3.5 Cross-sectional regression analysis
3.5.1 Model specifications
In order to more closely identifying the drivers of the stock market reactions following rating announcements, a cross-sectional regression analysis is conducted. The multivariate ordinary least squares (OLS) regression follows:
CARj,[−1;1]=β0+
n
X
i=1
βiV ARi+ (3.5)
whereCARj,[−1;1]is the abnormal return of firmj∈ {1, . . . , m}, during the [−1; 1] event window,
β0 is the regression constant, βi are the regression coefficients for the independent variables with
i∈ {1, . . . , m},V ariare the independent variables with i∈ {1, . . . , m}, andis the error term.
In order to explain the CARs during the [−1; 1] event window, company and event specific variables are tested. The model tests whether the results from the univariate analysis are consistent in a cross-sectional analysis. In the univariate regression, when a rating review preceded an action, the review caused a higher ACAR than the actual rating action. Similar effects are observed by
Table 3.6: Stock market reactions on downgrade actions by rating reason; ∗ , ∗∗ , ∗∗∗ denote significance at the 10, 5 and 1% level, respectively.
Event Median t-test SIGN BMP-test Corrado-test Sample window ACAR CAR (t-value) (Z-Score) (Z-score) (Z-scroe) size Panel A: Average abnormal stock returns by downgrades due to operating performance
[−2;−1] 0.00% 0.19% −0.01 −0.65 0.26 0.61 101 [−1;0] −0.22% −0.20% −0.71 −0.61 −0.57 −0.09 101 {0} −0.25% −0.24% −1.16 −1.40 −1.31 −1.43∗ 101 [0;1] −0.76% −0.82% −2.36∗∗ −2.84∗∗∗ −2.37∗∗ −2.32∗∗ 101 [−1;1] −0.73% −0.62% −1.77∗ −1.68∗ −1.64 −1.14 101 [−2;2] −0.96% −0.51% −1.65 −1.68∗ −1.70∗ −1.75∗ 101
Panel B: Average abnormal stock returns by downgrades due to capital structure
[−2;−1] 1.22% 0.37% 2.59∗∗ −1.84∗ 2.49∗∗ 2.09∗∗ 48 [−1;0] 0.47% 0.52% 1.31 −1.49 1.27 0.84 48 {0} −0.28% 0.00% −1.11 −0.57 −0.80 −0.21 48 [0;1] −0.55% 0.31% −1.04 −0.52 −0.82 0.37 48 [−1;1] 0.20% 0.10% 0.41 −0.55 0.58 0.51 48 [−2;2] 0.68% 0.68% 0.94 −1.24 1.10 1.06 48 Panel C: Differences of downgrades due to operating performance and capital structure
Event D_Median two-sample Wilxocon rank-sum window D_ACAR CAR t-test (t-value) test (Z-score) [−2;−1] −1.22% −0.17% −2.33∗∗ 1.95∗ [−1;0] −0.69% −0.72% −1.33 1.28 {0} 0.03% −0.25% 0.08 1.32 [0;1] −0.21% −1.13% −0.36 0.38 [−1;1] −0.93% −0.73% −1.35 1.05 [−2;2] −1.64% −1.19% −1.67 1.47
Table 3.7: Stock market reactions on upgrade actions by rating reason;∗ , ∗∗ ,∗∗∗ denote signifi- cance at the 10, 5 and 1% level, respectively.
