social institutional environment

Get Complete Project Material File(s) Now! »

Empirics and data

In most cases where there is conceptual disagreement over an issue one will encounter di-verging views on how to measure the phenomenon in question. Institutional approaches to economics are no exceptions. Glaeser et al. (2004) criticize different subjective measures like Polity IV for measuring outcomes of choices (e.g. by a dictator) rather than underlying facts and for being quite volatile.
Following the reasoning put forth in the background section, the author of this thesis has decided to use several alternative measures that tracks the level of corruption, to what ex-tent a country has contract institutions or extractive institutions and to what extent the citi-zens have confidence in the state to protect their private property and to enforce contracts.
Contract-Intensive Money (CIM) is a concept developed in Clague, Keefer, Knack and Ol-son (1999) and acts as a proxy for contractual enforcements and private property rights. CIM is the variable used for the base regression below and is a widely available, objective measure defined as the ratio of non-currency money to the total money supply and will ap-proach 1 as confidence in the banking system increases and as the willingness to do busi-ness transactions in cash decreases. As stated, this concept is aimed at measuring formal in-stitutions. It is however the belief of the author that it will also constitute a reasonably good proxy for social institutions such as trust. In any case, as reasoned by Knack and Zak (2001), good formal institutions might offset some of the negative effects from a bad social institutional environment.
European settler mortality is presented in Acemoglu et al. (2001) and documents the living conditions for European settlers arriving to a new colony. Where many settlers perished extractive institutions were put in place and where living conditions were favorable, con-tract institutions emerged, so that a high settler mortality rate will indicate an extractive in-stitutional system. Settler mortality measures the amount of deaths per thousand arriving settlers, including their replacements, which makes possible the peculiarity that some coun-tries have a rate of well over a thousand.
Alesina, Devleescauwer, Easterly, Kurlat and Wacziarg (2003) compile a comprehensive index of ethnic fractionalization for a wide array of countries. The figure is a good proxy for political disturbances, corruption and other related variables. It will be used here to control the robustness of the findings.
Data have been collected for 82 developing countries from 1960 to 2006, the selection be-ing identical to that of Rajan and Subramanian (2008). The following data sources have been used to compile the dataset:
• M2 and Currency in circulation from IMF (2008)
• Settler mortality rates from Acemoglu et al. (2001)
• Ethnic fractionalization from Alesina et al. (2003)
• Growth rate of GDP per capita, ODA as a percentage of GDP, Nominal GDP, Im-port and Export shares of GDP, Inflation (GDP deflator), CO2-emissions, popula-tion growth, external debt as a percentage of GDP and military spending as a per-centage of GDP from World Bank (2007).
The dataset is cross-sectional in nature and has an added time dimension (panel data). There are drawbacks to this method like any other, but it is clearly superior to cross-sectional estimations in at least one aspect: it addresses endogeneity issues by correcting for country and year specific effects.

Descriptive statistics

From 1960 to 2006, the countries in the sample grew at a mean rate of 1,56 percent as compared to 1,02 percent for the period after 1980. On average the countries in the sample received foreign aid amounting to 5,8 percent of domestic GDP. In the period after 1980 the corresponding figure was 6,7 percent.
The mean CIM measure for the recipient countries was 0,54 for the full period as com-pared to 0,56 for the period after 1980, while approaching 0,60 in recent years, indicating that public confidence in authorities has increased steadily over the reported period.
Hence, growth rates slowed down somewhat, while institutional quality increased quite steadily over time. Do note how this may indicate some support for the neoclassical as-sumptions outlined in the theoretical part. Poor countries may have grown faster for low levels of capital to labor ratios, that is exhibited some tendencies to diminishing returns to capital.
The development is illustrated in Figure 1, which outlines development in institutional quality and mean growth rates among the countries in the sample over the full reporting period. As can be seen from the picture institutional quality was strengthened over time, but does not appear to have any apparent relationship to elevated growth rates.
It is apparent how growth rates show some signs of a u-shaped development with a trough around the mid 1980’s, with peaks in the beginning of the graph in the late 1970’s and sub-sequently in recent years. Naturally, it might be the case that the dataset represents a snap-shop of a bigger picture with growth rates are oscillating around the 2 percent level.
Figure 2 graphs the sample mean income in PPP adjusted GNI per capita against develop-ments in institutional quality since 1980. As is obvious from the figure, both values increased over time. This is important information to keep in mind in empirical analysis of these figures, since problems with endogeneity are likely present and cannot be fully con-trolled for.
Hence there is a clear common trend among institutional quality and income levels but no ap-parent relationships between institutions and growth rates of incomes. It is apparent how short run growth rates are a product of a much more sophisticated set of variables than just institutions.

