Systemic risk is a very complex and challenging concept to measure. According to Benoit et al (2017, p. 131) it is “one of the most elusive concepts in finance”. The European Systemic Risk Board defines systemic risk as “a risk of disruption in the financial system with the potential to have negative consequences for the internal market and the real economy” (Constâncio 2010). As such, systemic risk affects all other types of risk that exist in financial systems.

Studies have proposed multiple methods to quantify systemic risk in advanced economies. (For a thorough analysis of the applicability of 54 such methods to the Central and Eastern European region we refer the reader to Karaś (2019).)

However, such methods are only partially transferable to countries with less advanced financial systems because these countries have much smaller stock markets than advanced economies.

To the best of our knowledge, no other work proposes the methodology outlined in this paper. Thus, the paper adds to recent advances in risk modeling. We propose a set of modifications for the computation of two well-known systemic risk measures: SRISK and CoVaR. The modifications make these risk measures applicable to a much wider range of credit institutions in emerging economies. Moreover, they allow for

• •

  including more information regarding the structural and ownership characteristics of the studied financial systems (using systemic importance scores (SISs) for system-wide risk aggregation); and

• •

  using information from domestic and global stock markets to enable a more effective analysis of both local and broader environments.

Further, we adopt a nonstandard approach to calculate CoVaR and $\mathrm{\Delta }$CoVaR, as we replace infrequent book-value data with daily stock returns. This replacement increases the sensitivity to behavioral phenomena and their effect on the quantification of systemic risk. It also improves transparency.

We empirically apply the proposed methodology to nine Central and Eastern European (CEE) countries (Bulgaria, Estonia, Czechia, Hungary, Latvia, Lithuania, Poland, Romania and Slovakia). The results indicate how systemic risk levels – specifically fragility (SRISK) and the potential of risk spillover ($\mathrm{\Delta }$CoVaR) – change throughout the period from 2006 to 2018. We also compile a set of rankings to facilitate comparison of the studied countries with respect to the price of systemic risk and its burden on the state. Our methodology is also used to calculate the same measures for Germany. This calculation enables a comparison of the scale of systemic risk experienced by the CEE region with the most advanced European economy.

Finally, we run regressions to test three hypotheses.

1. (1)

   SRISK and $\mathrm{\Delta }$CoVaR provide different information about systemic risk in the studied CEE countries.

2. (2)

   The difference between the results obtained with SRISK and $\mathrm{\Delta }$CoVaR relates to the degree of complexity of the studied financial systems.

3. (3)

   SRISK- and $\mathrm{\Delta }$CoVaR-based risk signals converge in periods of higher risk (ie, crises).

The paper proceeds as follows. Section 2 presents a review of systemic risk in CEE countries. In Section 3, we discuss the relevance of SRISK and CoVaR. In Section 4, we outline our proposed modifications. Section 5 presents the technical aspects of the proposals. Section 6 is our empirical analysis, and Section 7 concludes.

At least two facts about the financial systems of the CEE region have crucial implications for systemic risk analysis. The first is a typical feature of all emerging markets: the banking sector is the most central part of the financial system; thus, it is also “the main source of risk for financial stability” (Karkowska 2013, p. 4). The second fact is that foreign owners control the CEE banking sector: they possess over 90% of the total banking assets in the region (Radulescu et al 2018, pp. 7–8). As a consequence, CEE countries have relatively homogeneous financial systems, in which foreign-owned banks are major credit suppliers (Dumicic 2018, p. 2).

Interestingly, foreign-owned banks do not list on the local stock markets until these markets become sufficiently large and deep. The Polish example illustrates this phenomenon well. In Poland, all larger banks are listed on the Warsaw Stock Exchange,¹¹ 1 See the WIG Banking Index: https://bit.ly/3jmbn5y. while in the other analyzed countries almost none of the foreign-owned banks are locally listed. As a consequence, a much broader set of systemic risk measures are applicable to the Polish financial system than for the other CEE countries.

The fact that so many CEE banking systems depend on foreign-owned banks to supply credit poses a significant macroprudential challenge. Regulators in CEE countries pioneered the use of a wide variety of macroprudential tools, and were using them even before these tools were known as “macroprudential”, that is, before the global financial crisis (Dumicic 2018, p. 10). However, most of the subsidiaries of big Western European banks were able to circumvent the regulations thanks to their close ties with parent institutions (Dumicic 2018). This evasion was likely facilitated by the existing internal markets identified in an empirical study by De Haas and van Lelyveld (2010).

