Bond Rating Analysis Summary

Bond rating from the early 20th century American, after 100 years of development in the United States has formed a complete system, and in maintaining the stability of the financial system plays an important role in China, with the bond market, the emergence and development, the bond rating companies have emerged, mainly in the current international credit, Moody's and Standard & Poor's three.

Scholars of the bond rating of the study from the following two aspects: (1) bond rating model, this research examines the impact of various factors and their ratings of high and low impact on the rating function of the form, the application of methods statistical and measurement techniques, (2) bond rating and bond yield to maturity, the relationship between stock returns, these studies primarily through event study to examine changes in bond ratings and its creditors, shareholders wealth effects. below will These two aspects of the research review.


First, the bond rating model
Bond rating modeling ideas can be expressed as the following general model:
Which r @ ti i's bond rating company, Xji is a factor that may affect the rating. To establish the greatest difficulty rating model is the lack of appropriate economic theory as a guide, which determines the functional form of the model and variable selection in large extent can only be a positive attempt on from the way in terms of the rating model statistical techniques are ordinary least squares (ODS) regression, multiple discriminant analysis (MDA), order Probit (or Logit) models and Artificial neural Network (ANN).

1. Fisher ordinary least squares regression regression techniques to study the earliest introduction of bond market risk premium, and successfully explained 75% of the bond risk premium changes. This has inspired researchers began to use regression techniques to study the bond rating.

Horrigan for the first time the bond rating of some financial indicators such as total assets, current ratio and return to profitability, etc. He also used a 0-1 variable to represent the bonds of the secondary (subordinate) position. Horrigan investigated in 1959-1964 rating remained stable during the sample companies, and use 9-1 to represent values ​​from Aaa to B, C rating, this approach has been widely used follow-up study. West (1970) criticized Horrigan (1966) rely in financial variables and the lack of financial support for the practice of theory, he used the Fisher (1959) the economic variables to examine bond ratings. section 1949, using the estimated coefficients to predict the 1953 and 1961 bond rating, West found that ratings by influencing other investment options to change the bond yields and corporate bond in the control of the decision of the firm-specific factors, risk premium, the bond rating will systematically affect the bond yield.

PogueandSoldofsky ODS using a modified regression model to case-by-classification (pair-wise classification), where the dependent variable taking two values: 1 - companies with higher ratings, 0 - lower-rated companies, they examined the four samples bonds: (1) Aaa and Aa bonds, (2) Aa and A bonds, (3) A and Baa bonds, (4) Aaa and Baa bonds. This two (two-group) classification procedures to avoid Horrigan and West of the multiple range (interval) problem of comparison, however, by a factor of more procedures can not explain the other bond rating the quality and extreme groups (such as Aaa and Baa), and a comparison between non-adjacent groups, will lost in between the bond rating information.

Technically, OLS regression model of flaw is: First, the use of continuous values ​​(9-1) can not reflect the risk of AA and the difference between A and BB and B the difference between the different risk and secondly, when the explanatory variable is ordinal number, the disturbance is usually not satisfied with zero mean and fixed variance of the OLS assumption, therefore, need to find a more appropriate statistical methods.

2. Multiple discriminant analysis model is the bond rating classification data, classification data, the use of specialized inspection techniques such as discriminant analysis, Probit and Logit regression would be more appropriate. Discriminant analysis is a multivariate data into different groups for the statistical methods, but also The most popular model in the bond rating models.

PinchesandMingo examined between Aa-B rating between the newly issued industrial bonds, and they two-step method to study: the first of 35 accounting and financial variables in the factor analysis to determine the orthogonal components of the data, and secondly, in Each factor in the selection of the most important variable for the second step discriminant analysis, and their discriminant analysis in the original sample with 69.7% accuracy in the test sample (hold-outsample) has 64.6% of accuracy. In Pinchesand Mingo study, they did not use a secondary (subordinated) variable secondary (quadratic) discriminant analysis in the new model, they used five variables, thereby increasing the Baa bond rating on the performance, but other levels of performance are reduced.

Ahman and Katz using quadratic discriminant analysis to examine the utility's bond rating, they use 30 variables and 14 variables were investigated in the classification accuracy of discriminant analysis in the original sample, they obtained 91% accuracy in test samples obtained 77% accuracy, so they think that public data can be multivariate statistical framework to accurately assess the bond rating.

