The theoretical studies on the effect of trade barriers on economic growth have revealed mixed and vague results. Comparative advantage is seen as the main reason for countries to engage in trade. Countries tend to benefit from the specialisation of goods in which they have a comparative advantage. Early endogenous growth theories claim that less developed economies tend to converge towards advanced economies through trade and technological diffusion (Barro & Sala-i-Martin 1997; Grossman & Helpman 1991a; Romer 1990). Their model implies that countries more open to trade experience a faster growth rate.

Krugman (1990) lists the benefits of trade liberalisation for developing countries. He first opines that production patterns in low-income countries are time-consuming and skewed towards agriculture and manufacturing. Low-income countries have small markets and low per capita incomes. A liberalised trade policy therefore allows low cost producers to expand their output well above the demands of the domestic market. Secondly, an open trade policy allows countries to enjoy a constant return to scale over a wider range. Lastly, bureaucrats have better discretion over which industries to support or allocate limited foreign exchange in a regime of quantitative restriction.

Barro and Sala-i-Martin (1997) in their analysis maintained that trade liberalisation allows followers to converge towards leaders because copying is cheaper than innovation. As innovation starts to dwindle, the cost of imitation will rise and the followers’ growth rate will fall. Such observations are generally found in cross-country data on economic growth and display a form of conditional convergence. In the same manner, Chang, Kaltani and Loayza (2005) claimed that trade liberalisation leads to technological development and an efficient allocation of resources.

On the other hand, Grossman and Helpman (1991b) noted that the effect of international trade on a country depends on its capacity to acquire foreign technology and adapt it to the local environment. This is supported by Rodriguez and Rodrik (2001) who detected little evidence of trade policies’ impact on economic growth.

Some theorists argue that trade openness is prejudicial to economic growth when countries specialise in sectors in which development and research are not the core activities (Almeida & Fernandes 2008).

Empirical evidence on the positive effects of trade openness on growth is abundant (Chang et al. 2009; Dollar & Kraay 2004; Fetahi-Vehapi, Sadiku & Petkovski 2015; Guei, Mugano & Le Roux 2017; Sikwila, Ruvimbo & Mosikari 2014).

In contrast, Yanikkaya (2003) in a cross-country empirical investigation claimed that trade barriers and economic growth in developing countries are positively and significantly correlated. Sarkar (2008) used a time series data analysis and showed that for the majority of the countries observed over the period 1961–2002, there is no long-run relationship between openness and economic growth. Rigobon and Rodrik (2005) found that trade has a significant negative impact on income levels. Similarly, Moyo and Khobai (2018) investigated trade liberalisation for 11 Southern African Development Community (SADC) countries and found that trade liberalisation has a negative impact on growth. Ulasan (2015) also disagrees with the trade-led growth hypothesis.

Iyoha and Okim (2017) investigated the impact of trade on economic growth in the ECOWAS region. Using the pooled ordinary least square (OLS), the fixed effect model, the fandom effect model and the dynamic panel regression model, they found that exports were consistently positively related to growth. They also found that the four regression equations had high coefficients of determination and F-statistic.

Lloyd, Ogundipe and Ojeaga (2014) examined the respective impacts of export diversification and composition of GDP growth and GDP per capita on the 15 ECOWAS countries. They employed the panel least-square technique for the periods 1975–2009 and 1990–2007, respectively and found that export diversification and manufacturing value-added index have a positive and significant impact on per capita income growth.

So far, studies have looked at the effect of exports and export diversification on economic growth of the ECOWAS countries. Data were assumed to be stationary or partially stationary and to exhibit a long-run relationship during the periods of study. This article will investigate the long-run relationship between the variables and the effects of trade openness on GDP per capita in the ECOWAS region.

With the purpose of assessing the effects of trade openness on economic growth in the ECOWAS region, this section covers the data employed, the scope of the study and the model specification.

The variables used in this study include the following: GDP per capita growth rate, labour force, investment, financial development, exchange rate, external debt and trade openness. These variables are selected because these are important determinants of economic growth.

All variables were sourced from IMF (2018), World Bank (2018) and World Integrated Trade Solution (WITS 2018). The study covered the period 1990–2016. Firstly, the study used a time series regression model on each of the 15 ECOWAS member countries to assess whether there is a long-run relationship between the explanatory variables. It then explored the relationship between economic growth and trade openness in the 15 ECOWAS member countries using a PMG model.

The article hypothesises that trade openness generates economic growth in the ECOWAS region. Founded on the neoclassical growth model, initiated by Solow (1956), the approach used in this study is that of the augmented neoclassical growth model developed by Mankiw, Romer and Weil (1992). This model was chosen for two reasons. Firstly, it augments the slow growth model by including the accumulation of physical and human capital. Secondly, the goal is to examine the relationship between trade openness and growth; hence, trade openness and several other policy-related variables are included in the equation.

The empirical research comprises two steps. The first step aims to test the presence of a long-run relationship between the variables for each country in the ECOWAS region. The second step examines the stationarity of the variable using unit root tests after the panel regression model has been determined.

