Cross-Country Linkages in Europe
A Global VAR Analysis
  • 1 0000000404811396https://isni.org/isni/0000000404811396International Monetary Fund

Contributor Notes

Author’s E-Mail Address: ysun@imf.org (corresponding author), fheinz@imf.org, gho@imf.org

This paper uses the Global VAR (GVAR) model proposed by Pesaran et al. (2004) to study cross-country linkages among euro area countries, other advanced European countries (including the Nordics, the UK, etc.), and the Central, Eastern and Southeastern European (CESEE) countries. An innovative feature of the paper is the use of combined trade and financial weights (based on BIS reporting banks’ external position data) to capture the very close trade and financial ties of the CESEE countries with the advanced Europe countries. The results show strong co-movements in output growth and interest rates but weaker linkages bewteen inflation and real credit growth within Europe. While the euro area is the dominant source of economic influences, there are also interesting subregional linkages, e.g. between the Nordic and the Baltic countries, and a small but notable impact of CESEE countries on the rest of the Europe.

Abstract

This paper uses the Global VAR (GVAR) model proposed by Pesaran et al. (2004) to study cross-country linkages among euro area countries, other advanced European countries (including the Nordics, the UK, etc.), and the Central, Eastern and Southeastern European (CESEE) countries. An innovative feature of the paper is the use of combined trade and financial weights (based on BIS reporting banks’ external position data) to capture the very close trade and financial ties of the CESEE countries with the advanced Europe countries. The results show strong co-movements in output growth and interest rates but weaker linkages bewteen inflation and real credit growth within Europe. While the euro area is the dominant source of economic influences, there are also interesting subregional linkages, e.g. between the Nordic and the Baltic countries, and a small but notable impact of CESEE countries on the rest of the Europe.

I. Introduction 1

While there is a broad recognition that countries in Europe are closely linked through trade and financial channels, the mechanism of how such channels transmit shocks, and how real and financial sectors interact as the shocks are transmitted are less clear. These questions have drawn active interest from researchers in recent years. This paper tries to provide some insight on these issues by using the Global VAR (GVAR) model to account for such regional interdependencies, with a strong focus on linkages between advanced European and CESEE countries. The GVAR model is proposed by Pesaran, Schuermann and Weiner (2004, henceforth PSW) and further developed in Dées, di Mauro, Pesaran, and Smith (2007, henceforth DdPS).

The economic and financial linkages between the European economies (advanced and emerging) have increased significantly over the past two decades. Following the collapse of the Soviet Union in the early 1990s, trade and financial ties between Central Europe and Southeastern Europe (CESEE) and advanced Europe strengthened rapidly. The EU accession process has been one of the main drivers of closer east-west integration. The establishment of the euro has further cemented integration of the euro area member countries. Moreover, some of the CESEE countries joined the euro in the late 2000s.

Trade between Western Europe and CESEE countries has increased rapidly: by 2011, Western Europe was the destination of 75 percent of exports from CESEE, while 68 percent of imports into CESEE were from Western Europe. This largely reflects the fact that CESEE has become both a part of the production chain of, and new markets for western European producers. Exports from CESEE also grew during the period.

Financial integration also proceeded apace. Western European banks had gained a dominant position in the banking systems of most CESEE countries: the share of foreign banks (in terms of assets of local banking system) in 2011 exceeded 70 percent in most countries in the region, with the notable exception of the European CIS countries and Turkey.2 As a result, Western European banks and companies have become the main foreign source of capital in terms of bank funding and FDI for CESEE countries.

For the CESEE countries, these close linkages brought clear benefits, but also carried risks Trade links and financial capital inflows from advanced Europe made it possible for the CESEE countries to boost their growth potential faster than they otherwise could achieve shortly after they left the Soviet bloc. Growth for this region before the recent crisis was very impressive. Real per capita income increased by 4 percent annually in the period of 1995–2007 for the CESEE region, much higher than most other emerging market regions, with the exception of China and India. The close linkages also carried risks. As CESEE economies rely closely on Western Europe for capital and trade, economic slowdowns and financial market turmoil in Western Europe quickly spill over to CESEE countries. When Western European parent banks came under pressure in the fall of 2008, this triggered a sudden stop of capital flows to the region, which contributed to a deep crisis. 3 More recently, the CESEE region has also suffered from spillovers from the euro area crisis. CESEE regional growth has been declining since mid-2011, following the recession in the euro area.

