Are the Twin or Triple Deficits Hypotheses Applicable to Post-Communist Countries?

This study empirically examines the validity of the twin and triple deficits hypotheses using bootstrap panel Granger causality analysis and an annual panel data set of six post-communist countries (Russia, Poland, Ukraine, Romania, the Czech Republic, and Hungary) from 1994 to 2012. Our findings, based on panel data analysis under cross-sectional dependence and country-specific heterogeneity, support neither the twin deficits hypothesis nor its extended version, the triple deficits hypothesis, for any of the countries considered. In other words, we find no Granger-causal relationship between budget deficits and trade (or current account) deficits or among budget deficits, private savings-investment deficits, and trade deficits.


Introduction
The twin deficit hypothesis proposes that budget deficits and trade (or current account) deficits of an economy are intertwined. Deterioration in the budget balance results eventually in a corresponding deterioration of the trade (or current account) balance of an economy.
Whether in the context of the recent Eurozone crisis or US Congressional wrangling over the debt ceiling, this unresolved postulate is invoked repeatedly in framing macroeconomic policy discussions.
The twin deficits hypothesis gained popularity in the US in the early 1980s at a time when large chronic current account deficits were accompanied by widening US budget deficits. In 1984, Martin Feldstein, during his chairmanship of President Ronald Reagan's Council of Economic Advisers (Frankel, 2006) termed the co-existence and tandem movement of budget deficits and trade (or current account deficits) "twin deficits." According to Feldstein (1992), side-by-side depictions of budget and trade deficits produced an image of inseparable "Siamese twins." In the US case, the twin deficits hypothesis re-emerges in wide political discussion whenever the US is experiencing worsening trade deficits. Some of this may be ascribed to its populist appeal. As Gregory Mankiw noted in a December 2005 speech during his stint as head of George W. Bush's economics team: "From the perspective of the Beltway mercantilists, the trade deficit is a huge national problem. They look at the trade deficit simply as lost jobs for Americans." (Mankiw, 2006: 680). Trade deficits can be problematic for most nations, of course, so it is hardly surprising that the twin-deficits linkage has found its way into macroeconomic policy conversations around the world.
In recent years, a "triple deficits" hypothesis that includes the private savings-investment gap has emerged. Simply put, the triple deficits hypothesis proposes a linkage among government budget balance, savings-investment balance, and foreign trade (or current account) balance of an economy. Accordingly, government budget deficit along with savings-investment deficit (i.e. the economy-wide resource gap) induces trade (or current account) deficits. "Triple deficits" refers to the case where the domestic imbalance (simultaneous budget and private savings-investment deficits) is accompanied by an external imbalance (trade or current account deficits). To the best of our knowledge, Szakolczai (2006) may be credited for introducing the term "triple deficits" into wide use.
The relationship of budget deficits, private savings-investment deficits, and trade (or current account) deficits is a natural topic of interest for academics and policymakers.
Understanding possible causal relationships among these variables is a pre-condition for designing robust macroeconomic policies and creating policies that promote macroeconomic stability and economic growth. It is also generally accepted that large and persistent deficits threaten macroeconomic stability and growth. Indeed, as the experiences of many countries

Hüseyin Şen and Ayşe Kaya
Are the twin or triple deficits hypotheses applicable to post-communist countries?
6 have shown, large and persistent budget deficits cause serious problems for future generations by leaving them with a repayment burden. Similarly, large and persistent budget and trade deficits are problematic for countries when they drain their currency reserves, cause them to take on excessive debt, or set the stage for an economic crisis.
Perhaps the largest perceived threat of dual budget and trade (or current account) deficits, however, is their ability to induce macroeconomic imbalances that damage the longrun economic development trend of a country. This concern was prominent among policymakers in European transition economies two decades ago, when their countries faced huge initial distortions and there was great potential to run sizable trade and budget deficits for many years.
