Spillover Effects between Greece and Cyprus: A DCC Model on the Interdependence of Small Economies

This paper discusses the volatility spillovers between the Greek Debt crisis and the Cypriot financial crisis. Cyprus was in the spotlight of financial markets due to significant problems stemming from the banking sector, which were dealt with by EU regulators with a bail-in on bank deposits. The current analysis aims to shed light on the reasons behind the implementation of this novel approach to bank distress. The study uses a Dynamic Conditional Correlation model on the returns of the stock markets of the two countries, which shows strong spillover effects during the period leading up to the 2013 Cypriot crisis, but a significant decrease of these effects from then on. The results confirm the close interdependence of the Greek and Cypriot economies before 2013 but also show that this interdependence was limited from that point onwards. This would indicate that since the risk of contagion to the Eurozone had diminished, regulators were able to test the bail-in solution in Cyprus in 2015. The current work contributes to the discussion on the interdependence of European economies. The paper’s findings can also be applied to other emerging European economies.


Introduction
The Great Financial Crisis (GFC) of 2008 triggered an unexpected worldwide turmoil and resulted in a series of economic shocks. International markets experienced a new economic framework, the consequences of which permanently changed the financial sector. During this turbulent period, Cyprus was forced to implement a new banking practice, the bail-in, after suffering significant negative spillovers from the Greek debt crisis, which started in 2010. Under the pressure of its European partners, the Cypriot government was forced to levy all bank deposits above 100,000 Euros by 40%.
The current study aims to measure, quantify and compare the co-movements between the Greek debt crisis and the Cypriot financial crisis. In addition, the paper tries to examine whether there exists a strong contagion phenomenon between these two economies. To achieve this, Engle's Dynamic Conditional Correlation (DCC) model (Engle, 2002) is employed, which is better suited to measure volatility in asymmetric data (Castagneto-Gissey and Nivorozhkin, 2016). The present work aims to show the bail-in solution was implemented in Cyprus only after authorities had ensured that any financial distress would not be transmitted to the rest of Europe. This is because the interdependence between Cyprus and Greece, its main trading partner in the EU, was significantly limited. This paper contributes to three aspects of the relevant literature. Firstly, it demonstrates how the DCC model can be used to quantify the volatility spillovers between two economies. Second, the paper investigates the relationships and the covariance between the stock markets of two developing EU economies, namely Greece and Cyprus. Third, the current research examines the contagion outcome of implementing the bail-in solution in a Eurozone economy.