A notable example may be the sharp deleveraging of the Romanian and Bulgarian foreign-owned banks, which affected the supply of credit to the real economy (Radulescu et al 2018, pp. 3–4). Despite their efforts, local regulators were not able to limit this process. Correspondingly, Jočienė (2015, p. 51) concludes that foreign-owned bank subsidiaries in Lithuania, Latvia and Estonia were adopting the business models of their foreign parents even when these were not optimal models for their local environments. Barkauskaite et al (2018) find that a dependence on the foreign parent in strategic decisions is an important macroprudential factor. As they point out, improvement is needed in the methods currently used by regulators for the quantification of systemic risk for such banks.

Since 2015, the European Union has required the regulators of its member states to calculate SISs, which identify other systematically important institutions (O-SIIs). These O-SIIs have been described by the European Banking Authority (EBA) as “institutions that, due to their systemic importance, are more likely to create risks to financial stability whilst maximizing private benefits through rational decisions [and] may bring negative externalities into the system and contribute to market distortions” (European Banking Authority 2014a). Foreign-owned banks, most often the subsidiaries of much bigger Western European and Nordic banks, compose the majority of such O-SIIs in the CEE region.

All central banks in the European Union analyze the external financial environment in their financial stability reports. The long historical experience of systemic risk monitoring leaves no doubt as to the usefulness of such information. The works of Bisias et al (2012), Hattori et al (2014) and Benoit et al (2017) provide an overview of an extensive number of studies that propose various systemic risk measures.²² 2 In the course of the larger study, we have identified 54 systemic risk measures used by regulators and researchers. However, only a few of them are applicable to CEE countries due to data limitations. (For details, see Karaś (2019). The work may be accessed at http://www.wir.ue.wroc.pl.)

Despite this profusion, only a few studies present sophisticated systemic risk measures for the CEE countries. Among them are the financial instability index (FII) proposed by Jakubík and Slačík (2013) and the composite indicator of systemic stress (CISS) (Hollo et al 2012) calculated by Kubinschi and Barnea (2016) for Czechia, Hungary, Poland and Romania. The European Central Bank (2019) computes the CISS on an ongoing basis for the advanced economies of the eurozone as well as for Czechia, Hungary and Poland. Unfortunately, the FII and CISS use only broader market indexes (from the money market, the equity and bond markets and the exchange rate market). Therefore, these two measures do not enable the analysis of microprudential bank data such as leverage or various exposures.

In the sample period, among the central banks of the CEE countries selected for this study, only two used more advanced methods of analysis. One is Poland, which employs a relatively elaborate early-warning model using the estimated output gap (Narodowy Bank Polski 2019). The other is Lithuania, which uses the gross-domestic-product-at-risk measure (Lietuvos Bankas 2019). Both approaches allow for combining data regarding the wider financial environment with microprudential indicators to obtain a single score for systemic risk (an indicator or index).

Quantile-based systemic risk measures that combine bank-level and market-level data are even more rarely used in CEE countries. Further, when such measures are calculated, data limitations restrict the sample of banks by excluding the foreign-owned banks that are not listed on the domestic stock exchanges. Table 1 presents details regarding all (to the best of our knowledge) the studies that use this type of systemic risk measure for the CEE region published in the last decade.

As seen in Table 1, the studies that use quantile-based measures for systemic risk quantification in CEE countries were markedly ineffective in capturing the effect of O-SIIs. In all cases except Poland, no more than 50% of the existing O-SIIs were included in the sample. The last column indicates how our methods increase the number of systemically important banks that may be analyzed. In some cases, the increase is by 50–60% (Bulgaria, Czechia, Lithuania, Latvia and Slovakia) and in others it is by about 70–75% (Hungary and Romania). Our methods also enable the first analysis of Estonia. In total, we add 31 new O-SIIs to SRISK and $\mathrm{\Delta }$CoVaR computation. Thus, out of 65 O-SIIs identified by the EBA, we analyze a total of 49.

