Corruption, EU Aid Inflows and Economic Growth in Ghana: Cointegration and Causality Analysis

In this paper, we examine the causal relationship between aid inflows and economic growth for Ghana during the period from 1970-2013, taking into account structural breaks. To better reflect causality, corruption and trade are included as control variables. To test for causality in the face of cointegration, a vector error correction model (VECM) is used in place of a vector autoregressive (VAR) model. This approach is complemented with Toda and Yamamoto’s method to indicate the causal direction. Our estimation results suggest GDP growth has one cointegrating vector relationship with corruption, EU aid inflows and trade in both the short and long runs. There is a long-run unidirectional causal relationship from EU aid inflows to GDP growth and a short-run unidirectional causal relationship from trade to GDP growth. Corruption (which is a governance issue) was ineffective in inducing GDP growth. The error correction terms are the source of causation in the long run. The results indeed confirm the popular conjecture that corruption in Ghana is endemic and stifles development. Therefore, the decision by the government to launch a national anti-corruption campaign in 2011, though long overdue, was justifiable. We urge all stakeholders to work together to deepen good governance to promote sustainable growth and serve as inducement for continued aid inflows from multilateral donors to sustain efforts at achieving the national development thrust of poverty reduction and sustainable development in Ghana.


Introduction
Corruption as a social issue is widespread and continues to dominate many discussions in academic and policy circles due to its devastating effects on devel-opment. The subject has also been revisited in recent years following the massive looting reported by the European Union anti-corruption watchdog. The agency reveals that corruption alone costs the EU over EUR 120 billion per year, which is just less than the EU's annual budget (European Commission, 2014). A similar report from the World Bank estimates that every year, between USD $20 and $40 billion is lost from developing countries due to corruption and bribery, but it emphasized that corruption and bribery also impact developed economies through globalization (UN, 2013). The scourge on developed economies is a result of the commitments they make in the form of Overseas Development Assistance (ODA), which is often captured as aid, grants and loans to promote development in disadvantaged economies. In Ghana, incidental and systematic corruption is perceived to be high and is considered to be responsible for the slow pace of development (Lamptey, 2013). The causes of corruption are in manifold, but in the case of developing economies such as Ghana, whose budgetary demands depend on the fluidity of financial pledges from development partners, its sources are increasingly important to investigate. Ghana currently ranks 98 th out of 144 countries on a global percentile measure of irregular payments of bribes in public contracts based on a World Economic Forum (WEF) executive opinion, which is a trend that suggests weak institutional structure (WEF, 2014).
Foreign aid comes in different forms for different purposes. Currently, the world's poorer countries' activities are funded with aid from foreign governments and international organizations. Foreign aid may include billion dollar reconstruction projects in wartorn countries, microfinance programs for impoverished women, international research to find more productive crops and less polluting energy sources, expansion of primary education in rural regions, financing for health budgets, support for economic reforms, debt relief and civil society development programs in Africa. The number of participants involved in providing foreign aid has increased in recent years (see Lancaster, 2007).  (Bhasin & Obeng, 2007).
There was a steady increase in the inflow of aid from Meanwhile, the subject of aid effectiveness has also been examined, especially in SSA. In an empirical paper presented by Gomanee, Girma and Morrissey (2002) at a conference on the effectiveness of aid and investment for economic growth on 25 economies in SSA, it was noted that for each 1 per cent in aid received as a share of GNI, there is one-quarter of a percentage point increase in growth among the sample.
The authors therefore concluded that the state of poorer African countries should not be attributed to aid ineffectiveness, yet the study failed to propose an alternate reason accounting for these countries conditions. According to Maizels and Nissanke (1984) (Knack, 2013).
With the emergence of the "New Public Administration" (NPA), there has been increasing focus on strengthening institutional frameworks to reduce corruption. Nevertheless, this goal has been viewed in policy circles with mixed reactions. For instance, Alesina and Weder (2002) assert that corrupt governments do not tend to receive less aid than clean governments. Conversely, Dollar and Levin (2004) observe that over time, aid has been directed more toward countries with sound institutions and policies.
Thus, in recent times, good governance has become a condition for the disbursement of development assistance to less developed nations (Fayissa & Nsiah, 2010). The debate on conditionality was further intensified following a World Bank publication in 1998 on the assessment of aid policies to poorer countries with institutional challenges. The publication's policy decision was to adopt a selectivity approach because the effectiveness of aid can be increased if more is allocated to countries with good policies. The argument is that "aid does not work" in the sense that the amount of aid alone has no effect on growth, but aid makes a positive contribution to growth in those countries with good policies (Burnside & Dollar, 2000). Additionally, policy reform conditionality does not work because donors have less power to influence policies and institutions in the recipients' economies, let alone to bypass the government in implementing expenditures (Collier & Dollar, 2004).
Hence, more aid should go to nations that are already implementing good policies to boost the poverty alleviation process. However, opponents have challenged selective aid allocation (Dalgaard, Hansen, & Tarp, 2004;Hansen & Tarp, 2001). Their contention is that aid has contributed to poverty reduction and has improved the welfare of the poor independent of the recipients' policies (Gomanee et al., 2002;Mosley, Hudson, & Verschoor, 2004).
The ineffectiveness of conditionality is also contested on the ground that the specific reforms advocated by donors are hardly ever implemented fully within the relatively short time period of the associated aid program (Koeberle et al., 2005;Lensink & Morrissey, 2000;Mosley et al., 2004). In an attempt to contextualize this problem in Ghana, Lloyd, Morrissey and Osei (2001) investigate the relationship between aid inflows, trade and growth and contend that exports, aid and public investment are all positively related to long-run growth. However, in the pre-1983 era, they find that exports and public investment had a negative impact on short-run growth, with no significant impact reported on aid. The authors assert that the results for the post-reform era (after 1983) show a significant improvement in the statistical significance of these variables, which they attribute to institutional reforms that enhanced the governmental machinery. However, we find a gap in the type of proxy used to measure governance (or institutional inputs); hence, we argue that such a linkage cannot be precisely made unless the deficiency in variable measurement is reconciled.
Our argument is borne out of the fact that inducing efficiency in the governmental machinery has always been the central focus of administrative reforms in Ghana. Consequently, making such a concluding remark without reference to elements of governance in their model makes their latter finding untenable.
Moreover, the goals of aid are achieved when the existing institutions are proactive. This paper addresses this flaw by incorporating corruption as a component of governance.
From the findings above, it is clear that the aidgrowth nexus is mediated by an avalanche of factors embedded in the quality of the existing institutions, trade, imports, exports and public investment. Thus, in this paper, we explore this connection using two of the identified factors (corruption and trade) in Ghana. Regardless of these inflows, growth in Ghana has not been as dramatic as expected. Figure Ofori-Sarpong, 1986) that affected total output, but aid inflows then increased as multilateral donors such as the IMF and the World Bank were consulted for assistance. Nevertheless, this aid came with conditions. Key structural reforms were required to access donor funds. The introduction of programs such as the ERP and the SAP became the common approach to help Ghana's ailing economy recover at that time.

