Sovereign credit rating determinants under financial crises

Global Finance Journal xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Global Finance Journal journal homepage: www.elsevier.com/locate/gfj

Sovereign credit rating determinants under financial crises ⁎

João C.A. Teixeiraa, , Francisco J.F. Silvaa, Manuel B.S. Ferreirab, José A.C. Vieiraa a University of the Azores, School of Business and Economics, Centre of Applied Economics Studies of the Atlantic, Rua da Mãe de Deus, s/n, 9501-801 Ponta Delgada, Portugal b University of the Azores, School of Business and Economics, Rua da Mãe de Deus, s/n, 9501-801 Ponta Delgada, Portugal

AR TI CLE I NF O

AB S T R A CT

JEL classification: G01 G24

This paper empirically examines the determinants of sovereign credit ratings using panel data on a sample of 86 countries for 1993–2013. It further investigates whether the countries' average credit rating differs by region and for crisis and noncrisis periods, and how the bursting of the dot-com bubble, the Asian crisis, and the 2008 international financial crisis affected the average rating of each region. The estimation results reveal that macroeconomic, external, government, and qualitative factors importantly affect sovereign credit ratings, and that average ratings differ across all geographical regions except for North America and the Eurozone. While the recent crisis reduced the average rating across all regions, the dot-com bubble burst had no effect, the Asian crisis affected only the average rating of Asian countries, and the downgrade resulting from the 2008 crisis was larger in the Eurozone.

Keywords: Sovereign credit ratings Sovereign debt Financial crises Ordered probit model

1. Introduction The three main rating agencies, Standard and Poor's, Moody's, and Fitch, use a combination of economic, social, and political factors to assess a country's capacity and willingness to honor its current and future debt obligations in full and on time (Chen, Chen, Chang, & Yang, 2016). Sovereign credit ratings play an important role in determining countries' access to international financial markets and the terms of that access, since these ratings are perceived as useful in predicting sovereign distress. Moreover, Teixeira, Silva, Fernandes, and Alves (2014) show that sovereign credit ratings also affect banks' profitability and capital ratios, and Cavallo, Powell, and Rigobon (2013) provide evidence that sovereign credit ratings do matter for investors. As the dot-com bubble burst, the Asian crisis, and the international financial crisis of 2007–2008 led to several sovereign rating downgrades, it is important to examine in detail the quantitative and qualitative factors that determine these ratings.1 Are there substantial regional differences in the average sovereign credit ratings produced by the three main rating agencies during the last two decades? Did the dot-com bubble burst, the Asian crisis, and the recent international financial crisis lead to a downgrade on the sovereign credit rating across all regions worldwide, or have some regions seen their rating unchanged during these crises? This paper addresses these questions using panel data on a sample of 86 countries, spanning 7 regions worldwide, for the period 1993–2013. We investigate the determinants of sovereign credit rating as follows. First, we estimate an ordered probit model where the dependent variable is the average of the sovereign credit ratings assigned by Moody's, Standard and Poor's, and Fitch (converted to a

⁎

Corresponding author. E-mail addresses: [email protected] (J.C.A. Teixeira), [email protected] (F.J.F. Silva), [email protected] (J.A.C. Vieira). 1 For instance, in parallel to a downgrade in European sovereign credit ratings, the European sovereign debt crisis also led to more segmentation of the European economy (see Cipollini, Coakley, & Hyunchul, 2015). In addition, Aktug, Nayar, and Vasconcellos (2013) provide evidence of a strong linkage between sovereign credit ratings and the banking sector. https://doi.org/10.1016/j.gfj.2018.01.003 Received 10 June 2017; Received in revised form 16 January 2018; Accepted 17 January 2018 1044-0283/ © 2018 Elsevier Inc. All rights reserved.

Please cite this article as: Teixeira, J.C.A., Global Finance Journal (2018), https://doi.org/10.1016/j.gfj.2018.01.003

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

numerical scale), and the explanatory variables are a set of macroeconomic, external, government, and country-specific qualitative factors. Among macroeconomic factors we consider GDP per capita, GDP growth, unemployment, inflation, and total investment; among external factors, external debt, deficit in the current account, reserves, terms of trade, and liquidity risk. Among government factors we take into account the level of government debt and the fiscal balance. Lastly, among qualitative variables we incorporate the country's history of default, level of corruption, and political stability. We then expand this model by including dummy variables for years, in order to examine whether the average rating across the sample differs across the periods of the Asian crisis (1997 and 1998), the dot-com bubble burst or dot-com crisis (2000 to 2002), the recent international financial crisis (2008 to 2013), and the noncrisis periods. In parallel, we add dummy variables for seven regions: North America, the Eurozone, the rest of Europe, Africa, Asia, Latin America, and Oceania. We also incorporate a set of variables representing interactions between the year dummies and the region dummies in order to evaluate whether the effect of each crisis on ratings differs by region. This paper contributes to the empirical literature on the determinants of sovereign credit rating as follows. First, it examines the effect of country-specific qualitative variables (history of default, level of corruption, and political stability). Although some of these variables have been previously considered, for instance by Depken, LaFountain, and Butters (2011) and Erdem and Varli (2014), we integrate them into our analysis along with macroeconomic, external, and government factors. Second, the panel incorporates countries from most regions worldwide and covers two decades, allowing us to examine the effects of three crises on credit ratings, and whether these effects vary across regions. The remainder of this paper is organized as follows. Section 2 presents our model of the determinants of sovereign credit rating and discusses the expected effect of each explanatory variable on the rating, in line with the empirical literature. Section 3 describes the data and the descriptive statistics of the variables. Section 4 presents the initial estimation results of the main model and subsequently examines the results of other models that assess whether crises and regions affect the ratings. Section 5 concludes.

