Une analyse statistique de l’influence de facteurs macroéconomiques (croissance du PIB, taux de chômage…) et financiers (rendement des taux à dix ans, indices actions…) sur les changements de notes de cinq pays – États-Unis, Royaume-Uni, France, Italie et Grèce – conduit à conclure que ces facteurs ont certes une incidence sur les évolutions des notations, mais que celle-ci plus ou moins significative.
The relation among macroeconomic and financial fundamentals and the rating activity has been widely explored in the last 15 years.
The first time it has been demonstrated that rating transitions vary according to the business cycle has been in the 1990s, with several studies  examining rating transitions matrixes’ changes face to changes in the environment, to conclude that the hypothesis that ratings are sensitive to several external factors was validated. In 2000 Nickell, Perraudin and Varotto demonstrated that: (i) lower-rated issuers proved were significantly more volatile (ii) volatility falls sharply in business cycle peaks especially for investment grade issuers (iii) banks are much more volatile than industrial companies at any rating level and finally that (iv) investment grade companies (excluding Aaa-rated entities) experience the same probability of downgrade in normal times and troughs, whereas non-IG issuers are much more likely to be downgraded or default uniquely during troughs.
Several studies have been conducted ever since, all coming to the conclusion that rating transitions were not untouched by shifts in the environment. Even the consulting firm McKinsey proposes a model originally developed by Wilson, which rests on the assumption that default and migration probabilities are conditional to macroeconomic variables. McKinsey’s CreditPortfolioView® calculates the conditional probability of default from a multifactor model composed by various macro variables, adaptable for a number of countries and industries.
Despite the number of analysis conducted by both academics and professionals (even rating agencies themselves) the topic still has to be explored. Other than the existence of a relation between macroeconomic environment and ratings, findings from papers that have previously treated the topic are widely discordant, making it difficult to draw conclusion on the nature, timing and magnitude of this impact. Some authors argue that exogenous factors do not succeed in explaining rating transitions, while others find discordant signs to the coefficients of the independent variables chosen, meaning that the same macro factor has been demonstrated to have both positive and negative impact depending on the paper and statistical model used.
Methodology and data
The approach used by the majority of researches has been to focus on the three main rating transitions (default and transition from investment grade to high-yield and backwards), in most cases uniquely in the USA. Some of them developed their analysis in collaboration with one rating agency, therefore used data from that agency only.
This research renounces to a detailed dataset to be able to perform a more generalised analysis, involving a sample of European countries undergoing different economic situations (the UK, France, Italy and Greece) and the USA, to be able to benchmark results to the ones obtained in other studies. All rating transitions registered by S&P, Moody’s or Fitch across the last 20 years in the 5 Countries selected were gathered and analysed. The number of exogenous factors was kept low in order to assess their impact one by one, avoiding clusters of factors that would complicate the interpretation of results. The influence of the economic cycle is simplified though the nominal GDP Growth and the unemployment rate and the financial landscape is depicted using the return on 10-year Treasuries and a comprehensive Equity Index. The focus, unlike in most other papers, is on issuer rather than issue ratings, since the overall creditworthiness of a company over the long term is expected to be much more likely to follow economic cycles than a single obligation having its own, negotiable terms.
Rating transitions data take into account the whole universe of rated companies  and had to be adjusted to avoid double-counting or non-significant transitions that would have biased the results. The following transitions were excluded from the dataset: (i) all changes in outlook, (ii) most foreign issuer ratings, (iii) ratings withdrawn and newly rated companies and (iv) sovereigns, government regional and local ratings.
3 ratios were calculated for any year and Country :
being “i” the year index and “j” the Country index. The net number of rating changes has been simply calculated as (number of upgrades – number of downgrades) in a given year and Country.
The % net change in rating sometimes results as being >100%. This is mainly a result of the widespread practice to downgrade a company by only one notch at a time, which could result in having the same company downgraded 5 or 6 times in a row within few days or months, as it is the case for several firms in France in 2012, Italy from 2011 to 2013, and in Greece in 2011.
Concerning the independent variables, the historical evolution of the unemployment rate and the gross domestic product % yearly change in constant prices from IMF’s database  and the 10-year return of Treasury bonds from Bloomberg were utilised. The largest index available in any Country was chosen, to better represent the widespread recourse to debt capital markets by listed companies of all sizes. Selected Indexes were : S&P Composite 1500 for the US, UK FTSE All Share, CAC All Share for France, FTSE Italia All Share and the Greek Athens General Index.
The application of a Panel Model (Heiss 2016) instead of simple regressions considering only 2 factors at a time allowed the increase in the number of observations and therefore the statistical significance of the model. The model aims at explaining which are the determinants impacting ratings transitions the most among the macroeconomic and financial factors selected: nominal GDP Growth, the Unemployment Rate, the return on 10-year Treasuries and a comprehensive Equity Index. The limited sample in terms of number of Countries under analysis lead to choose a fixed effect approach rather than a random effects model.
Two models were fitted , with the aim of analysing both upgrades alone and the net % effect of net rating changes, calculated as (number of upgrades – number of downgrades)/total number of rated companies.
Both models can be expressed as : (see Formula 1)
The analysis was performed on 22 x 5 = 110 observations for each variable . The initial sample analysed was an unbalanced panel with n=5 and T=14-22, meaning that 5 Countries were analysed and data coverage was between 14 and 22 years.
