Modelling inter‐temporal aid allocation: a new application with an emphasis on Papua New Guinea

Abstract

This paper models the inter‐temporal allocation of foreign development aid to Papua New Guinea (PNG). A formal theoretical model of aid allocation is developed, in which aid to any one country is determined jointly with aid to all other recipient countries. This is recognized in the econometric application of this model, which involves simultaneously modelling aid to a number of countries in addition to PNG. Results based on data for the period 1969–99 indicate that both recipient need and donor interest variables determine the amount of foreign aid to PNG and most other countries under consideration.

Notes

* Simon Feeny (to whom correspondence should be addressed), School of Economics and Finance, RMIT University, GPO Box 2476V, Melbourne, 3001, Australia. Mark McGillivray, World Institute for Development Economics Research, Helsinki, Finland.

The authors are grateful for useful comments provided by an anonymous referee.

Development aid, or official development assistance (ODA), is defined by the DAC as grants or loans to developing countries which are: (a) undertaken by the official sector; (b) have the promotion of economic development and welfare as the main objective; (c) at concessional financial terms (a loan must have a grant element of at least 25%). In addition to financial flows, technical co‐operation is included in ODA. Grants, loans and credits for military purposes are excluded. Transfer payments to private individuals (e.g. pensions, reparations or insurance payouts) are in general not counted. Only countries that belong to part I of the DAC's list of developing countries can receive ODA. The DAC, whose membership comprises all major western aid donor countries, collects and reports aid flows on behalf of its member countries. See OECD (1999) for further details.

In a similar study, Feeny & McGillivray (2002) used the same approach to model aid allocated to PNG. However, only two equations were used. The first explains Australian aid to PNG and the second explains Australian aid to all other developing countries.

Trumbull & Wall (1994), Wall (1995), Tarp et al. (1999), Lahiri & Raimondos‐Møller (2000), Feeny & McGillivray (2002) and McGillivray et al. (2002) are relatively recent exceptions.

It could be argued that the bilateral aid decision‐makers also derive utility from the impacts of other programmes funded by the agencies in which they are located, such as the multilateral aid programme. However, as the bilateral aid decision‐makers have little or no control over the allocation of these funds, this impact is exogenous with respect to the preferences of these people, and including such a variable in the utility function makes no difference to the behavioural and estimating equation derived.

A number of studies have tested for what are referred to as the small and middle‐income “biases” in aid allocation, where aid decreases with population and increases with per capita income over given ranges of these variables. See, e.g. Arvin (1998) and Arvin & Drewes (2001) for recent evidence.

In recent years donors have increasingly adopted a policy of selectivity, providing more aid to countries with perceived better policies. The sample used in the paper extends only to 1999. Subsequent attempts to model aid allocation, using data from the early 2000s onwards, might need to take this into account.

These points were first noted in McGillivray & White (1993). Alternatively, the decision variable could be aid shares, with aid measured as a percentage or ratio of the total bilateral aid budget. Econometrically, using this measure or absolute aid makes very little difference with only the constant term being affected.

This was also supported by preliminary econometric testing. This involved estimating equation (11) as a simultaneous system of equations for a number of aid recipients, testing cross‐equation restrictions that β_0,1=β_0,2=… β_0,m , β_1,1=β_1,2=… β_1,m , and so on. These restrictions were clearly rejected. Further details are available from the authors.

This is also a point originally made in McGillivray & White (1993). The relevant variables omitted from the recipient need model are the donor interest variables, and vice versa. Unless it can be shown that none of the donor interest variables omitted from the recipient need model are orthogonal with the recipient need variables omitted from the donor interest model, which is unlikely in the extreme, then it in turn follows that the error terms of both models are not independent of their respective explanatory variables. The t‐ratios, F‐tests and R ²s resulting from separate estimation of the models are therefore invalid and the conclusions based on these statistics are likely to be misleading.

Large recipients of Australian aid for which a long time series was not available include China, Bangladesh, Vietnam, Cambodia, the Solomon Islands and Vanuatu. China and Bangladesh have also been two major recipients of DAC aid, but data availability did not permit a long time series. China started receiving aid from the DAC in 1979 and Bangladesh in 1972 (formerly West Pakistan). However, all of these recipients are included in the equation explaining aid to all other countries. Note also that Israel was no longer classified as DAC part I developing county from 1997 onwards. However, it continues to receive DAC “aid” (but not ODA) as a part II country on the DAC list, so it is included in the sample. See OECD (1999) for further details.

This equation will clearly be subject to a number of econometric issues, arguably the most serious being aggregation bias. Estimates of its parameters should therefore be treated with more than the usual degree of caution. However, its role is purely econometric, being to provide efficient estimates of the parameters of the other 11 equations in (14).

It should be noted that there is less than universal acceptance of income growth as an indicator of need. For a discussion of this issue see, e.g., Mosley (1981) and McGillivray & White (1993). Our choice was guided by the literature on aid allocation; most studies treat this variable as an indicator of need.

See Gulhati & Nallari (1988) for a good discussion of inertia in aid allocation.

Among the structural breaks considered but ruled out in the final analysis were increased support from the early 1990s for countries classified as low income or least developed or for those located in sub‐Saharan Africa and a possible diversion of aid from countries not belonging to part II of the DAC list following the collapse of the Soviet Union. Further details can be obtained from the authors.

The studies which have tested for the relevance of these categories of variables include McKinlay & Little (1977, 1978a,b, 1979), Maizels & Nissanke (1984), Gounder (1999) and Gounder & Sen (1999), and have done so by separately estimating recipient need and donor interest models of aid allocation. The former are comprised by recipient need variables only and the latter by donor interest variables. Conclusions regarding the overall significance of these vectors tend to be based on the adjusted R ² of each model. However, this approach is inherently problematic econometrically due to the reasons outlined in note 9.

For single equation estimation explaining Australian aid to PNG, the coefficients on the growth in income per capita variable and balance of payments variable are insignificant. This provides further evidence that the error term of this equation is correlated with the error terms from other equations. The parameter estimates are therefore inefficient and the corresponding t‐statistics are invalid.

For single equation estimation explaining DAC aid to PNG, the coefficients on the balance of payments, multilateral aid, investment and lagged aid variables are insignificant. In contrast to system estimation, this leads to the rejection of recipient need as a determinant of aid allocation. This provides further evidence that the rejection of recipient need as a determinant of aid allocation found by the previous empirical literature is likely to be due to single equation estimation.

Additional information

Notes on contributors

Simon Feeny *

* Simon Feeny (to whom correspondence should be addressed), School of Economics and Finance, RMIT University, GPO Box 2476V, Melbourne, 3001, Australia. Mark McGillivray, World Institute for Development Economics Research, Helsinki, Finland. The authors are grateful for useful comments provided by an anonymous referee.