Karkowska (2013) reports that systemic risk in the European banking sector is driven initially by bank-specific risk premiums and subsequently by market determinants, such as volatility and stock-exchange capitalization. Large financial leverage, significant maturity mismatches, turbulence in the financial markets and the interactions between market liquidity and financing significantly increase the degree of systemic risk.

Since adverse shocks spill over between financial institutions, systemic risk should not be analyzed using risk measurements focused on individual institutions. Especially in relatively small financial systems (such as those in the CEE region), the risk assumed by one larger institution may cause significant spillover effects. Further, herding behavior, which becomes more likely as the market size decreases, may cause exposure to adverse systemic events, even if individually financial institutions are characterized by a low level of risk. As Adrian and Brunnermeier (2011, p. 2) point out, “regulation based on the risk of institutions in isolation can lead to excessive risk-taking along systemic risk dimensions”. Therefore, there is a strong need to analyze risk in the financial system as a whole (see Duffie 2019; Engle and Ruan 2019).

At the same time, insight into the financial condition of the institutions that comprise the financial system is essential for quantifying systemic risk because institutional fragility entails an increased sensitivity to systemic events. This sensitivity refers especially to the build up of undercapitalization, which makes institutions vulnerable to liquidity fluctuations and spillover effects. Brownlees and Engle (2017, p. 57) argue that “because of the extensive use of leverage in the financial sector, this industry is particularly vulnerable to downward market movements”.

Many fragility-focused measures have received significant recognition in the literature. Comparative studies point to significant advantages of SRISK relative to other measures (see Benoit et al 2014; Acharya et al 2014; Laeven et al 2014, 2016; Brownlees and Engle 2017; Jajuga et al 2017). One such advantage is that SRISK uses information about the build up of undercapitalization. Another is SRISK’s use of the technique for calculating quasi-leverage proposed by Brownlees and Engle (2017), which retains the daily frequency. In effect, SRISK is adaptable to rapidly changing market conditions. The fact that this measure also contains information about the size of the bank further improves its informational properties.

Based on a series of robustness checks, Brownlees and Engle (2017) conclude that the precrisis levels of SRISK are a good predictor of external capital injections, bailouts and worsening macroeconomic conditions (specifically, industrial production and the unemployment rate). Even more interesting is the fact that SRISK is a predictor of other institution-focused risk measures that are based on proprietary regulatory data, such as stress tests. Moreover, since the measure is based on publicly available information, it is very transparent. Finally, because SRISK assumes that the triggering systemic event takes the form of a prolonged market decline, its conditionality is very intuitive (cf, Engle and Zazzara 2018).

The other systemic risk measure selected for this study is $\mathrm{\Delta }$CoVaR. Its authors (Adrian and Brunnermeier 2016) describe it as complementary to high-frequency measures based on the marginal expected shortfall (MES), such as SRISK. Similarly, Kuziak and Piontek (2018) confirm that $\mathrm{\Delta }$CoVaR does not correspond to the condition of financial institutions or their financial soundness indicators. This empirical observation confirms that $\mathrm{\Delta }$CoVaR is a measure of risk spillover and not a measure of fragility as such.

Interestingly, Benoit et al (2014) show that CoVaR reflects the bank’s tail risk (its VaR) conditional on market returns, while SRISK is associated mostly with leverage. Measures based on MES, such as SRISK, are able to determine the institutions most exposed to a potential financial crisis (fragility), whereas CoVaR indicates the institutions that contribute most to a potential crisis (risk spillover potential).

Benoit et al (2014) also report that, in their comparison of various measures, the lowest concordance was that between SRISK and $\mathrm{\Delta }$CoVaR (only 9.9%, with similarities between the other measures of about 50%), demonstrating that the differences in measurement output come from the characteristics of the two measures and not from model risk or instability. Additionally, they show that SRISK and $\mathrm{\Delta }$CoVaR${}^{\text{€}}$ ($\mathrm{\Delta }$CoVaR expressed in euros) produce different rankings of systemically important financial institutions (SIFIs), which also confirms that these two measures have a different focus and capture different information about systemic risk. In the empirical part of this paper, we produce similar rankings that confirm this conclusion for our sample of CEE countries.

