3.2.

Descriptive Statistics

Based on the comprehensive dataset of 183 corporate bond observations, this section
presents a detailed analysis of the key variables' statistical characteristics. Table 3.2
summarizes the descriptive statistics for all variables in the research model.

Table 3.2: Descriptive Statistics of Variables in the Research Model

Std.
Variable Observations Mean Minimum Maximum
Dev.

BY 183 1.26 0.47 0.43 2.37

MTBV 183 1.99 0.22 1.27 2.45

ROE 183 3.69 1.23 2.03 5.67

FREQ 183 40.43 17.25 1.00 62.00

MAT 183 8.79 3.44 1.00 20.00

CR 104 6.63 0.88 3.80 9.10

RAT 183 4.15 2.03 1.00 7.00

Note: All values except FREQ, MAT

and RAT are expressed in
percentages.

The analysis of corporate bond yields (BY), the dependent variable, reveals several
noteworthy patterns. The average yield of 1.26% demonstrates relatively modest returns
in the corporate bond market during the study period. The standard deviation of 0.47%
suggests moderate yield volatility, with yields ranging from a minimum of 0.43% to a
maximum of 2.37%. This spread of approximately 194 basis points indicates significant
variation in risk premiums across different bond issues.
Examining issuer characteristics, the market-to-book value ratio (MTBV) shows an
average of 1.99, indicating that companies in sample generally trade above their book
values. The relatively small standard deviation of 0.22 suggests consistent market
valuation across issuers. Return on Equity (ROE) averages 3.69% with a standard
deviation of 1.23%, reflecting moderate profitability levels among bond issuers. The
range from 2.03% to 5.67% indicates substantial variation in issuer profitability.

The frequency of bond issuance (FREQ) demonstrates considerable variation in market

participation among issuers. With a mean of 40.43 issues and a standard deviation of
17.25, the range extends from single issuers to those with 62 different bond issues,
highlighting diverse financing strategies among corporations. The maturity structure
(MAT) shows an average of 8.79 years, with issues ranging from 1 to 20 years, indicating
varied long-term financing needs.

Coupon rates (CR), based on 104 observations excluding zero-coupon bonds, average
6.63% with a standard deviation of 0.88%. The range from 3.80% to 9.10% reflects
varying risk premiums and market conditions at issuance. Credit ratings (RAT) display a
mean of 4.15 on 1-7 scale, with a standard deviation of 2.03, indicating diverse credit
quality among issuers.

Notably, the sample shows complete data for most variables (183 observations), except
for coupon rates where zero-coupon bonds are excluded. This comprehensive dataset
provides a robust foundation for subsequent statistical analyses and hypothesis testing.
The observed variations across all variables suggest that sample effectively captures the
diversity of the corporate bond market, enhancing the reliability of subsequent regression
analyses.
To evaluate the relationships between variables in the research model, a Pearson
correlation analysis was conducted. Table 4.1 presents the correlation matrix, revealing
several significant relationships among the variables.

Table 4.1: Pearson Correlation Coefficient Matrix

Variabl BY MTB ROE FREQ MAT CR RAT

e V

BY 1.000

MTBV -0.082 1.000

ROE 0.219 -0.297 1.000

FREQ 0.019 0.258 -0.164 1.000

MAT -0.063 0.161 -0.281 0.171 1.000

CR 0.017 0.056 -0.280 -0.038 0.660 1.000

RAT 0.357 -0.068 0.133 0.010 -0.064 -0.012 1.000

First, regarding correlations with the dependent variable (bond yields):

 Credit rating (RAT) exhibits the strongest correlation with bond yields (r = 0.357),
suggesting that lower credit quality (higher RAT values) is associated with higher
yields

 Return on Equity (ROE) shows a moderate positive correlation (r = 0.219),

indicating that more profitable firms tend to have higher yields

 Other variables show relatively weak correlations with yields, with coefficients
close to zero

Second, examining relationships among independent variables:

  The strongest correlation is observed between maturity (MAT) and coupon rate
(CR) at 0.660, suggesting longer-term bonds tend to offer higher coupons

