Available via license: CC BY 4.0
Content may be subject to copyright.
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
ISSN 1808-057X
DOI: 10.1590/1808-057x20241913.en
O A
This is a bilingual text. This article has also been translated into Portuguese and published under the DOI https://doi.org/10.1590/1808-
057x20241913.pt
Impact of the COVID-19 outbreak on credit ratings: Application
of the through-the-cycle approach
Dante Domingo Terreno1
https://orcid.org/0000-0003-4400-8058
Email: dante.terreno@unc.edu.ar
María Eugenia Donadille1
https://orcid.org/0000-0001-7683-7112
Email: mdonadille@unc.edu.ar
1 Universidad Nacional de Córdoba, Facultad de Ciencias Económicas, Departamento de Contabilidad y Ciencias Jurídicas, Córdoba, Argentina
Received on 04/17/2023 – Desk acceptance on 05/22/2023 – 2nd version approved on 09/11/2023
Editor-in-Chief: Andson Braga de Aguiar
Associate Editor: Andrea Maria Accioly Fonseca Minardi
ABSTRACT
e objective of this study was to analyze how the COVID-19 crisis has aected the determinants and predictability of
the domestic credit rating issued by Fitch Ratings in Argentina. Additionally, it aims to evaluate the eects of credit rating
agencies using the through-the-cycle method. Given the subjective nature of credit rating categorization, researchers have
developed models for explaining and predicting credit ratings. is subjectivity is signicant during economic events.
erefore, it is important to investigate whether the factors that determine and predict credit ratings remained consistent
before and during the COVID-19 crisis. is paper contributes signicantly to understanding how the application of the
through-the-cycle method aects the determinants and predictability of credit ratings in economic crises.eapplication of
the through-the-cycle method by credit rating agencies as a criterion during the COVID-19 crisis resulted in a breakdown of
the usual correlation between determinants and credit rating. Understanding whether variables are permanent or transitory
components is crucial for investors and borrowers to anticipate credit rating changes during economic downturns.e
dependent variables are the long-term domestic credit rating categories. e independent variables are derived from the
Fitch Ratings credit rating methodology and the literature, which includes quantitative and qualitative variables. e
statistical methods used are ordinal logistic regression, generalized ordinal logistic regression, and support vector machines.
e COVID-19 crisis was considered a transitory event due to the application of the through-the-cycle approach by rating
agencies. During the pandemic, specic determinants of credit ratings are not considered due to their transitory nature.
e study identies interest coverage ratio and competitive position as transitory components. is approach led to less
predictability but a more stable credit rating.
Keywords:credit rating, through-the-cycle approach, COVID-19, nancial information.
Correspondence address
Dante Domingo Terreno
Universidad Nacional de Córdoba, Facultad de Ciencias Económicas, Departamento de Contabilidad y Ciencias Jurídicas
Bv. Enrique Barros, s/n – X5000HRV
Ciudad Universitaria – Córdoba – Argentina
Impact of the COVID-19 outbreak on credit ratings: Application of the through-the-cycle approach
2Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
Impacto do surto de COVID-19 nos ratings de crédito: aplicação da abordagem
through-the-cycle
RESUMO
O objetivo deste estudo foi analisar como a crise da COVID-19 afetou os determinantes e a previsibilidade do rating de crédito
doméstico emitida pela Fitch Ratings na Argentina. Além disso, pretende-se avaliar os efeitos das agências de classicação de
risco de crédito usando o método through-the-cycle (ao longo do ciclo). Dada a natureza subjetiva da categorização dos ratings
de crédito, os pesquisadores desenvolveram modelos para explicar e prever esses ratings. Essa subjetividade é signicativa durante
eventos econômicos. Portanto, é importante investigar se os fatores que determinam e preveem os ratings de crédito permaneceram
consistentes antes e durante a crise da COVID-19. Este artigo contribui signicativamente para a compreensão de como a
aplicação do método through-the-cycle afeta os determinantes e a previsibilidade dos ratings de crédito em crises econômicas.
A aplicação do método through-the-cycle pelas agências de classicação de risco de crédito como um critério durante a crise da
COVID-19 resultou em uma quebra da correlação usual entre os determinantes e os ratings de crédito. Entender se as variáveis
são componentes permanentes ou transitórios é fundamental para que os investidores e tomadores de empréstimos antecipem as
mudanças nos ratings de crédito durante as recessões econômicas. As variáveis dependentes são as categorias de rating de crédito
doméstico de longo prazo. As variáveis independentes são derivadas da metodologia de rating de crédito da Fitch Ratings e da
literatura, que inclui variáveis quantitativas e qualitativas. Os métodos estatísticos utilizados são a regressão logística ordinal,
a regressão logística ordinal generalizada e as máquinas de vetores de suporte. A crise da COVID-19 foi considerada um evento
transitório devido à aplicação da abordagem through-the-cycle pelas agências de classicação de risco de crédito. Durante a
pandemia, os determinantes especícos dos ratings de crédito não são considerados devido à sua natureza transitória. O estudo
identica o índice de cobertura de juros e a posição competitiva como componentes transitórios. Essa abordagem levou a uma
menor previsibilidade, mas a um rating de crédito mais estável.
Palavras-chave: rating de crédito, abordagem through-the-cycle, COVID-19, informações nanceiras.
1. INTRODUCTION
Rating agencies play a crucial role in the nancial
markets. ey provide an independent opinion about an
issuer’s fundamental creditworthiness and ability to meet
its debt obligations in full and on time. e opinion is
expressed in the form of a credit rating. According to Kang
and Liu (2007), nancial markets have widely adopted
credit ratings because they can predict the likelihood
of defaults by reecting changes in credit quality levels.
Credit rating agencies issue global and domestic credit
ratings. Local credit ratings exclude sovereign eects,
transfer risk, and the possibility that investors may be
unable to repatriate interest and principal payments
(FixScr, 2014). ese ratings reect the perceived level
of risk and ability to fulll obligations within a specic
country. Countries with middle-income economies have
more domestic credit rating agencies and more developed
domestic bond markets.
However, rating agencies do not disclose the
methodology used for determining credit ratings, which
remains opaque and subjective. As such, it is dicult
to independently reproduce credit ratings with 100%
accuracy (Shin & Han, 2001). Given this subjectivity,
research has sought to identify the variables that underpin
credit ratings in order to anticipate those ratings or detect
situations where credit rating agencies apply lax criteria.
According to certain investors, rating agencies should
update their ratings more quickly (Chodnicka-Jaworska,
2022; Altman & Rijken, 2004). One widely accepted reason
for this is the agencies’ through-the-cycle methodology
and rating migration policy (Altman & Rijken, 2004). In
contrast to one-year default prediction models, rating
agencies using the through-the-cycle approach focus on
the probability of default in a stress scenario, with the
reference point being the permanent credit quality of a
borrower. erefore, this method requires the separation
of permanent and transitory components (Löer, 2004).
e time horizon considered by agencies for credit ratings
can be viewed as a period ranging from ve to ten years
(Gonzales et al., 2004). e purpose is to provide greater
stability to credit ratings (Altman & Rijken, 2004).
e COVID-19 outbreak resulted in an unprecedented
decline in global economic activity and increased global
nancial risks, which adversely aected global nancial
markets (Gormsen & Koijen, 2020; Phan & Narayan,
2020). e pandemic has led to extensive research on
its eects, assessing its impact on the economy and
Dante Domingo Terreno & María Eugenia Donadille
3
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
nancial market (Fernandes, 2020; Sharif et al., 2020), rm
bankruptcy (Bernardi et al., 2021), corporate performance
(Hu & Zhang, 2021), and credit risk downgrades (Altman
et al., 2022).
