Ratings matter: announcements in times of crisis and the dynamics of stock markets

In this paper we propose a novel approach in analysing the impact of changes in sovereign credit ratings on stock markets. We study the evolution of a segmented form of the stock market index for several crisis-hit countries, including both European and Asian markets. Such evolution is modelled by a homogeneous Markov chain, where the transition probabilities from one starting level of the index to a new (lower or higher) level in the next period depend on some explanatory variables, namely the country’s rating, GDP and interest rate, through a generalised ordered probit model. The credit ratings turn out to be determinant in the dynamics of the stock markets for all three European countries considered Portugal, Spain and Greece, while not all considered Asian countries show evidence of correlation of market indices with the ratings.


Introduction
Thomas Friedman acknowledges in 1995 from the columns of the New York Times that "In the 1990's the most important visitor a developing country can have is from Moody's Investors Service Inc.". He continues warning that "Moody's is the credit rating agency that signals the electronic herd of global investors where to plunk down their money, by telling them which countries' bonds are bluechip and which are junk". He concludes with the famous often cited statement "That makes Moody's one powerful agency. In fact, you could almost say that we live again in a two-superpower world.
There is the U.S. and there is Moody's. The U.S. can destroy a country by leveling it with bombs; Moody's can destroy a country by downgrading its bonds." More recently, since the sub-prime crisis and the bankruptcy of Lehman Brothers on the 15 th of September 2008, the role of these agencies has come to the spotlight in the debates of politicians, economists and the general public (Sangiorgi and Spatt, 2017).
Credit rating agencies play an important role in reducing information asymmetries in capital markets, providing significant information to (institutional) investors to evaluate risks. This makes them fundamental players in a globalised financial world (Levich et al. 2012), and the economic consequences of their activity should be analysed. The increasing number of private and public issuers and debt products, and the growing requirements applied to investors, banks and other financial institutions about risks, portfolios' compositions and capital adequacy, explain the growing importance of their role (Al-Sakka and Gwilym, 2010).
Rating agencies gained a particular negative reputation following downgrades of crisis-hit countries in Europe, becoming the synonym of austerity measures imposed in order to balance public budgets.
All this social cost has been undertaken to regain the confidence of the markets in the ability to pay public debt. This confidence was, and still is, summarized by a rating of a few letters assigned by the main rating agencies -Standard & Poor's (S&P), Moody's and Fitch -that determine whether a country is trustworthy or not. The credit ratings expressed by these companies have become so entrenched and common in the public knowledge that even the way we talk and think about credit risk has absorbed the rating scale of the famous 'AAA' and the infamous 'junk'.
Rating agencies were initially seen as private companies with a decentralized opinion, immune to governmental interference, and acting in a competitive market where their credibility was their strongest asset (White, 2010). As a matter of fact, by the end of the 90's the market was dominated by the 'Big Two': S&P and Moody's represented 90% of the market share (Bruner and Abdelal, 2005), with Fitch as a distant third. Moody's has been publishing ratings since 1909 and S&P since 1923; according to Mora (2006), their long life and the fact that investors pay to subscribe to their reports implies that their activity and function -providing an estimation of default probability of borrowers -must be of some value. In recent years, the importance of rating agencies increased due to the globalization of financial markets and the widespread attention to their ratings in banking and financial regulation. Consequently, the debate about the behavior of rating agencies -focused on issues like lack of competition and transparency, or the conflict of interests with issuers -and their role has gained increasing interest (Stolper, 2009;Mathis et al., 2009;Hunt, 2009 andOpp et al., 2013;Fuchs and Gehring, 2017).
The greatest impact of these agencies is, by far, the way they influence the access to institutional investors all over the world, acting as de facto gatekeepers to huge sums of capital. Indeed, credit ratings have been crucial in assigning default probabilities to debt issuers and, consequently, the access of governments to capital inflows. As Ferri et al. (1999) outline: "(...) besides affecting the cost at which issuers can borrow, ratings determine the extent of potential investors. Specifically, regulations and statutes either forbid institutional investors to invest in assets carrying ratings below a certain level or they require extra capital to be posted. These investments are referred to as 'belowinvestment-grade' or 'speculative assets'." Bruner and Abdelal (2005) provide an interesting discussion about the new status enjoyed by these agencies, the power that derives from it and the quasi-symbiotic relationship established between the public sphere of governments and these private companies.
One of the motivations for this work is the fact that the credit rating of a sovereign country heavily influences private-sector issuers of debt, what Bruner and Abdelal (2005) call the 'sovereign ceiling'.
Since only rare exceptions of private companies' foreign currency debt is issued with a higher credit rating than their sovereign's (Borensztein et al., 2013), the rating assigned to sovereigns affects all the other debt issues in the world. In fact, a sovereign debt rating downgrade is followed by a rising in the cost of debt for private issuers, with consequences both in financial and real markets (Almeida et al., 2017;Bedendo and Colla, 2015).
In the present study we explore data on the European and Asian crises to investigate whether there is evidence of an impact of the changes in sovereign credit ratings on the stock market indices of several countries, and in particular whether and how the credit ratings affect the probabilities of the indices increasing or decreasing in subsequent periods, as modelled by an homogeneous Markov chain.
The next section presents some background on research in this field, while Section 3 proposes a model for assessing the impact of ratings changes on stock market dynamics. Section 4 presents the application to the European and Asian data, while the last section concludes.