 Market-to-book value (MTBV) shows moderate negative correlation with ROE (-

0.297), indicating that firms with higher market valuations tend to have lower
profitability

 Maturity shows moderate negative correlation with ROE (-0.281), suggesting

more profitable firms tend to issue shorter-term bonds

 FREQ demonstrates modest positive correlation with MTBV (0.258), indicating

that firms with higher market valuations tend to issue more bonds

The correlation patterns suggest that multicollinearity is unlikely to be a serious concern

in the regression model, as most correlations among independent variables are modest.
The exception is the relationship between maturity and coupon rate, which may warrant
attention in model specification. The strongest relationship with bond yields is observed
for credit ratings, suggesting this variable's particular importance in explaining yield
variations.

These correlation patterns provide initial insights into the relationships between variables,
though more sophisticated regression analysis will be needed to fully understand the
determinants of corporate bond yields while controlling for multiple factors
simultaneously.

4.2 Regression Analysis and Model Testing

To examine the determinants of corporate bond yields, a multiple regression analysis was
conducted. Table 4.2 presents the results of this analysis.

Table 4.2: Multiple Regression Results for Corporate Bond Yield Determinants

Variable Coefficient Std. t-value P-value

 Error

MTBV -0.0003122 0.0022379 -0.14 0.889

ROE 0.0007826 0.0004072 1.92 0.058*

FREQ 0.0000326 0.0000450 0.73 0.470

MAT -0.0000963 0.0001427 -0.68 0.501

CR 0.0007328 0.0006866 1.07 0.288

RAT 0.0007672 0.0002209 3.47 0.001***

Constant 0.0015584 0.0068285 0.23 0.820

Note: *** p<0.01, ** p<0.05, *

p<0.1

Model Statistics:

 R-squared: 0.1696

 Adjusted R-squared: 0.1182

 F-statistic: 3.30 (p-value = 0.0053)

 Number of observations: 104

The regression analysis reveals several important findings:

1. Model Fit:

 The model explains approximately 16.96% of the variation in corporate bond

yields (R² = 0.1696)

 The F-test indicates that the model is statistically significant (p = 0.0053)

2. Individual Variable Effects:

a) Credit Rating (RAT):

  Shows the strongest and most significant effect (β = 0.0007672, p = 0.001)

 A one-unit increase in rating score (indicating lower credit quality) is associated

with a 0.077 percentage point increase in yield

 This finding aligns with theoretical expectations that lower credit quality
commands higher risk premiums

b) Return on Equity (ROE):

 Exhibits marginally significant positive effect (β = 0.0007826, p = 0.058)

 Each percentage point increase in ROE is associated with a 0.078 percentage point
increase in yield

 This somewhat counterintuitive relationship might reflect higher risk associated

with more profitable firms

c) Other Variables:

 Market-to-book value (MTBV), trading frequency (FREQ), maturity (MAT), and

coupon rate (CR) show no statistically significant effects on yields

 The negative coefficient for maturity, while not significant, suggests a slight
tendency for longer-term bonds to have lower yields

The relatively low R-squared value suggests that factors beyond these firm-specific
characteristics likely play important roles in determining bond yields. These might
include macroeconomic conditions, market liquidity, or other risk factors not captured in
the current model.

The significant F-statistic indicates that despite the modest R-squared, the model as a
whole provides meaningful explanatory power for understanding corporate bond yield
variations. The dominance of credit rating effects suggests that credit risk remains a
primary consideration in the pricing of corporate bonds in this market.
The regression equation can be written as:

Bond Yield = 0.0015584 - 0.0003122(MTBV) + 0.0007826(ROE) + 0.0000326(FREQ) -

0.0000963(MAT) + 0.0007328(CR) + 0.0007672(RAT)

3. Multicollinearity Test:

To assess potential multicollinearity among independent variables, Variance Inflation

Factor (VIF) analysis was conducted. Table 4.3 presents the results of this diagnostic test.