2. The determinants of sovereign credit ratings In this section, we describe the econometric specification of our model and discuss the expected relation between each explanatory variable and the country's rating. Cantor and Packer (1996) were among the first scholars to assess the importance of eight macroeconomic variables explaining rating assignments by Moody's and Standard and Poor's, using a cross-section analysis of 49 countries. Our paper provides a more comprehensive analysis of these determinants and relates to the literature that uses panel data, in particular the studies by Mellios and Paget-Blanc (2006), Bissoondoyal-Bheenick (2005), Bissoondoyal-Bheenick, Brooks, and Yip (2006), Afonso, Gomes, and Rother (2011), Erdem and Varli (2014), and Maltritz and Molchanov (2014).2 It also contributes to the empirical literature that discusses geographical differences in sovereign credit ratings, as do Depken et al. (2011) and Afonso et al. (2011), and the literature that examines the effect of financial crises on sovereign credit ratings, as do Ferri, Liu, and Stiglitz (1999) and Amstad and Packer (2015). Maltritz and Molchanov (2014) provide a good summary of a related literature that studies the determinants of countries' credit risk, as measured by sovereign yield spreads. A recent article by Dilly and Mählmann (2016) highlights, however, that the determinants of sovereign credit ratings are not exactly the same as the determinants of sovereign yield spreads. They show that initial ratings disagree with bond spread levels during booms: rating agencies hold a systematically more optimistic view. Moreover, Becker and Milbourn (2011) provide empirical evidence that in recent years increased competition from Fitch has coincided with lower quality of ratings from the incumbents Moody's and Standard and Poor's: rating levels went up, the correlation between ratings and market yields fell, and the ability of ratings to predict default deteriorated. There are also a few recent studies that analyze the effect of sovereign rating revisions on economic growth and other macroeconomic variables, as do Chen et al. (2016), and others that examine the linkage between financial openness and sovereign credit ratings, as do Andreasen and Valenzuela (2016). The empirical literature has identified a set of country-specific factors that play an important role in determining the sovereign credit rating (see, among others, Afonso et al., 2011; Bissoondoyal-Bheenick, 2005; Cantor & Packer, 1996; Erdem & Varli, 2014; Maltritz & Molchanov, 2014; Mellios & Paget-Blanc, 2006). However, given the differences in time periods, datasets, and countries, the empirical evidence of these studies does vary. Nevertheless, it is possible to identify a set of common factors. The literature seems to agree on the relevance of four vectors of specific factors: macroeconomic, external, governmental, and qualitative. As macroeconomic factors, we include GDP per capita, GDP real growth rate, unemployment, inflation, and total investment. As external factors we consider external debt, ratio of reserves to imports, deficit in the current account, terms of trade, and liquidity risk. Among government factors we take into account public debt and fiscal balance. Among qualitative factors we incorporate past default, the corruption index, and the political stability index. Following Bissoondoyal-Bheenick (2005) and Mellios and Paget-Blanc (2006), we measure the dependent variable rating as the arithmetic average, for each end of the year, of the sovereign credit ratings assigned by the rating agencies Moody's, Fitch, and Standard and Poor's, converted onto a numerical scale from 1 (worst) to 21 (best), as Table 1 shows. Other studies use an analogous rating transformation but with different numerical scales; Afonso et al. (2011) use a scale from 1 to 17, while Erdem and Varli (2014) use a scale from 1 to 46. 2 Other related studies include Feder and Uy (1985), Haque, Kumar, Mark, and Mathieson (1996), Larrain, Helmut, and Maltzan (1997), Haque, Mark, and Mathieson (1998), Jüttner and McCarthy (2000), Monfort and Mulder (2000), Mulder and Perrelli (2001), Afonso (2003), Pellegrini and Gerlagh (2004), and Afonso, Gomes, and Rother (2009).

2

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

Table 1 Numerical conversion of the sovereign credit rating. 21

20

19

18

17

16

15

14

13

12

11

Moody's Fitch S&P

Aaa AAA AAA

Aa1 AA+ AA+

Aa2 AA AA

Aa3 AA− AA−

A1 A+ A+

A2 A A

A3 A− A−

Baa1 BBB+ BBB+

Baa2 BBB BBB

Baa3 BBB− BBB−

Ba1 BB+ BB+

Moody's Fitch S&P

10 Ba2 BB BB

9 Ba3 BB− BB−

8 B1 B+ B+

7 B2 B B

6 B3 B− B−

5 Caa1 CCC CCC+

4 Caa2 CC CCC

3 Caa3 C CCC−

2 Ca RD CC

1 C D C/D

The rating provided by the three main rating agencies, Moody's, Fitch, and Standard and Poor's (S&P), is converted into a numerical scale from 1 to 21, where 1 indicates the worst rating and 21 the best rating.

Given the nature of the dependent variable, a discrete variable that reflects an order in terms of probability of default, with a unit change in between, we estimate the determinants of sovereign credit rating by using an ordered probit model. In line with Afonso et al. (2011), Depken et al. (2011), and Erdem and Varli (2014), we assume that each rating agency makes a continuous evaluation of a country's credit-worthiness, embodied in an unobserved latent variable Ratingit∗. Therefore, the ordered probit model is defined as

Ratingit∗ = βXit + αi + εit ,

(1)

where Ratingit is an observable variable with certain cut-off points μi, and Xit is the set of four vectors of explanatory variables of each country i in year t. αi is the country's time-invariant specific term, and εit is the normal error term. Hence, if we assume a particular probability distribution for Ratingit∗, we will have an ordered probit model and can calculate the following μi ´s :

⎧1 ⎪ 2 if ⎪ Ratingit = 3 if ⎨ ⎪⋮ ⎪ 21 ⎩

if

Ratingit∗ ≤ μ 1

μ 1 < Ratingit∗ ≤ μ 2 μ 2 < Ratingit∗ ≤ μ 3 if

20 < Ratingit∗

(2)

The parameters of Eqs. (1) and (2) are estimated using log-likelihood maximization. Following Afonso et al. (2011), we use a random effects ordered probit estimation, which considers both αi and εit to be normally distributed. We now explain the specification of each explanatory variable and the expected relation between it and the sovereign credit rating, in line with the predictions of the empirical literature. Table 2 summarizes the variable definitions, and Table 3 shows the Table 2 Variables definition. Variables

Definition

Rating

Arithmetic average, for each year, of the sovereign credit rating assigned by the Moody's, Standard and Poor's, and Fitch, after converting the rating onto a numerical scale from 1 to 21, where 1 corresponds to the worst rating and 21 to the best rating.