Literature Review
Research on measuring cross-market dependence, correlations between stock markets or financial contagion is not new but can be traced back to the past. Financial contagion is commonly defined as negative shocks or spillovers transmitted across countries, especially during times of crisis. In terms of policy responses, spillover effects are important in evaluating the applicability of different measures taken by authorities. Castagneto-Gissey and Nivorozhkin (2016) examine the transmission paths from the Russian stock market to 18 major global markets after the implementation of the 2014-2015 sanctions against Russia. They find limited evidence of negative spillovers in returns, but also present volatility spillovers, particularly in emerging economies.
The first attempts to examine the contagion effect confirmed its existence after a financial crash (Calvo and Reinhart, 1996;Lee and Kim, 1993;King and Wadhwani, 1990). There is an ongoing debate on this phenomenon since some researchers confirm the increased correlation following financial crises while others present doubts. The Asian Crisis was the first widely examined case of international contagion, albeit regional. Dungey and Martin (2007) confirmed volatility co-movements among the Asian economies in the both the stock and the currency markets. Similarly, Huidrom et al. (2017) find spillover effects among emerging markets, using a vector autoregressive model. On the other hand, Corsetti et al. (2005) find evidence that financial contagion in other markets does not exist in their sample. Zhou and Gao (2010) analyse tail dependence of six major real estate securities markets to monitor the co-movements by using symmetrised Joe-Clayton (SJC) copula. The results showed that these six markets display varying tail dependence, both in terms of intensity and in terms of dynamics. McDonald et al. (2015) construct financial stress indices for Eurozone countries, by implementing multivariate analysis (VAR models). In this manner, they are able to model the interactions between the root causes of systemic risk in the Eurozone. They find that systemic risk in the sample economies is mostly responsive to own-country financial shocks, even though shocks from neighbouring countries may also be propagated to a certain extent. Polyzos et al. (2018) show that systemic risk could also be stem from governance issues, related to each banking institution. On the other hand, Zimmer (2014) propose a copula-based approach to model co-movements in house prices and finds that conflicting results between the US other OECD countries. He shows that US house prices in different areas exhibit simultaneous co-movements, while this is not true for the rest of the world. In general, the literature recognises relationships in financial markets as non-linear (Anufriev et al, 2018). Pantos et al. (2019) show that volatility spillovers are also present in electricity markets.
Even though a wide range of methodologies has been used, economists do not seem to agree on a single empirical procedure for the identification of contagion. Several studies try to model the various channels that may transmit the spread and to quantify contagion using various econometric techniques. Among these techniques, Engle (2002) proposes the dynamic conditional correlation Generalised AutoRegressive Conditional Heteroskedasticity model (DCC-GARCH) to overcome the limitations of previous methodologies on financial contagion. The main issue is the heteroscedasticity problem when estimating the time-varying conditional correlations. Several other authors attempt to extend this methodology and propose various modifications (Aielli, 2013;Samitas et al., 2020;Cho and Parhizgari, 2009;Pesaran and Pesaran, 2007;Cappiello et al., 2006;Rigobon and Sack, 2003;Billio and Pelizzon, 2003).
Following the GFC period in 2008, many recent studies use dynamic conditional correlations to examine financial contagion. Hwang (2014) employs a DCC-GARCH model to examine the transmission of the negative effects of GFC from the US to four Latin American stock markets and confirm the contagion effect, as attested by the increased magnitude and volatility of conditional correlations during the GFC period. Kim and Kim (2013) also test for negative spillovers of the GFC towards Korea and other neighbouring financial markets, using DCC-GARCH. They demonstrate that the GFC shocks were transmitted domestic financial markets (increased correlation coefficients) and thus further weakened them. Ahmad et al. (2013) use dynamic conditional correlations and examine financial contagion of PIIGS 2 on BRIICKS 3 countries. The results indicate that there exists a contagion effect, both from BRIICKS to PIIGS and vice versa, albeit not among all the countries in the sample.
Following the same framework, Kenourgios and Dimitriou (2014) and Karanasos et al. (2016) propose the FIAPARCH-DCC model to test for possible contagion effects of the GFC. Both studies find significant spillover effects, as well as volatility dependence across neighbouring stock markets and among regional financial and non-financial sectors.
However, Dimitriou et al. (2013) study BRIICs using the FIAPARCH-DCC methodology and are not able to find any specific pattern of contagion.
Some studies also take into account the Greek crisis. Tamakoshi and Hamori (2013) employ an asymmetric DCC model on five significant banking institution in Europe, which were exposed to Greek sovereign bonds. They find a significant burst in time-varying correlations between the returns of these banks' shares in the period following the EDC period. Following the same framework, Kenourgios (2014) studies both U.S. and European stock markets during the GFC and the EDC in terms of volatility contagion. The results indicate the existence of contagion in cross-market volatilities, which are significantly increased during these periods. The DCC approach has also been used by numerous other researchers when examining contagion during the GFC and the EDC (Chiang et al., 2014;Kazi and Wagan, 2014;Liow, 2012).

Data
In order to measure the conditional correlations between Greece and Cyprus and present the significance of the evidence, the data must first be split into two major subgroups. The sample is divided into two periods. The first period covers the 2008 Global Financial Crisis

Methodology
The paper uses the DCC model of Engle (2002) in order to test the behaviour of correlations between the Greek and the Cypriot stock markets. A major advantage of this model is the ability to test for dependence different time series. Until now, the literature includes a variety of models to investigate the contagion phenomenon and spillover effects. The literature review shows that in most cases the DCC model permits researchers to obtain robust results, particularly in case where there are asymmetries in the data. The DCC model is an appropriate specification for quantifying the interdependence among markets because it is flexible and allows time-varying correlations and covariance matrixes. Engle (2002) proposes the Dynamic Conditional Correlation (DCC) model, which he presented as a generalisation of Bollerslev's Constant Conditional Correlation (CCC) model (Bollerslev, 1990). The covariance matrix Ht of Bollerslev's model has the following form:

