Development aid and the loss of competitiveness of the manufacturing sector

1 UNIVERSITY OF GRONINGEN – FACULTY OF ECONOMICS AND BUSINESS

Development aid and the loss of competitiveness

of the manufacturing sector

Daniel Harding s2248433

Daniel.Harding1@gmail.com

Supervisor: Marcel Timmer Co-assessor: Robert Inklaar

13-06-2016

Abstract

Keywords: Aid, manufacturing, growth, Dutch disease, exchange rate

3

1. Introduction

Development aid has become a standard component of national expenditure for most developed countries. In the 1970s, most developed countries have come to an international agreement of setting the target of investing 0.7% of GNI in development aid. According to UN calculations, it would be possible to achieve the Millennium Development Goals if this target was reached. The figure below gives an indication to what extent this target has been achieved in the year 2015.

Figure 1 provides an overview of spending on development aid per country as a percentage of GNI. Data have been obtained from the OECD – data on ODA 2015, visualized using the compare your country tool.

4

This has risen the debate of whether the money invested in developing countries is spent effectively. It is a hard to answer, seeing as the literature is highly divided and the

effectiveness of aid may can be determined using numerous different measures. However, for now it seems like development aid has not been as effective as it was hoped when the UN target of investing 0.7% of GNI in development aid was set. On the one hand, development aid has been successful at improving certain socio-economic indicators such as improving access to education and increasing life expectancy by reducing the prevalence of diseases. For example, due to strong collaboration with NGOs Bangladesh has been able to reduce infant mortality by half and achieved near-universal enrolment in primary education for girls (Goldin et al, 2002). On the other hand, when it comes to structurally lifting a country out of poverty by stimulating growth, numerous meta-analyses of the literature seem to conclude that aid does not stimulate growth (Doucouliagos & Paldam, 2009; Rajan & Subramanian, 2008), and may possibly even have an adverse effect on growth.

The question of what mechanisms lie behind the relationship between aid and growth has gotten a lot of attention within the literature. Numerous control variables and conditions have been examined, which will be further elaborated in chapter 2 of this paper, but so far there seems to be no clear consensus within the literature. One of the more recently examined explanations for the potential adverse relationship between aid and growth is inspired by the natural resource curse literature. Development aid, much like natural resources, can provide upwards pressure on the exchange rate, making a country‟s exports more expensive to other countries. Seeing as the country‟s tradable sector is often considered to be the engine of growth, an appreciation of the exchange rate could reduce the size of the tradable sector, and thereby hamper growth. This phenomenon is termed “the Dutch disease”.

In this paper, the Dutch disease of development aid will be examined. The purpose is to determine whether development aid can influence the growth of the manufacturing sector through an appreciation of the exchange rate. If Dutch disease is present, countries that receive more aid should have a more overvalued exchange rate and a smaller export sector. The main contributions to the literature will be the following. Firstly, in contrast to the majority of the literature where most studies run a cross-country panel regression, this paper uses a cross-industry, inter-country comparison model, which should reduce problems of endogeneity. Secondly, data will be obtained from a database covering a recent time period,

5

this paper will aim to determine whether it is possible to mitigate the negative effects of Dutch disease by specific allocation of development aid within the manufacturing sector. The findings can be summarized as follows. A significant and positive relationship was found between development aid and exchange rate overvaluation, however, no significant

relationship could be found between development aid and the hypothesized slowdown in growth of the main export sectors. Furthermore, no significant evidence could be found for investments of development aid within the manufacturing sector and the mitigation of Dutch disease effects. Lastly, the results suggest that exchange rate overvaluation may be the channel through which aid may influence growth, but also aid volatility may also play an important role at explaining the results. However, these findings are suggestive and not conclusive.

6

2. Literature review

2.1 General effectiveness of development aid

The literature on the effectiveness of development aid seems to be endless, and the results and methods used are diverse. Currently, it does not seem possible to draw a conclusion as to whether development aid has been effective in stimulating growth. Aid may have been successful at reducing the prevalence of certain diseases, reducing the amount of child death, and other socio economic indicators, however, the eventual purpose of aid is also to alleviate the country from poverty. To achieve this, growth is essential, and it seems like the literature cannot find a positive relationship between aid and growth.

To get started, a good overview on the literature of the effect of aid on growth is presented by (Doucouliagos & Paldam, 2006, 2007a, 2008, 2009), who conduct a meta-analysis of over 97 econometric studies. They find that the results are inconclusive, but they also note that the results are subject to bias.

This bias could emerge because the development aid literature seems to be an area where the author‟s motivation and ideology play a bigger role than other literature. On top of that, a big portion of the research is funded by charitable institutions themselves, which could therefore result in a conflict of interest. It is thus not surprising that the authors find that the results across studies are asymmetric, hinting at a reluctance to publish negative results. This may be the reason why about 74% of the studies on the relationship between aid and growth

published positive results, but a meta-analysis of these studies cannot prove these positive results to be significant.

To get a better picture on the literature on development aid, the way in which Doucouliagos & Paldam conducted the meta-analysis will be elaborated below. Within the studies they

7

Figure 2: The figure graphically depicts the different families of models used to examine the effectiveness of development aid. Taken from Doucouliagos & Paldam ( 2009)

Family A models are based on the Harrod-Domar model of savings and investment. Simply put, aid finances investment in the developing economy. Investment then results in capital accumulation, and capital accumulation results in growth. Essentially aid thus relieves the resource constraint on growth, which is a lack of investment.

Early studies on family A models such as Griffin and Enos (1970) identify a phenomenon that is now called “the savings challenge”. These studies found that gifts of aid reduced the

recipient country‟s savings rate. Aid seemed to discourage recipient countries from saving. Instead, these countries were seeing increasing their public expenditure and reducing efforts to collect taxes. Due to the fungibility of development aid, recipient countries engaged in so-called “aid-switching”, in which aid replaces the function of savings and may explain the observed phenomenon mentioned above (Snyder, 1990). Following this, numerous studies have been conducted on the relationship between aid and the rate of savings or investment. The meta-analysis on these studies by Doucouliagos & Paldam shows that aid does not

completely crowd out savings and investment, however, it increases capital accumulation only by 25% of aid. The remaining 75% of aid results in an increase in public consumption,

possibly due to aid-switching.

8

literature‟s findings are highly diverse, and a meta-regression finds that the results across studies are insignificant. Hence, we cannot conclude whether aid affects growth.

The last family of models, family C, have emerged after earlier models have failed to provide an answer to the effectiveness of aid. Family C models propose that the effectiveness of aid depends on a certain condition. In some conditions aid seems to be effective, in other conditions it is not. This family of model thus aims to find the condition which makes aid effective.

