This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
This study aims to observe the determinants of sovereign credit ratings on emerging market sovereign debt spread and the level of how those credit ratings affect investment grades. This study builds on pervious literature on the topic as shown in the literature review below.
The recent financial crisis of 2008 and its aftermath is still being felt today where the risk of sovereign default is a real possibility across many nations. Many investors, fund managers and governments look to independent sources to assess market risk and verification of credit worthiness. Capital markets also require these risk assessments to ensure transparency and confidence for those with capital to invest. Thus credit rating agencies have an important role in the financial markets as their assessments may have a significant impact on a nation's ability to borrow and to stay liquid during financial troubles and the price of its investments.
Reinhart (2002) explains that sovereign credit ratings play a critical role in determining which countries have access to international capital markets where the ratings serve as a summary measure of a countries likelihood of default. His paper found that countries with lower sovereign credit ratings are not only unable to borrow from international markets but those sovereign rating also influences the terms at which the private sector can borrow from international sources. Jaramillo (2010) also agrees with Reinhart (2002) by describing that Sovereign credit ratings play an important part in determining access to international markets and the terms of that access. She shows that the threshold between investment-grade and speculative-grade ratings having important market implications in that investment grades have a far better advantage then speculative grades.
Jaramillo (2010) states that sovereign debt ratings are intended to be forward looking qualitative measures of the probability of default by rating agencies and are summary assessments of a government's ability and willingness to repay its debts in full and on time. Thus a higher rating represents lower risk of default and a lower rating represents a higher risk of default.
Many studies have found that generally higher credit ratings are associated with higher bond yields and credit default swap spreads due to perceived risk of defaults and economic difficulties. As credit ratings are an important aspect of assessing sovereign risk, it will then be likely that these ratings will have a major impact on sovereign bond spreads. Hull et al (2004) explains that a credit default swap spread is the cost per annum for protection against default where an investment grade rating will have a lower risk and thus a lower spread (cost) and vice versa for protection against default. Jaramillo et al (2011) shows that sovereign investment grade status is often associated with lower spreads in international markets. They found for the 35 emerging markets investigated that investment grade status reduces spreads by 36 percent, then what is implied by macroeconomic fundamentals. They also found that an upgrade in ratings within the investment grade category causes a 5-10 percent reduction in spreads while there is no impact for movements within the speculative grade asset class, ceteris paribus. Many past studies have shown that sovereign credit rating changes have in the majority of cases had a significant impact on a countries bond spread which affects a nation's ability to borrow.
Studies into sovereign credit ratings have shown that they create spill over effects where changes in ratings of risk cause capital to move into other asset classes. Flores (2011) investigation of 18 emerging markets found that changes to sovereign credit ratings caused spill over effects where regional effects where strongest as counties within the same region are likely to have strong financial and trade ties with the country being rated. The paper notes that this supports previous studies within the financial contagion literature that contagious financial crises are usually regional.
Flores (2011) also shows that the vast majority of the credit rating literature does not attempt to account for market anticipation of credit rating changes which is a factor that needs to be addressed when conducting research into how credit ratings affect sovereign debt and spreads. Afonso et al (2011) analysis tests this anticipation of sovereign yields and credit default swaps to see if the market has already absorbed the information contained in changes in the ratings before credit rating announcements. The papers agree that taking account for market anticipation of credit rating changes provides a more accurate estimate of the immediate market responses to rating announcements and that these announcements do impact sovereign yields.
Brooks et al (2004) investigated the impact of sovereign ratings on market returns and found that rating downgrades have a negative wealth impact on market returns. Thus the theory here is that negative factors which cause a ratings downgrade impact market returns negatively. They also found no evidence that emerging markets react more severely to rating changes on their markets compared to developed nations.
The above studies have shown that sovereign credit ratings can have a significant impact on a nation's economy. Thus the determinant on sovereign credit ratings is important to see which variables are key factors in ratings. Jaramillo (2010) explains that identifying the main determinants of investment grade status can help guide government policies towards keeping a high rating or trying to achieve an upgrade.
The three major agencies that rate sovereign debt is Standard and Poor's, Moody and Fitch. Hill et al (2009) studied the variations in sovereign credit quality assessments across these rating agencies where they find that while credit rating agencies often disagree about credit quality, it is usually only confined to one or two notches on the finer scale. In their paper the determinates of credit rating are based on seven factors which are per capita income, GDP growth, inflation, fiscal balance, external balance, external debt and default history. Most of the literature uses these variables in determining credit ratings such as Cantor and Packer (1996). Much of the literature cites Cantor and Packer (1996) as a starting point in determining sovereign credit ratings. Based on these variables, changes in sovereign debt ratings should be linked to these macro factors which in turn affect these ratings that should impact bond spreads and credit default swaps.