Macroeconomic variables GDP per capita GDP growth Unemployment Inflation Investment External variables External debt Deficit in the current account Reserves Terms of trade Liquidity risk Government variables Government debt Fiscal balance Qualitative variables Default Corruption Political stability

Gross Domestic Product per capita at constant prices in U.S. dollars. Real growth rate of the GDP at constant prices. Unemployment rate. Inflation rate. Total investment as a percentage of GDP. Total external debt as a percentage of GDP. Deficit in the current account as a percentage of GDP. Ratio between reserves (including gold) and imports. Ratio between exports and imports. Ratio between short-term external debt and reserves. Gross government debt as a percentage of GDP. Fiscal balance as a percentage of GDP. Dummy variable that takes the value of one if the country has defaulted in the past and zero otherwise. Index from Transparency International, where zero corresponds to the highest level of corruption and 100 to the lowest level. Index from the World Bank that measures the political stability and absence of violence and terrorism in the country, where zero corresponds to the lowest level of political stability and 100 to the highest level.

3

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

Table 3 Predicted effect of explanatory variables on average sovereign credit ratings. Explanatory variables

Predicted effect on sovereign credit rating

Macroeconomic variables GDP per capita GDP growth Unemployment Inflation Investment External variables External debt Deficit in the current account Reserves Terms of trade Liquidity risk Government variables Government debt Fiscal balance Qualitative variables Default Corruption index Political stability

Positive Positive Negative Uncertain Positive Negative Uncertain Positive Positive Negative Negative Positive Negative Positive Positive

predicted effect of each variable on the sovereign credit rating. Among the macroeconomic variables, most empirical studies have found that a higher GDP per capita improves the country's rating. In our study, this variable is measured in constant prices in U.S. dollars. Cantor and Packer (1996), Bissoondoyal-Bheenick (2005), Mellios and Paget-Blanc (2006), Depken et al. (2011), and Erdem and Varli (2014) point out that countries with a higher GDP per capita have a higher fiscal base, which increases their ability to pay their debt and consequently reduces their intrinsic probability of default. Additionally, Afonso et al. (2011) suggest that we should expect economies with a higher GDP per capita to have more stable institutions that rely less on excess debt and therefore show less vulnerability to external shocks. Greater growth in real GDP also tends to improve a country's credit rating. Higher GDP growth reduces the relative burden of debt and can mitigate problems related to insolvency by improving the rating agencies' perception of the country's economy (Afonso et al., 2011; Cantor & Packer, 1996; Depken et al., 2011; Mellios & Paget-Blanc, 2006). The empirical evidence of Bissoondoyal-Bheenick (2005), Afonso et al. (2011), and Erdem and Varli (2014) clearly suggests that unemployment worsens sovereign credit ratings. According to Afonso et al. (2011), countries with a low unemployment rate tend to bear less cost of social benefits associated with unemployment and, at the same time, tend to broaden the fiscal base that derives from employment, increasing government income. The effect of inflation on a country's rating tends to be uncertain. On one hand, Bissoondoyal-Bheenick (2005), Cantor and Packer (1996), Mellios and Paget-Blanc (2006), Depken et al. (2011), and Erdem and Varli (2014) consider that its effect is negative, as high inflation can signal the worst economic performance and the worst rating. On the other hand, Afonso et al. (2011) point out that inflation can improve the country's rating if it can reduce the public debt in the national currency, leaving more resources to serve the external debt, but also can have a negative effect since it can signal macroeconomic problems, especially if it is caused by the monetary financing of deficits. Investment is measured as total investment as a percentage of GDP. The empirical evidence suggests that higher investment improves a country's rating. According to Mellios and Paget-Blanc (2006), investment measures the country's ability to grow in the future and therefore should be a decreasing function of default: the higher the investment, the lower is the probability of the country's default. Among external variables, the first one that we consider is external debt, measured as the country's total debt as a percentage of GDP. Its predicted effect on the country's rating is negative, as high external debt clearly implies a higher risk of default and a higher dependence on international creditors (see the empirical evidence of Afonso et al., 2011; Cantor & Packer, 1996; Erdem & Varli, 2014; and Mellios & Paget-Blanc, 2006). As regards deficit in the current account, measured as a percentage of GDP, the empirical evidence suggests an uncertain effect on sovereign credit ratings. On one hand, Bissoondoyal-Bheenick (2005), Mellios and Paget-Blanc (2006), Depken et al. (2011), and Erdem and Varli (2014) point out that a higher deficit in the current account leads to a higher reliance on external creditors, and if this deficit persists it can affect the country's long-term sustainability. In addition, Afonso et al. (2011) proposed that a higher deficit can lead to higher consumption levels that can lessen the country's long-term sustainability. On the other hand, they also argue that a higher current account deficit can reflect rapid accumulation of investment, which may increase growth and sustainability over the medium term. Reserves are measured as the ratio of reserves, including gold, to total imports. Mellios and Paget-Blanc (2006) document a positive effect of this variable on the country's rating, as higher reserves mean more funds available to cope with debt payments. Erdem and Varli (2014) find a similar relation, although they use total reserves instead of reserves divided by imports. Terms of trade are measured as the ratio of exports to imports. According to Hilscher and Nosbusch (2010), we should expect a 4