In this paper, we attempt to explore the regional linkages between Western Europe and CESEE using the GVAR framework. The main innovation of the paper is that we aim to capture both trade and financial linkages. Out study also has slightly different country coverage and the key variables studied compared to similar regional studies. A key innovation of this paper is that we use composite weights to reflect both trade and financial linkages between the countries of Europe. As explained later, a key step of GVAR analysis is to construct, for domestic variables of each country or region in the system, corresponding foreign variables, usually a weighted average of corresponding variables of its partners. For example, if the variable of interest is real GDP of country A, then its corresponding foreign variable (foreign real GDP) is constructed as a weighted average of the real GDP of its partners. The weighting scheme usually reflects the strength of economic ties of a particular country with its foreign partners. In the literature, the selection of weights often varies. Many GVAR studies - including PSW, DdPS (2007), Galesi and Lombardi (2009), and Feldkircher and Korhonen (2012) use weights based on trade flows; Vansteenkiste (2007) uses geographical distance based weights, whereas Hiebert and Vansteenkiste (2007) adopt weights based on sectoral input-output tables across industries. Galesi and Sgherri (2009) use financial weights based on bank lending data across countries. By using weights that reflect both trade and financial flows across countries, the results can better capture the rich transmission channels that exist among countries and regions in Europe.

In the paper, we focus on co-movements between output growth, inflation, real credit growth, and long-term interest rates. The objective is to show how real or financial shocks are propagated across countries within Europe. The variables in our model are real GDP growth, inflation, real credit growth, and long term interest rates. The country sample includes all Western European countries and also a fairly representative set of CESEE economies.

The paper focuses on a larger set of CESEE countries than similar studies. For example, Galesi and Sgherri (2009) present results on financial spillovers in Europe that includes a smaller group of CESEE countries. Their paper focuses on the relevance of international spillovers following a historical slowdown in U.S. equity prices in 2008, with a model that contain equity prices, GDP, interest rates, and credit to corporations. Galesi and Lombardi (2009) focus on international inflation linkages in a dataset that includes a few European countries (some of which from CESEE).4

The model has yielded interesting results. There are strong co-movements in output growth, interest rates, and somewhat weaker co-movements in inflation and credit growth. Shocks to euro area output growth reverberate strongly across European countries including Nordic countries and CESEE countries. Shocks to the UK long-term interest rate have a strong impact on long term interest rates in the euro area, the Nordic countries, but weak impact on CESEE countries. The impact of the interest rate on output is felt in all countries. Shocks to euro area inflation have a weak pass through to CESEE countries and other western European countries5; so is the impact of shocks to credit growth in the euro area on credit growth in CESEE.6 There are also interesting sub-regional ties. For example, the Baltic States appear to be very sensitive to shocks from the Nordic countries, which is not surprising given their very close financial and trade linkages with the Nordic countries. Shocks to central Europe countries appear to have a small impact on Western Europe. The impact of shocks to the Baltic countries on other countries is negligible (except for the Nordics and Russia).

The rest of the paper is structured as follows: Section 2 describes the analytical basics of the Global VAR framework and the data used in the analysis. Section 3 presents the estimation results. Section 4 analyzes country-specific and regional shocks by using the generalized impulse response functions and generalized forecast error variance decomposition from the GVAR model, and Section 5 concludes.

II. The GVAR Model – Model Structure and Data set Used

A. GVAR Model—A Non-Technical Summary

The GVAR model as developed in PSW and DdPS is a multi-country model. As the name suggests, the model is based on VAR models of individual countries. Its structure, however, makes it a good tool to study inter-country linkages for the chosen group of countries.

The main benefit of a GVAR model compared to individual country specific VAR model is that it allows full interactions of every country in the studied group to be captured explicitly, and in two aspects. First, the interactions among countries through trade, finance, or other channels are reflected in the construction of foreign variables specific to each individual country (see more on this below). Second, the estimation of a single, often fairly large, VAR model based on individual VAR models makes it possible to demonstrate how shocks specific to an individual country affect other countries, as the model is estimated globally at the group level.