=
(1) where = {√ℎ , } is the diagonal matrix of the conditional standard deviations and = { , } is the correlation matrix. The expressions h are estimated with univariate GARCH models for each return series. In this paper, the Glosten-Jagannathan-Runkle GARCH (GJR-GARCH) model of Glosten et al. (1993) is employed. The GJR-GARCH model is appropriate for capturing any asymmetry and excess kurtosis in the data, particularly in cases where indices fail the assumptions of normal distribution. The GJR -GARCH model assumes the following form: Many authors support the assumption that the GJR-GARCH model captures the increased impact on variance at time t of negative shocks at time t−1 when compared to positive shocks. This asymmetry is known as the leverage effect. The negative shock produces increased risk and this means that this particular model is able to capture a higher number of lags in conditional variance [GJR-GARCH (p, q)]: The suggestion here is to model εt = σtzt where zt is i.i.d. The best model is selected by using the Bayesian Information Criterion (BIC) and Akaike (AIC). The DCC model differs only in allowing Rt to be time-varying. Therefore, the Engle's DCC model is expressed as follows: where α and β are the scalar parameters like an ordinary GARCH model. Qt * is a diagonal matrix with a square root of the i th diagonal of Qt on its i th diagonal position. Namely, in the Qt matrix, the model estimates the elements of correlations, which are calculated by the coefficients. As discussed in Engle (2002)

Descriptive Statistics
The summary statistics of the data are demonstrated in Table 1. Both indices (Greece and Cyprus) are negative skewed in GFC period while they are positive skewed in the EDC period. Likewise, both indices have kurtosis higher than 3 in the GFC period. However, in the EDC period only Cyprus exceeds kurtosis higher than 3 while the Greek index scores 2.7289. In both periods, the Cyprus market demonstrates the lowest and highest average return. However, at the same time, this market demonstrates the highest volatility, as attested by the increased values in standard deviation. The Jarque-Bera test statistic shows that neither of the two indices is normally distributed. Consequently, based on these preliminary findings, an AR(1)-GJR-GARCH model is apposite in order to capture asymmetry and excess kurtosis in both indices. Furthermore, both indices exhibit ARCH effects, with the null hypothesis of no ARCH effect being rejected uniformly for up to 5 lags.

Empirical Results
The estimations of the DCC model are presented in Table 2 and Table 3 in a two-stage process. Table 2 presents the univariate estimations AR(1) -GJR GARCH (1,1) for both indices. The g coefficient, which shows the leverage effect, is significant only in the case of Greece in the GFC period. This guarantees the absence of normality in the index.
However, in all other cases the absence of normality is not strong enough.    Table 3 shows the Dynamic Conditional Correlations of the two stock markets. The unconditional correlation is statistically significant only in the case of the EDC period (0.883). The ARCH parameter α was higher in the GFC period (0.06) which means that shocks were significantly stronger in the first period than in the second (0.034). On the other hand, the GARCH parameter β was higher in the EDC period which shows the extent of increased volatility in the market. It is evident that if terms a and b are found to be positive and their sum is lower than one (a+b<1), this implies the existence of dynamic conditional correlations. As can be seen, the results support the presence of correlations over time and the existence of a contagion effect. Furthermore, the analysis shows significant increase during the crash period among the indices.  In the case of conditional correlation for the GFC period (Figure 7), the values start from the negative region, but display an upward trend until the peak value of 0.80. However, in the EDC period (Figure 8), it can be observed that the behaviour of the correlation is completely different; from 2012 onwards, the correlations show a negative trend until they reach their lowest point in 2014, before only slightly rising again. The two crises are faced differently by the Eurozone, since the Greek case was still underway when the Cypriot crises erupted. In addition, the nature of the problems and the structure of the economies were vastly different. Thus, the Cyprus case was a great opportunity to test run the bail-in solution, turning the depositors into bank shareholders.
The risk for Europe however lay on Greek instability, which could be deepened by the interdependence of the two economies. However, by 2013 this interdependence was significantly decreased (Figures 6 and 8) and, thus, it was now easier to implement the bailin in a shielded environment. It is clear that Cyprus, being a small economy, whose financial ties with the Eurozone were hindered, was an ideal case for a test implementation of bail-in, which, it should be noted, has since been adopted as the go-to solution for banking distress.

