Export-output causality: the Irish case 1953-93.
Debate on the nature of the relationship between exports and economic growth is ongoing with various explanations propounded as to the importance of trade in economic performance. As countries open up to trade, international communication of ideas and technology becomes increasingly possible. This may have the effect of intensifying competition, increasing the incentive for both imitation and innovation, and accelerating the rate of technical progress that can lead to efficiency gains through more competitive cost structures and productivity improvement. Foreign exchange constraints may also be eased since increased exports provide a source of foreign exchange for countries that wish to purchase imports of final products or inputs that embody domestically unavailable technology. Under the scenario where increased exports lead to cost reductions and increased efficiency, the underlying causal direction is from export growth to output growth.
The potential benefits of export growth for economic development have been widely discussed [Keesing, 1967; Krueger, 1980; Bhagwati, 1988; Greenaway and Sapsford, 1994] and empirically tested for many less-developed countries [Balassa, 1978; Feder, 1982; Bahmani-Oskooee and Alse, 1993]. However, for more industrialized or developed economies, the potential benefits of export growth may be less important because the positive externalities enjoyed by less-developed countries are significantly higher than for developed countries whose infrastructural development is more advanced [Afxentiou and Serletis, 1991]. Benefits from increased competition are lessened since advanced countries are more competitive. Also, new technology will have less impact because, to retain competitiveness, continuous improvements in technology are required. Such reasoning may provide some explanation for the principal focus of recent export-output studies on less-developed countries.
An alternative causal explanation of the link between exports and output is manifest in Verdoorn's law which holds that output growth has a positive impact on productivity growth. Kaldor  attributed this relationship to factors that included economies of scale, learning curve effects, increased division of labor, and the creation of new processes and subsidiary industries. In this case, productivity growth in the industry sector is considered the principal determinant of output growth. Improved productivity and reductions in unit costs make it "easier to sell abroad" [Kaldor, 1967, p. 42], implying a causal relationship from output growth via productivity growth to export growth.
Empirical research on the causality issue has yielded results that mirror the contradictory causality theories. The main objective of this paper is to estimate if a causal relationship exists between Irish exports and output and which direction it takes. The evidence presented adds to the exports-output causality debate and provides a method that can be used to estimate the extent (if any exists) of causal relationships between exports and output for broader samples of countries. The results contribute to the ongoing debate on the direction of causality between exports and output and to the wider debate of the causes of the variation in growth rates between countries. The econometric techniques used are based on the statistical theory of cointegration and Granger causality tests and incorporate error-correction modeling.
The Irish Dimension
Ireland represents an interesting application for investigating the nature of causality, given its status as a small open economy (where exports in 1997 represented 76 percent of GDP(1)) and the path that Irish industry and trade policies have followed in recent decades. Unlike other larger economies, Ireland's industry and trade policies have largely been in tandem since the late 1950s. There are many reasons for this situation. Beginning in the mid 1950s, the Irish government realized that Irish industrialization had reached a natural limit, given the resources and size of the country. The publication of Economic Development [Department of Finance, 1958] was the first unified policy program which laid out the policies deemed necessary for economic growth. Further industrialization could only be ensured through targeting export markets. This entailed a reorientation of the Irish development strategy from a highly protectionist import substitution policy to an export-oriented trade policy with foreign direct investment playing a central role [O'Sullivan, 1993].
Relative to other industrialized countries, Irish industrialization was late in changing as the reorientation from primary products toward manufacturing became apparent only from the 1960s onward. As a result, industry and trade policies were inextricably linked as the focus switched from the traditional exports from the agriculture sector to exports from the industry sector. The extent of the reorientation is clear from the changing composition of Irish exports shown in Table 1. Coupled with the changing composition of Irish exports was the increasing importance of exports as a share of Irish output, increasing from 32 percent of Irish GDP in 1960 to 76 percent in 1997.
TABLE 1 Composition of Irish Exports Percentage for the Selected Years 1960-90 Sector 1960 1973 1979 1986 1990 Agriculture 30 10 5 3 2 Manufacturing 30 44 52 65 69 Other 40 46 43 32 29 Total 100 100 100 100 100 Source: Central Statistics Office, Ireland, [various issues].