A GVAR model is constructed in three stages. First, for each country, the conventional VAR model is extended with the addition of a set of (weakly exogenous) foreign variables. These variables are usually constructed as weighted averages of same type of variables of all its trading or financial partners. For example, if GDP is one of the variables in a country’s original VAR model, then a foreign GDP variable - e.g. denoted as GDP* - will usually be constructed as a weighted average of GDP of the rest of the countries in the group. The choice of weights, as discussed below, should in principle, reflect the trade, financial, or geographical relationships among countries in the group. With the VAR models thus extended, the individual country models are estimated in a second step. The lag structure and the selection of foreign variables vary country by country, and this flexibility allows the country VAR to be modeled more accurately. In a third step, all individual country’s VAR models are collected and estimated as a single VAR model, and the dynamic properties of the model is used to analyze how shocks are propagated across countries. The GVAR model allows a sub-group of countries to be model together as a region, so when discussing the GVAR model structure, country and region is interchangeable. The technical detail of the GVAR model is summarized in Appendix I.

B. The Data Set (2000–2011)

Data are collected for 33 European economies, including both Western European and CESEE countries. Western European countries include: all the Western European euro area countries, the Nordic countries—Denmark, Norway, Sweden, and other advanced economies—the United Kingdom (UK), Switzerland, Iceland, Israel.7 The CESEE countries include: Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Russia, Slovakia, Slovenia, and Turkey.8

To control the dimensions of the GVAR to make it manageable, and also to sharpen the focus of the interactions of advanced European countries with CESEE, the Western European countries are grouped into three groups (see Table 1). The first group is the Western European euro area countries (“EURO-West” in the tables and charts below) which includes all euro area countries except Finland—included in the Nordic group, and Estonia, Slovakia and Slovenia which are modeled individually as other CESEE countries.9 The second group includes four Nordic countries (“NORD”). The UK, Switzerland, Iceland, and Israel10 constitute the third group (“ADV”).

Table 1.

Countries and Regions in the GVAR Model

article image

The models are estimated over the period 2000Q2-2011Q4. The variables include real GDP growth

(dyit), inflation (πit =

Pit

pi,t–1), long term interest rate (rit) (definitions vary country by county), and real credit growth (dCRit).11 Data sources are described in detail in Appendix Table A1. More specifically,

dyit=400*(ln(GDPit)ln(GDPit1)),pit=ln(CPIit)rit=14*ln(1+RitL/100),dCRit=Δln(CRitCPI)

where

GDPit is (seasonally adjusted) real Gross Domestic Product, CPlit is the Consumer Price Index (for mos is long-term interest rates (which may be government bond rate or bank lending rate depending on countries), for country i and period t.12 Before constructing the country specific foreign variables dyit*, πit*, rit*, and dCRit*, a key step is to build appropriate weights. These weights are calculated in this paper by using the trade flow and cross-border bank exposure data. The sample also includes the oil price which is treated as an exogenous variable for all countries except for the EURO-West group (the role of the oil price variable is to control for the global business cycle.)

Since the construction of the foreign variables is based on the weight matrix W= (wij), it is important that the weights should reflect as close as possible the underlying economic linkages among countries. As noted in DdPS, “The weights, … could be used to capture the importance of country j for country ith economy. Geographical patterns of trade provide an obvious source of information for this purpose and could also be effective in mopping up some of the remaining spatial dependencies.” In fact, the choice of weights affect the quality of the foreign variables which is a critical factor determining whether GVAR is more advantageous than traditional VAR.

We build the weights by combining bilateral trade and financial flows. Compared to similar GVAR studies, e.g. Galesi and Sgherri (2009) which use financial weights based on bank lending data only, or PSW and DdPS which uses just trade weights, we believe the combined trade and financial weights capture more accurately the trade and financial linkages between CESEE and advanced Europe.

The weights are calculated as follows. First, for each country i, bilateral annual trade flows (including both exports and imports) with its trading partners are collected.13 Then the financial data are collected. The financial data uses the external positions of international banks as published in the Bank for International Settlements (BIS) locational banking statistics. 14 For CESEE countries, as noted earlier, the funding from advanced Europe—mostly channeled through subsidiaries of advanced European banks were one of the driving forces of the boom and bust cycle. The sum of trade flow and foreign exposure positions are then used to derive the weight matrix. For the model estimated below, fixed weights based on the average weights for the period 2005–11 are used (see Table 2).15 Given that the recent crisis has resulted in fairly large swings in the trade weights and BIS exposure data in the region, the choice of fixed weights averaged across the cycle would hopefully reflect better the normal relations among countries. We have also used time varying weights for the study, and the results are generally qualitatively similar, and are available upon request.

Table 2.

Weight Matrix (average of weights for the period 2005-2011) 1/

article image

Bilateral weights are shown in columns and sum up to one. Weights are average annual weights for the period of 2005-2011. Weights for specific year are calculated based on the total of trade flow and BIS reporting banks’ external position between countries for that year. Pink numbers indicate they are larger than zero but smaller than 0.05.

