Split Ratings and Debt-signaling in Bond Markets- A [email protected]
-
Upload
angelcomputer2 -
Category
Documents
-
view
214 -
download
0
Transcript of Split Ratings and Debt-signaling in Bond Markets- A [email protected]
-
7/25/2019 Split Ratings and Debt-signaling in Bond Markets- A [email protected]
1/6
Split ratings and debt-signaling in bond markets: A note
Ashraf Ismail , Seunghack Oh 1, Nuruzzaman Arsyad 1
Sampoerna School of Business, Building D Mulia Business Park, Jl. Letjen MT. Haryono Kav. 58 60, Jakarta 12780, Indonesia
a b s t r a c ta r t i c l e i n f o
Article history:
Received 17 May 2013
Received in revised form 23 December 2014
Accepted 25 December 2014
Available online 3 January 2015
JEL classication:
C26
C58
G12
G14
F34
F36
Keywords:
Credit ratings
Asymmetric information
Debt signal
Bond markets
Split ratings occur when national and international credit rating agencies assign different ratings to the same
rm. Employing various proxies for asymmetric information and data from advanced and emerging bond mar-
kets, we review the evidence that split ratings are caused by asymmetric information between rms and credit
rating agencies. We then apply the debt-signaling model to the split ratings problem, by testing for a systematic
relationship between the debt-to-equity ratio and the magnitude of split ratings across countries.We nally test
forthe existenceof an optimaldebt-signal, which impliesthat higherdebt-to-equity ratioswill reducethe ratings
split to an optimal minimum, after which accumulating more debt widens the ratings split. Our results suggest
that rms in emerging markets can use thedebt-signal upto a maximal point, after which it becomes inefcient.
2014 Elsevier Inc. All rights reserved.
1. Introduction
Split ratings occur when national and international credit rating
agencies (CRAs) assign different ratings to the same rm (Shin &
Moore, 2003). We focus on the information asymmetry between rms
and CRAs, and examine the mechanism by which CRAs update their
beliefs regarding bond risk in response to information publicly released
by rms. First, we hypothesize that due to asymmetric information,
CRAs possess distinct prior beliefs regarding the quality of rms,
which may be updated in response to new information. To penetrate
market noise, rms attempt to reveal their quality via a debt-signal,
which if successful, will cause the posterior beliefs of CRAs to converge
in such a manner that contracts the ratings split. Second, using several
proxies for asymmetric information, we test for the existence of an op-
timal debt signal, which implies a quadratic relationship between
debt-to-equity ratios and split ratings.
Our primary contribution to the literature is the idea that debt-
signaling can help emerging market rms overcome the more pro-
nounced split ratings margins from which they suffer. We further
contribute to the literature by arguing that the debt-signal has an
upper bound, after which it is detrimental to a rm's credit rating. Our
results are signicant for two reasons. First, if the magnitude of split
ratings is correlated with the degree of asymmetric information, then
higher levels of split ratings will hinder the price discovery process.
This is particularly true in emerging bond markets where information
about rm performance is relatively scarce. Second, if the optimal
debt-signal reduces the cost of price discovery then it could promote
the convergence of bond yields for comparable rms across emerging
and advanced markets. Hence, our research has important implications
for bond market efciency and integration.
In Section 2, we develop our hypothesis regarding the importance of
debt-signaling for less efcient bond markets, and we compare several
proxies for asymmetric information, including the debt-to-equity
ratio, market-based proxies such as the price-to-earnings ratio and the
price-to-book ratio, as well as the standard deviation of forecasted EPS
(which serves as an opinion-based proxy). In Section 3, we measure
the effect of asymmetric information upon nationalinternational split
ratings. We rst analyze our cross-country data set, which consists of
313 rms drawn randomly from 14 countries, to detect evidence of
split ratings. We then employ a step-wise regression method to test
for the relationship between split ratings and debt-signaling. We con-
sider baseline models that control for rm size, industry and country
effects, and proceed sequentially to examine models that contain
Review of Financial Economics 24 (2015) 3641
Corresponding author. Tel.: +62 812 8363 6965.
E-mail addresses:[email protected](A. Ismail),[email protected](S. Oh),
[email protected](N. Arsyad).1 Tel.: +62 812 8363 6965.
http://dx.doi.org/10.1016/j.rfe.2014.12.003
1058-3300/ 2014 Elsevier Inc. All rights reserved.
Contents lists available at ScienceDirect
Review of Financial Economics
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / r f e
http://-/?-http://-/?-http://dx.doi.org/10.1016/j.rfe.2014.12.003http://dx.doi.org/10.1016/j.rfe.2014.12.003http://dx.doi.org/10.1016/j.rfe.2014.12.003mailto:[email protected]:[email protected]:[email protected]://dx.doi.org/10.1016/j.rfe.2014.12.003http://www.sciencedirect.com/science/journal/10583300http://www.elsevier.com/locate/rfehttp://www.elsevier.com/locate/rfehttp://www.sciencedirect.com/science/journal/10583300http://dx.doi.org/10.1016/j.rfe.2014.12.003mailto:[email protected]:[email protected]:[email protected]://dx.doi.org/10.1016/j.rfe.2014.12.003http://-/?-http://-/?-http://crossmark.crossref.org/dialog/?doi=10.1016/j.rfe.2014.12.003&domain=pdf -
7/25/2019 Split Ratings and Debt-signaling in Bond Markets- A [email protected]
2/6
information proxies, including the debt-to-equity ratio, market-based
proxies and an opinion-based proxy, and regress the proxy variables
on the split ratings gap.