In an Irish context, any discussion of the export-output relationship cannot ignore the role of foreign-owned industry. In 1991, overseas firms' contribution to the net output of Irish manufacturing industry output was 70 percent. The presence of overseas industry is evident across all industry sectors but is most readily identified with the high-technology industries of pharmaceuticals, office and data processing, electrical engineering, and instrumental engineering. Gross value-added of the high-technology sector (as a percentage of total manufacturing output) increased from 20 percent in 1980 to 39 percent by 1990 when it represented 58 percent of the net output of all overseas industry [National Economic and Social Council, 1993]. The high-technology sector is highly export-oriented and enjoys higher export shares and levels of productivity - computed as net output per head - compared to indigenous Irish industry.
On closer inspection, however, aggregate figures on strong economic performance reveal disturbing factors that may influence the relationship between Irish exports and output. For example, the lack of significant linkages between overseas and domestic firms means that much of the growth in output and productivity observed in foreign-owned industry is not a result of improvements in Irish technology, marketing, or other skills but, instead, stems from the expertise of associated firms located elsewhere. Research and development investment in Irish subsidiaries is significantly lower than in other comparable economies and the proportion of skilled workers compares unfavorably with other more-developed countries [Hitchens and Birnie, 1994]. This implies that the contribution of the foreign-owned sector, in terms of spillovers into the indigenous sector, is less than potentially possible, generally limiting the performance of the manufacturing sector and reducing the potential impact of such an export-driven sector on the Irish economy.
Additionally, the financial conduct of foreign-owned industry (evident in high and growing amounts of repatriated profits, dividends, and royalty payments) is indicative of a level of indifference to improving the learning process in Irish subsidiaries [O'Sullivan, 1995]. Furthermore, the strong productivity performance of the foreign-owned sector (relative to Irish indigenous industries and to advanced industrialized countries) may reflect the success of the low rate of Ireland's corporation tax in attracting foreign direct investment into Ireland. Transfer-pricing by multinational companies (whereby artificially low input prices on imports and high output prices on exports are recorded) provides the opportunity to take advantage of the incentive to pay as much tax as possible on profits in Ireland. The extent to which transfer-pricing is practiced means that output measures are artificially inflated and overstate the extent of productivity improvements.
Because of the extent of transfer-pricing and outflows of interest on foreign debt, it has been suggested that GNP, rather than GDP, is a more appropriate measure of Irish output [Kennedy, 1990]. However, in assessing the impact of exports on domestically produced output, as opposed to national output, GDP is the measure of output preferred for this study. An estimate of the effects of transfer-pricing on output can be derived by adjusting output data to take account of profit outflows, royalties, and dividends. Such data are available from 1984 onward [National Economic and Social Council, 1992]. Interestingly, the Irish Economy Expenditures Survey (carried out by the Irish Industrial Development Authority in 1983 [O'Malley and Scott, 1994]) indicated that 86 percent of profits, royalties, and dividends arose from the manufacturing sectors of pharmaceuticals, office and data processing machinery, electrical engineering, instrument engineering, and soft drink concentrates. These have been identified as sectors where "foreign firms ... export virtually all of their output" [O'Malley and Scott, 1994, p. 151]. Thus, any artificial inflation of output value resulting from transfer-pricing is matched by similar artificial inflation of export value for these sectors. Consequently, GDP is the output measure selected for use here.
Evidence of the Export-Output Relationship
Many studies of the links between exports and growth confirm a statistical relationship between export growth and output growth [Michaely, 1977; Krueger, 1978; Balassa, 1978; Feder, 1982]. The export growth correlation appeared to be particularly pronounced in the case of industrialized countries. Michaely  and Tyler  considered that a minimum level of development was required to observe a significant relationship between export growth and output growth. However, based on cross-country correlations between exports and output (or productivity), the empirical approach yields no information for the causality question.
Studies that used the Granger or Sims procedures to investigate causality do not provide conclusive support for the export growth relationship. For instance, Chow  used the Sims procedure to examine the causal relationship between export growth and output growth for manufacturing industries and found bidirectional causality for Hong Kong, Israel, Singapore, Taiwan, and Brazil, unidirectional causality for Mexico, and no causality for Argentina. In comparison, Jung and Marshall , who used the technique of the Granger causality tests, found support for the export-led growth hypothesis for just four of a sample of 37 developing countries.(2,3) A statistically significant relationship from output growth to export growth was found for three countries. Six countries exhibited evidence of an export-reducing growth relationship, while a further three supported a growth-reducing exports relationship.