Measuring the contribution of a given financial institution to contagion often relies on confidential data about positions and risk exposures that banks provide to the regulator. However, such risk may also transmit indirectly, for instance, through price effects and liquidity spirals. According to Adrian and Brunnermeier (2011, p. 8), the indirect spillover effects are “quantitatively more important”, as a fire sale of assets may lead to losses for all the market participants who hold similar exposures. In the same sense, in a low liquidity environment, each market participant creates a state in which their subsequent individually optimal responses to adverse events are likely to cause externalities for others. Arguably, such exposures are more significant in smaller, more shallow markets typical of emerging economies. $\mathrm{\Delta }$CoVaR captures this risk by looking at the increase in tail co-movement between the institution and the system.

The characteristics that align CoVaR-based measures with the theoretical foundation described above include the following (Adrian and Brunnermeier 2011, pp. 9–10).

Conditionality:

the conditioning of the risk measurement on an institution’s value-at-risk (VaR) with a threshold that facilitates a choice of a (very) low probability scenario.

Focus on tail distribution:

the adverse-scenario-based conditioning that typically shifts the mean downward, increasing the variance and potentially also increasing the higher moments, such as negative skewness and kurtosis.

Noncausality:

not distinguishing between causal co-movement and co-movement driven by a common factor, and thereby capturing direct and indirect spillovers.

Directionality:

the ability to measure how the system affects the risk of the institution and how the risk of the institution affects the system, while noting that these relations may be asymmetric.

$\mathrm{\Delta }$CoVaR is a statistical measure, so it does not require explicit reference to structural economic models, which makes it readily applicable to almost any financial system, even one with scarce data. The corresponding assumptions are that public market data, such as stock returns, reflect information about publicly traded financial institutions, and also depict the behavioral phenomena that accompany trading.

The measures selected for this study are also relatively well suited to distinguish between the risk triggers which are of systemic importance and those that are not. They enable us to limit the input data to O-SIIs, which means that any kind of distress related to a nonsystemic bank is automatically disregarded. Also, as the measurements are based on stock market reactions, they depict the sentiment of market participants toward the expected effect of various events. An example might be sale of a large branch of one systemically important bank to another. If the markets view such an event with optimism, eg, because this solution means a future healthier business position for the seller, then we observe a positive reaction in the markets. This is the case even if the event initially stresses the books of the seller by putting a strain on the asset side of its balance sheet.

A good illustration of such an event was the withdrawal of Deutsche Bank (one of the Polish O-SIIs) from Poland. Despite a drastic drop in assets, the downscaling did not act as a systemic trigger for the banking system, because the bank’s loans were secured by another global SIFI that was expanding into the Polish market at the time (Santander Bank). The German and Polish stock markets did not react badly to news of the withdrawal and did not record any increase in systemic risk. The reaction of the risk measures selected for this study reflected the actual situation: the takeover did not trigger systemic risk in the Polish or German financial systems.

Another interesting example relates to one of the SIFIs present in the CEE region: Crédit Agricole (CA). CA initially took over the Greek Emporiki Bank, only to later sell it to another systemically important Greek bank, Alpha Bank. This was significant for some CEE countries, such as Romania, in which Emporiki Bank had a major presence. The move allowed CA to grow in Albania, Bulgaria and Romania.

As the sovereign debt crisis unfolded, CA became exposed to Greek government bills and bonds. Because the Greek branches had by that point been unprofitable for some time, CA decided to sell them at a loss (for €1 per stock) to Alpha Bank. This decision was actually welcomed by the stockholders of CA and in the following weeks investors saw a price recovery for CA’s stock.

The systemic risk measures applied in this paper did not react to this sale as they would to a systemic risk trigger, despite the fact that CA had realized a book-value loss from the transaction. There was no actual systemic risk that either Greece or the parent institution could transfer to the CEE region. Thus, the systemic risk measures reacted appropriately.

In order to effectively apply SRISK and $\mathrm{\Delta }$CoVaR to the CEE region, we propose three modifications. The first allows for the expansion of the sample of institutions. The second consists in adding a broad domestic market index to the input data, which increases the sensitivity of risk measures to domestic macroeconomic conditions. The third proposal introduces a new way to establish the model of the financial system by modifying how the weights of banks are calculated. Proposals 1 and 3 refer to both risk measures, while Proposal 2 refers specifically to $\mathrm{\Delta }$CoVaR. Further, we follow an approach similar to Benoit et al (2014) in which we only use stock market data for the calculation of CoVaR and $\mathrm{\Delta }$CoVaR.³³ 3 The precise methodology we use is an extension of the approach used in our earlier studies (see Karaś and Szczepaniak 2017).