Cantor and Parker (1996) and most of the literature agree that per capita income, GDP growth, inflation, fiscal balance, external balance, external debt and default history explain the level of rating of the country. Alfonso et al (2007) also studied what is behind debt ratings with the same macro indicators and an added government effectiveness variable across the three major agencies for the period 1995 and 2005. They found the same relevant explanatory variables as in Cantor and Parker (1996) and their models show good overall predicative power.
Paget-Blanc et al (2006) used an ordered logistic model for 86 countries during 2003 to find the determinants of ratings and found that sovereign ratings are mostly influenced by per capita income, government income, real exchange rate changes, inflation rate and default history. Their study also highlighted the importance of a nation's corruption, as measured by Transparency International's Corruption Perceptions Index, which appears as a proxy for both economic development quality of the governance of a country which was significant.
Archer et al (2007) studied the determinants of Sovereign ratings by using a linear regression with panel corrected standard errors for 50 developing countries between 1987 and 2003. They found that find that regime type and most other political factors have little effect and that trade, inflation, growth, and bond default strongly affect sovereign ratings.
Bheenick (2005) uses a Ordered Response Model to determine credit ratings as he argues that ordinary least square estimation techniques as used by others such as Cantor and Parker (1996) as their study implies that the risk differential between credit ratings are equal such as a downgrade from B to B- is the same as C to C-. The paper states that the main finding was that quantitative measures are only a part of the input into sovereign rating decisions and on average, GNP per capita and inflation seem to be the most relevant economic variables.
These studies explain from the determinates that any negative change in these macro factors while holding everything else constant will cause a decline in ratings and any positive change will cause an improvement in ratings. This should cause a significant change in yields as ratings are linked to macro factors. Thus we should expect to find a negative change in a nations rating which reflect pessimistic macro factors. We should also expect to find a positive change in a nations rating which reflect optimistic macro factors.
The table below from Jaramillo (2010) describes the credit ratings issued by the three major agencies cassified into investment grade and speculative grade.
The methodology for this study will be based on a number of studies of the determinants of credit ratings such as Jaramillo (2010) and Afonso (2007). The sovereign credit ratings are obtained from three major credit rating agencies, Standard and Poor's, Moody and Fitch. The agencies use a number of ratings on domestic and foreign currency. For this study as with most in the literature long-term foreign currency sovereign ratings will be used.
In Afonso et al (2011) paper, a dummy approach is used for the ratings but in Reinhart (2002) an index approach is used where the ratings are transformed into numerical scores. This type of study will allow for a more informative study then using a dummy for an upgrade or downgrade. Numerical scoring will better represent the impact of a credit rating if it is upgraded or downgraded by a number of notches in a given period. As in Cantor and Packer (1996) and studies after them the sovereign credit rating information will be transformed into a discrete variable that codifies the decision of the rating agencies.
To study the determinants of credit ratings we will construct a variable (credit rating) that takes numeric values for each rating using a linear transformation, this assumes that the risk differential between each credit ratings are equal. Most of the literature uses this method when transforming credit ratings into numerical values. This will become the dependent variable and denoted as Sov.
We will also create a non-linear transformation of the ratings to assume that credit ratings do have an unequal risk differential between each credit rating. For example a credit rating in the investment category from A to A- has a 1 point differential while a credit rating in the speculative category from B to B- has a 2 point differential. This would represent that fact that as credit ratings from the top end of the scale is far better than those on the lower scale in a non-linear fashion then compared to a linear fashion.
As in Jaramillo (2010) who placed credit ratings into a linear scale and used this ordinal measure as the dependent variable to study to study the determinants of investment grade ratings for emerging market economies, the same can be done here. A credit rating within the investment category is given a numerical value of 1 while a speculative grade is given 0. This can help to explain what variables are significant for investment grade status and the relationship between investment grade. The data set will use end of year ratings for each country. The table below illustrates the credit rating measures applied by the three rating agencies showing the numerical values that will be used as the dependent variable to regress for the linear, non-linear and investment/speculative grade.
We expect that the non-linear ratings should show coefficients that are larger than the linear ratings as it has a larger scale. The linear or non-linear ratings may find different variables that are significant for example unemployment may be more significant in the non-linear ratings as it is more of an important factor lower down the scale which will be better represented by the non-linear scale.
Jaramillo et al (2011) finds that ratings do not differ significantly across the three agencies and found that in their study that the ratings across the agencies coincide for 94 percent of all observations. The literature finds that there is little disagreement in ratings from the agencies. For this study it is decided to use one data set for the ratings which is the Fitch set.
Afonso (2002) explains that assessing the credit risk of governments is not an easy task as there can be many factors that can be significant. He explains further that one must take into account both solvency facts and aspects such as the stability of the political system, social cohesion and the degree of interdependence with international economic and financial systems. The literature on the determinants of credit ratings uses many different variables in their models with many different varying results.