More recent research has indicated that cointegration tests are required to investigate causality since the presence of cointegration between the variables under examination invalidates the conclusions reached using the Granger test of causality. It is only possible to infer a causal long-run relationship between nonstationary time series when the variables concerned are cointegrated. If cointegration analysis is omitted, causality tests present evidence of simultaneous correlations rather than causal relations between variables. Afxentiou and Serletis  tested real exports and GNP data (1950-85) for cointegration in their examination of export-output causality for 16 developed countries including Ireland. They concluded that exports and GNP did not cointegrate except in the cases of Norway, Iceland, and the Netherlands. They found evidence of bidirectional causality for the U.S. and GNP to export causality for Norway, Canada, and Japan with no other significant causality results.
In a study of 65 countries (1965-85), Pomponio  found little evidence of export-output causality. However, when he incorporated investment into his analysis, tri-causal relationships were found to be significant in some cases. Seventeen countries displayed significant positive relationships from manufactured exports and investment to manufactured output, while three displayed a negative relationship. Fourteen countries exhibited significant positive causality from exports and output to investment, and one country had a negative result. A further 15 countries showed significant positive causal effects from investment and manufactured output to exports, while four displayed a negative result. Significance in all cases was associated with the fastest growing and intermediate growing economies in the sample.
Granger  has explained that, for cointegrated time series, standard Granger or Sims tests may provide invalid causal information due to the omission of the error-correction term from the tests. If the error-correction term is excluded from causality tests when the series are cointegrated, then no causation may be detected when it exists. The use of error-correction modeling provides an additional channel through which causality in the Granger sense may be assessed.
Research that incorporates an error-correction term yields interesting results. Marin  found that exports of manufactured goods Granger-cause productivity (manufactured output per employee) for the developed economies of the United Kingdom, Germany, the U.S., and Japan. These findings contrast with those of Afxentiou and Serletis , who used aggregate data, in each case apart from the U.S. In a study using Portuguese data, Oxley  found evidence that output growth caused export growth. Using similar techniques of cointegration and error-correction modeling, Bahmani-Oskooee and Alse  found support for a bidirectional relationship between export growth and output growth in the case of eight less-developed countries.
This study assesses if Irish export and output (GDP) data are cointegrated. Also, given evidence of cointegration, error-correction modeling is employed to investigate if causality is observed. The approach adopted here is the Johansen procedure which is an appropriate framework for the analysis of causality allowing for simultaneous investigation of both cointegration and Granger causality.
Data, Integration, and Cointegration
To provide valid empirical evidence on causality, it is meaningful to address the time series properties of the variables because any empirical analysis from which valid inferences can be made must ensure that the series considered are of the same order of integration [Phillips, 1986; Ohanian, 1988]. This avoids the potential problem of spurious relationships and incorrect inferences.
As a preliminary step to cointegration analysis, it is, therefore, necessary to test the stationarity of the variables. The data used are the natural logarithm of the real GDP (LGDP) level and the natural logarithm of the real exports (LX) level from 1953 to 1993. Annual data were used as quarterly GDP data are unavailable for Ireland. Data were compiled from the National Income and Expenditure Accounts published by Ireland's Central Statistics Office [various issues]. The time period selection for the study represents an attempt to not only use a sample period of relevance to the Irish experience, but to also use a sample size large enough to ensure reliable test results. Sample size alone is the reason that the initial observation is for 1953 despite misgivings that the export-output relationship may be investigated more appropriately from the late 1950s onward. Both Dickey-Fuller and augmented Dickey-Fuller (ADF) tests were used as stationarity tests. The ADF test is based on the following regression equations:
[Delta][LX.sub.t] = [[Psi].sub.1] + [[Theta].sub.1][LX.sub.t-1] + [summation of] [[Gamma].sub.i][Delta][LX.sub.t-i] where i = 1 to P + [[Epsilon].sub.t], (1)
[Delta][LGDP.sub.t] = [[Psi].sub.2] + [[Theta].sub.2][LGDP.sub.t-1] + [summation of] [[Beta].sub.i][Delta][LGDP.sub.t-i] where i = 1 to P, + [[Kappa].sub.t] (2)
where: [Delta] is the first difference operator; [Delta]LX.sub.t] = [LX.sub.t] - [LX.sub.t-1]; [Delta][LGDP.sub.t] = [LGDP.sub.t] - [LDGP.sub.t-1]; [[Psi].sub.1], [[Psi].sub.2], [[Theta].sub.1], [[Theta].sub.2], [[Gamma].sub.i], and [[Beta].sub.i] are the coefficients; and [[Epsilon].sub.t] [[Kappa].sub.t] are the error terms. The null hypothesis being tested is that [LX.sub.t] and [LGDP.sub.t] have unit roots, for example, that [[Theta].sub.1] = [[Theta].sub.2] = 0. The alternative is that the variables are integrated of order zero I(0). The hypothesis is rejected when [[Theta].sub.1] and [[Theta].sub.2] are negative and significantly different from 0, for example, when the t-statistics are greater (in absolute values) than the MacKinnon  critical values.