The study used the PMG model developed by Pesaran, Shin and Smith (1999). This technique is useful since it estimates a regression for each observation and then averages them across groups so that the short-run coefficient, the error term and the intercept are different across units but similar across groups. The advantages of this approach over traditional methods are well documented in the literature (Zahonogo 2016).

The estimators of the PMG method are less sensitive to outliers. They are also consistent and efficient when parameter homogeneity holds. Under this condition, the article addresses the issue of endogeneity by augmenting the PMG parameters with lags of regressors. This helps to reduce the bias of the estimators and to ensure that the regression residuals are not serially correlated.

The PMG model takes into account trade openness, heterogeneity of the coefficients and other controlled variables and can be specified for the ECOWAS countries during the period 1990–2016 as follows: Yit=β0i+β1i OPENNit+β2i LBit+β3i INVit+β4i FDit+  β5i ERit+β6i DEBTit+μi+∈i[Eqn 1] where Y_it = growth rate of GDP per capita at time t

OPENN_it = trade openness (measured by the ratio of exports and imports to GDP) at time t

LB_it = labour force at time t

INV_it = investment (proxied by gross capital formation) at time t

FD_it = financial development (proxied by domestic credit provided by financial institutions) at time t

ER_it = official exchange rate at time t

DEBT_it = external debt stock at time t

m_i = country-specific effect

∈_i = error term.

The ARDL dynamic specification for this relationship is expressed as follows: Yit=λYit−1+δ10i OPENNit+δ20i LBit+δ30i INVit+  δ40i FDit+δ50i ERit+δ60iDEBTit+δ11i OPENNit−1+  δ21iLBit−1+δ31iINVit−1+δ41iFDit−1+δ51iERit−1+  δ61iDEBTit−1+μi+∈it[Eqn 2]

The ARDL bounds testing approach to cointegration is based on the following error correction model: ΔYit=ϕi(Yit−1−β0i−β1iOPENNit−1−βiLBit−1−   β3iINVit−1−β4iFDit−1−β5iERit−1−β6iDEBTit−1)+   δi01ΔOPENNit+δi02ΔLBit−1+δi03ΔINVit−1+   δi04ΔFDit−1+δi05ΔERit−1+δi05ΔDEBTit−1+μi+∈it−1[Eqn 3] where β_oi = μ_i / (1 – λ_i)

β_1i = δ_10i + δ_11i / (1 – λ_i)

β_2i = δ_20i + δ_21i / (1 – λ_i)

β_3i = δ_30i + δ_31i / (1 – λ_i)

β_4i = δ_40i + δ_41i /(1 – λ_i)

ϕ_i = - (1 – λ_i).

Trade openness is measured by the ratio of imports and exports to GDP. It is expected to be positively correlated with GDP per capita.

Labour force comprises people aged 15 years and above who supply labour for the production of goods and services. It is important to note that the study uses the log of labour force in the model estimation. Labour force is expected to be positively correlated with GDP per capita. It is an important contributor to economic growth through domestic technology capability building and foreign technology spillover (Banerjee & Roy 2014; Pelinescu 2015). This variable is added to the model for two reasons: one is the size of the informal sector in ECOWAS. The second reason is that low level of education does not prevent farmers and the youth from actively participating in the economy.

In many empirical studies, investment has been used as a contribution to capital accumulation. In this study, it is measured as gross capital formation as a percentage of GDP. It is an indicator of basic economic infrastructure, such as machinery, roads and land improvement (World Bank 2018). Blanchard and Johnson (2014) also note that greater investments have a positive impact on economic growth. This coefficient, therefore, is expected to be positively correlated with GDP per capita.

Financial development is measured by the domestic credit offered by financial institutions as a percentage of GDP. The lack of a well-developed financial system and a dependence on primary products do not encourage sustainable growth in developing countries. Instead, the latter results in low real income and a tendency for borrowers to default. Hence, financial development is expected to be negatively correlated with GDP per capita.

Exchange rate refers to the price of South African currency (Rand) value against the US dollar. Hence, it measures the competitiveness of a currency. The exchange rate is expected to be negatively correlated with economic growth as suggested by the literature (Yan, Lin & Li 2016).

This article also captures the effects of debt on GDP per capita. Debt is the total external debt stock. The study uses the log of total external debt stock. Debt is expected to be inimical to economic growth.

The ARDL model is used to derive the long-run relationship in each of the 15 ECOWAS countries. The long-run relationship of the underlying variables was observed through the F-statistic or the Wald coefficient. The ARDL model used an OLS to evaluate the long-run relationship between trade openness and other controlled variables.

The ARDL bounds approach can be used irrespective of whether the underlying variables are I(0), I(1) or mutually cointegrated. This implies that the bound testing cointegration procedure does not require classification of the variables into I(0) and I(1) and is robust when there is a long-run relationship between the underlying variables. The representation of ARDL error correction becomes relatively more efficient when the F-statistic establishes a single long-run relationship between the underlying variables (Pesaran et al. 1999).