Building on the evidence provided by the literature we identify a set of potential determinants of credit ratings including additional variables. The reasons to include these variables and why they may determine credit ratings are given below:
Real GDP growth - Alfonso (2007) explains that higher economic growth should strengthen the government's ability to repay outstanding obligations. A growing economy is also more likely to absorb excess labour supply, decrease unemployment, and increase living standards and to downplay possible social conflicts and political instability. Higher economic growth should reduce the chance of default, thus economic growth should be positively correlated with the rating levels.
GDP per capita - The same reasoning used for Real GDP growth can be used for using GDP per capita where the size of the population is captured to get the average GDP per person of a nation. This can be used as a measure of prosperity where a higher GDP per capita should be able to finance its obligations. We would expect that higher GDP per capita should be positively correlated with the rating levels.
Inflation - Some papers argue that inflation can have an uncertain impact. Alfonso (2007) explains that higher inflation can reduce the real stock of outstanding government debt in domestic currency, leaving overall more resources for the coverage of foreign debt obligations but also notes that it is symptomatic of problems at the macroeconomic policy level. High inflation may even represent high demand for goods and service due to increased prosperity in the short-medium term. Baldacci et al (2008) explains that inflation is a key indicator of macro-economic stability and that high inflation can occur from monetization of debt and signal the need for higher interest rates which increases the cost of capital. He concludes for this reason that higher inflation should increase sovereign risk. We would generally expect to see inflation negatively correlated with the rating levels but may be positive.
Unemployment rate - This is a variable that is not widely used in the determinants of credit ratings but is used in Jaramillo (2010) and Alfonso (2007) with mixed results on significance. Alfonso (2007) explains the theoretical reason for unemployment being used as a variable is that a country with lower unemployment tends to have more flexible labour markets making it less vulnerable to changes in the economic environment. Lower unemployment should reduce fiscal burden of unemployment and social benefits and broaden the base for labour taxation. This should improve countries financial obligations, lowering its default risk. We should expect unemployment to be negatively correlated with the rating levels.
Current account balance - Baldacci et al (2008) explains that this variable is important indicator of economic performance and can provide information on the ability to repay foreign debt, it shows the balance of trade, if there is an earnings outflow or inflow. Alfonso (2007) argues it is difficult to determine how this variable is correlated with credit ratings. He explains that a higher current account deficit could signal an economy's tendency to over consume, undermining long-term sustainability, but it could reflect rapid accumulation of fixed investment, which should lead to higher growth and improved sustainability over the medium term. Jaramillo (2010) describes that a large current account deficit suggests a high dependence on foreign capital which could be seen as venerable for a economy or a sign of strength and high demand. Most of the literature suggests that we should expect that the current account balance should be positively correlated.
External debt to GDP - This variable is important in determining the level of debt owed by a country to foreign creditors. Alfonso (2007) explains that the higher the overall economy's external indebtedness, the higher becomes the risk for additional fiscal burdens. The higher the debt, the more vulnerable a country will be in changes in the economic environment. Also more resources are directed at paying foreign creditors then on domestic projects. Thus higher debt levels will increase a countries default risk. We expect this variable to be negatively correlated with the rating levels.
Central government debt to GDP - This variable is the total debt owed by a government. This is debt usually taken to finance government operations, projects and deficits. Jaramillo (2010) describes that the higher the debt burden of a government, the larger the transfer effort the government will have to make over time to service its obligations, and therefore a higher risk of default. As with external debt we would expect this variable to be negatively correlated with the rating levels as higher total government debt would make it more difficult to finance debt.
Savings to GDP - This variable is not used in the literature but can be an important in determining the level of savings a country has which can be used to buffer a negative economic climate, to pay off debt or funding for investments. But it is difficult to assess if a higher savings rate would have a positive impact on rating levels as a high savings rate would make debt burdens more manageable, thus improving a countries ability to meet debt obligations. But as Mody et al (2012) has shown that uncertainty is significantly associated with higher household savings. They reported that economic uncertainty such as the period 2007-09, a country can materially increase its saving rates which will contribute to lower consumption and GDP growth. Based on Keynes paradox of thrift that if savings increases, where individuals save more of their income, which is a prudent measure in times of uncertainty, then aggregate demand will fall, consumption will decrease and growth will slow. This should cause a negative impact on rating levels as this would make it difficult for a county to meet debt obligations. Overall we expect this variable to be negatively correlated with the rating levels.
Exports to GDP - In Jaramillo (2010) this variable is used as she explains that a higher ratio suggests a greater capacity to obtain hard currency to repay foreign currency denominated debt. The paper also notes that in most previous studies, exports are included only as a metric for external debt, while her paper introduces it as a independent variable. We would expect this variable to be positively correlated to rating levels.