This approach maximizes the sensitivity of the measures to the behavioral aspects of systemic risk and decreases dependence on book-value data.

For a general case, a national financial system $s$ is defined as a composite of $i$ O-SIIs, where $i=1,\mathrm{\dots },N$. The rate of return of financial system $s$ at time $t$ is defined as

where $r_{i,t}$ is the rate of return on the equity of institution $i$, and $w_i$ is the weight of financial institution $i$ in system $s$, which is equal to

where ${\mathrm{SIS}}_i$ is the systemic importance score of institution $i$, and ${\mathrm{SIS}}_s$ is the total SIS of all O-SIIs in the national financial system $s$.

In the case of O-SIIs that are not listed in the domestic stock market, the rate of return $r_{i,t}^{\mathrm{domestic}}$ and the market capitalization ${\mathrm{MV}}_{i,t}^{\mathrm{domestic}}$ are established as

where the conversion ratio is equal to

$A_{i,t}^{\mathrm{domestic}}$ is the book value of the O-SII’s assets and $A_{i,t}^{\mathrm{parent}}$ is the book value of the parent institution’s assets.

For CoVaR, the set of O-SIIs is supplemented by the domestic stock index. In this case, financial system $s$ consists of $i$ financial institutions and the stock index. Here, the rate of return $r_{s,t}$ of system $s$ at moment $t$ is defined by (5.6), where $r_{i,t}$ is the logarithmic rate of return on institution $i$ and $w_i$ is the institution’s share in the system, that is, the ratio of the institution’s score value ${\mathrm{scr}}_i$ to the total score of the entire system: ${\mathrm{scr}}_s={\sum }_{i=1}^{N-1}{\mathrm{scr}}_i$. In the case of the stock index, $r_{d,t}$ is the logarithmic rate of return on the domestic index, while the index share $w_d$ is the value of the institutions’ average share (mean $w_i$) in each analyzed financial system

where $r_{i,t}$ is the logarithmic rate of return on institution $i$ at time $t$, $r_{d,t}$ is the logarithmic rate of return on the domestic index at time $t$, $w_i={\mathrm{scr}}_i/{\mathrm{scr}}_s$ is the share of institution $i$ in the system $s$ and $w_d={\sum }_{i=1}^{N-1}{w}_i/(N-1)$ is the share of the domestic index in the system $s$.

When determining the currency-denoted value of system-level CoVaR, we establish the index capitalization ${\mathrm{MV}}_d$ as the average capitalization of financial institutions for each considered system:

where ${\mathrm{MV}}_i$ is the capitalization of institution $i$.

The VaR of financial institution $i$ in system $s$ at the level of confidence $(1-q)$ equals

where $F_i$ is the cumulative distribution function of $r_{i,t}$.

Since $P(r_{i,t}\le {\mathrm{VaR}}_{i,t}^q)=q$, ${\mathrm{VaR}}_{i,t}^q$ can be determined as a $q$-quantile of the distribution $F_i$ and the VaR of an individual financial institution is given by

where ${\sigma }_{i,t}$ is the volatility of the rates of return at time $t$.

The expected shortfall (ES) of system $s$ is defined as the average of all the losses that are greater than or equal to the VaR:

where $c$ is a quantile of the distribution of $r_s$ equal to ${\mathrm{VaR}}_{s,t}^q$ for $q=1\%$.

The marginal expected shortfall (MES) is given as a partial derivative computed as

The long-run marginal expected shortfall (LRMES) is defined as the expectation that the financial institution’s multiperiod return on equity is conditional on the systemic event,

where $\gamma$ is the correcting factor relative to the length of the assumed horizon.

The long-run market decline corresponds to a 40% loss over a horizon of six months (cf, Engle and Zazzara 2018). By assumption, we ignore the negative values of SRISK (ie, any potential capital surpluses). Thus, SRISK is defined as

where $D_{i,t}$ is the value of debt at time $t$, $V_{i,t}$ is the market value of equity at time $t$ and $k$ is the regulatory minimum capital ratio.