Table 3.

Trade Weight Matrix (average of weights for the period 2005-2011) 1/

article image

Bilateral weights are shown in columns and sum up to one. Weights are average annual weights for the period of 2005-2011. Weights for specific year are calculated based on the trade flow between countries for that year. Pink numbers indicate they are larger than zero but smaller than 0.05.

Table 4.

Financial Weight Matrix (average of weights for the period 2005-2011)

article image

Bilateral weights are shown in columns and sum up to one. Weights are average annual weights for the period of 2005-2011. Weights for specific year are calculated based on the BIS reporting banks’ external position between countries for that year. Pink numbers indicate they are larger than zero but smaller than 0.05.

Given that the financial linkages are generally between advanced Europe and CESEE rather than among CESEE countries themselves, there is a significant difference in the trade weights and financial based weights (see Table 3 and 4).16 In fact, the financial weights accentuate the pattern shown in the trade weights. For example, Euro-west has a very high share in terms of financial weights with CESEE countries, while Nordic also has a very high financial share with the Baltics. Both these shares are higher than trade shares. While there is some intra-regional trade among CESEE countries, the financial links among CESEE are not strong, with only Turkey having some financial links with other CESEE countries. Euro countries and Nordic countries have the most financial exposure towards CESEE countries. On the other hand, countries in the ADV group have very large financial exposure in the EURO-West group countries and vice versa. They also have strong exposure in Russia and Turkey, but less so in the Nordic countries.

Clearly the cross-country relationships are better revealed when both the trade and financial linkages between advanced European countries and CESEE are considered together. Either one studied alone will not give a full picture. The different trade and financial linkages provides justification for combining these weights in the GVAR setup.

Within the group of CESEE countries, inter-linkages between individual countries are usually very low (below 5% in most cases) in spite of the geographical proximity in many cases. There are only a few exceptions with somewhat larger bilateral links. For example, the Czech Republic is an important partner for the Slovak Republic with a weight from the Czech Republic to Slovakia at 11%, but the influence is smaller the other way round—the weight from the Slovak Republic to the Czech Republic is only 5%, though it is still higher than most other countries. Also, Russia is an important partner for Lithuania (weight at 16%) and Turkey (10%). The Baltic countries trade closely with each other (weights between Baltic countries are close to or above 10%).17

The weight matrix itself yields interesting information on cross-country linkages. It shows the dominant role of the euro area as the main partner for the rest of the countries. The weight for the euro area as a foreign partner ranges between 64% - 91% for all countries in the sample, except for the Baltic countries. For the Baltic countries, the Nordic countries (in this study including Finland) are clearly the most important partners, with their joint weights ranging between 39%-67% for the three countries, exceeding the influence from the euro area.18 The link between the rest of the advanced economies (“ADV” group) and CESEE countries is relatively weak: its weights are generally below 5%, except for Russia and Turkey where ADV’s weights are 11% and 17% respectively.

III. Estimation of the GVAR Model

A. Specification and Estimation of the Country-Specific Models

We start by assuming that foreign variables are weakly exogenous, and the VAR relationships (i.e. coefficients of individual country models) are stable over time. The result of unit root tests and of weak exogeneity tests are shown in Appendix II, and the issue of structural breaks is discussed later after the initial model is estimated.

Obvisouly no single structure can be imposed across the countries given both data constraints and different country circumstances. In fact, as noted earlier, the GVAR approach has the advantage to handle flexibly different specifications for different countries. The foreign inflation variable is excluded from entering the model for most of the countries except for Lithuania since they are I(0) (see Appendix II, and Appendix Tables A3-A5 for the unit root test results). Also since foreign interest rates are I(2) in ADV, Croatia, the Czech Republic, Hungary, Poland, Romania, Russia, Slovakia, Slovenia, and Turkey, they are excluded from entering the VARX model in those countries. Overall, most of the countries have the same set of domestic variables, except for a few countries where the interest rate is not included (Estonia, Croatia, Latvia, Lithuania, Slovenia, and Turkey)19. The interest rate for Turkey is more volatile and the VARX including the interest rate with the chosen domestic variables yielded a poor fit for interest rate. To avoid compromising the fit of the GVAR model, it is not included in Turkey’s model.