One of these conditions that is frequently examined in the literature is the condition of “good policy”. The idea is that only countries with good policies in place will be able to benefit from aid. Good policies could be defined as a high degree of openness to trade, and having an acceptable inflation rate and budget surplus. An influential paper by Burnside & Dollar (2000) found that aid in countries with good policies does positively affect growth, however, replication by later papers with different samples and control variables has failed to produce the same results.

Another condition that is included in family C models and that has frequently examined in the literature is the condition of aid itself, which may be subject to diminishing returns. The model used to examine this condition includes both a linear and a squared aid term in the regression. After running the regression, the linear aid term returns a positive coefficient, whereas the squared aid term returns a negative coefficient. This means that the equation creates an inverted parabola, and aid will only have a positive effect on growth as long as the parabola is above the x-axis. Thus, essentially the effect of aid on growth depends on aid itself: the amount of aid must fall within a certain threshold where the parabola is above the x-axis for the effect to be positive. A meta-analysis for this type of model has also failed to provide significant results, since the authors could only find a significant result for a specific dataset.

Currently, other conditions on which aid might be effective are being investigated, however, Doucouliagos & Paldam (2009) note that the literature is currently too scarce to conduct a meta-analysis on these studies. Only time will tell whether the results are robust.

9

thus the question remains what mechanism could prevent aid from having a positive effect on growth.

As mentioned earlier, one mechanism could simply be that the country does not have the right conditions for aid to be effective. For example, a country may not have the right monetary and fiscal policies to effectively absorb aid, or it may have a political system that is less optimal for absorbing aid (i.e. democracy versus dictatorship). Another explanation for the lack of aid effectiveness that is recently being examined in the literature is "the Dutch disease".

2.2 The Dutch disease

The Dutch disease is a term coined by The Economist to describe the state of the Dutch economy in the 1970s. In 1959 gas reserves were discovered in the Netherlands, which resulted in a large increase in exports. Normally an increase in exports is a positive development for a country‟s economy, however, the increase in exports of gas was paired with a decrease in corporate investment and an increase in unemployment from 1.1% to 5.1% during the 1970s. The Economist explained this puzzle by pointing to an appreciation of the Guilder, which was the Dutch currency at the time. Gas exports resulted in a large influx of foreign currency, which increased the demand for the Guilder and increased its value relative to other currencies. This meant that other industries within the Netherland suffered because the appreciation of the Guilder made them less competitive on international markets. Paired with policies that aimed to keep Dutch interest rates low to prevent the Guilder from

appreciating, this resulted in an investment outflow, reducing future economic potential. Since then, this phenomenon has been researched by numerous papers, and is also found to be more generally applicable, not just to discoveries of natural resources but to any event that results in a sudden increase of foreign capital flows. A more formal definition is provided by Magud & Sosa (2010): “Dutch disease refers to the effects of discoveries or price increases of natural resources that result in real exchange rate appreciation, factor reallocation, and de-industrialization. Similar effects may stem from other shocks entailing an increase in foreign exchange inflows, such as capital inflows, aid, and remittance”.

10

The mechanism through which aid could influence exchange rates is elaborated by Rajan & Subramanian (2011). They propose that aid is mainly invested in non-tradables within the receiving country, such as education, infrastructure or healthcare, which increases the wages in these sectors. This wage increase then attracts skilled labour out of the other sectors towards the aid-receiving sectors, resulting in an aggregate increase in wages. As a

consequence, the unit labour costs of exporting increase, which will reduce the international competitiveness of the exporting sector.

Besides an increase in the unit labour costs of exporting, a second mechanism related to this is that aid results in an appreciation of the real exchange rate. This mechanism is as follows. Higher wages caused by aid will be spent on tradables and non-tradable goods, but will only result in an increased price of non-tradables because prices of tradable goods are determined internationally, something which developing countries cannot realistically influence, whereas prices of non-tradable goods can inflate due to domestic increased demand. Thus, this means that the price of non-tradables relative to tradables increases, which is synonymous to the real exchange rate. As a consequence of the real exchange rate appreciation, the export sector‟s international competitiveness and profitability will decrease since their exports now become more expensive to other countries.

This effect can cause a serious threat to the development of a country seeing as the literature seems to suggest that exports and a corresponding development of the industrial sector are key to the development of a country (McMillan & Rodrik, 2011; Johnson, Ostry, and

Subramanian, 2006). The export sector is to a big extent associated with learning by doing, a process through which the productivity of a country increases and thereby results in an increase in growth. Thus, a decrease in the size of the export sector could potentially harm growth. This will be further elaborated in the literature review below.

2.3 Conditions on which Dutch disease has an impact on aid effectiveness

Most literature on the Dutch Disease makes use of family B and C models, in which aid influences growth through the mechanisms mentioned above. Much like the general literature on development aid, the literature on the Dutch disease is divided as well.

11

Dutch disease‟s effect on long-term growth. The former literature may not capture the complete picture. To illustrate this it may be useful to look at the diagram below.

Figure 3: The figure is taken from Rajan & Subramanian (2011). It highlights the channels through which foreign aid may influence growth.

Literature that only examines the real exchange rate may not capture the aggregate effect on a country‟s development. While it is true that an appreciation of the exchange rate could harm the export sectors, it could be that the positive effects of development aid simultaneously cause an increase in the productivity of the export sector, causing the export sectors to maintain their competitiveness despite an increase in exchange rates, which is what happens during the development process of a country. Therefore, it is necessary to look a step further than exchange rates, by looking at growth. If the exchange rate appreciates and export sectors are able to maintain their competitiveness due to an increase in productivity, this should be reflected in growth, which should remain stable (in the case the increase in productivity exactly offsets the increase in exchange rate). Therefore, this paper will look at growth to capture the aggregate effect of aid, with exchange rate appreciation as an intermediate channel.

The literature review on the Dutch disease will continue as follows. Literature on the Dutch disease distinguishes numerous factors that could influence whether DD effects can be mitigated. The main factors being: learning-by-doing spill-overs between both the tradable and non-tradable sector, exchange rate misalignment due to inappropriate macroeconomic policies, volatility of aid, and the sort of development aid. These factors will be further elaborated below.

2.3.1 Learning-by-doing spillovers between the tradable and non-tradable sector

12

source of growth in most Dutch disease models. The mechanism is that trade results in a positive externality called learning by doing. Research by Van Wijnbergen (1984) states that economic growth is mainly explained by learning by doing, a process by which performing a certain activity results in a learning process which increases productivity over time. In Van Wijnbergen‟s model the assumption made is that learning by doing is to a big extent associated with the tradable sector, and therefore a reduction in the tradable sector due to exchange rate appreciations would result in a corresponding slowdown in productivity and growth.