Table 4.3: Variance Inflation Factor (VIF) Test Results

Variable VIF 1/VIF

MAT 1.9 0.515254

4

CR 1.9 0.522248
1

ROE 1.2 0.815684

3

MTBV 1.1 0.859914

6

FREQ 1.1 0.871007

5

RAT 1.0 0.974807

3

Mean VIF 1.4

0

The VIF test results indicate that multicollinearity is not a concern in the model:
 1. Individual Variable Analysis:

 Maturity (MAT) shows the highest VIF at 1.94

 Coupon Rate (CR) follows closely with VIF of 1.91

 All other variables have VIF values well below 1.5

 Credit Rating (RAT) shows the lowest VIF at 1.03

2. Overall Assessment:

 All VIF values are substantially below the conventional threshold of 10

 The mean VIF of 1.40 indicates low overall correlation among predictors

 Tolerance values (1/VIF) are all well above 0.5, further confirming absence of
serious multicollinearity

The results suggest that the estimated coefficients in the regression model are reliable and
not distorted by relationships between the independent variables. The relatively high VIFs
for maturity and coupon rate reflect their natural relationship in bond structuring but are
not concerning for model estimation.

4.3 Testing Regression Model Assumptions

To verify the assumption of homoscedasticity in the regression model, the Breusch-

Pagan/Cook-Weisberg test was conducted with the following hypotheses:

H₀: Homoscedasticity (constant error variance) H₁: Heteroskedasticity (varying error

variance)

Table 4.4: Breusch-Pagan/Cook-Weisberg Test Results

Indicator Value

Chi-square(1) 1.31
 Prob > chi² 0.2520

Test Results Analysis:

 The test yields a chi-square statistic of 1.31

 The corresponding p-value is 0.2520, which is greater than the conventional

significance level of 0.05

 This provides insufficient evidence to reject the null hypothesis of constant

variance

Interpretation:

 The result indicates that the model does not violate the homoscedasticity
assumption

 The p-value of 0.2520 suggests strong evidence that the error terms have constant
variance

 This finding confirms the reliability of the regression coefficient estimates and
their standard errors

The absence of heteroskedasticity strengthens the validity of statistical inferences and

hypothesis tests, indicating that the model's predictions are equally reliable across all
levels of the predicted values.

2. Autocorrelation Test

To examine the presence of serial correlation in the model, Durbin's alternative test for
autocorrelation was conducted with the following hypotheses:

H₀: No serial correlation H₁: Presence of serial correlation

Table 4.5: Durbin's Alternative Test Results

Indicator Value

Chi-square(1) 6.253

Prob > chi² 0.0124

Number of gaps 10

Test Results Analysis:

 The test yields a chi-square statistic of 6.253 with 1 degree of freedom

 The p-value of 0.0124 is less than the conventional significance level of 0.05

 The data contains 10 gaps in the sample series

Interpretation:

 There is sufficient evidence to reject the null hypothesis of no serial correlation

 The result indicates the presence of first-order autocorrelation in the model

 This suggests significant temporal dependencies in the bond yield series

Implications:

 The presence of autocorrelation suggests that consecutive observations in the yield

series are not independent

 This finding is common in financial time series data

 While the coefficients remain unbiased, their standard errors might be affected

 This may require the use of robust standard errors or alternative estimation
methods to ensure reliable statistical inference
The detection of autocorrelation suggests that future research might benefit from
employing dynamic specifications or time series techniques that can better account for
these temporal dependencies in the data.

4.4 Robustness Check

To ensure the reliability of findings, particularly given the presence of autocorrelation, we

re-estimated the model using robust standard errors. Table 4.6 presents a comparison
between the original and robust estimation results.