Specification

Including settler mortality in the model does not radically alter the results, but decimates the sample by about 50 percent. The variable has therefore been excluded from the final empirical specification. Year dummies and country dummies are included in the regressions to filter out year specific and country specific effects, where in particular the former have been substantial.
Following the reasoning outlined in (4), that is, provided aid is affected by institutional quality, this makes possible the calculation of the marginal effect of ODA by taking the de-rivative of the growth equation with respect to aid:
The sum of aid’s impact on growth is the magnitude of its original impact in addition to the combined impact with institutions.

Findings and discussion

At this point the expected sign of the ODA / GDP (aid) variable is to be regarded as inde-terminate, but recalling that an aid disbursement has been predicted to diminish with dete-riorating institutional quality in (4) a positive ODA x Institutions interaction term is a rea-sonable assumption. Interestingly, it is negative and significant in estimation 1 in the ap-pendix, where the results of (5) are displayed. Estimation 2, displaying the outcome of the 1980-2006 regression, reveals similar results.
Recalling (6), these results are peculiar. At this point, some might recall the finding of Chauvet and Guillamont (2001): aid is more likely to be effective in difficult economic set-tings. Regardless of that being true to reality or not, it refers to short-term exogenous shocks, natural disasters and volatile terms of trade, not to bad governance and definitely not to long-term corruption or fear of expropriation. Interpreted literally, the results would mean that aid is more effective (if at all) in a state of deteriorating institutional quality.
Inspection of table 1 below can offer some insight into the potential problems with corre-lations in between some or many of the independent variables, commonly referred to as multicollinearity, and – as will be returned to – can introduce unacceptable levels of uncer-tainty (or even complete randomness) to regression results.
Near collinearity potentially poses a threat to the reliability of the results if ‘some or all the X variables are so highly collinear that we cannot isolate their individual influence on Y’ (Gujarati, 2003, p. 349). This is a vital remark, since numerous related studies in the past that have dealt with interaction terms most probably have been plagued by serious multicollinearity, which is a big contributing reason to the fact that many previous findings have proven impossible to replicate (Roodman 2007b).
Gujarati (2003) further suggests that multicorrelation is potentially a serious problem when pairwise correlations are in excess of 0.8. Hence, the correlation of the cross-term with the ODA figure and the institutions variable is unreasonably high. This is, however, not a sufficient criterion for establishing whether multicollinearity is problematic or not (Gujarati, 2003).
To test whether these problems are present the variables denoting ODA over GDP and CIM have been centered: the mean values for each variable have been subtracted from the value of each observation.
Centering the variables lowers correlation to more acceptable levels of 0.67 for ODA and 0.62 for Institutions. This is indeed an important justification for performing this study in the first place. Since collinearity has been an issue in much of the old research, documented lack thereof constitutes an interesting point of departure.
The new results are presented in table 2. These results are surprisingly robust to some modifications in the model and dataset, such as exclusion of potential outliers, different time periods and inclusion of various other control variables where the latter can be seen from the additional estimations 5 and 6 in the appendix. The aid variable is insignificant and the ODA x Institutions variable remains significant and negative.
Naturally, it is not the case that states with fewer checks and balances, higher corruption among officials and arbitrary private property legislation are generally better at achieving aid-fueled growth, but more probably, some type of endogeneity problem makes further analysis impossible, more of which will be addressed further down. As can be seen from the appendix the aid variable is more often than not negative and significant when the CIM-ODA interaction term is excluded or replaced. In light of these figures, clearly (6) makes no sense.
As predicted by Roodman (2007b) the results are quite fragile to changes in the key vari-ables. For example, meddling with the interaction-term sometimes produce very different results – as does excluding it altogether. However, keeping the corrected variable stabilizes the regression results and makes replication possible over different time periods, albeit with the unexpected sign.
By having a look at estimation 3 in the appendix it is striking how the expected results are produced when institutions are proxied by ethnic diversity instead of CIM. These results are not replicable to other time periods as is demonstrated in estimation 4. They are also sensitive to changes in the control variables. Quite possibly the interaction term, in this case is highly correlated with the other key variables and produces random results.
Hence, the results are rather conclusive with the findings in Rajan and Subramanian (2008), Roodman (2007a) and most other general conclusions outlined in the background and pre-vious studies sections. Notably, the testing cannot ascertain whether or not the effects of aid on growth is significantly different from zero. In light of (2), there are no signs that for-eign aid efforts have been successful in the process of financing enough investment to bringing about economic growth.
Further, the institutions variable itself is insignificant, which is what most of the literature reviewed in the background part would predict. Institutions have been found to have a positive effect on income levels, but is rather static over time and hence not a realistic ex-planatory factor for short-run growth rates, which are normally quite volatile.
This is easily grasped by recapitulating the figures in the descriptive statistics part above: even though institutional quality clearly improved over the past decades, the figure appear sluggish in comparison to the very volatile growth rates on a year-to-year basis in figure 1, while their correlation to income levels seem much clearer in figure 2. Also recall that Acemoglu et al. (2001) could explain contemporary income levels and state performances quite well with the help of centuries-old figures over colonial origins.
The control variables openness, past growth rates and inflation rates all exhibit signs of be-ing rather good predictors of short-term growth rates, which is quite well in line with intui-tion as well as with findings in previous research. Admittedly, it might be the case that panel data regressions are not the best available method to apply on the problem at hand and normal cross-sectional testing is also common in the literature, e.g. in Rajan and Subramanian (2008), wherein it is explicitly argued that cross-sectional tests are best suited for the problem.
Switching to cross-sectional testing over various time periods, such as row 2000-2006 in table 1 or 1995-1999 (estimation 9) in the appendix does not produce drastically different results except for the institutions-aid cross-term being insignificant. Further, openness, in-flation and population also fail to predict growth in this setting. Experimenting with time periods results in few changes, apart from the ODA over GDP variable sometimes emerg-ing as negative and significant – a common endogeneity problem often encountered in previous research3.
This latter result – that the cross-section cannot discern any systematic effects on growth from aid – is comparable to the findings in Rajan and Subramanian (2008) and is quite de-pressing reading, since changes in the Y variable from the X variables should register itself with such techniques. This undoubtedly strengthens the proposition in much of the previ-ous research that it is indeed hard to establish effects from foreign aid disbursements on growth levels.
In summation: at this point the author, not least supported by the results of the cross-sections, is in agreement with much of the previous research insomuch as any robust ef-fects on growth from aid are concerned. The obvious question for further investigation is why this is. The author proposes that the various reasons suggested in the previous re-search may be the divided in two sub-categories: reasons exogenous to the foreign aid is-sue, and reasons embedded in the system, henceforth referred to as endogenous factors.

