The Current Account Sustainability of European Transition Economies

This article presents an analysis of the sustainability of the current accounts of a group of central and eastern European countries. Given the link between national savings (public and private) and investment, the current account may yield instabilities in fundamental macroeconomic variables. Hence, this analysis is of paramount importance given the 2008–11 debt crises faced by many European economies, and the addition of new countries to the economic and monetary union. By means of unit root tests and fractional integration it is shown that, in general, the ratio of the current account to gross domestic product is a stationary and mean reverting process. However, in some cases, shocks tend to have long‐lasting effects, implying that there is no evidence of a potential debt default in this group of countries.


Introduction
The analysis of current account deficits and their sustainability has gathered momentum since the 2008-10 financial crisis. Fears of countries defaulting on their external debt have increased during the last few years after some European countries have shown relatively high rates on both internal and external debt. The flow of funds approach, from basic macroeconomic theory, predicts that net exports should equal public saving plus private saving net of investment. Hence, there is an important connection between external and internal debt.
A popular approach to analyzing the degree of persistence in the current account deficit and therefore whether a current account is sustainable is the use of unit root and stationarity tests (see, for example, Coakley et al., 1996;Milesi-Ferretti and Razin, 1996;Taylor, 2002). If shocks have transitory effects, the current account is a stationary process and external debt is mean reverting. In this situation, according to Trehan and Walsh (1991) and Husted (1992), the country is solvent and therefore these are necessary conditions for sustainability. Hence, if the current account balance is mean-reverting and stationary, external debt will not grow forever after a shock. In a seminal contribution, Bohn (2007) provides evidence that stationarity is not necessary for the transversality condition, derived from the inter-temporal budget constraint, to hold. Still, the external deficit may satisfy the transversality condition for higher orders of integration than zero. However, the time series properties of the current account are quite informative.
Policy-makers may want to learn when and how current account mean reversion happens. Within the literature, stationarity of the current account balance over gross domestic product (GDP) is seen as a strong form of sustainability.
If shocks have permanent effects, external debt is a unit root process, and may even be explosive so that external debt will not revert to equilibrium after a shock. In this situation, deficits tend to increase in the long run and the application of economic reforms will be necessary to avoid a situation of excessive debt accumulation. Indeed, reform may also be necessary when the process is not stationary, but mean-reverting, when the speed of mean reversion is relatively slow. Such considerations are particularly important for European transition economies -particularly those central and eastern European countries (CEECs) which are candidates, or even have aspirations, to join the eurozone (see below for more details).
There is a much smaller literature on the analysis of current account sustainability for recently emerged economies than for industrial economies (Holmes, 2006). For CEECs, Holmes (2004) finds evidence of current account sustainability for some by means of applying (linear) unit root tests to panel data, in contrast to the general findings of studies looking at industrialized countries (Holmes, 2006;Stein, 2007;Christopoulos and León-Ledesma, 2010;Cunado et al., 2010).
This article follows the approach of Christopoulos and León-Ledesma (2010), who apply nonlinear unit root tests to test for current account sustainability (their application was to the United States). In addition, panel unit root tests are used in order to gain power by taking into account cross-sectional information. Finally, in order to gain some flexibility when analyzing the order of integration of variables (Gil-Alana and Robinson, 1997, among others), the Robinson (1995) test for fractional integration is also applied.
The remainder of the article is organized as follows. The first section discusses the issue of current account sustainability for CEECs. The second section explains the methods applied to analyze current account sustainability, and the third and fourth sections present the results and concluding remarks, respectively.

I. Current Account Sustainability and Central and Eastern Europe
As pointed out by Roubini and Wachtel (1999), current account deficits are of particular importance for transition economies, given the general upward trend in the real value of their currencies. With an appreciating real exchange rate and currency boards, like the ones maintained by most CEECs, this may destabilize the current account even more.
