If you’ve been following development blogs for a few years, you probably remember The Great Fungibility Debate of 2010 that was sparked by a paper in the Lancet written by researchers at the Institute for Health Metrics and Evaluation. To briefly recap, the Lancet paper finds that, in the long run, foreign health aid is fully fungible. That is, in the long run, for every dollar of health aid a country receives, it spends about 1 dollar less of its own resources on health. This result attracted much attention in the blogosphere.
Now that the dust has settled, I want to use this space to draw attention to one issue that has been largely absent from the debate and that I believe has a big effect on how we should evaluate the fungibility of aid. For me, this issue goes back to a line in a 2000 paper by McGillivray and Morrissey who argue that, because a large portion of the aid that is reported by donors is not reflected in the public sector accounts of recipients, such aid measures “…are inappropriate for analyzing fungibility”. What I tried to do in my own work is draw out the consequences of such off-budget aid for the assessment of fungibility in more detail, and suggest some possible ways of accounting for off-budget aid when testing for the fungibility of aid.
Off-budget aid affects how we should test for the fungibility of aid
Roughly speaking, to test for the fungibility of, say, health aid using cross-country donor-reported aid data you would be interested in the coefficient of health aid in a regression where public health expenditure is the dependent variable (let’s call this coefficient beta). If this coefficient is close to 1 there is no fungibility, as every dollar of health aid leads to a 1 dollar increase in public health expenditure. The closer the estimated coefficient gets to 0, the more fungible health aid is.
This standard assessment of the degree of fungibility is, however, no longer valid in the presence of off-budget aid. One way to see this is with a simple example. Suppose a donor builds a hospital and does so in a way that bypasses the recipient government (i.e., using off-budget aid). If the recipient government does not do anything, we end up with one more hospital than in the absence of health aid, so there is no fungibility. But since the money used to build the hospital is off budget, it does not show up in the recipient government’s health expenditure. As a result, we would estimate a beta of 0 rather than 1, even though health aid is not fungible.
More technically, our beta coefficient is a weighted average of the coefficients of on- and off-budget health aid. While 1 and 0 are the appropriate benchmarks for testing whether on-budget aid is fungible or not, for off-budget aid no fungibility implies a coefficient of 0, whereas full fungibility implies a coefficient of -1. As a result, the presence of off-budget aid drags down our estimated beta coefficient, making it look as if aid is fungible if we follow the standard assessment described above (which is what has been done in the literature thus far).
Data I pieced together from various papers and reports for a number of countries suggest that off-budget aid is substantial, often making up more than half of total aid received, so not dealing with off-budget aid in a correct manner potentially matters a lot when you’re trying to find out whether aid is fungible.
How can we deal with off-budget aid when assessing the fungibility of aid?
One way to deal with the presence of off-budget aid is to try and separate on- and off-budget aid and compare the coefficient of each with the appropriate benchmarks (1 and 0 for on-budget aid, 0 and -1 for off-budget aid). In a paper recently published in the World Bank Economic Review, I construct data that split up education and health aid into different modalities, including some modalities that are more on budget (sector programme aid) and some that are more off budget (technical cooperation). The key finding is that technical assistance (the proxy for off-budget aid) is not very fungible in either sector (the estimated coefficient is close to zero or, at most, slightly negative). This matters because technical assistance makes up a large share of total education and health aid, so it has a big influence on the degree of fungibility of total education and health aid. The paper further shows that implicitly assuming that all health and education aid is on budget would lead one to falsely conclude that education and health aid are mostly fungible.
A second approach to deal with the presence of off-budget aid uses simple algebraic manipulations to transform our coefficient beta into a coefficient that we can use to assess the fungibility of health aid by comparing it to the standard benchmarks. This transformation depends on the relative variance of off- versus on-budget aid (call it delta) and the correlation between on- and off-budget aid (rho). In essence, this allows us to assess the degree of fungibility under varying assumptions about the role of off-budget aid (i.e. different values of delta and rho). In a recent CSAE working paper, a revised version of which is forthcoming in the Journal of Development Studies, I use data from the Lancet article mentioned at the start of this blog to do exactly this. I find that the full fungibility conclusion arrived at in the Lancet paper depends heavily on the implicit assumption that all health aid is on budget. Taking into account off-budget aid typically reveals a lower degree of fungibility and, under plausible assumptions about the role of off-budget aid, often only modest or zero fungibility is found, even in the long run.
The two approaches can also be combined, as I show in the WBER paper, to check, for instance, what happens when the assumption that technical assistance is completely off budget is relaxed to some extent.
Both these papers suggest that taking into account off-budget aid has a massive impact on our assessment of the degree of fungibility. This implies that papers in this literature that have relied on donor-reported aid measures (in particular, data provided by the OECD DAC) may seriously overestimate the actual degree of fungibility.
In terms of going forward, taking into account off-budget aid presents both challenges and opportunities for research. Challenges because it makes the assessment of the fungibility of aid more complicated. Opportunities because, as better data become available, we might be better able to distinguish on- and off-budget components of aid, which would enable us to test for the fungibility of aid in a more appropriate manner.
I’ve only been able to scratch the surface in this blog post so, if you’re interested, by all means have a look at these two papers on my website and let me know what you think.