Exogenous factors

A natural reason for this might be the difficulty involved in modeling growth over large sample sizes. Having a look at the goodness of fit (R-squared) values on any similar regres-sion offers some support for this hypothesis in that they are virtually always quite low, normally in the range from 0.25 to 0.40. It can of course be the case that some ingredient is missing before enough information on the effects of foreign aid are fully known.
Further, the assumption made in the theoretical framework may have been a bit rash. Since institutions do proxy related topics, notably corruption, they are not likely to be totally ir-relevant, but may on the other hand not be ideal either. As mentioned, ‘perfect institutions’ in the sense that CIM approaches 1 does not per se equal ‘perfect democracy’ or even ‘ab-sence of corruption’, it merely indicates confidence in the system, though some degree of totalitarianism or corruption may still be present. Perhaps institutions could be measured by checks and balances or malapportionment. This remedy would require a significant leap of faith though and would most probably bring similar problems with it – also bear in mind that two further concepts aimed at measuring related issues – settler mortality and ethnic diversity – failed to produce different results.
As argued above, multicollinearity has been a persistent problem in the previous research and has rather likely driven the results in many previous studies. After correcting the sam-ple this is probably not the case here, although the results are essentially the same as would have been obtained with uncentered variable as can be seen from estimations 1 and 2 in the appendix. It is nonetheless interesting to observe the wildly fluctuating signs of estimations 3 and 4 in the appendix, where institutions are measured by ethnic diversity.
A fact that does not strengthen the proposition that institutions are good for aid is the sta-bility of the results in their refusal to deliver positive and consistent results. Since the vari-able acts rather robust so long as the non-collinear setting is used it certainly casts a shadow of doubt on the theoretical background to the problem, although not refuting it altogether. As noted, it is still probable that some sample problems are present in table 2 and taken to-gether with the various results in the appendix, they are a pretty good illustration of the warnings in Roodman (2007b) referred to above.
In short, the problems with multicollinearity have likely been remedied, but some other forces persist in producing erroneous and sometimes absurd results.

