Determinants of Credit Default Swap Spreads : A Four-Market Panel Data Analysis

This paper attempts to elucidate whether firm performance and macroeconomic conditions play a significant role in explaining credit default swap (CDS) spreads. Our panel dataset covers 112 reference entities in four markets (South Korea, Hong Kong, France, and Germany) for the period 2001-12. Overall, our results suggest that market value indicators (Tobin’s Q, stock market returns, and the interest rate) appear to be more important than book value indicators (i.e., ROA, ROE, and the GDP growth rate) in determining CDS spreads. Moreover, Asian CDS markets are shown to be more sensitive to both GDP and stock market volatility, than the two European markets. Finally, the 2007-09 global financial crisis may have significantly affected the CDS market as a whole, but it generally did not affect the individual markets. These results are robust to various model specifications. This paper contributes to the understanding of CDS determinants at firm-, economy-, and market-level. JEL classifications: G10, G15, G32


Introduction
First introduced circa 1994 by JP Morgan, credit derivatives have substantially expanded over the past decade.Since their development, credit default swaps (CDSs) have attracted a wide range of users, from ~ 10 ~ banks and other financial institutions to corporate and supranational bodies.According to Depository Trust and Clearing Corporation (DTCC) statistics, the gross notional amount of CDS trading was $25.9 trillion as of the year-end of 2011, and the net notional amount stood at $2.7 trillion.As their primary function, CDSs provide lenders with a form of protection against the occurrence of a credit eventborrower (reference entity) default.The formation of a CDS contract involves the following conventional setting: When a lender (the protection buyer) purchases a CDS from an insurance company or another financial institution (the protection seller), the loan becomes an asset that may be swapped for cash in the event of loan default.If no credit event occurs, the protection buyer makes premium payments until the contract matures; however, if a credit event occurs, then the protection seller pays the buyer for the loss, and the contractual relationship ends.
Undoubtedly, the key feature of a financial product is the "price", which is represented by the "spread" in the case of CDSs.In a CDS contract, the amount of the protection premium, which is the annual amount that the protection buyer must pay the protection seller over the length of the contract, can be calculated from the "CDS spread".As with any other insurance product, the CDS spread can be regarded as the price of risk.Any changes in factors that could alter the perceived level of risk will cause an adjustment in the CDS spread.Given the emerging significance of CDSs as a risk management product in financial markets over the past decade, further knowledge about the determinants of CDS spreads will certainly provide more insight into the probability of default and ensure that both financial regulators and risk managers better understand the use of CDS contracts.In actual fact, this remains the main motivation of our current research.
The present paper aims to further contribute to the understanding of CDS pricing by expanding the search for CDS determinants to a multi-level dimension.Fundamentally, we employ a structural model and examine factors at three levels: microeconomic (firm); macroeconomic (economy); and across four markets in two continents (market), that theoretically contain information about CDS spreads.To facilitate a comparative study, we select four international CDS markets from two continents -Asia and Europe.The vast majority of studies in CDS employ US dataset (e.g., Houweling  ).We purposely choose Hong Kong, South Korea, France, and Germany for our study due to two main reasons: 1) Hong Kong is a major Asian financial hub which provides important services in international finance, whereas some South Korean conglomerates such as Samsung, Hyundai play a significant role in the global economy.2) France and Germany1 are the two largest Euro-denominated CDS markets.With this approach, we aim to capture any region/market-specific variations-that is, we wish to examine whether our results are affected by a market's geographical location.The hypothesized outcome is that, either: (i) no differences exist, i.e., all CDS markets are the same (homogeneous), or (ii) due to geographic and cultural reasons, the two Asian markets are similar, whereas the two European markets are alike; in other words, we can classify the four markets into either the Asian market group or the European market group.
The present paper has three distinctive features: first, as microeconomic variables, we include both market value and book value firm performance measures; second, we simultaneously examine both levels of and changes in CDS spreads; and third, we conduct two separate studies of our data by initially aggregating the CDS data from the four markets, and is then followed by a comparative analysis of the four markets.Our overall empirical results show that (i) Tobin's Q has a significantly negative impact on CDS spread levels and changes in all samples except for the Korean subsample, whereas ROA plays a

