THE MISSING MIDDLE – OR IS THERE AN OBVIOUS RESOURCE GAP FOR LMICS? Stefan Dercon*, # and Nick Lea# * Oxford University and #Department for International Development s-dercon@dfid.gov.uk, n-lea@dfid.gov.uk May 2015 Much has been made recently of a ‘resource gap’ for LMICs as ODA declines. In this note, we show that some of the alleged patterns are the result of the selection of specific sample countries. Using data description with similar techniques that others have used, we find no evidence of a systematic decline in total available resources (taxes, natural resource rents and ODA) across countries in the LMIC range. Focusing on social sector spending (health and education) we also cannot find any evidence of social spending being crowded out when countries get richer as they progress into middle income. The lack of a systematic pattern does not mean that specific LMICs do not face specific resource gaps or risks of crowding out of social spending, but that generalisations towards all LMICs or all countries below $3000 per capita would be misleading. 1. There is little doubt that most of the world’s extreme poor live in MICs, about 70 percent to be more precise, but most of these in countries that are just above the threshold for LMICs. MICs are typically countries with a GNI per capita of more than $1,045 but less than $12,746. This changing geography of poverty leads many to argue that more financial flows need to be directed towards MICs, and that the focus of development assistance on LICs is misplaced. Less widely known is that about 90% of the world’s poor live in countries with a GNI per capita below $1700. In fact, the majority of the world’s poor live in countries with GNI per capita between about $1000 and $1500 – there are a lot of poor people in countries like Pakistan, India, and Nigeria… These are countries that relatively recently became MICs – and took a lot of poor people with them into that status. 2. The case for higher flows to MICs is also made in other ways, by pointing to sudden changes in expenditure, or resource cliffs, related to reaching MIC status. A simple argument against MICs getting more resources because they still have need is that they are also likely to have more access to other resources: domestic tax and international non-aid flows. Recent work by Kharas, Prizzon and Rogerson1 has argued that this is indeed the case. Figure 4 and 5 from their paper show that total resources (tax plus ODA) tend to decline when countries get richer, and tax revenue is only able to compensate once a country is well into HIC status. In particular, near the threshold of $1000, tax resources (as a percentage of GDP) remains relatively insensitive to growth and only increase slowly as incomes rise towards $4,000. In short, there is a problem. In 1 Kharas, H., Prizzon, A., Rogerson, A., (2014), ‘Financing the post-2015 sustainable development goals – a rough roadmap’, Overseas Development Institute. 1 this range, as figure 5 shows, total external resources (including remittances and FDI) are nose-diving, showing that tax is unable to compensate for decreasing external resources. Figure 1 - (Figure 4 from Kharas, Prizzon and Rogerson) Figure 2 - (Figure 5 from Kharas, Prizzon and Rogerson) 3. Multilateral agences such as GAVI, GFTAM and others are arguing that this should mean that graduation has to be slower. For example, a background paper for the Equitable Access Initiative has the figure below2, which apparently is widely circulating amongst health multilaterals, as amunition to make the case for more support in these middle income countries. Besides the fact that the graph is always wrongly labelled (the X-axis, despite what it say in all the published versions should be labelled GDP per capita (logarithms) and not levels (linear scale), it is used to suggest an area between 2 Global Fund, (2015), “Evolving the Global Fund for Greater Impact in a Changing Global Landscape”, Report of the Development Continuum Working Group, signed by 27 authors, p.12. The figure is sourced as based on Brookings Analysis. The common author of this and the earlier quoted paper is Homi Kharas, of Brookings. Correspondence with him and another co-author of the Kharas et al. (2014) suggests that they are not necessarily agreeing with the implication for which these papers are used, i.e. that these middle income countries need more aid resources and graduation from aid must be delayed. 