To test the debt-signaling hypothesis, we employ a dummy variable
to capture the bloc effect,which measures the impact of incremental
changes in the magnitude of information proxies upon split ratings in
emerging bond markets compared to their advanced counterparts. In
this manner we estimate the effectiveness of each information proxy
as an explanation for split ratings across bond markets. More generally,
this method allows us to distinguish between the explanatory power of
debt-signaling compared to other proxies. To test for the existenceof an
optimal debt-signal, we rst specify an ordered probit model and em-
ploy polynomial regression to detect the hypothesized quadratic rela-
tionship between the debtequity ratio and split ratings, and we then
compare the results with the outcome of the original linear regression.
We nally analyze the results from our interaction terms in the context
of a multiple regression test on the debt-to-equity ratio and other prox-
ies, to determine whether their signs or signicance are altered by the
presence of other independent variables.
InSection 4, we review and interpret the results of our empirical
analysis, with regard to their implications for credit ratings and bond
market efciency. We nd evidence that supports the optimal debt-
signal hypothesis in emerging markets, but not in advanced bond mar-
kets.Consequently, our research explores the relationship between cap-
ital structure, asymmetricinformation, and splitratings, and addresses a
major challenge that emerging economies must overcome in order to
create more efcient and integrated bond markets.
2. Hypothesis development
Credit rating agencies exist in order to mitigate information
asymmetries between investors andrm insiders regarding a rm's val-
uation (Langhor & Langhor, 2008), but this role is often complicated by
the fact that CRAs can assign different ratings to the same rm or bond.
The prevalence of split ratings is displayed in Table 1, which isa sample
of 313 randomly selectedrms collected from ve advanced and nine
emerging economies.Table 1shows that rms in emerging markets
have higher degrees of split ratings than rms in advanced markets.
Ofthe 90rms not assigned split ratings, 81 are in advanced economies
and only nine rms are in emerging markets. At the other end of the
spectrum, only 13 rms in advanced markets are assigned split ratings
of two or more rating levels, while 56 rms in emerging markets aresimilarly rated. These stylized facts support our claim that emerging
marketrms tend to have a higher incidence of split ratings.
Explanations for split ratings range from Ederington (1986) who ar-
gues that split ratings are random errors, toCantor and Packer (1997),
who emphasize the use of different rating models by CRAs, toMorgan
(2002)who hypothesizes that split ratings are correlated with the
opacity of a rm's assets. Each of these arguments is problematic. For
instance, the random error argument is contradicted by evidence that
Moodys systematically assigns lower ratings than S & P ( Livingston,
Naranjo, & Zhou, 2007).Dandapani and Lawrence (2007)found that
one third of all split ratings can be explained by different rating
methods, which means that the majority of split ratings variation can-
not be similarly explained, nor does their approach explain why split
ratings are more prevalent in emerging markets than in advanced
markets. Finally,Livingston et al. (2007) nd that six out of seven vari-
ables used to measure asset opacity are signicant for explaining the
ratings split, but asset opacity is specic to the rm and not the CRA,
and perhaps for this reason,Shin and Moores (2003)test of the asset
opacity hypothesis in Japanese markets is inconclusive.
Our argument rests on the claim that asymmetric information is the
most importantnancial market imperfection.2 We postulate that if all
parties were equally well informed there would be less variation among
CRAs, and the distinction between advanced and emerging markets in
terms of split ratings would disappear.3 But due to asymmetricinforma-
tion, each CRA forms distinct prior beliefs, or conditional probability
distributions regarding the quality ofrms, which are expressed as
a variation in credit ratings. By beliefs we mean that CRAs have prior
probability distributions regarding the value of a specic rm, and in
response to the debt-signal CRAs may update their prior beliefs, gener-
ating aposteriorprobability distribution that is distinct from the prior
distribution. Alternatively, new information may have no effect on
CRA beliefs regarding the value/risk of a specic rm, which means
that the CRA's prior and posterior distributions are identical and so the
rm's credit ratings will remain unchanged. In principle however, CRA
beliefs may continue to evolve in response to new information until a
stationary signaling equilibrium is obtained, which implies that the un-
derlying stochastic process is stable over time, so that CRA beliefs and
credit ratings are consistent and mutually reinforcing.
Debt-signaling has two primary outcomes; pooling and separating.
