Disclaimer: This assignment was written by a student and is not an example of our proffesional work. You can view professional samples here.

Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not reflect the views of BusinessTeacher.org. Information in this assignment should not be used to form the basis for any kind of financial or investment advice, as the content may contain inaccuracies or be out of date..

Predicting Rating Changes with Market Quotes

3686 words (15 pages) Business Assignment

5th Oct 2020 Business Assignment Reference this

Tags: Business AssignmentsFinance

1. Introduction

The introduction of agency ratings for the companies and securities by independent institutions (credit rating agencies) using simple symbols – letters and/or numbers as a measure of probability of expected loss has facilitated the process of investment decision making. These simple symbols can help investor to evaluate several investment opportunities based on their credit risk. Moreover, regulators also use the credit ratings for their requirements in the investment decision and capital allocations (for instance, requirement of Basel II for banks).

However, one of the main drawbacks of the credit ratings is “stickiness”, because rating agencies face tradeoff between stability and accuracy. Due to the popularity of credit ratings, any changes may have significant consequence on the price of the securities issued by that firm. As a result, rating agencies will downgrade the credit rating only in the case when they are assured that the creditworthiness of the company will not improve back after a short period of time. (Cantor and Mann, 2003).

Moreover, during financial crisis, agency ratings were highly criticized for their ratings and not being able to predict the negative consequences on time. Therefore, as investors need more timely measurement of the probability of default, the idea of predicting agency ratings by market quotes become more popular.  Furthermore, it is supported by many empirical studies: there is a strong evidence that market reacts to the changes in the credit risk of the issuers faster than agencies (Hull et al., 2004), hence, they can be used to anticipate future credit rating events.

As it was pointed out earlier, the ability to predict credit ratings by publicly available information (without waiting for and relying on official published information, as in the case of balance sheet data), will be helpful for several reasons. Firstly, investors at the hedge funds, pension funds or banks will be able to monitor the companies, the securities of which are included in their investment portfolio. If there is a requirement for credit ratings on the composition of the portfolio (for instance, only investment grade securities), by predicting the agency rating beforehand (especially downward case), they can have more time to reconsider their investment strategies and make more rational decisions.

This coursework aims to present one of the methods for predicting credit rating with introducing market implied ratings. Market implied ratings will be based on bond spreads. The coursework is organized in the following way. Section 2 is devoted to literature review, where several methods of applying market quotes are discussed. Section 3 presents methodology and Section 4 describes data set. Finally, Section 5 discusses the expected results.

2. Literature review

Various papers in the academic literature have been devoted to studying the methods of predicting credit rating changes by market quotes.

One of the models that is used for predicting changes in the credit ratings is implied default probabilities model. These are models based on Merton’s framework (1974), where the main idea was to view equity of the firm as a call option with debt as an exercise price. From equity and its volatility, the firm’s asset values are estimated and with debt it is further used to obtain distance to default values. These values are then mapped to probability of default and rating score. These models use both accounting data and market data (share prices).
The type of models is known as structural models for default probabilities.  

Delianedis and Geske (2003) found out that risk neutral probabilities estimated from Merton’s model can anticipate credit rating changes for upgrades, downgrades and default months before the actual event.

The most well-known model based on default probabilities is Moody’s KMV model, they estimated expected default frequencies (EDF) on stock prices that further matched with historical default frequencies of agency ratings to make equity-based implied ratings (Korablev and Dwyer, 2007). They found out that EDF leads credit ratings in predicting defaults. They also indicated that EDF-implied ratings for European firms were similar to CDS-implied ratings. 

Similar analysis was performed by Liu et al. (2007), who presented in their paper Fitch Equity Implied Rating and Probability of Default model. The Probability Default model could anticipate agency rating 74% cases one month before the announcement date.

There are also some papers, which used regressions as the main method in attempting to predict future credit ratings. For instance, Di Cesare (2006) analyzed whether CDS, bond spreads and stock prices can predict upcoming change in the credit rating for 42 international large publicly listed banks using Probit model. He found out that CDS market was the most efficient indicator in anticipating future rating changes for negative event, stock prices for positive events. Bond market performed as the least efficient indicator for negative events, but it improved for positive events. He supposed that it was related to liquidity issue of bond market: when date of changing rating is near, the bonds liquidity increases, which leads to decrease in liquidity premium. In case of downgrade, it is offset by increase in premium for expected default. For upgrade they move in the same direction, hence, anticipation power for positive event by bond market is stronger.  