In the following analysis, we consider the validity of the twin deficits hypothesis and its cousin the triple deficits hypothesis in the context of six European transition economies (Russia, Poland, Ukraine, Romania, the Czech Republic, and Hungary). To the best of our knowledge, this is the first analysis attempting to examine the double and triple deficits hypotheses for these transition economies. We also employ bootstrap panel Granger causality analysis, a recent technique proposed by Kónya (2006) that allows for simultaneous analysis of Granger causality between two or three variables. The bootstrap panel Granger causality approach is based on a seemingly unrelated regression (SUR) estimation that considers cross-sectional dependence across countries. In practice, it means we can test Granger causality for each country by taking into account the possible contemporaneous correlation across countries. The approach is also based on a Wald test with country-specific bootstrap critical values, so it does not require a joint hypothesis for all members of the panel.
The rest of the study is divided into four parts. Section 2 briefly outlines the macroeconomic developments of the countries in our sample. Section 3 introduces the theoretical motivation and previous empirical findings on the twin and triple deficits hypotheses. Section 4 describes the data set, variables and methodology of the study, while Section 5 reports empirical results and discussion. Section 6 presents concluding remarks.

Macroeconomic backgrounds of six post-communist countries
In the aftermath of the collapse of the Soviet Union, a number of formerly socialist countries embarked on the long and painful transition process to market-based economies similar to their Western counterparts. The speed of the transition process has varied considerably across our six sample transition countries. In Poland and Russia, the process was quite rapid, and nearly as fast in Czechoslovakia. Hungary, a relatively more liberalized country, had less need for rapid change, so progress was slower. Romania and Ukraine faced resistance to reforms from pressure and interest groups. Nevertheless, all of these countries eventually In any case, virtually all Eastern European countries performed poorly in their first decade of transition. Policies tended to focus on the low-hanging fruit of industrial growth revival, rather than the harder-to-reach challenge of correcting macroeconomic imbalances.
Thus, monetary and fiscal policies in these countries created large demand for inadequate goods and services. With persistent excess demand, these countries encountered serious macroeconomic problems, including output gaps, unsustainable external debt, and high inflation.
Given the lack of monetary policy tools, transition countries initially adopted pegged exchange rate regimes. As time went on, they migrated to intermediate exchange rate regimes and eventually managed floats that recognized the potential destabilizing effects of international capital inflows and minimized negative effects on exports. In the initial one or two years of transition, all of our sample countries adopted conventional fixed pegs. In subsequent years, they migrated to export-oriented exchange rate regimes, such as crawling pegs, crawling bands or managed floats without pre-announced exchange rate trajectories.
Although none of these arrangements provides a stable exchange rate regime, our four sample countries that joined the EU (Hungary, Poland, Romania, and the Czech Republic) all eventually adopted independent floating regimes. Russia, in contrast, employed a managed float throughout most of the observation period. Ukraine has tried several exchange rate regimes, including fixed peg and independent float.
Russia's 1998 financial crisis negatively affected all our sample countries, most notably in the form of a collapse in Russian imports. Ukraine was hit hardest, due to its close trade ties with Russia. All sample countries devalued their currencies or abandoned their existing exchange rate regime following large ruble devaluations and all experienced subsequent declines in growth. The Ricardian approach, in contrast, asserts that increased budget deficits (regardless of whether they stem from tax cuts, higher spending or both) cause forward-looking economic agents to increase their savings in anticipation that the government will increase taxes in the future to meet rising deficits and pay off accumulated debt. These economic agents respond to budget deficits by accumulating wealth further rather than increasing their spend- Equation (1) represents the national income from the perspective of total expenditure. National income can also be expressed in terms of total income as in Equation (2). By definition, nations dispose of their income (GDP) for the period "t" as consumption (C), savings (S), or taxes (T). Accordingly, As total expenditure in the economy equals total income, we obtain Equation (3).
Breaking down total savings in an economy (S) into private (Sp) and government (Sg) savings yields Equation (5).
(T -G) + (Sp + Sg -I) = NX (5) Since private savings are the part of disposable income saved rather than consumed, we obtain Equation (6).
On the other hand, government savings are equal to the difference between government revenues and government expenditures, such that: Using the decomposed forms of Sp and Sg [Equations (5) and (6)] and then substituting into Equation (5), we re-write Equation (5) in the following form: After making necessary arrangements in Equation (8), we obtain Equations (9) and (10).