Real exports and output are cointegrated if some linear combination of the two series exists which is stationary, I(0), even though the variables themselves may be nonstationary. For example, if variables [LX.sub.t] and [LGDP.sub.t] are both I(1) and the sequence, [Z.sub.t], defined as [Z.sub.t] = [LX.sub.t] - [Alpha][LGDP.sub.t], is I(0), this implies that [LX.sub.t] and [LGDP.sub.t] are cointegrated. The implication of cointegration is that, over the long run, an equilibrium relationship exists between the variables making the difference between them stationary. In a bivariate case, the cointegrating parameter, [Alpha], is unique. Granger  explained that if two variables are I(1) and cointegrated, then Granger causality must exist in at least one direction as one variable aids prediction of the other. According to the Granger representation theorem [Engle and Granger, 1987], when a vector on n I(1) time series [X.sub.t] are cointegrated with a cointegrating vector, a, an error-correction representation, exists:
A(L)[Delta][X.sub.t] = - [Gamma][Alpha][X.sub.t-1] + [Beta](L)[[Epsilon].sub.t], (3)
where: A (L) is a matrix polynomial in the lag operator L with A(0) = [I.sub.n]; [Gamma] is a (n x 1) non-null vector of constants; [Beta](L) is a scalar polynomial in L; and [[Epsilon].sub.t] is a vector of white-noise errors. In the short run, any deviation from the long-run equilibrium ([Alpha][prime]X = 0) will impact on changes in [X.sub.t] and lead to movement back to equilibrium. If some element of the vector X is being driven by the equilibrium error (so that the relevant element of [Gamma] is nonzero), then there is such a feedback response. However, if the nth element of y is zero, the nth element responds only to short-term shocks to the stochastic environment [Agenor and Taylor, 1993].
The tests for cointegration reported here adopt the procedures profiled in Johansen and Juselius . Here, the procedure is applied to the bivariate systems with [LX.sub.t] and [LGDP.sub.t] as the dependent variables in the vector autoregressive (VAR) representation involving up to p lags of the variables in [X.sub.t]:
[X.sub.t] = [[Pi].sub.1] [X.sub.t-1] + [[Pi].sub.2][X.sub.t-2] + ...... + [[Pi].sub.p][X.sub.t-p] + [[Epsilon].sub.t], (4)
where: [X.sub.t] is a (2x1) vector of I(1) variables; [[Pi].sub.i] are (2x2) matrices of parameters; and [[Epsilon].sub.t] [similar to] I N(0, [Sigma]). The long-run equilibrium for the system is given by:
[[Pi].sup.*] X = 0, (5)
where [[Pi].sup.*] is the matrix of long-run coefficients given by:
I - [[Pi].sub.1] - [[Pi].sub.2] - ... - [[Pi].sub.p] = [[Pi].sup.*]. (6)
The rank (r) of [[Pi].sup.*] determines the number of cointegrating vectors that exist between exports and output. Cointegration exists in the bivariate case if r is equal to 1. If the matrix [Pi] is the product of two (2x1) matrices:
[Pi] = [Gamma] [Alpha][prime]. (7)
Then, if exports and output are cointegrated, the unique cointegrating vector is a and the coefficients in 7 represent the speed of adjustment of the system to disequilibrium.
The hypothesis to be tested is that there is, at most, one cointegrating vector between exports and output. Johansen suggests two tests to determine the number of cointegrating vectors. They are the [[Lambda].sub.max] test and the trace test. Johansen's maximal eigenvalue and trace statistics for testing [H.sub.o]: r [less than or equal to] 1 against [H.sub.a] r = 0 are given, respectively, by:
[[Lambda].sub.max] - n ln (1 - [Lambda]i), (8)
Trace test = -n[Sigma] ln(1 - [Lambda]i), (9)
where [Lambda]i are the squared canonical correlations between the [X.sub.t-p] and [Delta][X.sub.t] series corrected for the effect of the lagged differences of the X process. The relevant critical values for the trace test and [[Lambda].sub.max] test can be found in Osterwald-Lenum .
