Integration of current account imbalances in the OECD

Integration of current account imbalances in the OECD

Economic Modelling 38 (2014) 288–295 Contents lists available at ScienceDirect Economic Modelling journal homepage: www.elsevier.com/locate/ecmod I...

305KB Sizes 2 Downloads 10 Views

Economic Modelling 38 (2014) 288–295

Contents lists available at ScienceDirect

Economic Modelling journal homepage: www.elsevier.com/locate/ecmod

Integration of current account imbalances in the OECD Seema Narayan ⁎ School of Economics, Finance and Marketing, Royal Melbourne Institute of Technology University, Melbourne, Victoria, Australia

a r t i c l e

i n f o

Article history: Accepted 17 January 2014 Available online 13 February 2014 JEL classification: F15 F32 F41

a b s t r a c t The Glick and Rogoff (1995) hypothesis suggests that common or global shocks do not influence current accounts of countries which are symmetric. This is tested for 37 pairs of current account imbalances out of 17 OECD countries. Using time series data that spans the pre-Global crisis period but including several sub-samples recognizing the advent of the Euro, this study shows that for nine pairs of countries common shocks do matter for current account imbalances. Results obtained from the Granger causality test do not support the presence of spillovers in some of these pairs. Therefore, this study documents some empirical evidence which departs from the Glick and Rogoff proposal. © 2014 Elsevier B.V. All rights reserved.

Keywords: Current account shocks Correlation Granger causality

1. Introduction 1.1. Background and literature One of the key predictions that Glick and Rogoff (1995) derived using a two-country intertemporal model of the current account was that if countries of the world are symmetric, then worldwide or global shocks should not have any impact on the current account.1 In particular, within their equilibrium current account model, Glick and Rogoff (1995) showed that both worldwide and country-specific shocks lead to fluctuations in investment, but only country-specific investment is related to the deterioration of the current account. Several studies have empirically investigated the GR prediction by incorporating the key elements of the GR model (see, inter alia, Glick and Rogoff, 1995; Gregory and Head, 1999; Iscan, 2000; Nason and Rogers, 2002). For instance, Glick and Rogoff (1995) found that the global components of productivity shocks have limited impact on the current accounts of the G7 countries. Gregory and Head (1999), on the other hand, separated the common and country-specific fluctuations in the G7 countries' investment, total factor productivity, and the current account for the period 1970:1 to 1991:2, using dynamic factor analysis and Kalman filter approach. ⁎ Tel.: +61 3 99255890; fax: +61 3 99255624. E-mail address: [email protected] 1 Glick and Rogoff (1995) defined symmetric countries as having same preferences, technology, and initial capital stock. In their model, they also assumed that the initial net foreign asset position is zero. For symmetric countries, world shocks do not matter because as all nations react in the same way to world shocks, the reaction of each country's permanent income is the same, and as a result current account remains unchanged (Nason and Rogers, 2002). 0264-9993/$ – see front matter © 2014 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.econmod.2014.01.019

Common fluctuations were found to have positive and significant impact on fluctuations in both productivity and investment in the G7 countries. These were also found to be strongly correlated with fluctuations in the US investment and productivity but, their impact on the current account was found to be limited. Only the current accounts of the US and France were significantly related with the common factor. Nason and Rogers (2002) incorporated elements of the GR model to construct the dynamic relationship between investment and the current account within a SVAR framework. In particular to account for the fact that worldwide shocks do not matter to the current account, they identified long-run fluctuations in investment as permanent change in the level of the worldwide shock and imposed the restriction that the current account does to respond to the permanent changes in investment in the long-run. Using Canadian data for the post-1975 period (following the oil price shocks of the 1970s), they found that common shocks do not matter for the movements in current account under their restriction. Iscan (2000) extended the analytical model of GR to an economy with traded and non-traded goods. This allowed him to distinguish between traded and non-traded global (and country-specific) shocks. He showed that global traded productivity shocks do not have any significant impact on the current account but influences investment significantly. He also found using G7 data that the global non-traded component of the productivity shock is small, and that it has no statistically significant impact on the current account or investment. This literature is relevant to the discussion of integration of current account as these studies show theoretical and/or empirical evidence that common (also, known as global or worldwide) shocks do not matter for the movement in the current account. The objective of this study is twofold. First, it tests whether shocks to current account are

S. Narayan / Economic Modelling 38 (2014) 288–295

common to pairs of countries. Second, where shocks are common, the study examines whether or not there is a unidirectional or two-way spillover effects in pairwise relationships. 1.2. Contributions to the literature Shocks to current account, which have multiple sources, such as oil price, economic slowdowns, and primary commodity price uncertainties, are common occurrences. This study does not distinguish the source of shocks to current account; rather, it applies a much broader definition of current account shocks—that is aggregate shocks to current accounts. Such shocks are extracted from an autoregressive model of the current account. This allows for direct testing of the GR prediction that current account shocks of two similar countries are uncorrelated. Here, weak or insignificant correlation between current account shocks of any pair of countries is sufficient to suggest that common current account shocks are irrelevant determinants of the current account. A strong correlation would suggest one of the two things: one, that there is a presence of spillover effects; or two, that common current account shocks in pairwise settings are relevant determinants of current account imbalances because the two countries are asymmetric—as per the GR theory. Spillover effects from one country to the other are important causes of economic integration and these effects are also as relevant with financial integration. As a result, finding evidence of correlation between pairs of current account shocks is one thing but whether they are actually common shocks or spillover effects is another thing. Therefore, for

