5.1 R&D levels analyses
Before I examine the role of prospect incentive in loss-induced R&D relationship, I
first examine the effect of losses on R&D investments without conditioning on prospect
44
incentive (i.e., estimate Eq. (1) without BIG and the interaction term, BIG×LOSS). The
second column of table 3 reports the estimates of Eq. (1) when the dependent variable is
R&D intensity, calculated as R&D expense divided by sales revenue. My result shows a
positive and significant (1% level) coefficient of LOSS, indicating that R&D intensity
increases when the performance of firm fails below zero earnings point. The sign and significance do not change after another explanatory variable (BIG) in Eq. (2) is included. In
other words, the "failure-induced" model of R&D proposed by BT can be modified to
"loss-induced" model of R&D, which in turn support my H1, arguing that zero earnings point
also serve as an aspiration level for firms.
To extend my results from H1, I add the dummy variable, BIG, to further separate
losses (gains) region into two sub-regions, small loss (profit) and big loss (profit). By doing
this, I can capture the effect of loss firms' prospect incentive on R&D by examining the sign
of (𝛽𝛽1+𝛽𝛽4) since firms increase their prospect incentive as their losses become larger
according to PT. My estimates result for Eq. (2) shows a positive and significant (1% level)
sign of (𝛽𝛽1+𝛽𝛽4), indicating that loss firms with greater risk-seeking intention prefer to
choose R&D investments as my H2 predicts.
The results for the control variables in Eq. (1) and (2) are largely consistent with prior
research. I find that R&D investments decreases in firm’s physical assets (TANG), which is
consistent with Aghion et al. (2004), Hsiao and Tahmiscioglu (1997) and Fazzari et al. (1988),
45
but increases in lagged R&D (R&D_1) and investment opportunities (TOBINQ), which are
consistent with Lev and Sougiannis (1996), Dunlap-Hinkler et al. (2007), Lach and
Schankerman (1989) and Bhagat and Welch (1995). While the sign of the LEV's coefficient
supports prior studies (Smith and Warner, 1979; Williamson, 1988), it is insignificant. Finally,
my results for the rest control variables (Z SCORE and OCF) are contradicted with literature.
5.2 Earnings analyses
Panel A of table 4 reports the estimates of Eq. (3) when the dependent variable is
EARN5, average earnings over five years. This table reports earnings level estimates for all firm-years in the sample. The positive coefficient of SIZE is consistent with Fama and French
(1995), however, it is insignificant in both specification of Eq. (3) estimations. My results for
the coefficient of R&D, tangible assets and advertising expenditures are Consistent with Lev
and Sougiannis (1996), they are all positive and the coefficient of R&D is significant at 1%
level in these estimations.
To judge the effect of prospect-based R&D on future earnings, I refer to the sign of (𝛽𝛽1+𝛽𝛽4+𝛽𝛽5+𝛽𝛽7), reflecting the marginal effect of the shift from small loss firms' R&D to big
loss firms' R&D on future earnings, since I presume that prospect incentive, rather than
problem-solving incentive, is an important driver of loss-induced R&D, especially big loss.
My results from panel A support H3, proposing that prospect-based R&D of loss firms
deteriorates future earnings, and the combined coefficient, (𝛽𝛽1+𝛽𝛽4+𝛽𝛽5+𝛽𝛽7), is negative
46
(-4.4622) and highly significant (0.001 level). Comparing with firms with positive earnings,
whose R&D benefits future earnings (a positive and significant coefficient on R&D), my
results not only show that loss firms' R&D decreases future earnings (a negative sign on (𝛽𝛽3+𝛽𝛽6)), but also show that this negative effect increases as losses become larger (a much
more negative sign on (𝛽𝛽1+𝛽𝛽4+𝛽𝛽5+𝛽𝛽7)). These results strongly supports the proposition that
loss firms' betting on R&D substantially decreases future earnings, and could be seen as a
proof of research (e.g., Bowman, 1980) focusing on a negative association between firms',
which are below the reference point, risk taking behavior and future return.
Panel B of Table 4 reports the results when I replace the dependent variable average future earnings over next five years (EARN5) with one-year-ahead earnings (EARN1), and
R&D intensity is measured with R&D capital to sales (RDC) rather than with R&D expense to sales (RDS) as per Panel A. Relative to the Panel A approach, this specification is much
more robust to dependent variable survivorship biases, as the dependent variable requires data for one rather than for five years. Consistent with Panel A, the combined coefficient,
(𝛽𝛽1+𝛽𝛽4+𝛽𝛽5+𝛽𝛽7), is negative (-1.6642) and highly significant (0.001 level). Results for the
other variables in this specification are also broadly similar to those reported in Panel A,
except that BM coefficient becomes negative and ADEX coefficient becomes negative. The
former is consistent with Fama and French (1995) and the latter contradicts Lev and
Sougiannis (1996) while they are both insignificant.
47
5.3 Further analyses
Given the argument that prospect-based R&D deteriorate future earnings is verified in
H3, I perform an additional tests to see whether higher financial reporting quality curb this
unproductive activity by reducing firms' inclination to overinvest in R&D (appropriateness of
R&D intensity).
To examine whether reporting quality curbs loss firms' inclination to overinvest in
R&D, I first add a new variable, RQ, to Eq. (2) and interact it with the main variables in Eq.
(2), thereby giving my Eq. (4) and its estimates results are shown in table 5. The combined
coefficient, (𝛽𝛽5+𝛽𝛽7), in Eq. (4) is of my primary interest. My results show a significant
negative sign (-0.0308) on (𝛽𝛽5+𝛽𝛽7), indicating that the effect of prospect incentive on R&D
is less positive for firms with higher quality financial reporting, providing support for H4. In
other words, high reporting quality mitigates the magnitude of moral hazard-induced R&D
(the combined coefficient, (𝛽𝛽1+𝛽𝛽4)), of which is likely to be overinvested, and thus
improving loss firms future earnings, providing support for H4.
5.4 Robustness tests
In further untabulated analyses I conducted several robustness tests for the results
documented in Tables 3 to 5. To illustrate, I follow Eberhart et al. (2008) to use total assets
and market capitalization of common stock to scale R&D (either RDE or RDC). The sign and
significance of coefficients of my main variables in Eq. (1) to (4) remains stable. With respect
48
to Table 4, I follow Ciftci and Cready (2011) to use operating income before depreciation
(OPBDP from Compustat), advertising (XAD from Compustat) and R&D expenditures
(XRD from Compustat) as an alternative to earnings. The combined coefficient, representing
prospect-based R&D is negative and remains highly significant.
Since my sample includes firms with missing value of R&D, I could suffer potential
sample selection bias. To address this problem, I exclude those observations instead of
resetting them to zero, as Ciftci and Cready (2011) name this sample “R&D firms”. With the
new sample I used, my untabulated results show that all of the hypotheses are still validated.