(
)
(
)
t
t
t
t
t
t
t
a
erd
m
ep
d
d
e
a
x
θ
θ
−
+
+
=
−
+
+
−
1
*
1
(1)
This leaves the slightly absurd theoretical possibility to be tested in this paper: that
foreign aid only has positive growth effects to the extent that it is used to alleviate
external problems and thus does not enter the economy. In other words, it is possible
that aid could have positive growth effects when it is used contrary to its intended
purpose. For a developing country with no or only little external debt, foreign aid
inflows would result fully in Dutch Disease problems and thus lead to an ambiguous
relation with economic growth. In developing countries with high levels of external
debt, on the other hand, foreign aid would provide much-needed foreign currency to
meet debt payments. In the latter situation, a substantial part of the aid would therefore
not enter the economy, leaving less severe Dutch Disease and alleviating debt problems,
and thereby making a positive growth-aid association more likely. The specific
hypothesis to be tested in the following is therefore:
Hypothesis: The association between foreign aid inflows and economic growth
rates depends positively on the country’s level of external debt.
3. Data
To be
able to test this proposition, we employ data from different sources. The data to
be employed in the next section all derive from the Center for Global Development
database (CGD, 2006), the Penn World Tables Mark 6.1 (Heston et al., 2002) and the
World Bank’s (2005) World Development Indicators database. Our dependent variable
in the following set of Barro-type regressions is the average annual growth rate across
7
each five-year period between 1960 and 2000 for which full data are available. To
explain these growth rates, we employ a fairly simple baseline specification including
the logarithm to initial GDP per capita in PPP-adjusted US dollars, the investment rate
as percent of GDP and government consumption as percent of GDP, all from the Penn
World Tables. In a set of investment regressions, we also include openness to trade and
the relative investment price (capital goods prices as a ratio of the overall price level)
from the same source. The specifications also include our variables of interest, foreign
aid as percent of GDP, taken from the Center for Global Development database, and the
external debt as percent of GDP, taken from World Development Indicators and
averaged across each five-year period, as well as period fixed effects to account for
global business cycles. In half of the regressions below, we also include an interaction
term between aid and external debt, based on the centred values of the variables. As
such, the size and significance of the interaction term is evaluated around the sample
mean (cf. Brambor et al., 2006). Table 1 gives descriptive statistics of the data; all
countries included are reported in the appendix.
Insert Table 1 about here
We estimate Barro-type regressions using generalized least squares with country
fixed effects as Hausmann tests strongly favour fixed over random effects.
2
We present
results in two different samples. First, we use the largest possible sample for which
2
Using the full sample and specification, the Hausmann test statistics of the growth regression and
investment regression are Chi
2
= 19.12 (p<.004), and Chi
2
= 24.92 (p<.009), respectively. Without the
interaction term between aid and debt, the test is marginally stronger for the investment regression and
somewhat stronger for the growth regression.
8
there are full data, consisting of 199 observations from 38 of the 49 countries currently
categorized as LDCs. As noted above, we restrict our attention to LDCs as these
countries both have large debt problems as well as very limited inflows of foreign direct
investments. However, the possibility exists that these results may be biased as
countries may tend to enter international datasets after periods of relatively good
performance, which could bias coefficients on both foreign aid and external debts
upwards.
3
Second, by restricting the sample to those countries for which there are data
for at least half of the period 1960-2000, an alternative sample comes to consist of 156
observations from 27 countries. We also run separate analyses excluding either the top
performers or bottom performers at any time in the LDC group.
4. Empirical results
Before turning to the formal empirical results, it can be
informative to look at some
simple statistics. First, the data from the 38 LDCs show the same picture as the overall
data from developing countries, as the simple correlation between economic growth and
foreign aid is far from being significant at -.02. On the other hand, the correlation
between growth and the ratio of external debt to GDP is -.13 and significant at p<.10. At
first sight, then, it would seem that the raw data indicates that outflows of interest
payments on debt could be more important than inflows of foreign aid. Whether this is
the case is further explored in Table 2 that reports the results of estimating growth rates
for the LDCs.
3
That this is not merely a potential problem within this sample is indicated by some simple sample
statistics. In the first period that countries enter, the sample average growth rate is .64 percent while in the
following period, average growth is -.71 percent and -.10 on average for all subsequent periods.
9