289

those pairs of shocks that are correlated, it becomes important to examine whether Granger causes each other and the direction of Granger causality. This tells us whether current account shocks of country i Granger causes current account shocks of country j or vice versa, or simply that spillover results from both countries. In other words, this study will investigate whether shocks are bidirectional or unidirectional. The outcome here will evince, whether there is indeed a presence of spillover of shocks and if there is, which of the two countries current account shock leads the bilateral relationship. Moreover, with structural economic changes an important question is: have the correlations changed overtime? To address this issue, we include several subsamples of data. The various study periods for each pair of countries examined are noted in Table 1. Notice that for the full-sample period of the data set, the country-specific sample sizes vary depending on the availability of consistent data. To make comparisons easier, a common sample, which spans the period 1995:Q1–2006:3, is included for all pairs. Moreover, the Euro-zone member country pairs are examined within the pre-Euro and Euro periods. Related to these subsample analyses are various studies that show changes in the magnitude of economic integration over time since the inception of the Euro in 1999. For instance, Furceri and Karras (2007) found evidence of business cycle synchronization for many countries in the Eurozone since the inception of the European Union. Similarly, Camacho et al. (2006) found evidence of business cycle synchronization for European Union member countries and recent new members from the Central and Eastern Europe. In the current study, the common sample and Euro period samples reveal evidence of greater economic integration. Greater

Table 1 A summary of pair-wise data sample.

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37.

Pairs

Full sample

Common sample

Pre-1995 Period

Pre-Euro period

Euro period

Australia–Japan Australia–US Austria–Germany Belgium–Germany Belgium–France Canada–US Denmark–Germany Denmark–Sweden Finland–Germany Finland–Sweden France–Germany Italy–Germany Japan–US Korea–US Korea–Japan Portugal–Germany Portugal–Spain Portugal–France Portugal–UK Spain–France Spain–Germany Spain–UK Sweden–Germany Turkey⁎–Germany UK–US UK–Germany Austria–US Belgium–US Denmark–US Finland–US France–US Germany–US Italy–US Portugal–US Spain–US Sweden–US Turkey⁎–US

1980:1–2006:3 1974:3–2006:2 1991:1–2006:3 1995:1–2006:3 1995:1–2006:3 1961:1–2006:3 1991:1–2006:3 1993:1–2006:3 1993:1–2006:3 1991:1–2006:3 1993:1–2006:3 1990:1–2006:3 1980:1–2006:3 1995:1–2006:3 1995:1–2006:3 1995:1–2006:3 1995:1–2006:3 1995:1–2006:3 1995:1–2006:3 1995:1–2006:3 1995:1–2006:3 1995:1–2006:3 1995:1–2006:3 1991:1–2006:3 1960:1–2006:3 1991:1–2006:3 1988:1–2006:3 1995:1–2006:3 1990:1–2006:3 1975:1–2006:3 1993:1–2006:3 1991:1–2006:3 1980:1–2006:3 1995:1–2006:3 1995:1–2006:3 1993:1–2006:3 1987:1–2006:3

√ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √

√ √ √ NA NA √ √ √ √ √ √ √ √ NA NA NA NA NA NA NA NA NA NA √ √ √ √ NA √ √ √ √ √ NA NA √ √

NA NA √ √ √ NA √ √ √ √ √ √ NA NA NA √ √ √ NA √ √ √ √ NA NA NA √ √ √ √ √ √ √ √ √ √ √

NA NA √ √ √ NA √ √ √ √ √ √ NA NA NA √ √ √ NA √ √ √ √ NA NA NA √ √ √ √ √ √ √ √ √ √ √

Notes: Common sample includes the period 1995:1 to 2006:3; The Euro period begins at 1999:1; The Pre Euro period includes the beginning period of the full sample and ends 1998:4.

290

S. Narayan / Economic Modelling 38 (2014) 288–295

economic integration is consistent with an important prediction of GR that countries need to be symmetric. However, greater economic integration comes at the cost of greater spillover effects between countries. And, in pairwise relationship, if these spillover effects are prominent, than the correlation between countries could become stronger. Hence, a priori, it is unclear whether the correlation should be stronger or weaker during periods of relatively greater economic integration. This is investigated in detail. To foreshadow the key results, the study generally finds evidence in favor of the Glick and Rogoff (1995) prediction in the pairwise groupings. However, nine pairs of current account shocks show strong correlation, especially in the period 1995–2006 and/or the Euro period. The Granger causality test shows limited evidence of unidirectional Granger causality between the current account innovations, which is an indication that spillover effects are not responsible for the strong correlation. Of the nine pairs that show strong case of correlation, only three (Austria–Germany, Spain–Portugal, and Australia–US) show signs of one way Granger causality. For other highly correlated pairs, in particular those which are made up of EU member countries (Portugal– Germany, and Spain–France), show no signs of significant spillover effects from one country to the other during the Euro period (1999– 2006). 2. Empirical analysis 2.1. Data Current account imbalances are calculated as gross national income (GNI) less the sum of investment, consumption, and government spending. These are defined as a ratio of GDP so that they are measured in real terms. The data are sourced from the OECD national accounts (SourceOECD) and from the International Monetary Fund's database— the International Financial Statistics. The data is quarterly and the individual sample sizes vary depending on the availability of consistent data. This study avoids using data for the post 2006 period in order to prevent the spurious effects of the global financial crisis on our correlation results. Indeed future work, when sufficient post crisis time series data becomes available can replicate our analysis and compare the