Standard and Poor’s presented their Market Derived Signals model, where they estimated linear regression model in predicting credit ratings (Bergman et al., 2009). The main parameters in the model were logarithm of CDS spreads, CDS document type (No Restructuring, Full Restructuring, Modified Restructuring, Modified Modified Restructuring), S&P long-term issuer credit rating, its CreditWatch/Outlook status, Global Industry Classification Standard sector and currency denomination. Hence, they estimated piecewise linear function with log spreads and numerical scores. While predicting the credit rating the value of dependent variable was rounded to nearest integer, which represented value of Market Derived Signals.

However, most of the studies based on regressions were inflexible, and out-of-sample acted relatively poor (Reyngold et al., 2007).

Another way of using market quotes in predicting credit rating is, transforming raw market data into market implied rating using mapping function and compare it to the real credit rating. This idea was first introduced by Breger et al. (2003): they tried to find the simplest way of classifying bonds into implied ratings, which would further be used to anticipate credit rating of the companies. They estimated the range of spreads for each implied rating by applying penalty function, which minimizes the distance between misclassified bond spread and its boundary. However, they studied prediction power of implied ratings on for individual cases.

This method became popular and was the base for further market implied rating models, because it is the most direct and free from assumptions model. 

Munves et al. (2007) described bond and CDS based implied ratings. Unlike Breger et al (2003), they estimated the boundaries between market implied ratings, by calculating geometric mean of the two adjacent ratings’ median spreads. However, with CDS data sometimes median spreads were inverted, for instance, median spread for Aa3 was higher than median spread of A1, in this case they interpolated the median spread curve from two neighbouring ratings.

Following the methods described in paper of Breger et al. (2003), Kou and Varotto (2008) carried out their studies on large sample of eurobonds for 11 years with two major agency ratings Moody’s and S&P. They found no significant difference in responsiveness to market indicators between two agency ratings. They came to conclusion that bond implied rating can anticipate credit ratings up to six months before announcement date for both negative and positive events.

The introduction of CDS market, highly attracted many researchers in studying the market-based indicators based on CDS prices, because they purely represent credit risk and they are highly liquid.

Reyngold et al. (2007) first applied the penalty function technique for estimating implied rating borders on CDS quotes. They improved Kou and Varotto (2008) penalty function by using sum of the squared differences to make it differentiable and normalized the penalty function by dividing it to number of observations at particular credit rating, so that boundaries would not cross unintuitively. They used smoothing techniques on CDS spreads to separate short term volatilities – “noises” in spreads from long term “signals” (exponentially weighted moving average with time window 365 days). This technique helps to overcome the problem of inverted spreads for market implied rating that was described in research work by Munves et al. (2007).

However, due to quadratic penalty function, the outliers are heavily penalized, hence, Castellano and Giacometti (2012) slightly modified the function making it more computationally efficient linear programming. They compared ratings from 3 different credit rating agencies in Europe and USA for two sub-periods 2004-2006 and 2007-2009. They found that during pre-crisis period, the difference between credit rating and implied rating were around two notches, during financial turmoil it became wider. They also found out that implied rating model predicted credit rating better for US market than for European market. They concluded that Fitch reacted to changes faster compared to Moody’s and S&P in US market, while S&P in European market.

Overall, most of the studies showed that either agency followed market (implied rating led) or not (divergence), but there were no signs that market followed agency (agency rating led) (Reyngold et al., 2007, Kou and Varotto, 2008, Liu et al., 2007).

Some of the results of empirical studies indicated that implied ratings anticipated agency ratings. Hull et al. (2004) investigated the relationship between CDS spreads and agency ratings and reported that only downgrades can be predicted by CDS spreads, no significant result for anticipating upgrade in agency rating. Norden and Weber (2004) also found that equity market and CDS market can anticipate only downgrades. 

On the other hand, many recent papers found that market quotes could predict future rating event for both sides. Kiesel and Spohnholtz (2017) found out that CDS market implied ratings anticipated credit ratings for both upgrades and downgrades up to 16 weeks before rating announcement date. The research performed by FitchSolutions indicated that CDS implied rating and equity implied ratings predicted more than 50% and 63% of the rating changes up to three months and 64% and 73% for one month before event respectively (Reyngold et al., 2007, Liu et al., 2007).