Literature Review
The literature on CDS spreads can be divided into two main strands.Studies in the first strand focus on reduced-form models and examine the random shocks that affect CDS pricing; such studies often employ an event study methodology.Studies in the second strand apply structural models under the assumption that CDS spreads are driven by the default risk of the CDS reference entity.Thus, researchers in this strand of literature believe that CDS spreads function as an indicator of default risk that is triggered when the reference firm's value falls below some threshold; in other words, the level of default risk is priced accordingly in CDS spreads.Although a vast body of studies have examined the determinants of credit spreads by utilizing both models, Collin-Dufresne, Goldstein, and Martin (2001) and Duffie and Singleton (1997) offer succinct summaries of the empirical findings to date.They maintain that the explanatory power of many theoretical models is rather limited and that further search for additional deterministic factors is desirable.

Reduced-Form Models
The development of reduced-form models began relatively recently in the 1990s.Some key researchers who have contributed to the development of such models include Lando (1994Lando ( , 1998)); Madan (2014).The underlying assumption of reduced-form models is rather different from that of structural models, in that the former treat default as an exogenously determined random shock, and as such, firm-specific factors or indeed any variables that could affect firm performance contain no information on the firm's default probability.Despite the development of reduced-form models, many researchers prefer to use structural models because they perceive the lack of an economic rationale for reduced-form models as a major obstacle to applying such models and explaining their results.

Structural Models
Structural models are based on the option pricing model originally developed by Black and Scholes (1973) and Merton (1974).Unlike reduced-form models, structural models provide an intuitive framework for the deterministic relationship between credit risk factors and CDS spreads.Recent studies based on structural ~ 12 ~ models, including Longstaff, Mithal and Neis (2005), Blanco, Brennan and Marsh (2005) and Tang and Yan (2006).These researches demonstrate that credit spreads have a negative relationship with interest rates and that while they vary with economic conditions, firm characteristics have significant explanatory power for credit spreads.Aunon-Nerin, Cossin, Hricko, and Huang (2002) study CDS determinants by examining both macroeconomic and firm-specific variables such as asset volatility, stock price, leverage, rating, and market capitalization, and they conclude that these variables explain up to 82% of CDS pricing.Equivalently, Abid and Naifar (2006) examine the explanatory power of a structural model by estimating variables such as ratings, CDS contract maturity, stock volatility, risk-free interest rates and the slopes of yield curves and report that these variables can help to explain more than 60% of CDS pricing.Other contributions to the study of CDS determinants using the structure model framework include: Acharya and Pedersen (2005), Tang and Yan (2007), and Chen, Lesmond, and Wei (2007).Specifically, Collin-Dufresne et al. (2001) investigate the determinants of CDS spread changes by using monthly US industrial bond data and find explanatory power for both firm leverage and implied volatility.Although with limited statistical evidence (25% explanatory power of observed credit spread changes), their paper highlights the elusive nature of some of the more fundamental problems in the search for factors that help to explain credit spreads.

The Hypotheses
Our study of CDS spread determinants is also based on the structural model approach, as we analyze both firm-specific and macroeconomic factors.To facilitate our investigations, we develop four testable hypotheses.Hypothesis One (H1) entails the testing of firm-specific factors in the CDS determination.Previous studies such as Ericsson et al. (2009) and Galil et al. (2014) include one firm performance related variable -leverage, in their work, our paper extends the investigation by introducing three firm performance variables.This approach adds further vigorousness to the CDS research, and it forms a major contribution of this paper to the understanding of CDS determinants.Hypothesis Two (H2) follows closely the spirit of those work such as Tang and Yan (2006); Duffie, Saita and Wang (2007); and Baum and Wan (2010), in which the influence of macroeconomic conditions is maintained and tested.Hypothesis Three (H3) shares the consideration of Stulz (2010) and Chiaramonte and Casu (2013), by speculating an impact of financial crisis on the CDS markets.Similar to Doshi et al. (2014), Hypothesis Four (H4) assesses whether regional factor plays a role in CDS determination.The four hypotheses are summarized as follows: H1: Microeconomic factors such as firm performance contain information about CDS and ∆CDS.H2: Macroeconomic conditions, as captured by GDP, stock market returns and the interest rate, have explanatory power for CDS and ∆CDS.