Event Median t-test SIGN BMP-test Corrado-test Sample window ACAR CAR (t-value) (Z-Score) (Z-score) (Z-scroe) size Panel A: Average abnormal stock returns by upgrades due to operating performance
[−2;−1] −0.11% −0.11% −0.46 −0.21 −0.09 −0.26 50 [−1;0] 0.03% −0.36% 0.10 −0.67 0.24 −0.05 50 {0} 0.06% −0.13% 0.36 −0.05 0.44 0.28 50 [0;1] −0.21% −0.27% −0.81 −1.03 −0.78 −0.42 50 [−1;1] −0.24% −0.32% −0.72 −1.39 −0.61 −0.55 50 [−2;2] −0.39% −0.50% −1.00 −1.23 −0.73 −0.65 50 Panel B: Average abnormal stock returns by upgrades due to capital structure
[−2;−1] 0.40% 0.18% 1.12 −1.01 1.13 1.04 46 [−1;0] −0.26% −0.14% −0.91 −0.74 −1.00 −0.67 46 {0} −0.19% −0.32% −1.05 −0.97 −1.17 −1.14 46 [0;1] −0.19% −0.44% −0.62 −1.15 −0.52 −1.01 46 [−1;1] −0.25% −0.05% −0.74 −0.44 −0.67 −0.71 46 [−2;2] 0.38% −0.55% 0.62 −0.02 0.57 −0.05 46 Panel C: Differences of upgrades due to operating performance and capital structure
Event D_Median two-sample Wilxocon rank-sum window D_ACAR CAR t-test (t-value) test (Z-score) [−2;−1] −0.51% −0.29% −1.19 0.90 [−1;0] 0.29% −0.22% 0.72 0.00 {0} 0.25% 0.20% 1.03 −0.85 [0;1] −0.02% 0.18% −0.05 −0.15 [−1;1] 0.01% −0.27% 0.03 0.66 [−2;2] −0.78% 0.05% −1.08 0.77
Norden and Weber (2004) and by Bannier and Hirsch (2010). A rating process usually starts with a review by one agency and ends with an action by a second agency. The rating always moves either down or up for all announcements. If no rating review is announced, the rating process starts with the first rating action. The variable F IRST indicates whether the review or actual rating announcement is the first information available to the market. The variableREV IEW indicates if the rating announcement is a rating review. The univariate results suggest that investors do not distinguish between rating announcements attributed to operating performance and those attributed to capital structure. The rating rationale is captured by the variableREASON and is defined as 1 for changes in operating performance as rating reason. The variableM OODY Sis introduced as a dummy variable for the rating agency.
Besides these variables, other variables can potentially explain the size of the CAR, and our model controls for several additional variables. The regression examines the effect of the “old rating” category. Norden and Weber (2004) find in their cross-sectional analysis that the old rating level and previous rating events significantly influence the magnitude of the abnormal performance. The variableRAT IN Gis based on a numerical 17 rating scale (AAA/Aaa=1, AA+/Aa1=2,. . . , CCC/Caa1 and below=17). For investors whose mandate limits them to holding only investment grade securities, downgrade events to non-investment grade can be highly problematic. Hite and Warga (1997) find that rating changes from investment grade to non-investment grade have a fatal effect on bond prices.
In addition, the variableIN V_BORDERis introduced for the investment grade border. The size of the firm can be an important factor explaining the CAR. The size effects are measured by the market capitalization (M CAP) on the last trading day in the year prior to the year of the rating change.
For the firm’s credit risk, two additional firm information categories are introduced. The variableDEBT represents all interest-bearing and capitalized lease obligations as the sum of long- and short-term debt on the last trading day in the year prior to year of the rating change. Rising credit risk is also captured in the variableDEBT%. The variableDEBT% is defined as the total debt ratio between the last trading day two years and one year prior to the rating change.
Finally, we introduce a variable to address one of the most discussed topics since the peak of the financial crisis: to what extent rating agencies failures are responsible for the crisis. Since rating agencies underestimated the credit risk associated with structured financed products and failed to adjust their ratings quickly enough, there is a broad consensus that credit agencies contributed to the financial crisis. In 2008 and 2009, credit rating agencies adjusted their firm’s issuer credit ratings. The new probability of credit risk is also important for equity investors. Therefore, we introduce the variables Y2008 and Y2009 as dummy variables for the years with most rating announcements.