Kalã²S Kai Agathã²S? Government Quality and Cultural Heritage in the Regions of Europe

This paper uses panel data on over 200 regions of Europe to study the spatial distribution of UNESCO sites and the capacity of regional governments to conserve heritage, using new designations in the World Heritage List as a proxy. We test whether the location of a region matters by controlling for the stock of World Heritage in the surrounding regions, and if low regional government quality is an obstacle to inclusion of sites into the List. We find some evidence of within-country regional competition for inscription, and of a positive impact of government quality on the chances of having a UNESCO designation.  

This paper explores factors that affect the distance between sovereign credit ratings and the ratings assigned to new foreign-currency bonds issued by sub-sovereign entities (such as private non-financial corporations, financial firms, and public sector enterprises) in 47 emerging markets and developing economies. Censored and double-hurdle regression models are used to estimate the relative contributions of bond-level, issuer-level, and macroeconomic factors that determine this distance, separately for those rated at or below the sovereign rating and those rated above. For the three-quarters or more of sub-sovereign bond ratings that are constrained by the sovereign rating ceiling, a Tobit regression model shows a smaller distance-suggesting stronger sovereign-corporate linkages-for public sector enterprises and financial firms relative to other firms. Riskier global financial conditions are also associated with sub-sovereign bonds being rated closer to the sovereign rating. For the small number of sub-sovereign bonds rated higher than the sovereign rating, a double-hurdle model shows that certain debt features-such as bonds backed by future-flow receivables or other collateral, or structured as Special Purpose Vehicles (SPV)-significantly raise the likelihood of piercing the sovereign rating ceiling and also increase the distance above the sovereign ceiling.

Introduction
The credit rating of an emerging market sovereign is a key factor influencing the ratings of subsovereign entities including non-financial corporations, banks, public sector enterprises, and local governments. 1 Several recent studies have demonstrated the impact of sovereign credit ratings on corporate ratings (Ferri and Liu, 2003;Borensztein, Valenzuela and Cowan, 2013), on bank credit ratings (Alsakka, ap Gwilym and Vu, 2014;Huang and Shen, 2014), on bank stock returns (Correa et al. 2014), and on real economic activities (Almeida et al., 2017). Other related studies have considered the implications of changes in sovereign yield spreads for corporate borrowing costs in developed countries (Bedendo and Colla, 2015) and in developing countries (Durbin and Ng 2005;Dailami, 2010;Dittmar and Yuan, 2008). 2 These suggest a significant spill-over impact of sovereign debt impairment on corporate performance, whereby corporates in emerging markets tend to be more exposed to sovereign rating changes than in developed countries (Ferri and Liu, 2003).
The sovereign rating often acts as a "ceiling" for the sub-sovereign ratings in most instances, although the ratings of the sub-sovereign entities have sometimes exceeded the sovereign rating.
Historically, the likelihood of sovereign defaults and the default premia of firms in the same jurisdiction were highly correlated in emerging market economies, justifying the practice of "sovereign ceiling" as widely used by rating agencies (Grandes and Peter, 2006). Since 2001, however, rating agencies have relaxed this policy with a view that the correlation between the two has weakened (Moody's, 2005;S&P, 2013). Following the 2008-09 global financial crisis, Standard and Poor's (S&P) reconsidered its sovereign ceiling policy and redefined the maximum notch difference between non-sovereign entity's rating and the sovereign rating depending on sector and region where the firm operates (S&P, 2013).
