ESSAY 1 CHAPTER THREE: RESEARCH DESIGN
3.4 Model Specification
I use the length of earnings string (ES_String) and MBE string (MBE_String) for the proxies of the earnings benchmark string. To test H1, I use the following regression model to examine the association between the bond yield spread and the length of the earnings benchmark string:
Spreadiq = α0 + α1ES_Stringiq /MBE_Stringiq + α2Sizeiq + α3Lossiq + α4StdRetiq + α5Tobin’s Qiq
+ α6Leviq + α7CFOiq + α8Liqiq+ α9Tangiq + α10ROAiq + α11IssueSizeiq + (α12Dacci + α13Ab_Dexpiq + α14Ab_Prodiq) + ∑αtYeariq + ∑αtIndustryiq + εiq (1)
Where:
Spread = the yield to maturity at the issuance date for the largest bond firm it issued in
year t minus the Treasury bond yield with similar maturity. I also measure spread
as the natural logarithm of the initial bond spread.
ES_String = the natural logarithm of the length of the earnings string measured by a sequence of quarters in which a firm’s EBIT is higher than that of the same fiscal quarter from the previous year.
MBE_String = the natural logarithm of the length of the MBE string measured by a sequence of quarters in which a firm’s actual earnings meet or beat analysts’ most recent consensus forecasts before the earnings quarterly announcement date.
Size = the natural logarithm of firm i’s total assets at the end of quarter q.
Loss =equal to one if there is a loss in the current fiscal year, and zero otherwise.
StdRet = the standard deviation of firm i’s daily stock returns during quarter q.
Tobin’s Q =firm i’s market value of assets/the book value of assets at the end of quarter q.
Lev =firm i’s total debt/total assets at the end of quarter q.
CFO =firm i’s operating cash flow/total assets at the end of quarter q.
Liq = the ratio of new liquid assets (total current assets – total current liabilities) to total assets at the end of quarter q.
Tang =the ratio of net property, plant, and equipment over total assets at the end of quarter q.
ROA =firm i’s return on assets at the end of year t measured by income before extraordinary items scaled by average total assets (quarter q-1 and 1).
IssueSize =the natural logarithm of the offering amount of the bond (in millions of dollars).
Dacc = firm i’s quarterly discretionary accruals measured by a modified Dechow and
Dichev (2002) cash flow accrual model.3
3 The Dechow and Dichev (2002) cash flow accrual model regresses working capital accruals on operating cash flows in the prior, current, and future periods. The Dechow and Dichev (2002) model improved the traditional Jones model by explicitly mapping cash flows into the accruals generating process, which, I believe, is particularly important in the debt market. Further, as suggested by McNichols (2002), I also include the change in sales revenue and a fourth quarter indicator variable into the model in order to increase explanatory power. Specifically, I estimate discretionary accruals by regressing the following model for each industry group: ∆WCt=b0 + b1CFOt-1 + b2CFOt + b3CFOt+1
Ab_Dexp = firms i’s quarterly abnormal expenditures measured by Roychowdhury’s (2006)
real earnings management model.
Ab_Prod = firms i’s quarterly abnormal production measured by Roychowdhury’s (2006)
real earnings management model.
If debt holders view firms that sustain a string of earnings benchmarks as risky borrowers, then I expect that the coefficient on ES_Stringiq/MBE_Stringiq will be significantly positive,
suggesting that debt holders demand a higher risk premiums from firms that sustain a string of earnings benchmarks. To test H2, I use the following regression model to examine the association between the credit rating and the length of the earnings benchmark string:
SP_Ratingiq = α0 + α1ES_Stringiq /MBE_Stringiq + α2Sizeiq + α3Lossiq+ α3StdRetiq + α4Tobin’s Qiq
+ α5Leviq + α6CFOiq + α7Liqiq+ α8Tangiq + α9ROAiq + α10IssueSizeiq + (α11Dacciq
+ α12Ab_Dexpiq+ α13Ab_Prodiq) + ∑αtYeariq + ∑αtIndustryiq + εiq (3)
Where: SP_Ratingiq is the firm’s S&P ratings from AAA (indicating a string capacity to
pay interest and repay principal) to D (indicating actual default). I transform S&P ratings into a 22-point scale with a smaller number indicating a better rating. Table 1 provides a detailed rating-
scale index. Consistent with Jiang (2008), the BBB rating-scale has the largest observations.4 If
credit rating agencies also view firms that sustain a string of earnings benchmarks as risky borrowers, then I expect the coefficient on ES_/MBE_Stringiq will be significantly positive.
Further, to examine whether the debt market reacts differently from the equity market to firms that sustain a string of earnings benchmarks, I also use the following regression model to
examine the association between firms’ abnormal stock returns and the length of the earnings benchmark string:
CAR (-1, 1)iq/ BHAR (-1, 1)iq = α0 + α1ES_Stringiq /MBE_Stringiq + α2Sizeiq + α3Lossiq
+ α4Tobin’s Qiq+ α5Leviq + α6CFOiq + α7Liqiq+ α8Tangiq + α9ROAi + α10IssueSizeiq + (α11Dacciq + α12Ab_Dexpiq + α13Ab_Prodiq)
+ ∑αtYeariq + ∑αtIndustryiq + εiq (4)
Where: CAR (-1, 1)iq is firm i’s cumulative abnormal returns measured over three days (-
1, +1) around the quarterly earnings announcement date. The abnormal returns are obtained from a Carhart (1997) four-factor model with the CRSP value-weighted index return as the market returns.5 The parameters of the model are estimated over the period day -300 to day -60. In addition, I also use buy-and-hold abnormal returns (BHAR) as an alternative proxy as it can effectively avoid a rebalancing bias in the monthly reference index returns (Lyon, Barber, and Tsai, 1999).
BHAR (-1, 1)iqis firm i’s buy-and-hold abnormal returns measured over three days (-1, +1) around
the quarterly earnings announcement date. The abnormal returns are also obtained from a Carhart (1997) four-factor model with the CRSP value-weighted index return as the market return. If the equity market rewards firms that sustain a string of earnings benchmarks, then I expect the coefficient on ES_Stringiq/MBE_Stringiq will be significantly positive.
5 The Carhart (1997) four-factor model extend Fama-French (1989) three-factor model by including a momentum factor. The momentum is estimated by subtracting the equally-weighted average of the lowest performing firms from the equally-weighted average of the highest performing firm, lagged one month.