In a recent study by Dubinova et al. (2021), it was
found that there was a shi in the correlation between
macro fundamentals and credit risk at the beginning
of the pandemic. However, no research has been
conducted to determine the factors that inuenced
credit ratings during the COVID-19 crisis. e question
remains whether the determinants and predictability of
credit rating categories in economic stability remained
consistent during the COVID-19 crisis. Rating agencies
typically use the through-the-cycle approach to assess the
probability of default in a stress scenario. It is important
to assess how this approach aected the determinants
and predictability during the COVID-19 crisis. is
study aims to address this research gap by investigating
these questions.
is study aims to examine the impact of the COVID-19
crisis on the determinants and predictability of the
domestic credit rating issued by Fitch Ratings in Argentina,
and to assess the consequences of the application of the
through-the-cycle method by credit rating agencies.
is work compares the results of the proposed models
for 2020-2021 (during the COVID-19 outbreak) with
the period 2018-2019 (before the COVID-19 outbreak).
Fix Scr is the aliate of Fitch Ratings in Argentina and
is responsible for issuing most domestic credit ratings.
Fix Scr (2020) documented that during the COVID-19
pandemic, ratings were based on the expected credit
prole by the end of 2021, rather than the worst moment
of the crisis.
e research takes place in Argentina, which is the
third-largest economy in Latin America, following
Brazil and Mexico. e country has abundant natural
resources, which has made it one of the region’s leading
food producers and exporters. However, over the past
decade, the economy has experienced macroeconomic
uncertainty, including ination, exchange rate uctuations,
and a decline in production levels (Aromí et al.. 2022;
Cepal, 2020). In this situation, the pandemic had a severe
impact on Argentina, with its gross domestic product
(GDP) falling by 10% in 2020 (IMF, 2021). As a result, the
COVID-19 crisis had a signicant impact on Argentine
businesses, making it an interesting case study to evaluate
the eects of the pandemic.
is study contributes to a better understanding of
the rating methodology used by credit rating agencies
and, consequently, to the predictability of credit ratings
during economic crises. e results of this study can
help investors decide whether to trust the credit ratings
assigned by credit rating agencies, anticipate any potential
downgrades in domestic credit ratings, and enable debt
issuers to determine their borrowing costs. e academic
signicance of this research lies in the renement of credit
risk assessment studies by distinguishing between the cycle
and point-in-time methods. Furthermore, the implications
of this research may be helpful to policymakers by helping
them maintain an ideal balance between rating stability
and rating timeliness.
is paper is divided into ve sections. e second
section provides the theoretical framework; the third
section presents the data and empirical methodology; the
fourth section presents the results, which include model
estimation and forecasting; and the h section concludes
the study with concluding remarks and suggestions for
future research.
2. THEORETICAL FRAMEWORK
Several studies have found that the market-based
model is more eective in explaining credit ratings
than the accounting-based model (Figlioli et al., 2019;
Novotná, 2013; Tanthanongsakkun & Treepongkaruna,
2008). However, Du and Suo (2007) suggest that Merton’s
theoretical default measure is not a sucient statistic of
stock market information on credit quality.
Financial ratios have a pronounced eect on credit
ratings, mainly interest coverage with earnings before
interest, taxes, depreciation, and amortization (EBITDA)
and leverage (Gray et al., 2006; Feki & Khou, 2015; Hung
et al., 2013). Other studies have also shown the relevance
of rm size, as measured by total assets, and liquidity
(Feki & Khou, 2015). Damasceno et al. (2008) found that
return on assets, total debt to total assets, and presence
in the capital market are essential factors in determining
a corporate credit rating. Access to external nancing is
also an essential factor (Murcia et al., 2014). Additionally,
Drobetz and Heller (2014) suggested that protability
does not signicantly aect the rating assessment.
The literature suggests that incorporating both
quantitative and qualitative factors can improve the
predictability of credit rating models. Lehmann (2003)
confirmed that including qualitative information
signicantly improves model performance for dierent
classication measures. More recently, Drobetz and Heller
Impact of the COVID-19 outbreak on credit ratings: Application of the through-the-cycle approach
4Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
(2014) suggested that strategic objectives and future
liquidity risks are the most important business risk factors
aecting credit ratings and that qualitative information is
relevant in explaining credit ratings. Soares et al. (2012)
found that corporate governance is the main determinant
of credit ratings, along with accounting data.
According to the literature, the eects of variables on
credit ratings are not direct and linear across all categories.
For instance, Gray et al. (2006) found that nancial ratios
aect credit rating categories dierently. In particular,
nancial ratios help distinguish between A- and BBB-
rated rms, but are less precise in separating AA- from
A-rated rms. Krichene and Khou (2016) noted that the
interest coverage ratio loses all signicance when it falls
below zero or exceeds 20. Likewise, the debt coverage
ratio loses all signicance when it falls below negative
one or exceeds one. Blume et al. (1998), motivated by the
strong skewness in the distribution of interest coverage,
support the hypothesis that there is a non-linear eect
for the interest coverage ratio.
e studies on credit ratings in Argentina found
more interest in the sovereign credit market than in the
corporate debt market. Freitas and Minardi (2013) found
that the announcement of rating downgrades signicantly
impacts Latin American stock prices.
Other papers have examined the impacts of COVID-19
on corporate credit ratings. In one such study, Altman et al.
(2022) estimated the impact of the COVID-19 pandemic
on credit risk changes. ey applied the Altman Z”-score
model to analyze several possible crisis scenarios. e
analysis showed that the subsequent downgrades from
the base case (in 2019) are non-linear for the initial
rating category or the economic sector. e severity
of the downgrades in dierent scenarios depends on
the characteristics of individual rms and cannot be
determined at a general or sectoral level.
Dubinova et al. (2021) showed that credit risk models
based on observable covariates typically suer from
instability problems from the pre-COVID-19 period to
the early pandemic months. In contrast, models based on
unobserved components and frailty dynamics appear to
capture credit dynamics better, even in extreme periods
such as the COVID-19 pandemic. Chodnicka-Jaworska
(2022) carried out a study of European banks during the
COVID-19 pandemic (2000–2021). is study conrms
the strong impact of the macroeconomic environment on
default risk and the direct inuence on the methodology
used by agencies. It also conrms the notion of a delayed
reaction of agencies to changes in the situation during
the pandemic. Furthermore, the study reveals a more
substantial impact on banks from developing countries
and outside the Eurozone.
3. DATA AND EMPIRICAL METHODOLOGY
3.1 Dependent Variable
e dependent variables are the domestic long-term rating categories. e scale of Fix Scr for Argentina is divided
into four ordinal rating categories, as shown in Table 1.
Table 1
Ordinal rating categories
Rating Number of rm-years Combined rating Ordinal rating
AAA 7 AAA/AA 4
AA 22
A 52 A 3
BBB 49 BBB 2
BB 5
BB/C 1
B 4
CCC 3
CC 5
C 3
Total 150
Source: Elaborated by the authors.
Dante Domingo Terreno & María Eugenia Donadille
5
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
e rating categories are combined based on the
following criteria:
y
AAA/AA: e rating categories AAA and AA are
combined in the same group; both imply very solid
credit quality and the lowest relative expectation of
default risk.
yA: is category implies very solid credit quality, but
changes in economic conditions may aect the ability
to meet obligations.
y
BBB: is rating category indicates adequate credit
quality, but changes in economic conditions have
the highest probability of aecting the ability to meet
obligations.
yBB/C: e rating category BB denotes a high risk of
default, and the rm is more vulnerable to changes
in economic conditions. Rating category B indicates
higher vulnerability than BB and is dependent on
sustained and favorable development of economic
conditions. Categories CCC to C denote a high risk
of default, with C indicating a high risk of default if
economic and business conditions do not change.