Determinants of credit ratings
The sovereign credit rating literature is extensive. Its start can be traced back to the work of Cantor and Packer (1996) who describe and quantify the determinants and impact of sovereign credit ratings given by the major U.S. based rating agencies, Moody's and S&P. They find that the ordering of risks that credit ratings imply is broadly consistent with specific macroeconomic fundamentals, particularly per capita income, GDP growth, inflation, external debt, level of economic development and default history.
Later studies examine the quantitative determinants of S&P and Moody's sovereign ratings before and during the global financial crisis, finding evidence that the most relevant economic variables are the Gross National Product (GNP) per capita and inflation (Bissoondoyal-Bheenick, 2005); GDP (per capita and growth rate), inflation and unemployment rate, government and external debt, fiscal balance 3 and political and historical factors (Afonso et al., 2009); and finally GDP growth and volatility, GNI, Import/Exports, Inflation, Rule of Law and Government Debt (Basu et al., 2013).

Performance of credit rating agencies
In recent years the role of credit rating agencies in international markets has been the subject of deep analysis, debate and also criticism, and a large body of work assessing their performance has been published. In fact, the agencies' role in reducing information asymmetries among financial agents -by condensing deep debt issuer's analysis into one grade that provides a global risk evaluation -is crucial in the current complex financial world. Ferri et al. (1999), in line with Radelet and Sachs (1998), demonstrate that the pro-cyclical nature of rating agencies' sovereign rating may have contributed to aggravate the East Asian financial crisis. However, later Mora (2006) contradicts the previous findings, concluding that rating agencies did not aggravate the Asian crisis. Altman and Rijken (2004) address the investor's perception of rating agencies being very slow to adjust their ratings. An explanation of this perception is the "through-the-cycle" rating methodology that agencies apply in their assessments, while investors have a "point-in-time" perception of creditworthiness. The authors conclude that rating agencies' methodology suppresses the sensitivity of the ratings to short-term fluctuations in credit quality.
Several authors investigate the agencies' influence and the independence of their evaluations. Bruner and Abdelal (2005) focus on the relationship between sovereign governments and the rating agencies' authority in the bond markets, discussing possible alternatives in order to improve their accountability, given the immense power they have gathered, especially with institutional investors.
Tennant and Tracey (2013) find a negative bias in S&P assignment of rating upgrades and downgrades to low and lower-middle income countries, while Fuchs and Gehring (2017) find evidence of home bias in the production of the sovereign ratings, as a result of political and cultural factors.
Finally, Sangiorgi and Spatt (2017) analyse from an economic and regulatory perspective several limitations of credit rating agencies as risk evaluators, comparing with other gatekeeper providers, as the external auditors. In spite of concluding that ratings play an important role as information provider for capital markets players, the authors identify several weaknesses associated with the conflict of interest by paying for information.