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

high ratio to improve the country's rating, since a higher value of exports compared to imports means that the country can generate more resources to pay its external debt. In addition, Erdem and Varli (2014) find that high total exports improve sovereign credit ratings. Liquidity risk is measured as the ratio of short-term external debt to reserves. As Mellios and Paget-Blanc (2006) point out, higher values of this ratio suggest more liquidity problems, so we should expect it to lower sovereign credit ratings. Among government variables, our model incorporates (gross) government debt and the fiscal balance, both measured as a percentage of GDP. Mellios and Paget-Blanc (2006) and Afonso et al. (2011) seem to agree that higher levels of government debt are associated with potential default and liquidity problems, so we should expect this variable to lower ratings. Higher fiscal deficits reduce the country's savings and induce negative macroeconomic shocks, which can ultimately cause political problems (see Afonso et al., 2011; Cantor & Packer, 1996). Among our qualitative variables, default is defined as a dummy variable that takes the value of 1 if the country has defaulted in the past and zero otherwise. Cantor and Packer (1996), Mellios and Paget-Blanc (2006), Depken et al. (2011), and Afonso et al. (2011) report that a country's history of default can affect its reputation; in particular they show that the debt of countries that have previously defaulted tends to be more risky, and this should translate into a lower sovereign credit rating. The level of corruption can also importantly affect sovereign credit ratings (Depken et al., 2011, Erdem & Varli, 2014, Mellios & Paget-Blanc, 2006). We measure corruption with an index from “Transparency International: The global coalition against corruption,” where zero corresponds to the highest level of corruption and 100 to the lowest level. We should expect countries with more corruption to show less capability to pay their debt, which means that we should expect a positive relation between this variable and the sovereign credit rating. We measure political stability also as an index, in this case obtained from the World Bank. This index measures political stability and the absence of violence and terrorism; zero corresponds to the lowest level of political stability and 100 to the highest level. As Mellios and Paget-Blanc (2006) and Erdem and Varli (2014) have pointed out, greater political stability should make the country able to pay its debt and, as a consequence, should raise its sovereign credit rating. 3. Data and descriptive statistics Data on sovereign credit ratings were collected from the web sites of the rating agencies Moody's, Fitch, and Standard and Poor's. For each country, the macroeconomic, external, and government variables were obtained from the World Economic Outlook of the International Monetary Fund (IMF) and the World Development Indicators of the World Bank. The history of default, the corruption index, and the political stability index were collected from the web site of the rating agency Standard and Poor's, the web site of the International Coalition against Corruption, and the data set of the World Bank, respectively. Table 4 lists the 86 countries in the data set by region and also gives the number of observations for each country; Table 5 shows the descriptive statistics that characterize the panel of countries. The panel data go from 1993 to 2013 and include 1606 observations. The countries are grouped into 7 regions: Asia, with 21 countries and 401 observations, the Eurozone, with 17 countries and 346 observations, the rest of Europe, with 17 countries and 321 observations, Latin America, with 13 countries and 250 observations, Africa, with 10 countries and 135 observations, Oceania, with 4 countries and 79 observations, and North America, with 4 countries and 74 observations. The mean sovereign credit rating is 14.22, a value that fits in the investment grade category, as it corresponds to a rating of BBB+ on Fitch's and Standard and Poor's scales and to a rating of Baa1 on Moody's scale. These values are in line with Mellios and PagetBlanc's (2006) mean rating value of 14 on the same scale, although their period covers only the years 1998 to 2002. The standard deviation of the rating variable is 5.09, which is relatively high, a result that is better illustrated in Fig. 1, which shows the dispersion of this variable. The rating ranges from a minimum of 1.5 to a maximum of 21, even though there is a higher concentration of observations with ratings of 21. Interestingly, seven countries have seen their average rating remain unchanged at the highest level during the sample period: Germany, Luxembourg, the Netherlands, Switzerland, the United Kingdom, the United States, and Australia. Table 5 also shows a substantial heterogeneity in wealth per capita and most other variables. The mean GDP per capita is 15,606 USD, whereas the median is only 8600 USD. Moreover, the richest country, Luxembourg, had a GDP per capita of 114,186 in 2011, while the poorest country, Malawi, had a GDP per capita of 114 USD in 1994. Table 6 lists countries that have defaulted in the past, with the year of default and average rating in that year. All countries that have defaulted had average ratings equal to or below 8; Argentina has the most events of default in the sample, and Cyprus and Greece are the most recent cases of default. Before we discuss the regression results, we further analyze some descriptive statistics of the dependent variable, in particular those of the countries' ratings by region and time-period, as presented in Table 7. The region with the highest mean rating is North America, with 19.04, followed by the Eurozone, with 18.31, and the rest of Europe, with 15.12. At the bottom of the ratings we find Latin America and Africa, with mean values of 9.91 and 9.09 respectively. In Latin America and Africa none of the countries obtained an average rating of 21, the maximum on our scale, during the sample period, while in North America the United States and Canada tend to maintain the maximum rating for several years. The lowest mean rating occurred during the recent international financial crisis, with 13.89, followed by the Asian crisis, with 13.90, and the dot-com crisis, with 14.18. In the next section, we use regression analysis to examine in more detail whether there are statistical differences in the mean rating by region and time-period. The effect of the 2007 crisis is particularly important in the Eurozone, where most countries saw their ratings downgraded from 2007 to 2013. This effect is evident in Fig. 2, which shows the ratings of the Eurozone countries in 2007 and 2013. In this region, 11 5

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

Table 4 Number of countries and country-year observations for each region. Region

Number of countries

Number of observations

Africa South Africa Cape Verde Cameroon Egypt Gambia Ghana Lesotho Mozambique Malawi Tunisia North America Aruba Bermuda Canada USA Latin America Argentina Brazil Chile Colombia Costa Rica El Salvador Ecuador Mexico Panama Peru Dominican Republic Uruguay Venezuela Asia Azerbaijan Bahrain China South Korea Philippines Hong Kong India Iran Israel Japan Kazakhstan Kuwait Lebanon Malaysia Russia Singapore Thailand Taiwan Turkmenistan Turkey Vietnam Oceania Australia Indonesia New Guinea New Zealand Eurozone Austria Belgium Cyprus Estonia Finland France Germany Greece Ireland Italy

10

135 20 11 11 17 12 11 12 11 11 19 74 12 20 21 21 250 21 21 21 21 17 18 17 21 17 17 17 21 21 401 14 18 21 21 21 21 21 15 21 21 18 19 17 21 18 21 21 21 17 21 13 79 21 21 21 16 346 21 21 20 17 21 21 21 21 21 21 (continued on next page)

4

13

21

4

17

6

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

Table 4 (continued) Region

Number of countries

Luxembourg Malta Netherlands Portugal Slovakia Slovenia Spain Rest of Europe Bulgaria Croatia Denmark Hungary Iceland Latvia Lithuania Norway Poland United Kingdom Czech Republic Republic of Moldova Romania San Marino Sweden Switzerland Ukraine Total of countries Total of observations

Number of observations 21 20 21 21 19 18 21 321 18 17 21 21 21 17 18 21 19 16 21 17 18 13 21 21 21

17

86 1606

The sample consists of 1606 observations of average sovereign credit ratings across 86 countries, grouped into 7 regions, from 1993 to 2013. The sovereign credit ratings are obtained from the web pages of the three main rating agencies, Standard and Poor's, Moody's, and Fitch. Table 5 Descriptive statistics. Distribution

Rating Macroeconomic variables GDP per capita (USD) GDP growth (%) Unemployment (%) Inflation (%) Investment (%) External variables External debt (%) Deficit in the current account (%) Reserves Terms of trade Liquidity risk Government variables Government debt (%) Fiscal balance (%) Qualitative variables Default Corruption index Political stability index

N

Mean

St. dev.