Rating categories BB to C are grouped together because
there are few observations in each category. However,
this group includes ratings with dierent levels of
nancial distress. It is dicult for rms with nancial
problems to be removed from the rating.
y
D: Rating category D denotes an issuer that has entered
bankruptcy, which does not need to be included in a
model because it can be objectively known.
3.2 Independent Variables
e independent variables were obtained from the
Fitch Ratings credit rating methodology and variables
used in previous research (Jiang & Packer, 2017; Drobetz
& Heller, 2014). Later, the selection of variables was based
on the signicance of the coecients and their ability to
reect the character of the credit rating category.
e variables can be divided into quantitative and
qualitative factors.
3.2.1 Quantitative factors
3.2.1.1 Firm size
Studies of bankruptcy have identied rm size as an
important explanatory variable. Larger rms generally
have access to a wider range of nancing sources and
more exibility to redeploy assets than smaller rms.
Until recently, the probability of bankruptcy was very low
for large rms (Wahlen et al., 2014).e larger the rm,
the greater the potential to diversify non-systematic risks,
which reduces the risk of the company’s bonds (Elton &
Gruber, 1995). Domestic rating agencies weigh size more
heavily as a positive credit risk factor than global agencies
(Jiang & Packer, 2017). Most studies measure size using
total assets, which are calculated as follows:
( )
logFirm size assets=
1
e asset value was adjusted using the IPC (Consumer
Price Index) from the scal year-end to the last month of
the study period (October 2021) to achieve homogeneity
in the values.
3.2.1.2 Leverage
e leverage ratio measures how much a rm is
nanced with debt. e greater the rm’s leverage ratio,
the greater its risk of failure. Conversely, a lower leverage
ratio leads to a better rating for the rm. is ratio can
be calculated as follows:
Total liabilities
Leverage Assets
=
2
3.2.1.3 Interest coverage ratio
e interest coverage ratio with EBITDA is part
of Fitch Ratings’ methodology, and there is a more
frequently cited determinant variable in the literature
(Feki & Khou, 2015). e interest coverage ratio
indicates the number of times a rm’s earnings or cash
ow could cover its interest expenses. is ratio can be
calculated as follows:
EBITDA
Interest cover age ratio Interest expenses
=
3
EBITDA: Earnings before interest, taxes, depreciation,
and amortization.
Nominal interest includes ination coverage as
Argentina is an inationary economy. Interest expenses
are calculated using the average of the last three years.
Intermediate periods are annualized.
3.2.1.4 Financial exibility
Graham and Harvey (2001) report that corporate
managers consider nancial exibility and maintaining a
good credit rating as the two most important determinants
of their debt nancing policy. An analysis of Fitch’s
Impact of the COVID-19 outbreak on credit ratings: Application of the through-the-cycle approach
6Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
reported ratings reveals that the main characteristic of
rms rated between BB and CCC is limited nancial
exibility. ese rms face diculties in rolling over
their obligations due to insucient cash ow. Financial
exibility (FF) can be measured at a dierent level based
on the number of times net earnings are negative over
the analysis periods. is is presented as a categorical
variable, as shown in Tab l e 2:
Table 2
Financial exibility
Variable name Financial exibility level FFaDummy variable
FF1 High 0 0 or 1
FF2 Medium 1 0 or 1
FF3 Moderate 2 0 or 1
FF4 Limited 3 0 or 1
a
; NITWO = One if net income was negative, zero otherwise.
Source: Elaborated by the authors.
3.2.2 Qualitative factors
Qualitative factors are information that is not measured by a number, but can represent either negative or positive
forces aecting the rm. e interpretation of qualitative data implies a certain degree of subjectivity and depends
on the context (Liberti & Petersen, 2019). However, qualitative data can be summarized in numerical information.
3.2.2.1 Sector risk
One of the rst steps in analyzing a rm is to determine
the characteristics of the economic sector or industry in
which it participates (FixScr, 2014). e main factors
considered are industry characteristics, competitiveness,
growth prospects, entry and exit barriers, regulations,
cyclical factors, price volatility, and counterparty risk.
is variable is dened as an ordinal variable based on
the level of risk in the rm’s sector, as shown in Tabl e 3:
Table 3
Sector risk
Sector level risk Ordinal variable
Low 1
Medium 2
High 3
Source: Elaborated by the authors.
3.2.2.2 Competitive position
Competitive position seeks to determine how the
rm is positioned within its specic sector and its
performance within it (FixScr, 2014). e main factors
considered are market share, geographic and product
diversication, business integration, supplier and buyer
power, and economies of scale. is variable is classied
as follows in Table 4:
Table 4
Competitive position
Competitive position Ordinal variable
High 1
Medium 2
Low 3
Source: Elaborated by the authors.
Finally, Table 5 summarizes the independent variables used in the analysis.
Table 5
Financial and qualitative variables
Name Description Type
FrmSz Firm size Numeric
Lev Leverage Numeric
EA Interest coverage ratio Numeric
FF1 Financial exibility-High Categorical
FF2 Financial exibility-Medium Categorical
FF3 Financial exibility-Moderate Categorical
FF4 Financial exibility-Limited Categorical
SR Sector risk Ordinal
CP Competitive position Ordinal
Source: Elaborated by the authors.
Dante Domingo Terreno & María Eugenia Donadille
7
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
3.3 Statistical Methods
3.3.1 Ordinal logistic regression
e most common statistical methodologies in credit
rating prediction are ordinal logistic or probit models
because rating categories can be represented as ordinal
variables (Amato & Furne, 2004; Drobetz & Heller,
2014). Ordinal logistic regression (OLR) is a statistical
method that models the relationship between an ordinal
multilevel dependent variable and independent variables.
e values of the dependent variable have a natural order
or ranking. e OLR model compares the probability of a
response less than or equal to a given category (j=1,…J-1)
to the probability of a response in a higher category. e
model can be expressed as follows (Liu, 2009):
( )
12
|,
n
Logit Y j x x x≤…
4
( )
( ) ( )
12
11 22
12
|,
|,
n
j nn
n
Y jx x x
ln X X X
Y jx x x
παββ β
π
≤…
= = + − − …−
>…
where x=[x1, x2,…,xn]T is a vector of n explanatory
variables, β = [β
1
,β
2
,…, β
n
]
T
is the corresponding
coecient vector, and α is the cut-o point for rating
category y. us, this model predicts cumulative logits
across J−1 response categories. e cumulative logits
can then be used to calculate the estimated cumulative
odds and the cumulative probabilities at or below the
j category.
One of the key points of OLR is the proportional
odds assumption, which assumes that the eect of the
explanatory variables on the independent variable is
constant across all categories. is assumption implies
that the coecients of the independent variable are
consistent across the categories, resulting in parallel
slopes at all response levels. is requirement is essential
for interpreting model coecients and the validity of
predictions.
e proportional odds assumption holds when the
regression x’ β is independent of j, such that β has the
same eect for each of the j-1 cumulative logits. It is
noteworthy that x’ β does not contain an intercept, since
the α
j
(threshold) acts as an intercept. Another assumption
is the absence of multicollinearity, which occurs when
the independent variables are too highly correlated. e
models are estimated using the maximum likelihood
method, and the observed information matrix calculates
variance estimates.