Equation (10)  13 direction by about the same amount, and thereby we can imply that the two balances are twinned or directly interrelated. In this case, a deterioration of budget balance leads to deterioration in the trade (or current account) balance. If private savings do not equal the investment balance, i.e. the shortfall of domestic saving as compared with domestic investments (Sp < I) and budget balance is negative (T < G), we are faced with triple deficits, where the sum of the two domestic deficits is equal to the trade deficit. From the policy perspective, this implies that if budget deficits exist along with a private savings-investment gap, triple deficits are unavoidable.
Equation (10) by itself says nothing about the causes and interconnections of the deficits. The commonly accepted view is that budget deficits are the fundamental cause of twin or triple deficits and that the cure is to reduce budget deficits [see e.g. Feldstein (1992), Ahmed and Ansari (1994), Khalid and Guan (1999), IMF (2011), Tang (2014)]. Here, twin or triple deficits are seen as a consequence of government overspending and all three deficits should cease to exist when the government cuts spending.

Empirical background
To our knowledge, Milne (1977) produced the earliest study of the relationship between fiscal deficits and trade deficits. Examining 38 countries, she concludes that fiscal deficits are an important factor in determining trade deficits. Several subsequent studies, including Bernheim (1988), Miller and Russek (1989), Abell (1990), and Latif-Zaman and DaCosta (1990), concentrated exclusively on the US in examining the validity of the twin deficits hypothesis. The empirical findings of these studies generated results in favor of the validity of the twin deficits hypothesis.
In 1990, notably a time of recession, work on the twin deficit hypothesis again expanded to other countries. This new wave of studies even considered the validity of the hypothesis for country groups (e.g. OECD and the EU). Studies deserving mention include Ahmed and Ansari (1994) for Canada, Magazzino (2012) for Italy, Bostancı and Tunç (2002) and Kıran (2011) for Turkey, Kim and Kim (2006) for South Korea, Baharumshah and Lau (2007) for Thailand, Sobrino (2013) for Peru, Mudassar et al. (2013) for Pakistan, Marinheiro (2008) and El-Baz (2014) for Egypt, Ogbonna (2014) for South Africa, Salvatore (2006) for the G-7 countries, Baharumshah et al. (2006) for the ASEAN-4 countries, Afonso et al.

Hüseyin Şen and Ayşe Kaya
Are the twin or triple deficits hypotheses applicable to post-communist countries? 14 The empirical findings of the studies are mixed on support for the twin deficits hypothesis. Studies confirming the validity of the hypothesis include Rosensweig and Tallman (1993), Ahmed and Ansari (1994) for Canada, Bostancı and Tunç (2002) for Turkey, Baharumshah and Lau (2007) for Thailand, Holmes (2010) for the US, and Vamvoukas (2010) for Greece. Studies finding no supporting evidence include Dewald and Ulan (1990) and Rahman and Mishra (1992)  Other studies find highly disparate results that change according to the statistical techniques used, the length and timing of the observation period, as well as country-specific features. For instance, Miller and Russek (1989) find different results for the same sample countries, whereas Khalid and Guan (1999) assert that the twin deficits hypothesis is only valid for developing countries. Ratha (2012) argues that, while the Keynesian proposition holds in the short run, the Ricardian equivalence proposition is present in the long run. A relatively recent study by Eldemerdash et al. (2014) found different results for Arab countries that produce oil and those that do not. Their findings suggest a positive relationship between fiscal and external balances for oil-producing countries, but no similar relationship between two non-oil countries.
Among the most interesting of all the studies here is that of Kim and Roubini (2008), who argue that in the case of the US, cuts in budget deficits increase current account deficits, resulting in twin divergences. In other words, budget deficit shocks in the US case tend to improve the current account and depreciate the real exchange rate in the short run.
Perhaps due to a lack of data, academic economists and other researchers have neglected the twin deficits hypothesis in case of the post-communist countries. Indeed, there are only a handful of studies analyzing the twin deficits hypothesis for these countries. To our knowledge, with the exception of a few single-country studies, the big-picture works are limited to the studies of Fidrmuc (2003), Gurgul and Lach (2012), Aristovnik and Djurić (2013), Tosun et al. (2014), and Gabrisch (2015). In all cases except Fidrmuc (2003), these studies yield results that favor the Ricardian view.
