5.4 RESULTS
6.4.3 Ordered Probit Models Testing for Asymmetry in Determinants
The results for the NKC models are displayed first in table 6.11. Once more, all the explanatory variables have been lagged.
Table 6.11: NKC Ordered Probit Models Testing for Asymmetry in Determinants
Upgrade Interactions
Downgrade Interactions Variable Coefficients Variable Coefficients
External Balance 0.0514 *** External Balance 0.0515 ***
External Debt -0.0031 ** External Debt -0.0031 **
Investment 0.0525 ** Investment 0.0534 **
Foreign Reserves 0.0389 *** Foreign Reserves 0.0377 ***
Per Capita Income 0.0002 *** Per Capita Income 0.0002 ***
Corruption -0.0310 * Corruption -0.0354 **
Regulatory Quality 2.8636 *** Regulatory Quality 2.7635 ***
Internet Users 0.0314 *** Internet Users 0.0293 ***
Upgrades -1.1920 Upgrades -0.1216
Downgrades -0.9764 *** Downgrades -3.6125
Upgrade*External Balance -0.0192 Downgrade*External Balance
0.2214 **
Upgrade*External Debt -0.0020 Downgrade*External Debt -0.0006 Upgrade*Investment -0.1883 Downgrade*Investment -0.4587 ** Upgrade*Foreign Reserves 0.1238 Downgrade*Foreign Reserves -0.0354 Upgrade*Per Capita Income -0.0004 Downgrade*Per Capita Income -0.0002 Upgrade*Corruption 0.0269 Downgrade*Corruption 0.2103 ** Upgrade*Regulatory Quality -1.1392 Downgrade*Regulatory Quality -1.5604
Upgrade*Internet Users -0.0307 Downgrade*Internet Users -0.1125 Total Panel Observations 160 Total Panel Observations 160
Log Likelihood -229.05 Log Likelihood -230.37
*, **, *** 10%, 5%, 1% level of significance, respectively. Source: Model estimations.
It can be seen in table 6.11 that all the interaction variables in the NKC upgrade model (the left-hand side) are statistically insignificant. This means that positive changes in the enclosed determinants are not appropriate incentives for NKC to upgrade a
specific African sovereign. Hence, local authorities cannot really do much to speed up the upgrade process of a specific sovereign, given the good performance of the enclosed variables. The downgrade variable is statistically significant in the upgrade interactions model. This confirms the asymmetry in the model: a previous downgrade in a specific sovereign’s credit rating has a negative influence on the current credit rating.
When considering the downgrade interactions model of NKC (the right-hand side), it can be seen that there are two variables that can help to soften the blow from a downgrade - the external balance and the corruption index variable, which are positive signs - and one variable that can exacerbate the downgrade phase - the investment variable, which has a negative sign. The results show that the current account to GDP (the external balance) could act as a buffer in order to smooth downgrade phases and protect the country. The same applies to the corruption index: if a country’s corruption index decreases, therefore indicating its becoming less corrupt; it will soften the effect of a downgrade on the country’s credit rating. The significance of the investment variable (FDI to GDP) shows that a decrease in investments could exacerbate the downgrade phase more by causing a decrease in the credit rating of an African sovereign.
The results for the Standard and Poor’s upgrade and downgrade interaction models are displayed in table 6.12. All the explanatory variables have been lagged.
Table 6.12: Standard and Poor’s Ordered Probit Models Testing for Asymmetry in Determinants
Upgrade Interactions
Downgrade Interactions Variable Coefficients Variable Coefficients
GDP Growth -0.1241 ** GDP Growth -0.1167 **
External Debt -0.0076 *** External Debt -0.0064 ***
Foreign Reserves 0.1193 *** Foreign Reserves 0.1107 ***
Internet Users 0.0570 *** Internet Users 0.0579 ***
Upgrades -4.3770 * Upgrades -0.8923 **
Upgrade Interactions
Downgrade Interactions Variable Coefficients Variable Coefficients
Upgrade*GDP Growth 0.3236 Downgrade*GDP Growth 0.7790
Upgrade*External Debt 0.0141 Downgrade*External Debt -0.0123 Upgrade*Foreign
Reserves
0.0175 Downgrade*Foreign Reserves
0.0983 *
Upgrade*Internet Users -0.0055 Downgrade*Internet Users 0.1596 Total Panel Observations 80 Total Panel Observations 80
Log Likelihood -108.54 Log Likelihood -106.96
*, **, *** 10%, 5%, 1% level of significance, respectively. Source: Model estimations.