After individual country models are specified, the lag length of the VARX(p, q) model is selected using Akaike Information Criterion (AIC) with a maximum length set at three for domestic variable (pmax) and two for foreign variables (qmax) to control the total dimension of the system. In the end, a majority of the domestic variables have a lag order of two. Then we proceed to conduct the co-integration analysis with a specification of unrestricted intercept in the co-integration relations.

The results of the lag order selection and co-integration tests are shown in Appendix Table A6. The co-integration results are based on trace statistic at the 95 significance level, with critical values from MacKinnon, Haug, and Michelis (1999). The trace statistic has better small sample power compared to the maximal eigenvalue statistic. The diagnostic test results for all equations are given in Appendix Table A7. With the exception of Turkey which the original co-integration analysis shows a full rank co-integration matrix, all other countries have reasonable results.

B. Testing for Structural Breaks

We also test for structural stability of the model. Following DdPS, a battery of parameter constancy tests are carried out. The test is mainly on the structural stability of the short-term coefficients, rather than the long-run coefficients which is unlikely to be feasible given the data constraints, as pointed in DdPS. Nevertheless, the stability of short-run coefficients matters more to the transmission of shocks across countries which is the main interest of this study.

The tests include Ploberger and Krämer’s (1992) maximal OLS cumulative sum (CUSUM) statistic, denoted by PKsup and its mean square variant PKmsq; tests for parameter constancy against non-stationary alternatives proposed by Nyblom (1989), denoted by R. They also include several sequential Wald-type tests of a one-time structural change at an unknown change point: the Wald form of Quandt’s (1960) likelihood ratio statistic (QLR), the mean Wald statistic (MW) of Hansen (1992) and Andrews and Ploberger (1994) and the Andrews and Ploberger (1994) Wald statistic based on the exponential average (APW). The heteroskedasticity-robust version of the above tests is also presented.

Table 5 summarizes the results of the tests by variable at the 5% significance level. The results show that structural instability is not a serious concern for the sample, although results vary by tests and by variables.20 These are quite encouraging results given that the sample period covers a very severe boom and bust for CESEE and also a crisis for advanced Europe where economic variables have undergone significant fluctuations. Looking into the details, we note, for example, the two PK tests do not reject structural stability in any of the cases.

Table 5.

Number of rejections of the null of parameter constancy per variable across the country-specific models at the 5% level

article image
Note: Percent of rejection in parenthesis. The test statistics PKsup and PKmsq are based on the cumulative sums of OLS residuals, SR is the Nyblom test for time-varying parameters and QLR, MW and APW are the sequential Wald statistics for a single break at an unknown change point. Statistics with the prefix ‘robust’ denote the heteroskedasticity-robust version of the tests. All tests are implemented at the 5% significance level.

For the other three types of tests, both the constant variance version and the heteroskedasticity robust version of the tests seem to reject only a small share (4-10 percent) of all possible cases. Together, the three Wald-type tests suggest that a slightly higher probability of breaks in error variances than parameter coefficients.

C. Contemporaneous Effects of Foreign Variables on their Domestic Counterparts

We present in Table 6, the contemporaneous effects of foreign variables on their domestic counterparts. For example, for CESEE countries, a 1% increase in foreign output growth in a given quarter leads to an average 0.4% increase in domestic output growth within the same quarter. For credit growth, significant elasticity is observed in Hungary, Lithuania, Romania, Slovakia, and Turkey for CESEE, and ADV and NORD in advanced Europe. For a few countries where foreign inflation and interest rates are directly included in the model, there are high contemporaneous effect as well. For example, there is a high elasticity between domestic and foreign inflation, π and π* for Lithuania, and we also observe a significant elasticity between domestic and foreign interest rates, r and r*, for EURO-West and NORD indicating close co-movements of interest rates in these two regions.

Table 6.

Contemporaneous Effects of Foreign Variables on Their Domestic Counterparts

article image
Note: Newey-West’s heteroskedastic-robust t-ratios are given in brackets.

D. Pair-wise Cross-Country Correlations: Variables and Residuals

Next we present results on how idiosyncratic shocks of the individual country models are correlated across countries. A low correlation is one of the main conditions for a well functioning GVAR model. A low correlation would suggest that the cross-dependence of idiosyncratic shocks is “sufficiently” small, therefore we can isolate the impact of country specific, idiosyncratic shocks from other shocks in the dynamic analysis we carry out later.