H3: The global financial crisis of 2007Q3-2009Q2 affected CDS and ∆CDS.
H4: A geographic effect plays a role in the determination of CDS and ∆CDS.

Method
As our dataset contains both cross-sectional and time-series dimensions, an unbalanced panel data estimation approach is adopted.Fixed effects with White Cross-Section Robust Standard Errors2 that

Variables
In

CDS Spreads
In this paper, we use "CDS" and "∆CDS" ( ) to represent CDS spread levels and CDS spread changes, respectively, which are the two independent variables in our regressions.

Firm Performance
As a major extension to Ericsson et al. (2009) and Galil et al. (2014), 6 three firm performance ratios are used in our study.The first ratio is return on assets (ROA), which is calculated by dividing a firm's net income (NI) by its total assets (TA). 7ROA measures firm profitability.As profitability increases, the probability of default decreases, and CDS declines.Alternatively, firm performance can be measured by 3 A key assumption of the least squares regression is that no omitted variables are correlated with the explanatory variables; otherwise, the estimates would be biased.The advantage of using fixed effects is that by assuming a constant i α , where as a unique constant for each firm, we can include the unobservable variable z in the equation , thus rendering the least squares method possible.In this setup, the slope coefficient β is the same for all i cross-sectional entities; however, the intercept terms i α vary across i but are constant over time.As we incorporate firm-specific effects on the CDS spread relationship, a model allowing for a different intercept for each individual reference entity would be the preferred estimation technique.Furthermore, our sample includes CDS spreads and firm-specific accounting data; hence, they would unlikely satisfy the standard assumption for estimating random effects on a random sample.
~ 14 ~ return on equity (ROE), which is equal to net income divided by total equity (TE). 8As ROE represents the return to shareholders on their equity, a higher ratio indicates a lower likelihood of default and therefore a lower CDS.Another commonly used indicator of firm performance is Tobin's Q (TBQ), which is defined as the ratio of the market value of the firm to the replacement cost of its assets. 9When Tobin's Q is greater than one, the current value of a firm's assets is higher than the replacement cost, i.e., the firm is performing well, and the probability of default and CDS decline.The same logic can be applied to relationship between CDS spread changes and changes in these firm performance indicators.For example, when ΔROA is positive, the firm is performing well, and ΔCDS will be negative; hence, we continue to expect a negative relationship.Since the calculations of the three firm performance ratios share data such as net income, total income and total equity, a high degree of correlation may exist among them.Table 1 reports the correlations among these ratios, and the correlations between ROA and ROE range from 55% to 81%.To avoid potential multicollinearity problems, we estimate the three firm performance indicators separately in Model 1 and Model 2.