Min.

Max.

25th

50th

75th

1606

14.22

5.09

1.50

21.00

10.33

14.17

19.00

1806 1806 1806 1806 1806

15,606 3.55 8.65 10.47 23.28

17,318 4.47 4.77 47.57 6.88

114 −30.90 0.65 −8.53 0.10

114,186 34.50 21.15 992.39 76.86

2480 1.69 5.30 2.03 19.23

8600 3.69 7.36 3.53 22.12

24,584 5.89 11.26 7.93 26.23

1806 1806 1806 1620 1806

104.02 −1.04 0.78 0.99 2.07

277.25 8.15 3.55 0.28 7.68

0.14 −40.11 0.00 0.13 0.00

2910.98 44.62 61.73 2.89 133.92

13.15 −4.58 0.18 0.86 0.14

34.52 −1.74 0.32 0.98 0.35

67.73 2.63 0.53 1.11 0.88

1806 1806

52.64 −12.34

33.06 109.57

2.42 −2072.44

243.54 0.17

30.80 −0.03

45.45 −0.00

69.41 −0.00

1806 1806 1806

0.01 51.66 56.86

0.100 23.06 26.43

0.00 8.00 0.69

1.00 100.00 100.00

0.00 32.00 37.02

0.00 47.00 58.17

0.00 71.60 78.68

The sample consists of 1606 observations of average sovereign credit ratings across 86 countries, grouped in 7 regions, from 1993 to 2013. The sovereign credit ratings were obtained from the web pages of the three main rating agencies, Standard and Poor's, Moody's, and Fitch. The macroeconomic, external, and government variables were obtained from the World Economic Outlook of the International Monetary Fund and the World Development Indicators of the World Bank. The history of default, the corruption index, and the political stability index were collected from the web site of the rating agency Standard and Poor's, the web site of the International Coalition against Corruption, and the data set of the World Bank, respectively.

7

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

0.20 0.18 0.16 0.14

Fraction

0.12 0.10 0.08 0.06 0.04 0.02 0.00

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21 Rating

Fig. 1. Distribution of sovereign credit rating. The figure shows the distribution of the average sovereign credit rating for the sample of 1606 observations, across 86 countries, from 1993 to 2013.

Table 6 Countries with history of default. Country

Region

Year of default

Sovereign credit rating

Argentina Argentina Argentina Argentina Argentina Cyprus Ecuador Ecuador Greece Indonesia Indonesia Indonesia Dominican Republic Russia Russia Uruguay Venezuela

Latin America Latin America Latin America Latin America Latin America Oceania Latin America Latin America Eurozone Oceania Oceania Oceania Latin America Asia Asia Latin America Latin America

2001 2002 2003 2004 2005 2013 2008 2009 2012 1999 2000 2002 2005 1999 2000 2003 2005

1.67 1.67 2.67 2.67 4.67 6.33 2.33 4.33 6.33 5.67 6.00 6.00 6.33 4.67 6.67 6.00 8.00

List of countries that have defaulted in the past, with year of default and average rating in the year of default. Table 7 Descriptive statistics of the sovereign credit rating by region and time-period. Distribution

Africa North America Latin America Asia Oceania Eurozone Rest of Europe Asian crisis (1997 and 1998) Dot-com crisis (2000 to 2002) International crisis (2018 to 2013) Noncrisis years

N

Mean

St. dev.

Min.

Max.

25th

50th

75th

135 74 250 401 79 346 321 146 235 516 709

9.09 19.04 9.91 13.40 14.66 18.31 15.12 13.90 14.18 13.89 14.57

2.81 2.57 3.00 4.34 5.91 3.10 4.98 4.8 5.24 5.07 5.09

5.00 13.00 1.67 4.67 5.33 6.00 1.50 4.7 1.5 2.30 2.7

14.33 21.00 17.67 21.00 21.00 21.00 21.00 21.00 21.00 21.00 21.00

7.00 19.33 8.00 9.67 8.00 16.33 11.83 10.46 9.5 9.75 10.33

8.00 19.67 10.33 13.67 19.00 19.00 15.00 13.00 14.00 13.33 15.00

12.00 20.67 11.67 16.83 20.33 21.00 20.67 19.00 19.33 18.67 19.33

The table shows the descriptive statistics of the average sovereign credit rating assigned by Standard and Poor's, Moody's, and Fitch, for each region and crisis period, as well as for the noncrisis years.

8

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

Fig. 2. Sovereign credit rating of Eurozone countries in 2007 and 2013. The figure depicts the average sovereign credit rating of the Eurozone countries for the years 2007 and 2013.