On the other hand, Torvik (2001) question the assumption that learning by doing is only associated with the traded sector, and creates a different model in which learning by doing results in imperfect spill-overs between the tradable and non-tradable sector. When

developing countries receive foreign aid, this shifts (skilled) labour towards the non-tradable sector due to the higher wages in this sector. In their model, the increase in labour in the non-tradable sector shifts learning-by-doing towards this sector. In the long run, this increased productivity in the non-tradable sector reduces the price of non-tradables, and hence the exchange rate is also restored (ratio of tradables to non-tradables). Essentially in this model spill-overs to the non-tradable sector allow the economy to shift to a new long-run

equilibrium.

Adam and Bevan (2004) examine this in an empirical setting in Uganda. They find that

productivity (learning by doing) spill-overs are present in the tradable and non-tradable sector. In their model, which is tested using Ugandan data, investment in public infrastructure results in productivity spill-overs that are skewed towards the non-tradable sector. This implies that Dutch Disease effects are present in the short-run, however, as predicted in the model by Torvik (2001) they find that in the medium/long run the steady state has shifted and the export sector finds a significant increase in growth.

2.3.2 The effect of monetary and fiscal policies on exchange rate misalignment

13

drive up the value of domestic currency (the nominal exchange rate), which also increases the ratio of non-tradable to tradable prices, resulting also in an increase of the real exchange rate. In the case that this appreciation of the exchange rate results in an overvaluation where it is above its long-term equilibrium, the effects on growth are clear. The literature on the relationship between exchange rate overvaluation and growth seems to be quite conclusive. For example, papers by Cottani, Cavallo, & Khan. (1990) and Rodrik (2008), find that overvaluation of the exchange rate can be damaging to growth: “Overvalued currencies are associated with foreign currency shortages, rent seeking and corruption, unsustainably large current account deficits, balance of payments crises, and stop-and-go macroeconomic cycles, all of which are damaging to growth.”

However, an appreciation does not always mean it is overvalued. It could simply reflect a new equilibrium. Edwards and Aoki (1983) examine this phenomenon and find that the Dutch Disease may not be a disease after all as long as the shift in exchange rate represents a new long-run equilibrium. In their scenario, they model the effect of a sudden export boom of oil. They find that in the long-run, the relative price of the traditional export sectors (non-oil exports) declines relative to the price of non-tradables and to the price of oil. This implies that in the long-run, a new equilibrium is created that allows the competitiveness of the traditional export sectors to remain stable due to their decrease in price. However, they find that an excess spending of income generated from the export boom in the short-term can result in an overvaluing of the price of non-tradables relative to tradables above their long-run

equilibrium. In this case, the economy diverges from its equilibrium, the exchange rate becomes overvalued and the other sectors will suffer as a result. Thus, incorrect policies (domestic overspending in the short-term) results in an overvaluation of the exchange rate for which the other sectors suffer.

14

obtain credit, which reduces private investment that is necessary to induce growth. It has also kept the inflation rate high because increased domestic demand drives up prices. However, Younger concludes that as long as aid is not too volatile, and the appropriate policies are adopted, development aid does not have to be harmful to the economy. The author argues that the policies imposed by the IMF are too strict. If development aid is not volatile, Ghana can adapt to a new long-run equilibrium, in which it makes more sense to spend aid rather than keeping it as foreign reserves. According to Younger, the correct policy is therefore to implement looser monetary policies, but tighter fiscal policies. In this way money is put back into financial system so that firms can obtain loans and increase investments, but by using tighter fiscal policies, the same firms are taxed higher so that the additional money in the economy is kept in check. In this way you include aid in the equation rather than exclude it Besides monetary and fiscal policies, it seems there are numerous other policies to counter the negative effects of aid on inflation or exchange rate appreciation. However, one should note that the correct policy will of course be dependent on characteristics of the country‟s

economy. For example, Van Wijnbergen (1986) distinguishes between two types of

economies: one with a high demand for non-tradable goods (a savings gap), and one with an excess supply of non-tradable goods. In the economy with the excess demand for non-tradable goods aid will result in an appreciation of the exchange rate if it is not carefully

complemented with the correct monetary policies. However, the economy with the excess supply of non-tradable goods will be able to absorb the increased demand caused by aid, and in this economy different monetary and fiscal policies would be appropriate.

McKinley (2005) creates a general overview of possible policies, policies that aim to properly spend and absorb aid so that the effects of the Dutch disease can be mitigated. Besides the conventional monetary and fiscal policies, a solution to the appreciation of the exchange rate is import liberalisation. By reducing taxes on imports, the increasing amount of imports can put downward pressure on the exchange rate. Spending aid directly on imports rather than investing it domestically could complement this. Lastly, the authors argue that there is a limit to import liberalization, so in the short run an appreciation of the exchange rate to stimulate an increase in imports is necessary and may be an inevitable result of aid. Governments may thus not always want to counter an increase in the exchange rate by playing with the nominal exchange rate.

15

non-tradables. In developing countries this policy could have potential because developing countries often do not operate on their production possibility frontier, meaning that they have the capacity to increase supply if demand increases. However, this excess capacity is often limited by certain bottlenecks, such as lack of infrastructure or education. Eliminating these bottlenecks would increase supply, but at the same time policymakers have to be careful, since the investments to remove these bottlenecks could have an adverse effect and cause inflation.

Lastly, a potential policy could be to invest aid in increasing the productivity of tradable goods. An increase in productivity of tradables will keep the price down, which can keep the exporting sector competitive despite an increase in the exchange rate.

Summarizing, there are numerous policy options available whose effectiveness depends on many different factors. For the scope of this research, it is not possible to control for the appropriate monetary and fiscal policies, since these are highly dependent on the country‟s characteristics, and economists‟ opinions about the appropriate policies differ as hinted by Younger (1992) and McKinley (2005). However, a policy such as import liberalisation is a policy whose effect is more straightforward, and thus can be controlled for.

2.3.3 Volatility of development aid

Besides incorrect policies, a related factor that could affect the valuation of the exchange rate is the consistency of the development aid provided. It is sometimes the case that gifts of development aid are volatile: the amounts and timing of given aid can vary, which could result in a shock to the economy if it is not accompanied by appropriate fiscal and monetary policies. Furthermore, as mentioned earlier in the paper of Younger (1992), the inconsistency of aid received could prevent developing countries from including aid into their long-run equilibrium.

