6. Sistemas de almacenamiento de energía
6.2. Ultracapacitores y sus aplicaciones en el mercado mundial
As explained above, the objective of this chapter is to study factors of stickiness variation over time. Therefore, dependent variables are the upgrade and downgrade thresholds (resp. α1 and α2)
and explanatory variables are reputation proxies and the business cycle proxy.
Models of reputation argue that the reputation of the CRA affect her behavior and also can be affected by the CRA’s behavior. Thus, there is an endogeneity problem due to simultaneous causality between reputation proxies and dependent variables.
Therefore, we rely on the paper of Keele and De Boef (2004) to choose the adequate model for a stationary data that suffers simultaneous causality between explanatory variables and dependent variables. Keele and De Boef (2004) prove that a single-equation Error Correction Model take into account both cointegrated and long memorized data. They provide significant evidence that the single-equation Error Correction Model is the best estimation model if we use stationary data.
Thus, we estimate a single-equation Error Correction Model in order to take into account both short term and long-term effects on the dependent variable and deal with cointegration problems.
∆α1t= ω0+ ω1∆GDPt+ ω2∗ ∆M ktCapt+ ω3∗ ∆CAARU P Gt−
ω4[α1(t−1)− ω5∗ GDPt−1− ω6∗ M ktCapt−1− ω7∗ CAARU P Gt−1] + ε (3.5)
∆α2t= ω0+ ω1∆GDPt+ ω2 ∗ ∆M ktCapt+ ω3∗ ∆CaarDt−
ω4[α2(t−1)− ω5∗ GDPt−1− ω6∗ M ktCapt−1− ω7∗ CaarDt−1] + ε (3.6)
Differentiated variables reflect the immediate and short term effect of a variable and the lagged variable captures to the long term effect. ω4 has to be negative and contained between 0 and -1 to
note a long term effect. Otherwise, if ω4is lower than −1 or neither ω5, nor ω6nor ω7are significant
we conclude that there is no long term effect and then no long term relationship between dependent and independent variables.
Dependent variables of the model are, α1t(upgradethresholdst) and α2t(downgradethresholdst),
the estimated thresholds of stickiness respectively for upgrades and downgrades estimated by fric- tion model during the quarter t. Explanatory variables are as follows. First, the quarterly Gross Domestic Product growth rate of USA14represents the business cycle proxy. We assume that boom
phase is presented by a positive quarterly Gross Domestic Product growth rate, while recession phase is presented by a negative quarterly Gross Domestic Product growth rate. Therefore, if the CRA ’s rating criteria is pro-cyclical we should find an increase in both downgrade and upgrade thresholds during boom phases and a decrease in both downgrade and upgrade thresholds in down- turns. Thus, the coefficient of this variable is expected to be positive in the downgrade threshold regression and negative in upgrade thresholds.
Second, as suggested Benabou and Laroque (1992) and Duff and Einig (2009), the CRA’s re- putation is the belief of investors that the CRA’s rating is accurate. Thus, we retain the Cumulative Abnormal Average Return (CAAR) of rated firms as a proxy for stock market reaction to rating changes and therefore the CRA’s reputation. Then, Caarupgt, CaarDtrepresents Cumulative Ave-
rage Abnormal Return of the window event [−1, +1] through the quarter t and across firms who experienced rating upgrades, downgrades, during this quarter. We expect the CRA to take into ac- count previous investor’s perception in order to improve the ratings standards or not. The CRA should change rating standards whenever Caarupg is close to zero or negative or CaarD is close to zero or positive. Thus, we should find a negative coefficient for both variables. Finally, the va- riable M ktCap represents the quarterly Market Capitalization return of McGraw-Hill company.
14http ://www.bea.gov/national/index.htm#gdp
It’s a proxy for S&P market capitalization because the financial services segment which comprises S&P’s rating business contributes to 39% of McGraw-Hill’s revenue. We use this variable as a proxy for the CRA’s reputation following Lôffler (2011). Indeed, the author finds that the stock price of Moody’s and McGrawHill reacts negatively to the default of an issuer in the investment grade. The reaction of the market proves that CRAs’ market capitalization could be used as a proxy for their reputation.
6
Results
Assuming that investors do not perceive the same way a default of a firm in the speculative grades or in the investment grades. We first conduct the study over all grades (both investment and speculative grades). Then, we split the database into two groups (investment grades and speculative grades) in order to detect whether the CRA adopts different credit rating standards over grades. We present below the results of the model that study factors of stickiness relative to all grades, investment grades and speculative grades
Table (3.2) presents results of the study of factors of stickiness variation over time. Panel A reports results of the Error Correction Model for the dependent variable upgrade thresholds (α1).