Broad money to GDP - This variable is added as a proxy for financial depth. Jaramillo (2010) is the only paper in the literature to use this variable. She uses this variable for financial depth and financial intermediation as she explains that countries that have access to a deep and diversified pool of finance are in a better situation than those whose private savings are low and whose financial system is repressed, giving the government more financial flexibility and its ability to sustain debt obligations. We would then expect this variable to be positively correlated to ratings levels.
Vix index - This variable is not used in the literature but used as a proxy for international economic conditions where a higher vix represents high volatility and a low vix represents low volatility. Positive global conditions should allow governments to better meet debt obligations as their economies are in stronger positions then poor global conditions such as a global contraction. Jaramillo et al (2011) and Floriana (2008) uses the VIX is a proxy for the risk appetite of international investors when studying the determinants of emerging bond spreads. Floriana (2008) found a long term relationship with the vix and emerging bond spreads where an increase in the vix resulted in higher spreads which reflects a decrease of risk appetite of investors on the global markets. We expect this variable to be negatively correlated to ratings levels.
Political risk - Cuadra and Sapriza (2006) in their paper report emerging economies tend to experience larger political uncertainty and more default episodes than developed countries which make political risk a key variable to use in this paper. Jaramillo (2010) uses the International Country Risk Guide as a proxy for political risk. She explains that the rule of law and respect for property rights provide confidence that political and civil institutions have a strong commitment to honouring financial obligations, thus political risk measures a countries wiliness to repay debt obligations. Alfonso (2007) uses six World Bank governance Indicators and found that government effectiveness was the only one that was significant. The index we will use in the index of economic freedom from the Heritage Foundation. This index has separate scores for business freedom, trade freedom, fiscal freedom, monetary freedom, investment freedom, property rights, corruption, and labour freedom. From these indexes we will use the overall score that incorporates all indexes to be used as a proxy for political risk. We expect this variable to be positivity correlated to rating levels as the higher the overall score which represents a lower risk of default.
The variables above will be used in determining credit ratings. Other papers use a mixture of variables that we have omitted as they are not suitable. For example Alfonso (2007) uses extra variables in their models that are averages, such as average GDP growth and average GDP per capita. Paget-Blanc et al (2006) uses mobile phone usage for a proxy for the level of technological development, no other paper uses technology to determine credit ratings. This is not used as we believe it would not be representative of development as a country can import phones with relativity low physical infrastructure and technology needed to operate wireless communication.
The variables used for this study are given in the table below where the definition of each variable is given and its unit of measurement. The source provider is given for the data set. From the data set, only emerging markets where selected and where there were no gaps in the data. From the data set, 15 countries where chosen to assess the determinants of credit ratings. The date range is from 1995 to 2010 and end of year observations are used.
A panel data framework will be used as done in most of the literature on the determinants of credit ratings, for example in Jaramillo (2010) a panel data framework with a logit model. The model specification is written as:
Where Sov denotes the credit rating value, ï¡ is a constant, ï¢X is a vector of time varying explanatory variables as shown above. represents disturbances that are independent across countries and across time. i denotes the country and t denotes the time period. The equation will be estimated using a pooled regression, a fixed effect and a random effect for the linear and non-linear ratings. A logit model as used in Jaramillo (2010) will be used for the investment/speculative ratings as the values are either 1 or 0.
The pooled regression is simply having both elements of time series and cross-sectional data (panel data) and regressing the data to interpret. Dougerty (2011) explains that in principle random effects estimation is more attractive because observed characteristics that remain constant for each individual, in our case countries are retained in the regression model while in the fixed effect model they are dropped. Another advantage which is noted is that with random effects over fixed effects, that we do not lose n degrees of freedom.
Afonso et al (2007) in their work set out the case that in normal conditions where all the estimators are consistent then the three methods used in terms of efficiency is that a random effect model is preferable to the fixed effects which is preferable to a pooled regression. A normal condition is if country specific error is uncorrelated with the explanatory variables. If not then we prefer the fixed effect. The Hausman and Breusch-Pagan Lagrange multiplier test will determine which model is preferable.
Dougerty (2011) explains that the Hausman tests if ï¡ are distributed independently of the X variables. If the random effects estimates are subject to unobserved heterogeneity bias, then it will not differ systematically from the fixed effect estimates and the fix will be the better method. Dougerty (2011) summaries that the Hausman test determines whether the estimates of the coefficients are significantly different in both regressions and that if any variables are dropped in the fixed effects, then they are excluded from the test.
Dougerty (2011) also explains that if the random effects model is the better model against the fixed effects, then it should be considered whether there are any unobserved effects at all. The model may be so well specified that the disturbance term consists only of the purely random component and there is no individual specific ï¡ term. Dougerty (2011) describes that in this case pooled regression should be used as there will be a gain in efficiency as we are not attempting to allow for non-existent within group autocorrelation and for inference we will be able to take advantage of the finite sample properties of OLS rather than having to take on the asymptotic properties of random effects.