We set the $k$ parameter at 8%, following Brownlees and Engle (2017) and Engle and Zazzara (2018). As presented before, the system-wide ${\mathrm{SRISK}}_{s,t}$ of financial system $s$ is computed as the sum of all ${\mathrm{SRISK}}_{i,t}$ for all O-SIIs in that system, where $i=1,\mathrm{\dots },N$.

CoVaR captures the total price of risk in the financial system that is conditional on an institution being in distress. Formally, ${\mathrm{CoVaR}}_{s,t}$ of financial system $s$ corresponds to the ${\mathrm{VaR}}_t^q$ of the market return that is obtained conditionally from event $c(r_{i,t})$ that is observed for financial institution $i$. In computation, we use the concept of the financial system and its VaR that are outlined in (5.1)–(5.7) as well as the following probability equation:

$\mathrm{\Delta }{\mathrm{CoVaR}}_{i,t}$ captures the marginal contribution of institution $i$ to overall systemic risk in a noncausal sense. The distress state $c$ of financial institution $i$ is defined as ${\mathrm{VaR}}_{i,t}^q$ for $q=1\%$, and distress is understood as every instance when losses are at least equal to ${\mathrm{VaR}}_{i,t}^q$:

To establish the value of $\mathrm{\Delta }{\mathrm{CoVaR}}^{\text{€}}$, we multiply the $\mathrm{\Delta }\mathrm{CoVaR}$ of each institution by its size (market capitalization) expressed in euros. We use daily euro exchange rates for each domestic currency.

The sample period for this study is from 2006 to 2018, which includes the global financial crisis, the sovereign debt crisis and the period of the subsequent economic downturn. The total sample of banks equals 53 (49 for the CEE region and 4 for Germany; see Table 2).

The data comprise approximately 170 000 entries. The data sources are Thomson Reuters DataStream and the publicly available financial statements of the sample banks, as well as official documentation related to the identification of O-SIIs published by national regulators. The results include system-level $\mathrm{\Delta }$CoVaR and SRISK presented in Figures 1–12, the rankings of countries based on these measures presented in Tables 3 and 4 and the regressions of $\mathrm{\Delta }$CoVaR and SRISK presented in Table 5. Additionally, we provide granular bank-level results in the online appendix.

This paper presented three proposals to modify two well-established measures of systemic risk, SRISK and $\mathrm{\Delta }$CoVaR. Thanks to these modifications we significantly increased the number of analyzed systemically important institutions in the CEE region. We also included information about banks’ systemicness, interconnectedness, complexity, significance and size. Additionally, we integrated the data of broad domestic market indexes into the calculation of $\mathrm{\Delta }$CoVaR.

Our methodology takes into account endogenous systemic risk factors, at both the bank level and the system level, as well as exogenous factors such as the exchange rate and the impact of the largest European stock markets. For this reason, the study is more comprehensive than its predecessors with respect to the CEE region.

The empirical results confirmed increased systemic risk for CEE countries in two periods, 2008–9 and 2011–13, at the same time identifying countries more prone to fragility and to contagion. The rankings confirmed the high potential cost of systemic risk materialization in the CEE region. We also showed that, in some cases, this burden was greater for small CEE countries than it was for Germany, which has, by far, the greatest value of systemic risk.

We verified three postulated hypotheses and confirmed that systemic risk in the CEE region has the same theoretical properties as it does for the advanced economies. These properties underline the necessity of analyzing this risk by using measures at least as sophisticated as the ones used by the most developed countries.

The results also show that during periods of calm, SRISK and $\mathrm{\Delta }$CoVaR each indicate different financial institutions as the weakest links in the given financial system. These results have a macroprudential consequence: the regulatory bodies in the analyzed countries should consider the information provided by both SRISK and CoVaR when monitoring systemic risk, especially when they want to put banks under scrutiny.

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

We are extremely grateful to special issue Co-Editor Mark Carey and anonymous reviewers for their substantial contributions, because the manuscript was significantly improved with their assistance. This paper forms part of a research project funded by the National Science Centre, Poland, under Agreement UMO-2018/29/N/HS4/02783.