The reasons for analyzing the current account sustainability are, then, twofold. First, the current account balance could be considered a proxy for the strength of the external position of a country since it is a measure of the foreign resources that come into the country to finance insufficient national savings. Second, the degree of persistence of current account deficits provides insights into the possibility that countries might default. Temporary or transitory current account deficits may promote economic growth as far as they are allocated in countries to productive investments. However, permanent deficits might imply increasing interest repayments and this may impose restrictions on future generations and/or, eventually, the impossibility of ultimate redemption of debt. In this situation, a short-term solution is the application of a tight monetary policy in order to increase national interest rates to attract foreign capital. This, of course, will affect real The current account sustainability of European transition economies exchange rates and the overall competitiveness of the country, reducing the current account. However, these measures increase the cost of debt and, to the extent that this happens, debt repayment will be even more difficult for the host country (see Christopoulos and León-Ledesma, 2010, among others). Also, this will increase the overall fiscal debt burden, thereby increasing internal debt problems which many European countries are already facing.
This situation will not be very promising, especially for those EU countries which joined the Union without an opt-out clause and who will therefore have to join the economic and monetary union (EMU) in the future. This implies satisfying the Maastricht criteria on public debt, interest rates, exchange rates and inflation differentials with the best three EU inflation performers. In particular, the Maastricht Treaty states that the country should be in the exchange rate mechanism II for two years without friction, which means no possibility of devaluation whatsoever. If countries run large and permanent current account deficits, devaluation of the currency could otherwise be a feasible option to increase competitiveness. This downward pressure over the value of a country's currency might provoke speculative attacks. To date, Estonia, Slovakia and Slovenia have become members of the EMU and have adopted the single currency as a unique legal currency. For these economies, the possibility of devaluation is non-existent, and hence, the reduction of persistent current account deficits would have to rely on structural measures.

II. Econometric Methods
The econometric approach is to apply a number of unit root tests for panel data, individual series and fractional integration in order to analyze the long-run behaviour of the current account ratio to GDP for a pool of CEECs. First, a group of panel unit root tests are applied. These tests take into account cross-sectional information, although is not possible to distinguish which series are I(0) when the null is rejected. Thus, in this paper the Levin, Lin and Chu (LLC) (Levin et al., 2002), Im, Pesaran and Shin (IPS) (Im et al., 2003) and Maddala and Wu (1999) and Choi (2001) (MWC) tests are applied. The first test imposes a common unit root under the null hypothesis, against the alternative of stationarity of all individual series, whereas the latter allow for individual stationarity under the alternative hypothesis. This supposes a less restrictive framework as, in the former case, the assumption of a common unit root under the null, or general stationarity under the alternative, may be too strong. Hence, IPS base their test on the assumption of different autoregressive parameters for each individual series.
An alternative approach is taken by MWC, who combine the different p-values of the individual auxiliary regressions, either for the Augment Dickey-Fuller (ADF) or Phillips-Perron (PP) (Phillips and Perron, 1988) tests, to obtain the following Fisher (1932)-type test: where p i is the asymptotic p-value of a unit root test for individual i.
Finally, it is also possible to apply the Kwiatkowski et al. (1992) (KPSS) test in a panel framework. Thus, Hadri's (2000) test is a panel version of the stationarity test KPSS. Therefore, for the Hadri test, the null hypothesis is stationarity of all the individual countries of the panel.
In order to analyze the order of integration of the ratio of current account to GDP for the individual countries, two groups of unit root tests are also considered: Ng and Perron (2001), which are based on linear models; and Kapetanios, Shin and Snell (KSS) (Kapetanios et al., 2003) and Sollis (2009), which are nonlinear. This article uses linear unit root tests as a starting point and benchmark for the subsequent analysis. However, because the assumption of constant parameters might be too simplistic, the possibility of changing parameters is also taken into consideration. Unlike models with structural breaks, which normally depend on some historical incident that changes permanently the mean or the slope of the relations investigated, here is incorporated the possibility of changing autoregressive parameters, depending upon the size of shocks. In general, small shocks, which may only have mild effects on a given economic variable, are unlikely to trigger any alarm and the national authorities may decide not to act to correct for any deviations. Nevertheless, when shocks are significant in nature and have an important effect on the target variable, the responsible authority may put in place a series of mechanisms and policy adjustments aimed at neutralizing the effects of the shocks. In this situation, we may observe that the further the variable deviates from the equilibrium value, the faster will be the reversion towards it. This relaxation of the constancy of parameters implies that the autoregressive parameter in auxiliary regressions for unit root tests should depend on the values of the so-called 'transition variable', which in general is a lag of the variable of interest.