Discussion
Greece, as a member of the European Economic Community (EEC) from 1981, enjoyed several advantages through development programs provided by the European Union.
During the last decade, government policies led to a substantial public deficit due to the inefficient management of the development programs. The 2004 Olympic Games and the non-productive public sector increased country's obligations. These needs were financed by bonds, the return on investment ratio of which was not adequate to cover the country's costs. Tax evasion as well as political corruption led the country to a financial dead end.
The 2008 Global Financial crisis revealed these problems in the Greek economy and alerted hedge funds and major credit rating firms which focused on the Greek economy and its declining debt-worthiness. Despite the fact that the Eurozone seemed to be well secured, credit default swaps (CDS) focused on Greece. The consequences of these events forced the Greek government to implement a series of harsh austerity measures in order to decrease its deficit and debt, which at the end of 2009, according to Eurostat, were 15.2% and 126.8% of GDP respectively (Figure 9).
The situation in Greece has since been characterised by an economic impasse, with rising unemployment and significant liquidity problems in the banking sector. However, some of the core issues of cooperation among EU members did not assist to achieve a swift response and thus volatility remained in the European economic environment. Investors who bet on the collapse of the Eurozone took advantage of the conflicting interests between its members and increased the pressure on countries with high debt and deficits. This resulted in a debt crisis for South European countries and Ireland, which was nightmarish not only for Greece but for the Eurozone as a whole. For Greece, the crisis was deepened by the inefficient banking sector (Christopoulos et al., 2020) as well as by corruption and the poor functioning of government institutions (Policardo and Carrera, 2018).
Additionally, markets were still restless due to the global recession that followed the US Subprime Crisis. Many other countries including Belgium, UK and France faced high debts and deficits. This resulted in an extended recession in the Eurozone. Following the Greek debt crisis, Cyprus was hit by the domino effect of negative consequences. As can been seen from Figure 10, the Cypriot economy passed into a recessionary stage after 2009. The country seemed to be well secured at the beginning of subprime crisis, but then a huge debt crisis was triggered which surpassed the average level of the Eurozone. Some of these reasons were non-performing loans, the exposure to the haircut of the Greek government bonds and the inability to raise liquidity from the markets to support the financial sector. This resulted in an increase in unemployment and a steep deterioration in output in the tourism and shipping sectors. Consequently, commercial properties declined by almost 30% and the banking sector faced liquidity problems from the exposure (22 billion Euros) to the Greek private sector. It is clear that the Cyprus crisis was different from the Greek crisis as the initial problem was the banking sector.