Panel B reports results of the Error Correction Model for the dependent variable downgrade thre- sholds (α2). We first explain that coefficients of differenced explanatory variables represent the
immediate effect of these variables on the dependent variable. For example in Panel A, the coef- ficient of the explanatory variable ∆GDP is equal to 4.4640 suggests that an increase of 1% in the GDP decreases the absolute value of α1 (upgradethresholds) by 0.04464. Second, coeffi-
cients of lagged explanatory variables represents the long term effect and is distributed over future 133
TABLE3.2 – Empirical results of factors of stickiness variation over time
Empirical results of the Error Correction Model for both upgrade and downgrade thresholds and for each group of the study (All grades, investment grades and speculative grades)
All grades Investment grade Speculative grade Panel A :Dependent Variable :α1(upgradethresholds)
Sample (adjusted) : 1996Q2 2011Q4
Variable Coefficient t-Stat Coefficient t-Stat Coefficient t-Stat ω0 -0.1301 -5.4738 *** -0.0100 -7.3815*** -0.1944 -5.943526*** ∆GDP 4.4640 2.6989 *** 0.0403 0.3851 5.9877 2.374567** ∆M KT Cap -0.1783 -1.9899 ** -0.0001 -0.0181 -0.3173 -2.380531** ∆Caarupg 0.2195 0.6850 -0.0525 -3.2648*** 0.1005 0.328887 α1(−1) -0.7355 -6.0288 *** -0.9920 -9.1243*** -0.9150 -7.006985*** GDP (−1) 8.7581 3.3992 ** 0.2495 2.2162 ** 7.2356 2.376803** M ktCap(−1) -0.2595 -1.4315 -0.0163 -2.0600** -0.2863 -1.320353 Caarupg(−1) 0.6248 1.0058 -0.0329 -1.4219 0.3274 0.705806 R-squ Adj 0.3833 0.6426 0.4786 F-statistic 6.5943 17.1809 9.2607 Long run equilibrium α1 = −0.17684 α1 = −0.01 α1 = −0.2124
+8.7581GDP +0.2494GDP +7.2355GDP −0.0162M ktCap
Panel B :Dependent Variable :α2(downgrade thresholds) Sample (adjusted) : 1996Q2 2011Q4
Variable Coefficient t-Stat Coefficient t-Stat Coefficient t-Stat ω0 0.0947 3.9354 *** 0.0119 6.492157 0.1414 4.8393 *** ∆GDP -3.1320 -1.9543** -0.1465 -1.085014 -3.4614 -1.7588 * ∆M KT Cap 0.1432 1.7180 * 0.0027 0.392073 0.2283 2.2049 ** ∆CaarD -0.0057 -0.0167 0.0275 0.724121 0.3355 1.4361 α2(−1) -0.6573 -5.3311*** -1.0341 -9.506998 -0.8136 -6.2177 *** GDP (−1) -6.6031 -2.4236** -0.3615 -2.285762 -4.3494 -1.6553 M ktCap(−1) 0.3602 1.9782 ** 0.0205 2.164275 0.3304 1.7851 CaarD(−1) 0.0581 0.0785 0.0986 1.73725 0.2908 0.7200 R-squ Adj 0.3077 0.5974 0.4265 F-statistic 5.0003 14.3549 7.6941 Long-run equilibrium α2 = 0.1440 α2 = 0.011 α2 = 0.1737 −6.603GDP +0.3602M ktCap ( 1%***. 5%** and 10%*)
time periods according to the coefficient of error correction, the coefficient of the lagged dependent variable. For example in panel A, the coefficient of the lagged explanatory variable, GDP−1, is
equal to 8.7581 suggests that an increase of 1% in the GDP decreases the absolute value of α1
(upgradethresholds) spreads over future time periods at a rate of 73.55% per time period i.e. the absolute value of α1 decreases by (8.7581 ∗ 0.7355 ∗ 1% = 0.06441) at the first year, then another
0.0473 at the second year, until the change in the GDP has no effect on the dependent variable. 134
Finally, the long run equilibrium equation summarizes the long term effect of explanatory variables on the dependent variable. For example in panel A, the α1(upgradethresholds) for all grades has a
constant equal to −0.1768 which suggests that the probability has to exceed a threshold of 17.68% in order to observe a rating upgrade. Besides, the coefficient of the GDP suggests that an increase in the GDP decreases the absolute value of the upgradethresholds.