3.3.2 Generalized ordinal logistic regression
e generalized ordinal logistic regression (GOLR)
model extends the OLR model by relaxing the proportional
odds assumption. When a particular predictor violates
the assumption, its eect will be estimated freely across
dierent categories of the dependent variable. e GOLR
model is expressed as follows (Williams, 2006):
( )
12
|,
n
Logit Y j x x x>…
5
( )
( )
( )
12
11 2 2
12
|,
|,
n
j j j nj n
n
Y jx x x
ln X X X
Y jx x x
παβ β β
π
≤…
= = + − − …−
>…
In this expression, all the eects of the independent
variables vary at each cut-o point. If some of these eects
are stable, they will be constrained to be equal, as in the
proportional odds assumption. us, the GOLR model
refers to the case in which at least one of the coecients
for a predictor varies across categories.
3.3.3 Support vector machines
e bibliographic review of Louzada et al. (2016) on
classication methods applied to credit scoring nds that
the support vector machine method has better predictive
performance than other methods.
Support vector machines (SVMs) seek to nd an
optimal hyperplane with a maximum margin that acts as
the decision boundary to separate two dierent categories.
Given a training set of labeled instance pairs (x〗i,y_i),
where i is the number of instances i = 1,2,3, …, m, x〗_i ∈R
and y_i ∈ {−1, +1}, the decision boundary to separate two
dierent categories in the SVM is generally expressed as:
*0wx b+=
6
e optimal separating hyperplane is the only one
with maximum margin, and all training instances are
assumed to satisfy the constraint:
( )
* 1
ii
ywx b+≥
7
Impact of the COVID-19 outbreak on credit ratings: Application of the through-the-cycle approach
8Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
e convex optimization problem is dened as follows:
,
(7)
8
s.t. 1 (8)
9
e optimal hyperplane is equivalent to the optimization
problem of a quadratic function, where the Lagrange
function is utilized to nd the global maximum. e slack
variable ϵi is introduced to account for misclassication,
accompanied by C as the penalty cost. e kernel trick is
used to modify the SVM formulation. Linear and radial
basis function (RBF) kernels are used:
iLinear:
(9)
10
ሺiiሻRadial basis function ሺRBFሻ: exp ቄെߛฮݔെݔ
ฮଶቅ (10)
11
is explanation can be extended to more than two
variables using the same reasoning.
3.4 Model Estimation
e dependent variable in this study is the credit rating
category, represented as an ordinal variable, as shown
in Table 1. e independent variables are size, leverage,
interest coverage ratio, nancial exibility, sector risk, and
competitive position, as shown in Table 5. To model this
relationship, the following expression is used:
( )
1..4
, , , , ,
j
R f FrmSz Lev EA FF SR CP
=
=
Rj represents the credit rating category (j = 1...4) and FF
is a categorical variable that captures nancial exibility.
Specically, it tests the eects of three levels of nancial
exibility (medium, moderate, and limited) relative to a
high level captured by the intercept FF1. RLO and RLOG
will be applied to obtain the magnitudes and signicance
levels of the regression coecients.
is study will also evaluate the predictive accuracy of
the model. Overtting is one of the biggest issues when
building an eective predictive model. is occurs when
a statistical model is too closely aligned with a limited
set of data points. erefore, it is crucial to measure
the predictive accuracy with out-of-sample data. e
most commonly used method for this purpose is cross-
validation, which involves randomly partitioning the
original sample into k equal-sized subsamples. A single
subsample is retained as the validation data for testing
the model, and the remaining k−1 subsamples are used
as training data. en, the cross-validation process is
repeated k times, with each of the k subsamples used
exactly once as the validation data. e k results can then
be averaged to produce a single estimation.
3.5 Data and Summary Statistics
e data were obtained from the FixScr national
rating reports of the company issuers of long-term
financial obligations in Argentina (FixScr, 2021).
e sample includes large and medium-sized rms
rated in 2018-2019 (before the COVID-19 crisis) and
2020-2021 (during the COVID-19 crisis). e nancial
data cover the interim nancial statements prior to the
issuance of the rating and two subsequent scal years. e
dataset contains 150 rm-year observations, 75 rm-years
per period. e qualitative factors were obtained from the
Fix Scr rating reports, rms’ annual reports, and other
publicly available information.
e rule for identifying outliers is based on considering
any data point that is more than 2.5 standard deviations
(x ± 2.5 σ) away from the mean in a sample. According
to this criterion, the variable interest coverage ratio was
winsorized at the 4th and 96th percentiles; the other
variables are not winsorized. e interest coverage ratio
has a signicant dispersion due to interest rate volatility
caused by ination.
Table 6 presents the number of rm-years for each
rating category. e A and BBB categories have the highest
number of rm-year observations, while the remaining
categories have relatively fewer observations; therefore,
they were combined.
Table 6
Firm-years for rating categories
Rating Number of rm-years Combined rating Number of rm-years Number of rms
during COVID-19
Number of rms
before COVID-19
AAA 7 AAA/AA 29 16 13
AA 22
A 52 A 52 26 26
BBB 49 BBB 49 21 28
12
Dante Domingo Terreno & María Eugenia Donadille
9
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
Rating Number of rm-years Combined rating Number of rm-years Number of rms
during COVID-19
Number of rms
before COVID-19
BB 5
BB/C 20 12 8
B 4
CCC 3
CC 5
C 3
Total 150 150 75 75
Source: Elaborated by the authors.
Table 7 shows the changes in credit rating categories
during the COVID-19 crisis relative to the pre-COVID-19
period. e data reveal that 32% of rms experienced
a change in their credit rating category, with a higher
proportion of low-rated rms being aected.
Table 7
Changes in rating categories for rm-years during the COVID-19 crisis relative to the pre-COVID-19 period
Rating Number of rms changing categories Number of rms keeping categories Number of rms pre-COVID-19
AAA/AA 2 11 13
A 7 19 26
BBB 11 17 28
BB/C 4 4 8
Total 24 51 75
% 32.00% 68.00% 100.00%
Source: Elaborated by the authors.
Table 8 provides insights into the means of the variables
before and during the COVID-19 crisis. e results show
that the mean values of FrmSz, Lev, EA, and CP remained
similar in both periods. However, the distribution of EA
and CP diers across rating categories. e variables FF
and SR increase in value in most categories except AAA/
AA, which is negatively related to the rating categories.
Table 8
Means for each rating category during and before COVID-19
Panel A: During the COVID-19 crisis
Rating FrmSz Lev EA FF SR CP
AAA/AA 8.1877 0.6657 4.2150 2.1250 2.0625 2.4375
A 7.5463 0.6614 5.5640 1.5769 2.1538 2.2692
BBB 7.0640 0.7536 3.9216 1.8095 2.3330 1.7619
BB/C 7.3916 0.7684 2.2935 3.3000 2.3000 2.0000
Total 7.5474 0.7123 3.9985 2.2029 2.2123 2.1172
Panel B: Before the COVID-19 crisis
Rating FrmSz Lev EA FF SR CP
AAA/AA 8.1744 0.6683 6.0026 1.3038 1.5385 2.8452
A 7.4300 0.6693 5.2825 1.6154 1.9231 2.1923
BBB 7.0791 0.7561 3.2703 2.0000 1.9286 1.8214
BB/C 7.0781 0.7936 1.3202 3.0000 2.1000 1.7000
Total 7.4404 0.7218 3.9689 1.9798 1.8725 2.1397
FrmSz = Firm size; Lev = Leverage; EA = Interest coverage ratio; FF = Financial exibility; SR = Sector risk; CP = Competitive
position.