Once the long-run relationship had been established, the article used the PMG model. The latter is a cointegration technique that deals with variables that are integrated of different orders such as I(0) and I(1) (Pesaran et al. 1999). This is a better cointegration technique than that of Engle and Granger, and the Johansen test because they cannot be used in cases where variables are integrated of different orders (Nkoro & Uko 2016). Another advantage of the PMG is that unlike the Johansen technique it uses a reduced form equation.

The PMG model estimates a dynamic heterogeneous panel by looking at the long-run equilibrium relations between the underlying variables.

The null hypothesis of no cointegration is given as:

ϕ₁ = ϕ₂ = ϕ₃ = ϕ₄ = ϕ₅ = ϕ₆ = 0 against the alternative hypothesis ϕ₁ ≠ ϕ₂ ≠ ϕ₃ ≠ ϕ₄ ≠ ϕ₅ ≠ ϕ₆ ≠ 0.

Figure 1 shows the distribution of GDP per capita growth rate and trade openness (imports plus exports as a share of GDP) during the observed period. The variables tend to follow a similar trend. The graph shows that the two variables present a downward trend from 1990 to 1992. GDP per capita growth in ECOWAS countries reached a peak in 1997 when trade openness was also at a peak level. Trade openness exhibits a downward trend from 2007 to 2009, which could be attributed to the global financial crisis. The growth rate of GDP per capita also declined during this period.

The descriptive statistics and correlation matrix are shown in Table 1.

Table 1 shows that the average level of GDP per capita growth is 1.24%. Trade openness has an average of 67.56 and reached its maximum at 311.35. Debt, financial development, investment, labour force, and exchange rate have a respective mean value of 4.72, 35.31, 19.05, 14.74 and 715.08.

The correlation analysis was conducted using the Spearman’s rank-order test. The correlation matrix shows a positively significant relationship between investment and GDP per capita, which follows a priori expectation. However, a negative and statistically insignificant relationship between trade openness and GDP per capita growth is observed. Based on the correlation results reported in Table 1, there is no problem of multi-collinearity among the explanatory variables as only a small correlation between them is observed.

Several unit root tests were performed to test the order of integration of the series (see Table 2). The Phillips-Perron (PP) Fisher chi-square test, the Augmented Dickey-Fuller (ADF) Fisher chi-square test developed by Maddala and Wu (1999), Im, Pesaran and Shin test (2003) and Levin, Lin and Chu test (2002) were used. Using the PP Fisher chi-square methods, the results show that only labour force is stationary at the 10% level. All other methods showed a stationarity of 1% or 5% at either levels or at first difference for all the variables. This implies the possible existence of a long-run relationship between the variables.

The bounds test results are shown in Tables 3 and 4, which reveal that there is a cointegration between the variables at 1% level in most countries. However, we failed to reject the null hypothesis of no cointegration for Ghana, Guinea-Bissau, Mali, Senegal and Togo.

A diagnostic test is also conducted for each of the ECOWAS member countries using the ARDL models. The results are shown in Table 5. The 1% significance level is chosen for the analysis. The residual diagnostic tests show that of the 15 countries observed, only Benin and Liberia show serial correlation. No other country shows evidence of heteroscedasticity, serial correlation and residual non-normality. Model misspecification is only found in Ghana. Cointegration is found in most of the countries studied and the ARDL models pass the diagnostic tests. Hence, there is evidence of a long-run relationship between the variables. In this case, there is a long-run relationship between GDP per capita and trade openness. This implies that the PMG is an appropriate model for the analysis.

The PMG results are shown in Table 6. The a priori expectation is that openness and GDP per capita have a positive relationship. However, the results of the study show that trade openness and GDP per capita have a negative relationship in the long run. This finding is supported by Vlastou (2010), who, using a sample of 34 African countries, found that trade openness has a negative impact on economic growth. Meierrieks and Kraft (2009) also showed that trade openness and growth in ECOWAS countries (Senegal, Togo and Cote d’Ivoire) are not significant.

A possible explanation for the negative impact of trade openness on GDP per capita is that ECOWAS countries are not able to take full advantage of exports diversification, which is a necessary condition to support economic growth.

From Table 6, debt, exchange rate and financial development are negatively correlated with GDP per capita. This is in line with the a priori expectation. Labour force is positively correlated with GDP per capita, which also satisfies the expectation of this study.

Surprisingly, investment has a negative relationship on economic growth, which is against the a priori expectation. Investment includes construction of roads, railways, schools, hospitals and land improvement. The economic costs of poor road maintenance, rehabilitation projects and the degradation of the environment could be the reasons why investment has a negative effect on GDP per capita.

The error correction term is significant and negative, indicating that there is a stable long-run relationship between the variables. The coefficient suggests that 89% of the disequilibrium level of output in the short run is corrected in the long run. In the short run, trade openness has a negative relationship with GDP per capita; however, it is not statistically significant.

The heterogeneous PMG results are presented in Table 7. The impact of trade openness on economic growth is negative and significant in eight countries. This suggests that trade openness is detrimental to economic growth in the short run. Most of the adjustment coefficients are negative and significant. Only Guinea-Bissau shows a positive adjustment coefficient, which is an indication of model instability.