results from the post-crisis with our pre-crisis period results. Since there is insufficient post-crisis time series data for meaningful analysis, we leave this for future research to address. This study is based on 17 OECD countries. From these 17 nations, a total of 37 pairs are constructed. The choice of at least the first 26 pairs is dependent on the trade relations between countries. Table 2 presents each of the 17 OECD countries (which we refer to as home countries or country i) and their key trading partner from the group of 16 OECD countries (denoted as country j) using 2004 bilateral trade share data. The trade shares are calculated on the basis of total exports and imports of goods and services. Notice that only those trading partners that contribute 10% or more to total trade of the home country are captured. While this pairing of 17 OECD countries with their major trading partners is to account for the important bilateral relationship between countries, it is also part of an attempt to understand the link between trade-related spillover effects and the correlation between shocks. In terms of the trade patterns, in 2004, all countries, except Germany, had at least one trading partner from the group of 17 industrialized countries with a trade share of 10% or more. In fact, the US contributed 76% to the total trade of goods and services of Canada in 2004. This represents the largest trade share within this group. Austria's trade share with Germany is ranked second, with Germany contributing some 40% to the total trade of Austria. In third placing is the Portugal–Spain pair, Spain contributed 30% to Portugal's total trade. Japan–US is ranked fourth, with the US contributed 21% to Japan's total trade. Meanwhile, Australia has two trading partners (Japan and the US) from the G17 group contributing more than ten percent to their total trade; and so does Belgium (Germany and France), Denmark (Germany and Sweden), Finland (Sweden and Germany), and Korea (US and Japan). Portugal has three other trading partners— Germany France and UK, while Spain has three trading partners– France, Germany and the UK–from the same group of industrialized countries in 2004. Moreover, given that the US is the world's largest nation, it is an important source of global shocks. In fact, Gregory and Head (1999) find that global fluctuations were highly correlated with fluctuations in US productivity and investment. As a result, all 11 nations which

Table 2 OECD trading partners. (Contributing 10% or more to total trade in goods and services of the home country). Home countries

Trading partner (1)

Trade in 2004

Trading partner (2)

(%) Australia Austria Belgium Canada Denmark Finland France Germany+ Italy Japan Korea Portugal Spain Sweden Turkey⁎ UK US

Japan Germany Germany US Germany Sweden Germany NA Germany US US Spain France Germany Germany US Canada

13 40 18 76 17 13 16 NA 16 21 17 30 16 13 13 16 17

Trade in 2004

Trading partner (3)

(%) US – France – Sweden Germany – – – – Japan Germany Germany – – Germany –

11 – 15 – 12 12 – – – – 14 15 14 – – 11 –

Trade in 2004

Trading partner (4)

(%) – – – – – – – – – – – France UK – – – –

– – – – – – – – – – – 13 12 – – – –

Trade in 2004 (%)

– – – – – – – – – – – UK – – – – –

– – – – – – – – – – – 10 – – – – –

Source: IMF, 2006a, 2006b. Direction of trade. Notes: This table shows the key trading partners contributing 10% or more to total trade of individual OECD country. The domestic country and the trading partners belong to the group of 17 OECD countries. Notice that we were able to find at least one trading partner for almost all countries + Germany's trade with the other OECD countries was less than 10%. ⁎ For Turkey only international trade in goods are included as services was not available.

S. Narayan / Economic Modelling 38 (2014) 288–295

were not previously paired with the US are paired up here. For these 11 pairs, given that the US is not an important trading partner, it is expected that the GR prediction on global shocks to dominate. Finally, as mentioned earlier, the data covers the full-sample; a common sample (1995:1–2006:3); the pre-1995 period; the Euro period (1999:1–2006:3); and the pre-Euro period. 2.2. Cross-correlation of the residuals—the dynamic approach A dynamic approach is used to calculating the pairwise correlation. This method involves calculating the cross correlation of residuals from autoregressive, AR(p), models of two countries, where p denotes the number of lags in the equation.2 The limitation of this approach is that the cross correlation result tends to depend on the autoregressive lag structure. Brockwell and Davis (1991) showed that this problem can be rectified if at least one of the series is white noise. Following Peiro (2002), a two-step approach is applied here to derive the autoregressive structure. First, AR(p = 1,2,3,4) models are estimated for each country.3 The AR(p) model for each country takes the following form: CAt ¼ α 1 þ α 2

p X

CAt−k þ CARt

ð1Þ

k¼1

Here, CAt is the current account imbalances at time, t. α1 is the constant; and CARt is the disturbance term or the residual from the OLS regression. As part of the second step, residuals of models with the lowest Schwarz Information Criterion statistics are used to calculate the correlation coefficient of the same variables between different pairs of countries. The cross correlation technique is used to calculate the correlation between the current account residuals. This is calculated for country i and j over time as:   corr CARi;t ; CAR j;t   rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi    ffi ¼ cov CARi;t ; CAR j;t = var CARi;t : var CAR j;t ;

ð2Þ

where CARi,t and CARj,t are the current account residuals of country i and j at time, t; corr is the pair-wise correlation; var is the variance; and cov is the covariance.4 Clearly, the disturbance term or the residuals contain all the other factors, except the t− k period current account, that determines the value of the dependent variable, CAt. If common shocks between two countries do not have any impact, then they are unlikely to be part of the residuals. In contrast, if CARi,t is highly correlated with CARj,t, then this would mean that common shocks have an impact on the current accounts of the two countries. The signs on the cross correlation coefficients imply the direction of the co-movement. A positive correlation coefficient suggests that the variables are moving in the same direction, such that when the current account of one country increases, the other also increases. On the other hand, a negative sign suggests that variables are moving in opposite directions, so that when the current account shock of one country is increasing, that of the other country is decreasing.