Source: Elaborated by the authors.
Table 6
Cont.
Impact of the COVID-19 outbreak on credit ratings: Application of the through-the-cycle approach
10 Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
e pairwise correlations between the variables during
and before the COVID-19 crisis are presented in Tab l e 9.
e results indicate that FrmSz has a higher correlation
with CP, Lev with EA, and EA with FF during and before
the COVID-19 period. Moreover, the pre-COVID-19
period shows higher correlations between Lev and FF.
Table 9
Correlation matrix
Panel A: During the COVID-19 crisis
FrmSz Lev EA FF SR PC
FrmSz 1.0000
Lev -0.1150 1.0000
EA -0.0948 -0.4383 1.0000
FF 0.1841 0.1539 -0.4688 1.0000
SR 0.2018 0.2221 -0.0968 0.1573 1.0000
CP 0.3383 -0.2253 0.1468 -0.0482 -0.2921 1.0000
Panel B: Before the COVID-19 crisis
FrmSz Lev EA FF SR PC
FrmSz 1.0000
Lev -0.0221 1.0000
EA 0.0981 -0.4247 1.0000
FF 0.0340 0.5443 -0.3939 1.0000
SR 0.0091 0.0743 -0.0842 0.0955 1.0000
CP 0.2426 -0.1983 0.2067 -0.1144 -0.2545 1.0000
FrmSz = Firm size; Lev = Leverage; EA = Interest coverage ratio; FF = Financial exibility; SR = Sector risk; CP = Competitive
position.
Source: Elaborated by the authors.
4. EMPIRICAL RESULTS
4.1 Model Estimation
Table 10 presents the results estimated by OLR.
Among the quantitative variables, only the FrmSz
coecient is statistically signicant during and before
the COVID-19 crisis. e positive sign of the FrmSz
coecient indicates a higher value for an increase in
the corporate credit rating. ese results highlight the
importance of rm size in credit rating classication in
dierent situations, suggesting that this variable has a
persistent character.
Among the qualitative variables, the FF4 coecient
with FF1 and the SR coecient are signicant in both
periods. Conversely, the negative sign of the FF4 and SR
coecients indicates that a higher value decreases the
corporate credit rating. Firms facing potential nancial
problems, as measured by low nancial exibility, continue
to be a relevant variable in the pandemic crisis. e
inherent risks of the sector also remain signicant.
e coecients of the variables EA, FF3 with respect
to FF1, and CP are statistically signicant only in the
pre-COVID-19 period. e positive sign of the EA and
CP coecients indicates that a higher value increases
the credit rating. In contrast, the negative sign of FF3
indicates that a higher value decreases the credit rating.
e interest coverage ratio is sensitive to economic activity;
however, its relevance was bypassed during the pandemic
because it was considered a transitory event. Under
normal circumstances, prior to the COVID-19 period,
the interest coverage ratio served as a crucial nancial
metric in credit rating assessments. e rm’s competitive
position was aected by the pandemic. Despite this eect,
credit rating agencies tended to downplay its signicance
by viewing the pandemic as transitory.
Dante Domingo Terreno & María Eugenia Donadille
11
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
Table 10
Ordinal logistic regression model estimates
Panel A: During the COVID-19 crisis Panel B: Before the Covid-19 crisis
Coefcients Coefcients
Value t-value Value t-value
FrmSz 2.6218 5.22 *** 3.6391 5.22 ***
Lev -2.6136 -1.37 -1.2871 -0.57
EA 0.1172 1.11 0.2401 2.48 **
FF2 -0.4292 -0.70 -1.2665 -1.72
FF3 -0.9618 -1.25 -2.2671 -2.48 **
FF4 -3.4886 -3.38 *** -2.8752 -2.76 ***
SR -0.7983 -2.05 ** -0.8645 -2.09 **
CP -0.3420 -1.12 0.7754 2.39 **
N 75 75
LR chi2(8) 53.13 *** 80.88 ***
Pseudo R2 0.2631 0.4220
FrmSz = Firm size; Lev = Leverage; EA = Interest coverage ratio; FF2 = Medium nancial exibility; FF3 = Moderate nancial
exibility; FF4 = Limited nancial exibility; SR = Sector risk; CP = Competitive position.
*** Signicant at 1%; ** Signicant at 5%.
Source: Elaborated by the authors.
Table 11
Collinearity diagnostics
Variables Panel A – Period: 2021-2020 Panel B – Period: 2019-2018
VIF Tolerance VIF Tolerance
FrmSz 1.2900 0.7752 1.2000 0.8333
Lev 1.4300 0.6993 1.6400 0.6098
EA 1.7000 0.5882 1.3200 0.7576
FF(2) 1.5500 0.6452 1.3600 0.7353
FF(3) 1.6700 0.5988 1.4300 0.6993
FF(4) 1.3400 0.7463 1.5000 0.6667
SR 1.3100 0.7634 1.1000 0.9091
CP 1.3600 0.7353 1.2700 0.7874
Mean 1.4600 1.3500
VIF =Variance Ination Factor; Tolerance = (1/VIF)
Source: Elaborated by the authors.
e VIF (variance ination factor) indicates the
degree to which the variance of the coecient estimate
is inated due to multicollinearity. As with tolerance, there
is no specic threshold value to denitively determine
the presence of multicollinearity. However, VIF values
exceeding 2.5 are oen considered a potential cause for
concern (Johnston et al., 2018). In Table 11, the VIF
values of the variables do not exceed the aforementioned
threshold, suggesting that multicollinearity may not be a
signicant issue in the model.
e likelihood ratio test of the proportional odds
assumption, shown in the notes to Tab l e 12, indicates the
violation of the proportional odds assumption; therefore,
it is applied to the GOLR. Panel A presents the results of
the GOLR estimates during the COVID-19 crisis. e
variables FrmSz, FF3, and SR violate the proportional
odds assumption; their coecients dier across rating
categories. FrmSz and SR have a signicant coecient for
discriminating the AAA/AA vs. A and A vs. BBB rating
categories, with FrmSz having a positive eect and SR
Impact of the COVID-19 outbreak on credit ratings: Application of the through-the-cycle approach
12 Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
hurting the higher rating categories. e FF4 variable
has a signicant coecient for all rating categories,
and the FF3 variable has a signicant coecient only
for discriminating the BBB vs. BB/C rating categories.
ese variables have a negative eect on credit rating
classication.