As previously mentioned, this type of state-dependent model should not be confused with models including structural breaks in which, although a type of nonlinear model, the nonlinearity in the latter is related to the deterministic component and not to the speed of adjustment towards equilibrium. Neglecting these sources of parameter inconstancy when testing for unit roots has been reported to affect the power of the tests (see KSS, among many others). If the underlying data generation is nonlinear, linear unit root tests may confuse a stationary process with a unit root when the nonlinearity is not accounted for. In the case of this article, let us suppose an inner regime and an outer regime where the ratio of current account to GDP may behave in a different manner.
According to the literature, nonlinearities may be relevant for analyzing the time series properties of the current account; Milesi-Ferretti and Razin (1998) study declines in current account deficits and exchange rate depreciations in a number of low and middle income countries. They find the existence of some domestic factors which may trigger current account reversals. In addition, Freund (2000) finds that current account reversal is a function of the business cycle, and that there is a threshold (5 per cent of GDP deficit) which triggers reversion of current account for a number of industrialized countries. Finally, Mann (2002) analyzes the fundamentals of American current account reductions, arguing that the speed of mean reversion may be faster if investors adjust their portfolios. This may be done by means of reducing their holdings of American assets and a depreciating dollar able to equilibrate imports and exports. 1 Above all, these works highlight the importance of nonlinearities in the process of adjustment, which needs to be accounted for.
In this article the KSS test is also applied. KSS has under the null a unit root process, but unlike the linear unit root tests takes into consideration the possibility of a globally stationary exponential smooth transition autoregressive (ESTAR) process under the alternative hypothesis. This makes it possible to characterize the target variable as a two regime process, for which the change in regimes is smooth rather than sudden. 2 Therefore, the variable may behave as a stationary process in the outer regime, but as a unit root in the inner regime. The unit root hypothesis can be tested against the alternative of a globally stationary ESTAR process using the following auxiliary regression: The null hypothesis H 0 : q = 0 that the process is a unit root in the outer regime is then tested against the alternative H 1 : q > 0 of stationarity. However, this test cannot be performed directly over q, given that it is not possible to identify this parameter under the null hypothesis of random walk. By means of a first-order Taylor expansion of Equation (2), KSS proposes the form: Δy Testing H 0 : b = 0 against H 1 : b < 0 is equivalent to testing for unit roots in the outer regime. Equation (3) may incorporate lagged Dy t . KSS considers three possibilities regarding the deterministic components: applying the test to the raw data, to the demeaned data, and to the demeaned and detrended data. Since we are analyzing the ratio of the current account to GDP against convergence to an equilibrium value, the KSS test is applied to the demeaned data. The nonlinear function used by KSS in order to take into account nonlinearities assumes that shocks have symmetric effects upon the variable -that is, the sign of the shocks does not matter, only the size. However, for many economic variables this assumption may be too simplistic. The speed of mean reversion may actually depend not only on the absolute deviation from the equilibrium, but also upon the sign of the shock. Intuitively, it makes sense to think that a negative shock on the current account balance may be more difficult to tackle than a positive shock. Hence, Sollis (2009) proposes a similar test to KSS in the sense that both assume that the speed of mean reversion depends on deviations from equilibrium. However, Sollis distinguishes asymmetric or symmetric effects under the alternative hypothesis. This asymmetric ESTAR model (AESTAR) is defined as: with g 1 Ն 0, and S t (g 2 , y t-1 ) = {1 + exp(-g 2 y t-1 )} -1 , with g 2 Ն 0. Again, Equation (4) may incorporate lags of the dependent variable to control for autocorrelation.