Rapport, Murphy and Parr using factor analysis examines 123 Moody and S & P rating the same industrial company commercial paper, they first used 37 variables, followed by stepwise discriminant analysis to the second variable is reduced to six, won 83 percent of the overall classification accuracy. Bhandari, Soldofskyand-BoeL ... using linear discriminant analysis on the up and down the bond rating classification, and use the estimated model for the same bond rating classification. They look at 1974, 1975 49 years of industrial companies, and selected 24 companies rated as no change compared results show that compared to downgrades of their models to better predict rating raised.

Despite the multiple discriminant analysis is an effective data classification methods, but its defect is unable to adapt to the ordinal nature of bond rating, financial ratios as explanatory variables often do not meet the multivariate normal distribution assumptions.

3. Sort Probit
Probit (or Logit) is designed to sort the data of statistical techniques, it avoids the OLS and multiple discriminant analysis was used to bond rating can not be overcome when ordering and fixed interval (ie, the value of the interval between adjacent rating is - like a) problem. Kaplan and Urwi using Probit techniques to study, they believe to be the different values ​​of explanatory variables (such as 9-1) represents an ordinal relationship, but the difference between the number of different sequences is not the same as their theoretical differences The explained variable (Y: default risk) and the observed variables (z: bond rating) and maximum likelihood estimation to estimate the results show that eLS Probit better than the performance. Is · kandarandEmery Probit using eLS and sorting to study the terms of the bonds, bond ratings, and found that the terms of the bonds set help predict bond rating. Kamstra, Kennedy and Suan (2001) Logit and Probit used to predict the sort bond rating, and with eLS multiple discriminant analysis and comparison results show that the combination of their sort Logit model to better predict bond rating, therefore, Probit, in theory, although better than the OLS and multiple discriminant analysis, but did not seem in better predictive ability.

4. Artificial neural network research post-began to use neural networks to study the bond rating. Duttaand Shekhar on multiple regression and neural network models were compared, they estimate the sample companies are divided into samples and test samples, the sample used to train the neural network estimates and estimates regression coefficients, test samples used to test the neural network with Links to free download http://www.hi138.com regression results to compare results showed that, in the learning stage, the classification accuracy of the higher regression model, and In the test phase of neural networks with higher accuracy. Surkan, etc. using a hidden unit and multi-level (hiddenlayered) neural network model results show that two yuan to hide hidden unit level model is better than the classification accuracy of the higher-level model.

Utansand Moody uses a MLP (multi-layerper-ceptron) network to study the bond rating, and it compared with the linear regression model. Kwon, HahandLee in the neural network used in the OPP (ordinal pair-wisepArtitioninS) method, and with the traditional neural network (CNN) and multiple discriminant analysis to compare the results show that OPP better than CNN, and CNN and OPP better than the multiple discriminant analysis.

Overall, the use of statistical and measurement approach to modeling bond ratings, there are ordinary least squares regression, multiple discriminant analysis, sorting Probit (or Logit) models and Artificial neural network in several ways. In statistics, although the latter three methods to better meet the rating distribution of the data, but found no evidence in that way consistently superior to other methods, therefore, a reasonable suggestion is used in the study should also these methods, and compare the performance of various methods.


Second, the wealth effect of bond rating changes in other study examined the bond rating and bond yield to maturity, the relationship between stock returns, using the main event study method.

For the bond rating changes will lead to wealth effects, early studies found inconsistent. Wcinstem using monthly bond yield data, Wakeman (1978) using monthly stock returns and the weekly bond yield data, found that bond rating changes did not lead to significant share price movements, while Katz uses the monthly bond yield to maturity changes, Gtier and Katzt using the average monthly bond prices, lngram, Brooks and Copelandt municipal bond yield to maturity using the monthly change in the bond rating changes that will lead to stock price significant response of these studies did not distinguish between the direction and size of the rating change, which may lead to the same conclusion could not find a reason. so different bond rating changes will lead to a different market response? follow-up studies to answer this question .