2.3. Preliminary examination of CARi,t and CARj,t This section focuses on the time series properties and the descriptive statistics of the current account shocks. First, to avoid spurious results, it is important that variables enter the correlation test in stationary form. As a result, the conventional Augmented Dickey–Fuller (ADF) test (Said and Dickey, 1984) and the Kwaitkowski et al. (1992) tests were used to determine the integrational properties of the residual series. The ADF test is constructed on the null that the series is non-stationary while the KPSS test is based on the null hypothesis that the series is stationary. The two tests suggest that CARt is I(0)or stationary for almost all counties for the full sample period. The sub-samples results also show similar result. For brevity, only the full sample results for selected countries are presented in Table 3. Table 4 presents the common descriptive statistics on the current account residuals for selected pairs. According to the Jarque–Bera test, the normality assumption for most series cannot be rejected. The only exceptions are residual series for Austria and Canada. For these countries, we notice excess Kurtosis, which means that the chance of extreme value in the future is higher for these series. Fig. 1 plots the residual series for selected nations. Visually, it is difficult to suggest any correlation between the series. However, they do portray highly volatile but stationary behavior, confirming the result derived above.

2.4. Results on key trading partners with the group of 17 OECD countries In this section, the pairwise correlations for the individual OECD countries are examined against their OECD partners. Table 5 presents the cross correlation results on the current account shocks for the first 26 pairs of countries. These results are provided for the full-sample; a common sample (1995:1–2006:3); the pre-1995 period; the Euro period; and the pre-Euro period. The correlations between current account shocks of the bulk of the pairs are weak (or statistically insignificant). The full-sample results show that 20 out of 26 pairs of countries have current account shocks which are not synchronized significantly. In other words, the correlations are statistically insignificant. The six pairs of countries that show statistically significant correlations include

Table 3 Time series properties of the current account residuals (CARi(j)). Country

Sample period

i¼1

mean of the variables.

ADF test

KPSS test

t-Statistic

t-Statistic

[Lag length]

[Bandwidth]

−9.9670⁎⁎⁎ [0] −6.9001⁎⁎⁎

0.0651 [1] 0.3076 [1] 0.0789 [1] 0.2533 [1] 0.1636 [4] 0.1185 [4] 0.1712 [0] 0.0912 [3] 0.1026 [1] 0.2020 [4]

Australia

1974:3–2006:2

Austria

1991:1–2006:3

France

1995:1–2006:3

Germany

1991:1–2006:3

[0] −6.2257⁎⁎⁎ [0] −8.8296⁎⁎⁎

Italy

1980:1–2006:3

[0] −12.9863⁎⁎⁎

Japan

1980:1–2006:3

Portugal

1995:1–2006:3

Spain

1995:1–2006:3

[0] −10.324⁎⁎⁎ [0] −6.7812⁎⁎⁎ [0] −4.5188⁎⁎⁎

Sweden

1993:Q1–2006:3

[0] −5.6773⁎⁎⁎

US

1960:Q1–2006:Q3

2

One can also apply a richer model, such as the autoregressive moving average (ARMA) model; however, the analysis with ARMA models does not change the results. Another option lies in the use of multivariate ARCH/GARCH type models; but these are suited for high frequency data and not for the quarterly data that we use in this paper. Therefore, the results reported here are derived from a relatively simple AR(p)model. 3 Given that the data is quarterly, the AR models are estimated with up to 4 lags. n  4 The variance and the covariance are calculated as follows: varðxi Þ ¼ 1 n−1 ∑ ðxi −xi Þ2 i¼1 n      and cov xi ; x j ¼ 1 n−1 ∑ðxi −xi Þ x j −x j , where x ¼ CAR; xa ; and (a = i, j) — are the

291

[0] −13.4307⁎⁎⁎ [0]

Notes: This table presents the unit root tests for the current account residuals of selected countries for the full sample. Both the ADF and KPSS tests are conducted with an intercept only. *** denotes the level of significance at 1%.

292

S. Narayan / Economic Modelling 38 (2014) 288–295

Table 4 Descriptive statistics on the current account residuals for selected pairs. Pairs

AUS

JAP

AUS

US

AUT

GER

BEL

GER

ITA

US

Mean Maximum Minimum Std. dev. Skewness Kurtosis Jarque–Bera Probability Observations

−5.8E−18 0.023 −0.020 0.010 0.190 2.663 1.108 0.575 103

−3.5E−05 0.009 −0.007 0.003 0.380 3.056 2.491 0.288 103

4.1E−18 0.025 −0.026 0.011 0.015 2.706 0.459 0.795 126

3.7E−05 0.014 −0.012 0.005 0.151 3.042 0.489 0.783 126

−1.6E−05 0.012 −0.016 0.005 −0.690 5.227 17.449 0.000 61

−3.3E−18 0.020 −0.017 0.007 −0.020 3.686 1.199 0.549 61

−5.1E−18 0.026 −0.026 0.010 0.071 3.408 0.357 0.836 46

−2.2E−18 0.019 −0.016 0.008 −0.118 3.385 0.391 0.822 46

2.4E−04 0.018 −0.018 0.007 −0.020 3.074 0.030 0.985 104

−4.4E−18 0.012 −0.012 0.005 −0.013 2.919 0.031 0.984 104

Pairs

FRA

SPA

SWE

GER

GER

POR

POR

SPA

CA

US

Mean Maximum Minimum Std. Dev. Skewness Kurtosis Jarque–Bera Probability Observations