The null hypothesis of a unit root can be specified as H 0 : g 1 = 0. However, under the null hypothesis, g 2 , r 1 and r 2 cannot be identified. To solve this problem, Sollis proposes the use of the following auxiliary equation using Taylor  Thus, testing for unit roots in Equation (5) implies testing H 0 : b 1 = b 2 = 0 by means of an F-type test, whose critical values are given by Sollis (2009, p. 121), since the standard F distribution is not valid for an unknown order of integration of the residuals in Equation (5).
If the null hypothesis is rejected, the possibility of symmetric versus asymmetric shocks may be of relevance. Thus, this latter hypothesis can be tested by means of standard hypotheses testing since the null of symmetry would imply that b 2 is not statistically significant.
In order to take into account the possibility of a slow speed of mean reversion towards equilibrium, the possibility of fractional orders of integration is also tested for. 3 The aforementioned unit root tests only consider integer numbers for the order of integration, say d, which may be too restrictive -particularly when the variable requires a long period of time to revert to its mean. Following the contributions in the field of spectral analysis, long memory and fractional integration, the tests of Robinson (1995), which take into account the possibility of values of d in the interval (0, 1) or even above 1, is also used. Robinson's method is based upon the original idea of Geweke and Porter-Hudak (1983) using a log-periodogram-type regression. It estimates the value of d in: where L is the lag operator and e t is I(0), without taking into consideration any autoregressive (AR) or moving average (MA) structure. 4 The closer the parameter d is to 1, the more persistent is the process, and the effect of shocks on the variable will last longer. If d ∈ (0, 0.5) the series is covariance stationary and mean-reverting. However, if d ∈ [0.5, 1) the series is no longer stationary, but is still mean-reverting. The case when d Ն 1 implies that the series is non-stationary and non-mean-reverting. Robinson (1995) proposes a multivariate semiparametric approach in order to estimate the differencing parameter d in Equation (6). This test may be applied to individual series or to a pool of variables; allowing in the latter, intercept and slope to be different for each individual of the pool.

III. Empirical Evidence
The variable of interest for the current analysis is the ratio of the current account to GDP. The data for this empirical analysis have been obtained from Eurostat. 5 Quarterly data have been used, from 1999:Q1 to 2011:Q3. The data have been seasonally adjusted using the X-12 filter. Figure 1 displays the ratios for the target countries. It appears that the deficits have been quite close to zero for most of the sample for countries like Bulgaria, Lithuania, 3 See Diebold and Rudebusch (1991), Hasslers and Wolters (1994) and Lee and Schmidt (1996) for the analysis of power of unit root tests in the context of fractional integration. 4 Extensions of this approach have been examined by Moulines and Soulier (1999), Velasco (2000), Phillips and Shimotsu (2002) and Andrews and Guggenberger (2003), among others. 5 «http://epp.eurostat.ec.europa.eu/portal/page/portal/eurostat/home/». Slovakia and Slovenia. However, this changed at the end of 2006 with the beginning of the global economic crisis. Figure 1 also shows that the deficits have improved at the end of the sample. As pointed out by Aristovnik (2006), the current account deficits for most of these countries are a result of long-term growth and structural, external and domestic policy factors -particularly the growth in trade deficits of merchandise products, slowdown in services trade, profit repatriation and appreciation of the exchange rate.