As suggested in DdPS, a simple diagnostic of the extent to which the country-specific foreign variables have been effective in reducing the cross-country correlations of the variables in the GVAR model could be the simple average pair-wise correlation for the endogenous variables, and those of the associated residuals over the estimation period. A low correlation of the residuals is a strong indication that the GVAR model has been quite successful at capturing the common effects driving the endogenous variables, and the GVAR model should be considered fairly effective in explaining cross-country interdependencies. Consequently, shocks to a domestic variable in an individual model can be considered idiosyncratic.

It can be seen from the statistics shown in Table 7 that the average cross-section correlations are generally high for the level of domestic variables. The results vary somewhat for individual countries and for specific variables. It is interesting to note that cross section correlations of real GDP growth are quite high (averaging 57%), as are interest rates (52%), while cross-section correlations are slightly lower for inflation and real credit growth (with averages at around 27% and 36% respectively). This suggests a significant co-movement for output growth and interest rates, while domestic inflation and credit growth are less synchronized. The cross-section correlation falls as we move from level to first difference, with the reduction most pronounced in real credit growth, interest rates, output growth, and inflation in that order. There are still noticeable correlations in the first differences, as the average correlations range between 20%- 27%, except for real credit growth which is at 6%.

Table 7.

Average Pair-wise Cross-Section Correlations: Variables and Residuals

article image
Note: VECMX residauls are based on co-integrating VAR models with countr-specific foreign variables.

In contrast, correlations of the residuals from the VARX models are very small. The detailed results show that, with few exceptions, the (absolute) correlations are generally less than 10%. For example, for the real output growth equation, correlation of residuals are between -6% and +7%, much smaller compared to the correlations in level which range between 17% and 68%. The relative reduction in correlations for inflation is more modest. Nevertheless, the correlations in the residuals of the inflation equation are not large: they are below 16% for all countries, and for 60 percent of the countries, the correlation is below 6%.

IV. Dynamic Analysis Using Generalized Impulse Response Functions and Generalized Forecast Error Variance Decomposition

In this section we look at the propagation of shocks between different regions of Europe over time, considering both real and financial shocks from the euro area and other parts of Europe. Based on the estimated GVAR model, we conduct a few experiments and analyze the model’s dynamic properties: i.e. the time profiles of the model’s response following a shock (e.g. a shock to a specific variable of a particular country or region) using the generalized impulse response functions. This will give insight on how shocks are propagated across countries.

We organized the type of shocks into three categories, one is real shocks—e.g. direct shocks to real GDP growth in different regions, the second is financial shocks—e.g. shocks to interest rate or credit growth, the third shock includes shocks to inflation. The motivation of experimenting on these shocks is to see how the impact of these shocks—originating in a particular region are felt and transmitted across countries. For example, a few of the shocks experimented below is on the shock to real GDP growth originating in the EURO-West region, the Nordics, the CE region, and even in the Baltic countries. Such experiments can reveal how output, credit growth, and other variables are affected with these shocks.21 On the other hand, motivated by questions such as how does pressure to strengthen western banks’ balance sheet affect credit and output growth in CESEE, or whether an interest rate shock originated in the U.K. (following shocks in the US) will affect interest rate in the rest of the Europe, we also conduct a few experiment on the impact of shocks to credit growth and interest rates in some region. The question of how significant is the inflation pass-through in the region is also investigated as in Galesi and Lombardi (2009).

We use the method of generalized impulse response functions (GIRF) proposed by Koop et al. (1996) and Pesaran and Shin (1998). The GIRF method is an alternative to the orthogonalized impulse response function, and it is invariant to the ordering of the variables and countries in the model. The GIRF approach has the advantage that in the absence of strong prior belief of the ordering of the shocks or countries, it still can provide useful information on the transmission dynamics of the model to individual shocks.

The GIRF is presented over a relatively long period (over 20 quarters). Nevertheless, we generally try to focus on responses over a shorter period, say two years, which is a reasonable time frame for credible results. To avoid lengthy discussion of response for individual country and rather to focus on common pattern of response for countries in the same region for CESEE countries, we recast some countries in CESEE into sub regions: central Europe (“CE” in the tables and figures below) which includes Hungary, Poland, the Czech Republic, the Slovak Republic, and Slovenia; Southeastern Europe (“SE”) which includes Romania and Croatia; the Baltics (“Baltic”) which includes Estonia, Latvia, and Lithuania. The country weight is based on each country’s GDP at PPP price. The two largest economies in the region: Turkey and Russia are not included in any of the aggregates. The regional weight matrix for the GIRF exercise is shown below (Table 8). The region based analysis provides a good summary of response to individual shocks. To keep the length of the main text in control, the detailed country level IRF figures are presented in the Appendix without discussion.22

Table 8.