Although, there were many studies carried on anticipating agency credit ratings by market quotes, most of them were based on CDS spreads, because CDS is the most liquid instrument and represent pure price of credit risk. However, there are still some advantages of considering bond spread implied rating models. One of the main benefits is that dataset comprised from bond spread is larger, because there are more historical data for bonds prices available. Moreover, as bonds have their own ratings, the model allows to estimate both bond level implied rating and issuer level implied rating. Therefore, in this research, the focus area will be coming up with the model that helps to predict accurately credit ratings with bond spreads.

3. Methodology

Bond implied ratings will be used as a main model for predicting agency ratings for the firms. Following the models discussed in the literature review, the following steps are taken: estimating boundaries of market implied ratings and classifying bonds according to their spreads, performing Hit and Miss matrix analysis to compare agency ratings to implied ratings, carrying out lead-lag analysis to find out whether bond market anticipates agency ratings or the other way round.

3.1. Mapping process

Ideally, as bond spreads reflects credit risk of the entity, the bonds with the same credit rating need to have the similar spreads. With this idea, bond spreads can be divided into hypothetical rating groups: all bonds with the same credit rating need to be placed into interval of spreads with some upper and lower bound spread values. However, there will be some outliers, which falls into another rating category. Therefore, main idea of the market implied rating model is to find upper and lower boundaries, that minimizes average distance between each outlier and its boundary. For this reason, the following penalty function can be used, which minimizes the border spread between bond rated A and BBB:


where, m – number of bonds rated A, n – number of bonds rated BBB.

Applying it to all rates and bonds up to time T:


subject to:

bk-1+=bk-,   ∀k>1

where, sj,k,t – bond j’s spread of credit rating k at time t,
bk+ and bk- – upper and lower boundaries.

Ratings are converted into numerical values from 1 to M, where 1 is rating with lowest risk. The constraint assures that lower boundary of the current rating is equal to the higher boundary of the previous implied rating.

Following Castellano and Giacometti (2012), the solving below linear programming model will lead to the same result: minb+, b-⁡∑k=1M∑j=1N∑t=1T1Nk⁡(zj,k,t+wj,k,t)

subject to:bk-1+=bk-,   ∀k>1





Identifying boundaries of the market implied ratings will allow  classifying bond to market implied ratings by their spreads. The borders will be re-estimated daily to account for changes in the market (for instance, risk aversion of investors), but for each day the bond spreads of past 30 days will be used to smooth out volatility. 

3.2. Hit and Miss Matrix

The agency credit rating and market implied rating would be identical, if they both reacted to changes with the same timing. For that purpose, we can perform Hit and Miss matrix analysis, where diagonal matrix values would show perfect match, off diagonal ones include reclassified number of bonds. The results for different periods may be compared to see whether the downturns in the economy affected the differences between market implied and credit agency ratings.

3.3. Forward analysis 

To further analyze off-diagonal values of Hit and Miss matrix, forward analysis will be carried out. For each downgrade or upgrade event, there will be some time interval before and after the event. If market implied rating decreased (increased) before credit rating downgrade (upgrade), implied rating led agency rating. On the other hand, if implied rating declined (rose) after agency rating downgrade (upgrade), agency rating led implied rating. However, if these changes were in different directions, two cases would be observed: convergence (ratings moved towards each other) or divergence (they moved in different way):

Figure 1. Possible outcomes of Forward analysis (Reyngold et al., 2007).

 Case IR > AR:




IR lead


IR lead


AR lead


Case IR < AR:




AR lead




IR lead

IR lead

4. Data

The data set includes bond prices, bond’s cash flow structure and agency ratings for issuers and corporate bonds. Bond prices, issuer’s and bond’s agency ratings are downloaded from Thompson Reuters Eikon. Bonds’ cash flow structure is obtained from Bloomberg.

So far, the data set includes prices and all available agency ratings (issuer and/or bond level) for 1288 bonds of 272 Russian firms over the period from January 2009 to March 2019.