The sovereign rating often serves as a benchmark for the capital-raising activities of both the public and private sectors by affecting their borrowing spreads, particularly when they access international capital markets and issue debt denominated in foreign currencies. The past episodes in emerging market economies confirm significant negative externality from sovereign downgrades to downgrades of public enterprises, banks and private sector entities, by raising their funding costs and potentially triggering covenants (Augustin et al., 2016;Chow, 2015; Williams, Alsakka and ap Gwilym, 2013). The deterioration in balance sheets of public enterprises and banks could negatively feed back to the sovereign via credit guarantees or other financial supports, augmenting their sovereign-corporate linkages. 3 Since the relaxation of sovereign ceiling policy, corporate ratings that "pierce" the sovereign ceiling has become more common (Almeida et al., 2017), and bonds issued by such sub-sovereign entities could be more sensitive to changes in the sovereign rating.
Despite a growing literature on sovereign-corporate linkages, there is very little evidence on how far the ratings of sub-sovereign bonds would be generally positioned above or below the corresponding sovereign rating. In particular, as the sovereign ceiling policy has recently been relaxed, further empirical analysis is needed to identify key factors that determine the strength of sovereign and sub-sovereign rating linkage when bond issuers are constrained or unconstrained by the sovereign ceiling. This paper revisits the effect of sovereign creditworthiness on sub-sovereign ratings using data on foreign currency bonds issued by sub-sovereign entities in 47 emerging markets and developing economies (EMDEs) during 1990-2013. We use censored and double-hurdle regressions to estimate relative contributions of each factor to determine the "distance" between sovereign and sub-sovereign ratings, separately for bonds rated at or below the sovereign (bounded bond ratings) and above the sovereign (non-bounded bond ratings). We also merge the bond-level dataset with balance sheet data from Bloomberg to estimate the importance of firm's financial situation in affecting the distance for a subset of matched debt issuers.
The results shed light on the asymmetry in factors which determine the distance between bounded and non-bounded bond ratings. For bounded bond ratings, their distance is mainly determined by bond issuers' ownership structure, size of their assets, and sector of issuer. The distance tends to be narrower as the country's trade linkages become stronger and global 3 In related work, Williams et al. (2013) and Huang and Shen (2014) found that the impact of sovereign credit ratings on bank credit ratings is significant but varies by countries' macroeconomic and institutional factors. Correa et al. (2014) found that a downgrade of sovereign credit rating has a large negative effect on bank stock returns for those banks that would receive stronger support from their governments. Ağca and Celasun (2012) also outline channels through which higher public external debt (which is typically associated with lower sovereign credit ratings; see Cantor andPacker 1996 andS&P 2013) can raise corporate default premia. 4 financial condition gets riskier. A country's capital account openness and domestic credit market development also influence the distance. Debt characteristics, including whether the bond issue is callable and its maturity, matter for both bounded and non-bounded bonds, although the bond securitization and collateralization mainly determines whether and how high the corporates are rated above the sovereign for non-bounded firms. Finally, a Heckman selection estimation confirms the robustness of our main results to sample selection bias in debt issuance.
As emerging markets liberalize their capital accounts, firms from emerging countries have been raising a substantial financing through bond issuances in international markets since 1990s (Gozzi et al., 2015). However, the number of rated firms with access to international markets is still limited for lack of information in assessing prospective debtors' credit risk. Our results underscore the relative importance of bond-level and firm-level attributes in estimating their prospective bond ratings compared to country-level and global factors.
The paper is structured as follows. Section 2 describes the data and presents some descriptive statistics. Section 3 explains our empirical strategy, and sections 4 and 5 provide empirical results. Section 6 concludes.