A logit model will be used for the investment/speculative ratings as it takes a value of 1 and 0. Afonso et al (2007) in their analysis of credit ratings also used a probit model for determinants as they believed it was a natural approach as rating is a discrete variable and reflects an order in terms of default, where the credit rating agencies is categorical in nature. Jaramillo (2010) on the other hand uses a logit model. The logit and probit model is essentially the same where the differences are that logit uses the cumulative standard logistic distribution and probit uses the cumulative standard normal distribution. Both models have very similar results when used for estimation. The logit model coefficients cannot be read as with OLS coefficients as they are in log-odds units, but can help us determine if the variables are significant and the relationship with investment/speculative rating levels.
Most of the literature on the determinants of credit ratings does not mention the possibility of multicollinearity but Bheenick (2005) notes that the use of large economic variables in a model can introduce the possibility of multicollinearity, that the independent variables are closely related to each other. Dougerty (2011) explains that a high correlation does not necessarily lead to poor estimates, any regression will suffer from some extent and it is a matter of degree. It is only of concern if it is believed that it is affecting the regression results seriously. Multicollinearity also does not mean that the model is misspecified.
The correlation matrix between the economic variables used is shown below. We find that generally the correlations between variables are not severely high. We find the highest correlation is between government debt and external debt at 0.7225. From the correlation matrix we determine that the values on average do not indicate that Multicollinearity should not be a significant issue affecting the regressions.
The table below shows the results of using the linear rating series with the three methods of random effects, fixed effects and pooled regression. Column A, C, E reports the unrestricted models where all variables are used to determine the effects on credit ratings. Column B, D, F reports only the variables that are statistically significant. This was conducted by systematically dropping the highest insignificant variable until only variables that are significant up to the 10% level are left. The z and t statistics are in parentheses and stars indicate if the level of significance is at the 1%, 5% or 10% level. Based on the significance of the coefficients, the restricted models are more suitable in determining credit ratings then the unrestricted models.
From the reported models we can identify if the random effect or pooled regression model will be the most appropriate. Using the Breusch-Pagan Lagrange multiplier test for random effects where the null hypothesis is that the variance across entities is zero we obtain a Prob <0.05 for the unrestricted models, thus we do not reject the null that there are no random effects, there is evidence of differences across counties. We find the same results for the unrestricted models. Thus we prefer to use the random effect model over the pooled regression.
Using the Hausman test in stata to determine if the random or fixed effects model is appropriate we use the null hypothesis that that the coefficients estimated by the random effects are the same as the ones estimated by the fixed effects. We use the unrestricted model A and C. Using the test we obtain a Prob>0.05 which indicates that using the 5% level that this is insignificant. As the p-value is insignificant we prefer to use the random effect model.
Using the Hausman test on the restricted model B and D, we report a Prob<0.05, thus we can reject model B as inconsistent and the fixed effects model is appropriate. Overall we would assume that the better model is model D based on these tests where the coefficients are all significant.
Wooldridge (2009) explains that we can test whether a group of variables has an effect on the dependent variable, thus we can see if variables are jointly significant by using the F test. Using a 5% level of significance we find all models have a Prob<0.05 showing that all the variables jointly are different from zero. We find from the results that generally the variables that are significant in the unrestricted models are still significant in the restricted models.
Following the same procedure as above we now regress using the non-linear ratings with the results reported below with the three methods of random effects, fixed effects and pooled regression.
Using the Breusch-Pagan Lagrange multiplier test for random effects to determine if pooled of random effects model is more appropriate. The test reports that both the unrestricted and restricted models have a Prob<0.05 which indicates that the random effects model is preferable to the pooled regression.
For the restricted models H and J we report a Prob>0.05 which indicates that the fixed effects model is more appropriate. Thus the restricted fixed effect model is the preferred model as with the linear ratings where the restricted fixed effect model is also preferred.
Below we now use the logit model to regress the investment/speculative ratings scale. The table below shows the results of using the pooled regression, the random effects and fixed effects model.
The reported unrestricted models is tested to see if variables are jointly significant where we obtain a Prob<0.05 for all three models. Jaramillo (2010) in her paper used the Hausman test to determine that the random effect logit model was the appropriate model to use. In this paper we find that the fixed effects is the appropriate model as we obtain a Prob<0.05 when using the Hausman test.
Real GDP Growth
As expected we find that GDP growth is positive expect for model E. We find with the linear ratings that this variable is not significant at the 5% level and do not make it into any of the restricted models. On the other hand for the non-linear ratings we find that GDP growth is significant for the fixed and random effects model for both the unrestricted and restricted model. This may be due to GDP growth being a significant factor on lower ratings but not higher rating and thus is picked up on the non-linear scale more effectively then the linear rating scale. This suggests that GDP growth is important for countries with low ratings where growth can make a much bigger impact then a country with a high rating.