Macroeconomic Conditions
A number of authors recognize the importance of macroeconomic conditions in the determination of credit spreads.For instance, Fama and French (1989), Korajczyk and Levy (2003) and Duffie et al. (2007) all document the contribution of macroeconomic conditions to credit spreads.In this paper, we examine the effects of a country's macroeconomic conditions from three perspectives: economy, financial, and interest , where MVE is the market value of equity and DEBT represents the firm's book value of debts (liabilities).MVE is the product of the bank's closing share price at the end of the financial year and the number of common stock shares outstanding, DEBT is the book value of the bank's short-term debt plus the book value of the bank's long-term debt and TA is the book value of the total assets of the bank.As stated above, all of these required inputs are readily obtainable from the bank's basic financial and accounting information.~ 15 ~ rates.
10 Fluctuations in GDP are important indicators of the macroeconomic condition of an economy (e.g., Tang and Yan, 2006).We study the economic health of a country by using two GDP variables -GDP growth rate (YGRT) 11 and GDP volatility (YVOL), which is the conditional volatility obtained from estimating a GARCH (1, 1) model. 12Our volatility measure is similar to that of Byrne and Davies (2005), Driver, Temple and Urga (2005), and Baum and Wan (2010).We expect a negative relationship between GDP growth and CDS spreads because when the economy is growing, business confidence increases, firm profitability rises, and hence CDS spreads decrease.By contrast, we expect the opposite relationship for GDP volatility, i.e., when fluctuations increase, economic uncertainty rises, the probability of firm default increases, and hence changes in CDS spreads increase. 13by using the stock index closing price (P) of the relevant country, and analogous to the GDP volatility computation, the stock market volatility (SVOL) calculation relies on the estimation of a GARCH (1,1) model.We expect that as stock market returns increase, economic confidence rises, and CDS decreases.Hence, a negative relationship should exist between stock market volatility and CDS.Moreover, we expect a positive relationship between CDS spreads and stock market volatility because when the business climate becomes more unstable, stock market volatility increases and CDS increases accordingly.As for the changes in the economic environment, we expect the same relationship for ΔCDS: when the change in economic volatility is positive, i.e., stability decreases, ΔCDS increases.
Many previous studies include risk-free interest rates in their analyses.For instance, both Longstaff and Schwartz (1995) and Blanco et al. (2005) show that the risk-free rate contains information about CDS spreads.To study its effects, we use the 5-year swap rate (SWP) 14 as a proxy for the risk-free interest rate, which determines the risk-adjusted drift of firm value.Therefore, an increase in the risk-free rate would tend to decrease risk-adjusted default probabilities and hence CDS spreads.We thus expect a negative relationship between the risk-free rate and CDS spreads.A positive change, i.e., rise, in the risk-free rate signals a decline in default probability, and hence, CDS spreads will fall.However, for a negative change, i.e., fall, in the risk-free rate, a negative ∆CDS follows, and vice versa; therefore, a negative relationship between the risk-free rate and ∆CDS is expected.

Crisis
The impact of the global financial crisis that began in 2007 has become an important consideration in recent research on the determinants of CDS spreads (see, e.g., Chiaramonte & Casu, 2013;Kress, 2011;Stulz, 2010;and Dickinson, 2008).To capture the potential effects of this global financial crisis on our hypothesized CDS spread relationships, the dummy variable Crisis is included in our estimations.This dummy variable is constructed by assigning "1" to the period from 2007Q3 through 2009Q2 and "0" for the rest of the sample.Many studies 15 on the crisis effect divide their full sample into pre-crisis and postcrisis periods and examine the differences between them.Owing to the relatively limited size of our sample, we believe that directly containing a crisis variable in our regression equations would be the preferred approach to minimize the small sample bias problem.For the purposes of our hypothesis testing, a positive relationship is expected between the crisis variable and our CDS and ∆CDS dependent variables.Table 2 displays the expected signs of the explanatory variables discussed above.

Models
Generally, the four hypotheses underlying our study can be represented by the following equation: where, n = 1,..,4 (number of markets).i is the number of reference entities: i = 9 for KOR; i = 8 for HKG; i = 54 for FRA; and i = 41 for GER.The sample periods are as follows:

Data
All the data for our estimations are collected from Bloomberg, and calculations are performed wherever necessary to compute the required variables.Although daily CDS and stock market index prices are available, our samples are constrained by the availability of GDP and firm-level balance sheet and income statement data, which are listed only annually (South Korea); semiannually (Hong Kong and France); and quarterly (Germany).In the first three cases, following Collin-Dufresne et al. (2001), Tang and Yan (2007), and Ericsson et al. (2009), we apply linear interpolation to obtain quarterly data for the first three markets.To ensure the dynamic nature of our dataset, we omit firms with inactive CDS spread changes for four or more consecutive quarters.In total, we have 112 single-name reference entities with 3931 ~ 18 ~ quarterly observations.The names of the reference entities and their credit ratings according to the three main ratings agencies are presented in the Appendix.Figure 1 graphically displays the relationships among our explanatory variables in the four markets over time.From the left sides of the four panels, we observe that stock market returns fluctuate more than GDP growth and swap rates.These changes are supported by a high degree of stock market volatility in the right-hand panels for each market, particularly for HKG and GER.Excess movements in the KOR and HKG stock markets during 2007-2008 are also clearly visible.This figure reveals that the global financial crisis affected these variables.interest rate leads to a decrease in CDS spread levels and changes, clearly supporting our theoretical understanding of their relationship.