4.1. Data filtering

Firstly, with price and cash flow structure of the bonds, yield to maturity (YTM) for all the bond in the data set needs to be calculated. Following Kou and Varotto (2008) approach, synthetic government bond will be priced with the government zero-coupon rates with maturity and cash flow structure corresponding to each bond in the sample. For Russian market, these rates are downloaded from cbr.ru website. Next, YTM is estimated for each synthetic government bond. Finally, bonds spreads are calculated by taking the difference between corporate bond’s YTM and corresponding government bond’s YTM. 

As time to maturity approaches, the bonds become less liquid, hence bonds with time to maturity less than one year are eliminated from sample. Moreover, we will drop from data set bonds with negative spreads.

5. Expected results

The outcome of the project is the model, which will be able to predict future changes of agency ratings with bond quotes. Based on the literature review presented above, it can be noticed that market quotes are highly informative in timely assessing the probability of expected loss. Therefore, due to stickiness of the credit ratings, market prices could predict the upcoming credit rating event.

Clearly, in Hit and Miss matrix, we did not expect 100% perfect match, otherwise it would not be possible predict credit ratings by market implied ratings. However, 85-90% of the difference between ratings should lie within two notches.

At the end, the model is expected to predict the credit rating changes both for upgrades and downgrades with bond spreads as accurate as CDS based models discussed above, and additionally it could be applied to Russian bond market.


  1. Bergman, S., Hampel, M., Rome, J., Shi, I., Raralli, L. and Yang, X. (2009). How Standard and Poor’s arrives at Market Derived Signals. Standard and Poor’s.
  2. Breger, L., Goldberg, L. and Cheyette, O. (2003). Market Implied Ratings. Risk Magazine.
  3. Cantor, M. and Mann, C. (2003). Measuring the Performance of Corporate Bond Ratings. Moody's Investors Service.
  4. Castellano, R. and Giacometti, R. (2012). Credit default swaps: implied ratings versus official ones. A Quarterly Journal of Operations Research, 10(2), pp. 163-180.
  5. Delianedis, G. and Geske, R. (2003). Credit Risk and Risk Neutral Default Probabilities: Information About Rating Migrations and Defaults. EFA 2003 Annual Conference Paper No. 962.
  6. Di Cesare, A. (2006). Do Market-Based Indicators Anticipate Rating Agencies? Evidence for International Banks. Economic Notes, 35(1), pp. 121-50.
  7. Hull, J., Predescu, M. and White, A. (2004). The relationship between credit default swap spreads, bond yields, and credit rating announcements. Journal of Banking and Finance, 28(11), pp. 2789–2811.
  8. Kiesel, F. and Spohnholtz, J. (2017). CDS spreads as an independent measure of credit risk", The Journal of Risk Finance, 18(2), pp.122-144.
  9. Korablev, I. and Dwyer, D. (2007). Power and level validation of Moody’s KMV EDF™ credit measures in North America, Europe, and Asia.  Moody’s KMV.
  10. Kou, J. and Varotto, S. (2008). Timeliness of spread implied ratings. European Financial Management, 14(3), pp. 503-527.
  11. Liu, B., Kocagil, A. and Gupton, G. (2007). Fitch Equity Implied Rating and Probability of Default model. Quantitative Research Special Report, Fitch, Inc.
  12. Merton, R. (1974). On the Pricing of Corporate Debt: The Risk Structure of Interest Rates. Journal of Finance, 29(2), pp. 449-470.
  13. Munves, D., Hamilton, D, Mann, C., Wooley, M., Dwyer, D., Gibbon, J., Qu, S., Fingerman, E. and King, J. (2007). Moody's Market Implied Ratings Description, Methodology, and Analytical Applications. Moody’s Investors Service.
  14. Norden, L. and Weber, M. (2004). Informational efficiency of credit default swap and stock markets: The impact of credit rating announcements. Journal of Banking & Finance, 28(11), pp. 2813-2843.
  15. Reyngold, A., Kocagil, A. and Gupton, G. (2007). Fitch CDS Implied Ratings (CD-IR) Model. Quantitative Research Special Report, Fitch, Inc.

Cite This Work

To export a reference to this article please select a referencing stye below:

Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.

Related Services

View all

DMCA / Removal Request

If you are the original writer of this assignment and no longer wish to have your work published on the UKDiss.com website then please: