Sample characteristics

Among the 140,378 observations, I find that 23,573 (16.79%) observations

have LOSS=1, as their annual income before interest and tax (EBIT) and the lagged

R&D is negative. The remaining 116,805 observations have LOSS=0, because their

EBIT and the lagged R&D is positive.¹⁴ Normal management in the final sample

accepts positive net present-value opportunities since doing so is irrelevant to the

predicted sign of reported income. Table 1 reports descriptive statistics for the sample.

To mitigate the influence of potential outliers, I winsorize all continuous variables at

the 1% and 99% levels. Respecting my dependent variables, the mean and median

values of RDE are 0.12 and 0 respectively, while RDC are 0.33 and 0.01 respectively.

The mean and median of EARN5 are -0.11 and 0.10 respectively, while EARN1 are

-0.28 and 0.09. Respecting my primary independent variables, the mean and median

values of LOSS dummy are 0.16 and 0 respectively, implying that losses are relatively

infrequent and transient as per Hayn (1995). As mentioned earlier, I create a dummy

variable, BIG, and incorporate it with LOSS to capture firms’ risk seeking incentive

instead of traditional median ROE approach (e.g., Miller and Bromiley, 1990;

14 To avoid biases favor my hypotheses, I discard 6,983 observations. These observations have

positive income before tax and the lagged R&D but their income becomes negative once the lagged R&D expenditures are deducted from the income. Baber et al. (1991) show that management in these observations tend to reject some positive net present value opportunities since doing so yields positive income.

35

Fiegenbaum, 1990; Jegers, 1991; Sinha, 1994; Gooding et al., 1996), thereby yielding

0.48 and 0 for the mean and median respectively. As for RQ, I multiply it by negative

one so that higher RQ indicates higher reporting quality. As a result, RQ has negative

values with the mean and median of -0.13 and -0.08 respectively. The distribution of

RQ is similar Biddle et al. (2009).

Table 2 reports the Pearson and Spearman correlations among my variables.

Consistent with my predictions, I find that there is a positive and significant

correlation between LOSS, BIG and RDE. It therefore provides us preliminary

evidence on my hypotheses (1) and (2) stating that losses drive firms to invest more in

R&D and prospect incentive strengthen this inclination even more. As for hypotheses

(3), I find that LOSS is significantly negative associated with EARN5 and EARN1,

and significantly positive associated with R&D, which is consistent with Hayn (1995)

and Joos and Plesko (2005). However, it is difficult to explain the association among

BIG and R&D, LOSS, EARN5, EARN1 since there exists opposite correlations

between BIG and the others and some of these correlations are insignificant. It also

yields little information for us to explain BIG alone since it captures profit firms’

prospect incentive and that is not of my main interest. Finally, respecting hypotheses

(4), I find that RQ is negatively associated with LOSS, which induces prospect

incentive, and positively associated with EARN5 and EARN1. Again, since my main

36

interest is “loss firms’ prospect incentive”, I cannot refer to BIG alone and therefore

cannot observe whether the prospect-based R&D is negatively associated with RQ.

My correlation results for other control variables are substantially consistent with

prior studies mentioned before. For example, a positive and significant correlation

between Z SCORE, lagged R&D, TOBINQ and RDE; a negative and significant

correlation between TANG, LEV and RDE. There is also a high correlation between

SIZE, BM, TANG and EARN5.

37 TOBINQ 0.31575 0.30419 -0.01990 0.40448 0.51592 -0.06388 0.31154 -0.38507 -0.25621 -0.06743 -0.51532 0.06777 -0.02464 -0.04301 -0.10892 LEV -0.34415 -0.34438 -0.11057 -0.22208 -0.60965 0.36016 -0.33979 -0.46442 0.09823 0.22826 0.39389 -0.04992 0.00480 0.01952 0.10431 OCF -0.12579 -0.14497 -0.33976 0.18994 0.17215 0.30574 -0.13356 0.13446 0.00578 0.21215 0.04944 -0.00886 0.04396 0.17574 0.08643