The results from the panel unit root tests reported in Table 1 are rather mixed. Applying the LLC, IPS and ADF test on the raw data, the null of a common unit root cannot be rejected. In contrast, the PP and Hadri tests point to the existence of stationarity in the data. Using the cross-sectional de-meaned data in order to take into account cross-section dependence, 6 the results are similar, with some evidence towards the alternative hypothesis at the 10 per cent significance level when applying the IPS test.   In order to distinguish which countries' ratios are stationary, Table 2 reports the Ng-Perron, KSS and Sollis unit root tests results. These do not seem very promising. The null of a unit root can only be rejected at the 5 per cent significance level for the Czech Republic, Estonia and Lithuania. Some evidence of rejection of the null is found in the case of Latvia with the Ng-Perron test at the 10 per cent level. In addition, for the Czech Republic and Lithuania the null of asymmetric effect is rejected.
As mentioned, unit root tests may not be able to distinguish between unit root processes and fractional integrated processes. Thus, Table 3 displays the Robinson pooled test for fractional integration. Interestingly, the null hypothesis that d = 0, cannot be rejected in any case at conventional significance levels, and the estimated d are below 1. Therefore, the ratio of current account to GDP turns out to be a mean-reverting process. This means that after a shock, the ratio tends to correct the effect of the shock and returns to the longrun equilibrium. From Table 3, it can be seen that the speed of mean reversion is different for each country, given that the estimated d differs from country to country. However, in order to take into account the possibility of a more general model, autoregressive, fractionally integrated, moving average (ARFIMA) models are estimated for the target countries. ARFIMA (p, d, q) models take the form: where F p (L) and Q q (L) are polynomials of orders p and q, respectively, with all zeros of F p (L) outside the unit circle, and all zeros of Q q (L) outside or on the unit circle, and e t a white noise process (Granger and Joyeux, 1980;Granger, 1980Granger, , 1981Hosking, 1981). This has been done by means of using the Fox and Taqqu (1986) approach. The results are presented in Table 4. For all countries, an ARFIMA (4,d,0) seems to be the most appropriate model, except for Poland, where the selected model is an ARFIMA (1,d,0). The selection of the model has been made according to the Akaike information criterion. Again, the results point to different degrees of persistence, and in most cases the variables seem to be mean reverting, except for Romania and Hungary. This is corroborated by the impulse-response functions obtained by means of Gourieroux and Monfort's (1997, p. 438) theorem, which are displayed in Figure 2. In general the variables show a high degree of persistence after a shock, although some interesting distinctions can be made. First, Slovakia and Slovenia tend to suffer less after a shock since the immediate effects are not very large, and the effects of the shock tend to vanish relatively quickly. Bulgaria, Estonia and Latvia seem to suffer a huge impact immediately after the initial shock and, although the speed of mean reversion seems to be quite rapid, it takes a significant number of periods for the effects to disappear. The Czech Republic, Poland and Lithuania only seem to suffer mild effects immediately after a shock, although the speed of mean reversion tends to be slow. Finally, Hungary and Romania do not seem to present mean reversion at all.  Robinson (1995), the test has been applied using 0.9 as power for the number of ordinates entering the log-periodogram regression. Interestingly, Slovakia and Slovenia are the countries which joined the eurozone first -in 2009 and 2007, respectively. Looking at the evolution of their current accounts as a percentage of their GDP, in Figure 1, the results are not surprising since they are the countries with the most equilibrated current account in the region, probably linked to their examination for euro membership.
Some of the target countries have been in the exchange rate mechanism II in order to fulfil this element of the Maastricht criteria. Given that eurozone countries are their main trading partners, joining the single currency and losing the possibility of devaluation or revaluation will not help to correct current account deficits. This implies, therefore, that additional and probably more demanding policy decisions will need to be taken to reduce future current account deficits. This problem is particularly important for those countries which tend to suffer more severely from the effects of shocks on their current accounts.
In general it is found that, although the degree of persistence varies from country to country, with the exception of Hungary and Romania there is no statistical evidence indicating a potential problem of current account sustainability in this group of CEECs investigated. This result contrasts with previous studies on industrialized economies (for example, Cunado et al., 2010) and complements those of Holmes (2004). The macroeconomic adjustments performed during the last decades by this group of countries, from communism to market economies in order to prepare for EU membership, have helped to control external debt.