2 around $500 and $3500 (the pink area) see declining ODA, still low tax revenue and a clear potential ‘gap’ in financing. $3500 is higher than their usual graduation cliffs. So it is argued that these countries are facing a financing problem over a range of GDP per capita with tax resources available only picking up when they get considably richer and that this means that grant and concessional financing should be offered to these countries corresponding to their need.3 Figure 3 - From background note on EAI 4. This analysis and its conclusion are misleading in four ways. First, all these graphs and especially the last picture are somewhat misleading, with the shape of the regression line driven by the selection of the sample and cut-off points. Second, it assumes that the prevailing tax policy is “optimal”, the best for the country involved, and should therefore be taken as given. Thirdly, it suggests a causality that declining ODA is the main cause for resource gaps in these countries. Fourthly, analysis using shares of GDP rather than levels add further confusion. To illustrate these points, we repeated the analysis on total tax plus ODA relative to GNI per capita. Figure 4 shows first ODA plus tax as a percentage of GDP plotted against (log) GNI per capita. We use the same data sources for tax data (ICTD data to account for anomalies in some of the IMF statistics; it includes all tax, duties and natural resource rents the state collects). We include the ‘cloud’ (scatterplot) as well as the smoothed curve (as in figure 1 above); and contrary to figure 3, we do not restrict the sample. Although we could not fully replicate the figure quoted in the Global Fund document, it was suggested to us that small economies (with population below 5m) and high income countries were excluded to 3 See for example, M.Suzman, “Crunch meeting in July to decide finance for world’s poor”, Financial Times Blog, April 8, 2015. http://blogs.ft.com/beyond-brics/2015/04/08/crunch-meeting-in-july-to-decide-finance-for-worlds-poor/ To quote “..[T]he economic success of many developing countries is making them ineligible for financing that provides life-saving benefits to the millions people within their borders who still live in extreme poverty.” “Their own domestic finance is unable in the short-term to make up the shortfall, particularly in the so-called “social sectors” such as health and education, which are critical to their continued economic progress.” 3 draw the pattern.4 Dropping ‘small’ countries with populations below 5m may seem innocuous, but this means not that they just drop ‘island economies’ as Salvado and Walz claim, but dropping up to 80 countries such as Costa Rica, Central African Republic, Congo, Liberia, Mauritania, Panama, Uruguay, Bhutan, Equitorial Guinea, Gabon, Gambia, Botswana, Lesotho, Jamaica and Namibia.5 Figure 4 - Tax and Natural Resource Revenue plus ODA (% GDP) 60% LSO COG WSM AFG 50% MWI 40% BDI ERI STP MOZ DJI AZE AGO CPV SAU RUS TON SYC ZWE GNQ DZA ZAF ZAR PLW MRT DMA BTNBOL VUT HUN TTO PRT TGO GAB MDA MKD MNE BRA GMB SLE CYP SRB KGZ VNM PNG GEOUKR BIHNAM RWA KNA BLR MLTGRC MLIBEN BWA MDV BLZ TZA HTI EGY MAR TCD VCT SUR POLATG TUR FJI NER LCA SWZ JAM NIC BFA BRB HRV COM SENKEN TJK JOR EST KOR UZB YEM SVN ARG GNB GRD LBN COL BGR ALB CZE ETH CAFUGA ZMB HND ARM TUN IRN LVA ECU KAZ LAO ROM MUS CIV URY VEN MDG THA NGA MYS LTU SVK GINNPL CHN INDGHA PAN CMR SDN PRY TKM CHL KHM MEX BHS LKAIDN SLVPER PAK DOM CRI PHL BGD GTM 30% 20% 10% MNG GUY MMR 0% 5 6 7 8 9 10 11 Log GDP per capita Note: lnGDP per capita=6.2 implies about $500; 6.9 is about $1000 and 7.3 is about $1500 Sample restricted to all countries below $30,000 GDP per capita. Curve smoothed using a polynomial up to the fifth power (and not an imposed quadratic as in figure 3) 5. We do not find the trends implied by the previous graphs: there does not seem to be obvious systematic declining pattern during early MIC status. There are some differences in the data, apparently, but they are relatively small – although our figure 4 uses far more data points and does not limit the sample. Looking at this figure, one may be tempted to say that the impression of declining total resources as a share of GDP is there. Actually, if there is declining pattern then it is in the range of GDP per capita between $500 and about $800, after which the curve is essentially flat.6 The quadratic trend in figure 3 – that argues for the declining resources – is not there, once all countries for which we have data are used, and no quadratic function is enforced. In short, there appears to be no systematic trend of declining resources at middle income status ‘when tax is still low and ODA is declining’. The pattern in the ODI study is also not quite borne out either. 