Sovereign ratings and sub-sovereign foreign currency bond ratings
The sample includes 5,033 long-term foreign currency sub-sovereign bonds rated by the three major international rating agencies (S&P, Moody's and Fitch Ratings)  Bond ratings by the three major international agencies for sub-sovereign foreign currency debt and other bond characteristics were obtained from Dealogic. Sovereign credit ratings were obtained from Bloomberg and Thomson Reuters. Debt in default was excluded from the sample.
The conversion from letter to numeric ratings on a 1-21 scale, with higher values indicating 5 better ratings, broadly follows Cantor and Packer (1996), Gande and Parsley (2005), and Hill, Brooks and Faff (2010) and is described in table 1.

around here]
As capital markets of emerging economies mature, local currency debt issuance has also started to increase. We however do not include local currency debt in our analysis for three reasons.
First, local currency sub-sovereign debt is typically not subject to sovereign risk, in particular to convertibility and transfer risks. Hence, the sovereign ceiling does not appear to apply to local currency bonds. 4 Second, local currency debt is typically not subject to the problem of 'original sin' typically faced by developing countries when they are unable to borrow abroad in their own currencies (Eichengreen, Hausmann, and Panizza, 2005). 5 Third, local currency debt is often rated by national rating agencies on rating scales that can differ substantially from those used by international rating agencies for foreign currency debt.
Figures 1a-1c show scatterplots of sovereign and sub-sovereign ratings, by public and private issuers and financial and non-financial issuers for the three rating agencies. These figures illustrate the constrained nature of the data, where most of sub-sovereign ratings are bounded by the sovereign rating, but with several bonds rated higher than the sovereign. These figures also illustrate the strong correlation between the sovereign and sub-sovereign debt ratings for public sector debt issuers, with ratings of public sector sub-sovereign issuers typically close to the sovereign rating (along 45 degrees line) compared to the private sector, and for financial issuers compared to non-financial issuers. These are consistent with the earlier discussion that public enterprises and sub-national authorities, as well as financial issuers, are likely to be closely related to the sovereign.
[Insert Figure 1 around here] Summary statistics for ratings of foreign currency bond issues are presented in table 2. In general, the average sub-sovereign rating is lower than the average sovereign rating across the three rating agencies. There is considerable variation in the distribution of sub-sovereign ratings. 4 Grandes and Peter (2006) estimate the elasticity of corporate spreads with respect to sovereign spreads in South Africa, finding that the sovereign ceiling does not appear to apply to local-currency denominated corporate bonds. 5 The original sin refers to the limited ability of poor countries to issue international debt in their own currencies. In consequence, these countries often resort to borrowing from international markets in the major foreign or "hard" currencies, such as the US dollar, euro, yen and British pound. 6 About 38-43 percent of sub-sovereign bonds had the same rating as sovereign rating and 36-46 percent had lower rating than the sovereign rating-adding up to 74-89 percent of sub-sovereign bonds rated at or below the corresponding sovereign rating. This confirms that the majority of sub-sovereign bonds are subject to a sovereign ceiling. The share of sub-sovereign bonds rated above the sovereign rating is relatively small at 11-26 percent of all bonds (12.3 percent for S&P, 25.9 percent for Moody's and 11.0 percent for Fitch), with average distance above the sovereign rating of 2.4 to 4.0 notches on our rating scale.
[Insert Table 2 around here]

Control variables
We use a range of global, country-level, and bond-level variables to account for the factors that can influence the sovereign and sub-sovereign rating relationship. The summary statistics for these are presented in table 3. The panel of charts in Figure 2 shows the evolution of the annual average of global and country-level macroeconomic variables since the early 1990s, together with the average distance between sovereign and sub-sovereign ratings. Pairwise correlations between the distance and macroeconomic variables are provided in the chart. [Insert Figure 2 around here] Second, the US 3-month treasury bill rate is used as the benchmark global interest rate, given that a majority of sub-sovereign bonds was issued in U.S dollars and the outsized importance of the U.S Federal Reserve in influencing global financial conditions (Rey 2015). A low global interest rate will lower the financing cost for firms to access capital, and hence may widen the distance between the sovereign and sub-sovereign ratings. The second chart in Figure 2 shows that the Fed Funds rate was high in the 1990s, fell during the early 2000s following the dot-com bust (Ferguson and Schularick 2007), rose again in the subsequent years, and was sharply lowered during the global financial crisis (Taylor 2009).

around here]
Country-level cyclical factors such as GDP growth and CPI inflation would also directly affect the performance of sub-sovereign entities (Dailami 2010). In our sample, average growth rate at the bond issuance is positive at 4.6 percent, while average inflation is 5.9 percent with a range of -1.4 to 85.7 percent. Higher GDP growth is usually associated with better performance of the private sector, which would help sub-sovereign entities service their debt obligations. But improved cash flows may also reduce the dependence of a sub-sovereign on the sovereign credit rating. The third chart in Figure 2 shows a general co-movement between the rating distance and GDP growth. High inflation reduces the real value of future earning of firms and jeopardizes macroeconomic stability, thereby reducing the willingness of sub-sovereign entities to plan future investments. Higher inflation is expected to be associated with a greater dependence of sub-sovereign ratings on the sovereign. Average inflation for our sample of countries has fallen from double-digit levels in the 1990s to 5-7 percent in recent years (fourth chart in Figure 2), and appears to show a negative relationship with average distance over time.
The country's level of financial development and external openness affect corporate performance due to the difference in access to domestic and external financing. Domestic credit (as a share of GDP) is an indication of the depth of the domestic financial sector, and ranges from 3.9 to 192.7 percent. Despite the variation across countries, for the group of emerging market countries, average domestic credit as a share of GDP has risen almost steadily since the mid-1990s (fifth 8 chart in Figure 2). More mature domestic financial markets may weaken the sovereign-corporate linkage, but a large increase in domestic credit could also strengthen macro-financial linkages, thereby increasing exposure to external shocks (Lane and Milesi-Ferretti 2011).
Regarding external openness, we control for both openness in trade and capital account. Capital account openness (measured by the Chinn-Ito index Ito 2006, 2008)) exhibits significant variation across countries, but on average, increased in the years leading up the global financial crisis, and then fell slightly in subsequent years (sixth chart in Figure 2). Greater capital account openness may weaken the sovereign-corporate linkage as it provides firms greater access to foreign capital to service foreign-currency debt obligations. Hence, the distance between the sovereign and corporates may widen when the capital account is relatively more open.
Trade openness (export and import values as a share of GDP) for the group of emerging market countries rose strongly in the years leading up to the financial crisis (seventh chart of figure 2).
After a recovery from a steep fall during the 2008-09 global financial crisis, trade growth has slowed in the post-crisis years. Although exports provide foreign currency revenues, stronger trade linkage may tighten the sovereign-corporate linkage as the private sector (mainly tradable sector) is more exposed to changes in demand of the trading partners.