16

17

3 Empirical model and testable hypotheses

As mentioned earlier, the purpose of this paper is to examine the effect of development aid on growth, with exchange rates as an intermediate channel. This makes it possible to understand the underlying mechanisms through which aid influences growth. Rajan & Subramanian (2011) construct a model that examines this relationship, which, with a few adaptations, will be the model used by this paper. The purpose of their model is to examine the channel

through which Dutch disease operates, which in this case is hypothesized to be exchange rate overvaluation. To determine whether the channel is at work, the model compares whether industries that are most sensitive to this channel grow differentially faster or slower in

countries where the channel is most likely to be operative. The channel we are interested in is exchange rate overvaluation, and the channel is most likely to be operative in industries that are sensitive to exchange rate appreciation. Thus, the hypothesis is that, if Dutch disease is present, countries that receive more aid should see relatively slower growth in industries whose competitive position is most sensitive to exchange rate appreciation.

( ) (

)

Taken from Rajan & Subramanian (2011), equation (1) is the core model. It examines the direct relationship between aid and growth differentials between industries. In this model,

Growthij is the annual average rate of growth of value added of industry i in country j over a

fifteen-year period, minus the annual average growth rate of the total manufacturing sector of country j over said period. By subtracting growth of an industry with total growth in the manufacturing sector, country specific effects such as inflation or macroeconomic shocks are being filtered out. This makes the use of a deflator or the use of country fixed effects

18

industry-level deflators and country fixed effects rather than subtracting with total

manufacturing growth. This paper does not use country fixed effects because by subtracting with the growth foot one already filters out country fixed effects.

The explanatory variable in equation (1) is the fraction of Aid to country j to GNI of country j, which is the average share of aid to GNI for that country over the fifteen-year period.

The variable E(i) is the exportability index and is a dummy variable that is equal to 1 if the industry has a revealed comparative advantage index above the median across industries (averaged across countries in the sample). This dummy variable captures the industry‟s sensitivity to Dutch disease effects. Seeing as one of the channels through which Dutch disease operates is exchange rate overvaluation, industries that are most important for exports should be more affected by overvaluation and grow differentially slower. Secondly, another channel through which Dutch disease operates is an increase in wages. Developing countries, like the ones in this paper‟s sample, mostly have a comparative advantage in labor-intensive sectors due to their relatively lower wages. Thus if wages increase due to Dutch disease, these industries should be more affected, and grow differentially slower. The methodology to obtain the exportability index is different from Rajan & Subramanian (2011) due to data availability, however, it should capture the same effect.

The choice to use a dummy variable rather than a continuous variable is because the model looks at growth differentials between industries. Therefore, by using a dummy variable we are able to distinguish between two types of industries, industries who are more and industries who are less likely to be affected by Dutch disease, and determine whether the former group grows differentially slower. Also, distinguishing between different groups of industries is in line with most literature on Dutch disease. For example, Corden & Neary (1982) splits the economy into different sectors: the booming (natural resource) export sector and the lagging (traditional) exports sector. Even though the mechanisms behind Dutch disease caused by natural resources are slightly different, the results are the same: Dutch disease causes a certain sector to lag behind. Essentially the dummy variable allows this paper to distinguish between the lagging export sector.

19

that non-export industries do actually export, however, they are simply play a less important role in the country‟s exports.

These variables complete the base model. In equation (1) the coefficient of interest is α, which captures the interaction between the amount of aid a country receives and an industry‟s

sensitivity to exchange rate appreciation caused by the Dutch disease. The hypothesis is that alpha will be negative: in countries that receive more aid, industries who are sensitive to Dutch disease effects should grow relatively slower.

3.2 Addition of control variables to the base model

( ) (

)

In equation (2) a vector of control variables X is added to the model, interacted with the exportability index (dummy). The interaction of the control variables with the exportability index is necessary due to the nature of the model, which looks at growth differentials. These growth differentials are computed in the model by the interaction with the indicator variable. For example, if we want to control for whether growth differentials are caused by the

volatility of aid rather than the level of aid, we need to add aid volatility interacted with the indicator variable to the model to distinguish whether volatility affects growth differently between the two groups of industries.

As for the control variables used in the regression, one was already mentioned in the example above: aid volatility. Following the methodology of Arellano et al (2009), the volatility of aid will be calculated as the standard deviation of aid inflows over the sample 15 year period divided by the mean, which is the coefficient of variation. This control variable will be included following the literature review, which concludes that aid volatility could result in a misalignment of the exchange rate, and thereby harm growth of the export sector. Thus if the variable is significant, it could be that part of the reduced growth in industries who are

20

Secondly, we want to control for whether it is aid that results in Dutch disease symptoms, or whether this could be the result of bad policies. As mentioned in the literature review, it could be that aid leads to overvaluation of the exchange rate due to bad monetary and fiscal policies, and not so much due to aid itself. For the scope of this research, it is not possible to control for appropriate monetary and fiscal policies since the appropriate policy depends on numerous different factors, and even then opinions among economists differ. However, the implications of trade liberalization policies compared to monetary and fiscal policies are more

straightforward: a reduction in tariffs on imports could increase the amount of imported products to the country, reducing the current account surplus and putting downwards pressure on the exchange rate. Also, another channel through which trade liberalization could affect the results are export tariffs: it could be that lower growth of the tradable sector is caused by increased export tariffs. Thus, this paper will control for a country‟s openness to trade by using the ratio of a country‟s exports to GDP as a control variable, making the assumption that countries with more liberal trade policies have a higher share of exports and imports to GDP.

21

exchange rate overvaluation, which may then lead to Dutch disease effects, causing export industries to grow relatively slower.

If regression (3) turns out to be significant, equation (4) will test whether exchange rate overvaluation indeed causes export industries to grow relatively slower, by regressing average annual growth against the term Overvaluation interacted with the exportability indicator. The coefficient beta in equation (4) is expected to be negative: exchange rate overvaluation should cause export industries to grow relatively slower. This may provide further evidence that exchange rate overvaluation is the channel through which the Dutch disease of development aid operates.

A last test to determine whether exchange rate overvaluation is the channel is performed by adding both aid and overvaluation in the same regression (equation 5). If the channel through which aid affects growth is exchange rate overvaluation, adding this term to the regression should take over some of the explanatory power of aid. The expectation is that adding overvaluation to the regression should reduce the magnitude of the aid coefficient alpha compared to its estimated magnitude in equation (1), and possibly lower its significance. The coefficient for overvaluation is expected to be negative similar to equation 4.

3.4 Examining whether Dutch disease effects can be mitigated by investments of aid in manufacturing

( ) (

)

22

The second reason to assume an increased share of aid in the manufacturing sector can prevent a loss of competitiveness is because it may prevent resource outflow. As mentioned earlier, the channel through which Dutch disease operates is that aid causes an increase in wages of the non-tradable sector, which starts a kind of bidding war against the tradable sector. For example, skilled labour may be bid out of the tradable sector due to the higher wages in the non-tradable sector. Thus, if the tradable manufacturing sector receives a

relatively higher share of aid, it may be better equipped to bid against the non-tradable sector, preventing resource outflows and loss of competitiveness.