Table 4.6: Comparison of Ordinary and Robust Regression Results

Variable Original Original p- Robust Robust p- Coefficient

SE value SE value

MTBV 0.0022379 0.889 0.0018192 0.864 -0.0003122

ROE 0.0004072 0.058* 0.0003462 0.026** 0.0007826

FREQ 0.0000450 0.470 0.0000442 0.462 0.0000326

MAT 0.0001427 0.501 0.0001527 0.529 -0.0000963

CR 0.0006866 0.288 0.0008723 0.403 0.0007328

RAT 0.0002209 0.001*** 0.0002574 0.004*** 0.0007672

Constant 0.0068285 0.820 0.007215 0.829 0.0015584

Note: *** p<0.01, ** p<0.05, * p<0.1

Model Statistics:

 R-squared: 0.1696 (unchanged)

 F-statistic: 3.83 (robust) vs 3.30 (original)

 Prob > F = 0.0018 (robust)

Key Findings from Robust Estimation:

 1. Stability of Coefficients:

 All regression coefficients remain identical after robust estimation

 This stability confirms the reliability of the estimated relationships

2. Changes in Statistical Significance:

 ROE becomes more significant under robust estimation (p = 0.026 vs p = 0.058)

 RAT maintains strong significance (p = 0.004 vs p = 0.001)

 Other variables remain non-significant

3. Standard Error Adjustments:

 Most variables show modest changes in standard errors

 MTBV shows reduced standard errors under robust estimation

 CR shows increased standard errors, suggesting more uncertainty in its estimate

The robustness check confirms the main findings of analysis while providing more
reliable statistical inference under potential violations of classical assumptions. The
strengthened significance of ROE and continued significance of RAT under robust
estimation particularly reinforces the importance of these variables in determining
corporate bond yields.

4.5 Discussion of Research Findings

Based on the comprehensive statistical analyses conducted, including robust regression

and diagnostic testing, draw several important conclusions about the determinants of
corporate bond yields in sample:

The model's explanatory power, while modest with an R-squared of 0.1696, is

statistically significant (F = 3.83, p = 0.0018), indicating that these firm-specific
characteristics do play a meaningful role in determining bond yields, though other factors
likely contribute substantially to yield variations.

1. Credit Rating Effects (RAT):

 Emerges as the most significant determinant (β = 0.0007672, p = 0.004)

 The positive coefficient indicates that lower credit quality (higher rating score)
leads to higher yields

 This relationship remains robust across both standard and robust estimations

 Confirms the fundamental role of credit risk assessment in bond pricing

2. Profitability Impact (ROE):

 Shows significant positive relationship with yields (β = 0.0007826, p = 0.026)

 The relationship strengthens under robust standard errors

 This somewhat counterintuitive finding might suggest that:

o More profitable firms may engage in riskier activities

o Higher ROE might be associated with greater operational leverage

o Market participants may perceive high profitability as unsustainable

3. Market Valuation and Trading Activity:

 Market-to-book value (MTBV) shows no significant effect on yields

 Trading frequency (FREQ) similarly demonstrates no significant impact

 These findings suggest that market perception and trading activity may not be
primary considerations in yield determination

4. Bond Characteristics:

 Neither maturity (MAT) nor coupon rate (CR) shows significant effects on yields
  The lack of maturity effect is particularly noteworthy, as it differs from traditional
term structure theories

 This might reflect market-specific factors or the particular nature of sample period

Model Diagnostics and Reliability:

 No significant multicollinearity (mean VIF = 1.40)

 Absence of heteroskedasticity (p = 0.2520)

 Presence of autocorrelation addressed through robust standard errors

 Results remain stable across different estimation methods

Implications:

1. For Investors:

 Credit ratings should be the primary consideration in yield expectations

 Firm profitability warrants attention, but traditional interpretations may need

revision

 Traditional term structure considerations may be less relevant in this market

2. For Issuers:

 Credit rating management should be a priority for managing borrowing costs

 Higher profitability may not necessarily lead to lower borrowing costs

 Bond structure (maturity, coupon) may have less impact on costs than traditionally
assumed

3. For Researchers:

 The modest R-squared suggests important roles for factors not included in this
model
  Future research might consider:

o Macroeconomic variables

o Market liquidity measures

o Industry-specific factors

o Time-varying effects

These findings contribute to understanding of corporate bond yield determinants while

highlighting the complexity of yield determination in practice. The dominance of credit
ratings and the unexpected relationship with profitability suggest that traditional
frameworks for understanding bond yields may need modification in current market
conditions.
Appendix