Research Hypotheses
Governance: Governance is the exercise of economic, political and administrative authority to manage a country's affairs at all levels (United Nation Develop-ment Programme [UNDP], 1997). However, the concept has been defined in different ways, which highlights how subjective it can be. The term is sometimes synonymous with corruption, which is the abuse of entrusted power for private gain (Hardoon & Heinrich, 2013;Kaufmann, Kraay, & Mastruzzi, 2006).
Corruption is considered to be inimical to development, although this relationship has received mixed reactions from both the grease-in-the-wheel and sand-in-the-wheel perspectives. Thus, the model relationship between governance (corruption) and aid inflows and for that matter, economic growth, has received mixed reactions (see Burnside & Dollar, 2000;Dietz, Neumayer, & De Soysa, 2007;Fayissa & Nsiah, 2010;Gyimah-Brempong, 2002;Lensink & Morrissey, 2000;Próchniak, 2013;Svensson, 2000). Therefore, we hypothesize that good governance (reduction in corruption) is associated with EU aid inflows, trade and economic growth in Ghana.
Aid Inflow: Aid inflow is the transfer of capital for the benefit of the recipient country or its popula- Corruption, EU Aid Inflows and Economic Growth in Ghana: Cointegration and Causality Analysis tion (see Lancaster, 2007). EU aid comes in different forms for different purposes (i.e., economic, military, or even emergency humanitarian assistance). Evidence on selective aid allocation as a sine qua non for good governance and economic expansion is widespread but not straightforward (see Easterly 2007;Fayissa and Nsiah, 2010;Hout 2007a;2007b;Kargbo, 2012;Knack, 2013;Nunnenkamp & Thiele, 2006;Ohler et al., 2012). However, in this study, we envision a strong association between EU aid inflows and the three regressors (corruption, trade and economic growth) in Ghana.