It can be seen in table 6.12 that all the interaction variables in the Standard and Poor’s upgrade model (the left-hand side) are statistically insignificant, as was the case with NKC. This means that positive changes in the enclosed determinants are not appropriate incentives for Standard and Poor’s to upgrade a specific African sovereign. Therefore, sovereigns cannot really influence the upgrade process of a specific sovereign, given the positive performance of the included variables. It is interesting to see that the individual upgrade and downgrade variables are significant in the upgrade interaction models. The negative sign of the downgrade variable relates to the previous asymmetry findings, that is, a previous downgrade will influence current ratings negatively. The negative sign of the upgrade variable could even support the mentioned asymmetry further. A previous upgrade will influence a current upgrade negatively, meaning that successive upgrades for African sovereigns are less likely.
When considering the downgrade interactions model of Standard and Poor’s (the right- hand side), it can be seen that only one variable can help to soften the blow from a downgrade, namely foreign reserves. The positive sign of the variable indicates the ability of foreign reserves to act as a buffer to smooth downgrade phases. The individual upgrade and downgrade variables are also both significant in the downgrade interactions model. Once more, the asymmetry in the Standard and Poor’s model is confirmed. The model shows that ANY rating change (whether it is an upgrade or a downgrade rating action) will have a negative effect on the credit ratings of African sovereigns.
The results for the Fitch upgrade and downgrade interaction models are displayed in table 6.13. All the explanatory variables have been lagged.
Table 6.13: Fitch Ordered Probit Models Testing for Asymmetry in Determinants
Upgrade Interactions
Downgrade Interactions Variable Coefficients Variable Coefficients
GDP Growth -0.2776 *** GDP Growth -0.2538 ***
External Debt -0.0129 *** External Debt -0.0125 ***
Foreign Reserves 0.0280 ** Foreign Reserves 0.0283 **
Regulatory Quality 4.0099 *** Regulatory Quality 3.6557 ***
Upgrades -0.4792 Upgrades -0.4486
Downgrades 1.1596 ** Downgrades 1.2364
Upgrade*GDP Growth 1.7864 * Downgrade*GDP Growth -0.0750 Upgrade*External Debt -0.0600 * Downgrade*External Debt -0.0005 Upgrade*Foreign Reserves -0.2820 Downgrade*Foreign Reserves -0.0009 Upgrade*Regulatory Quality -1.0520 Downgrade*Regulatory Quality 0.2895
Total Panel Observations 82 Total Panel Observations 82
Log Likelihood -104.98 Log Likelihood -108.48
*, **, *** 10%, 5%, 1% level of significance, respectively. Source: Model estimations.
It can be seen in table 6.13 that the Fitch upgrade interaction model (the left-hand side) shows two upgrade interaction variables as statistically significant (on the 10 percent level of significance), namely, GDP growth and external debt. The positive sign for the interaction GDP growth variable shows that an increase in economic growth in previous upgrade periods acts as an incentive for Fitch to consider upgrading a specific African sovereign in the current period. The negative sign for the interaction external debt variable signals the negative effect of increased external debt (external debt to GDP) during previous upgrade phases on current credit ratings. Therefore African sovereigns do have some influence on the upgrade process of Fitch.
The individual downgrade variable is also significant in the upgrade interaction models. Surprisingly, the sign of the downgrade variable is positive and not negative as was the case in the Fitch model in the previous section. The positive sign means that a previous downgrade has a positive influence on the current rating of an African sovereign. This could perhaps mean that the probability of getting downgraded twice by Fitch is less likely.
When considering the downgrade interaction model of Fitch (the right-hand side), it can be seen that none of the interaction variables nor the individual upgrade and downgrade variables is statistically significant. There are therefore no identified variables that can soften the blow or exacerbate a downgrade by Fitch.
The results for the Moody’s upgrade and downgrade interaction models are displayed in table 6.14. All the explanatory variables have been lagged.
Table 6.14: Moody’s Ordered Probit Models Testing for Asymmetry in Determinants
Upgrade Interactions
Downgrade Interactions Variable Coefficients Variable Coefficients
Fiscal Balance -0.4230 *** Fiscal Balance -0.2591 **
External Balance 0.5918 *** External Balance 0.2637 ***
Per Capita Income 0.0040 *** Per Capita Income 0.0024 ***
Upgrades -21.7242 Upgrades -2.2801 **
Downgrades -1.4746 Downgrades -1.1935
Upgrade*Fiscal Balance -5.9656 Downgrade*Fiscal Balance
0.0747
Upgrade*External Balance 0.0686 Downgrade*External Balance 0.0349 Upgrade*Per Capita Income 0.0037 Downgrade*Per Capita Income -0.0002
Total Panel Observations 35 Total Panel Observations 35
Log Likelihood -17.02 Log Likelihood -27.34
From table 6.14, it can be seen that the Moody’s upgrade interaction model (the left- hand side) shows no statistically significant upgrade interaction variables. The individual upgrade and downgrade variables are also insignificant in the upgrade interaction model. Therefore, African sovereigns do not really have any influence on the upgrade process of Moody’s.