The last reason to assume an increased share of aid in the manufacturing sector can prevent a loss of competitiveness is because it may result in less exchange rate appreciation. This assumption more contentious however: the assumption is that a bigger share of aid invested in the tradable manufacturing sector means that a smaller share of aid is invested in the non-tradable sector. Correspondingly, the price of non-non-tradables will inflate relatively less, resulting in less appreciation of the RER. The assumption is contentious because

manufacturing is not the only tradable sector, thus a higher share invested in manufacturing does not per say mean less investment in the non-tradable sector.

If the hypothesis holds, the coefficient for aid invested in the manufacturing sector should be smaller than total aid, seeing as a bigger share of aid in manufacturing is hypothesized to reduce the effects of Dutch disease and cause smaller growth differentials. A formal test will be used to compare coefficients.

Besides testing Rajan & Subramanian‟s model using a different database and sample,

equation (5) will be the main contribution to the literature because earlier literature by Rajan & Subramanian (2011) only looks at the effect of overall aid on growth. This paper aims to decompose aid into the sectors in which it is spent, to determine whether spending more aid in the tradable manufacturing sector could reduce the symptoms of Dutch disease.

3.5 Advantages of using a cross-industry, within-country comparison model

23

The main advantage is that this resolves the problem of omitted variable bias that plagues cross-country regression, namely, that a country-specific variable that is not included in the regression explains the observed correlation. Furthermore, by comparing variation within the same industry (manufacturing), it is also possible to rule out factors that cause a specific industry to grow faster or slower to explain the results.

The second advantage of running a within-country regression is that it reduces problems of causality. Namely, a big problem of cross-country regressions is that it does not control for the rationale for why aid is given: it could be that countries that have lower growth are given more aid, rather than more aid resulting in lower growth according to the Dutch disease

hypothesis. By comparing inter-industry growth differentials the rationale for why aid is given becomes less important, since we are comparing differences in growth rather than the level of growth. It is harder to think of a rationale to give aid to countries with higher growth

differentials. Assuming growth differentials are caused by exchange rate appreciation, one would have to find a reason to give aid to countries with a higher real exchange rate. While it is not impossible to find reasons for this, it is less likely. Still, the use of an instrumental variable would provide better proof, and the lack of instrumental variable in this paper could be considered a limitation.

The last advantage of their methodology is that it allows one to determine whether exchange rate overvaluation is the channel through which aid influences growth. Firstly, they examine the channel from aid to (excess) exchange rate appreciation, and secondly, they examine the channel from exchange rate appreciation to growth (differentials).

24

4. Data description and outlier detection

4.1 UNIDO INDSTAT4 (rev 3)

The dataset used to collect data on value added growth is INDSTAT4 (rev 3) by the United Nations Industrial Development Organization (UNIDO). The data have been gathered from questionnaires sent by UNIDO to national statistical authorities. The dataset contains data disaggregated on the country and industry level, from the time period 1990 to 2012. The industries are separated at the 3-digit and 4-digit level of ISIC codes (revision 3), and comprises more than 150 manufacturing sectors and sub-sectors. The countries contained in the database are both developed and developing countries, and add up to a total of 139 countries.

4.1.1 Growth rate of value added

The main variable used in the regression is value added growth. Value added captures the value of output minus the value of input. This prevents misattributing value to an industry that is added by intermediate goods produced in other industries. The data on value added are in USD, current prices. To correct for deflation, value added growth per industry has been compared with value added growth in the total manufacturing sector, as described in chapter 3.

The data on value added contain many gaps and therefore certain countries had to be excluded from the regression. This paper, following the methodology of Rajan & Subramanian (2011), only includes countries in the regression that contain at least seven annual value added growth observations in the period from 1994 to 2008. This time period has been chosen since most observations fall within this timeframe. Also, seeing as the model is a cross-sectional model and therefore takes the average of the annual value added growth over this time period, a time period had to be chosen that should capture the long-term effect of development aid. The goal was to choose a time span of 10 to 15 years (to capture the long term effect) with as many observations as possible. This turned out to be the period from 1994 to 2008 that fulfilled these criteria.

25

initial year of the time period. Following these criteria, a set of 18 developing countries remained. See appendix table 1 for the sample of countries.

As for the set of industries used in the regression, for each country, only those industries that had at least 7 growth observations over the time period between 1994 and 2008 have been averaged and included as an observation. Furthermore, a set of 4-digit level industries have been removed from the sample as well. The INDSTAT4 database contains industries at the 3-digit and 4-3-digit level, where 4-3-digit industries are sub-industries of the 3-3-digit industries. Seeing as we are comparing cross-industry growth, it is not possibly to both include 3-digit and its respective 4-digit level sub-industries in the regression, seeing as we would be

comparing a sub-industry to itself (plus the aggregate of the other sub-industries). Therefore, the sub-industries that also have a 3-digit ISIC code have been removed from the data, leaving a total of 60 industries remaining in the sample. The choice to remove the 4-digit industries rather than the 3-digit industries is due to data availability: significantly more observations were available at the 3-digit level.

The remaining sample thus consists of 18 countries, each having 60 industries, which equals a total of 1080 observations. However, as mentioned in the methodology, only those industries that have at least seven annual average growth observations over the 15 year sample period will be included in the regression. Due to the many gaps in the data, 547 observations had to be deleted leaving a total of 533 observations remaining.

4.1.2 Initial industry share

The same value added data have been used to compute Initial industry share. This has been calculated by dividing the value added of each industry by the total value added in

manufacturing for the first year in the sample period that contains an observation. Due to the many gaps in the data, the first observation in the sample period may be different for each industry. Therefore, the shares do not add up to exactly 100% since this would only be possible if the first year of observation would be the same for each industry. However, this should not matter for the results since we are only interested in the initial year in which growth is calculated, which is the same observation for the calculation of initial industry share.

26

that are very small, and may only consist of one factory in the country. Including these

industries in the sample is not meaningful since they often grow disproportionately faster than the other industries in our sample and their small size makes their growth more susceptible to other factors. Thus, only industries with an initial industry share above 0.5% of total

manufacturing value added have been included in the sample. A total of 207 industries had a share below 0.5%, and were deleted from the sample, leaving a total of 326 observations remaining.

4.2 World bank

4.2.1 Exportability index

To create an indicator variable for the importance of each industry in exports, data on the revealed comparative advantage of each country in our sample have been consulted from the World Bank. The database gives each industry an index for their comparative advantage: the higher the value, the higher the relative advantage of exporting the good for said country. To compute the indicator Exportability index, first, the average value of revealed comparative advantage across countries from this paper‟s sample has been calculated (for each industry). Then, only the industries above the median of the average across countries are considered as high in “exportability”, and were given an indicator value of 1. These industries are often also the industries with the highest share of value in total exports. The results of the calculations are the following. The industries that were found to be above the median level of comparative advantage are in this paper‟s sample are: food, textiles, hides & shins, footwear , stone & glass, metals, fuels, chemicals, minerals, and animal foods. Unlike the industries in our sample, the above mentioned industries from the dataset of the world bank are not coded according to ISIC rev 3. Therefore, a rough translation of the industries mentioned above to the more disaggregated industries from this paper‟s sample had to be made. The final values for the indicator variable be found in table 1 under Exportability index 1.

27

is considered to be a capital intensive industry. This could possibly influence the results since the natural resource sector may be less susceptible to Dutch disease effects. Since the natural resource sector is capital intensive it has less labour costs, and therefore it will be less affected by an increase in wages caused by DD effects. Furthermore, one could argue that an

appreciation of the exchange rate may not influence the competitiveness of the sector as much since prices for natural resources could be considered more flexible. Compared to normal goods, natural resources only have to be extracted and transported, and as long as the world price (converted to national currency) is high enough to cover these costs, it can lower its prices and continue exporting.

Therefore, another indicator variable (Exportability index 2) will also be used as a robustness check. This indicator is 1 for the industries that have been most closely associated with the growth of developing countries as they have moved out of agriculture (Rajan & Subramanian, 2011): textiles, clothing, leather and footwear. Thus the indicator only contains the labour-intensive textile industries and does not include the capital labour-intensive natural resource sector which may be less susceptible to DD effects. Furthermore, as a second robustness check the natural resource industries will be dropped from the sample.

ISIC

code Industrial sector

Exportability index 1

Exportability index 2

153 Dairy products 1 0

154 Grain mill products; starches; animal feeds 1 0

155 Other food products 1 0

171 Beverages 0 0

172 Tobacco products 0 0

191 Spinning, weaving and finishing of textiles 1 1

202 Other textiles 1 1

210 Knitted and crocheted fabrics and articles 1 1

221 Wearing apparel, except fur apparel 1 1

222 Dressing & dyeing of fur; processing of fur 1 1 241 Tanning, dressing and processing of leather 1 1

242 Footwear 1 1

251 Sawmilling and planing of wood 0 0

269 Products of wood, cork, straw, etc. 0 0

273 Paper and paper products 0 0

281 Publishing 0 0

289 Printing and related service activities 1 0

291 Reproduction of recorded media 0 0

28

331 Refined petroleum products 1 0

351 Processing of nuclear fuel 0 0

359 Basic chemicals 1 0

369 Other chemicals 1 0

1520 Man-made fibres 1 0

1600 Rubber products 0 0

1730 Plastic products 0 0

1810 Glass and glass products 1 0

1820 Non-metallic mineral products n.e.c. 1 0

1920 Basic iron and steel 1 0

2010 Basic precious and non-ferrous metals 1 0

2230 Casting of metals 1 0

2310 Struct.metal products;tanks;steam

generators 0 0

2320 Other metal products; metal working

services 0 0

2330 General purpose machinery 0 0

2430 Special purpose machinery 0 0

2520 Domestic appliances n.e.c. 0 0

2610 Office, accounting and computing machinery 0 0 2710 Electric motors, generators and transformers 0 0 2720 Electricity distribution & control apparatus 0 0

2930 Insulated wire and cable 0 0

3000 Accumulators, primary cells and batteries 0 0

3110 Lighting equipment and electric lamps 0 0

3120 Other electrical equipment n.e.c. 0 0

3130 Electronic valves, tubes, etc. 0 0

3140 TV/radio transmitters; line comm. apparatus 0 0 3150 TV and radio receivers and associated goods 0 0 3190 Medical, measuring, testing appliances, etc. 0 0 3210 Optical instruments & photographic

equipment 0 0

3220 Watches and clocks 0 0

3230 Motor vehicles 0 0

3320 Automobile bodies, trailers & semi-trailers 0 0

3330 Parts/accessories for automobiles 0 0

3410 Building and repairing of ships and boats 0 0

3420 Railway/tramway locomotives & rolling stock 0 0

3430 Aircraft and spacecraft 0 0

3520 Transport equipment n.e.c. 0 0

3530 Furniture 0 0

3610 Manufacturing n.e.c. 0 0

3710 Recycling of metal waste and scrap 0 0

3720 Recycling of non-metal waste and scrap 0 0

29 4.2.2 Aid as a percentage of GNI

Data on development aid have been obtained from the World Bank. The variable “Net ODA received as a percentage of GNI” was available for all countries in the sample, with only four missing observations. Missing observations are no concern since the average aid over the time period 1994 to 2008 is computed for the regression. The same data were used to compute aid volatility, calculated as the standard deviation of aid inflows over the sample period, divided by the average. The results can be found in the table below.

Country Aid (% of GNI)

Albania 6,529208 Azerbaijan 3,014983 Eritrea 23,03056 Ethiopia 12,25749 Georgia 6,802759 India 0,321337 Indonesia 0,726729 Jordan 5,582162 Kenya 4,491502 Kyrgyzstan 13,23045 Malawi 23,64041 Mauritius 0,650551 Moldova 5,916726 Morocco 1,456853 Mongolia 14,26681 Senegal 10,41364 Uganda 13,60565 Yemen, Rep. 3,570069

Table 2: Aid as a percentage of GNI, averaged over the period from 1994 to 2008

4.2.3 Trade openness

30

more sophisticated measures for trade openness such as the Sachs and Warner index were not available for the sample period.

4.3 Darvas & Zsolt (2012a) - Real effective exchange rate

Data about the real effective exchange rate (REER), which will be one of the measures for overvaluation, were obtained from the working paper by Darvas & Zsolt (2012a). The REER measures the real value of a country‟s currency against a basket of trading partners, and is calculated by weighing the nominal effective exchange rate with the consumer price index, which is a measure for the relative cost of goods between a country and its trading partners. Data for the REER were available for each country in the sample, containing observations for all the years. The database contains both annual and monthly data, considering different amounts of trading partners. This paper uses annual REER data, with the basket of 172 trading partners. To fit the data into this paper‟s cross-sectional regression, the average of the REER over the sample period was computed for each country.