Impact of rating changes on the economy and the financial markets
Various scholars have investigated the relationship between changes in credit rating quality and economic and financial national indicators, using different datasets and methodologies. Kräussl (2005) confirms that changes in sovereign rating statistically influence size and volatility of emerging capital markets, but results depend on the type of changes (downgrade/upgrade), while Hooper et al. (2008) examine the impact of rating changes on stock and foreign exchange market returns and volatility, for 42 countries during the period 1995-2003. They conclude that the overall impact is significant, but asymmetric, with amplified effects for downgrades, in emerging markets and during periods of crisis. This conclusion is also confirmed by Brooks et al. (2004), where the effects on stock indices are significant when a downgrade occurs. In an event-study setting, they also find that emerging markets are not sensitive to rating changes. Kim and Wu (2008) come to the conclusion that long-term ratings are most important for the development of financial sectors in emerging economies. They remark that foreign currency long-term ratings provide the most important contribution for international capital inflows and, consequently, domestic financial market development. This argument is supported by observing that the three forms of capital inflows studied (foreign direct investment, international banking flows and portfolio flows) significantly increased as foreign currency long-term ratings of emerging market sovereigns improved.
Other authors investigate the response of yield spreads of credit default swaps (CDS) to rating announcements, both in emerging markets (Ismailescu and Kazemi, 2010) and in developed countries (Afonso et al., 2012), while De Santis (2014) analyses the developments of euro area sovereign yield spreads, all finding that there is a significant response to changes in both the credit rating notations and in the outlook, with strong spillover effects. More recently, Lee et al. (2018), using a large sample of S&P rating notch and watch changes in the U.S. market, find that the CDS price has predicting power for negative events (downgrading) and contains specific information that contributes to price discovery.
Al-Sakka and Gwilym (2013) examine how the foreign exchange markets reacted to sovereign credit events prior to (2000)(2001)(2002)(2003)(2004)(2005)(2006) and during the crisis (2006)(2007)(2008)(2009)(2010). Their results support the view that rating agencies' signals do affect the own-country exchange rate. They also identify strong spillover effects to other countries' exchange rates in the region. The authors also mention that market reactions and spillovers are far stronger during the financial crisis period than pre-crisis.  Kaminsky and Schmukler (1999) study the effects of news, among which rating announcements, on extreme markets movements. Based on data from nine Asian countries, their analysis concentrates on the twenty largest market changes (jitters) registered in each country between the beginning of 1997 and mid 1998, finding evidence of large spillover and contagion effects, as well as herding behaviour of the investors. Similar results are found in Baig and Goldfajn (1999), based on five Asian countries, reenforcing patterns of contagion during the crisis, as well as large cross-country correlations in the currency and equity markets, even after controlling for own-country news and other fundamentals.
Although the present work can be categorised in this same area of research -since it aims to measure the effect of rating changes on the stock markets -we develop a different approach from what was previously studied, as we explain in the following section.

A model for stock markets dynamics
The objective of this paper is to study whether the evolution of the sovereign credit ratings effectively acts upon the dynamics of stock markets, and if so, to quantify such influence , after controlling for other relevant variables.
The natural high volatility of the market indices does not match the substantially lower frequency of rating change announcements. Therefore, in order to capture the actual (possible) impact of the announcements on the market, we propose to focus not on the actual value that the stock market index is attaining in each period of observation, but rather on the qualitative performance of the index, represented by a certain number of classes or levels (ordered from 'very low' to 'very high').
The low frequency of the announcement points to the use of a discrete-time setting. The segmented form of the stock index allows to represent its dynamics through a discrete-state process. In order to model the transitions of the stock index levels across different classes over time, a possible choice is therefore a discrete-time Markov chain.
The Markov structure implies that the level of the index in the next time period only depends on its current level, and not on the levels in the previous periods, considering that all needed information about the market is included in the current value of the index. This is a fairly common assumption in the modelling of financial series, which reflects the so-called weak form of market efficiency, where past patterns of prices have no forecasting value (see for instance Dixit and Pindyck, 1994). As a matter of fact, among the stochastic processes most widely used to model financial time series, many benefit from the Markov property, like for instance the random walk and the Brownian motion (see for example Taylor and Karlin, 2014). A further discussion of the Markovian properties of financial time series can be found in Chen and Hong (2012).