−1.8E−05 0.009 −0.008 0.004 0.411 2.571 1.538 0.463 43

−2.7E−18 0.023 −0.029 0.013 −0.505 2.666 2.026 0.363 43

−1.1E−17 0.021 −0.016 0.008 0.246 2.731 0.564 0.754 43

7.3E−05 0.019 −0.017 0.008 −0.114 3.426 0.417 0.812 43

8.3E−05 0.019 −0.017 0.008 −0.119 3.505 0.572 0.751 44

3.2E−17 0.023 −0.018 0.008 0.209 3.675 1.157 0.561 44

−7.8E−05 0.023 −0.018 0.008 0.235 3.629 1.106 0.575 43

−2.7E−18 0.023 −0.029 0.013 −0.505 2.666 2.026 0.363 43

1.1E−18 0.037 −0.018 0.008 0.555 4.444 25.152 0.000 182

1.7E−18 0.014 −0.012 0.005 0.089 2.938 0.272 0.873 182

Notes: These descriptive statistics relates to full sample data only. AUS stands for Australia; AUT—Austria; BEL—Belgium; CA—Canada; FRA—France; GER—Germany; ITA—Italy; POR— Portugal; JAP—Japan; SPA—Spain; and SWE—Sweden.

.03

AUS-JAP

.03

AUS-US

.025

AUT-GER

.020 .02

.02

AUS

AUS

.015

.01

AUT

.010

.01

.005 .00 .000

.00 -.01

JAP

-.005

US

GER

-.010

-.01

-.02 -.015

-.02 80 82 84 86 88 90 92 94 96 98 00 02 04 06

.03

-.020 -.03 74 76 78 80 82 84 86 88 90 92 94 96 98 00 02 04 06 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06

BEL-GER .04

CA-US

.020

FRA-GER

.015

BEL

.02 .01

.02

.00

.01

GER

CA

.03

.010 .005 .000 FRA

-.005 -.01

.00 -.010

GER

-.02

-.01

-.015

US

-.03

.03

-.020

-.02 95

96

97

98

99

00

01

02

03

04

05

1965 1970 1975 1980 1985 1990 1995 2000 2005

06

POR-GER

.03

POR-SPA

93 94 95 96 97 98 99 00 01 02 03 04 05 06

.03

.02

SPA-FRA

.02

.02

SPA

SPA GER

.01

.01

.01 .00

.00

POR

.00

FRA

-.01

-.01

-.02

-.02

POR

-.01

-.02

-.03

-.03 95

96

97

98

99

00

01

02

03

04

05

06

95

96

97

98

99

00

01

02

03

04

05

06

Fig. 1. Current account residuals for selected pairs.

95

96

97

98

99

00

01

02

03

04

05

06

S. Narayan / Economic Modelling 38 (2014) 288–295

293

Table 5 Pair-wise cross correlation results between current accounts shocks. Pairs

Lags⁎(pi − pj)

1.

Australia–Japan

1–3

2.

Australia–US

2–1

3.

Austria–Germany

1–2

4.

Belgium–Germany

1–1

5.

Belgium–France

1–1

6.

Canada–US

1–1

7.

Denmark–Germany

4–2

8.

Denmark–Sweden

4–4

9.

Finland–Germany

4–2

10.

Finland–Sweden

4–4

11.

France–Germany

1–1

12.

Italy–Germany

1–2

13.

Japan–US

1–3

14.

Korea–US

1–1

15.

Korea–Japan

1–3

16.

Portugal–Germany

1–3

17.

Portugal–Spain

3–4

18.

Portugal–France

3–4

19.

Portugal–UK

3–2

20.

Spain–France

3–1

21.

Spain–Germany

3–1

22.

Spain–UK

3–3

23.

Sweden–Germany

4–4

24.

Turkey⁎–Germany

3–3

25.

UK–US

3–2

26.

UK–Germany

3–1

Full sample

Common sample (1995:1–2006:3)

Pre-1995 period

Pre-Euro period

Euro period

Correlation

Correlation

Correlation

Correlation

Correlation

a

a

a

a

(t-Statistic )

(t-Statistic )

(t-Statistic )

(t-Statistic )

(t-Statistica)

−0.017 (−0.176) 0.115 (0.125) 0.291⁎⁎ (2.336) −0.035 (−0.231) 0.154 (1.033) 0.019 (0.251) 0.049 (0.367) 0.006 (0.041) 0.240 (1.870) 0.130 (−0.919) 0.0590 (0.426) 0.253⁎⁎ (2.004) −0.084 (−0.849) 0.012 (0.122) 0.290⁎⁎⁎ (3.061) 0.122 (0.796) 0.352⁎⁎ (2.409) −0.105 (−0.673) 0.236 (1.575) 0.350⁎⁎ (2.350) 0.100 (0.645) −0.078 (−0.499) 0.309⁎⁎ (2.278) 0.067 (0.570) −0.092 (−1.248) −0.000 (0.002)

−0.349 (−0.033) 0.373⁎⁎ (2.669) 0.208 (1.424) −0.035 (−0.231) 0.154 (1.033) −0.009 (−0.062) −0.100 (−0.539) 0.011 (0.077) 0.273 (1.905) −0.080 (−0.537) 0.067 (0.452) 0.330⁎⁎ (2.341) 0.033 (−0.217) −0.021 (−0.138) 0.228 (1.574) 0.122 (0.796) 0.352⁎⁎ (2.409) −0.105 (−0.673) 0.236 (1.575) 0.350⁎⁎ (2.350) 0.100 (0.645) −0.078 (−0.499) 0.334⁎⁎ (2.375) 0.120 (0.808) −0.034 (−0.230) 0.018 (0.118)