We use this framework to analyze the adverse selection problem
faced by CRAs in emerging markets, since CRAs cannot always distin-
guish between high quality and low quality rms. We also use thisframework to examine the hypothesis that debt-signaling can help
high quality rms distinguish themselves from low quality rms
(Klein, O'Brien, & Peters, 2002; Leland & Pyle, 1977; Myers & Majluf,
1984; Ross, 1977). A pooling equilibrium illustrates adverse selection,
since pooling implies that CRAs (and investors) assign the same valua-
tion to high quality and low quality rms, thereby under-estimating
the credit worthiness of high qualityrms and over-estimating the
credit worthiness of low quality rms. In a separating equilibrium by
contrast, CRA beliefs regardingrm quality will converge in a manner
that distinguishes between high quality and low quality rms, which
causes a contraction in the ratings split.
We argue that debt-signaling is more likely to yield a separating
equilibrium in emerging markets than in advanced markets. Debt is an
effective signal in noisy markets because it is senior to equity in thecash-ow waterfall, and once equity is exhausted during the bankrupt-
cy work-out process, what remains of the rm's assets reverts to debt-
holders. Debt thereby increases a rm'snancing costs as well as the
likelihood of default if a rm becomes illiquid. The debt-signal is thus
Table 1
Split ratings.
Category Whole samples Advanced economies Emerging economies
Number ofrms with non-split 90 81 9
Number ofrms with split by one level 87 68 19
Number ofrms with split by two levels 67 33 34
Number ofrms with split by more than two levels 69 13 56
Total number ofrms 313 195 118
2 International CRAs provide ratings for a limited number of listed rms in emerging
bond markets and may lack locally specic knowledge, while national CRAs suffer from
ratings criteria that vary widely, so it is not obvious which of these entities is better in-
formed about arm's operations. For this reason, we follow the literature on split ratings
by positing asymmetric information as an explanation for split ratings without specifying
which entity is better informed.3 Shen, Huang, and Hasan (2012)go further to argue that the higher degree of asym-
metric information in emerging markets leads CRAs to adopt different rating methods
across markets.
37A. Ismail et al. / Review of Financial Economics 24 (2015) 3641
-
7/25/2019 Split Ratings and Debt-signaling in Bond Markets- A [email protected]
3/6
incentive compatible for high quality rms in emerging markets be-
cause it is too costly for low quality rms to mimic.
Our distinction between advanced and emerging markets is consis-
tent withKlapper and Love (2004)who nd that variation in the trans-
parency of corporate governance regimes across emerging markets is
correlated with rm valuations.Morck, Yeung, and Yu (2000) nd
that weak property rights protection in emerging markets makes in-
vestment arbitrage opportunities less attractive; and in the context of
bond ratings, this distinction is also supported by Bhojraj andSengupta (2003), whond that disclosure quality and bond yields are
negatively correlated. In advanced bond markets, corporate governance
regimes reveal higher quality information aboutrms and the legal sys-
tems protect investor rights more effectively, thereby neutralizing the
efcacy the debt signal.
We nally hypothesize the existence of an optimal debt signal. Opti-
mal debt theory posits that raising a rm's debt-to-equity ratio has the
benet of increasing the dollar value of its interest tax shield, but also
increases a rm's bankruptcy costs (Miller, 1977; Myers, 1984).The op-
timal debt level is dened as the point at which the present value of the
benet from increasing the debt-to-equity ratio equals the present
value of its cost. By extension, we posit the existence of an optimal
debt signal, which we dene as the debt level that minimizes the mag-
nitude of the ratings split. Optimality also implies that increasing a
rm's debt level beyond the optimal point should cause the ratings
split to widen, since doing so increases bankruptcy costs beyond any
value added to the rm.
Our expectations are that debt-signaling will help explain the varia-
tion of split ratings across countries, but it will be less effective in
explaining split ratings in the more efcient bond markets of advanced
economies. In what follows, we test our hypothesis that the debt-to-
equity ratio is the most effective proxy for explaining the difference in
the ratings split between advanced and emerging markets, and we also
test the idea that the relationship between split ratings and debt-
signaling is quadratic,as suggested by the optimal debt-signal hypothesis.
3. Data and methodology
In order to estimate the differential impact of asymmetric informa-tion upon the credit rating process, we must rst select an appropriate
proxy measure. In this regard,Clarke and Shastri (2000)review three
general measures of asymmetric information: 1) opinion-based proxies
derived from analyst forecasts, 2) proxies based upon nancial state-
ment ratios, such as price-to-book or debt ratios; 3) microstructure
proxies usually involving a decomposition of the bidask spread; and
beyond Clarke and Shastri, 4) the literature on asymmetric information
also relies upon the volatility of returns or trading volume, or some
permutation thereof, as measures of asymmetric information (Bates,
Conghenour, & Shastri, 1999).
Each of these proxies has weaknesses, such as the inherent subjec-
tivity of analysts' estimates, and the lack of a consensus regarding how
to disaggregate the relevant components of the bidask spread, as
well as the potential for measurement error in a rm'snancial ratios.However, proxies based upon nancial ratios also have a numberof dis-
tinct benets. First, nancial ratios are widely available and are open to
critical evaluation. Second, international accounting standards providea
unied framework in which to interpret the precise meaning and impli-
cations ofnancial ratios. Third, when we control for rm size, we can
easily test whether nancial ratios are correlated with the magnitude
of split ratings. For all these reasons, nancial ratios provide effective
proxy measures of asymmetric information that can be applied across
industries and countries.
