The Impact of Foreign Aid on Economic Growth in Papua New Guinea

Abstract

This article investigates the impact of foreign aid on economic growth in Papua New Guinea (PNG) using time-series data for the period 1965 to 1999. Following the most recent literature, the article examines whether aid effectiveness is conditional on levels of economic policy and governance. An empirical model is estimated using the Autoregressive Distributed Lag (ARDL) approach to cointegration proposed by Pesaran and Shin [1995]. Results provide little evidence that aid and its various components have contributed to economic growth in PNG. There is some evidence that aid is more effective during periods when the country has undertaken a World Bank Structural Adjustment Program (SAP). An alternative interpretation is that a SAP may be more effective at spurring growth when supported by foreign aid.

Notes

Simon Feeny, School of Economics and Finance, RMIT University, GPO Box 2476V, Melbourne, VIC 3001, Australia. Tel: + 61 3 9925 5901; fax: + 61 3 9925 5986. E-mail: [email protected]. A large proportion of this work was carried out during an internship at the World Institute for Development Economics Research, United Nations University, Helsinki, Finland. The author is grateful to the Institute for its financial and institutional support. He is also grateful to Mark McGillivray, Guanghua Wan, Asghar Adelzadeh, Marko Nokkala, Matt Hammill and three anonymous referees for helpful comments and advice.

Earlier studies include Islam [ 1992 , 1999 ] for the case of Bangladesh and Mbaku [1993] for the case of Cameroon. However, these studies fail to account for the time-series properties of the data and to test for cointegration. Their findings that aid is ineffective at spurring growth are therefore likely to be misleading.

Exceptions are Collier and Dollar [ 2001 , 2002 ] and Collier and Hoeffler [ 2002 ] which use the World Bank's Country Policy and Institutional Assessment (CPIA) measure of policy. The CPIA has 20 equally weighted components divided into the following four categories: macroeconomic management and sustainability of reforms; structural policies for sustainable and equitable growth; policies for social inclusion; public sector management. The CPIA measures are only available from 1998 so are not used in this article.

Lensink and White [ 2000 ] outline the problems in detail. In summary they argue that the impact of trade policy is entirely sensitive to the measure used and evidence on the positive effects of outward orientation is therefore weak. Similarly, the evidence regarding the budget surplus as a determinant of growth is also weak. Further, it is only high levels of inflation which are harmful to growth suggesting that this variable should enter the growth equation in a non-linear manner. Lensink and White are also critical of the construction of a single index.

Note that the models of Gounder [ 2001 , 2002 ] are based on an augmented neoclassical Solow type model. In such a model there is a steady state rate of growth which, in econometric terms, can be represented by cointegrating relationships. As pointed out by an anonymous referee and demonstrated by Mankiw et al. [ 1992 ], models of this type should include the log of the lagged level of GDP per capita together with other explanatory variables which describe the level of steady state income.

Diversion relates the theft or expropriation of resources from productive units. It might relate to illegal activity such as theft corruption, or the payment of bribes, or it can be legal such as confiscatory taxation by the government, frivolous litigation, or the lobbying of the government by special interests. Diversion acts in similar way to a tax on output.

Ideally a separate variable would be employed in the empirical model to represent labour. The trend variable is used to capture the growth in the labour force since accurate labour force data are not available in the case of PNG. The dependent variable is the same as that used by Gounder [ 2001 , 2002 ]. Lloyd et al. [ 2001 ] use real private consumption and real GDP in levels as dependent variables in their analysis of growth in Ghana. Although these variables are expressed as growth rates in the error correction version of the model, the standard ARDL model estimates the determinants of the levels of these variables rather than the growth rates. Growth in private consumption is not modelled for PNG. The vast majority of the population are in the informal sector which casts doubts over the accuracy of such data. A number of alternative specifications were tested with the regressors expressed in growth rates, levels and as ratios to GDP. The specification of equation (1) was found to be superior to other specifications in terms of cointegrating relationships between variables, R-squareds, and diagnostic tests. Although the model specification largely follows that of the previous literature, like any time-series study, the empirical model is constrained by the availability of data to capture determinants of long-run growth. An anonymous referee relates the specification in (2) to a demand model of annual growth (or business cycle model of growth) rather than a long-run model of growth. Trade is widely perceived to be pro-cyclical and the findings of a high estimated speed of adjustment towards equilibrium and a positive and significant coefficient on the trade variable lend some support to this notion.

