The effects of exports, aid and remittances on output: the case of Kiribati

Abstract

Country specific time-series models of the determinants of output for the small developing island countries in the Pacific region are relatively few. This article explores the applicability of the framework underlying Solow (1956) to analyse the determinants output in Kiribati for the period 1970 to 2005. It is found that technical progress in Kiribati has been negative virtually offsetting the positive effects of factor accumulation. Aid and remittances have negative effects and exports have only a small positive effect in the short-run.

Acknowledgements

This article is partly based on the masters thesis of the T. B. Takirua. The authors are grateful to a referee of this journal for suggesting several improvements and to Professors John Lodewijks and Anis Chowdhury of the University of Western Sydney for comments on an earlier version of this article. Author alone are responsible for the remaining errors.

All GDP data are from the UN database at http://unstats.un.org/unsd/snaama/selectionbasicFast.asp Data on employment, aid and remittances are estimated by T. B. Takirua from various Government of Kiribati's publications.

Notes

¹ Kiribati consists of 33 atoll islands with a total land area of 811 square kilometers, lying astride the equator situated in a 3.6 million square kilometers of its Exclusive Economic Zone. Its population in 2005 was about 103 000 and real GDP, in Australian dollars, in 2005 was about 45 millions, implying a per capita income of about $420. The average growth rate of GDP for the period 1971 to 2005 was almost zero and it shows a large variation. The mean and SD of the rate of growth of output, respectively are, −0.0051 and 0.1793, implying a high coefficient of variation for output.

² These abbreviations are well-known. We have used Eviews 5 for these tests and it has a detailed explanation of these tests.

³ In our view, in some respects GETS has been a better approach than the bounds test of Pesaran et al . (1996) because there is no indeterminate range for the critical values of the cointegration test. Although originally there were no cointegration tests for GETS, recently Ericsson and MacKinnon (2002) have developed cointegration tests and these are similar to the well-known cointegration test of MacKinnon (1991) for the Engle–Granger procedure. However, when our article is almost completed, Professor Michael Sumner brought to our attention the Turner (2006) in which a test similar to the MacKinnon (1991) was developed for cointegration in the bounds test. Unlike in the original Pesaran et al . asymptotic critical values, Turner has estimated critical values with adjustment for the sample size, which is a welcome addition. In general, Turner's critical values are more than the Pesaran et al . values, but converge to the latter as the sample size increases.

⁴ Cointegration with unknown structural breaks is both hard and often misused. As far as we are aware, such tests were developed by Gregory and Hansen (1996) for only the Engle–Granger equations and there are no such tests for GETS. There is also a test for JMLVECM models when the break date is known a priori; see Jusilius (1996) and Joyeux (2007).

⁵ Recently Sumner (2004) systematically outlined some difficulties in selecting the options in JMLVECM. However, Rao (2007) explains how GETS estimates can be used to select these options. This is not to say that JMLECM is the best technique. In practice, it is better to use alternative cointegration techniques to prepare summaries of data for interpretation of the data. If alternative techniques yield similar summaries then our confidence in them improves.

⁶ In the eigenvalue and trace tests, the null of no cointegration is rejected. The computed test statistics, with the 95% critical values in the brackets are 20.325 (18.33) and 28.235 (23.80), respectively. The null that there is at least one cointegrating vector could not be rejected. The computed test statistics for these two tests are 7.91 (11.54) and 7.91 (11.54).

⁷ Sumner (2004) warns about some difficulties with using JMLVECM in small samples.

⁸ A referee has pointed out that there could be some endogenous bias in the estimates of the effects of aid. While this may be true, we believe that the instrumental variables method of estimation minimizes this bias. It is difficult to develop and estimate a model, with a limited number of observations, in which aid and output are endogenous variables.

⁹ Professor Anis Chowdhury and the senior author of this article have shared their personal experiences on how remittances are generally wasted by the recipients in Bangladesh and India. Therefore, our finding that remittances may have a negative effect in Kiribati is not altogether unexpected.

¹⁰ There are several other ad hoc specifications in which growth of output is simply regressed on aid etc and the equation is estimated with OLS. We shall not examine all such ad hoc specifications. An elaborate specification to determine the effects of aid in a time-series model is Fenny (2005) for the PNG, in which it was found that aid has no significant effect on growth. Although at times Fenny has used nine variables in the cointegrating vector, the two basic conditioning variables, viz. capital and labour were not used in his specifications.

¹¹ We are grateful to Professor John Lodewijks for pointing out the need for some explanation of the negative TFP. In addition to what is stated in the text, the available data for Kiribati is of poor quality and we suspect that the negative coefficient of trend may be an overestimate. Therefore, our finding that TFP is negative needs further attention and beyond the scope of the present article. It is worth pointing out that using a similar framework we found that TFP is Fiji is less than 1% and in the Solomon Islands it is zero.

¹² Professor Anis Chowdhury has pointed out that these negative effects need some justification in addition to our conjecture that this may be a cultural phenomenon. Kiribati is one of a few countries with budget surpluses. During the period 1990–2005, the average budget surplus, as a proportion of GDP, was 5.5%. On the assumption that the government of Kiribati treats aid as a part of its revenue, aid might have partly contributed to these budget surpluses and surpluses are deflationary. It is difficult to rationalize the negative effects of remittances with other than some cultural factors. We stress that our findings are significant at the margin and therefore should not be taken as definitive.

¹³ A simulation study by Rao (2006a), based on the closed form solution of Sato (1963) for output in Solow (1956), shows that the effects of increased investment ratio on the growth of output are significantly positive for up to 15 periods. Therefore, policies to improve the growth rate in the short-run deserve attention in the developing countries because it takes considerable time to implement reforms to increase the long-run equilibrium growth rate.