Causality and Error-Correction Models
A time series, X, is said to Granger-cause another series, Y, if the inclusion of lagged values of X improve the forecast of Y (evident in a smaller mean square error) on the forecast derived from the use of lags of Y alone. Here, exports are said to cause output, with respect to given information including LX and LGDP, if the prediction of output is improved by using past values of exports, given that relevant information is totally contained in the present and past values of these variables. The rationale is similar to test for causality from output to exports.
The approach to test for a short-term causality relationship between exports and output is to run two-way Granger causality tests. Given economic theory on causality between exports and output, there is no a priori reason to exclude any one of the causal directions in this case.
The standard Granger causality test analyzes bivariate weakly stationary stochastic processes. If the original series are nonstationary, they must be transformed into stationary variables. This is carried out by differencing the variables until they are stationary. If [Delta][LX.sub.t] and [Delta][LGDP.sub.t] denote the transformed stationary values of exports and output, the formal VAR system is given by:
[Delta][LX.sub.t] = [summation of] [[Phi].sub.1i] [Delta][LX.sub.t-1] where i = 1 to Z + [summation of] [[Phi].sub.2i] [Delta][LGDP.sub.t-i] where i = 1 to T + [[Xi].sub.1t], (10)
[Delta][LGDP.sub.t] = [summation of] [[Omega].sub.1i] [Delta][LGDP.sub.t-i] where i = 1 to R + [summation of] [[Omega].sub.2i] [Delta][LX.sub.t-i] where i = 1 to H + [[Xi].sub.2t]. (11)
If [Delta][LX.sub.t] Granger-causes [Delta][LGDP.sub.t] but not vice versa, all coefficients, [[Phi].sub.2i], should be statistically insignificant from 0 and there should be at least one coefficient, [[Omega].sub.1i], that is statistically significantly different from 0. Since the data are transformed to be stationary variables, it is possible that the causality structure may be affected [Geweke, 1984]. By differencing the data, any information about the long-run relationship between the trend components of the original series is removed so that the Granger causality tests describe only short-run relationships between exports and output.
It is possible, however, that additional long-run relationships exist between the variables. Standard Granger causality tests, augmented with error-correction terms and derived from the long-run cointegrating relationships, are used to assess the long-term effects. Such tests are undertaken on I(0) variables to ensure that valid inferences may be made from the tests [Engle and Granger, 1987]. The augmented Granger causality test is structured as:
[Delta][LX.sub.t] = [[Phi].sub.0] + [summation of] [[Phi].sub.1i] [Delta] [LX.sub.t-i] where i = 1 to Z + [summation of] [[Phi].sub.2i] [Delta][LGDP.sub.t-i] where i = 1 to T + [Delta][[Mu].sub.t-1] + [[Xi].sub.1t], (12)
[Delta][LGDP.sub.t] = [[Omega].sub.0] + [summation of] [[Omega].sub.1i] [Delta][LGDP.sub.t-i] where i = 1 to R + [summation of] [Q.sub.2i] [Delta][LX.sub.t-i] where i = 1 to H + [Lambda][[Eta].sub.t - 1] + [[Xi].sub.2t], (13)
where [[Mu].sub.t - 1] and [[Eta].sub.t - 1] are the error-correction terms that are found from the long-run cointegrating regressions:
[LX.sub.t] = [Delta] + [Phi][LGDP.sub.t] + [[Mu].sub.t], (14)
[LGDP.sub.t] = [Iota] + [Lambda][LX.sub.t] + [[Eta].sub.i]. (15)
Including the error-correction terms in the equations offers an extra channel through which causality may be observed. The error-correction coefficients [Delta] and [Lambda] are expected to capture the adjustments of [Delta][LX.sub.t] and [Delta][LGDP.sub.t] to their long-run equilibrium, while [Delta][LX.sub.t - i] and [Delta][LGDP.sub.t - i] are expected to capture the short-run dynamics of the models [Jones and Joulfaian, 1991]. Focusing on (12), LGDP is said to Granger-cause LX not only if the [[Phi].sub.2i] is are jointly significant but also if [Delta] is significant. Interestingly, in contrast to the standard Granger test, the error-correction model allows for the finding that LGDP Granger-causes LX once the coefficient on the error-correction term is significant and even if the [[Phi].sub.2i] is are not jointly significant.