As for the triple deficits hypothesis, the existing literature in this matter is indeed scarce.
To our knowledge, the relevant studies are Szakolczai (2006), Akıncı andYılmaz (2012), Şen et al. (2014), and Tang (2014). All offer evidence favoring the validity of the triple deficits hypothesis, but all are also single-country studies. 8 Moreover, the study of Szakolczai (2006) is not empirical. To sum up, many studies have attempted to establish a nexus between the budget balance and the trade (or current account) balance, but no clear consensus has emerged.
While some studies such as Latif-Zaman and DaCosta (1990), Baharumshah and Lau (2007), and Xie and Chen (2014) assert that budget deficits and current account deficits are "twins", "identical twins", or even "reverse twins" [Anoruo and Ramchander (1998), Kim and Kim (2006), El-Baz (2014)], others such as Enders and Lee (1990), and Kim and Roubini (2008) We employ annual data on budget balance, private savings-investment balance, and trade balance to construct our three variables used in bootstrap panel Granger causality analysis.
Our data set is restricted by the availability of comparable data, especially at the onset of transition; we limit the scope of our data to the period 1994 to 2012 and six of the larger postcommunist economies (Russia, Poland, Ukraine, Romania, the Czech Republic, and Hungary).
All the data related to the variables have been directly taken from the relevant sources in proportion to GDP. The data on budget balance (cash surplus/deficit basis and at general government level) are taken from the World Bank World Development Indicators Database. The data for Poland and Russia 1994-2000, Ukraine 1994-1998, Romania 1994-2001, and Hungary 1994 are extracted from the respective IMF country reports. As can be 8 See Appendix A for details.

Hüseyin Şen and Ayşe Kaya
Are the twin or triple deficits hypotheses applicable to post-communist countries?
16 seen in column 1 of the Table 1 below, most of our sample countries ran sizable budget deficits during the observation period. The third approach, proposed by Kónya (2006), allows both heterogeneity and cross-sectional dependence to be taken into account.
This study employs the approach proposed by Konya (2006), which has three advantages over the first two approaches. First, this approach is based on a SUR estimation that allows us take into account cross-sectional dependence across countries. Second, it does not require the joint hypothesis for all members of the panel because it is based on a Wald test with country-specific bootstrap critical values. Finally, it requires no pre-testing for panel unit roots or co-integrating relationships. A general drawback of the unit root test is its low testing power, which can lead to incorrect judgments with regard to co-integrating relationships.
Here, we take into account the possible existence of direct relationship between budget deficits and trade deficits, and/or among budget deficits, private savings-investment deficits, and trade deficits. For this purpose, we employ the bootstrap Granger causality approach developed by Kónya (2006) where NXB, BB, and SIB denote trade balance, budget balance, and private savings-investment balance, respectively. N is the number of countries of panel (I = 1, 2, 3,…, N), t is the time period (t = 1, 2, 3, …, T), and "l" is the lag length. The error terms, ε 1Nt , ε 2Nt and ε 3Nt , are supposed to be white-noises (i.e. they have zero means, constant variances and are individually serially uncorrelated) and may be correlated with each other for a given country.
We assume that NXB, BB and SIB are stationary or cointegrated so, depending on the time-series properties of the data, they may denote the level, first difference or some As the results from our Granger causality test may be sensitive to lag structure, determining optimal lag length(s) is crucial as to the robustness of the findings. To determine optimal lag structure, we follow Kónya's approach, whereby maximal lags are allowed to vary across variables, but remain the same across equations. We estimate the system for each possible trinity of p 1 p 1 p 1 , p 2 p 2 p 2 and p 3 p 3 p 3 by assuming from one to four lags, and then choose the combinations which minimize the Akaike Information Criterion (AIC) and Schwartz Information Criterion (SIC).