3.1. Data sources
Table 2 summarizes our sampledistribution by country andindustry
(nancial sector vs. non-nancial sector). Our data set consists of
corporate ratings for 313 rms drawn fromve advanced markets and
nine emerging bond markets, including Canada, Japan, South Korea,
Taiwan, the United States, India, Pakistan, Indonesia, Kazakhstan,
Malaysia, Thailand, the Philippines, South Africa, and Sri Lanka. Of the
total rms in ourdataset,195 (62%)are drawn from advanced markets,
while 118 are drawn from emerging markets. Further, 133 rms (42%)
are nancial rms and the remainder are non-nancial rms. Each
rm in our sample is rated by both national and international credit
rating agencies.4 International credit ratings were drawn from Fitch,
Moody's, and Standard and Poor's, and were collected from their
website as of March 2012. National credit ratings were drawn from sev-
eral national credit rating agencies websites as of March 2012. Firm-
level data was collected from Datastream, and onlyrms listed in equity
markets were included.
3.2. Variable denitions
A split rating is operationally dened by the difference in the letter-
level between national and international CRAs, where the dummy
variable is assigned 0if there is no split rating, 1if the difference
between the ratings is one letter-level (for example AA and A), 2if
the difference is two letter-levels (for example between AA and BBB),
and3if the difference is more than two letter levels (for example be-
tween AA and BB or B). Because the split ratings variable is ordinal, we
use an ordered probit model to analyze the data.
Table 3displays the nancial ratio proxies and the opinion based
proxy for asymmetric information. Financial ratios include the price-
to-book ratio (PBR), the price-to-equity ratio (PER), and the debt-to-
equity ratio (DER), which are widely used in the literature to measure
the difference between market information and accounting information
associated with a specicrm.Table 3shows that at least for PBR, PER,and DER our sample is generally representative of the population of
rms in each economy.
Table 2
Sample distribution.
Country Financial Non-nancial Total
Canada 19 10 29
Japan 31 49 80
South Korea 12 25 37
Taiwan 0 9 9
United States 5 35 40
Advanced 67 (34%) 128 (66%) 195 (62%)
India 19 18 37
Indonesia 11 16 27
Kazakhstan 4 0 4
Malaysia 9 6 15
Pakistan 6 0 6
Philippines 1 1 2
South Africa 3 3 6
Sri Lanka 6 0 6
Thailand 7 8 15
Emerging 66 (56%) 52 (44%) 118 (38%)
Total 133 (42%) 180 (58%) 313
4 DBRS forCanadian rms (http://www.dbrs.com), Japan CreditRatingAgencyand R&I
rating agency for Japanese rms (http://www.jcr.co.jp/english/andhttp://www.r-i.co.jp/
eng/), KIS and NICE Rating Agency for South Korean rms (http://www.kisrating.com/
eng/ratings/hot_disclosure.asp and http://eng.nicerating.com), Taiwan Rating for
Taiwanese rms (http://www.taiwanratings.com), the EganJones rating for US rms
(www.egan-jones.com), CRISIL (http://crisil.com/ratings/credit-ratings-list.jsp), ICRA
(http://www.icra.in/CurrentRating.aspx), and CARE (http://www.careratings.com) for
Indian rms, Pendo Credit Rating Agency for Indonesian rms (http://new.pendo.
com/index.php),RFCAfor Kazakhstan rm (www.rfcaratings.kz), RAM credit rating for
Malaysian rms (http://www.ram.com.my), JCR-VIS for Pakistani rms (http://jcrvis.
com.pk/), PRSfor Philippine rms (www.philratings.com.ph), GCRfor South African rms
(http://globalratings.net/), RAM Ratings (Lanka) for Sri Lankanrms (www.ram.com.lk)
TRIS rating agency for Thairms (http://www.trisrating.com).
38 A. Ismail et al. / Review of Financial Economics 24 (2015) 3641
http://www.dbrs.com/http://www.jcr.co.jp/english/http://www.r-i.co.jp/eng/http://www.r-i.co.jp/eng/http://www.kisrating.com/eng/ratings/hot_disclosure.asphttp://www.kisrating.com/eng/ratings/hot_disclosure.asphttp://eng.nicerating.com/http://www.taiwanratings.com/http://www.egan-jones.com/http://crisil.com/ratings/credit-ratings-list.jsphttp://www.icra.in/CurrentRating.aspxhttp://www.careratings.com/http://new.pefindo.com/index.php),RFCAhttp://new.pefindo.com/index.php),RFCAhttp://new.pefindo.com/index.php),RFCAhttp://new.pefindo.com/index.php),RFCAhttp://www.rfcaratings.kz/http://www.ram.com.my/http://jcrvis.com.pk/http://jcrvis.com.pk/http://www.philratings.com.ph/http://globalratings.net/http://www.ram.com.lk/http://www.trisrating.com/http://www.trisrating.com/http://www.ram.com.lk/http://globalratings.net/http://www.philratings.com.ph/http://jcrvis.com.pk/http://jcrvis.com.pk/http://www.ram.com.my/http://www.rfcaratings.kz/http://new.pefindo.com/index.php),RFCAhttp://new.pefindo.com/index.php),RFCAhttp://www.careratings.com/http://www.icra.in/CurrentRating.aspxhttp://crisil.com/ratings/credit-ratings-list.jsphttp://www.egan-jones.com/http://www.taiwanratings.com/http://eng.nicerating.com/http://www.kisrating.com/eng/ratings/hot_disclosure.asphttp://www.kisrating.com/eng/ratings/hot_disclosure.asphttp://www.r-i.co.jp/eng/http://www.r-i.co.jp/eng/http://www.jcr.co.jp/english/http://www.dbrs.com/ -
7/25/2019 Split Ratings and Debt-signaling in Bond Markets- A [email protected]
4/6
Withregard to opinion-based proxies, we usethe standard deviation
of the forecasted EPS (earnings per share). The standard deviation of
EPS is useful because the higher the degree of asymmetric information,
the more likely analysts and investors will differ in their assessments of
arm's future earnings, which translates into a higher standard devia-
tion of forecasted EPS. We eliminate all rms from our sample that are
monitored by only one equity analyst, because this would result in
zero standard deviation of forecasted EPS. Therefore, in regressions
that include the standard deviation of forecasted EPS as an explanatory
variable, our sample is reduced to 247 rms.