The two regressions were performed both with and without Greece, for which the sample was too small to provide statistically significant data  and rating transitions were too biased by the net predominance of the banking sector. Financial institutions have been proved in several other papers to undergo very different rating cycles when compared to other sectors, which could have probably contributed to draw data from Greece further away from the other Countries’.
Nominal GDP Growth has a significant and strong positive effect, much stronger for rating downgrades. Unemployment has a strong negative correlation when Greece is excluded from the dataset, but appears significantly positive when Greece is included in the regression. The 10-year return on Treasuries has a strong and significant negative effect and the equity index does not result significant.
Nominal GDP Growth presents a positive, significant relation throughout all regressions. The effect is much stronger for downgrades, since the coefficient varies between 1.5 and 2.5 approximately for upgrades, while it jumps to 9 and 11 for upgrades and downgrades together. As an example, this implies that when GDP Growth increases by 1% ratings upgrades tend to increase by 2% on a Country average, while the overall ratings level improves by around 10%.
This finding is consistent with the conclusions drawn from most previous papers, which proved that macroeconomic factors tend to affect downgrades much more than upgrades: the impact of the economic cycle tends to be stronger on the downside rather than on the upside. As a means of example, during a boom period upgrades improve by 2%, whilst during a crisis the overall ratings level deteriorates by 10%.
The analysis confirms that GDP Growth has a significant positive effect on ratings, concluding that rating transitions follow the economic cycle.
The unemployment rate is the very reason why regressions were performed both with and without Greece. In fact the unemployment rate does impact ratings negatively, but this relation does not hold in Greece. The strong positive correlation that results when all 5 Countries are included in the analysis is an anomaly, given the conclusion just reached that ratings follow the economic cycle. If the relation depicted in the first and third regression was true it would mean that for an increase in the unemployment rate of 1% rating levels would improve approximately by 2.8%, which in theory does not make any sense.
What still has to be confirmed is whether the unemployment rate is a good indicator for the direction of the economy, able to depict the economic well-being of a Country. From a preliminary analysis conducted on the dataset it seems like it isn’t, since the unemployment rate tends to increase or keep a high level up until 2 or 3 years after the end of the recession as defined from the GDP Growth. Starting from the assumption that nominal GDP Growth is a good indicator to identify a Country’s economic cycle , linear regressions joining the nominal GDP Growth rate and the unemployment rate in the 5 Countries between 1980 and 2015 were performed. None of the regressions showed significant correlation or statistically relevant results, the R2 ranging from 0.00015 for France to 0.13826 for Greece. The unemployment rate still provides very useful information on the economic well-being of a Country, but the amount of time it takes to adjust after a boom or deep recession should be taken into consideration.
The existence of a negative relation between the unemployment rate and ratings is validated, but its statistical relevance is hard to capture given the delay in the adjustment of the unemployment rate to the economic cycle. Greece clearly represents an anomaly, notwithstanding the statistical relevance of results drawn from the regressions in which it is included (especially the first regression concerning upgrades only). The coefficient to retain in this case is therefore the -2.5 obtained from the second regression performed, concerning upgrades with the exclusion of Greece.
10-Years Treasuries’ Yield
10-year Treasuries show a significant negative relation, biased by the presence of Greece in the dataset that puts pressure on the statistical significance of the regression.
This relation is logical, since Treasury yields increase because the general creditworthiness of that Country deteriorated, and this affects directly the creditworthiness of companies domiciled in that Country. This effect on the downside is exactly what happened in Greece in 2011, when the number of downgrades peaked, as the negative coefficient of 3.8 in the “net rating changes” regression clearly depicts. Once again, the coefficient is much higher for downgrades and upgrades together than for upgrades alone, ranging from -3.2 and -8.3 in the two most significant regressions that exclude Greece.
The equity index’s correlation to rating transitions results being low and not statistically significant. This result could be justified by the fact that the analysis is not only based on defaults, but on all rating transitions. Analysing defaults only or focusing uniquely on a period of crisis, the relation would be expected as being statistically significant, since defaults or big deteriorations in the creditworthiness of issuers would have been more likely to be anticipated by stock performance. On the contrary, in a more comprehensive analysis including all rating transitions on a longer time frame that includes booms as well as recessions, this is apparently not the case.
The result still remains unpredicted, since some of the other researches argue that stock and credit markets are generally positively correlated, with a delay in credit market’s impact on the stock market. This implies that the relation to check for was the opposite, represented by the effect rating transitions have on stock indexes. This relation is based on the theoretical framework demonstrating that when bonds prices fall their yield automatically increases, therefore the company’s cost of capital increases and enterprise value falls, and consequently the stock price.
Both factors chosen to represent the macroeconomic environment, nominal GDP Growth and the unemployment rate, result being statistically significant and in line with theoretical beliefs. Out of the two financial outlook factors, only the 10-year Treasuries yield is statistically significant and negative, whilst the equity index’ p-value is way too high to consider it significant.
Overall statistical results are very satisfying, notwithstanding the fact that the R2 referring to the whole model is fairly low, ranging between 0.21556 and 0.32722 in the four models. In this case this means that the model is able to explain between 22% and 33% of ratings’ variability. Considered the purpose of the model, which is to assess whether macroeconomic and financial factors impact rating transitions and not to identify all endogenous and exogenous factors driving them, this result can be considered very satisfactory. In fact, most of the indicators driving rating transitions are intrinsic to the firm, and knowing that only 3 broad indicators explain from 22% to 33% of rating transitions means that these factors have an extraordinary impact.
Chargement du panier
Le meilleur de l’expertise métier et de la pédagogie