2020. Dropping Tanzania from our sample has no impact, however.

Excess mortality data are from https://​ourworldindata.​org/​excess-mortality-covid, a collaboration between the Oxford Martin Programme on Global Development​ at the University of Oxford and the Global Change Data Lab. This is the chief source used by organizations, such as The Economist and The New York Times, for their graphics and data analysis. The WHO plans to publish data on 2020 excess mortality through its “World Health Data Hub” by the end of 2021 but notes that “[o]nly 16 of the 106 Member States in these regions [Africa, Eastern Mediterranean, South-East Asia, and Western Pacific] have sufficient data” (https://​www.​who.​int/​data/​stories/​the-true-death-toll-of-covid-19-estimating-global-excess-mortality).

The same holds if we instead use UNGA votes from 2020, except that UNGA voting is statistically significant in the overall conditional allocation equation. It is not statistically significant when broken out by regular or COVID though the point estimate is much smaller (and t-stat closer to zero) for COVID lending. COVID lending continues to show no link to UNGA voting.

Restricted samples could introduce sample selection issues, e.g., if results change and such changes are driven by the change in sample rather by the additional controls per se. As it turns out, the key results persist across specifications/samples.

Results available from authors. We omit them here because the omitted variable bias story outlined above is not likely to explain the (generally) insignificant results in COVID loan selection equation (or in the conditional allocation equations, for that matter).

In Table 5UNSC Member is never significant. Also, following Bechtel and Ziegler (2019) we explore a quadratic specification for UNGA Voting. Neither the linear nor the quadratic term is individually significant and the implied peak is at the 98^th percentile, i.e., there is little evidence of an inverted-U.

With this substantially smaller sample, we have to drop the UNSC Member dummy as its marginal effect is not identified in the probit equation. The sample restriction also requires us to drop the IDA and blend dummies as they are perfectly collinear with the constant term.

The classic enlightened self-interest argument for aid holds that economic development itself is an international public good. This follows if richer countries are better trading partners, have lower rates of population growth, contribute less to climate change, are less prone to conflicts and violence that spill outside their borders, or are less likely to be the source of emerging infectious diseases. As Kanbur (2002) points out, this hinges on the assumption that aid—in this case, World Bank lending—promotes economic develop, a point still debated in the theoretical (Brecher & Bhagwati, 1982) and empirical literatures (Doucouliagos & Paldam, 2008). The causal role of economic growth in demographic transitions is still the subject of some debate (e.g., Dyson, 2010), even more so the existence of an environmental Kuznets curve (Stern, 2004). Links between income and conflict are complex at best, e.g., with some finding a positive link between income and terrorism (Blomberg et al., 2004) and others finding a negative link between income shocks and civil conflict (Miguel et al., 2004). In short, the link between regular World Bank lending and international public goods is much more indirect and tenuous than the link between COVID World Bank lending and international public goods.