GDP Per Capita Income: GDP per capita is an
indicator of a country's standard of living (Cypher & Dietz, 2009

Data Description
Data for the analysis are obtained from reputable orga-  Table 1 for data summary).

Model Specification
Y is a random walk with an intercept: (2) t Y is a random walk with an intercept and a time trend: ( ) In addition, one can use the Phillips-Perron (PP) approach to detect the unit root. Intuitively, the PP test is the same as the ADF, except the PP test uses a non-parametric statistical method to handle serial correlation in the error term and does not include the lagged differences in the model. We describe the PP model as follows: t Y is a random walk and assumes the following form: ( ) Y is a random walk with an intercept and a time trend: In each of the cases outlined above, the null hypothesis δ= 0 implies that there is a unit root and the time series is non-stationary. The alternative hypothesis δ<0 implies that the time series is stationary. In the case where the null hypothesis is rejected, it presupposes that t Y is a stationary time series at I (0). Otherwise, sequential differences are taken until the null hypothesis is rejected. Thus, to test for unit roots against the alternative of a one-time structural break, Zivot and Andrews propose three approaches. However, the third approach is considered to be superior, as the loss in power is substantially low (see Sen 2003;Waheed et al., 2006). The test equation is adopted and outlined below: where t DU is a dummy variable for a mean shift occurring at each possible break date (TB), while t DT is a corresponding trend shift variable. The null hypothesis in this model is that 0 α = and an alternate hypothesis is that 0 α < . The former implies that the series contains a unit root with drift that excludes a structural break, whiles the latter implies that the series is a trend-stationary process with a one-time break occurring at an unknown point in time. This approach is able to fix all possible points as potential time breaks, with subsequent estimations through regression to determine the break points (Shahbaz et al., 2014).
After the stationarity tests, the cointegration method is linked to the vector error correction model (VECM). We describe the vector error correction model (VECM) having included a dummy variable for a break in the series below: where ∆ denotes the first difference order, for example, Although determining the exact order of cointegration is necessary, it might not be sufficient to establish the causal relationship among the variables of interest.
As a consequence, there is a need to use the traditional Granger causality approach to unearth this possible relationship. This approach is the most common way to test for a causal relationship between two variables and thus involves estimating a simple vector auto regression (VAR) equation, as shown below: where the disturbances µ 1t and µ 2t are assumed to be uncorrelated. The two equations above, (9) and (10), posit that variable X is decided by lagged variable Y and X, except that the dependent variables are interchanged in each case. Granger causality means that the lagged Y significantly influences X in equation (9) and vice-versa in equation (10) To resolve these shortcomings, Toda and Yamamoto (1995) present an alternative approach that accounts for the described limitations. Among other things, this test can be used irrespective of whether t Y and t X are cointegrated of the order I(0), I(1) or I(2) or whether they are non-cointegrated. The name of the method is the Toda and Yamamoto (T-Y) augmented Granger causality test, and it is based on the following equations: where d is the maximal order of integration of the variables in the system, h and k are the optimal lag lengths of t Y and t X and are error terms that are assumed to be white noise with zero mean, a constant variance and no autocorrelation. We are required to determine the lag order of integration, which by default occurs in the model, and to construct a VAR in their levels with a total of ( ) k d + lags.

Structural Break Test
We begin the analysis with a validation test using Chow's (1960)
equilibrium relationships, an appropriate lag length should be selected to facilitate the estimation of the long-run equilibrium relationship.