4.4 Penn World Tables – Overvaluation of real exchange rate

Data to calculate overvaluation of the real exchange rate according to Rodrik (2008) have been gathered from the Penn World Tables. Rodrik provides a measure for exchange rate overvaluation which is an index for the domestic price level of a country corrected for the Balassa Samuelson effect. The Balassa Samuelson effect predicts that as countries grow richer, their real exchange rate should appreciate due to increasing prices of non-tradable goods. Essentially Rodrik‟s measure for overvaluation is the country‟s real exchange rate adjusted for the Balassa Samuelson effect. This adjustment is done by subtracting the real exchange rate by the predicted real exchange rate, which is what the exchange rate should be according to the predicted Balassa Samuelson effect, given the country‟s real GDP per capita. The methodology to derive the index is the following. Firstly, the logarithm of the real

31

overvalued, and positive when it is undervalued. The logarithm of overvaluation will be used as explanatory variable in the regressions.

As for data availability, data in the Penn World Tables was available for all countries in the sample except for Eritrea.

4.5 OECD Creditor Reporting System – Commitments of aid decomposed per sector

The OECD creditor reporting system contains disaggregated data on aid, which allows one to see for which purpose aid is being granted, or which sector it is being invested in. To test the last set of hypotheses from this paper, data about the share of aid invested in the

manufacturing sector had to be gathered from this database. The share of aid invested in industrial sector is calculated by dividing it by the total (all sectors). Data were available for each country in the sample, with only a few years of observations missing. Also for this measure the average had to be taken over the sample period to fit into the cross-sectional regression.

4.6 Outlier detection and summary statistics

Figure 4: a scatterplot of average annual growth against aid as a percentage of GNI, including outliers

32

Looking at the scatterplot above, there seems to be at least one clear outlier, observation 251. This observation is a small industry in Mongolia (close to 0.5% industry share), which probably suffered a big shock unrelated to Dutch disease. Therefore, it will be removed from the sample.

Another test to detect outliers is Cook’s distance, which measures the influence of an observation on the estimated coefficients. Essentially, it runs the regression leaving out one observation at a time, and measures the change in the estimated coefficients. The bigger the change, the more influential the observation. According to the STATA manual, a cook‟s distance higher than 4/N can be problematic. In this paper‟s sample, 7 observations (including observation 251) had a value higher than 4/N and were dropped from the sample. The

remaining sample consists of 319 observations. To get a clearer picture of the remaining 51 industries in the sample and the amount of observations per country and industry, consult table 1 & 2 in the appendix.

What can we say about the remaining amount of observations? Looking at table 1 in the appendix, it seems like observations are quite balanced across countries. Only Uganda seems to have rather few observations, but this should be no cause for concern seeing as the

observations across countries will be pooled together, which is made possible by subtracting the growth foot of each country from each observation and thereby correcting for country fixed effects. As for the remaining amount of observations across industries, it seems to be a little less balanced. Unsurprisingly, the industries exporting primary goods such as food, minerals, basic iron and steel and tobacco, have more observations than the more

sophisticated industries such as electronics and machinery. However, again this should be no cause for concern since the indicator variable splits the industries into two different groups (export- and non-export industries), meaning that the regression will look at the aggregate amount of industries within each group, and not at specific industries. The aggregate amount of industries should sufficiently high to provide reasonably accurate averages of their average annual growth rate.

33

Variable Mean Standard

deviation

Median Maximum Minimum Number of observations Growth rate of value added ij 0.07977 0.2211 0.035664 1.505604 -0.91279 318

Aid as percentage of GNI j 6.887177 6.78071 5.582162 23.64041 0.321338 318

Exportability index 1 i 0.509434 0.500699 1 1 0 318

Exportability index 2 i 0.176101 0.381506 0 1 0 318

Real effective exchange rate j 95.94401 10.79263 94.28522 130.1536 83.68697 318

Overvaluation j -0.35711 0.434395 -0.33 0.61 -1.06 302

Trade openness j 74.712 29.30985 67.07515 126.8457 31.02129 318

Volatility of aid j 0.393929 0.133323 0.398247 0.640748 0.185454 318

Aid invested in manufacturing (% GNI) j

0.639087 0.699636 0.323823 2.229648 0.050448 318

Table 3: Summary statistics of the variables that will be used in the regression, excluding outlier

To get an indication of the relationship between variables consult the correlation matrix on the next page. Correlation above 0.4 could lead to problems of collinearity. This is the case for a couple of variables. Firstly, the two indicator variables are correlated (exportability index 1 and 2), and also aid invested in manufacturing and overvaluation are correlated, but seeing as these variables will not be used in the same regression this should be no concern. However, aid and the real effective exchange rate, and aid and the volatility of aid also have a high correlation and will be used in the same regression. This could lead to problems of

multicollinearity. Hence, after each regression the variance inflation factors will be calculated to determine whether these correlations could be problematic. Lastly, there is a high

34 Growth rate of value added Aid as percentage of GNI Exportability index 1 Exportability index 2 Trade openness Real effective exchange rate Aid invested in manufacturing (% GNI) Overvaluation Volatility of aid Growth rate of value added 1 Aid as percentage of GNI 0.0022 1 Exportability index 1 -0.0287 0.1572 1 Exportability index 2 -0.1409 0.0943 0.4507 1 Trade openness -0.0178 0.118 0.0191 0.0815 1 Real effective exchange rate 0.0117 0.4443 0.0234 0.0664 0.165 1 Aid invested in manufacturing (% GNI) -0.0104 0.9393 0.1637 0.0876 0.1195 0.3978 1 Overvaluation -0.0255 -0.4263 -0.0284 -0.0056 0.58 0.1452 -0.4826 1 Volatility of aid -0.0167 -0.4554 -0.0987 -0.0225 -0.1946 -0.3534 -0.3813 -0.1012 1

35

5. Results

The figure below provides a visualisation of the data on the basis on the model. Essentially the model used in this paper computes the growth difference between export industries (those with an indicator value of 1) and non-export industries (indicator value of 0), and determines whether the difference becomes bigger as aid increases. The figure below depicts the average growth for the export industries, minus the average growth for non-export industries for each country. Each dot thus represents the average growth differential between export and non-export industries for a country, plotted against the amount of aid said country receives. A negative value means that export industries grow slower than non-export industries. This is by no means an official measure, it is only meant to give a general idea of the data and what the model will examine.

Figure 5 represents the average growth differential between export and non-export industries for a country, plotted against the amount of aid said country receives. The average growth differential on the y-axis can be converted to percentages by multiplying by 100. The data used to compute this graph includes outliers.