Table 2 presents the ADF test results (with two lags of the dependent variable) for the log levels and first differences of the logs of real GDP and real exports. based on the ADF results, the null hypothesis of a unit root cannot be rejected for LX and LGDP. By comparison, even at the 1 percent level, the first differences of LX and LGDP accept the hypothesis of a unit root implying that LX and LGDP are I(1) while [Delta]X and [Delta]LGDP are I(0).
TABLE 2 Unit Root Test for Irish Exports and GDP Variable Without Trend With Trend LX 1.14 -1.22 LGDP -1.29 -2.30 [Delta]LX -4.74(*) -4.72(*) [Delta]LGDP -4.03(*) -4.13(**) Note: * denotes significance at the 1 percent level and ** denotes significant at the 5 percent level. The critical values of the ADF t-statistic for the null hypothesis of a unit root are -2.94 at the 5 percent significance level and -3.62 at the 1 percent level when testing without trend. The critical values of the ADF t-statistic for the null hypothesis of a unit root are -3.54 at the 5 percent significance level and -4.22 at the 1 percent level when testing with trend.
In using bivariate VARs of exports and output, the lag length choice in the VAR may affect inferences made from the causality tests. To determine the optimal lag length of the VAR (in log-levels), two criteria are considered. They are the Akaike information criterion [Harvey, 1990] and Akaike's final prediction error [Hsiao, 1979]. Both criteria suggest that a lag length of 2 is optimal and it is used in the following causal estimation. The results of applying the Johansen procedure to test for cointegration are presented in Table 3. The test results indicate that there is one cointegrating relationship between Irish exports and output.
[TABULAR DATA FOR TABLE 3 OMITTED]
A Wald test (F-test) was carried out to determine if the direction of causality between exports and output could be established. Results of the causality tests are presented in Table 4 where the computed F-statistics are reported. based on the estimates, causality runs from exports to output and provides empirical support for the export-led growth hypothesis for the Irish case.
TABLE 4 Short-Run Causality Tests: Granger Causality Test Results F-statistic [Delta]LX [implies] [Delta]LGDP 4.35(*) [Delta]LGDP [implies] [Delta]LEX 1.88 Note: * denotes significance at the 5 percent level. At the 5 percent significance level, the critical F-statistic (2, 32) is 3.30. The F-statistic was computed as: [F.sup.*] = ([SSE.sub.c] - [SSE.sub.u])/m*]/[SSE.sub.u]/(T-k) [similar to] [F.sub.m*,T-k)], where: [SSE.sub.c], [SSE.sub.u] = residual sum of squares of the constrained and unconstrained models, respectively; T = total number of observations; m* = number of restrictions; and k = number of parameters estimated in unconstrained regression.
The question of whether long-run causality is evident between Irish exports and output was addressed by considering the significance of the coefficients on each error-correction term in the Granger equations. On the basis of results in Table 5, there is some statistical support for the hypothesis that exports Granger-cause output growth because the coefficient [Lambda] on the error-correction term [[Eta].sub.t - 1] is significant at a 7.75 percent significance level. The coefficient also displays an appropriate negative sign which is indicative of the existence of a valid equilibrium relationship between the variables in cointegrating regression (15).
This paper investigates the relationship between exports and output using the techniques of cointegration and causality. The findings of the cointegration analysis indicate that there are long-run common trends between Irish exports and output. The hypothesis of export-led growth cannot be rejected since lagged values of exports are jointly significant in explaining changes in GDP. This indicates a short-run Granger causal relationship from exports to output. The significance of the error-correction term implies a long-run causal relationship in the same direction - from export growth to output growth. These findings for Ireland contrast with the findings of Afxentiou and Serletis . Reasons for the differences may be attributed to the different time periods considered and to the different methods used. For the data considered here, Ireland's outward-oriented policies appear to have contributed to a strong output performance. The findings are interesting in light of recent research in this area for industrialized countries. A number of issues are also raised that require further examination.
TABLE 5 Long-Run Causality Tests: Error-Correction Analysis Coefficient T-statistic ECM [implies] [Delta]LGDP -0.140 -1.800 (0.08)(*) ECM [implies] [Delta]LX 0.002 0.024 (0.98) Note: * denotes significance at the 10 percent level. The figures in parentheses denote p-values.
Recent research indicates that a number of developing countries display statistically significant support for the export-led growth hypothesis. Marin's  research focuses on manufacturing industries and reveals a systematic relationship between exports and output for four developed countries. The research here is based on aggregate output and export data. However, as a next step, it appears valid to assess the exports-output relationship at more disaggregate levels since this would be indicative of whether the relationship is observed at levels of further disaggregation. Such studies could be particularly informative, given the statistical significance of results not only for Ireland but also (in the context of Marin's  research) for other developed countries. If positive externalities from trade benefits are lower for developed countries, then the extent of an export-output relationship may be lost in aggregate analysis and may be detected only at sectoral or even industrial levels.