Appropriate Lag Length Selection
There are many ways to choose the optimal lag length in statistics, but the most commonly used methods are the Akaike information criterion (AIC) and the Schwarz information criterion (SIC). We use these methods to select the appropriate lag length for the model. The estimation results indicate that LR, FPE, AIC, SIC and HQ show significant results at 1, 2 and 4 lag length periods.
This result is straightforward and makes our work quite easy. The numbers with asterisks are the smallest values in each of the criteria. Before selecting the lag length, two issues must be addressed. One must understand that a lag length that is too short in VAR may not capture the dynamic behavior of the variables. Conversely, it is also argued that a lag length that is too long may distort the data and lead to a decrease in the explanatory power. Guided by these principles, the optimal lag length selected for these four variables is based on the SIC (Schwarz information criterion), which indicates a lag length of 1 period (see Table 5).

Figure 3
Breakpoints for the series GDP per capita, EU aid inflows, corruption and trade. All series except corruption are significant at the 5 per cent level. The vertical line shows the break date on the x-axis.    -4.93, and 4.58 at 1%, 5% and 10% levels of significance, respectively. **Denotes statistical significance at the 5% level. K denotes the lag order in parentheses.
Corruption, EU Aid Inflows and Economic Growth in Ghana: Cointegration and Causality Analysis

Relationships
The empirical analysis in this paper is subject to a linear relationship; hence, less emphasis is paid to whether the variables have a time trend. We reached this conclusion due to the existence of one cointegrating relationship among the variables (see Table 6). Table 6 shows the trace and max-eigen statistics.
The trace statistic of 97.892 is greater than the critical value of 68.819. This result implies that the null hypothesis 0 r = can be rejected, while the alternate hypothesis 0 r > accepted. The max-eigen value confirms the trace test results. The max-eigen value statistic of 56.027 is larger than the critical value of 33.877. Thus, the null hypothesis 0 r = is rejected, and the alternate hypothesis 1 r = accepted at a 5 per cent significant level. The long-and short-run dynamics are estimated using the VECM.

Long-term Cointegration Analysis
The cointegration test is conducted for Ghana's economic growth using GDP as the dependent variable lagged on selected independent variables. Several tests for causality are performed here: (1) long-run causality -the significance of the error-correction terms is determined by a t-test; (2) short run causality -the joint significance of the coefficients of the lagged terms of each independent variable are determined by Wald Chi-square tests; and (3) Toda and Yamamoto causality tests -the joint significance of the four sources of causation is determined.
We begin by evaluating the robustness of the VECM against the normality residual test of Jarque-Bera, the ARCH test of serial correlation and the heteroskedasticity test. Using the histogram normality test, the Jarque-Bera is 14.69 with a p-value of 0.00065, which indicates that we reject the null hypothesis of the normally distributed residual at a 5 per cent confidence level. This result suggests that the residual is not normally distributed. However, the other two tests passed the robustness checks. The ARCH test of heteroskedasticity indicates that the p-value of the observed Rsquared is 0.4872, which is greater than the 5 per cent confidence level. Thus, we accept the null hypothesis that there is no ARCH in this model. This result is desirable, as the model does not have an ARCH effect. Using the Breusch-Godfrey serial correlation LM test, the p-value of the observed R-squared is 0.7103, which is greater than the 5 per cent confidence level.
Thus, we accept the null hypothesis that there is no serial correlation. On the basis of these two tests, the model is acceptable, as the residuals are almost Gaussian white noise. From the test result, we find the p-value of the chi square to be less than 5 per cent (0.0429), which implies that the null hypothesis can be rejected. We interpret this result to mean that corruption, EU aid inflows and trade jointly cause GDP growth in the short run.
However, according to the Granger theorem, when variables are cointegrated, there must be an error correction (EC) that describes the short-run adjustments of the cointegrated variables as they move toward their long-run equilibrium positions (see Table 7). Corruption, EU Aid Inflows and Economic Growth in Ghana: Cointegration and Causality Analysis and 0.328 percentage points, respectively, to return to equilibrium. However, the error correction term of EU aid inflows is positive. This result implies that when EU aid inflow to Ghana is too low in equilibrium, it will begin increasing in the following year by 0.06 percentage points to correct the equilibrium error. In general, the model confirms both long-and short-run causation but is unable to determine the direction of causality. To determine the direction of causality, Toda and Yamamoto's test model in equations (11) and (12) is used in place of the traditional Granger test (see Table 8).