Table (3.2) shows that the GDP has an immediate impact on both thresholds in panels A and B. Moreover, GDP has a long term equilibrium relation with the dependent variable. The GDP decreases the absolute value of the upgrade threshold (α1 < 0) and decreases the downgrade
threshold (α2 > 0). Hence, a recession increases the absolute value of upgrade thresholds and
downgrade thresholds. Then, the CRA is more likely to stick to its rating during recession.This finding is in contrast with those of previous studies (Nickell et al. (2000) and Amato and Furfine (2004)) who find that the CRA is more likely to downgrade ratings in a recession period. However, this finding is in line with the study of Ferri et al. (1999).
In addition, the pooled sample and the speculative grade sample show that the CRA’s market capitalization has an immediate and negative impact in the upgrade threshold and an immediate and positive impact in the downgrade threshold. Then, an increase in the market capitalization increases immediately the absolute value of upgrade thresholds and downgrade thresholds. This effect largely disappears in the long run window, although, coefficients of M ktCap(−1) are significant for both upgrade thresholds in the investment grade and downgrade thresholds in the pooled sample.
Regarding the effect of investors’ perception on the CRA’s rating timeliness, we only find that upgrade thresholds of firms in the investment grade are immediately influenced by the the cumu- lative average abnormal return (Caarupg). Indeed, the higher the impact on the market the lower the absolute value of upgrade threshold is in the next quarter. Hence, the CRA varies the rating
standards in order to follow the impact of CAAR on the market only when they upgrade a firm that belong to the investment grade. Surprisingly, the downgrade threshold model is not significant in the case of investment grades. Downgrade thresholds are not related to reputation nor to business cycle. We finally note that stickiness is not symmetric between upgrade thresholds and downgrade thresholds. This finding leads to believe that the CRA reacts more timely when the default proba- bility of the rated firm increases than when it decreases.
To conclude, the main result of our study is that the CRA’s rating timeliness is significantly and negatively affected by both recession phases and the CRA’s reputation. Hence, we reject our first hypothesis. However, we do not reject our second hypothesis.
7
Conclusion
This chapter aims at understanding the determinants of rating timeliness changes over time. The change in the number of downgrades and upgrades between recession periods or boom periods, can be related to firms specificity or to variation in CRAs’ credit rating standards. Thus, we analyze timeliness over time in terms of thresholds (upgrade threshold and downgrade threshold) which will be analyzed by macroeconomic conditions proxy and reputation proxy. We find that CRAs’ rating is more timely in boom periods than in recession periods. Moreover, the CRA’s reputation affects negatively credit rating timeliness. We provide evidence that CRAs vary their rating stan- dards according to investors’ perception of previous rating changes only in the case of upgrades of investment grade. However, we do not find a significant impact of investors’ perception of pre- vious rating changes on rating timeliness for speculative grades. The main result of our study is that CRAs vary their credit rating timeliness over time due to reputation concerns.
8
Appendix
The latent rating change ∆DPRit∗ is modeled by ∆HP DP L as follows :
∆DPRit∗ = ξ1∆HP DP L + εit ∆DPRt = ∆DPRt∗ − α1 ∆DPRt∗ < α1 0 α1 < ∆DPRt∗ < α2 ∆DPRt∗ − α2 ∆DPRt∗ > α2
The stickiness is estimated by the upgrade threshold α1 < 0 and the downgrade threshold α2 > 0.
The threshold parameters, α1 and α2 are estimated based on the maximum of the likelihood
function of the model.
The likelihood function of this model is the following
F (∆DPRt/ξ, α1, α2, σ) = Π∆DPRt<01/σφ( ∆DPRt1+α1−ξX1 σ ) .Π∆DPRt=0[φ(α2−ξX0σ ) − φ(α1−ξX0σ )] .Π∆DPRt>01/σφ( ∆DPRt2+α2−ξX2 σ )
The log-likelihood is given by LogL() =P log(F ) : 137
LogL(∆DPRt/ξ, σ, α1, α2) = −1/2 ∗ (n1+ n2) ∗ log(2π) − (n1+ n2) ∗ Log(σ) − 1 2σ2 ∗P(∆DPRt1+ α1− ξX1)2 − 1 2σ2 ∗P(∆DPRt2+ α2− ξX2)2 +P[Log(φ(α2−ξX0σ )) − Log(φ(α1−ξX0σ ))] φ(.) = √1
2πexp[−1/2(.)] density function of the standard normal function
n1 =
P
∆DPRt<0, X1 = ∆HP DP L if ∆DPRt < 0
n2 =P∆DPRt>0, X2 = ∆HP DP L if ∆DPRt > 0