4-Market
Furthermore, the magnitude and negative sign of all six variables mentioned above demonstrate that the economic significance of the firm performance measures (TBQ and ΔTBQ) is rather strong: the factor loading on these two variables is three times larger than the factor loading on the risk-free interest rate (SWP and ΔSWP) and over 80 times larger than the factor loading on the business climate indicators (STRN and ΔSTRN).It is also very interesting that all of the above variables are market value indicators rather than book value indicators, implying that risk managers and policy makers should pay more attention to market data in forecasting the default risk of the reference entities.The following differences in results between Models 1 and 2 are notable: 1) ROA contains information on CDS in the expected manner; 2) the recent global financial crisis of 2007-2009 had marginal effects on CDS; and 3) adjustments in CDS in response to the explanatory variables are rather slow; hence, significant AR(1) coefficients persist in Model 1.  and YVOL (ΔYVOL) contain information on CDS (ΔCDS).Moreover, ΔSRTN and ΔSVOL are found to affect ΔCDS, whereas ROA is found to possess explanatory power for CDS.All statistically significant coefficients take the expected signs in both the CDS and the ΔCDS spread specifications.Listed in Table 5, the estimation results for HKG rather differ from those for KOR.Specifically, the reported results show that TBQ (ΔTBQ), SRTN (ΔSRTN), SVOL (ΔSVOL), and SWP (ΔSWP) are important determinants of CDS (ΔCDS).Again, we obtain the expected signs for all the statistically significant coefficients, and ROA continues to be the variable with explanatory power for CDS.

Hypothesis Testing Synopsis
Synthesizing our overall results, we gather the following statement regarding our four hypotheses: (i) TBQ (ΔTBQ) has a significantly negative impact on CDS (ΔCDS) in all samples except for Korea, whereas ROA plays a significantly negative role in explaining CDS mainly in the full sample and in some individual subsamples.This result lends general support to H1.
~ 24 ~ (ii) SRTN (ΔSRTN) and SWP (ΔSWP) have a significantly negative impact on CDS (ΔCDS) in all samples except for Korea, whereas YGDP (ΔYGDP) is significant with the expected sign only for Korea and France.Thus, H2 cannot be rejected.Moreover, followed by SWP (ΔSWP) and STRN (ΔSTRN) in most cases, TBQ (ΔTBQ) has the greatest magnitude in terms of economic significance and a negative sign, which reemphasizes the importance of the market value indicators in developing risk management strategy.
(iii) A significantly positive relationship is observed between Crisis and CDS mainly in the full sample, which suggests that H3 is weakly supported.
(iv) YVOL (ΔYVOL), and in particular SVOL (ΔSVOL) are found to have a significant positive impact on CDS (ΔCDS) only in Asian economies, implying that the two Asian markets are more sensitive to market volatility.In other words, market players might have more confidence in the two European economies, which marginally supports H4.

Firm Performance Dummy
To account for all of the firm performance information embedded in ROA, ROE, and TBQ and to enhance the robustness of our analysis, we construct a performance dummy (PDMY).This dummy variable is created by assigning the value "1" whenever either of the two accounting ratios has a change that is greater than zero, and "0" otherwise.In addition to making use of all three performance ratios, this performance dummy offers with us the opportunity to resolve the occasions in which one of the ratios is either not changing or even changing in the opposite direction of the other two ratios.Such a situation is likely to occur when variables measured in both book and market values are used. 16Equations (3a) and (3b) below present the structure of the hypothesized relationship.
Both the dependent and the explanatory variables are identical to those in Equations (1a) to (2c) above, except that firm performance variables such as ROA, ROE and TBQ are now replaced by the two dummy variables PDMY and ∆PDMY.
The estimation results of Equations (3a) and (3b) are presented in Table 8.Generally, we can see that the results share a similar pattern with the prior results.However, one difference emerges.In the HKG, FRA and GER markets, the prior results indicate that some firm performance indicators are statistically significant variables in explaining both CDS and ∆CDS; however, these factors no longer possess explanatory power when they are instead captured by PDMY and ∆PDMY.Nevertheless, given the consistently significant appearance of the stock market and interest rate variables, overall, the results support the claim that our regression results are robust to the use of alternative explanatory variables-in the case of performance indicators-and the results from the FRA market further support our contention.Overall, we find that in many cases, the differences in the strength of our estimations used to explain CDS spread levels and changes are small and that the coefficients take the expected signs.