In order to test thedebt-signaling hypothesis, we measure the corre-
lation between DER and the magnitude of a rm's ratings split. Since
capital structure is considered an important measure of a rm's future
protability, DER should serve as an effective signal to the market
up to an upper bound, and after that, increasing a rm's debt level is
interpreted negatively by the market, since it increases bankruptcycosts beyond any added benet. To capture the optimal debt-signaling
idea we also include DER^2 as a variable, which is used to test for the
possibility of a quadratic relationship between DER and split ratings.
We also employ several control variables, such as rm size and
dummy-nance, that control for the industry effect arising from the
more opaque balance sheets ofnancial rms, and dummy-countries,
which are included to reduce any bias due to possible differences in
the ratings methodologies of national CRAs. We control for rm size be-
cause large rms areoften under more scrutiny from themarket,thus in-
creasing the informationow to investors. Finally, we use Japan as the
baseline for the dummy-country variables, and we categorize Canada,
Japan, South Korea, Taiwan and the United States as advanced economies
due to their high levels of per capita income, while the remaining coun-
tries in our sample are categorized as emerging economies.
3.3. Methodology
To test our hypothesis, we compare the results of the ordered
probitmodel with theresults of an Ordinary Least Squares (OLS) regres-
sion. Consequently, we use both OLS and the probit model when we
compare split ratings between advance and emerging economies,
which allows us to test the optimal debt-signal hypothesis more thor-
oughly. Using the emerging markets dummy variable (DEM), we divide
our samples into advanced market and emerging market sub-samples.
We further include interaction terms that consist of the emerging mar-
ket dummy variable multiplied by our independent variables, which we
use to distinguish between the effects of independent variables upon
our two sub-samples. If no signicant difference is discovered between
the two sub-samples, this would imply that asymmetric information
has a similar effect upon split ratings in both advanced and emerging
bond markets. However, if an interaction term is signicant,that implies
that the corresponding information proxy is signicant for one sub-
sample and not the other, and our results can thereby provide an effec-
tive measure of the effect of asymmetric information upon split ratings
across bond markets.
We specify our ordinal probit model as follows:
RATi 0 1EXPi 2EXPi DEMi 3CONi
(i) RATi denotes the magnitude of the ratings split, where RAT
equals 0if there are no split ratings, 1if the ratings are split
by one letter-level, 2 if the ratings are split by two letter-
levels, and 3if the ratings are split by more than two letter-
levels.
(ii) EXPidenotes the set of independent variables. Independent var-
iables are price-to-book ratio (PBR), price-to-earnings ratio
(PER), the debt-to-equity ratio (DER), debt-to-equity ratio-
squared (DER^2), and the standard deviation of forecasted EPS
(SDF).
(iii) DEMi is the dummy variable for emerging markets, where
DEM = 1 denotes rms from emerging markets, and DEM = 0
denotes advanced markets.
(iv) EXPi DEMiis the interaction term between explanatory vari-
ables and the emerging market dummy.
(v) CONidenotes the set of control variables. Control variables used
for this model are (i) the log of totalassets to capture thesize ef-
fect, (ii) dummy-nance to capture the difference between -nancial and non-nancial rms, and (iii) country-dummy
variables to reduce the biascaused by different rating methodol-
ogies used by national rating agencies.5
Finally, to ensure the integrity of our results, we calculate the corre-
lation coefcients for our independent variables, which are displayed in
Table 4. We nd no notable correlation coefcient in this matrix; and
hence multicollinearity is not a problem for our methodological ap-
proach. However, to further avoid the multicollinearity problem, we
use a step-wise regression procedure that allows us to evaluate the ex-
planatory power of each independent variable.
Table 3
Population and sample comparison (standard deviations are in parentheses).