Bond-level characteristics:
Among the various factors which could affect the distance between sovereign and sub-sovereign ratings, perhaps the most relevant are bond issuers' ownership structure (public vs. private) and sector of issuer (financial vs. non-financial). As shown in table 3, financial firms and stateowned entities represent 57 percent and 37 percent of the overall sample respectively. Due to the relatively strong macro-financial linkages in emerging economies, banks and other financial firms (especially large institutions) typically benefit from explicit and implicit guarantees from the central bank or government in case of financial distress. Hence, debt issued by financial firms is likely to be rated closer to the sovereign rating, compared with debt of non-financial firms.
Similarly, public enterprises in emerging markets and developing countries, even if nominally free of government control, tend to have privileged access to government support during times of distress, thus are also likely to be rated closer to the sovereign rating.
Similar to the findings of Gozzi et al. (2015), differences in bond characteristics (listed below) can affect the sovereign-corporate relationship: 9  Debt issue size: The debt issue size captures the importance of the issuer in its market and its access to a broad range of financing sources. For the same borrower, though, a larger debt may carry higher risk and hence result in a lower rating. The average logarithm of issue size is 18.9, about 163 million in constant 2010 US dollars.
 Maturity: Bonds with longer maturity may receive a lower rating than shorter maturity debt as long-term bonds accompanies larger uncertainty than short-term bonds. Average maturity is about 7 years in the sample.
 Fixed vs. floating rate: Fixed rate bonds typically carry a higher yield than floating rate bonds to compensate the bond-holder for holding the debt to maturity at a given interest rate.
Fixed rate notes account for the bulk (82 percent) of sub-sovereign bonds.

Firm balance-sheet variables:
To test the robustness of our main results obtained from bond-level data to inclusion of balance sheet variables (see section 5), the data on international bond issuance was matched with firmlevel balance sheet data from Bloomberg based on International Securities Identification Numbers (ISINs). The inclusion of balance-sheet data reduces the sample size to a subset of bonds for which such matched data are available. The balance sheet variables include the logarithm of total assets (in thousands of constant 2010 US dollars), net profit as a percent of assets, financial leverage (total assets to common equity ratio), and capital expenditure as a share of assets. 10

Sub-sovereign bonds rated at or below the sovereign rating (bounded bond ratings)
In the first specification, we consider the sample of sub-sovereign bonds that are rated at or below the corresponding sovereign rating and consider the determinants of the distance of the sub-sovereign rating below the sovereign rating. The specification in equation (1) below is estimated using a Tobit model which is appropriate due to the clustering of a large number of observations with distance at zero. * = + + + Θ The dependent variable * = ( − ) is the difference between the sovereign and sub-sovereign ratings by the same rating agency. i refers to the debt issue, j refers to country, k refers to industry and t refers to time period. The dependent variable takes on a value of zero for sub-sovereign bonds rated at the corresponding sovereign rating and a positive value for bonds rated strictly below the sovereign rating.
The baseline specification (1) uses the sovereign and sub-sovereign ratings of the three rating agencies separately to account for the differences in the weights that rating agencies assign to the various explanatory variables. 6 Equation (1) Borensztein et al. (2013) discuss the "sovereign ceiling lite hypothesis", where corporates can be rated above the sovereign rating, and thus a ceiling does not impose an absolute constraint on corporate ratings (see also Almeida et al., 2017). In order to account for the corner solution properties of our bond rating data, we use a double-hurdle model (Cragg, 1971) for the determinants of the distance of the sub-sovereign bond rating above the sovereign ceiling. This distance is modeled as a function of debt characteristics, firm-level attributes, and country and global factors in a second-stage equation, conditional on a first stage equation for the probability that a bond is rated above the sovereign rating. An advantage of the double-hurdle model is that it allows the effect of the covariates to vary across the two stages.

