3.1 Data
The sample of product recall announcements was collected from LexisNexis
Academic during 1991 to 2008. We applied the structured content analysis approach
(Jauch, Osborn and Martin, 1980) to identify if the recall announcements belong to
voluntary or involuntary strategy. Following Siomkos and Kurzbard (1994), we
searched for voluntary product recalls using the keywords “voluntary”, “improve” and
“precaution” in the announced news, and the involuntary product recalls using the
keywords “order”, “mandatory” and ”refuse”. If there are multiple recall
announcements for the same product, only the first one is included in the sample. In
addition, we deleted the recall announcements when other important news is released
around the product recall announcement date to avoid their confounding effects on
stock market reactions1
To measure the abnormal returns, we require the sample firms to be listed on
the New York Stock Exchange (NYSE), American Stock Exchange (AMEX) and
NASDAQ. The daily stock return data is obtained from the Center for Research
Security Prices (CRSP) database. We draw on the Compustat database for a firm’s
financial information. The final sample includes announcements of 355 voluntary
recalls, and 55 involuntary recalls.
.
3.2 Event Study Analysis
Event study methodology has been widely applied in marketing and
management research (Mathur and Mathur, 2000; McWilliams and Siegel, 1997;
Sorescu, Shankar and Kushwaha, 2007). Based on the Market Efficiency Hypothesis
(Fama, 1998), the financial impact of product recalls should be quickly reflected in
stock price changes when the information is disclosed to the public. Thus, The
1
valuation impacts of an event is measured by the abnormal changes in share prices
surrounding the time of the event announcements.
We measure abnormal stock returns by subtracting expected returns from real
returns. The expected stock returns are measured by estimating the following equation
with the ordinary least squares technique: rit = +ai b Ri mt + , where eit r = the stock it
return for firm i on day t; a and i b are the regression parameters for firm i; i R mt
is the return of the market portfolio on day t; and e is the residual term. The it
abnormal returns AR associated with recall announcements are measured by the it
following equation: ARit = −rit (aˆi+b Rˆi mt), where AR is the abnormal return for it
firm i on day t, and ˆa and ˆi b are parameters estimated for the i a and i bi in
the preceding equation by using a 140-day estimation period prior to the
announcement day (Chen, Ho, Ik and Lee, 2002). We use the cumulative abnormal
returns on the announcement date and one day before the announcement date, CAR
(-1,0) as the dependent variable in the empirical analyses.
3.3 Variables
The first independent variable is a binary variable, VOLUNTARY, that takes a
value of 1 for voluntary product recalls, and zero for involuntary recalls. The second
independent variable, CSR, is a firm’s CSR performance. We collect information of
CSR ratings from KLD STATS. CSR is an indicator variable that equals 1 if a firm’s
number of CSR strengths minus the number of weaknesses is nonnegative, and 0
otherwise (Siegel & Vitaliano, 2007). Furthermore, based on the CSR classification in
KLD STATS provided in Mattingly and Berman (2006), we measure technical and
institutional CSR. Specifically, TCSR is a dummy variable that is 1 if the number of
strengths minus the number of weakness in the items of TCSR is nonnegative, and 0
otherwise. The variable ICSR is similarly defined.
We control several important factors on the market reactions to product recall
announcements in the regression analyses. REPUTATION measures a firm’s
reputation based on the rating scores in the annual America’s Most Admired
Companies survey appeared in Fortune magazine in the product recall year (Chen et
al., 2009). To capture the effect of serious incidents on the value assessment of
investors, we use a dummy variable INJURY that equals 1 if injury is reported in the
recall news, and 0 otherwise.