Country Average PBR Average PER Average DER
Population Sample Population Sample Population Sample
Canada 1.90 1.47 (1.05) 16.24 13.57 (11.14) 0.85 1.13 (0.94)
Japan 0.90 0.98 (0.45) 21.90 28.20 (52.00) 1.81 1.75 (2.22)
South Korea 1.10 1.18 (0.94) 9.10 17.74 (15.82) 1.20 1.76 (1.77)
Taiwan 1.82 1.31 (1.17) 15.27 27.83 (31.19) 0.36 0.49 (0.39)
United States 2.70 3.62 (5.75) 17.40 35.66 (75.62) 2.12 0.84 (1.03)
Advanced 2.19 1.99 (4.37) 17.59 26.28 (52.24) 1.89 1.52 (1.98)
India 2.00 1.84 (1.63) 11.70 12.07 (7.31) 1.30 1.82 (1.68)
Indonesia 3.10 1.73 (0.87) 13.70 18.40 (21.89) 0.81 0.86 (0.63)
Kazakhstan 0.52 3.03 (2.60) 5.50 5.00 (7.12) 1.49 1.33 (1.21)
Malaysia 1.80 2.13 (0.83) 13.80 17.51 (11.93) 0.95 1.17 (0.84)
Pakistan 2.36 0.92 (0.44) 9.91 4.23 (2.24) 2.71 1.57 (1.07)
Philippines 2.87 2.20 (1.27) 17.37 11.90 (4.81) 0.83 0.59 (0.60)
South Africa 1.26 2.05 (1.12) 17.50 13.43 (2.03) 0.40 0.22 (0.25)
Sri Lanka 1.67 1.33 (0.64) 10.74 14.38 (16.46) 0.83 1.51 (1.34)
Thailand 2.40 2.41 (2.48) 12.40 11.80 (3.55) 1.37 1.69 (1.29)
Emerging 2.17 1.91 (1.49) 12.74 13.72 (13.06) 1.16 1.35 (1.28)
Note: Average populations for groups (advanced and emerging) is calculated as a weighted average, with GNI as the weight. Data for populationswas gathered from various sources in-
cluding IMF Financial Statistics, Bloomberg, the Tokyo Stock Exchange, the Japanese Ministry of Finance, and the Thailand Stock Exchange.
5
Seethe footnotesfor Tables 5 and 6 foran explanationof therole of country dummies.
39A. Ismail et al. / Review of Financial Economics 24 (2015) 3641
-
7/25/2019 Split Ratings and Debt-signaling in Bond Markets- A [email protected]
5/6
4. Results and discussion
Our results are organized in Tables5 and 6. Table 5 shows the results
of the ordered probit test for our information proxies regressed upon
our sample's split ratings, andTable 6displays the results of the probit
and OLS tests for the optimal debt-signal. Overall, our results support
the hypothesis that debt-signaling is effective for emerging market
rms, but less so for rms in advanced bond markets. Our results also
provide evidence for optimal debt-signaling in emerging bond markets.
In order to more thoroughly test our hypothesis we employ a step-
wise ordered probit regression method. As can be seen in Table 5,the
independent variables (PBR, PER, DER, and the standard deviation of
forecasted EPS) are included in separate regression models. We start
with a baseline model that consists only of control variables (column
1) and successively add independent variables in a step-wise fashion
in order to discriminate between the effects of each independent vari-
able. Our analysis is divided into models that contain no interaction
terms and models that contain interaction terms, which captures the
difference between the independent variables' explanatory power for
advanced and emerging bond markets respectively.
4.1. Split ratings and asymmetric information
Columns 2 and 3 in Table 5show that PBR is only signicant for
emerging markets, since PBR is not signicant by itself, but the
PBR DEM interaction term is signicant at the 10% level. Accordingto columns 4 and 5, PER is not signicant for either the advanced or
emerging economies. By contrast, columns 6 and 7 support the debt-
signaling hypothesis, since DER is not signicant for the entire sample,
while the DER DEM interaction term is signicant at the 1% level,
which implies that DER is signicant only for emerging markets. The
negative sign of the interaction term also meets our expectations,
since a higher DER is associated with a lower ratings split in emerging
markets. Columns 8 and 9 show that the standard deviation of forecast-
ed EPS is not signicant for explaining split ratings, in either the whole
sample or in emerging markets. Most importantly however, the multi-ple regression test displayed in column 10 afrms the negative sign of
DER at the 1% signicance level for emerging markets. Consequently,
only DER is signicant at the 1% level foremerging markets, which sup-
ports the debt-signaling hypothesis.
4.2. The optimal debt-signal
Table 6shows the results of the ordered probit and OLS regression
tests for the optimal debt-signal hypothesis. Column 1 ofTable 6dis-
plays the results of the ordered probit test for the entire sample using
DER and DER^2, which reveals that neither variable is statistically
signicant. The results presented in column 1 may be biased towards
rms in advanced markets, and for this reason the next regression
(presented in column 2) includes the DER DEM and DER^2 DEM
interaction terms. Their inclusion allows us to distinguish between the
effect of debt-signaling upon emerging and advanced bond markets.