Sub-sovereign debt rated above the sovereign ceiling (non-bounded bond rating)
The first stage equation investigates factors affecting the probability that a sub-sovereign bond rating is higher than the sovereign rating using the probit model defined in equation (2).
The dependent variable is a binary variable that takes on a value 1 when the sub-sovereign rating is higher than the sovereign rating and 0 otherwise. Using equation (2) as the first stage, the second stage of the double-hurdle model regresses the distance of the sub-sovereign bond rating above the sovereign rating on the same set of explanatory variables, as defined in equation (3) below: * = + + + Θ The dependent variable * = ( − ) is the notch difference of the sub-sovereign rating above the corresponding sovereign rating by the same agency. AMEs for the relevant groups are presented for both the first and second stage regressions. 7 The standard errors of the regression were clustered at the level of countries in all regressions.
12 Additionally, we include an indicator for bonds backed by future-flow receivables, as such securitization structures may help to pierce the sovereign rating ceiling (Ketkar and Ratha 2008).
We also include an indicator for Special Purpose Vehicle (SPV), which includes bonds involving securitization of assets other than future flow receivables, and an indicator for other collateralized bonds.

Discussion of results
This sub-section discusses the results of the benchmark regressions as defined in equations (1)- (3) in the previous section.

Results for bounded bond ratings
The results for the bounded bond ratings are presented in columns (1)-(3) of table 4. The coefficient presents how much each covariates contribute to narrow or widen the distance between sovereign and sub-sovereign ratings from the average distance below the sovereign rating of 3.2 to 3.6 notches as shown in Table 2.
The indicator for public entities is statistically significant at 1 percent for all three agencies, with the coefficient ranging from -1.4 to -1.8 notches. This indicates that public enterprises are rated much closer to the sovereign ratings as observed in figures 1a-1c. The financial sector dummy is also negatively signed and highly statistically significant across the three agencies, with the coefficient in the range of -0.6 to -1.3 notches. This shows that the distance is smaller for financial firms than non-financial firms as observed in figures 1a-1c.
[Insert Table 4 around here] Among the macroeconomic variables, the global risk VIX index is negatively signed and statistically significant for all rating agencies, showing narrower distance between sub-sovereign and sovereign ratings if bonds are issued during periods of riskier global financial conditions. Spikes in the VIX index have been associated with increased co-movement in sovereign credit spreads (Longstaff, Pedersen, and Singleton, 2011), likely reflecting increased information asymmetries and herding behavior of investors during global "risk off" episodes. In such circumstances, international investors are likely to pay more attention to sovereign 13 creditworthiness when investing in firms in emerging market economies, which may explain the tighter relationship between sovereign and corporate debt ratings.
The coefficient for capital account openness is positive and significant for S&P and Moody's.
This suggests that corporate bonds tend to be rated closer to the sovereign when issued in countries with relatively closed capital accounts. This may reflect the lack of external financing sources for sub-sovereign entities in the closed economies, making their ratings more reliant on the sovereign. Conversely, greater capital account openness can widen the pool of available financing sources for firms and weaken the sovereign-corporate linkages.
The coefficient of domestic credit-to-GDP ratio is negatively signed, suggesting that bonds tend to be rated closer to the sovereign if the issuing country has higher credit as a share of GDP, thus entailing stronger macro-financial linkages. This is consistent with the findings of other studies that find that a large increase in credit is associated with greater likelihood of future crises [Insert Figure 3 around here] A one standard deviation increase in the VIX index is associated with a reduction in the distance by 0.18-0.26 rating notches. This suggests that despite the strong statistical relationship to the distance in most specifications, the magnitude of this effect is smaller than domestic credit. Only in a severe crisis situation when the VIX index spikes by the equivalent of three or four standard deviations, would the change in distance approach a change comparable in magnitude to that of domestic credit. The response of distance (0.3 rating notches increase) to a one standard deviation increase in capital account openness is slightly larger than VIX index. Going from a fully closed capital account to a fully open capital account would imply an increase in the distance by 1.2 rating notches for both S&P and Moody's. The response of distance to a one standard deviation increase in short-term US interest rate is 0.4 rating notches increase, larger than that for capital account openness and VIX, but is marginally significant only for S&P. The magnitude of response of distance to a one standard deviation increase in trade is larger but significant only for Moody's in the baseline model, as discussed earlier.
Among debt characteristics, log issue size is negatively signed for all three agencies, although statistically insignificant. The coefficient of debt maturity is negatively signed and statistically significant at 10 percent level for two out of the three agencies. On the other hand, the callable bond dummy is positive and statistically significant across all the specifications, indicating that they are typically rated with a greater distance. This "penalty" in terms of a larger distance may reflect the nature of the bond contract where the issuer has a right to prepay the bond prior to maturity.
There are concerns that a few countries where each represents more than 500 sub-sovereign bonds in the sample in Appendix table 1, drive the baseline results. Brazil, Russia, and South Korea account for more than half (54 percent) of our observations. The first three columns of Appendix We also consider the regional variations in the rating distance. We estimate the baseline regression with five regional dummies, taking the East Asia and Pacific region as the base 15 category. The coefficient estimates and statistical significance of the regional dummies for S&P [Insert Figure 4 around here]