When considering the downgrade interactions model of Moody’s (the right-hand side), it can be seen that, once more, none of the interaction variables are significant in the model. There are therefore no identified variables that can soften the blow or exacerbate a downgrade by Moody’s. The individual upgrade variable in the downgrade interaction model is, however, statistically significant. The negative sign of the upgrade variable shows that a previous upgrade will have a negative effect on the current ratings of an African sovereign during downgrade phases. There is a decreased likelihood of another upgrade after a previous upgrade.
6.4.4 Overall Synthesis of Results
The results for NKC African Economics show that the research unit tends to rate African sovereigns independently of the business cycle; in other words, through-the- cycle. The ratings of NKC are therefore inherently more stable, which is what most market participants prefer (Kiff et al., 2013). The results of NKC also correspond to the findings of Kaminsky and Schmukler (2002) and Dimitrakopoulos and Kolossiatis (2015). NKC is therefore slow to react and sticky. As for asymmetry, there are signs that the research entity places a greater emphasis on previous downgrade actions when determining future ratings for African sovereigns. Prior rating downgrades have a negative influence on ratings on the future, which could be an indication of cliff effects, as proposed by Kiff et al. (2013); and rating momentum, as proposed by Topp and Perl (2010). The external balance and the corruption index could soften the blow of future downgrade actions for the entity, given a downgrade period. Negative movements in investment could exacerbate a downgrade phase even further, given a downgrade period.
The results for Standard and Poor’s show that the agency assigns ratings to African sovereigns procyclically: in other words, they assign point-in-time ratings, which are
inherently more accurate. There is a higher probability of attaining a higher rating during booms and a lower rating during recessions. This procyclical behaviour corresponds to the findings of Topp and Perl (2010) and Amato and Furfine (2003): here, both studies also concluded that Standard and Poor’s are procyclical in their rating assignment process. In addition to the procyclicality, there is also evidence of asymmetry in the rating process of Standard and Poor’s. Previous downgrades have a negative influence on future rating movements (rating momentum and cliff effects). There is also evidence that previous upgrade actions have a negative influence on present rating actions as well. These results correspond to the findings of Freitag (2015) that if a rating change occurs at present, there is an increased probability of rating changes in the future. It was also shown that foreign reserves can serve as a buffer to smooth downgrade phases. The significance of the foreign reserves’ variable (FDI to GDP) agrees with the results of Broto and Molina (2014), which confirm the importance of a country’s stock of reserves.
There is evidence that Fitch also allocates point-in-time ratings like Standard and Poor’s. There is thus an increased probability of getting downgraded in a recession and upgraded in a boom phase. It should be noted that Fitch has not been included in many studies on the topic, but the results correspond to the findings of Topp and Perl (2010) and Amato and Furfine (2003) for Standard and Poor’s. Asymmetry is also present in the rating assignment of Fitch, as future rating movements are influenced by downgrade actions. In contrast to NKC and Standard and Poor’s, there are two variables that influence the actions of Fitch during upgrade phases, namely GDP growth and external debt. An increase in the GDP growth of an African sovereign during a previous upgrade phase could have a positive effect on the current rating of the specific country. An increase in the external debt of a country during a previous upgrade phase could have a negative effect on the current rating of a specific African sovereign. It was also shown that previous downgrades have a positive influence on the current rating of African sovereigns. This could indicate that the probability of successive downgrades, cliff effects and rating momentum, by Fitch is low.
In the Moody’s model, it was shown that Moody’s assigns ratings procyclically as well. There is a higher probability of being upgraded during boom phases and downgraded during recessions. Interestingly, it was also shown that previous upgrades have a
negative effect on current ratings. This perhaps confirms the asymmetry in the models that most rating agencies tend to downgrade African ratings more than upgrade.
The procyclicality of Standard and Poor’s, Fitch and Moody’s could show that the three major rating agencies are assigning ratings by making use of a point-in-time methodology. According to Ferri et al. (1999), the rationale for procyclical behaviour for agencies is to protect their reputations. The ratings of NKC are perhaps not procyclical, like the ratings of the other rating agencies, because, as a non-formal rating agency, they do not have a reputational risk yet. The significance of the downgrade variables in all models corresponds to the research by Mora (2006) and Broto and Molina (2014) who concluded that upgrade phases are stickier than downgrade phases. Broto and Molina (2014) state that, “Downgrade periods are deeper and faster than those of upgrade phases, so that the probability of a future downgrade given a past downgrade is higher than that of a future upgrade given a past upgrade.”
