• No results found

4. Data, Sample and Variable Definitions

4.4 Empirical specifications

4.4.1Conservatism and uncertainty in the bond market

To test my first hypothesis that conservatism increases uncertainty among bond investors especially when reports reflect low performance, I regress firms’ quarterly bond return volatility, as a proxy for bond investors’ uncertainty, on the lagged measure of conservatism as well as its interaction with an indicator variable that takes a value of one for accounting loss-quarters. I control for several firm-level variables that could affect the volatility of bond returns as well as quarter and firm fixed effects. To test H1a, I use the following specification and expect 𝛽1 to be significantly positive:

π‘…π‘’π‘‘π‘‰π‘œπ‘™π‘–π‘—π‘‘ = 𝛽0+ 𝛽1 πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ’1+ 𝛽2π‘…π‘‚π΄π‘—π‘‘βˆ’1+ 𝛽3πΏπ‘’π‘£π‘—π‘‘βˆ’1+ 𝛽4π‘†π‘–π‘§π‘’π‘—π‘‘βˆ’1+ 𝛽5π΅π‘€π‘—π‘‘βˆ’1+

𝛽6πΆπΉπ‘‚π‘‰π‘œπ‘™π‘—π‘‘βˆ’1+ 𝛽7π‘†π‘‘π‘œπ‘π‘˜π‘…π‘’π‘‘π‘—π‘‘+ 𝛽8π‘…π‘Žπ‘‘π‘–π‘›π‘”π‘—π‘‘+ 𝑒𝑖+ 𝑣𝑑+ πœ€π‘–π‘—π‘‘ (1) where 𝑖, 𝑗, and 𝑑 are bond, firm, and quarter indices, respectively. To test H1b, I augment equation (1) with its interaction with a loss variable (as defined below) and expect 𝛽2 to be significantly positive:

π‘…π‘’π‘‘π‘‰π‘œπ‘™π‘–π‘—π‘‘ = 𝛽0+ 𝛽1πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ’1+ 𝛽2πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ’1βˆ— πΏπ‘œπ‘ π‘ π‘—π‘‘βˆ’1+ 𝛽3πΏπ‘œπ‘ π‘ π‘—π‘‘βˆ’1+ πΆπ‘œπ‘›π‘‘π‘Ÿπ‘œπ‘™π‘  +

𝑒𝑖+ 𝑣𝑑+ πœ€π‘–π‘—π‘‘ (2) where πΆπ‘œπ‘›π‘‘π‘Ÿπ‘œπ‘™π‘  is the same set of control variables included in equation (1). π‘…π‘’π‘‘π‘‰π‘œπ‘™ equals the

quarterly measure of bond return volatility defined as the standard deviation of daily bond returns during a quarter. I define πΏπ‘œπ‘ π‘  as the fraction of the past 20 quarters with negative earnings before extraordinary items.

I control for several known determinants of return volatility following Bao and Pan (2013) and Rajgopal and Venkatachalam (2011). Vuolteenaho (2002) has shown that returns can be decomposed into news about expected return and expected cash flows. Therefore, I control for the volatility of firms’ prior cash flows (𝑉𝐢𝐹𝑂) because variation in cash flows is likely to be mirrored in variation in returns. I also, control for leverage (𝐿𝑒𝑣) and bond ratings (π‘…π‘Žπ‘‘π‘–π‘›π‘”) because both

23

capture firms’ financial distress risk which can affect the volatility of bond returns. I expect firms’ volatility to increase with leverage and decrease with ratings. I also expect returns of smaller firms to be more sensitive to corporate news as they have less ability to absorb negative corporate shocks and fewer assets to compensate lenders during the bankruptcy process. I predict the volatility of bond returns to be negatively associated with firm size (𝑆𝑖𝑧𝑒). I control for firms’ growth opportunities by including the book-to-market ratio (B𝑀) because growth firms are more likely to have volatile performance and returns. Since bondholders have fixed claims against borrowers’ value, they are likely to be more sensitive to corporate news when borrowers are less profitable. Thus, I expect, firms’ profitability (𝑅𝑂𝐴) to reduce the volatility of bond returns. Additionally, I control for corporate news that is known to the stock market which could have an impact on the bond market by including the compounded stock returns of the given quarter (π‘†π‘‘π‘œπ‘π‘˜π‘…π‘’π‘‘).

4.4.2Informational cost of conservatism

My second hypothesis (H2) predicts that the uncertainty induced by accounting conservatism increases bond yield spreads. I test this prediction using the following empirical specification:

π‘†π‘π‘Ÿπ‘’π‘Žπ‘‘π‘–π‘—π‘‘= 𝛽0+ 𝛽1πΌπ‘›π‘“π‘œπΆπ‘œπ‘ π‘‘π‘—π‘‘βˆ’1+ 𝛽2π‘…π‘‚π΄π‘—π‘‘βˆ’1+ 𝛽3πΏπ‘’π‘£π‘—π‘‘βˆ’1+ 𝛽4π‘†π‘–π‘§π‘’π‘—π‘‘βˆ’1+ 𝛽5π΅π‘€π‘—π‘‘βˆ’1+