Inclusion of additional covariates for firm balance sheet characteristics
In columns (4)

Results for non-bounded bond ratings
The results for the double-hurdle model in equations (2) and (3) are presented in table 5. The first three columns of table 5 provide the AMEs of the first-stage probit equation, while the next three columns presents the AMEs of the second stage regression for distance above the sovereign rating.
[Insert Table 5 around here] The results in table 5 suggest that securitization structures, such as future-flow receivables and SPVs, and collateralization, play an important role in achieving a sub-sovereign debt rating higher than the sovereign rating. The future-flow receivable securitization indicator shows significantly positive and large marginal effects on both the likelihood of being rated above the sovereign rating and the distance above the sovereign rating for all three rating agencies. The distance above the sovereign rating for sub-sovereign debt securitized by future-flow receivables is 1.6, 1.1 and 0.8 notches higher than other bonds for Moody's, S&P and Fitch respectively, after controlling for other determinants of the distance. The coefficients are statistically significant at the 1 percent level in both stages, indicating a robust relationship.
Bonds structured as SPVs, which are typically backed by existing assets, also tend to have a higher likelihood of being rated above the sovereign ceiling and to achieve a higher distance above the ceiling, although the AMEs are much smaller than that for future-flow securitizations.
In the first-stage regression, the coefficients of the SPV indicator are statistically significant at least at 5 percent level in most cases, except for Moody's. A similar pattern is observed for other collateralized bonds, where the coefficients are positive and statistically significant at 1 percent for S&P and Moody's in the second-stage equation. However, the magnitude of the effect is smaller than the effects for the future-flow and SPV indicators.
Among the global variables, the 3-month U.S. policy interest rate is negatively signed across most specifications, while the VIX index is positively signed and statistically significant at 1 percent. These suggest that (a) a looser monetary policy in the U.S. allows debt issuers to achieve higher rating than the sovereign; and (b) the likelihood of sub-sovereigns rated above the sovereign is higher amidst risker global financial condition. The country-specific variables such 17 as capital account openness, the domestic credit-to-GDP ratio, GDP growth, and inflation don't exhibit consistent signs and are insignificant in most specifications.
The public entity dummy is negative and significant in the first stage equation, showing lower likelihood of being above the sovereign ceiling. As suggested by figures 1a-1c, bonds issued by public entities tend to be rated close to the sovereign rating even when they pierce the sovereign ceiling. On the other hand, financial sector issuers tend to be rated above the sovereign rating, with greater distance above the sovereign ceiling, although statistically significant at 5 percent only for Moody's. As discussed earlier, this could reflect market expectations of implicit or explicit support from the country's central bank. Besides, the results suggest that debt-specific variables (issue size, maturity, fixed rate, callable) have relatively small marginal impacts or are insignificant in most cases for both margins.
As we did in sub-section 4.1, we exclude the top three countries and re-run the double-hurdle model with the remaining observations. The results are reported in Appendix    Table 6 around here]

Inclusion of balance sheet variables in double hurdle model
Overall, the double-hurdle estimation suggests that certain debt characteristics -in particular, a bond backed by future-flow receivables, SPVs, and collateralized bond -raise the probability of piercing the sovereign ceiling and increase the distance above the ceiling. The magnitude of the effect of future-flow securitization structures is the largest compared to SPVs and other collateralized bonds. Other debt characteristics, country-level variables, and global variables serve the role of useful controls but are in most cases themselves not robustly related to the sovereign ceiling.