Table 12
Generalized ordinal logistic model estimation
Panel A: During the COVID-19 crisis
From AAA/AA to A From A to BBB From BBB to BB/C
Threshold coefcients
Variables Value t value Value t value Value t value
FrmSz 7.2188 4.06 *** 5.6913 3.55 *** 0.4609 0.53
Lev (+) -2.0810 -1.14 -2.0810 -1.14 -2.0810 -1.14
EA (+) 0.2163 1.47 0.2163 1.47 0.2163 1.47
FF2 (+) -0.3130 -0.40 -0.3130 -0.40 -0.3130 -0.40
FF3 1.7888 1.63 -0.0448 -0.41 -4.8764 2.82 ***
FF4 (+) -4.7530 -3.13 *** -4.7530 -3.13 *** -4.7530 -3.13 ***
SR -2.8124 -3.34 *** -3.2259 -2.67 *** 1.8664 1.83
CP (+) -0.6930 -1.82 -0.6930 -1.82 -0.6930 -1.82
Intercept -48.8929 -3.77 *** -34.1051 -3.26 *** -0.6215 -0.10
N 75
LR chi2(12) 104.74 ***
Pseudo R2 0.5186
Panel B: Before the COVID-19 crisis
From AAA/AA to A From A to BBB From BBB to BB/C
Threshold coefcients
Variables Value t value Value t value Value t value
FrmSz (+) 4.9710 4.94 *** 4.9710 4.94 *** 4.9710 4.94 ***
Lev (+) -1.2755 -0.48 -1.2755 -0.48 -1.2755 -0.48
EA 0.0815 0.55 0.4688 2.35 ** 2.1105 3.55 ***
FF2 -1.0854 -0.89 -1.1888 -1.26 -4.6889 -3.03 ***
FF3 (+) -1.6717 -1.51 -1.6717 -1.51 -1.6717 -1.51
FF4 (+) -3.1856 -2.51 ** -3.1856 -2.51 ** -3.1856 -2.51 **
SR (+) -0.9839 -2.14 ** -0.9839 -2.14 ** -0.9839 -2.14 **
CP (+) 0.9943 2.69 *** 0.9943 2.69 *** 0.9943 2.69 ***
Intercept -39.1758 5.00 *** -36.7298 -4.80 *** -34.0436 -4.56 ***
N 75
LR chi2(12) 101.17 ***
Pseudo R2 0.5279
Note: Likelihood ratio test of proportional odds assumption: during COVID-19: chi2 = 71.21(Prob > chi2 = 0.00); pre-COVID-19:
chi2 = 45.60 (Prob > chi2 = 0.00). (+) The same coefcient for all rating categories.
FrmSz = Firm size; Lev = Leverage; EA = Interest coverage ratio; FF2 = Medium nancial exibility; FF3 = Moderate nancial
exibility; FF4 = Limited nancial exibility; SR =Sector risk; CP = Competitive position.
***Signicant at 1%; **Signicant at 5%.
Source: Elaborated by the authors.
Panel B presents the results of the GOLR estimates in
the pre-COVID-19 crisis period. e variables EA and
FF2, with respect to FF1, violate the proportional odds
assumption, while the remaining variables maintain the
same coecient for all rating categories. e variables
FrmSz, FF4, SR, and CP show a signicant coecient for
all rating categories, with the FrmSz and CP coecients
being positive, and the FF4 and SR coecients being
negative. Additionally, the AE variable has a signicant
positive coecient in discriminating between the A vs.
BBB and BBB vs. BB/C rating categories.
Dante Domingo Terreno & María Eugenia Donadille
13
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
Applying GOLR shows that during the COVID-19
crisis, rm size and sector risk are relevant factors in
determining the credit rating of rms with a rating above
BBB, with rm size having a positive eect and sector risk
having a negative eect. Limited and moderate nancial
exibility are also relevant in determining the credit rating
of rms with a BB/C rating. is means that rms with
limited and moderate nancial exibility, as determined
by the proxy used, experienced losses for three and two
years, respectively.
Before the COVID-19 crisis, rm size, limited nancial
exibility, sector risk, and competitive position were
relevant in all rating categories. Firm size and competitive
position have a positive impact on the rating categories,
while limited nancial exibility and sector risk have a
negative impact. e interest coverage ratio variable is
relevant for the categories below the BBB rating.
In conclusion, both methods produce similar results.
However, the main dierence is that GOLR captures
the nonlinearity of the relationship between covariables
and independent variables, as pointed out by Gray et al.
(2006). e dierence between the determinant before
and during the COVID-19 crisis is due to the through-the-
cycle approach used by Fitch Ratings, which considers the
COVID-19 crisis as a transitory event. According to this
approach, the permanent components are rm size, sector
risk, and nancial exibility, while the transitory components
are interest coverage ratio and competitive position.
Firm size and potential sector risk are considered
permanent factors due to their intrinsic long-term
behavioral characteristics. Financial flexibility, as
measured by cumulative negative net income, is also
considered a permanent factor due to the low possibility
of short-term reversal. Although the literature suggests
that the interest coverage ratio is a key determinant of
credit rating, it is viewed as a transitory factor because
of the probability of short-term reversal if the rm
maintains its renancing capacity. e pandemic aected
the competitive position of some rms, which is regarded
as a transitory factor due to the possible reversal of such
a situation.
4.2 Prediction
4.2.1 In-sample
First, the predictive power of the model is examined
using in-sample data. Predictive accuracy is measured
by comparing the predicted credit rating to the actual
rating for each rm and calculating the ratio of correctly
classied rms to the total number of rm observations.
Table 13 shows that the model has higher predictive
accuracy during the pre-COVID-19 period than during
the pandemic, using both the OLR and GOLR methods.
However, GOLR shows a greater dierence between
the pre-COVID-19 period and the COVID-19 crisis
(78.67 > 65.33) than OLR (69.33 > 64.00). is suggests
that the predictability of the model decreased during
the pandemic compared to the pre-COVID-19 period.
is can be explained by the cycle approach applied by
the rating agencies.
Table 13
Predictive accuracy with in-sample data
Panel A: During the COVID-19 crisis
Rating Ordinal logistic Generalized ordinal
regression logistic regression
AAA/AA 50.00% 68.75%
A 88.46% 80.77%
BBB 52.38% 57.14%
BB/C 50.00% 41.67%
Total 64.00% 65.33%
Panel B: Pre-COVID-19 crisis
Rating Ordinal logistic Generalized ordinal
regression logistic regression
AAA/AA 76.92% 76.92%
A 76.92% 80.77%
BBB 67.86% 78.57%
BB/C 37.50% 75.00%
Total 69.33% 78.67%
Source: Elaborated by the authors.
Impact of the COVID-19 outbreak on credit ratings: Application of the through-the-cycle approach
14 Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
Moreover, Panel A shows that during the COVID-19
crisis period, the OLR and GOLR methods have similar
predictive accuracy rates (64.00% ≈ 65.33%) for the
aggregate categories. e GOLR method outperforms
OLR in terms of accuracy for the AAA/AA rating category
(68.50%>50.00%), while the dierence in accuracy for the
A and BB/C rating categories is smaller (80.77% < 88.46%;
41.67% < 50.00%, respectively). In Panel B of Table 13, the
GOLR method has higher predictive accuracy than ORL
for the pre-COVID-19 crisis period (78.67% > 69.33%).
e individual rating categories also show high predictive
accuracy in nearly all rating categories, particularly in the
BB/C category (75.00%>37.50%). However, the AAA/
AA category has the same predictive accuracy in both
methods.
4.2.2 Out-of-sample
e statistical method used to predict the accuracy
includes OLR, GOLR, and SVM. e out-of-sample
validation is implemented by resampling the cross-
validation through 10 folds and repeating it ve times.
e radial basis function (RBF) kernel used for SVM is
more accurate than the linear method. Two parameters are
associated with the RBF kernel: cost of misclassication
(C) and gamma (γ). Technically, the gamma parameter is
the inverse of the standard deviation of the RBF kernel.
High gamma values usually produce highly exible
decision limits, and low gamma values oen result in a
more linear decision limit. e optimal values of the cost
and gamma parameters (C and ƍ) were obtained using
ve-fold cross-validation.
Table 14 presents the results of the study on the predictive
accuracy of the OLR, GOLR, and SVM statistical methods.
e results show that in the pre-COVID-19 period, the
three statistical methods used had greater predictive
accuracy compared to the COVID-19 crisis period. e
GOLR has the highest accuracy rate (70.40%>56.61%),
especially in the AAA/AA and BB rating categories. In
the A and BB/C categories, OLR and SVM both show
high accuracy.