Source: World Bank & ECB
Cyprus had a very low tax rate and has thus attracted many foreign investors, including many Russians. As credit rating firms gradually downgraded their ratings for the Cypriot economy and the liquidity problem came to surface, Russia offered an emergency loan of 2.5 billion Euros (at 4.5% interest rate) to Cyprus in order to cover its financial gap through the international markets. Unfortunately, this solution did not solve the problem since the loan did not include any funds for the recapitalisation of the banking sector after the haircut of the Greek government bonds. was inherently unstable and would bring about significant debt crises due to asymmetric effects (Beckworth, 2017).
The core issue of the Eurozone in the Greek Debt crisis was whether a small country, that covers the 2.5% of Eurozone's GDP, can affect the entire European region. This possible scenario forced the Eurozone and the IMF to focus more on this direction. In the meantime, most of the developed economies were struggling to recover from the subprime crisis and hedge the risk from the exposure. The involved and exposed stakeholders tried to confront the threat at an early stage. Greece government adopted many austerity measures (such as 10% cut to bonuses, freezes in public-sector salaries and increases in VAT) in order to increase savings and reduce the high government deficit. Unfortunately, the measures were not enough, and the recession deepened even more while consumption decreased rapidly, and the government was unable at this stage to stabilise tax revenue. All the upcoming rescue packages did not change the financial condition in Greece; tax collection inefficiency as well as delays in public sector's reconstruction were the biggest challenges.
Eurozone's presented significant inability to successfully resolve the problem in Greece creating serious doubts about the effectiveness of the program. Shortly after, the Eurozone felt the pressure from the credit rating firms. Hence, in January 2012, Standard & Poor's downgraded France (from AAA rating to AA+) and this was the first shock in the Eurozone area.
As for the Cypriot Financial crisis, the applied bail-in model, affected only the domestic economy while the spillover effects to other countries were significantly low. It was assumed that the program of Cyprus was ineffective in the first place, because even three years after the applied measures, the Cypriot economy presented negative GDP growth and persistently high unemployment. On the other hand, major economies and investors had a great opportunity to implement a new model in a small country with low transmission effects. The economy of Cyprus had a significant, well-organised banking sector, compared to the size of the country, and foreigners (including many Russian investors) had placed large amounts of money in the local economy. In addition, the country invested a lot in the exploration of natural gas in the maritime exclusive economic zone and the agreements with Israel and USA were the next great challenge to lead the economy to development.
The austerity measures implemented in Greece did nοt provide any flexibility to increase the GDP and simultaneously decrease the deficit to a sustainable level. Also, this was the first time that a Eurozone country faced such a severe financial crisis that was intercorrelated with the unified currency. The threat of financial contagion led the members of the Eurozone as well as investors and governments to study and carefully monitor the possibility of a domino effect from Greece to other countries or channels of the economy; especially that period after the 2009 and beginning of the European Debt crisis. In case of a "Grexit", some may have anticipated great losses to several major economies, which would be difficult to calculate that period. In the pessimistic scenario, the EU could face several attempts from its members to withdraw from the Eurozone area, with the rest of the PIIGS countries being the first in line. The pending (at the time) decision for Brexit deepened this risk (Polyzos et al., 2020). Despite claims and reassurances from EU policymakers that the financial condition in the Eurozone was tranquil, stock markets were strongly interconnected with rumours and negative information. Thus, a possible domino effect was feasible and for a persistent long period before the stability gain ground. It is reasonable to conclude that the Greek Debt crisis was similar the ones in Italy and Portugal, while the banking crisis in Cyprus resembles those of Ireland, Spain and Iceland.

Conclusion
In this study, a DCC Model was applied to investigate the existence of interdependence during the Greek Debt crisis and the Cypriot Financial crisis. The paper's findings, in line with existing literature (Suleman et al., 2017), show increased volatility during the outbreak of the two crises. The current work also shows that the correlation between the two stock markets was not only strong, but also increasing up to approximately 2013. However, after the emergence of the Cyprus banking crisis, this correlation was significantly decreased.
These findings are in line with similar literature on the topic (Samitas and Kampouris, 2019).
Following this, the European authorities chose to implement the bail-in solution to the Cypriot crisis, in order to test the results in a protected environment. Cyprus, as a small country and economy, seemed to not have the power to produce spillover effects on bigger economies, except through the Greek economy. Since the correlation with Greece was decreased, the path was henceforth open. It can therefore be assumed that Cyprus was used as a test case to measure the effectiveness of bail-ins as a solution to banking sector stress, without the risk of further impact to the Eurozone. In addition, the implementation of the bail-ins served as a deterrent for banks and local authorities alike, in order to avoid risky behaviour or loose banking oversight.
In terms of suggestions to policymakers, the outcome of the experiment seems to be successful. The bail-in solution was tested and, despite the hard consequences for both the Cypriot and the Greek economies, the resulting effects on the rest of Eurozone were minimal. The bail-in was deemed successful and thus was adopted by the European