When it comes to the investment and speculative ratings we find that this variable is positive but not significant in determining investment status.
Afonso et al (2007) in their study with the same rating set (Fitch) found that GDP growth was significant for only there restricted fixed effects model while Jaramillo (2010) found this variable was not significant in determining credit ratings. We find using the preferred fixed effects model J where GDP is significant that if GDP growth increases by 1 full percent (e.g. 3% to 4%) that rating levels will increase by 0.1558.
GDP Per Capita
The linear and non-linear ratings shows that GDP per capita is positive for both fixed and random effects but not for the pooled regression as found with GDP growth. For the linear ratings we find that per capita GDP is significant for both random and pooled regressions. With the non-linear ratings we find that both random and fixed effects are significant. Afonso et al (2007) in his estimation found this variable being significant for all models, while Jaramillo (2010) found this variable was only significant for the pooled regression.
The investment and speculative ratings show a positive relationship but an insignificant variable in determining investment status. We find using the non-linear ratings and the preferred fixed effects model that if GDP per capita increases by $1 that rating levels will increase by 0.0012.
Savings Per GDP
This variable was found to be significant for all models and negative when using linear, non-linear and investment/speculative ratings. When using the restricted models we find that this variable is significant at the 1% level while for the logit models we find it significant at the 5% level.
As expected this variable was believed to be negative as countries under stress typically see increases in savings as found in Mody (2012) when looking at savings during the recession of 2007-2009 which was attributed to the precautionary savings motive. This may also support Keynes paradox of thrift that savings may be good for individuals but on the national level causes a weaker economy as total savings will decrease. In times of uncertainty and economic weakness, people save more, this in turn causes more economic weakness and uncertainty.
This variable is not used in the literature in determining credit ratings, but is found to be a very significant variable in this study. This variable shows that lower savings rate lead to higher ratings, which suggests that more income is being spent in the economy which can drive growth, employment, investment and lead to a stronger economy which should allow governments to better meet debt obligations.
When using the non-linear ratings and the preferred fixed effects, an increase of 1% (e.g. 10% to 11%) in the savings rate causes ratings level decrease by -0.2056. While for linear ratings the decrease is -0.1409.
We find that inflation is positive for the linear and non-linear ratings which were expected to be negative. This may suggest as Afonso et al (2007) explains that higher inflation can reduce the real stock of outstanding government debt in domestic currency, leaving overall more resources for the coverage of foreign debt obligations. Inflation may also represent strong demand and a strong economy pushing prices up.
The restricted models for the linear ratings show that inflation is significant for the pooled regression but not the fixed or random effects model. While for the non-linear models we find that it is significant for the random and pooled regression model but not the fixed effects. Inflation is not significant for the preferred fixed effects model when determining credit ratings. The investment and speculative ratings show that this variable is insignificant in explaining investment grades.
Afonso et al (2007) in his estimations found inflation to be negatively correlated with ratings and found to be significant for all models used. Jaramillo (2010) found inflation also being negatively correlated by insignificant for all models used in her paper.
We expected the unemployment variable to be negative but find that unemployment is positive for the fixed and random effects model but negative for the pooled regression when using the linear and non-linear ratings. We also find that unemployment is significant for the fixed and random effects model but not the pooled regressions.
The logit model shows a negative relationship with investment grade but insignificant in explaining investment grade.
It is difficult to explain why unemployment is positive and significant for the fixed and random effects model when using the linear and non-linear ratings. Afonso et al (2007) who also found a positive sign for unemployment explains that countries with structural reforms that raise unemployment in the short run but improve fiscal sustainability could provide an explanation for this, but further research would be necessary to validate this hypothesis. Jaramillo (2010) also found a positive sign for a pooled and random effect model but does not comment for this result as it is not a significant variable.
Another reason for the positive sign may be due to fact emerging markets are used in this study as where higher unemployment may cause wages to fall which will promote foreign investment and outsourcing due to labour advantage. This extra investment and funds can help improve conditions to meet debt obligations.
When using the non-linear ratings and the preferred fixed effects, an increase of 1% (e.g. 10% to 11%) in the unemployment rate causes ratings level increase by 0.2604. While for linear ratings the increase is 0.2103.
Exports to GDP
We find that exports is only significant for the pooled regressions but insignificant for the fixed and random effects model when using the linear and non-linear ratings. The positive relationship for both linear and non-linear pooled regression was expected as this would suggest that this helps a country's capacity to obtain hard currency to pay debt obligations. Jaramillo (2010) found this variable positive and significant for her random and fixed effects model. The logit model shows that export is positive but insignificant in determining investment grade.
Using the restricted pooled linear ratings model we find that an increase of 1% (e.g. 10% to 11%) in exports to GDP causes ratings level to increase by 0.0417 while for the non-linear ratings the increase is 0.0860.