𝛽6π‘‚π‘“π‘“π‘’π‘Ÿπ΄π‘šπ‘‘π‘–π‘—π‘‘+ 𝛽7π‘…π‘Žπ‘‘π‘–π‘›π‘”π‘–π‘—π‘‘+ 𝛽8π‘€π‘Žπ‘‘π‘’π‘Ÿπ‘–π‘‘π‘¦π‘–π‘—π‘‘+ 𝛽9π‘†π‘’π‘›π‘–π‘œπ‘Ÿπ‘–π‘‘π‘¦π‘–π‘—π‘‘+ 𝛽10π‘πΆπ‘œπ‘£π‘–π‘—π‘‘+

𝛽11𝐴𝑠𝑠𝑒𝑑_π΅π‘Žπ‘π‘˜π‘’π‘‘π‘–π‘—π‘‘+ 𝛽12π‘…π‘’π‘‘π‘’π‘’π‘šπ‘Žπ‘π‘™π‘’π‘–π‘—π‘‘+ π‘‚π‘“π‘“π‘’π‘Ÿ_π‘€π‘›π‘‘β„Žπ‘–π‘—+ πΌπ‘›π‘‘π‘’π‘ π‘‘π‘Ÿπ‘¦π‘–+ πœ€π‘–π‘—π‘‘ (3)

where i and t represent bonds and offering periods, respectively. The dependent variable, Spread,

is the yield spread of new bonds defined as the offering yield minus the yield of treasury securities issued during the same month with the closest maturity date to that of the corresponding bond.

πΌπ‘›π‘“π‘œπΆπ‘œπ‘ π‘‘ equals the volatility of bond returns induced by conservatism and the interaction of conservatism (πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£) and Loss variables from the firm-specific regression (2). The parameters of equation (1) are estimated separately for each new bond issuance over the same 5 to

24

20 prior quarters that are used to calculate the conservatism measure. Then, InfoCost is defined as the predicted volatility of returns explained by the most recent values of πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£ and its interaction with πΏπ‘œπ‘ π‘ , i.e., InfoCostjt = 𝛼̂𝑗1 πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘+ 𝛼̂𝑗2πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ— πΏπ‘œπ‘ π‘ π‘—π‘‘+ 𝛼̂𝑗3πΏπ‘œπ‘ π‘ π‘—π‘‘. I predict the coefficient on πΌπ‘›π‘“π‘œπΆπ‘œπ‘ π‘‘ (𝛽1) to be significantly positive, confirming that uncertainty induced by conservatism increases the yield spread of new bonds.

In equation (3), I also include several firm and bond specific control variables that are expected to impact bond interest rates. I expect larger and more profitable firms to issue bonds at lower interest rates as the extent of bondholder-shareholder conflict is potentially less severe. Similarly, higher leverage is associated with higher expected default and bankruptcy risk which should increase bond interest rates. I include ROA, Lev, BM, and Size as defined in the previous section to control for firm-level determinants of the cost of borrowing (Fama and French 1993, Gebhardt et al. 2005, Zhang 2008, Ahmed et al. 2002). Another subtle reason for including firm profitability (ROA) is to control for the mechanical association between earnings skewness and the mean earnings. As earnings become more negatively skewed, its mean also shifts to the left which could cause an increase in bond spreads. However, my main conjecture is that conservatism increases yield spreads by increasing uncertainty rather than by simply lowering operating performance measures. Controlling for earnings profitability ensures that any association between conservatism and yields is not due to a decline in operating performance. I also control for other terms of bond contracts as they are simultaneously determined with interest rates. I expect higher bond ratings (π‘…π‘Žπ‘‘π‘–π‘›π‘”), higher bond seniority (π‘†π‘’π‘›π‘–π‘œπ‘Ÿπ‘–π‘‘π‘¦), and asset-backed bonds (𝐴𝑠𝑠𝑒𝑑_π΅π‘Žπ‘π‘˜π‘’π‘‘) to have lower interest rates as they reflect lower risk of default and lower agency conflict between borrowers and lenders (Blume et al. 1991, Gebhardt et al. 2005, Bessembinder et al. 2008, Zhang 2008, Ahmed et al. 2002). I also, include bond issue size (π‘‚π‘“π‘“π‘’π‘Ÿ_π΄π‘šπ‘‘), time to maturity (π‘€π‘Žπ‘‘π‘’π‘Ÿπ‘–π‘‘π‘¦), and total number of included covenants (π‘πΆπ‘œπ‘£), and an indicator for redeemable (π‘…π‘’π‘‘π‘’π‘’π‘šπ‘Žπ‘π‘™π‘’) bonds (Blume et al. 1991, Gebhardt et al. 2005, Zhang 2008). I do not

25

have any prior expectations about their effect on interest rates, because prior literature finds that, beyond contractual values, these bond agreement terms could indicate other offsetting factors. For instance, while intensity of covenants could reduce the flexibility of borrowers to expropriate wealth from lenders (Beatty et al. 2002), it could also signal the severity of agency conflict between the two parties (Li et al. 2016). Similarly, while longer duration and larger size of bonds could increase lender exposure to borrowers’ risk, it could signal the credibility of the borrower as well.