−0.548⁎⁎⁎ (−4.349) −0.026 (−0.228) NA

NA

NA

NA

NA

0.033 (0.175) 0.239 (0.886) −0.236 (−0.875) NA

0.497⁎⁎⁎ (3.084) −0.035 (−0.231) 0.221 (1.220) NA

NA

0.168 (0.870) 0.122 (0.522) 0.101 (0.515) −0.063 (−0.268) 0.157 (0.726) 0.231 (1.257) NA

−0.025 (−0.980) −0.135 (−0.736) 0.309 (1.755) −0.167 (−0.913) 0.086 (0.466) 0.361⁎⁎ (2.082) NA

NA

NA

NA

NA

NA

NA

NA

0.648⁎⁎⁎ (4.586) −0.166 (−0.531) 0.348 (1.174) NA

0.124 (0.670) 0.491⁎⁎⁎ (3.036) −0.190 (−1.045) NA

0.358⁎⁎ (2.017) 0.159 (0.866) NA

0.350⁎⁎ (2.394) −0.100 (−0.539) NA

0.111 (0.478) NA

0.381⁎⁎ (2.224) NA

NA

NA

NA

NA

NA NA NA NA NA −0.089 (−0.282) NA NA NA

NA NA NA NA NA NA NA 0.130 (0.879) NA −0.253 (−0.907)

Notes: For brevity, we report the lag length for the full sample only. p denotes number of lags for the autoregressive models of country i and country j.⁎⁎ (⁎⁎⁎) denote the level of significance at 5% (1%). a The t-statistic corresponds to the test of zero correlation.

Austria–Germany; Italy–Germany; Korea–Japan; Portugal–Spain; Spain–France; and Sweden–Germany. Three new pairs are found in the sub-sample analyses. These include: Australia–Japan during the pre-1995 period and Australia–US in the post-1995 period. The third pair, Portugal–Germany, is found to be significant in the pre-Euro period. Moreover, all the significant pairs, except Australia–Japan, show a positive correlation. This indicates that for most pairs, common current account shocks are moving in the same direction. The correlation is negative for the Australia–Japan pair, implying that common shocks to the current accounts are moving in opposite directions. To get some indication of where these high correlations are coming from, we turn to subsample results. Here, notice that for most of the

significant pairs, there are significant variations in the degree of synchronization between current account shocks across the pre- and post-1995 or the pre-Euro and Euro sub-samples. For instance, for Austria–Germany; Italy–Germany, Portugal–Spain, Spain–France, and Sweden–Germany, the correlations are statistically insignificant in the pre-Euro period. However, in the Euro period, they become statistically significant. 2.5. Results on the relation between the US and 12 other OECD countries The cross correlation results between the US and OECD countries are presented in Table 6. The pair-wise cross correlation between current

294

S. Narayan / Economic Modelling 38 (2014) 288–295

Table 6 Pair-wise cross correlation results on the dynamic relations between the US and 12 other OECD countries. Pairs

1.

Austria–US

2.

Belgium–US

3.

Canada–US

4.

Denmark–US

5.

Finland–US

6.

France–US

7.

Germany–US

8.

Italy–US

9.

Portugal–US

10.

Spain–US

11.

Sweden–US

12.

Turkey–US

Full-sample

Common sample (1995:1–2006:3)

Pre-1995 period

Pre-Euro period

Euro period

Correlation

Correlation

Correlation

Correlation

Correlation

#

#

#

#

(t-Statistic )

(t-Statistic )

(t-Statistic )

(t-Statistic )

(t-Statistic#)

0.024 (0.201) 0.008 (0.050) 0.019 (0.251) −0.118 (−0.929) −0.027 (−0.298) 0.135 (0.972) −0.056 (−0.431) 0.024 (0.242) −0.212 (−1.407) 0.032 (0.206) 0.140 (0.993) 0.011 (0.782)

−0.147 (−0.999) NA

0.346 (1.804) NA

−0.009 (−0.062) −0.132 (−0.894) −0.105 (−0.710) 0.156 (1.056) −0.089 (−0.602) 0.067 (0.453) NA

0.058 (0.670) −0.044 (−0.166) 0.015 (0.129) −0.278 (−0.579) 0.041 (0.141) −0.002 (−0.015) NA

0.205 (1.328) 0.220 (0.782) NA

−0.113 (−0.613) −0.118 (−0.639) NA

NA

NA

0.143 (0.970) 0.101 (0.679)

0.134 (0.191) 0.064 (0.329)

−0.104 (−0.571) 0.015 (0.140) −0.046 (−0.207) 0.185 (0.994) 0.019 (0.162) 0.145 (0.486) 0.404 (1.398) 0.313 (1.396) 0.130 (0.852)

−0.032 (−0.171) −0.129 (−0.701) 0.109 (0.593) −0.132 (0.715) −0.043 (−0.234) −0.262 (−1.463) −0.081 (−0.436) 0.129 (0.702) 0.033 (0.177)

Notes: # The t-statistic corresponds to the test of zero correlation.

account shocks of all 12 OECD countries and the US are statistically insignificant. Consistent with the above findings, these results suggest that common current account shocks between the US and the other 12 OECD countries are irrelevant in the determination of their current accounts. Another observation that can be made from Table 6 is that although statistically insignificant (at the 5% level), Austria, Belgium, Denmark, Spain, Sweden, and Turkey experienced a fall in the correlation between current account residuals in the Euro period in comparison to the pre-Euro period. In contrast, Germany, Italy, and Portugal experienced an increase in the correlation. These may be related to the trade relations between these countries.