6
The calculation of a so-called transition matrix allows to obtain, at each point in time, the probability of the stock market index moving to an upper or lower level (or remaining at the same level) in the following period, conditional on the current position and some other relevant explanatory variables.
Given the natural ordering of the index levels, the probabilities of transition can be computed according to an ordered qualitative dependent variable models, and then used to reconstruct the corresponding discrete-time transition matrix. For the modelling of stock markets dynamics, in the present work we follow an homogeneous specification of the Markov chain, which corresponds to assuming that the dynamics of the stock market index in different periods during the debt crisis is comparable, specifically the probability of the index jumping from one level to the other remains the same, whether we are at the beginning, the middle or at the end of the crisis period. This hypothesis is rather widespread, as mentioned earlier, and can easily be tested in a cursory way, for instance, by introducing a series of dummy variables for a certain number of time frames (e.g. beginning, middle, end of crisis), and testing their significance.
In case the time effects turn out to be significant -and according to the objectives of the study -more The model for market dynamics can be derived from a latent variable model, where the latent variable y * represents the unobservable underlying "quality/reliability" of the market -or the confidence of the investors in the market. The observed value of the level of the market index, say y, is considered to be a "signal" of the unobserved "quality" y * . 7 determined by a set of explanatory variables x as y * = xβ + e, where the error is assumed to follow a standard Normal distribution, conditional on the set of regressors x. The market index level y is linked to the underlying "quality" of the market y * as follows: The α j are a set of unknown ordered cut-off points, where α j−1 < α j , α 0 = −∞ and α J = +∞.
The conditional distribution of y given x can be derived from the standard normal assumption for e, obtaining the following response probabilities 2 : From the parameter estimates, the above probabilities can be obtained, together with the estimated partial effects of the regressors on each of the J probabilities. Both the probabilities and the partial effects need to be evaluated at some sensible values for the regressors, for example at the mean value of each variable, or at some specific value of interest.
The standard OP model can be generalised, allowing the effects of the regressors to vary with the levels, i.e. specifying the response probabilities as: The following interest rates were used: for all European countries the 6-month EURIBOR rate 5 , while for South Korea and Indonesia the OECD country-specific 3-Month Interbank Rates 6 . The time series of the EURIBOR Interest Rate is shown in Figure 2, while the country-specific Interbank Rates for Asia are reported in Figure 3.

Analysis of variables
The chosen weekly frequency of the index values reflects the objective of measuring the possible effect of rating changes on market dynamics. As noticed earlier, given that the rating changes present a rather low frequency (see Figure 1), it seems more likely that ratings affect the medium term level of the stock market indices, rather than the much more volatile daily values. Therefore, the weekly or monthly changes of the stock indices seem to be better candidates for the present study. Given that monthly changes would give rise to very few observations in the period of interest, and most likely would not capture the actual dynamics of the market at a time of great instability, we chose weekly observations.
The original series of stock exchange indices were studied before proceeding to their discretisation. Some summary descriptive statistics about the indices are reported in Table 3 Table   2, for the sake of the example. All transitions lie on the three main diagonals; going to a finer definition of the levels, for instance taking J = 10, would exacerbate even further the sparseness of the matrix, and is therefore not a suitable choice for the present case. Table 3 reports some statistics about the classified series, showing very similar patterns in terms of transitions in all countries.
Finally, the possible correlation of the ratings with time trends is investigated. At the bottom of  Table 4 reports the estimation results of the ordered probit models for the five countries under consideration, where the response variable is the ordered level of the stock market index. The regressors common to all models are the country's credit rating, the GDP and the previous level of the index.