6.5 CONCLUSION
The purpose of this chapter was to investigate the effect of the business cycle and upgrade and downgrade actions on sovereign credit ratings in Africa in order to analyse procyclicality and asymmetry trends of the three major credit rating agencies (Standard and Poor’s, Fitch and Moody’s) and NKC African Economics (a South African based research entity). The annual rating levels, business cycle indicators, upgrade and downgrade actions and other determinants, as identified in chapter 2 for the four rating agencies, were used for 27 African countries from 2007 to 2014. Two measures were used for the business cycle to ensure the robustness of the results. Ordered probit models were used.
All the models made use of the actual rating levels of the different rating agencies as the dependent variable. A linear and logistic transformation of the dependent variable was used in the first two sets of the models. The first set of models used the variables identified in chapter 3 as independent variables, as well as the two business cycle indicators mentioned before, in order to identify possible procyclicality trends in the data. The second set of models replaced the business cycle indicators with upgraded
and downgraded binary variables. The last set of models illustrated an interaction between the binary variables and significant determinants. The last two sets of models were used to identify the potential asymmetry trends in the data.
The evidence from the first set of models supports the research by Freitag (2015), that there are cases where rating agencies take the business cycle into account. NKC rate sovereigns independently of the business cycle, and therefore through-the-cycle. This means that ratings by NKC are inherently more stable as it takes issuer-specific characteristics into account. Standard and Poor’s, Fitch and Moody’s take the business cycle into account when assigning ratings to African sovereigns. This means that there is an increased probability of African sovereigns getting upgraded during boom phases and downgraded during recession phases by the mentioned rating agencies. The ratings of the major rating agencies are therefore more volatile, but, at the same time, more accurate since they take current information into account.
The second sets of models showed that downgrades have a greater influence on the direction of future rating movements than upgrade actions for NKC, Standard and Poor’s and Fitch. It was also shown that prior rating downgrades by the mentioned rating agencies have a negative influence on ratings in the future. Therefore, there is an increased likelihood of African sovereigns being downgraded successively. Prior rating movements do not influence future rating movements of Moody’s.
Results from the third sets of models indicated that there is divergent behaviour between upgrade and downgrade phases between the respective agencies. It is clear that rating agencies react differently to different determinants during downgrade and upgrade periods. Firstly, none of the significant determinants included in the Moody’s models can aid in softening downgrade phases or help to speed up upgrade phases for the included African sovereigns. Also, none of the significant determinants included can help to speed up the upgrade process for NKC and Standard and Poor’s.
There is one variable that can help to speed up the upgrade process, given an upgrade in terms of Fitch, namely, GDP growth; and one variable that can have a negative influence on the future rating movements of African sovereigns given an upgrade, namely, external debt. This shows that sovereigns do have some influence in terms of
upgrades when it comes to the ratings of Fitch. There are variables that can aid in smoothing the path of downgrades in terms of NKC and Standard and Poor’s. The external balance and the corruption index (NKC) and foreign reserves (Standard and Poor’s) can aid in softening the blow for future rating movements from a previous downgrade. It was also shown, in terms of NKC, that investment can exacerbate a previous downgrade action even further.
CHAPTER 7
SUMMARY OF FINDINGS AND RECOMMENDATIONS
7.1 INTRODUCTION
The research question in this thesis was to determine whether credit rating agencies invest enough time and effort into accurately rating African sovereigns in order to reflect the economic, financial and political milieu of the continent. Four supporting objectives were established in order to answer the initiating question of the study. The first was to establish the determinants of sovereign credit ratings of African countries. The aim was to evaluate if the identified determinants in the literature are relevant to African economies, and if there were different significant determinants among the respective rating agencies. This was addressed in the first empirical essay in chapter 3.
The second objective was to investigate the determinants of sovereign credit ratings in Africa according to regional and income classifications, which was covered in the empirical essay in chapter 4. The aim of this objective was to test if there was a difference in the significant determinants for different regions and income classifications of sovereigns on the African continent.
The third objective, which was addressed in the empirical essay in chapter 5, was to identify credit rating leaders and followers in the African sovereign rating market. The aim of this objective was to establish if the rating actions of the respective rating agencies had any influence on each other.
The fourth and last objective was to analyse the procyclicality and asymmetry trends of African sovereign credit ratings. The aim of this objective was to determine if the