For the linear fixed and random effects model and the non-linear fixed effects model we find it is negative which we did not expect as it is difficult to explain this type of relationship. As it is insignificant for these models and does not make it into the preferred restricted fixed effects model, we believe this variable is of little importance to explain the negative sign.
The current account is found to be insignificant for all restricted models for linear and non-linear ratings which show that it does not determine credit ratings. This variable was positive for the random and fixed effects model which was expected as a current account surplus should allow a country to better meet their debt obligations. The current account is not significant for the preferred fixed effects model when determining credit ratings. For the logit models this variable also shows a positive relationship but is also insignificant in determining investment grade.
Jaramillo (2010) also found this variable to be insignificant for all her logit models used. Afonso et al (2007) found this variable also to be insignificant when using the Fitch data set and found this to be a negative sign for all models used.
External debt is found to be insignificant for linear and non-linear restricted fixed and random effects model but significant for the pooled regressions. This variable shows a negative sign which was expected except for the linear unrestricted fixed effect model. But overall the negative relationship should explain the fact that higher foreign debt makes it more difficult to meet debt obligations which reduces credit ratings levels.
The logit models also produce a negative relationship with investment grades but are found to be insignificant for the fixed and random effects model, but significant for the pooled regressions.
Using the restricted pooled linear ratings model we find that an increase of 1% (e.g. 10% to 11%) in external debt to GDP causes ratings level to decrease by 0.0437 while for the non-linear ratings the decrease is 0.0860. External debt is not significant for the preferred fixed effects model when determining credit ratings. Jaramillo (2010) and Afonso et al (2007) both found external debt to be significant and negative for all the models used.
As expected Government Debt has a negative relationship with ratings level. All three ratings scales, show that this variable is negative and significant for all models used. Afonso et al (2007) found this variable negative and significant for all models used while Jaramillo (2010) who used domestic debt instead which is similar was also negative and significant. The literature on the determinants of credit ratings find this variable is a core variable.
The debt held by the government seems to be a big impotence in the ratings levels where using the linear ratings preferred fixed effects model, a 1% increase in government debt (e.g. 10% to 11%) decreases ratings level by 0.0297. For the non-linear ratings this is a decrease of -0.0916.
This variable is positive for all models used with linear, non-linear and investment grade ratings. This was the expected sign as greater financial depth should give the government more financial flexibility and its ability to meet debt obligations.
Broad money is significant for all models expect for the fixed effects model using linear and non-linear ratings. Broad money to GDP is not significant for the preferred fixed effects model when determining credit ratings but significant in determining investment grades. Jaramillo (2010) is the only paper to use this variable and found that it is positive and significant for all her models.
The political risk variable is positive for all models which were expected as a higher value represents a better economic environment. We also report that this variable is significant for all models for the linear and non-linear ratings.
This paper is the only one that uses The Heritage Foundation freedom index without the combination of other political measures averages together.
Using the linear ratings preferred fixed effects model, a 1 point increase in the political risk variable (e.g. 80 to 81) increases ratings level by 0.2471. For the non-linear ratings this is a increase of 0.4565.
The VIX index is found to be significant for all linear and non-linear models expect for the linear restricted pooled regression. This variable is not used in the literature for the determinants of credit ratings and is found to be negative as expected. This variable suggests to us that better global conditions are better for ratings level as countries are in a stronger position then poor global conditions. The logit model shows with investment grades that this variable is insignificant.
Using the linear ratings preferred fixed effects model, a 1 point increase in the vix index (e.g. 14 to 15) decreases ratings level by -0.0756. For the non-linear ratings this is a decrease of -0.1212.
Using linear ratings and the preferred restricted fixed effects model we find that the main significant variables were savings to GDP, per capita GDP, unemployment, government debt, political risk and the VIX index.
When using the non-linear rating scale and the restricted fixed effects model we find that the main significant variables were real GDP, per capita GDP, savings to GDP, unemployment, government debt, political risk and the VIX index.
The logit fixed effects model that used the investment and speculative grade scale, found the main variables were savings to GDP, government debt, broad money and political risk.
This paper used 15 emerging markets to identify the determinants of sovereign credit ratings and from the estimations found a number of core variables that impact rating levels. From this papers estimations we argue that the core determinants of ratings which were found to be significant for all ratings scales used are savings to GDP, government debt and political risk.
The use of savings to GDP to determine credit ratings has not been used in past studies but in this paper found to be one of the most significant determinants of credit ratings. This could be due to the fact that it reflects important conditions about the economy and its ability to pay debt obligations. Mody (2012) study and many others have shown that increasing savings typically occur during poor economic conditions where individuals save more of their income and thus reduce their consumption due to economic uncertainty and to buffer negative conditions. Keynes argued that if everyone increases their savings in the economy then the economy will contract and thus total savings can fall which is overall harmful for the economy. It is also possible that this variable is capturing other factors in the economy such as political uncertainty, poor employment factors and negative growth prospects which may cause individuals to increase savings. This may create an environment that increases a countries default risk and lower rating levels.