out of the 17 countries had lower ratings in 2013, and the downgrade was more pronounced in Cyprus, Greece, Portugal, Spain, and Ireland. These were the countries that during the crisis benefited from rescue packages from the IMF, the European Central Bank, and the European Commission, although in Spain this package was directly oriented to the banking system. Only Germany, Finland, Holland, and Luxembourg kept their mean ratings from 2007 to 2013. 4. Results In Section 4.1, we discuss the estimation of model 1, which considers the macroeconomic, external, government, and qualitative explanatory factors discussed in Section 2. In Section 4.2, we discuss the estimation of models 2 and 3, which evaluate the effects of region and crises on the results. Model 2 expands the previous model by incorporating three dummy variables associated with the crises. Each variable takes the value of 1 during the years of the corresponding crisis and zero otherwise. We define the Asian crisis as occurring during 1997 and 1998, the dot-com crisis or dot-com bubble burst during 2000–2002, and the recent international financial crisis during 2008–2013. This model allows us to discuss the direct effect of each crisis on the sovereign credit rating. It also incorporates a set of dummy variables for regions, which enable us to examine whether there are statistical differences among the rating levels of those regions. Model 3 further expands the analysis conducted in model 2 by examining whether the effect of each crisis on the rating differs by region; it incorporates interactions between these two variables. We estimate all three models using the ordered probit estimator. 4.1. Sovereign credit ratings and the effect of macroeconomic, external, government, and qualitative factors In this section, we examine the regression results of model 1, as depicted in Table 8.3 First, all coefficients associated with the macroeconomic variables are statistically significant at the 1% level except the GDP growth coefficient, which is significant at a 5% level. As we expected, GDP per capita, GDP growth, and investment improve the sovereign credit rating, whereas unemployment worsens it. Countries with higher inflation tend to have worse ratings, a result that is in line with the empirical evidence of Bissoondoyal-Bheenick (2005), Cantor and Packer (1996), Mellios and Paget-Blanc (2006), Depken et al. (2011), and Erdem and Varli (2014). Second, among the external variables, three out of the five coefficients are statistically significant. The estimated coefficient associated with external debt and liquidity risk is statistically significant at a 1% level, and the estimated coefficient associated with deficit in the current account is statistically significant at a 5% level. While the negative effect of external debt on rating is in line with the empirical evidence, the positive sign of the estimated coefficient of liquidity risk is at odds with the literature, as Mellios and Paget-Blanc (2006) predict that higher values of this ratio suggest more liquidity problems and, therefore, lower ratings. In our model, the variables reserves and terms of trade have no effect on sovereign credit ratings. Third, among the government variables, the estimation results suggest no effect of the fiscal balance variable but a negative effect of the level of government debt on ratings. This last result is in accordance with the empirical evidence of Mellios and Paget-Blanc 3

The model presented incorporates only the estimated coefficients that turned out to be statistically significant.

9

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

Table 8 Sovereign credit rating model. Model 1

Macro. variables

External variables

Govern. variable Qualitative variables

GDP per capita GDP growth Unemployment Inflation Investment External debt Deficit current account Liquidity risk Government debt Default Corruption index Political stability Cut1 Cut2 Cut3 Cut4 Cut5 Cut6 Cut7 Cut8 Cut9 Cut10 Cut11 Cut12 Cut13 Cut14 Cut15 Cut16 Cut17 Cut18 Cut 19 Average distance between cuts Log-likelihood P-value of likelihood Ratio test (LRT) Number of countries Number of observations

Coeff.

Std. E.

0.00006*** 0.0198** −0.131*** −0.011*** 0.042*** −0.001*** −0.009** 0.057*** −0.019*** −2.132*** 0.034*** 0.022*** −7.055*** −6.711*** −6.258*** −4.393*** −3.130*** −2.374*** −1.425** −0.782 −0.193 0.710 1.490** 2.172*** 2.660*** 3.378*** 4.203*** 4.709*** 5.546*** 7.060*** 8.159*** 0.845 −2381.466 0.000 86 1606

0.000005 0.009 0.015 0.003 0.009 0.0004 0.007 0.018 0.002 0.315 0.006 0.004 0.736 0.720 0.699 0.641 0.629 0.625 0.623 0.622 0.621 0.619 0.618 0.616 0.616 0.617 0.620 0.622 0.626 0.630 0.631

The sample consists of 86 countries, grouped in 7 regions, from 1993 to 2013. See Table 2 for the definition of variables. Model 1 is given by: Ratingit∗ = βXit + αi + εit, where the dependent variable is the arithmetic average, for each year, of the sovereign credit rating assigned by the Moody's, Standard and Poor's, and Fitch, after converting the rating onto a numerical scale from 1 to 21, where 1 corresponds to the worst rating and 21 to the best rating. Xit is composed of four vectors of macroeconomic, external, government, and qualitative variables. The first column shows the estimated coefficient and the second column the standard error. The reported coefficients and their standard errors are obtained using the random effect ordered probit estimator. ***, **, and * denote statistical significance at the 1%, the 5%, and the 10% level, respectively.

(2006) and Afonso et al. (2011), and supports the idea that higher levels of government debt are associated with potential default and liquidity problems and, therefore, worse ratings. Fourth, all three qualitative variables importantly affect countries' ratings, as the estimated coefficients associated with these variables are statistically significant at a 1% level. In accord with the empirical literature, we find that countries with a history of default tend to have lower ratings, as do countries with more corruption and less political stability. An analysis of the estimated thresholds (cuts) reported in the second part of Table 8 reveals that there are two rating categories that have wider ranges. While the average distance between thresholds is 0.845, the distance is 1.865 in category CCC+ and 1.514 in category AA (levels 5 and 19, respectively, on our scale of 1 to 21).4 Therefore, the jump to the next category is more difficult at a rather low level, CCC+, and at a high level, AA. In the middle levels, the transition is relatively smooth. 4.2. Sovereign credit ratings and the effect of region and crises Models 2 and 3 are presented in Table 9. Model 2 incorporates the effect of crises, given by the dummy variables created for each crisis, and of region, given by the dummy variables associated with each region. Among regions we omit the Eurozone from the model; the estimated coefficients report comparisons with this region. In model 2, the estimated coefficients of the dummy variables associated with the Asian crisis and the recent international 4

The distance 1.865 is cut4 minus cut3, and the distance 1.514 is cut18 minus cut 17.

10

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

Table 9 Sovereign credit rating models with crisis and regional effects. Model 2

Macro variables

External variables

Gov. var. Qualitative variables

Region dummy variables

Asian crisis effect

Dot-com crisis effect

International crisis effect

GDP per capita GDP growth Unemployment Inflation Investment External debt Deficit in current account Liquidity risk Government debt Default Corruption index Political stability Africa North America Latin America Asia Oceania Rest of Europe Asian crisis Asian crisis × Africa Asian crisis × North America Asian crisis × Latin America Asian crisis × Asia Asian crisis × Oceania Asian crisis × Rest of Europe Dot-com crisis Dot-com crisis × Africa Dot-com crisis × North America Dot-com crisis × Latin America Dot-com crisis × Asia Dot-com crisis × Oceania Dot-com crisis × Rest of Europe International crisis International crisis × Africa International crisis × North America International crisis × Latin America International crisis × Asia International crisis × Oceania International crisis × Rest of Europe Log-Likelihood P-value of Likelihood Ratio Test (LRT) Number of countries Number of observations

Model 3

Coeff.