Robustness check
In the analysis above, a large sample of sub-sovereign bonds was used to estimate the influence of sovereign ratings. However, as we only observe ratings when sub-sovereign bonds were issued, the sample may differ from the population of all bonds. In this section, we address the issue of possible sample selection bias using a two-step Heckman selection model to check the robustness of our baseline results.
Our identification strategy relates to Francis, Aykut, and Tereanu (2014) which reshaped loan and bond-level data into country-quarter panel to correct the selection bias by estimating the probability of corporate bond/loan issuance in the first stage using the average bond-level characteristics as an exclusion restriction. For running the selection regression, we similarly need to collapse the bond-level dataset at higher unit, at industry level for instance, to have observations for each month and country where a borrower did not issue any bonds. We decide to categorize bond issuers into eight industry types (based on their ownership type (public vs. private), industry (financial vs. non-financial), and the governing law they used for issuance, and take average sovereign and sub-sovereign ratings for each country-industry-month pairs to build a balanced panel. Should at least one firm in the same industry for each country issue bonds in a particular month using the same governing law, we observe the distance between the sovereign and sub-sovereign ratings for each country-industry-month pairs.
In several of the emerging market countries covered in our sample, only a few foreign currency bonds were issued during 1990-2013. For this reason, we restrict our sample to 22 emerging 19 countries where more than 30 foreign currency bonds were issued during the sample period to minimize sampling bias caused by a large number of censored observations. 10 In the selection equation, we use an institutional quality dummy (i.e., governing law is English common law or French civil law) as an exclusion restriction. La Porta et al. (1998) predict that common law system in the British tradition offers stronger investor protection than the system in the French civil law tradition, which in turn, promotes the development of market-based financial systems than bank-based financial systems (Levine and Demirguc-Kunt, 1999). Based on these works, English common law is expected to increase the probability of issuing a bond, but not directly affect the sub-sovereign rating itself. 11 In tables 7 and 8, the Heckman selection model for distance both below and above the sovereign rating are estimated separately for S&P (in columns (1)  is positive at around 0.6 for the distance below and above the sovereign, with both significant at the 1 percent level.

around here]
The second stage results in columns (4)-(6) present similar results to the benchmark specifications for each rating agency. The coefficient of public entity dummy is always negative and significant, while the coefficient of financial sector dummy is negative for those below the sovereign (table 7) and positive for bonds above the sovereign ( In all specifications, the selectivity effect lambda ( ) is positively signed (and significant in many cases) suggesting positive selection, i.e. that the observed sample of sub-sovereign bonds is likely to be rated higher than a (counterfactual) random sample. However, the main results remain broadly unchanged, which suggests that selection issues do not materially change our earlier finding.