Analyses of the Goodness-of-Fit and Redundant Fixed-Effects Test Statistics Figure 2. Goodness-of-Fit analysis
To gauge the goodness-of-fit of our regressions, we also performed an analysis of the adjusted R 2 .Figure 2 displays values of the adjusted R 2 associated with all of the regressions that we have estimated.From the top half of the figure for models with CDS as the dependent variable, we observe that GER has the lowest goodness-of-fit results, while FRA yields the highest goodness-of-fit values.Moreover, all four performance measures produce comparable adjusted R 2 values within an approximately 10% range.The bottom half of the figure for models with ΔCDS as the dependent variable shows that GER again has the lowest adjusted R 2 and that KOR yields the best goodness-of-fit results.Indeed, the goodness-of-fit difference between these four performance measures and the five market categories has widened substantially to approximately 37%.  2, regressions employing CDS spread levels as the dependent variable (Model 1) produce a set of relatively high adjusted R 2 values that range from 0.681 to 0.798, whereas those using CDS spread changes as the dependent variable (Model 2) generate relatively lower adjusted R 2 values that range from 0.056 to 0.428. 18However, according to the redundant fixed-effects test F-statistics, 19 although Model 2 17 See, for example, ∆ROE, which has the largest difference (0.428 -0.056) = 0.372. 18As further support for our claim, the models estimated by generates lower adjusted R 2 values, it seems to produce less biased results than Model 1.Therefore, to ensure the robustness of our results, we maintain that a model with a reasonably good adjusted R 2 value and relatively smaller F-statistics would be preferred and considered more reliable to derive our conclusion.Accordingly, estimation results using ∆CDS as the dependent variable together with changes in certain explanatory variables appears to yield findings that fulfill these criteria.

Conclusions
This paper attempts to study the determinants of CDS spread levels and changes by using a panel dataset covering 112 reference entities from four markets over the period 2001-2012.Employing a structural model, we establish eight equations incorporating variables that could affect the default risk of a reference entity and hence CDS spreads.Our empirical results suggest that both firm performance and macroeconomic conditions possess significant explanatory power for CDS spreads; however, market value indicators (i.e., Tobin's Q, stock market returns and the interest rate) appear to be much more important than book value indicators (i.e., ROA, ROE, and GDP growth rate) in determining CDS spread levels and changes.Followed by the interest rate and stock market returns, Tobin's Q demonstrates the strongest economic significance among the market value indicators.Therefore, both H1 and H2 cannot be rejected.H3 also cannot be rejected because the global financial crisis of 2007 significantly affect global CDS markets as a whole, but it generally did not affect the individual markets under study.The results also show that only the Asian CDS markets in the sample are sensitive to both GDP and stock market volatility, whereas the two European markets are free from such an impact.This finding lends clear support to H4, which argues for the existence of geographic effects.
On the basis of our empirical results, we can assert that any government policy that could help provide a stable stock market and generate economic growth would facilitate the functioning of CDS markets and thus enhance the use of CDSs as a risk management tool for the investment community.In particular, both risk managers and financial regulators are encouraged to devote greater attention to the market value indicators of firm performance and macroeconomic conditions.Considerable weight should be given to Tobin's Q, the risk-free interest rate and stock market returns in risk pricing.For actors dealing with CDSs in Asian markets, economic and stock market volatility should also be covered closely.Notwithstanding, while some fundamental determinants of CDSs remain elusive, further research on this subject by extending the number of markets and embracing some market-specific geopolitical variables that help produce a more precise picture of the effect of market factors on CDS pricing across countries would certainly be beneficial.This paper contributes to the research of CDS determinants in two ways.Firstly we highlight the importance of market/geographic effect, and secondly, to our best knowledge, we are the first to test for the explanatory power of our three firm performance -market and book value indicators in the formation and movements of CDS spreads.