Pinches and Singletont found that bond rating changes (including up and down) the information content are relatively small. Wansley and Clauretie found that companies in the S & P Credit Watch appears and then confirm the rating, the market reaction is not obvious. Griffin and Sanvicente (1982) and Zaima and McCArthy (1988) found that rating downgrades to the shareholders to pass significant information, rating up there is no significant information content, while, Zaima and Mc Car-thyt proposed two competing hypotheses to explain the bond rating changes: information content and wealth redistribution hypotheses hypothesis hypothesis that the information content of ratings downgrades of the company's value should be decreased, while the rating increase the company's value should rise, the wealth redistribution hypothesis that bond rating downgrades will lead to decrease of wealth, and shift part of the wealth of shareholders .

Holthausen and Leftwich found from investment grade to speculative-grade rating downward market reaction than the average market reaction to downgrade more. Hand, Holthausen and Leftwich and Gohand Ederington that bond rating downgrade is more, the greater the stock price reaction: Secondly, in the speculative-grade rating downgrade within the stock price reaction is greater than the reduction in investment-grade rating within. GobandEd erington that the impact of lower ratings may be a function of fundamentals. Specifically, they distinguish changes due to financial leverage and financial outlook deterioration caused by the downgrades, increased financial leverage that led to the downgrades caused a rise in stock prices, suggesting that shifted wealth from creditors to shareholders, while the mountain due to deterioration of the financial outlook downgrades have resulted in a stock price decline. In addition, Gob and Ederington also found that if the bond rating changes before the negative excess stock returns, the market response to downgrades of bond more strongly, but with different Holthansen and Leftwich, they found differences in bond ratings decline, and did not cause different stock price reaction.

Bond rating will not only to individual company's creditors and shareholders to transmit information, the transfer may also produce information (information transfer) effects, eg, Ak-lfigbe, Jeff and Ann Marie looked bond rating changes within the industry effects, found that downgrade negative valuation effect will be spread throughout the industry, they found that the effect depends on industry characteristics downgrade, specific business and industry competitors when rating the greater the reduction in the company's stock price reaction, downgraded the company in a dominant position in the industry , downgraded the company closely linked to its competitors, downgraded due to deterioration caused by the financial outlook, the more obvious effects of this industry.

Finally, some studies find that the market reaction to the existence of bond rating changes lag for example, Pinches found in the bond rating change within one year after, there was no significant excess returns, but there is weak evidence that changes in the ratings after the publication of a regression of stock price trends. Glascock, Davidson and Henderson found that changes in the ratings after the publication of a quarterly stock returns exists within the 4% to 5% of the reversal. Holthansen and Leftwich found a quarter after ratings downgrades in the memory of negative abnormal returns rate, indicating that there is a drift rate of return rather than reverse the trend. Dichev and Piotroski examines the changes in bond ratings after the long-term changes in stock prices, bond rating increase after that there is no significant excess returns, while bond rating lowered after one year, there are 14% to 15% of the excess rate of return, the second and third years of the excess returns were 4% and 6%, and this phenomenon is small and speculative-grade bonds, the company is more obvious, they pointed out This phenomenon is due to lack of response from the market caused.

Overall, the study's findings suggest that bond ratings provide a useful market information, market reaction to the bond rating changes depends not only on whether the rating also depends on the direction and magnitude of changes in technology, the event study is to examine bond rating of the main method of the wealth effect, therefore, important to note that in the event of measurement problems and their possible impact on the conclusions of this problem in the long window studies may be more serious.


Third, the sum of this bond rating of the West made a comprehensive review, and its research methods and conclusions of the corresponding review. Rating study examined the rating of statistical modeling techniques and rating wealth effect (or the market reaction). use of statistical and measurement methods to modeling bond ratings, there are ordinary least squares regression, multiple discriminant analysis, sorting Probit or (Logit) models and artificial neural network in several ways in which, after the three methods in statistical to better meet the rating data distribution, so the interpretation and prediction is better than ordinary least squares regression, and between these three methods and no significant differences between the pros and cons. The market reaction to rating changes, bond rating for the market research that provides useful information to the market's reaction to the bond rating changes depends not only on whether the rating also depends on the direction and magnitude of change.

Currently, the bond rating companies and in the development of China's capital market plays an increasingly important role. Then the researchers, an important issue is how China's national conditions, to establish some effective bond rating model, ratings for the market to provide useful information. In this regard, Western research can provide us with useful guidance and reference.

Note: This article involved in the charts, notes, formulas, etc. Please read the original text in PDF format. Links to free download http://www.hi138.com