When comparing the methods used, the
GOLR method has the highest accuracy in the pre-
COVID-19 crisis period (70.40% > 65.10% > 47.66%).
However, the SVM methods show slight superiority
(59.10% > 57.00% > 56.61) during COVID-19. Moreover,
there is no uniformity in all categories; for example, during
the COVID-19 crisis, OLR has the highest accuracy in
rating category A (80.69%) and SVM has the highest
accuracy in rating category BB/C (73.13%).
Table 14
Predictive accuracy cross-validation data
Rating
Ordinal logistic regression Generalized ordinal logistic regression Support vector machine
Periods Periods Periods
COVID-19 Pre-COVID-19 COVID-19 Pre-COVID-19 COVID-19 Pre-COVID 19
AAA/AA 41.31% 83.24% 58.59% 89.29% 47.42% 83.24%
A 80.69% 66.86% 50.59% 63.70% 68.30% 66.86%
BBB 44.64% 59.89% 65.88% 70.00% 48.57% 58.71%
BB/C 48.13% 47.66% 58.33% 69.70% 73.13% 47.66%
Total 57.00% 65.10% 56.61% 70.40% 59.10% 64.60%
C 200 200
ƍ0.01 0.01
Source: Elaborated by the authors.
In summary, the conclusions obtained from the out-
of-sample and in-sample data are similar. Predictability
decreases during the pandemic crisis due to the use of
the through-the-cycle approach by the rating agencies.
is conclusion is consistent with the results of the
determinants of credit rating. Overall, the generalized
ordinal logistic regression method showed superior
predictive performance in this study.
5. CONCLUSIONS
First, this study seeks to identify the determinants of
domestic credit rating in Argentina, explicitly issued by
Fitch Ratings, before and during the COVID-19 crisis.
e results indicate that the determinants of credit rating
during the pre-COVID-19 period are rm size, sector
risk, nancial exibility, interest coverage ratio, and
Dante Domingo Terreno & María Eugenia Donadille
15
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
competitive position. However, during the COVID-19
crisis, interest coverage ratio and competitive position
are not found to be relevant determinants. e dierence
in the determinants before and during the COVID-19
crisis is due to the through-the-cycle approach used by
rating agencies, in which the COVID-19 crisis is viewed
as a transitory event.
According to previous results, rm size, sector risk,
and nancial exibility are classied as permanent
components, while interest coverage ratio and competitive
position are classied as transitory components. Firm
size is particularly relevant in the upper categories (A/
AA), and moderate and limited nancial exibility is a
parameter for rating rms in the lower categories (BB/
CCC). e interest coverage ratio variable is relevant
for the categories below the BBB rating, conrming the
non-linearity of the variables across rating categories.
Although the literature suggests that the interest coverage
ratio is a determinant of credit rating, it was considered
a transitory component during the COVID-19 period.
We also examine how the predictability of credit
ratings was aected by COVID-19. e models used to
assess this predictability suggest that the accuracy of credit
ratings decreased during the COVID-19 crisis period. is
result is consistent with the through-the-cycle approach
adopted by credit rating agencies, which considers the
pandemic event to be transitory. is approach led to less
predictability and more stable credit ratings, as pointed
out by Löer (2004). is conclusion is consistent with
the results of the determinants of credit ratings.
e main limitation of this work was the small number
of observations in some categories. ese categories were
grouped together to address this, but this approach may
have led to some loss of information.
is paper contributes signicantly to understanding
the impact of the application of the through-the-cycle
method on the determinants and predictability of credit
ratings. e ndings of the study can help investors and
nancial analysts make more informed decisions and assess
the creditworthiness of companies during an economic
crisis. is study contributes to the development of the
domestic bond credit market in emerging economies,
improving the readability and transparency of credit rating
agencies. e conclusions make an important academic
contribution, as they will allow researchers to improve
credit risk assessment studies, considering the dierences
between the through-the-cycle approach and the point-
in–time approach. Future research on credit ratings
should incorporate contextual variables that capture the
application of the through-the-cycle approach.
REFERENCES
Altman, E. I., Sieradzki, R., & lon, M. (2022). Assessing
Corporate Credit Risk Transitions and Bankruptcy
Prediction on SMEs as a result of the COVID-19 Pandemic
[Working Paper]. Social Science Research Network,
Rochester, NY. Retrieved from http://dx.doi.org/10.2139/
ssrn.4048333
Altman, E. I., & Rijken, H. A. (2004). How rating agencies achieve
rating stability. Journal of Banking & Finance,28(11), 2679-
2714. https://doi.org/10.1016/j.jbankn.2004.06.006
Amato, J. D., & Furne, C. H. (2004). Are credit ratings
procyclical?Journal of Banking & Finance,28(11), 2641-2677.
https://doi.org/10.1016/j.jbankn.2004.06.005
Aromí, D., Bermúdez, C., & Dabús, C. (2022). Incertidumbre
y crecimiento económico: enseñanzas de América
Latina.Revista CEPAL. Retrieved from https://repositorio.
cepal.org/handle/11362/48085
Bernardi, A., Bragoli, D., Fedreghini, D., Ganugi, T., &
Marseguerra, G. (2021). COVID-19 and rms’ nancial
health in Brescia: a simulation with Logistic regression and
neural networks.National Accounting Review,3(3), 293-309.
https://doi.org/10.3934/NAR.2021015
Blume, M., Lim, F., & MacKinlay A. (1998). e declining quality
of US corporate debt: Myth or reality? Journal of Finance, 53,
1389 – 1413. https://doi.org/10.1111/0022-1082.00057
Cepal, N.U. (2020). Balance preliminar de las economías de
América Latina y el Caribe 2020.Santiago: CEPAL. Retrieved
from https://repositorio.cepal.org/bitstream/handle
/11362/46501/4/BP2020_Argentina_es.pdf
Chodnicka-Jaworska, P. (2022). Impact of COVID-19 on
European banks’ credit ratings.Economic Research-
Ekonomska Istraživanja, 1-20. https://doi.org/10.1080/133167
7X.2022.2153717
Damasceno, D. L., Artes, R., & Minardi, A. M. A. F. (2008).
Determinação de rating de crédito de empresas brasileiras
com a utilização de índices contábeis.Revista de
Administração-RAUSP,43(4), 344-355.
Drobetz, W., & Heller, S. (2014). What factors drive corporate
credit ratings? Evidence from German SMEs and large
corporates [Working Paper]. Social Science Research Network,
Rochester, NY. Retrieved from http://dx.doi.org/10.2139/
ssrn.2392377
Dubinova, A., Lucas, A., & Telg, S. (2021). COVID-19, Credit
Risk and Macro Fundamentals. Tinbergen Institute
Discussion Paper 2021-059/III. Social Science Research
Network, Rochester, NY. Retrieved from http://dx.doi.
org/10.2139/ssrn.3875628
Du, Y., & Suo, W. (2007). Assessing credit quality from the
equity market: can a structural approach forecast credit
Impact of the COVID-19 outbreak on credit ratings: Application of the through-the-cycle approach
16 Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
ratings?Canadian Journal of Administrative Sciences/Revue
Canadienne des Sciences de l’Administration,24(3), 212-228.
https://doi.org/10.1002/cjas.27
Elton, E. J., Gruber, M. J., & Blake, C. R. (1995). Fundamental
economic variables, expected returns, and bond fund
performance. e Journal of Finance,50(4), 1229-1256.
https://doi.org/10.1111/j.1540-6261.1995.tb04056.x
Feki, A., & Khou, W. (2015). e determinants of issuers’ long
term credit ratings: American S&P500 index. International
Journal of Accounting and Economics Studies,3(1), 78–85.
https://doi.org/10.14419/ijaes.v3i1.4631
Fernandes, N. (2020), Economic eects of coronavirus outbreak
(COVID-19) on the world economy [Working Paper]. Social
Science Research Network, Rochester, NY. Retrieved from
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3557504
Figlioli, B., Antonio, R. M., & Lima, F. G. (2019). Stock Price
Synchronicity and Current and Potential Credit Ratings.