Many papers have used government debt to determine credit ratings and found to be a key variable in nearly all studies. This paper also finds that the negative relationship of government debt is a key component to ratings level where it is significant in all models used. This was an expected result as a higher debt burden would make it difficult to meet debt obligations as more resources are dedicated to debt repayments. Higher debt levels would also mean that it is venerable to higher interest rates which can increase the debt burden and make it more difficult to borrow money from the markets.
The use of the freedom index by the Heritage foundation for a proxy for political risk has also shown to be a variable that is highly significant for all models used. Other papers such as Jaramillo (2010) find generally their political risk proxies to also be a major component in explaining sovereign credit ratings. The political environment of a country is an important factor in the willingness of a country to pay debt obligations, how it can best build and allocate resources and how it decides to dedicate those resources in the pursuit of paying debt obligations. Cuadra and Sapriza (2006) explain in their paper that emerging economies tend to experience larger political uncertainty than developed countries, which makes political risk it important factor in this paper I determining sovereign credit ratings.
Though the variables described above where found to be the core determinates when looking through the estimations with all three ratings scales, there are other variables that cannot be completely dismissed. Real GDP growth and per capita growth was significant for the preferred fixed effects model when using linear and non-linear ratings. This tells us based on those models that growth of the economy and its per capita GDP improve the ability of a country to reduce its default risk as more resources are available to meet debt obligations.
We also find unemployment significant and having a positive relationship to ratings level for the preferred fixed effects model. This is a counter-intuitive relationship but the use of emerging markets in this study may be able to explain that a higher unemployment rate can cause a fall in labour costs, giving a labour cost advantage for foreign countries to invest in, which can help strengthen a countries overall economy.
Broad money is also a variable that is highly significant in all models except for the linear and non-linear fixed effects model, thus slightly miss's what we would argue is the core variables for all our ratings level. The logit model shows that this significant in determining investment grades. As a proxy for financial depth of a country, a country with a strong and developed financial system can have improved access to debt markets and a diversify pool of funds, this allows a country to be more financially flexible reducing default risk.
The VIX index was significant for the linear and non-linear models but not for the logit model. Based on the linear and non-linear ratings we find that the VIX index as a proxy for global risk has an impact in ratings. This is a difficult problem for countries to buffer themselves from as this is outside of their scope to manage. A global recession can have an impact on financial flows, global interest rates and available funds.
To add to the core variables, we note to a lesser extent that GDP grow, GDP per capita, broad money and the vix index also has an importance in determining credit ratings.
The findings in this paper can highlight certain policies that can be recommended and how it can improve rating levels. Savings to GDP can be used as an indicator for economic conditions as high rates represent poor conditions that force individuals to save and thus set favourable conditions such as low interest rates that lower borrowing costs, helps credit expansion and increase consumption, this should lower savings to GDP as economic conditions are strong.
Government debt has a big impact on ratings and a lower debt to GDP is more favourable to a higher debt level. Thus a government policy of running a budget surplus or balance budget would be better than a budget deficit growing debt to GDP. This would help a government in lowering the debt burden and ultimately lower borrowing costs.
Our estimations show that a better political environment improves credit ratings. To lower political risks governments should have policies that encourage openness in trade and investments. Corruption should be tackled and property rights respected. Also flexible and fair labour laws should be supported. This should create a more stable and favourable political environment which can help towards gaining international funds, a stronger economy and help in lowering borrowing costs. These policies can also help in developing a stronger financial system where investors will be confident in using.
When it comes to improving GDP and per capita GDP, policies aimed at fostering growth such as lower taxation rates, low interests and incentives for trade should make an impact in gaining a higher rating level. Higher GDP can also help a country buffer itself from negative global conditions as a country can look to its internal market for demand for goods and services rather than relying on the global market.
The results in this paper are limited and could be improved if different sets of data could be used. The use of better proxies for political risk, financial depth and global risk could improve results in the estimations. Also additional variables could improve results such as a variable that can represent technological development which is difficult to agree upon. The use of more countries can also give a better representation of the population and improved results. The difficulty in studying the determinants of credit ratings is how to transform ratings intro numerical terms that can be estimated as many different scales can be used as well as many different types of econometric models.
Further study in the area of the determinants of credit ratings can focus on many different aspects in which variables matter and what models would be most appropriate. The major credit ratings agencies use a number of different ratings other then the long term foreign currency credit ratings used in much of the literature. Determinants can be used against the domestic rating system, short term outlook and ratings watch system. Using other measures and variables can help in determining which factors matter for ratings and how countries and individuals can best use that information.