2.6. Granger causality in the short-run From the above results, it was found that current account shocks in Austria–Germany, Italy–Germany, Portugal–Spain, Spain–France, Sweden–Germany, Portugal–Germany, Australia–Japan and Australia– US are significantly correlated. One thing to note is that the pairs found with significant correlations in the Euro period comprise of the home country and their largest trading partner. For instance, the largest trading partner of Austria, Italy, and Sweden, is Germany; for Spain, it is France; and for Portugal, it is Spain. These results may be suggesting that the integration of economic and financial activities in the Euro period also led to the fine tuning of the transmission of current account shocks between these pairs of countries. Spillover shocks, in particular, transmit from one country to another through trade, financial integration, investment flows, information flows and migration of labor (see, Peiro, 2002). In some literature, such as the business cycle literature, spillover costs are referred to as ‘common shocks’ (see, for instance, Kose et al., 2003). The literature on business cycle synchronization and financial market integration provides evidence of spillover of shocks from one country to another by way of comovement of exchange rate markets, stock markets, business cycles, macroeconomic variables and shocks. Within this literature, there are various studies that associate comovement of business cycles or markets with trade and financial

linkages — the sources of spillover shocks (see, inter alia, Artis and Zhang, 1997, 1999; Babetskii, 2005; Frankel and Rose, 1998; Gruben and Koo, 2002; Imbs, 2004; Kose et al., 2003). Hence, it seems that strong trade links between the pairs are driving the strong correlation through the spillover effect. To examine whether or not the suspected spillover effects are taking place between the countries, this study applies the Granger causality test. This test specifically examines whether current account shocks of country i Granger cause the current accounts shocks of country j in the short-run and vice versa. Variables enter the Granger causality test in stationary form. As mentioned earlier, the ADF and KPSS tests suggest that CARt is I(0)or stationary for all for the full sample period as well as the sub-sample period of interest here (see, Section 2.3 and Table 3). Eqs. (2) and (3) are estimated with autoregressive lag lengths p and q for countries i's and j's residuals, respectively: CARi;t ¼ δ1 þ ϕ1

p X p¼1

CAR j;t ¼ δ2 þ ϕ3

q X q¼1

CARi;t−p þ ϕ2

p X

CAR j;t−p þ ε1;t

ð3Þ

p¼1

CAR j;t−q þ ϕ4

q X

CARi;t−q þ ε2;t

ð4Þ

q¼1

Here, δt, and ϕt are parameters that will be estimated; εt is the disturbance term; and CARi,t and CARj,t are the current account residuals of country i and j, respectively. These residuals were derived in the previous section. The optimal lag lengths are selected by the Schwarz Information Criterion. The F-statistics and the p-values associated with each model are reported in Table 7. The results reported here correspond to pairs which showed signs of significant correlations. Two key results emerging for the pairs which are European Union members (Portugal–Germany, Spain–France, Austria–Germany, and Portugal– Spain) from this exercise are as follows. First, only Austria–Germany and Portugal–Spain show signs of unidirectional Granger causality in the short-run. Second, since the inception of the Euro in 1999, only Portugal–Spain shows signs of Granger causality. This suggests that for

S. Narayan / Economic Modelling 38 (2014) 288–295 Table 7 Granger causality results. Lag′

1. 2. 3. 4. 5. 6. 7. 8.

AUT→ GER→ ITA→ US→ SPA→ POR→ SPA→ FRA→ SWE→ GER→ POR→ GER→ AUS→ US→ AUS→ JAP→

GER AUT US ITA POR SPA FRA SPA GER SWE GER POR US AUS JAP AUS

1–1 1–1 2–12 12–2 12–12 12–12 12–12 12–12 2–1 1–2 1–1 1–1 12–6 12–6 1–1 1–1

Full sample

Lag′

F-stat

Prob.

4.2829⁎⁎ 1.0214 1.2549 1.2974 9.4391⁎⁎⁎

0.0430 0.3165 0.3287 0.3060 0.0056 0.1837 0.2512 0.3138 0.1999 0.7364 0.5767 0.5541 0.7398 0.1236 0.5806 0.7327

2.1174 1.7617 1.5248 1.6561 0.4253 0.3167 0.3561 0.7538 1.5197 0.3073 0.1173

1–1 1–1 12–12 12–12 12–12 12–12 12–12 12–12 1–1 1–1 1–1 1–1 2–2 2–2 1–1 1–1

Euro period F-stat

Prob.

2.1306 0.6435 1.6567 1.3426 9.4391⁎⁎⁎

0.1555 0.4292 0.2768 0.3753 0.0059 0.1837 0.2512 0.3138 0.0318 0.2767 0.9475 0.7918 0.7858 0.0492 0.9831 0.3278

2.1174 1.7617 1.52481 5.1072⁎⁎ 1.2305 0.0044 0.0711 0.2425 3.2451⁎⁎ 0.0005 0.9800

Notes: Subsample analysis cover the period 1999:Q1–2006:Q4 for 1–6, while the period 1995–2006:Q3 is covered for 7–8. Here, AUS stands for Australia; AUT—Austria; FRA— France; GER—Germany; ITA—Italy; POR—Portugal; JAP—Japan; SPA—Spain; and SWE— Sweden. The optimal lag lengths are selected by the Schwarz Information Criterion. The two lag lengths relate to the countries in the pair, respectively.