Std. E.

Coeff.

Std. E.

0.00007*** 0.012** −0.137*** −0.013*** 0.041*** −0.002*** −0.009** 0.055*** −0.019*** −2.100*** 0.033*** 0.016*** −3.998*** −0.370 −4.544*** −3.234*** −2.793** −2.451*** −0.216**

0.000007 0.009 0.015 0.003 0.009 0.0004 0.007 0.019 0.002 0.315 0.006 0.004 0.908 1.225 0.843 0.754 1.193 0.753 0.105

0.0001*** 0.003 −0.124*** −0.012*** 0.031*** −0.002*** −0.005 0,116*** −0.018*** −1.890*** 0.034*** 0.019***

0.000008 0.010 0.016 0.003 0.009 0.0005 0.007 0.020 0.002 0.318 0.006 0.004

−0.037

0.087

−0.367***

0.082

−0.087 −0.203 −0.450 0.455 −0.523* −0.581 −0.176 0.345 −0.007 −0.475 −0.324 −0.605 −1.007 −0.325 −1.794*** 1.222*** 1.771*** 1.564*** 1.767*** 1.370*** 1.075*** −2316.480 0.000 86 1606

0.236 0.515 0.636 0.328 0.308 0.508 0.329 0.211 0.410 0.518 0.291 0.261 0.423 0.286 0.212 0.290 0.446 0.256 0.228 0.375 0.250

−2351.237 0.000 86 1606

The sample consists of 86 countries, grouped in 7 regions, from 1993 to 2013. See Table 2 for the definition of variables. Models 2 and 3 are based on a basic model given by Ratingit∗ = βXit + αi + εit, where the dependent variable is the arithmetic average, for each year, of the sovereign credit rating assigned by Moody's, Standard and Poor's, and Fitch, after converting the rating onto a numerical scale from 1 to 21, where 1 corresponds to the worst rating and 21 to the best rating. Xit is composed of four vectors of macroeconomic, external, government, and qualitative variables. Model 2 further incorporates dummy crisis variables, which take the value of 1 for the years of each crisis, and zero otherwise, and the dummy variables associated with each region. Model 3 further evaluates the effect of a particular crisis on the dependent variable for each region. In models 2 and 3 the region Eurozone is left out of the estimation. For each model, the first column depicts the estimated coefficient and the second column the standard error. The reported coefficients and their standard errors are obtained using the random effect ordered probit estimator. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. To simplify the exposition, we do not report the estimated coefficients associated with the cut-off points.

financial crisis are statistically significant and negative, showing that, on average, countries had lower ratings during these crises than during noncrisis years. This negative effect is stronger for the recent international financial crisis. Interestingly, the dot-com crisis, on average, seems not to have impaired sovereign ratings. The findings for model 2 show no statistically significant differences between the average ratings of countries in the Eurozone and North America, which have the highest ratings, followed by the rest of Europe, Oceania, Asia, Africa, and Latin America, in that order. In general, these results are in line with the discussion in Section 3 above. Next, in model 3, by incorporating a set of variables that are the product of the dummy variable crisis by the dummy variables