Endogenous factors

The problem with this approach is obvious: since aid efforts have been well-documented for 50 years, then after trillions of dollars it should have been possible to see some results under a certain set of circumstances. Hence, another reason for the absence of results may be that aid in the broad sense is simply not effective.
Aid to extremely corrupt regimes is not likely to be effective in boosting growth since fights over resources and rent-seeking behaviors is likely to emerge – a phenomenon ex-plored e.g. in Djankov et al. (2008). But a growing body of evidence also suggest that aid in practice might not have the prerequisites to be effective in the first place due to crowding out effects and other problems intrinsic to the system (Roodman 2007; Rajan & Subrama-nian, 2005). Indeed, about a hundred peer-reviewed papers have been published on the aid effectiveness issue and if there is a consensus, it is that aid has not been effective (Dou-couliagos & Paldam, 2005).
Borrowing a term from Mosley (1987), this may be characterized as the micro-macro paradox standpoint; that regardless of successful execution of specific development projects, econ-omy-wide progress has not materialized. That is, regardless of institutions being perfect or not, aid may simply not be effective when considering its net effects – which is really the only sensible position.
A commonly voiced potential reason for this is supported by the conclusions in Rajan and Subramanian (2005) and is described as the tendency for aid inflows to trigger a real ex-change rate overvaluation that is not observed in the case of private remittances. This re-duces the competitiveness of the export sector and may thereby hamper growth, in despite of proper implementation in a country with a sound structure of formal and informal insti-tutions. This phenomenon is commonly referred to as the Dutch disease of foreign aid, but has unfortunately been largely ignored in the aid effectiveness literature up to this point (Doucouliagos & Paldam, 2005).
The endogenous type of critique is of course more fatal than exogenous explanations, sim-ply because it quite explicitly claims that the problems with foreign aid effectiveness may be beyond repair – no matter how benevolent recipient governments get or how well moni-tored disbursements are. Such insights have led to a situation where an increasing number of scholars make pledges to denounce foreign aid efforts altogether – the most famous of which is expressed in Subramanian (2007).
One might also suspect that poor people are more prone to holding a large share of their wealth in money. Bjørnskov (2009) identifies a related problem: foreign aid is likely to breed inflation through a Dutch disease effect, which will skew the income distribution and hence make the poor relatively worse off. By thinking about inflation as a tax on holding money, this means local elites will be relatively better off than the poor, even in the absence of corruption.

Extended analysis

Pondering the existence of such endogenous problems might offer some insight into the lack of significant results presented in this section and in the pile of previous research. Whether high quality institutions make aid allocation more efficient or not, the net effects might still be offset by the triggering of an exchange rate overvaluation and a subsequent weakening of the domestic manufacturing sector.
Bearing in mind that changes in the interaction term could produce substantially different results, it is not difficult to understand why David Roodman (2007a; 2007b) choose to title his critiques of the contemporary foreign aid research The Anarchy of Numbers and A Guide for the Perplexed. The main point made is that just about any desired result can be supported by regression results if the performer does not mind tweaking the estimation with uses of collinear cross-terms, triple cross-terms, squared cross terms, aid squared terms or a com-bination of these4.
This analysis has put effort into fairness and consistency in the choice of interaction term and has ensured it free from collinearity to the furthest extent possible. Still, the variables have not shown any signs of acting according to theory.
As has been covered thus far, foreign aid disbursements can have detrimental effects on completely different areas of society than the one targeted by the effort, which is likely a contributing reason to the micro-macro paradox defined earlier. Failure to pay attention to the macroeconomic net effects of a disbursement can have unfortunate effects for the recipi-ent country’s export sector and can misallocate a chunk of the skilled labor force from their productive work to something less desirable. Therefore, any study that does not deal with the big picture must be considered with caution.

Table of Contents
Master thesis in economics
Magisteruppsats inom nationalekonomi
1 Introduction
1.1 Purpose
1.2 Outline
1.3 Definitions
1.4 Background
2 Previous studies
3 Theory
4 Empirics and data
4.1 Descriptive statistics
4.2 Specification
4.3 Findings and discussion
5 Conclusions
5.1 Suggestions
References
GET THE COMPLETE PROJECT