Conclusion
This paper has investigated the relative contributions of bond-level, issuer-level, and macro-level conditions in determining how far sub-sovereign bond ratings would be generally positioned above or below the sovereign credit rating. The empirical results suggest that key determinants of the distance between the two are clearly different between the case when bond ratings are bounded and non-bounded by the sovereign rating ceiling. For non-bounded bond ratings, the analysis found that bonds are usually structured with some forms of securitization (future-flow securitization, SPV) or collateralization, which mainly determines how high the corporates are rated above the sovereign.
In most cases, however, sub-sovereign bond ratings are constrained by the sovereign ceiling. In this traditional case, our analysis suggests that the rating agencies are paying attention to corporate ownership structure and sectoral affiliation when evaluating the creditworthiness of sub-sovereign debt. The distance between the rating of sub-sovereign bonds issued by public enterprises and financial firms is closer to the sovereign rating, compared to other firms. This closer link suggests that public enterprises, banks and other financial firms benefit from the possibility of government support during times of financial stress. For bonds rated below the sovereign ceiling, riskier global financial conditions are also associated with international bonds being rated closer to the sovereign rating. Lack of capital account openness is associated with a smaller distance, reflecting reduced access to foreign financing. Well-developed domestic 21 financial markets also tend to be related to a smaller distance, likely due to stronger macrofinancial linkages.
The overall results remain stable when the three countries with the largest bond issuance are excluded from the sample. For a subset of firms for which we matched bond-level data to firms' balance sheet data, the main results are robust to the inclusion of balance sheet variables, with firms having larger asset size and more profitable firms rated closer to the sovereign rating. Our main results are robust to possible sample selection bias in bond issuance.
Given the increasing access of emerging market sub-sovereign entities to global debt markets, it is important to note that their cost of foreign currency debt financing is influenced by sovereign credit ratings, especially for public enterprises and banks. In a similar spirit as the predictive model of "shadow" rating used to estimate the credit risk of unrated countries (Basu et al., 2013;Ratha, De and Mohapatra, 2011), the findings of this study could provide useful insights to evaluate the creditworthiness of unrated firms. Finally, this study calls for further careful assessment of the sub-sovereign rating decisions based on bond-level and corporate financial factors after incorporating the externality which stems from country and global risk factors.    Notes: The dependent variable is the distance of the sub-sovereign debt rating below the sovereign rating. All regressions include additional controls for current account balance and U.S. 10-year government bond yield. Heteroscedasticity-consistent robust standard errors are clustered at the country level. * Significant at 10%; ** significant at 5%; *** significant at 1%  (44) 2,167 (40) Notes: The two stages of the double-hurdle model presented in columns (1)-(3) and in columns (4)-(6) are estimated simultaneously. The dependent variable in columns (1)-(3) is an indicator that the sub-sovereign debt rating is higher than the sovereign rating. The dependent variable in columns (4)-(6) is the distance between the sub-sovereign debt rating and sovereign rating, conditional on being above the sovereign rating. All regressions include additional controls for current account balance and U.S. 10-year government bond yield. Heteroscedasticity-consistent robust standard errors are clustered at the country level. * Significant at 10%; ** significant at 5%; *** significant at 1%. Notes: The two stages of the double-hurdle model presented in columns (1)-(3) and in columns (4)-(6) are estimated simultaneously. The dependent variable in columns (1)-(3) is an indicator that the sub-sovereign debt rating is higher than the sovereign rating. The dependent variable in columns (4)-(6) is the distance between the sub-sovereign debt rating and sovereign rating, conditional on being above the sovereign rating. All regressions include the full set of control variables in the baseline model in table 5. Heteroscedasticity-consistent robust standard errors are clustered at the country level. * Significant at 10%; ** significant at 5%; *** significant at 1%. Notes: The regression is estimated using a two-step Heckman procedure. All regressions include additional controls for current account balance and U.S. 10-year government bond yield. * Significant at 10%; ** significant at 5%; *** significant at 1% Notes: The regression is estimated using a two-step Heckman procedure. All regressions include additional controls for current account balance and U.S. 10-year government bond yield. * Significant at 10%; ** significant at 5%; *** significant at 1%

Figure 4: Regional variation in distance below the sovereign rating obtained from baseline Tobit regression
Note: Coefficients obtained from baseline Tobit regression in table 4 with inclusion of regional dummies. East Asia & Pacific region is taken as the base region. Statistical significance of regional coefficients (* significant at 10 percent; ** significant at 5 percent; *** significant at 1 percent) indicated next to the bars -1.5 -1 -0.5 0 0.5 1 1.  (3) Notes: The dependent variable is the distance of the sub-sovereign debt rating below the sovereign rating. All regressions include additional controls for current account balance and U.S. 10-year government bond yield. Heteroscedasticity-consistent robust standard errors are clustered at the country level. * Significant at 10%; ** significant at 5%; *** significant at 1% Appendix  (37) Notes: The two stages of the double-hurdle model presented in columns (1)-(3) and in columns (4)-(6) are estimated simultaneously. The dependent variable in columns (1)-(3) is an indicator that the sub-sovereign debt rating is higher than the sovereign rating. The dependent variable in columns (4)-(6) is the distance between the sub-sovereign debt rating and sovereign rating, conditional on being above the sovereign rating. All regressions include additional controls for current account balance and U.S. 10-year government bond yield. Heteroscedasticity-consistent robust standard errors are clustered at the country level. * Significant at 10%; ** significant at 5%; *** significant at 1%.  (3) Notes: The two stages of the double-hurdle model presented in columns (1)-(3) and in columns (4)-(6) are estimated simultaneously. The dependent variable in columns (1)-(3) is an indicator that the sub-sovereign debt rating is higher than the sovereign rating. The dependent variable in columns (4)-(6) is the distance between the sub-sovereign debt rating and sovereign rating, conditional on being above the sovereign rating. All regressions include additional controls for current account balance and U.S. 10-year government bond yield. Heteroscedasticity-consistent robust standard errors are clustered at the country level. * Significant at 10%; ** significant at 5%; *** significant at 1%.