The suitability of the Johansen methodology for examining cointegration and causality suggests that inferences made in previous studies from estimation using ordinary least squares may not be valid. If these methods can highlight for Ireland - one developed country - that aggregate exports are an important factor in economic growth, then further examination could be carried out for other developed countries to consider the question of causality. The benefits of an outward-oriented trade policy in terms of overall economic performance could, thus, be assessed.
The fact that the export-output debate could yield different causality results for different countries is of critical interest since a correct policy direction for one country would be ineffective or inefficient for another country. Interest arises in addressing a number of countries separately because of the unique nature of individual economies. This can, therefore, aid the interpretation of econometric analysis. Results for Ireland would not be indicative of trends for other developed countries. Only further research can verify the extent of support for or opposition to the export-led growth hypothesis.
An earlier version of this paper was presented at the Forty-First International Atlantic Economic Conference in Paris, France, March 13-18, 1996. The author would like to thank Liam Gallagher and Van Newby for helpful comments.
1. This figure is based on current price data taken from the National Income and Expenditure Accounts published by Ireland's Central Statistics Office [various issues].
2. Interestingly, aggregate export and output data for Taiwan, Brazil, and Mexico are included in Jung and Marshall's  analysis and provide no evidence of a causal relationship despite Chow's  findings at the manufacturing industries level. Both studies of Jung and Marshall and Chow find no causal relationship for Argentina.
3. Darrat  and Hsiao  find a similar lack of support for the export-led growth hypothesis.
Agenor, P. R.; Taylor, M. P. "The Causality Between Official and Parallel Exchange Rates in Developing Countries," Applied Financial Economics, 3, 3, September 1993, pp. 255-66.
Axfentiou, P. C.; Serletis, A. "Exports and GNP Causality in the Industrial Countries: 1950-1985," Kyklos, 44, 2, 1991, pp. 167-79.
Bahmani-Oskooee, M.; Alse, J. "Export Growth and Economic Growth: An Application of Cointegration and Error-Correction Modeling," The Journal of Developing Areas, 27, 4, July 1993, pp. 535-42.
Balassa, B. "Exports and Economic Growth: Further Evidence, "Journal of Development Economics, 5, 2, June 1978, pp. 181-9.
Bhagwati, J. "Export-Promoting Trade Strategy: Issues and Evidence," World Bank Research Observer, 3, 1, January 1988, pp. 27-57.
Central Statistics Office. National Income and Expenditure Accounts, Dublin, Ireland: Central Statistics Office, various issues.
Chow, P. C. Y. "Causality between Export Growth and Industrial Development: Empirical Evidence from the NICs," Journal of Development Economics, 26, 1, June 1987, pp. 55-63.
Darrat, A. F. "Trade and Development: The Asian Experience," Cato Journal, 6, 2, Fall 1986, pp. 695-99.
Department of Finance, Economic Development, Dublin, Ireland: Stationery Office, 1958.
Engle, R. F.; Granger, C. W. J. "Co-Integration and Error-Correction: Representation, Estimation, and Testing," Econometrica, 55, 2, March 1987, pp. 251-76.
Feder, G. "On Exports and Economic Growth," Journal of Development Economics, 12, 1/2, February/April 1982, pp. 59-73.
Geweke, J. "Inference and Causality in Economic Time Series Models," in Z. Griliches; M. D. Intriligator, eds., Handbook of Econometrics, 2, Amsterdam, The Netherlands: North-Holland, 1984.
Granger, C. W. "Some Recent Developments in a Concept of Causality," Journal of Econometrics, 39, 1/2, September/October 1988, pp. 199-211.
Greenaway, D.; Sapsford, D. "Exports, Growth, and Liberalization: An Evaluation," Journal of Policy Modeling, 16, 2, April 1994, pp. 165-86.
Harvey, A. C. The Econometric Analysis of Time Series, 2nd ed., London, United Kingdom: P. Allan, 1990.
Hitchens, D.; Birnie, J. The Competitiveness of Industry in Ireland, Avebury, Ireland: Aldershot, 1994.
Hsiao, M. W. "Tests of Causality and Exogeneity between Export Growth and Economic Growth," Journal of Economic Development, 12, 2, December 1987, pp. 143-59.