Regional Weights for the GIRF Exercise

article image
Note: Weights are based on GDP at PPP price.

We also present results of the Generalized Forecast Error Variance Decomposition (GFEVD) which give a picture of how forecast error variance can be traced (though not exclusively) to shocks to different variables (and regions). The GFEVD is based on the GIRF, and is a natural extension to the conventional (orthogonalized) forecast error variance. As GFEVD do not necessarily add to 1 due to contemporaneous correlations among innovations, we present relative contribution, based on rescaled GFEVD, of different variables (from different countries and regions). Such relative contribution can still provide an indication of how important shocks to different variables from a particular region or country are, compared to shocks to other variables from the rest of the region or countries.

In the results that follow, we can see that the impulse responses settle down reasonably well. This is because the estimated GVAR model is stable: the modulus of every eigenvalue of the GVAR is on or within the unit circle (Figure 1). Some of them are complex, which result in oscillating features in the impulse responses. However, bootstrap simulation based on the estimated model generally points to rapidly widening bands for the IRF (not shown in the paper). Therefore, the mean results presented here are only indicative and results over 6-8 quarters should be treated with caution.

Figure 1.
Figure 1.

Modulus of the Eigenvalues of the Estimated GVAR model

Citation: IMF Working Papers 2013, 194; 10.5089/9781484345474.001.A001

1. Spillover of Real Shocks: Shocks to Real GDP Growth

A. Negative Shock to EURO-West Real GDP growth

The first experiment we implement is a 1 percentage point negative shock to the EURO-West group’s real GDP growth which showed large responses in output across the region.23 The generalized impulse response of real GDP growth to the shock is shown in Figure 2.24 The negative shock in the EURO-west results in negative growth for all the countries and regions in the sample. The response generally follows the same profile: there is an immediate impact on growth, the impact then oscillates and dissipates in about 12 quarters. GDP growth in the CESEE countries drops by 0.65–1.25 percentage points (p.p.) in the same quarter.25 This behavior is largely consistent with the GDP growth spillovers observed in 2011 and 2012. The Nordic countries also experience a fairly significant decline in growth rate in the same quarter (about 0.5 p.p.), while the ADV group also similarly impacted - the growth rate declines by about 0.5 p.p. in growth rate.

Figure 2.
Figure 2.

Generalized Impulse Response Function of Real GDP Growth to a Negative One p.p. Shock to Real GDP Growth in the Euro-West Group

Citation: IMF Working Papers 2013, 194; 10.5089/9781484345474.001.A001

Notes: GIRF calculated based on the estimated GVAR model, see paper.Source: Author’s calculations.

The GFEVD results are presented in Table 9. The table shows that shocks to variables in the EURO-West group together have the highest share of contribution to forecast error variance (over half of the rescaled total variance in the first four quarters). Among the EURO area variables, shock to real GDP growth is the dominant source of innovation, although oil price which is treated as an endogenous variable to the euro area is also an important source of shocks. Given that the oil price is the only explicit link of the region with the global economy in our model, it suggests that shocks from outside Europe are important.26 Within each country or region, shock to real GDP growth is the main source of innovation compared to shocks to other variables in the same region, although contribution of shocks to other variables rises over time.

Table 9.

Generalized Forecast Error Variance Decompositions: a Negative One s.d. Shock to EURO-West real GDP Growth

article image
Note: Based on percentage of the k-step ahead forecast error variance of a one s.d. shock to the EURO group’s real GDP growth. Original percentages do not sum to 100 due to non-zero covariance between the shocks, according to Pesaran and Shin (1998). Figures in the tables are rescaled to 100, as suggested by Wang (2002).