Bold correlations are NOT significant at 5%. R&D is either RDE or RDC. The former is the R&D expense (XRD from Compustat) to sales revenue (SALE from Compustat) in year t, the latter is R&D capital to sales ratio. R&D capital is calculated assuming 20% amortization rate. For firms with zero or missing R&D expenditures, RDE (RDC) equals 0. LOSS is a dummy variable equal to 1, if the sum of income before interest and tax (EBIT from Compustat) and the lagged R&D is negative in year t, and 0 otherwise. BIG is a dummy variable equal to one, if the earnings to one-year lagged total assets is above (below) the median value of profit (loss) firms’ earnings to one-year lagged total assets in year t and industry j, and zero otherwise. Z SCORE is bankruptcy risk as in Altman’s Z-score of firm i at the end of year t-1. TANG is the net book value of property plant and equipment (PPENT from Compustat), scaled by total assets (AT from Compustat) of firm i at the end of year t-1. RDE_1 is R&D expense (XRD from Compustat) to sales revenue (SALE from Compustat) in year t-1. TOBINQ is the

38

ratio of the market value to the book value of total assets of firm i at the end of year t-1. LEV is the ratio of long-term debt (DLTT from Compustat) to the sum of long-term debt plus the market value of equity (DLTT + CSHPRI × PRCC_F) of firm i at the end of year t-1. OCF is net cash flow from operating activities (OANCF from Compustat) divided by sales revenue in year t. SIZE is the natural logarithm of total assets of firm i at the end of year t-1. BM is calculated as stockholders’ equity (SEQ from Compustat) divided by market value of equity (CSHPRI × PRCC_F from Compustat). ADEX is advertising expenditures (XAD from Compustat) to sales ratio. For firms with zero or missing advertising expenditures, ADEX equals 0. EARN5 is the average earnings (EBIT + XRD) over five subsequent years from t+1 to t+5 divided by sales revenue in year t. EARN1 is the earnings before interest, tax and R&D in year t+1 divided by sales revenue in year t. RQ is financial reporting quality derived by Dechow and Dichev (2002) and modified by Mc Nichols (2002) of firm i at the end of year t-1.

39

The table reports results of estimating Equation (1) and (2) using pooled OLS regressions. The sample contains 19,788 firm-year observations over 1993-2006. *, **, *** statistically significant at 10%, 5%, 1%, respectively.

T-Statistics are reported in parenthesis. P-values are one-tailed for statistical tests with directional predictions and two-tailed otherwise. All continuous variables are winzorized at 1% and 99%. RDE is R&D expense (XRD from Compustat) to sales revenue (SALE from Compustat) in year t. For firms with zero or missing R&D expenditures, RDE equals 0. BIG is a dummy variable equal to one, if the earnings to one-year lagged total assets is above (below) the median value of profit (loss) firms’ earnings to one-year lagged total assets in year t and industry j, and zero otherwise. LOSS is a dummy variable equal to 1, if the sum of income before interest and tax (EBIT from Compustat) and the lagged R&D is negative in year t, and 0 otherwise. Z SCORE is bankruptcy

40

risk as in Altman’s Z-score of firm i at the end of year t-1. TANG is the net book value of property plant and equipment (PPENT from Compustat), scaled by total assets (AT from Compustat) of firm i at the end of year t-1.

RDE_1 is R&D expense (XRD from Compustat) to sales revenue (SALE from Compustat) in year t-1. TOBINQ is the ratio of the market value to the book value of total assets of firm i at the end of year t-1. LEV is the ratio of long-term debt (DLTT from Compustat) to the sum of long-term debt plus the market value of equity (DLTT + CSHPRI × PRCC_F) of firm i at the end of year t-1. OCF is net cash flow from operating activities (OANCF from Compustat) divided by sales revenue in year t. SIZE is the natural logarithm of total assets of firm i at the end of year t-1.Industry indicators are created based on Fama and French 48-industry definitions.

Table 4

Earnings analyses.

𝑅𝑅𝑇𝑇𝑅𝑅𝑇𝑇𝑖𝑖𝑖𝑖 = 𝛽𝛽0+ 𝛽𝛽1𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖 + 𝛽𝛽2𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑖𝑖+ 𝛽𝛽3𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖+ 𝛽𝛽4𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖× 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑖𝑖+ 𝛽𝛽5𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖 × 𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖+ 𝛽𝛽6𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑖𝑖× 𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖 + 𝛽𝛽7𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖× 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑖𝑖× 𝑅𝑅𝑅𝑅𝑖𝑖𝑖𝑖 + 𝛽𝛽8𝐿𝐿𝑇𝑇𝑍𝑍𝑅𝑅𝑖𝑖𝑖𝑖 + 𝛽𝛽9𝑇𝑇𝐵𝐵𝑖𝑖𝑖𝑖+ 𝛽𝛽10𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑖𝑖𝑖𝑖−1+