Empirical results and discussion
Taking into account cross-sectional dependence and country-specific heterogeneity in empirical analyses is essential as our sample countries are highly integrated and highly globalized in their economic relations. If cross-sectional dependency does exist, the use of the seemingly unrelated regressions (SUR) approach should be more efficient than an ordinary  Zellner (1962) should be more reliable than those obtained from county-specific OLS estimations. The Monte Carlo experiment by Pesaran (2006) emphasizes the importance of testing for the cross-sectional dependence in a panel data study. It also illustrates the substantial bias and size distortions that arise when cross-sectional dependence is ignored.
A further issue to decide is whether to treat slope coefficients as homogenous to impose the causality restriction on the estimated parameters. The causality from one variable to another variable by imposing the joint restriction for the panel is the strong null hypothesis and the homogeneity assumption for the parameters is unable to capture heterogeneity due to country-specific characteristics.
Thus, we start our empirical analysis with testing for cross-sectional dependency, followed by slope homogeneity across countries. We then select the appropriate panel causality method for determining the direction of causality between budget balance, private savings-investment balance, and trade balance in our six post-communist countries.
To investigate the existence of cross-sectional dependence, we implement four tests: the LM, CDlm, CD and LMadj tests. 9 The test results are presented below in Table 2. As shown from the table, the null hypothesis of no cross-sectional dependence across the countries is strongly rejected at 1% level of significance, implying that the SUR method is more appropriate than country-by-country OLS estimation. The findings in Table 2   The results reported in Table 3 suggest that there exists a significant, but negative, Granger causality running from budget deficits to trade deficits at 10% level of significance only for Poland and Romania. We do not find any significant relationship from budget deficit to trade deficit for Russia, Ukraine, the Czech Republic or Hungary.
On the other hand, Table 3 indicates that there is a significant and positive Granger causality running from trade deficit to budget deficit at 10% level of significance for three countries (Russia, Romania, and Hungary).
The possible explanation of these findings might be that widening trade deficits may have decreased aggregate demand in these countries, resulting in a reduction in output and an increase in unemployment.    We find no evidence in favor of the twin/triple deficits hypothesis for the countries considered. This means that there is no Granger causality running from budget deficits to trade deficits, and no Granger causality was found running from budget deficits and private

Hüseyin Şen and Ayşe Kaya
Are the twin or triple deficits hypotheses applicable to post-communist countries?
24 savings-investment deficits to trade deficits. Based on these findings, we conclude that the Ricardian equivalence proposition of the twin and triple deficits hypotheses holds for these six post-communist countries over the observation period.
Overall, it appears that budget deficits and trade deficits are causally independent variables in our sample. It is worth mentioning that our findings are broadly parallel to the empirical findings of several earlier studies, including Dewald and Ulan (1990) and Rahmann and Mishra (1992)  There may be several explanations for these findings. First, the existence of an output gap may be a factor. With some minor exceptions, all sample countries in the first decade of the transition displayed actual output levels in proportion to potential GDP well below their potential levels. This suggests the existence of an output gap. If so, increases in aggregate demand following expansionary fiscal policies may have been masked by increases in domestically produced goods and services, rather than through imports. The second plausible explanation may be a substantial exogenous increase in investment. These investment booms might have been generated through foreign technical assistance, technological innovation, successful market-oriented reforms, or a combination of all three. Successfully implemented free-market reforms, in particular, would have conferred the economic benefits of growth, enhanced trade competitiveness, and inflows of much-needed foreign capital. Third, there was the external assistance these countries received at the earlier stages of transition from international financial organizations such as the IMF, World Bank, as well as bilateral donors. Moreover, the countries that have already joined the EU all received substantial financial and technical supports from the EU throughout their accession processes. Finally, Russia and Ukraine are commodity-exporting countries 13 and so their export earnings and demand depend mostly on external factors. Over the observation period, there were several currency devaluations that effectively restrained imports to Russia and Ukraine.
This study broadly relates to possible explanations for the divergent results among the many empirical papers. The different findings may largely arise from the differences in methodology and data. In some previous studies, the possibility of structural breaks was ignored in the series. In others, the variables considered were treated as integrated of order "one", referring to the existence of a unit root. Further, analysis of data sets that focus on a short period of time may not yield reliable evidence. Lack of longer-term data for countries, as in the case of this study, limits the possibility for clear-cut, differentiated results. Not to put a fine point on The association between fiscal balance and current account balance is limited. Accordingly, an improvement in the fiscal balance of 1% of GDP is found to improve the current account balance by 0.2-0.3 percentage point of GDP. The association between the two is as strong in emerging and low-income economies as in advanced countries; significantly stronger in country-years where output is above potential than in cases where below potential.