The trend line is negative. This is according to the hypothesis, which predicts that industries most important for exports will be more sensitive to Dutch disease effects, and therefore grow differentially slower as aid increases. It seems like the negative slope of the trend line is mainly caused by the three bottom observations, representing the countries Senegal, Uganda and Mongolia. Whether the negative relationship between aid and growth differentials is

-0,3 -0,25 -0,2 -0,15 -0,1 -0,05 0 0,05 0,1 0 5 10 15 20 25 Av e ra ge g ro w th d iffe re n ti al Aid (% of GNI)

36

significant, also with the removal of outliers, will have to be determined in the regression results below.

In the following section the results of the regression estimations will be presented. All regressions are run using robust standard errors seeing as the Breusch-Pagan test for

heteroskedasticity returned a significant result for each of the models. Furthermore, for each regression the variance inflation factor (VIF) has been calculated to determine problems of multicollinearity. All the regressions returned a VIF value below 3, which is far below 10, the threshold value that most would consider to be problematic. Hence, there is no reason to assume the standard errors to be inflated due to multicollinearity.

5.1 The relationship between aid and growth differentials within the

manufacturing sector

Table 5. The effect of aid on sectoral growth

Dependent variable: Average annual growth rate of value added ij

(1) (2)

Panel A: interactions with Exportability 1 index Aid /GNI (j)* Exportability index 1 (i) -0.00145

[0.00718]

Aid in manufacturing /GNI (j)* Exportability index 1 (i) -0.0297

[0.0787]

Test for difference in coefficients Chi2=0.16 (Insignificant)

Observations 319 319

R-squared 0.0101 0.0087

Panel B: interactions with Exportability 2 index Aid /GNI (j)* Exportability index 2 (i) -0.01657

[0.01803]

Aid in manufacturing /GNI (j)* Exportability index 2 (i) -0.18867

[0.2037]

Test for difference in coefficients Chi2= 0.86 (Insignificant)

Observations 319 319

R-squared 0.0165 0.0188

Table 5: Coefficients estimated using an OLS regression with robust standard errors. Standard errors are reported in brackets below the estimated coefficients. Significance of the results is depicted using ***, ** and * denoting significance at the 1%, 5% and 10% level respectively. Column (1) is based on regression equation (1), column (2) is based on equation (6), refer to chapter 3. As an additional robustness check regressions have been run twice (Panel A and Panel B), each making use of a different indicator for exportability (exportability index 1 and 2 respectively, refer to table 1 for further details).

37

The question answered in the regression output above is whether development aid causes industries that are most important for exports to grow significantly slower than the remaining (non-export) industries. The interaction term between aid and either of the exportability indices in column 1 return a negative value, which does imply that export sectors grow slower than non-export sectors, albeit insignificantly so. The way to interpret these values is that a one percent point increase in aid causes the sectors most important for exports to grow 0.145 and 1.657 percent per year slower than the reference (non-export) sectors, making use of exportability index 1 and 2 respectively. Despite the sign of the coefficients being negative according to the hypothesis, both coefficients are insignificant, and therefore we cannot reject the null hypothesis. We cannot conclude that aid causes export sectors to grow significantly slower.

In column 2 the same base model is used, except that the variable (total) aid as a percentage of GNI has been replaced with aid invested in manufacturing as a percentage of GNI. Hypothesis (4) states that aid invested in manufacturing could potentially counter the effects of Dutch disease. Therefore, a bigger amount of aid invested in manufacturing should result in a smaller growth differential compared to the growth differential of total aid. This does not seem to be the case, seeing as the coefficient is actually larger for aid invested in

manufacturing. However, a test to check whether the coefficients are significantly different from eachother returned an insignificant result. This was expected due to the high correlation between total aid and aid invested in manufacturing from the correlation table. No meaningful conclusions can thus be drawn.

38

5.2 The relationship between aid and growth differentials within the

manufacturing sector: robustness to controls

Table 6. The effect of aid on sectoral growth: robustness to

control variables

Dependent variable: Average annual growth rate of value added ij

(1) (2) (3)

Panel A: interactions with Exportability 1 index

Aid /GNI (j)* Exportability index 1 (i) -0.00145 0.0022 -0.00090

[0.00718] [0.0047] [0.00652]

Trade openness (j) * Exportability index 1 (i) -0.00096

[0.00094]

Volatility of aid (j) * Exportability index 1 (i) -0.0398

[0.1026]

Observations 319 319 319

R-squared 0.0101 0.019 0.016

Panel B: interactions with Exportability 2 index Aid /GNI (j)* Exportability index 2 (i) -0.01657 -0.0042 -0.0098

[0.01803] [0.0103] [0.0152]

Trade openness (j) * Exportability index 2 (i) -0.0025

[0.00184]

Volatility of aid (j) * Exportability index 2 (i) -0.3239*

[0.1714]

Observations 319 319 319

R-squared 0.0165 0.025 0.0194

Table 6: Coefficients estimated using an OLS regression with robust standard errors. Standard errors are reported in brackets below the estimated coefficients. Significance of the results is depicted using ***, ** and * denoting significance at the 1%, 5% and 10% level respectively. Regression is based on equation (2) in chapter 3. As an additional robustness check regressions have been run twice (Panel A and Panel B), each making use of a different indicator for exportability (exportability index 1 and 2 respectively).

39

model. Same as for trade openness, adding this variable to the model reduces the magnitude of the aid coefficient. However, the coefficient for aid volatility seems to be much larger than the aid coefficient when included in the same regression, which could imply that it provides more explanatory power than the level of aid when explaining growth differentials. Despite its magnitude, aid volatlity also returns a p-value that is insignificant at the 5% level, so the results are only suggestive, not conclusive.

5.3 The relationship between aid and growth differentials within the

manufacturing sector: robustness to sample

Table 7. The effect of aid on sectoral growth: robustness to sample

Removing natural resource sectors

Removing 4-digit industries

(1) (2)

Panel A: interactions with Exportability 1 index

Aid /GNI (j)* Exportability index 1 (i) -0.00096 0.00097

[0.00631] [0.0029]

Observations 305 181

R-squared 0.009 0.0007

Panel B: interactions with Exportability 2 index

Aid /GNI (j)* Exportability index 2 (i) -0.00401 0.00256

[0.00557] [0.00356]

Observations 305 181

R-squared 0.013 0.0021

Table 7: Coefficients estimated using an OLS regression with robust standard errors. Standard errors are reported in brackets below the estimated coefficients. Significance of the results is depicted using ***, ** and * denoting significance at the 1%, 5% and 10% level respectively. Regression is based on equation (1) in chapter 3. As an additional robustness check regressions have been run twice (Panel A and Panel B), each making use of a different indicator for exportability (exportability index 1 and 2 respectively). In column 1 sectors producing primary natural resource products have been dropped from the sample, namely, industries 331, 1820, 1920 and 2010. In column 2 all 4-digit industries have been dropped from the sample.