4.4.3Contractual benefits of conservatism

My third set of hypotheses predicts that earnings covenants reduce bond yield spreads and that this effect gets reinforced with level of accounting conservatism. To test H3a and H3b, I use the following two empirical specifications:

π‘†π‘π‘Ÿπ‘’π‘Žπ‘‘π‘–π‘—π‘‘ = 𝛽0+ 𝛽1π‘πΈπ‘Žπ‘Ÿπ‘›_πΆπ‘œπ‘£π‘–π‘—π‘‘+ πΆπ‘œπ‘›π‘‘π‘Ÿπ‘œπ‘™π‘  + π‘‚π‘“π‘“π‘’π‘Ÿ_π‘€π‘›π‘‘β„Žπ‘–π‘—+ πΌπ‘›π‘‘π‘’π‘ π‘‘π‘Ÿπ‘¦π‘–+ πœ€π‘–π‘—π‘‘ (4) and,

π‘†π‘π‘Ÿπ‘’π‘Žπ‘‘π‘–π‘—π‘‘ = 𝛽0+ 𝛽1π‘πΈπ‘Žπ‘Ÿπ‘›_πΆπ‘œπ‘£π‘–π‘—π‘‘+ 𝛽2 πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ’1+ 𝛽3πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ’1βˆ— π‘πΈπ‘Žπ‘Ÿπ‘›_πΆπ‘œπ‘£π‘–π‘—π‘‘+

πΆπ‘œπ‘›π‘‘π‘Ÿπ‘œπ‘™π‘  + π‘‚π‘“π‘“π‘’π‘Ÿ_π‘€π‘›π‘‘β„Žπ‘–π‘—+ πΌπ‘›π‘‘π‘’π‘ π‘‘π‘Ÿπ‘¦π‘–+ πœ€π‘–π‘—π‘‘ (5) where πΆπ‘œπ‘›π‘‘π‘Ÿπ‘œπ‘™π‘  are same control variables used in equation (3). π‘πΈπ‘Žπ‘Ÿπ‘›_πΆπ‘œπ‘£ equals the intensity of operating performance covenants defined as the number of earnings-based covenants included in new bond contracts. H3a predicts that the coefficient on π‘πΈπ‘Žπ‘Ÿπ‘›_πΆπ‘œπ‘£, 𝛽1, in equation (4) is significantly negative.

To capture the reinforcing effect of accounting conservatism and intensity of earnings covenants, I interact πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£ and π‘πΈπ‘Žπ‘Ÿπ‘›_πΆπ‘œπ‘£ in equation (5). A significantly negative coefficient on the interaction term, 𝛽3, will confirm the prediction in H3b. Thus, I refer to the combined effect of conservatism and earnings covenants (πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£ βˆ— π‘πΈπ‘Žπ‘Ÿπ‘›_πΆπ‘œπ‘£) as the β€œcontractual benefit” of conservatism.

26

4.4.4Overall effect of conservatism on bond yield spreads

Lastly, I test the overall effect of conservatism on bond yield spreads (H4) by regressing yield spreads (π‘†π‘π‘Ÿπ‘’π‘Žπ‘‘) on conservatism (πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£) and other control variables, where indices and control variables are identical to those in equation (3).

π‘†π‘π‘Ÿπ‘’π‘Žπ‘‘π‘–π‘—π‘‘ = 𝛽0+ 𝛽1πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ’1+ πΆπ‘œπ‘›π‘‘π‘Ÿπ‘œπ‘™π‘  + π‘‚π‘“π‘“π‘’π‘Ÿ_π‘€π‘›π‘‘β„Žπ‘–π‘—+ πΌπ‘›π‘‘π‘’π‘ π‘‘π‘Ÿπ‘¦π‘–+ πœ€π‘–π‘—π‘‘ (6) and,

π‘†π‘π‘Ÿπ‘’π‘Žπ‘‘π‘–π‘—π‘‘ = 𝛽0+ 𝛽1πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ’1+ 𝛽2πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£π‘—π‘‘βˆ’1βˆ— πΏπ‘œπ‘ π‘ π‘—π‘‘βˆ’1 + 𝛽3πΏπ‘œπ‘ π‘ π‘—π‘‘βˆ’1+ πΆπ‘œπ‘›π‘‘π‘Ÿπ‘œπ‘™π‘  +

π‘‚π‘“π‘“π‘’π‘Ÿ_π‘€π‘›π‘‘β„Žπ‘–π‘—+ πΌπ‘›π‘‘π‘’π‘ π‘‘π‘Ÿπ‘¦π‘–+ πœ€π‘–π‘—π‘‘ (7) H4a predicts that the coefficient on πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£, 𝛽1, in equation (6) is significantly positive. To test H4b that the effect of conservatism on bond yield spreads is stronger when accounting reports convey bad news, I interact πΆπ‘œπ‘›π‘ π‘’π‘Ÿπ‘£ and π‘™π‘œπ‘ π‘  in equation (7). A significantly positive coefficient on the interaction term, 𝛽2, will confirm the prediction in H4b.

Related documents