Note that DER DEM is signicant at the 1% level and its coefcient
has a negative sign, while DER^2 DEM is signicant at the 5% level
and its coefcient has a positive sign. The signs and signicance of the
interaction terms indicate that the relationship between DER and the
split ratings in our sample is quadratic (U-shaped), which thereby pro-
vides evidence for optimal debt-signaling. Our hypothesis is also sup-
ported by the multivariate ordered probit test in column 3, that
includes all independent variables as well as the interaction terms,
whose signs and signicance are therein conrmed.
The results of the OLS test inTable 6provide further evidence for
the optimal debt-signal. DER and DER^2 are signicant for the entire
sample, with the expected U-shape curvature. Note that the OLS test
also yields a negative sign for DER*DEM at the 5% signicance level,while the coefcient for DER^2 DEM is positive, which again suggests
Table 4
Correlation matrix.
Log
(rm size)
PBR PER DER Std. deviation of
forecasted EPS
(SDF)
Log (rm size) 1 0.151 0.069 0.004 0.153
PBR 1 0.044 0.149 0.040
PER 1 0.039 0.043
DER 1 0.060
Std. deviation offorecasted EPS (SDF)
1
Table 5
Ordered probit regression: information asymmetry and split ratings.
Variables 1 2 3 4 5 6 7 8 9 10
Log (rm size) 0.053 0.055 0.058 0.051 0.049 0.000 0.003 0.104T 0.225** 0.049
(0.039) (0.039) (0.048) (0.039) (0.049) (0.029) (0.031) (0.059) (0.039) (0.048)
D. Finance 0.398*
(0.158)
0.408*
(0.161)
0.391*
(0.151)
0.391*
(0.158)
0.411**
(0.148)
0.292
(0.151)
0.285
(0.148)
0.434*
(0.182)
0.220
(0.148)
0.418**
(0.158)
PBR 0.010
(0.032)
0.041
(0.031)
0.065
(0.247)
PBR DEM 0.175
(0.095)
0.168
(0.099)
PER 0.001
(0.001)
0.001
(0.001)PER DEM 0.007
(0.014)
DER 0.011
(0.038)
0.065
(0.057)
0.085
(0.054)
DER DEM 0.358**
(0.122)
0.361**
(0.120)
Std. dev. of forecast (SDF) 0.000
(0.000)
0.000
(0.000)
SDF DEM 0.014
(0.019)
Pseudo R2 0.29 0.29 0.29 0.29 0.29 0.28 0.30 0.29 0.22 0.31
N 313 313 313 313 313 313 313 247 247 313
for 10%, * for 5%, ** for 1%. Standard errors are in parentheses.
Note: We consider only letter rating, not the negative or positive sign (for example: AA+, AA, and AAare considered equal). Split rating by one level represents one letter difference
(for example AA and A), split rating by two levels represents two letter differences (for example AA and BBB). We also include country dummy variables in each model, with Japan as
the baseline, to capture the effect of national rating methodologies. The coefcients for country dummies are not reported for the sake of brevity.
40 A. Ismail et al. / Review of Financial Economics 24 (2015) 3641
-
7/25/2019 Split Ratings and Debt-signaling in Bond Markets- A [email protected]
6/6
a U-shaped relationship between DER and the split ratings dependent
variable. Thus, while DER and DER^2 are not signicant for Canada,
Japan, Korea, Taiwan and the United States, the optimal debt signal is
signicant for emerging markets. Empirically, these results imply that
increasing the DER reduces split ratings to a minimum, after which ac-
cumulating more debt will exacerbate the ratings split.
Based on our results, we can conclude that (1) the debt-to-equity
ratio is an effective proxy for asymmetric information between rms
and CRAs; (2) in advanced bond markets, asymmetric information is
relatively low, so the debt-signal is not effective in reducing the ratingssplit; (3) in emerging markets, information about rm quality is more
scarce, and so the debt-signal has a measurable effect upon the ratings
split;and nally, (4) the resultsof the probit and OLStests imply theex-
istence of an optimal debt-signal in emerging markets that can mini-
mize the ratings split between national and international CRAs.
5. Conclusion
Because DER is only signicant for emerging markets, our research
suggests that in emerging bond markets, the debt-signal can cause
CRA beliefs to converge so as to diminish the ratings split. However,
this result is not replicated in advanced bond markets. Our research fur-
ther suggests that rms can adopt an optimal capital structure that di-minishes the effect of asymmetric information on the price discovery
process for their respective bonds. Our results also provide avenues
for future research. For instance, if rms in emerging markets can
adopt an optimal debt-signal, this should not only minimize the ratings
split, but should also have some measurable impact upon bond yields. If
the optimal debt-signal is indeed effective in mitigating asymmetric in-
formation in emerging markets, then it should narrow the difference
between bond yields for comparable rms across bond markets. In
this manner, our research can contribute more fundamentally to the
xed-income literature in the future.
References
Bates, T., Coughenour, T. J., & Shastri, K. (1999). Spinoffs, spreads, and information asym-metry.Working paper. University of Pittsburgh.
Bhojraj, S., & Sengupta, P. (2003).The effect of corporate governance mechanisms onbond. Ratings and yields: The role of institutional investors and outside directors.