Previous studies have discussed the difference in the interpretation of parameters when investment and/or aid are included in the same regression. If aid and investment are included in a regression and the coefficient on the aid variable is significant, the interpretation is that aid affects the level of efficiency, in the sense that it has effects on growth above and beyond its effects on the incentives to invest. If investment is omitted from the regression and the coefficient on the aid variable is significant, we do not know whether aid impacts on growth directly or through the incentive to invest [ Sala-i-Martin, 1997 ]. However, it must be true that if investment is an important determinant of economic growth and is not included in the empirical model, results will suffer from omitted variable bias.

Gomanee et al. [ 2002 ] also recognise that imports, government fiscal behaviour and government policy represent transmission mechanisms operating from aid to growth. In this article, government policy is proxied for by a dummy variable taking the value of one in periods of structural adjustment and it is very difficult to establish the impact of aid on this variable empirically. Moreover, the parsimonious model used in this article does not include government consumption as an explanatory variable and imports are included in the trade variable. It is recognised that further investigation into these transmission mechanisms remains an important area for future research.

It is well documented that the use of generated regressors in empirical models yields inefficient standard errors (see, for example, Wooldridge [2002, 115–16]). Although there are procedures which are able to correct the standards errors, this article defers this issue for an area for future research. However, it is noted that in a number of the models estimated, the coefficients on the foreign aid variable are negative but insignificant. Therefore more efficient standard errors are likely to lead to more negative inferences with regard to the impact of aid on economic growth than currently suggested by this article.

The empirical model was also estimated using donor measures of aid. Although broad conclusions do not change, there was less evidence of cointegrating relationships between variables in some of the specifications.

The IRIS III Data set provides annual values for indicators of the quality of governance. It was constructed by Stephen Knack and the IRIS Centre, University of Maryland. The data can be obtained from the PRS group.

This approach is similar to the approach taken by Lloyd et al. [ 2001 ]. They include interactive dummy variables for the period from 1983 when Ghana undertook a major economic recovery programme under the auspices of the World Bank. An alternative approach to investigate the impact of policy on aid effectiveness would be to employ the budget balance, inflation and a measure of trade openness separately in the model. The aid variable could then be interacted with each of these variables, or a policy index could be formed using the regression coefficients on these variables. However, this could not be done due to a lack of degrees of freedom in the ARDL model.

There are a number of problems associated with ADF tests. The outcomes of the tests can vary depending upon the inclusion of a constant, time trend, number of lagged differences to control for autocorrelation and the level of statistical significance that a researcher wishes to abide by. The power of the tests is very low. Unit root tests do not have the power to distinguish between a unit root and near unit root processes. The tests will all too often indicate that a series contains a unit root. Moreover, they have little power to distinguish between trend stationary and drifting processes [Enders, 1995, 251–52]. The presence of structural breaks will also bias ADF tests towards the non-rejection of a unit root.

An alternative approach would be to obtain the generated regressor in the first stage and then to test this variable for endogeneity. However, using this approach, the generated regressor was also found to be endogenous and finding instruments for this variable is very problematic.

The standard errors of the long run coefficient estimates are computed using Bewley's [ 1979 ] regression approach.

An additional model specification could disaggregate foreign aid into bilateral and multilateral aid. However, a time-series of these data are not readily available from the PNG Budget Papers. Data are available from the World Bank's World Development Indicators. However, the ARDL technique found no evidence of cointegration between the variables in a model where aid was disaggregated into bilateral and multilateral aid.

Additional information

Notes on contributors

Simon Feeny

Simon Feeny, School of Economics and Finance, RMIT University, GPO Box 2476V, Melbourne, VIC 3001, Australia. Tel: + 61 3 9925 5901; fax: + 61 3 9925 5986. E-mail: [email protected]. A large proportion of this work was carried out during an internship at the World Institute for Development Economics Research, United Nations University, Helsinki, Finland. The author is grateful to the Institute for its financial and institutional support. He is also grateful to Mark McGillivray, Guanghua Wan, Asghar Adelzadeh, Marko Nokkala, Matt Hammill and three anonymous referees for helpful comments and advice.