B. Shock to Real GDP Growth in Nordic Countries

In contrast to strong region wide responses to output shocks in the Euro-West region, the shocks to real GDP growth in the Nordic region is less severe region wide, but is felt strongly in the Baltics. Given the very close relationship of the Nordic countries to the Baltic countries, we conduct the next experiment on a positive shock to real GDP growth in the Nordic countries. For the Nordic countries, there is a gradual decline in growth rate after the initial impact (see Figure 3).27 As expected, the impact of growth shock from the Nordics to the Baltic countries is quite significant. In the same quarter, the growth rate in the Baltic increases by 1 p.p., and rises and reaches 1.5 p.p. in the third quarter before declining afterwards. While the Nordic economies are only about 10 percent of the size of the EURO-West group, with the close links between the two regions (recall Nordic is only about 15% of the weight for the EURO-West group), there is still some noticeable impact on EURO-West group’s growth. There is an immediate effect of 0.2 p.p. increase in growth rate for the Euro-west group, which rises further to about 0.3 p.p. in the next quarter. The profile of response is similar in other CESEE countries. The same quarter impact to growth for central Europe, Russia, and Turkey ranges is around 0.15—0.2 p.p., and the effect rises further in the next 2-3 quarter before the impact diminishes. The shock to the Nordic region’s GDP growth also has a small impact on the ADV group: the immediate effect is only 0.1 p.p. This reflects the relative distant linkages between the two groups: the Nordic group’s weight is only 6% for the ADV group.

Figure 3.
Figure 3.

Generalized Impulse Response Function of Real GDP Growth to a One p.p. Shock to Real GDP Growth in the Nordic countries

Citation: IMF Working Papers 2013, 194; 10.5089/9781484345474.001.A001

Notes: GIRF calculated based on the estimated GVAR model, see paper.Source: Author’s calculations.

Table 10 presents the GFEVD results for this experiment. With shocks originating from real GDP growth in the NORD group, it follows that such innovation is one of the main source of influence for forecast error variance. Other important sources of influence are shocks to interest rate in the ADV group, oil price shocks, and shocks to output in the EURO-West group. These results suggest that real GDP growth in the Nordic group is sensitive to these external shocks given its close link to the EURO-West group, as well as to the other advanced economy.

Table 10.

Generalized Forecast Error Variance Decomposition: a One s.d. Shock to Real GDP Growth in the Nordic countries

article image
Note: Based on percentage of the k-step ahead forecast error variance of a one s.d. shock to the NORD group’s real GDP growth. Original percentages do not sum to 100 due to non-zero covariance between the shocks, according to Pesaran and Shin (1998). Figures in the tables are rescaled to 100, as suggested by Wang (2002).

C. Shock to Real GDP Growth in Central Europe

As the Central European economies grow in size and importance, a shock to their growth is likely to have a larger impact on its trading partners, including the western European countries. In particular, serving as a market for Western European countries, any shocks in domestic demand in Central Europe could have affected demand for Western European goods and services. In this section and the next, we experiment how shocks to CE countries (which include Czech R., Hungary, Poland, Slovakia, and Slovenia in this study) affect other countries in the region.

As shown in Figure 4, a one p.p. shock to CE group real GDP growth has some discernible impact on its trading partners. Its own real GDP growth declines gradually and settling down in about six quarters after the shock.28 Among the other regions, the Euro-West group sees a 0.1 - 0.2 p.p. increase in growth in the first two quarters, with the impact dissipating quickly afterwards. For the Nordic countries, there is a rise in growth rate of 0.1 p.p. on impact which then declines and dissipates in the following periods. Similar profile is also evident for growth in ADV countries. The impact on CESEE countries is relatively larger and longer lasting. For example, the SE group countries will experience a rise of below 0.15 p.p. in growth rate on impact, and 0.25 p.p. in the second quarter. The impact on the Baltic countries is even more visible: GDP growth is expected to rise by 0.2 p.p. on impact, and over 0.4 p.p. in the second quarter before declining afterwards.

Figure 4.
Figure 4.

Generalized Impulse Response Function of Real GDP Growth to a One p.p. Shock to Real GDP Growth in the Central European countries (Czech R., Hungary, Poland, Slovakia, and Slovenia)

Citation: IMF Working Papers 2013, 194; 10.5089/9781484345474.001.A001

Notes: GIRF calculated based on the estimated GVAR model, see paper.Source: Author’s calculations.

The GFEVD results (Table 11) suggest that CE real output growth is very sensitive to shocks to EURO-West group’s output, oil price shocks, and shocks to ADV group output and interest rate. CE’s domestic inflation and output are main source of domestic shocks.

Table 11.

Generalized Forecast Error Variance Decomposition: a One p.p. Shock to to Real GDP Growth in the Central European countries (Czech R., Hungary, Poland, Slovakia, and Slovenia)

article image
Note: Based on percentage of the k-step ahead forecast error variance of a one s.d. shock to the NORD group’s real GDP growth. Original percentages do not sum to 100 due to non-zero covariance between the shocks, according to Pesaran and Shin (1998). Figures in the tables are rescaled to 100, as suggested by Wang (2002).