Journal of Business,76(No. 3), 455475.Cantor, R.,& Packer, F. (1997). Differencesof opinion andselectionbias in thecreditrating
industry. Journal of Banking and Finance,21(No. 10), 13951417.Clarke, J., & Shastri, K. (2000). On information asymmetry metrics.Working paper, Katz
Graduate School of Business. University of Pittsburgh.Dandapani, K., & Lawrence, E. R. (2007). Examining split bond ratings: Effect of rating
scale.Quarterly Journal of Business & Economics,46(2), 6582.Ederington, L.H. (1986).Why split ratings occur. Financial Management,15(1), 3747.Klapper, L.F., & Love, I. (November2004). Corporate governance,investor protectionand per-
formance in emerging markets. Journal of Corporate Finance,vol. 10 (5). (pp. 703728).Elsevier, 703728.
Klein, L. S., O'Brien, T.J., & Peters, R. S. (2002). Debt vs. Equity and asymmetric informa-tion: A review.The Financial Review,37, 317350.
Langhor, H.M., & Langhor, P. T. (2008).The rating agencies and their credit ratings. NewYork: John Wiley & Sons, 111.
Leland,H., & Pyle,D. (1977).Informational asymmetries, nancial structure, andnancialintermediation. Journal of Finance,32, 371387.
Livingston, M., Naranjo, A., & Zhou, L. (2007).Asset opaqueness and split bond. FinancialManagement,36(3), 4962.
Miller, M.H. (1977).Debt and taxes.The Journal of Finance,32(2), 261275.Morck, R., Yeung, B.Y., & Yu, W. (2000). The information content of stock markets: Why
do emerging markets have synchronous stock price movements? Journal ofFinancial Economics,59, 215269.
Morgan, D.P. (2002).Rating banks: Risk and uncertainty in an opaque industry. The
American Economic Review,92(4), 874
888.Myers, S.C. (July, 1984).The capital structure puzzle.Papers and proceedings, forty-second
annualmeeting. The Journalof Finance,vol. 39, No. 3. (pp. 575592). American FinanceAssociation.
Myers, S.C., & Majluf, N.S. (1984). Corporate nance and investment decisions whenrmshave information that investors do not have. Journal of Financial Economics, 13 ,187221.
Ross, S.A. (1977).The determination ofnancial structure: The incentive signaling ap-proach.Bell Journal of Economics,8, 2340.
Shen, C.H., Huang, Y.L., & Hasan, I. (2012).Asymmetric benchmarking in bank credit rat-ings. Journal of International Financial Markets Institutions and Money,22, 181193.
Shin, Y.S., & Moore, W.T. (2003). Explaining credit ratings differences between Japaneseand US agencies.Review of Financial Economics,12, 327344.
Table 6
Ordered probit and OLS regressions: Debt-signaling and split ratings.
Ordered probit
Y = 0 if there is no split
=1 if split by 1 letter
=2 if split by 2 letters
=3 if split by N 2 letters
OLS
Y = national rating international rating
1 2 3 4 5 6
Log (rm size) 0.053 0.062 0.139* 0.296* 0.297* 0.283*
(0.039) (0.040) (0.068) (0.119) (0.124) (0.125)
D.nance 0.385* 0.381* 0.448** 0.566 0.480 0.432
(0.161) (0.183) (0.171) (0.328) (0.334) (0.348)
DER 0.056 0.152 0.069 0.546* 0.135 0.168
(0.083) (0.099) (0.130) (0.225) (0.215) (0.215)
DER^2 0.006 0.010 0.003 0.067* 0.026 0.030
(0.008) (0.008) (0.017) (0.029) (0.031) (0.031)
DER DEM 1.135** 1.159** 1.460* 1.507**
(0.312) (0.357) (0.570) (0.578)
DER^2 DEM 0.155* 0.163* 0.186* 0.192*
(0.067) (0.069) (0.079) (0.079)
PBR 0.129 0.144
(0.117) (0.093)
PBR DEM 0.205 0.260
(0.150) (0.173)
Pseudo R2/adjusted R2 0.29 0.31 0.34 0.57 0.59 0.60
N 313 313 313 313 313 313
for 10%, * for 5% and ** for 1%. Standard errors are in parentheses.
Note: We consider only letter rating, not the negative or positive sign (for example: AA+, AA, and AAare considered equal). split rating by one level represents one letter difference
(for example AA and a), split rating by two levels represents two letter differences (for example AA and BBB). We also include country dummy variables in each model, with Japan as
the base, to capture the effect of national rating methodologies. The coefcients for country dummies are not reported for the sake of brevity.
41A. Ismail et al. / Review of Financial Economics 24 (2015) 3641
http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0090http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0090http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0090http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0090http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0100http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0100http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0100http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0100http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0020http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0020http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0020http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0020http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0020http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0020http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0020http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0120http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0120http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0120http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0055http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0055http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0055http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0055http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0055http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0055http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0055http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0085http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0080http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0075http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0070http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0125http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0060http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0050http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0055http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0045http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0040http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0120http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0120http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0115http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0110http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0020http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0105http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0100http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0100http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf9995http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0095http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0090http://refhub.elsevier.com/S1058-3300(14)00052-4/rf0090