Figure 1 .
Figure 1.Individual market -GDP growth, stock market return and swap rate movements (level and volatility)

Table 1 .
ROA, ROE, and Tobin's Q correlations Arnold and Vrugt (2008)07)resne et al.(2001),Acharya and Johnson (2007)andArnold and Vrugt (2008)show that stock market returns and volatility are important indicators of changes in the business climate.According to the contingent-claims framework, the features of a CDS resemble those of a short put option.As volatility increases option values, the link between CDS spreads and volatility becomes apparent.A positive stock return signifies a healthy business climate, and default risk is hence lower, or the probability of recovery is higher.By contrast, a more volatile stock market increases the likelihood of firm default.Overall, we believe that the functioning and movements of the stock market are important factors that we want to test in the CDS relationship.Unlike Collin-Dufresne et al. (2001) and Galil et al. (2014), who use a stock market volatility index, we calculate stock market returns (SRTN)

Table 2 .
Explanatory variable and expected signs on estimated coefficients Acharya and Johnson (2007)(level and change) of the reference entity i at time t; Micro represents the microeconomic conditions or, more precisely, a vector of firm-specific variables indicative of the financial performance of firm i at time t; Macro is a vector of variables (GDP, stock market, swap rate) that captures the macroeconomic conditions of the relevant market; and Crisis serves as a dummy variable that takes the value "1" for the period 2007Q3-2009Q2 and zero otherwise.To perform our hypothesis testing, two models are derived from equation (M).The formation of Model 1 is similar to that inTang and Yan (2007),Acharya and Johnson (2007)and Pires, Pereira and Martins (2010) in which CDS spread levels are investigated as the dependent variable and the levels of explanatory variables are examined.Model 2 is considered a dynamic version of Model 1 in which ∆CDS is used as the dependent variable and changes in both firm performance indicators and macroeconomic measures are studied.

Table 3
Galil et al. (2014)tion results from the 4-Market aggregated sample.The results from both Models 1 and 2 statistically demonstrate that both TBQ and ΔTBQ have explanatory power for CDS and ΔCDS, respectively, with the expected signs.Moreover, in all six equations, the highly statistically significant coefficients for STRN and ΔSTRN signify that both CDS and ΔCDS decline when stock market returns increase and when positive changes in stock market returns occur.Our results are in agreement withGalil et al. (2014), who report a negative and significant relationship.As explained above, we are not surprised by the existence of these negative slope coefficients, as they indicate that CDS spread levels and changes decrease whenever business performance or the macroeconomic environment improves.Further explanatory variables that produce consistently significant estimation results are SWP and ΔSWP.Again, we obtain the expected signs.The negative coefficients for these variables suggest that a rise in the ~ 20 ~

Table 4 .
Estimation results for KOR Notes: Associated t-ratios in parentheses.Significant statistics are in bold.***, ** and * denote statistically significant levels of 1%, 5% and 10% respectively.Intercept estimates are not shown.

Table 4
presents the results for KOR.The results of both models show that SWP (ΔSWP), YGRT (ΔYGRT)

Table 6
presents the results for FRA.As shown, ROA (ΔROA), TBQ (ΔTBQ), YGRT (ΔYGRT), SRTN (ΔSRTN), and SWP (ΔSWP) have significant relationships with CDS (ΔCDS).Moreover, ΔROE contains information on ΔCDS.While we obtain the expected signs for the significant coefficients except for YVOL, this is the first time where we find explanatory power for all firm performance indicators within the same model-Model 2. The empirical results for GER are provided in Table7.Only TBQ (ΔTBQ), SRTN (ΔSRTN) and SWP (ΔSWP) exhibit strong deterministic relationships with CDS (ΔCDS).Again, all of the statistically significant estimated coefficients take the expected signs.

Table 7 .
Estimation results for GER

Table 8 .
Estimation results for firm performance dummy variable Notes: Associated t-ratios in parentheses.Intercept estimates are not shown.Significant statistics are in bold.***, ** and * denote statistically significant levels of 1%, 5% and 10% respectively.
Galil et al. (2014)03)results suggest that reference entities with lower credit ratings such as those in KOR and HKG exhibit greater explanatory power than those in other markets.Although this observation diverges from the results of Avramov, Jostova and Philipov (2007) andEricsson et al. (2009), it is in line with those ofHuang and Huang (2003)andGalil et al. (2014). Asshown in Figure The summary of redundant fixed effects test F-statistics is available upon request.