-----. "Causality Tests in Econometrics," Journal of Economic Dynamics and Control, 1, 1979, pp. 321-46.
Jones, J. D.; Joulfaian, D. "Federal Government Expenditures and Revenues in the Early Years of the American Republic: Evidence from 1792 to 1860, "Journal of Macroeconomics, 13, 1, Winter 1991, pp. 133-55.
Johansen, S.; Juselius, K. "Some Structural Hypotheses in a Multivariate Cointegration Analysis of the Purchasing Power Parity and the Uncovered Interest Parity for the U.K.," unpublished paper, Institute of Mathematical Statistics, University of Copenhagen, 1990.
Jung, W. S.; Marshall, P. J. "Exports, Growth and Causality in Developing Countries," Journal of Development Economics, 18, 1, May/June 1985, pp. 1-12.
Kaldor, N. Strategic Factors in Economic Development, New York State School of Industrial and Labor Relations, Ithaca, NY: Cornell University, 1967.
Keesing, D. B. "Outward-Looking Policies and Economic Development," Economic Journal, 77, June 1967, pp. 303-20.
Kennedy, K. A. "The Context of Economic Development," The Economic and Social Research Institute, paper to Conference on the Development of Industrial Society in Ireland, Nuffield College, Oxford, United Kingdom, December 7-9, 1990.
Krueger, A. O. "Trade Policy as an Input to Development," American Economic Review, 70, 2, 1980, pp. 282-7.
-----. Foreign Trade Regimes and Economic Development: Liberalization Attempts and Consequences, Cambridge, MA: Ballinger, 1978.
MacKinnon, J. "Critical Values for Co-integration Tests," in R. F. Engle; C. W. Granger, eds., Long-run Economic Relationships, Oxford, United Kingdom: Oxford University Press, 1991.
Marin, D. "Is the Export-led Growth Hypothesis Valid for Industrialized Countries?," The Review of Economics and Statistics, LXXIV, 4, November 1992, pp. 678-88.
Michaely, M. "Exports and Growth: An Empirical Investigation," Journal of Development Economics, 4, 1, March 1977, pp. 149-53.
National Economic and Social Council, "A Strategy for Competitiveness, Growth and Unemployment," NESC, Dublin, Ireland: NESC, 1993.
-----. "The Association between Economic Growth and Employment Growth in Ireland," NESC, Dublin, Ireland: NESC, 1992.
Ohanian, L. E. "The Spurious Effects of Unit Roots on Vector Autoregression: A Monte Carlo Study," Journal of Econometrics, 39, 3, November 1988, pp. 251-66.
Osterwald-Lenum, M. "A Note with Quantiles of the Asymptotic Distribution of the Maximum Likelihood Cointegration Rank Test Statistics," Oxford Bulletin of Economics and Statistics, 54, 3, August 1992, pp. 461-72.
O'Malley, E.; Scott, S. "Profit Outflows Revisited," in S. Cantillon; J. Curtis; J. FitzGerald, eds., Economic Perspectives for the Medium Term, Dublin, Ireland: The Economic and Social Research Institute, 1994, pp. 149-56.
O'Sullivan, P. "An Assessment of Ireland's Export-Led Growth Strategy via Foreign Direct Investment: 1960-1980," Weltwirtschafiliches Archiv, 129, 1, 1993, pp. 139-58.
O'Sullivan, M. "Manufacturing and Global Competition," in J. W. O'Hagan, ed., The Economy of Ireland: Policy and Performance of Small European Country, New York, NY: Gill and Macmillan, 1995.
Oxley, L. "Cointegration, Causality and Export-led Growth in Portugal, 1865-1985," Economics Letters, 43, 2, 1993, pp. 163-6.
Phillips, P. C. B. "Understanding Spurious Regressions in Econometrics, "Journal of Econometrics, 33, 3, 1986, pp. 311-40.
Pomponio, X. Z. "A Causality Analysis of Growth and Export Performance," Atlantic Economic Journal, 24, 2, June 1996, pp. 168-76.
Tyler, W. "Growth and Export Expansion in Developing Countries: Some Empirical Evidence," Journal of Development Economics, 9, 1, August 1981, pp. 337-49.
|Printer friendly Cite/link Email Feedback|
|Publication:||Atlantic Economic Journal|
|Date:||Jun 1, 1998|
|Previous Article:||Small firms and economic growth in Europe.|
|Next Article:||Sticky prices as a coordination success.|