Model specification and empirical results
In the case of the European countries, the EURIBOR 6-month interest rate is also considered, while country-specific 3-month interbank rates are used for South Korea and Indonesia. For what concerns the specification of the ordered probit, heteroskedasticity-robust standard errors have been used. The potential endogeneity of the Rating variable has been tested through joint estimation of the main ordered probit regression and a reduced form linear regression for the Ratings, where a time effect acted as instrument (see Roodman, 2011). In all countries the time effect is the Year, but in the case of Greece, due to the shorter time span of the two sub-periods of analysis, a Quarter and a Semester effect have been used respectively in Period I and II. The endogeneity test has been performed by testing the significance of the correlation of the errors of the two equations, which in all cases resulted non significant, confirming the exogeneity of the Ratings. While Table 4 reports only the p-value of the test, the full instrumental variable estimation results are reported in Appendix, in Tables 11 and 12.
Whenever regressors were individually not significant at 10%, their joint significance has been tested through exclusion restrictions, allowing to reach a reduced model including only significant variables.
In all tests the restrictions were considered valid (see again Table 4). The parallel lines assumption (PLA) of the ordered probit has been tested in the reduced model through an approximate likelihoodratio test of equality of coefficients across response categories. Results, at the bottom of the Table, show that models for most countries do not violate the hypothesis, however in the case of Greece (period II) and South Korea the Rating variable needs to be modelled allowing for level-varying effects.
After this first exploratory model, a final specification was obtained reducing the set of regressors only to those that were significant, and using a generalised ordered probit (GOP) relaxing the PLA for the Ratings in the cases of Greece period II and South Korea. The estimation results of the final model are reported in Table 5. Notice that in the case of the GOP some of the level-dependent coefficients turned out to be not significant, and were subsequently constrained to be equal to zero when using the model for prediction. The imposed constraints were tested via likelihood-ratio tests, which confirmed the correctness of the restrictions.
As an overall picture, looking at Table 5 companies to have access to the international capital markets, and thus being less exposed to the rating changes of the home country. Stulz (1995) argues that globalization of firms reduces the cost of capital. Kaminsky et. al. (2002) find that the effect of upgrades and downgrades is present also for emerging markets, but it appears not to be large in economic terms.
When a sovereign downgrade happens, rating agencies used to apply the"sovereign ceiling policy": the credit rating of the firm is set at the level or below the sovereign rating of the domicile country.
This practice has been recently revised, and for S&P firms can have a rating up to four notches above the sovereign rating. 7 For all remaining countries, the effects of the Ratings are highly significant; in the case of levelvarying effects, we observe for Greece a much larger value for the last threshold, i.e. the ratings influence even more the capacity of the index to reach the top class. 13 of remaining at the same index level taking values roughly between 70% and 85%. The Asian countries show a similar slow dynamics at the mean value of the regressors. In the case of Greece, however, the resilience of the index seems to be weaker, showing very high probabilities of falling to lower levels whenever the index is in the top classes. Rather than persisting at the top levels, the index shows probabilities between about 50% and 75% of falling back one class, and almost zero chances to improve the level.
To further explore the effects of the Ratings, the transition matrices are also evaluated at different percentiles of the variable, while keeping the GDP and Interest Rates at their mean values. For the sake of the example, Table 9 shows the probabilities evaluated at the 10 th and the 90 th percentiles of A summary representation of the changes in the transition probabilities when a country moves from a lower to a higher sovereign rating is given in Figure 4 for all considered countries (except Indonesia, where the variable was not significant). The graphs display the probabilities that the stock market index shows an increase or a decrease in the following period, for specific values of the Ratings.
The probabilities are computed at a selection of percentiles of the rating distribution for each country (reported in Table 10), all remaining regressors being evaluated at their mean.
The results are rather striking, showing that when the countries are rated higher the respective stock market index has a much higher probability of increasing, reducing at the same time the chances to fall to lower levels. For the second period of Greece, only when the B−/B threshold is passed we observe some changes in the resilience of the index; below such level the probability of improving is extremely low, while the index falls to lower levels with probabilities between 40% and 60%; however, soon after the threshold, the probability of improvement of the index increases steeply, and the chances of falling back are reduced to close to zero.

Conclusions
In this research we examine the impact of sovereign debt rating changes on security prices in stock exchanges in five countries, proposing a new empirical approach based on Markov Chains. This 14 work is important due to the relevance of credit ratings agencies as a risk assessment provider for portfolio managers, for bankers and regulators (in the context of Basel Committee recommendations and ECB supervision) and also for governments and policy makers. In addition, the importance of institutional investors in capital markets, its portfolio composition restrictions on securities that have an investment-grade rating, and the impact of stock-market behavior on the cost of funding, both for companies and governments, explain why the impact of sovereign debt ratings on stock markets is particularly important.
In the end, we can conclude that European countries -Portugal, Greece and Spain -presented a similar behaviour with respect to the variable credit rating, with a correlation sign that was expected and that suggests that credit ratings are indeed relevant to the economic performance of the stock exchange.  Note: Transitions between the class of origin of the index in a given week (rows) and the respective class on the following week (columns). Classes are defined dividing the observed range of the index into 5 equally spaced intervals. Data between 01/01/2009 and 28/03/2014.

In person
All over the European Union there are hundreds of Europe Direct information centres. You can find the address of the centre nearest you at: https://europa.eu/european-union/contact_en

On the phone or by email
Europe Direct is a service that answers your questions about the European Union. You can contact this service: -by freephone: 00 800 6 7 8 9 10 11 (certain operators may charge for these calls), -at the following standard number: +32 22999696, or -by electronic mail via: https://europa.eu/european-union/contact_en

Online
Information about the European Union in all the official languages of the EU is available on the Europa website at: https://europa.eu/european-union/index_en

EU publications
You can download or order free and priced EU publications from EU Bookshop at: https://publications.europa.eu/en/publications. Multiple copies of free publications may be obtained by contacting Europe Direct or your local information centre (see https://europa.eu/europeanunion/contact_en).