4 Figure 4 has some restriction – countries above $30,000 per capita were excluded but in Appendix Figure A.1, all countries were included and the pattern is the same. 5 R.Salvado and J.Walz (2015), “Poor people or poor countries? Aid transition and the “missing middle” for LMICs” BMGF Development Policy and Finance Discussion Paper. In the appendix Figure A.2 we can confirm that dropping these 80 countries can produce a U-Shape… 6 So these are larger numbers than just tax revenue as percent of GDP as in WDI or IMF data – natural resources rents and any missed duties are included. If we use the standard ‘tax revenue’ data (without resource rents), then there is a slight difference in the pattern compared to figure 4. Fitting an up to a fourth power polynomial to these data does not become essentially a flat line from GDP per capita of about $800 but the percentage of resources in GDP decline from about $500 per capita until about $60,000 per capita! But the latter result is then largely due to the increasing share of these other contributions to government resources when countries get richer. 4 6. It is nevertheless striking that there is large variation in available resources between countries at similar income levels – which is likely to reflect specific policy choices. Countries with particularly low levels of total resources as a percentage of GDP shed some insight on possible explanations for this wide dispersion. For countries above $1000 GDP per capita but with percentages below 20%, some striking patterns emerge: in the range near $1000 and $3000, many of the larger countries in this group with low resources from tax and similar sources are Asian: such as Myanmar (MMR), Cambodia (KHM), Bangladesh (BGD), Pakistan (PAK), India (IND), Philippines (PHL), Indonesia (IDN). Those with higher GDP per capita but low resources are often Central or Latin American economies. These low rates reflect policy choices: these economies are known to have low tax collection efforts or success. India for example still has a percentage of households paying income taxation less than a few percent. Hence, to suggest that international resources should compensate for lower resources ‘because tax collection hasn’t picked up’ seems an inappropriate policy conclusion! 7. There is no obvious evidence that an emphasis on domestic tax effort (rather than aid) should raise concerns that, on average, this will be at the cost of the social sector spending. Figure 5 shows how ODA as a percentage of GDP steadily and sharply declines as income grows – unsurprisingly. While this implies that domestic resources become increasingly essential for social sector funding, figure 6 shows that as a percentage of GDP, public expenditure on health and education increases steadily – peaking at GDP per capita of around $3000 per capita. In short, LMICs, even if they may have less resources from ODA, are able to raise spending to these sectors steadily. While it is hard to judge the quality or equity of these expenditures, the removal of ODA and increased reliance on taxes does not of itself seem to crowd out social spending in a systematic way. There are countries which choose to spend relatively little but from the data pattern it is not possible to argue that this is because aid fell. 5 Figure 5 – Net ODA received (% GDP) 40 LBR 35 AFG 30 25 KIR 20 STP ZAR 15 10 BDI 5 0 5 6 WSM GMBMOZ CPV SLE NER TJK RWA VUT ETHCAF BFA CIV BTN TZA MLI UGA SEN KGZ PNG GINTGOBENLSO KHM KENYEM ZMB NICMDA COGGEO GHA JOR DMA MDG NPL CMRLAO BOL ALBMNG SYC BIHNAM HND KNA SRB ARM TUN MDV SDN ISR LCA FJIBLZ LBN MAR MKD VCT MUS BGD PAK SLV GRD ECU LKA SUR EGY KAZ GTM AZE BWA BHR PRY NGA UKR DOM TUR BRB GNQ ZAF HRV MLT CYP AGO PAN SVN PER COL IDN ATG BLR JAM BHS DZA IND CRI ARG BRAHUN CHL VEN URY IRN TTO SGP KWT PHL MYS HITA KG POL RUS LVA LTU SVK EST CZE PRT GRC ESP JPN GBR NZL ARE FRA ISL DEU BEL AUT FIN IRL NLD CAN USA SWE DNK AUSCHEQAT NOR LUX 7 8 9 10 11 12 Log GDP per capita Figure 6 – Health plus education public expenditure (% GDP) 30 25 MHL LSO 20 TUV KIR DNK FSM SLB NZL NLD ISL SWE BEL FRA SWZ PLW NOR AUT GBR CRI USA DEU DJI MDA FIN NAM CAN MLT TMP BWA MWI CHE JPN ARG SVN ESP STP WSM PRT ITA IRL AUS SRB RWA JOR GHA KGZ BOL CZE ZAF BDI URYEST GRC CYP ISR BLZTHAMDV TUN UKR COM HRV LTU POL BRB COL BRAHUN SVK KOR LUX MNG VCT DMA PRY JAM VNM BLR TGO TZA SENKEN NIC LCA MYS DZA PAN COG VEN TON MEX RUS LVA BGR VUT MKD CPV CHL SYC HND BTNMAR MRT BEN SLV GUY TUR LBR MOZ GMB NER SAU ECU MLI ROM KNA NPL GRD FJI BFA ETHGNB CIVYEM OMN ATG IRN BHS ALB AGO TTO BHR BRN KWT PER MUS EGY LBY ZMB KAZ SLE TJK CMR ZAR MDGUGA GTM ARM DOM CHN LBNGAB SGP IND PHL IDN QAT LAO GIN AZE GEO BGDTCD ERI AFG KHM GNQ PAK CAF ARE LKA HTI MMR 15 10 5 0 5 6 7 8 9 10 11 12 Log GDP per capita Note: lnGDP per capita=6.2 implies about $500; 6.9 is about $1000; 7.3 is about $1500; 8 approx $3,000 8. Even if one wants to doubt some of the analysis above, the evidence points towards generally increasing per capita spending in health and education. Some may still want to argue that figure 3 or figure A.2 is the right way to look at the problem of resource availability (i.e. excluding the set of countries excluded). But in any case drawing a conclusion from figure 3 that this must mean that we should step in is very misleading as well. This analysis is conducted in percentages, but what matters, say, for spending on vaccines, is the size of the resource envelope, the total sum available and not the percentage of GDP it represents. Drawing a curve in relationship to GDP per capita, is easy enough. Figure 7 and 8 give this, and to drill home the point, first for all countries with a population larger than 5m and also for the same group, but excluding high income countries – hence replicating the Global Fund analysis but now in levels. We find that there is a strong increase in resources per capita when GDP rises. At most, there is decline between GDP per capita around $300 and $550, to pick up again afterwards. Throughout the entire MIC range it increases persistently. So surely this is not is there a cliff or a ditch, as total resources keep on increasing for these countries rather fast, so there is more money for social services and development, even if there is 6 possibly some decline in the share of GDP.7 It definitely does not lead to the finding that taking grants or concessional funding from development spending in lower income countries to these lower MICs is the obvious equitable answer. Figure 7 – Tax Revenue plus ODA per Capita in $ (only countries above 5m population) Figure 8 – Tax Revenue plus ODA per Capita in $ (only countries above 5m population) 9. The analysis in this note shows again that generalising about standard categories such as LIC or LMIC is not helpful. There is not an obvious case for the existence of a resource gap in LMICs (or countries between $1000 and $4000) where ODA has fallen but taxation has not yet been able fill the gap. Similarly, LMICs, facing lower aid do not appear to be systematically spending less on social sectors – used here as a possible crude category of spending that may benefit the poor. If there are patterns, then they 7 It follows from figure 6 and 8 that repeating Figure 8 for education plus health expenditure also shows increasing expenditure throughout the lower middle income range. 7 are likely to relate to specific countries with specific tax and spending policies. For DFID this means that general policies ‘favouring’ LMICs or relaxation of ‘graduation’ criteria for such groups of countries are not justified on the basis of income. It does not mean that some of these countries should or could require specific attention, and of course, it is true that there is still a lot of poverty in these places. But any attention to such countries should be based on different or more nuanced information than a GNI per capita cut-off or a current level of need, but also a clear assessment of ability to contribute from domestic resources, overall access to financial resources well beyond such ODA and an appropriate understanding why social sector spending is not appropriate. Some of these questions are hard questions to ask our partners, but not asking them would mean ignoring opportunities to achieve real progress for the poor and misleadingly assume that the poor human development outcomes in some middle income countries are always just an issue of lack of aid. Appendix Figure A.1 Unrestricted version (all countries in the world) of Figure 4 8 Figure A.2 Figure 4 but excluding countries below 5m population 9