Kıran (2011)  Three tests are available to investigate the existence of cross-sectional dependence: the Lagrange multiplier test statistic of Breusch and Pagan (1980) for cross-sectional dependence, and the cross-sectional dependence test statistics of Pesaran (2004), which are based either on Lagrange multiplier or pair-wise correlation coefficients.
The Lagrange Multiplier (LM) test developed by Breusch and Pagan (1980) requires estimation of the following panel data model: for I = 1,2, 3, …, N; t = 1,2, 3, …, T , where I is the cross section dimension; t is the time dimension; X it is kx1 vector of explanatory variables, and α I and β I are the individual intercepts and slope coefficients allowed to differ across states.
In the LM test, the null hypothesis of no cross-sectional dependence H 0 : Cov (μ it , μ jt ) = 0 for all t and I ≠ j is tested against the alternative hypothesis of cross-sectional dependence H 1 : Cov (μ it , μ jt ) ≠ 0 for at least one pair of i≠ j.
For testing the null hypothesis, the LM test statistic for cross-sectional dependence of Breusch and Pagan (CD BP ) is given as: where ρ � ij 2 is the estimated correlation coefficient among the residuals obtained from individual OLS estimation of Equation (1). Under the null hypothesis, the LM statistic has an asymptotic chi-square distribution with N (N-1)/2 degrees of freedom. Pesaran (2004) indicates, however, that the CD BP test has a drawback when N is large, implying that it is not applicable when N→∞. To overcome this problem, the following LM statistic for cross-sectional dependence (CD LM ) was developed by Pesaran (2004). The CD LM statistic is given as: Under the null hypothesis of no cross-sectional dependence with T→∞ and then N→∞, CD LM asymptotically follows a normal distribution. Unfortunately, the CD LM test is likely to indicate size substantial distortions when N is large relative to T. Pesaran (2004) therefore proposes an alternative test for cross-sectional 14 These definitional equations heavily borrow from Şen et al. (2015).
dependence (CD) that can be used where N is large and T is small. The CD statistic is calculated as follows: Pesaran (2004) states that under the null hypothesis of no cross-sectional dependence with T → ∞ and N → ∞ in any order, the CD test is asymptotically normally distributed.  qualify this by noting that when the population average pair-wise correlation is zero, the CD test has less power. Therefore, they propose a bias-adjusted LM test that uses the exact mean and variance of the LM statistic. The bias-adjusted LM statistic is calculated as follows: where u Tij and v Tij 2 are the exact mean and variance of (T-k) ρ � ij 2 , which are provided by . Under the null hypothesis of no cross-sectional dependence with T → ∞ first followed by N → ∞, the results of the CD adj test follow an asymptotic standard normal distribution.
The standard F-test is the most widely used way to test the null hypothesis of slope homogeneity H 0 : β I = β for all I against the hypothesis of heterogeneity H 1 : β I ≠ β j for a non-zero fraction of pair-wise slopes for i≠j. This requires that the explanatory variables are strictly exogenous and the error variances are homoscedastic. To relax the assumption of homoscedasticity in the F-test, Swamy (1970) developed a slope homogeneity test that examines the dispersion of individual slope estimates from a suitable pooled estimator.  state that both the F-test and Swamy's test require panel data models where N is relatively small compared to T. Therefore, they propose a standardized version of Swamy's test (hereafter, ∆ � test) for testing slope homogeneity in large panels. The ∆ test is valid when (N, T) → ∞ without any restrictions on the relative expansion rates of N and T when the error terms are normally distributed. Swamy's statistic can then be modified as: where β � I is the pooled OLS estimator; β � WFE is the weighted fixed effect pooled estimator of the Equation (1); M τ is an identity matrix of order T and σ � i 2 is the estimator of σ i 2 .