International Journal of Economics and Finance,11(10), 1-16.
https://doi.org/10.5539/ijef.v11n10p1
FixScr (2021). Calicaciones. Retrieved from https://www.xscr.
com/calicaciones
FixScr (2020). Corporates Argentina: Vulnerabilidad ante
COVID-19 Revisión Sectorial y por Emisor. Retrieved from
https://www.xscr.com/site/download? le=oJEvh8OT6l mrl
TUOQMz99E50l29BjBy9.pdf
FixScr (2014). Argentina: Metodología de calicación de
riesgos. Retrieved from https://www.xscr.com/site/
download?le=Manual+Empresas_Jun+2014.pdf
Freitas, A. D. P. N., & Minardi, A. M. A. F. (2013). e impact of
credit rating changes in Latin American stock markets.BAR-
Brazilian Administration Review,10, 439-461. https://doi.
org/10.1590/S1807-76922013000400005
Gonzalez, F., Haas, F., Persson, M., Toledo, L., Violi, R., Wieland,
M., & Zins, C. (2004). Market dynamics associated with
credit ratings: a literature review. ECB Occasional Paper Nº
16. Social Science Research Network, Rochester, NY. Retrieved
from http://dx.doi.org/10.2139/ssrn.752065
Gormsen, N. J., & Koijen, R. S. (2020). Coronavirus: Impact on
stock prices and growth expectations. e Review of Asset
Pricing Studies,10(4), 574–597. https://doi.org/10.1093/
rapstu/raaa013
Graham, J. R., & Harvey, C. R. (2001). e theory and practice of
corporate nance: Evidence from the eld. Journal of nancial
economics,60(2-3), 187–243. https://doi.org/10.1016/S0304-
405X(01)00044-7
Gray, S., Mirkovic, A., & Ragunathan, V. (2006). e determinants
of credit ratings: Australian evidence. Australian
Journal of Management,31(2), 333–354. https://doi.
org/10.1177/031289620603100208
Hung, K., Cheng, H. W., Chen, S. S., & Huang, Y. C. (2013).
Factors that aect credit rating: An application of
ordered probit models. Romanian Journal of Economic
Forecasting,16(4), 94-108.
Hu, S., & Zhang, Y. (2021). COVID-19 pandemic and rm
performance: Cross-country evidence. International review of
economics & nance,74, 365–372. https://doi.org/10.1016/j.
iref.2021.03.016
International Monetary Fund (IMF) (2021). World Economic
Outlook: Managing Divergent Recoveries. Washington D.C.,
April. Retrieved from https://www.imf.org/-/media/Files/
Publications/WEO/2021/April/English/text.ashx
Jiang, X., & Packer, F. (2017). Credit ratings of domestic and
global agencies: What drives the dierences in China and how
are they priced? BIS Working Paper No. 648. Social Science
Research Network, Rochester, NY. Retrieved from https://ssrn.
com/abstract=2996280
Kang, Q., & Liu, Q. (2007). Credit rating changes and CEO
incentives. Social Science Research Network, Rochester, NY.
Retrieved from https://ssrn.com/abstract=2996280
Johnston, R., Jones, K., & Manley, D. (2018). Confounding and
collinearity in regression analysis: a cautionary tale and an
alternative procedure, illustrated by studies of British voting
behaviour.Quality & quantity,52, 1957-1976. https://doi.
org/10.1007/s11135-017-0584-6
Krichene, A. F., & Khou, W. (2016). On the Nonlinearity of the
Financial Ratios-Credit Ratings Relationship. Applied Finance
and Accounting,2(2), 65-70. https://doi.org/10.11114/afa.
v2i2.1604
Lehmann, B. (2003). Is it worth the while? e relevance of
qualitative information in credit rating [Working Paper].
Social Science Research Network, Rochester, NY. Retrieved
from http://dx.doi.org/10.2139/ssrn.410186
Liberti, J. M., & Petersen, M. A. (2019). Information: Hard and
so.Review of Corporate Finance Studies,8(1), 1–41. https://
doi.org/10.1093/rcfs/cfy009
Liu, X. (2009). Ordinal regression analysis: Fitting the
proportional odds model using Stata, SAS and SPSS. Journal
of Modern Applied Statistical Methods, 8(2), 632-645. https://
doi.org/10.22237/jmasm/1257035340
Löer, G. (2004). An anatomy of rating through the cycle.
Journal of Banking & Finance,28(3), 695–720. https://doi.
org/10.1016/S0378-4266(03)00041-4
Louzada, F., Ara, A., & Fernandes, G. B. (2016). Classication
methods applied to credit scoring: Systematic review
and overall comparison. Surveys in Operations Research
and Management Science,21(2), 117-134. https://doi.
org/10.1016/j.sorms.2016.10.001
Murcia, F. C. S., Murcia, F. D.-R., Rover, S., & Borba, J. A. (2014).
e determinants of credit rating: Brazilian evidence. BAR-
Brazilian Administration Review,11(2), 188-209. https://doi.
org/10.1590/S1807-76922014000200005
Novotná, M. (2013). Multivariate Analysis of Financial and
Market-Based Variables for Bond Rating Prediction. Journal
of Economic Computation and Economic Cybernetics Studies
and Research,47(2), 67–83.
Phan, D. H. B., & Narayan, P. K. (2020). Country responses
and the reaction of the stock market to COVID-19—A
preliminary exposition. Emerging Markets Finance and
Tra de,56(10), 2138-2150. https://doi.org/10.1080/154049
6X.2020.1784719
Dante Domingo Terreno & María Eugenia Donadille
17
Rev. Contab. Finanç. – USP, São Paulo, v. 35, n. 95, e1913, 2024
Sharif, A., Aloui, C., & Yarovaya, L. (2020). COVID-19
pandemic, oil prices, stock market, geopolitical risk and
policy uncertainty nexus in the US economy: Fresh evidence
from the wavelet-based approach.International review
of nancial analysis,70, 1-9. https://doi.org/10.1016/j.
irfa.2020.101496
Shin, K. S., & Han, I. (2001). A case-based approach using
inductive indexing for corporate bond rating. Decision
Support Systems,32(1), 41-52. https://doi.org/10.1016/S0167-
9236(01)00099-9
Soares, G. D. O. G., Coutinho, E. S., & Camargos, M. A.
(2012). Determinantes do rating de crédito de companhias
brasileiras.Contabilidade Vista & Revista,23(3),
109-143. Retrieved from https://www.redalyc.org/
pdf/1970/197026238005.pdf
Tanthanongsakkun, S., & Treepongkaruna, S. (2008). Explaining
Credit Ratings of Australian Companies—An Application of
the Merton Model.Australian Journal of Management,33(2),
261-275. https://doi.org/10.1177/031289620803300203
Wahlen, J. M., Baginski, S. P., & Bradshaw, M. (2014).Financial
reporting, nancial statement analysis and valuation. Cengage
learning.
Williams, R. (2006). Generalized ordered logit/partial
proportional odds models for ordinal dependent
variables. e Stata Journal, 6(1), 58–82. https://doi.
org/10.1177/1536867X0600600104