Co-Integrating Equation D(Y) D(COR) D(AID) D(TRD) D(DUMMY)
The F-statistics of the modified Wald test are calculated. In Table 8, the T-Y causality test results show that there is a unidirectional causality from EU aid inflows to economic growth in Ghana. A similar direction is evident between trade openness and economic growth. EU aid inflows and trade openness supplement domestic incomes and thus lead to economic growth in the short and long run. Nevertheless, the slow pace of Ghana's development is inferred from the insignificance of corruption, as aid and trade alone may not necessarily be sufficient to lead to growth that addresses the needs of all, but institutional inputs such as good governance may play a key facilitating role in achieving sustainable growth. This result is consistent with related studies in the literature (see Gomanee et al., 2002;Hansen & Tarp, 2001;Kilby & Dreher, 2010;Lloyd et al., 2001;Maizels & Nissanke,1984).

Policy Implications and Conclusion
This paper examines the causal relationship between EU aid inflows and GDP in Ghana while controlling for corruption (governance) and trade openness. The causal relationship between GDP, corruption, trade openness and EU aid inflows is based on pre-stated hypotheses. The statistical inference deduced from the Johansen model after testing for multivariate cointegration between GDP and the regressors indicates that there is one cointegrating vector relationship. The dynamics of the variables in the short run indicate that the source of causality runs through aid inflows, trade and the dummy variable to GDP. There is a long-run unidirectional causal relationship from EU aid inflows to GDP growth. Conversely, corruption remained insignificant and ineffective to power growth, but it shares a relatively stronger correlation with EU aid inflows. An essential implication of this result is that aid inflows and trade alone may not necessarily be sufficient to lead to growth that addresses the needs of all, but institutional inputs such as good governance may play a facilitating role in achieving sustainable growth. This analogy confirms the increasing attempts by the government of Ghana to meet the selectivity criteria espoused by most multilateral donors through reforms, which are often used as conditions for aid. However the quality and pace of reform has been below expectations; hence, development has been sluggish.
The long-run unidirectional causality from EU aid inflows to economic growth is confirmed by the negative coefficient and significance of the error correction term. This result supports the growth-led theses on income convergence in the literature (Cypher & Dietz, 2009;Forson et al., 2013;Kargbo, 2012;Mankiw et al., 1992). Therefore, the efforts of the government and policy makers should be directed at meeting selectivity criteria to serve as an inducement for the continued inflow of aid to sustain efforts to meet the MDGs and other developmental priorities. The short-run unidirectional causality from trade to growth is also consistent with the existing literature (see Federici & Montalbano, 2010;Haddad et al., 2012;Stensnes, 2006;Ulaşan, 2012).
The results in this paper also hold significant implications for the conjecture that corruption is endemic in Ghana and is the cause of the slow pace of development. Consequentially, the decision of government to launch the National Anti-Corruption Action Plan (NACAP) in 2011, though long overdue, is rational based on the evidence presented in this paper. However, such an attempt will only work if the various stakeholders (MMDAs, parastatals etc.) show much-needed commitment to eliminate the menace of corruption from its roots. Incidental and systemic corruption is high in Ghana, and due to its nature, it often goes unnoticed or treated with impunity. The scourge of petty corruption on national development is thus underestimated in most cases.
Nevertheless, the activities of NACAP should be directed toward fighting incidental corruption in the MMDAs, customs and police services, as these are the core breeding points of petty corruption. In addition, despite the laudability of the initiative, there are some further concerns that need clarification. For instance, Corruption, EU Aid Inflows and Economic Growth in Ghana: Cointegration and Causality Analysis to what extent will this agency be independent so that it can tackle corruption without fear or favors? How different will its mandate be from existing bodies? Are there mechanisms in place to check conflicts of interest and duplications of functions? We recommend further qualitative research to explore and present an in-depth account to these questions. In addition, relevant draconian measures should be reintroduced and applied where necessary to reduce the level of impunity and at the same time give more meaning to upholding the rule of law and accountability in Ghana.
A central limitation of this paper is the approach used to determine the presence of a single structural break. It is possible that the series may contain two structural breaks at a time. Therefore, considering only one break when there are indeed possibilities for more than one may lead to a loss of power of the test. Future research can extend this analysis by simultaneously considering two breaks in the series.