𝛽𝛽11𝑇𝑇𝑅𝑅𝑅𝑅𝐴𝐴𝑖𝑖𝑖𝑖 + 𝑇𝑇𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝑖𝑖𝐼𝐼𝐼𝐼 & 𝑌𝑌𝑌𝑌𝑌𝑌𝐼𝐼 𝑇𝑇𝐼𝐼𝐼𝐼𝑖𝑖𝐼𝐼𝑌𝑌𝑖𝑖𝐼𝐼𝐼𝐼𝐼𝐼 + 𝜀𝜀𝑖𝑖𝑖𝑖 (3) Panel A: Regressions of average earnings over years t+1 to t+5 (EARN5)

Variable Pred Eq. (3)

41

(0.47)

Adjusted R-squared 0.0206

N 19,788

Panel B: Regressions of earnings in year t+1 (EARN1)

Variable Pred Eq. (3)

This table presents the coefficient estimates from pooled cross-section regressions of Eq. (3). The sample contains 19,788 firm-year observations over 1993-2006. *, **, *** statistically significant at 10%, 5%, 1%, respectively. T-Statistics are reported in parenthesis. P-values are one-tailed for statistical tests with directional predictions and two-tailed otherwise. All continuous variables are winzorized at 1% and 99%. EARN5 is the average earnings before interest, tax and R&D (EBIT plus XRD from Compustat) over five subsequent years from t+1 to t+5 divided by sales revenue (SALE from Compustat) in year t. EARN1 is the earnings before interest, tax and R&D in year t+1 divided by sales revenue in year t. BIG is a dummy variable equal to one, if the earnings to one-year lagged total assets is above (below) the median value of profit (loss) firms’ earnings to

42

one-year lagged total assets in year t and industry j, and zero otherwise. LOSS is a dummy variable equal to 1, if the sum of income before interest and tax (EBIT from Compustat) and the lagged R&D is negative in year t, and 0 otherwise. RDE is R&D expense (XRD from Compustat) to sales revenue in year t. RDC is R&D capital to sales ratio. R&D capital is calculated assuming 20% amortization rate. For firms with zero or missing R&D expenditures, RDE (RDC) equals 0. SIZE is the natural logarithm of total assets of firm i at the end of year t-1.

BM is the stockholders’ equity (SEQ from Compustat) divided by market value of equity (CSHPRI times PRCC_F from Compustat). TANG is the net book value of property plant and equipment (PPENT from Compustat), scaled by total assets (AT from Compustat) of firm i at the end of year t-1. ADEX is advertising expenditures (XAD from Compustat) to sales ratio. For firms with zero or missing advertising expenditures, ADEX equals 0. Industry indicators are created based on Fama and French 48-industry definitions.

Table 5

43

This table presents the coefficient estimates from pooled cross-section regressions of Eq. (4) and (5). The sample contains 19,788 firm-year observations over 1993-2006. *, **, *** statistically significant at 10%, 5%, 1%, respectively. T-Statistics are reported in parenthesis. P-values are one-tailed for statistical tests with directional predictions and two-tailed otherwise. All continuous variables are winzorized at 1% and 99%. RDE is R&D expense (XRD from Compustat) to sales revenue in year t. RDC is R&D capital to sales ratio. R&D capital is calculated assuming 20% amortization rate. For firms with zero or missing R&D expenditures, RDE (RDC) equals 0. BIG is a dummy variable equal to one, if the earnings to one-year lagged total assets is above (below) the median value of profit (loss) firms’ earnings to one-year lagged total assets in year t and industry j, and zero otherwise. LOSS is a dummy variable equal to 1, if the sum of income before interest and tax (EBIT from Compustat) and the lagged R&D is negative in year t, and 0 otherwise. RQ is financial reporting quality derived by Dechow and Dichev (2002) and modified by Mc Nichols (2002) of firm i at the end of year t-1. Z SCORE is bankruptcy risk as in Altman’s Z-score of firm i at the end of year t-1. TANG is the net book value of property plant and equipment (PPENT from Compustat), scaled by total assets (AT from Compustat) of firm i at the end of year t-1. RDE_1 is R&D expense (XRD from Compustat) to sales revenue (SALE from Compustat) in year t-1. TOBINQ is the ratio of the market value to the book value of total assets of firm i at the end of year t-1.

LEV is the ratio of long-term debt (DLTT from Compustat) to the sum of long-term debt plus the market value of equity (DLTT + CSHPRI × PRCC_F) of firm i at the end of year t-1. OCF is net cash flow from operating activities (OANCF from Compustat) divided by sales revenue in year t. SIZE is the natural logarithm of total assets of firm i at the end of year t-1.