the pairs–Portugal–Germany and Spain–France–spillover effects are not driving the high correlation that we found in the previous section. This finding contradicts the GR result but signals economic asymmetries which may have been caused by financial and fiscal problems of Portugal and Spain.5 For the rest of the pairs (Sweden–Germany, Australia–US, Australia– Japan, and Italy–US), there is evidence of unidirectional Granger causality in two. For Sweden–Germany, we find that current account shock of Sweden Granger cause Germany's current account shock in the Euro period. For Australia–US, the current account shocks of US are found to Granger causes current account shocks of Australia over the recent period 1994–2006. Hence, for the two pairs, the high correlation may have been related to the spillover effects between the countries. 3. Concluding remarks It is readily recognized in the current account literature that global or worldwide shocks do not have any impact on the current account imbalances provided that the world economies are symmetric. This paper tested whether this is also true for common shocks between pairs of countries. The 17 OECD countries were paired with their key trading partners from a sample of OECD countries and the US. In total, 37 pairs of countries were examined. It was found that the cross correlation between current account shocks for pairs of countries was statistically insignificant for the bulk of the pairs studied. This is taken to be consistent with the GR prediction that global shocks are irrelevant in the determination of current account movements. However, some pairs of countries were at odds with the GR prediction, in that they showed strong correlation between current account shocks. Some of these pairs were made up of Euro nations; for instance, Austria–Germany, Italy–Germany, Portugal–Spain, Spain–France, and

5 For discussion and empirical evidence on the twin deficit problem in Portugal and Spain (see, Trachanas and Katrakilidis, 2012) and on the long-run unsustainability of the current account imbalances of these two nations in the long run (see, Chen, 2011).

295

Portugal–Germany. Interestingly, these pairs also showed much stronger correlations in the Euro period. These high correlations for these pairs are either suggesting that common shocks (or spillover effects) matter or that there are asymmetries between the two countries. To investigate whether there were spillover of costs and benefits from one country to the other, the Granger causality was conducted test for these pairs. From this, limited evidence of unidirectional Granger causality was found. For Sweden–Germany and Australia–US, there is evidence to suggest that spillover shocks matter for their current accounts. For Austria–Germany and Spain–Portugal, evidence of Granger causality in the pre-Euro period was found but in the Euro period, only the latter pair was affected by spillover shocks. Spain–France and Portugal–Germany show interesting results. They are highly correlated but show no Granger causality. This is taken as an indication that the correlation between these pairs was not driven by spillover effects. This, in turn, suggests that strong correlations were driven by global shocks. This finding points towards economic divergence between these EU member countries. Exactly what factors are causing this economic divergence between the EU countries is the subject for future research. References Artis, M.J., Zhang, W., 1997. International business cycles and the ERM. Int. J. Finance Econ. 2 (1), 1–16. Artis, M.J., Zhang, W., 1999. Further evidence on the international business cycle and the ERM. Oxf. Econ. Pap. 51, 120–132. Babetskii, I., 2005. Trade integration and synchronization of shocks. Econ. Transit. 13 (1), 105–138. Brockwell, P., Davis, R., 1991. Time series: theory and methods, 2nd edition. SpringerVerlag, New York. Camacho, M., Quiros, G.G., Saiz, L., 2006. Are European business cycles close enough to be just one? J. Econ. Dyn. Control 30, 1687–1706. Chen, S.-W., 2011. Current account deficit and sustainability: evidence from the OECD countries. Econ. Model. 28, 1455–1464. Frankel, J., Rose, A., 1998. The endogeneity of the optimum currency area criteria. Econ. J. 108, 1009–1025. Furceri, D., Karras, G., 2007. Business cycle synchronization in the EMU. Appl. Econ. 37. Glick, R., Rogoff, K., 1995. Global versus country-specific productivity shocks and the current account. J. Monet. Econ. 159–192. Gregory, A.W., Head, A.C., 1999. Common and country-specific fluctuations in productivity, investment and the current account. J. Monet. Econ. 423–451. Gruben, W.C., Koo, J., 2002. How much does international trade affect business cycles synchronization? Working paper 0203. Federal Reserve Bank of Dallas. Imbs, J., 2004. Trade, finance, specialization and synchronization. Rev. Econ. Stat. 86 (3), 723–734. IMF, 2006a. Direction of Trade Statistics — by commodities. IMF, 2006b. Direction of Trade Statistics — by services. Iscan, T.B., 2000. The Terms of Trade, productivity growth and the current account. J. Monet. Econ. 45, 587–612. Kose, M.A., Prasad, E.S., Terrones, M.E., 2003. How does globalization affect the synchronization of business cycles? Am. Econ. Rev. 93 (2), 57–62. Kwaitkowski, D., Phillips, P.C.B., Schmidt, P., Shin, Y., 1992. Testing the null hypothesis of stationarity against the alternative of a unit root. How sure are we that economic time series have a unit root? J. Econ. 54, 159–178. Nason, J.M., Rogers, J.H., 2002. Investment and the current account in the short run and the long run. J. Money Credit Bank. 34 (4), 967–986. Peiro, A., 2002. Macroeconomic synchronization between G3 countries. Ger. Econ. Rev. 3 (2), 137–153. Said, E., Dickey, D.A., 1984. Testing the Unit Root in autoregressive moving average models of unknown orders. Biometrika 71, 599–607. Trachanas, E., Katrakilidis, C., 2012. The dynamic link between fiscal and current account deficits: evidence from five highly indebted European countries accounting for regime shifts and asymmetries. Econ. Model. 31, 502–510.