تجزیه و تحلیل عوامل موثر بر گسترش بازده اوراق قرضه با مدل بیزین میانگین

This paper analyzes determinants of country default risk in emerging markets, reflected by sovereign yield spreads. The results reported so far in the literature are heterogeneous with respect to significant explanatory variables. This could indicate a high degree of uncertainty about the “true” regression model. We use Bayesian Model Averaging as the model selection method in order to find the variables which are most likely to determine credit risk. We document that total debt, history of recent default, currency depreciation, and growth rate of foreign currency reserves as well as market sentiments are the key drivers of yield spreads.

Analysis of country default risk is an important issue in international finance. The topic is of particular importance in emerging markets, as the default risk is the primary factor in determining the cost of international capital. Thus, there exists a comprehensive literature on identifying the determinants of country default risk. Empirical papers mostly rely on regression models where a default indicator is regressed on potential explanatory variables. These papers found a great variety of variables significant in explaining default risk. One can observe, however, high variation between the reported regression models and the determinants they include – variables are found to be significant in some papers, whereas in others they lack significance.1 This indicates a high degree of model uncertainty. Based on this observation, we contribute to the literature by applying Bayesian Model Averaging (BMA) – which explicitly accounts for model uncertainty – to analyze the determinants of country default risk. The existing empirical literature on country default risk can be grouped by the indicator variable used in the regressions. Popular choices include credit ratings, yield spreads and an indicator variable showing whether a country has defaulted in the recent past. Credit ratings and yield spreads are highly correlated. Spreads, however, tend to lead changes in sovereign bond ratings (see Cantor and Packer, 1996 and Larrain et al., 1997), indicating that bond markets are highly efficient in collecting and processing information relevant to country default risk. A comprehensive strand of the theoretical literature proves the relation between yield spreads and (risk-neutral) default probabilities by applying ‘reduced-form’, also called ‘intensity-based,’ models (see, e.g., Jarrow and Turnbull, 1995, Jarrow et al., 1997, Duffie and Singleton, 1999, Duffie et al., 2003 and Longstaff et al., 2005). In addition to yield spreads, dummy variables that indicate whether a country is in default are used in some studies. We argue that yield spreads are advantageous in capturing default risk to using default dummies. Such dummy variables for defaults provide very crude approximations of the “true” credit risk. Furthermore, there is no unique agreed-upon definition of default.2 We thus use yield spreads of emerging market bonds to assess default risk. This measure is popular among researchers, with a comprehensive body of literature having been developed following a seminal study by Edwards (1984). The choice of specific explanatory variables to be included in an empirical investigation should, as far as possible, be guided by theoretical considerations. When it comes to model specification, however, economic theory gives no full guidance in selecting potential determinants of country default risk. The problem is especially acute in our context. First, there exists a number of variables that potentially influence countries’ ability to make debt service payments required to avoid a default. Second, an influential strand of theoretical literature (see, e.g. Eaton et al., 1986 and Eaton and Fernandez, 1995) emphasizes the importance of a country’s willingness to make payments since – in contrast to private enterprises – it is hard to enforce payments of sovereign countries. Thus, politics may play an important role in determining countries’ default risk and, hence, political variables could also be considered along with economic determinants. The results documented in the existing literature suggest that there is a high degree of uncertainty about the determinants of emerging markets’ yield spreads, and, thus, country default risk. A similar problem is addressed by Sala-i-Martin et al. (2004), who point out that “artistic” economic theory can suggest a very large number of potential explanatory variables. Bayesian Model Averaging (BMA) becomes an attractive modeling choice in such an environment. BMA explicitly acknowledges that the “true” model is not known, and, therefore, analyzes the entire model space (i.e., every possible model that could be constructed from a set of potential independent variables). Data mining concerns are, therefore, mitigated,3 as the results are based on the entire model space, rather than on a single model. Bayesian model selection techniques have been used in some prominent studies. Avramov (2002) and Cremers (2002) apply Bayesian techniques to the stock return predictability issue. Vrontos et al. (2008) use BMA to address model uncertainty in hedge fund pricing. Koop et al. (2008) use BMA to account for model uncertainty when investigating the consumption-wealth relationship. Sala-i-Martin et al., 2004 and Fernandez et al., 2001b, and Masanjala and Papageorgiou (2008) apply the Bayesian approach to selecting the determinants of economic growth.4Bandiera et al. (2010) apply BMA for estimation and prediction of sovereign defaults.5 We analyze yield spreads for 35 emerging countries included in the EMBI+ index for the years 1996–2010. Panel nature of the dataset enables us to control for world-wide events, such as the Asian financial crisis of 1997–1998. In addition to a number of economic variables typically used in yield spread analysis, we include data on political variables obtained from Heritage foundation. Market sentiment can also play an important role in determining sovereign yield spreads, even though it may not impact default probability directly. In order to address this, we include several measures of global and regional sentiment. As we are interested in basic/fundamental and long-term determinants of default risk, we perform our analysis using annual data. In addition, many economic and political variables are only available in annual frequency. We analyze 34 potential regressors using 374 observations. We document that total debt, history of recent default, currency depreciation, growth rate of foreign currency reserves and global market sentiment (measured by US stock market returns) are among the most important variables determining credit risk. On the other hand, variables commonly found to be significant in determining default risk, such as debt service ratio, budget balance, and inflation rate are found to have low influence on developing countries’ default risk. The rest of the paper is organized as follows. Section 2 presents an overview of some of the existing literature on yield spread determinants. Section 3 presents a brief background on BMA methodology. Section 4 presents the variables and the estimation results. Section 5 concludes.

We analyze the determinants of default risk of developing countries reflected by their sovereign yield spreads. Previous results of the literature indicate a high degree of model uncertainty. Thus, we apply Bayesian Model Averaging (BMA), which explicitly accounts for model uncertainty. BMA is suitable for the analysis of country default risk for a number of reasons. First, no full theoretical guidance is available regarding the determinant of default probability. Second, alternative model specifications proposed in the literature have produced conflicting results. Finally, the ratio of the number of observations to the number of candidate independent variables is low. There is, therefore, a chance that regression coefficients of irrelevant independent variables will not converge to zero, as it is normally the case with large samples. Analysis of default risk is an important issue in international finance, as the default risk is a key factor in determining the cost of external borrowing. We measure the default risk of a country by EMBI+ sovereign yield spreads. We include 34 candidate independent variables, and document that the ratio of external debt to GDP, default history, currency depreciation, growth rate of foreign currency reserves, and market sentiment proxied by S&P 500 returns are among the most important variables in determining yield spreads. On the other hand, some of the economic variables which have traditionally been included in the analysis as well as political and governance variables are shown to have low to medium probabilities of being included in the regression model. We believe our results highlight the importance of accounting for possibility of alternative model specifications. Whereas some highly intuitive determinants of default risk used in prior studies are confirmed to have a significant impact, others are found to have very low influence. We believe that accounting for model uncertainty is especially important when many of the candidate independent variables essentially measure few broad concepts (solvency, liquidity, and macroeconomic conditions, to be specific). Our results could be used as guidance for researchers who wish to model country default risk and use sovereign yield spreads as a proxy for such a risk. Some directions for future research may include the analysis of credit rating determinants and their impact on the yield spreads.