11

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

associated with each region, we investigate whether the effects of the crises on ratings differ by region. Since the omitted region is still the Eurozone, the estimated coefficient associated with each crisis variable measures the effect of a crisis on the Eurozone rating.5 As regards the Asian crisis, we can immediately observe that only the coefficient associated with the variable Asian Crisis x Asia is statistically significant, which means that the downgrade in the average rating during this crisis was limited to the Asian region. In contrast, the recent international financial crisis caused a downgrade in the average rating across all regions, as the estimated coefficients regarding these regions for this crisis are all negative and statistically significant. But the downgrade differed in magnitude across regions. The highest drop occurred in the Eurozone, followed by the rest of Europe, Africa, Oceania, Latin America, Asia, and North America. These results are consistent with the empirical evidence of Amstad and Packer (2015), who show that during the financial crisis of 2007–2009 the highest drop in the average rating—three notches, from AA+ to A+—occurred in the Eurozone. Finally, none of the coefficients for interactions between the dot-com crisis variable and the regional dummy variables are significant; that is, none of the regions observed a drop in its average rating during 2000–2002. 5. Conclusion This paper contributes to the literature in two main ways. First, it establishes an integrated set of factors that determine sovereign credit ratings, including important country-specific qualitative factors: history of past default, level of corruption, and political stability. Second, the sample not only incorporates a wide list of countries from most regions worldwide, but also covers the last two decades, allowing us to examine the effect of the Asian crisis, the dot-com crisis, and the recent international financial crisis on sovereign credit ratings. We find that all three qualitative factors play important roles in defining the ratings granted by agencies. Among the macroeconomic factors the important ones are GDP per capita, GDP growth, unemployment, inflation, and investment; among external factors, the amount of external debt, deficit in the current account, and liquidity risk. One government factor is important: level of public debt. We confirm the existence of regional differences in ratings. The Eurozone and North America lead the sovereign credit ratings, followed by the rest of Europe, Oceania, Asia, Africa, and Latin America. We also find that countries had, on average, lower ratings during the Asian crisis and the international financial crisis than during noncrisis years, and this negative effect is stronger for the recent international financial crisis. The dot-com crisis did not cause a downgrade in the average rating. The downgrade in the average sovereign credit rating during the Asian crisis was limited to Asian countries, whereas the recent international financial crisis caused a downgrade in the average rating across all regions, though with a greater magnitude in the Eurozone countries. The findings documented in this paper can be an important tool for governments and investors. Our conclusions raise important challenges to government and political institutions that are trying to attract foreign investment, as sovereign credit ratings depend considerably on factors that are under their control. Investors need to know which factors can potentially cause a downgrade on the rating of a certain country or region in order to invest selectively and better assess the systematic risks of their investments. We believe our results can be extended in future research by comparing the determinants of sovereign credit ratings with the determinants of government yield spreads, as yield spreads are also an important proxy for a country's risk of distress. It would also be important to examine in more detail the dynamics of the variations in sovereign credit ratings in the Eurozone region, especially in countries that benefited from rescue packages from the IMF, the European Central Bank, and the European Commission. Acknowledgments We would like to thank Maria G. Batista and Nancy D. Mann for helpful comments and suggestions. This paper is financed by Portuguese National Funds through FCT – Fundação para a Ciência e a Tecnologia, I.P., project number UID/ECO/00685/2013. References Afonso, A. (2003). Understanding the determinants of sovereign debt ratings: Evidence of the two leading agencies. Journal of Economics and Finance, 27, 56–74. Afonso, A., Gomes, P., & Rother, P. (2009). Ordered response models for sovereign debt ratings. Applied Economics Letters, 16, 769–773. Afonso, A., Gomes, P., & Rother, P. (2011). Short- and long-run determinants of sovereign debt credit ratings. International Journal of Finance and Economics, 16, 1–15. Aktug, R. E., Nayar, N., & Vasconcellos, G. M. (2013). Is sovereign risk related to the banking sector? Global Finance Journal, 24, 222–249. Amstad, M., & Packer, F. (2015). Sovereign ratings of advanced and emerging economies after the crisis. BIS Quarterly Review, 77–91. Andreasen, E., & Valenzuela, P. (2016). Financial openness, domestic financial development and credit ratings. Finance Research Letters, 16, 11–18. Becker, B., & Milbourn, T. (2011). How did increased competition affect credit ratings? Journal of Financial Economics, 101, 493–514. Bissoondoyal-Bheenick, E. (2005). An analysis of the determinants of sovereign ratings. Global Finance Journal, 15, 251–280. Bissoondoyal-Bheenick, E., Brooks, R., & Yip, A. (2006). Determinants of sovereign ratings: A comparison of case-based reasoning and ordered probit approaches. Global Finance Journal, 17, 136–154. Cantor, R., & Packer, F. (1996). Determinants and impact of sovereign credit ratings. Federal Reserve Bank of New York Economic Policy Review, 2, 37–54. Cavallo, E., Powell, A., & Rigobon, R. (2013). Do credit rating agencies add value? Evidence from the sovereign rating business. International Journal of Finance and Economics, 18, 240–265. Chen, S., Chen, H., Chang, C., & Yang, S. (2016). The relation between sovereign credit rating revisions and economic growth. Journal of Banking & Finance, 64,

5 In Model 3, while the estimated coefficient associated with each crisis variable measures the effect of the crisis on the average rating of the Eurozone (region omitted in the regression), the effect of the crisis on the rating of any other region is obtained by the sum of the estimated coefficient of the crisis variable and the estimated coefficient associated with the variable crisis multiplied by region. For example, the effect of the Asian crisis on the rating of the Asian region is given by the coefficient −0.087 of the variable Asian Crisis and the coefficient −0.523 of the variable Asian Crisis × Asia. The overall effect is thus −0.610.

12

Global Finance Journal xxx (xxxx) xxx–xxx

J.C.A. Teixeira et al.

90–100. Cipollini, A., Coakley, J., & Hyunchul, L. (2015). The European sovereign debt market: From integration to segmentation. European Journal of Finance, 21, 111–128. Depken, C., LaFountain, C., & Butters, R. (2011). Corruption and creditworthiness: Evidence from sovereign credit ratings. In R. Kolb (Ed.). Sovereign debt: From safety to defaultHoboken, NJ: John Wiley & Sons (Chapter 9). Dilly, M., & Mählmann, T. (2016). Is there a ‘boom bias’ in agency ratings? Review of Finance, 20, 979–1011. Erdem, O., & Varli, Y. (2014). Understanding the sovereign credit ratings of emerging markets. Emerging Markets Review, 20, 42–57. Feder, G., & Uy, L. (1985). The determinants of international creditworthiness and their implications. Journal of Policy Modeling, (1), 133–156. Ferri, G., Liu, L. G., & Stiglitz, J. E. (1999). The procyclical role of rating agencies: Evidence from the East Asian crisis. Economic Notes by Banca Monte dei Paschi di Siena SpA. 88. Economic Notes by Banca Monte dei Paschi di Siena SpA (pp. 335–355). Haque, N., Kumar, M., Mark, N., & Mathieson, D. (1996). The economic content of indicators of developing country creditworthiness. IMF Staff Papers, 43, 688–724. Haque, N., Mark, N., & Mathieson, D. (1998). The relative importance of political and economic variables in creditworthiness ratings. (IMF Working Paper 98–46). Hilscher, J., & Nosbusch, Y. (2010). Determinants of sovereign risk: Macroeconomic fundamentals and the pricing of sovereign debt. Review of Finance, 14, 235–262. Jüttner, J., & McCarthy, J. (2000). Modeling a rating crisis. Working paperMacquarie University. Larrain, G., Helmut, R., & Maltzan, J. (1997). Emerging market risk and sovereign credit ratings. (OECD Development Center Technical Paper 124). Maltritz, D., & Molchanov, A. (2014). Country credit risk determinants with model uncertainty. International Review of Economics and Finance, 29, 224–234. Mellios, C., & Paget-Blanc, E. (2006). Which factors determine sovereign credit ratings? European Journal of Finance, 12, 361–377. Monfort, B., & Mulder, C. (2000). Using credit ratings for capital requirements on lending to emerging market economies: Possible impact of a new Basel accord. (IMF Working Paper 1–69). Mulder, C., & Perrelli, R. (2001). Foreign currency credit ratings for emerging market economies. (IMF Working Paper 1–191). Pellegrini, L., & Gerlagh, R. (2004). Corruption's effect on growth and its transmission channels. Kyklos, 57, 429–456. Teixeira, J. C. A., Silva, F., Fernandes, A., & Alves, A. (2014). Banks' capital, regulation and the financial crisis. North American Journal of Economics and Finance, 28, 33−58.

13