Chen et al. (2009) shows that the stock market reactions to recall
announcements may be related to the recalling firm’s size and financial leverage. We
thus control those two factors in the analyses. The variable SIZE is measured by the
logarithm of book value of total assets for the fiscal year of the product recall
announcements (Chen et al., 2002). The financial leverage, LEV, is measured as the
book value of total debt divided by the book value of total assets for the fiscal year
preceding the announcement (Lang, Ofek and Stulz, 1996). In addition, product
recalls may have a greater impact on firms that have more growth opportunity
because the valuable growth options may disappear due to the recall events. Thus, we
include the growth opportunity variable, TOBIN’S Q, measured by the ratio of the
market to book value of a firm’s assets, where the market value of assets equals the
book value of assets minus the book value of common equity plus the market value of
common equity (Chen et al., 2002). To capture possible trends of the impact of recalls
on returns, a time variable, Year trend effect is included (Chen et al., 2009) in the
analyses, measured by the number of years between 1991, the earliest sample year,
and the year of the recall. Finally, we control for the industry-specific effects by
including the industry dummy variables based on the two-digit Standard Industry
Classification (SIC) codes.
4. Results
4.1 Descriptive Statistics
Table 1 illustrates the distribution of the entire sample. The majority of the
recall announcements occurred after 2000. For the industrial distribution, the product
recalls in the sample were made by firms operating in 20 industries based on the
2-digit SIC codes. The majority of recalls announcements happened in the
manufacturing industries (SIC codes from 20 to 39), and the Transportation
equipment industry (SIC code 37) accounts for around 30% of the entire sample.
--- Insert Table 1 Here ---
Panel A of Table 2 presents descriptive statistics for the variables in the
analyses. For CSR performance and firm reputation, we find there is no significant
difference between the voluntary and involuntary subsamples. The results indicate
that injuries occur significantly more frequently for involuntarily recalled products
than voluntarily ones. The evidence for TOBIN’S Q suggests that firms involving
voluntary recalls have significantly greater growth opportunity than those with
involuntary recalls. The non-parametric statistics for the difference in median are very
similar. Finally, no significant difference is found in the leverage ratio between
voluntary and involuntary recalls. Panel B presents the Pearson correlations among
the variables.
--- Insert Table 2 Here ---
4.2 Analyses of Impact of CSR Performance on Recall Strategies
Table 3 presents the comparison in CAR based on the recall strategy and CSR
performance. Panel A shows that while voluntary and involuntary recalls are both
associated with negative announcement returns, voluntary recalls receive marginally
greater abnormal returns than involuntary recalls.2
Panel C considers both recall strategy and CSR performance simultaneously.
The findings show that for voluntary recalls, the mean CAR for high-CSR firms is
positive (0.04%), although statistically insignificant. On the contrary, the CAR for
low-CSR firms is significantly negative (-0.54%). The difference in mean CAR
between high- and low-CSR firms is statistically significant at the 5% level. For
involuntary recalls, firms in both high- and low-CSR groups receive negative mean
CAR, and the difference is not statistically significantly different from zero. The Pane B compares CAR based on
CSR performance. The results show that high-CSR firms have an insignificant mean
CAR of -0.12% (t = 0.55), and, in a sharp contrast, low-CSR firms experience a
significantly negative mean CAR of -0.58% (t = -3.12). The difference in CAR is
statistically significant at the 10% level. The results suggest a firm’s CSR
performance may be considered as a positive signal to investors in their assessment of
the valuation impacts of product recalls.
2
overall evidence in Panel C suggests that CSR plays an important role in investors’
value assessment for voluntary recalls, but not involuntary recalls.
--- Insert Table 3 Here ---
In Table 4 we further decompose CSR ratings into TCSR and ICSR
components based on the activity classification in Mattingly and Berman (2006).
Panel A shows that there is no significant difference in the mean CAR between the
high- and low-ICSR subsamples. However, the results show a very different pattern
when focusing on TCSR. The mean CAR for the high-TCSR firms is significantly
positive, but that for the low-TCSR firms is significantly negative, and the difference
is strongly significant at the 1% level (t = 3.27). The results imply that the information
content of product recalls is more related with the involvement in TCSR activity of
the recalling firms.
Panel B presents the impacts of TCSR and ICSR on the abnormal returns of
recall strategy. Consistent with the evidence in Panel A, ICSR rating does not yield
important difference in the mean CAR for both voluntary and involuntary recall
announcements. However, for TCSR rating we find that high-TCSR firms have
significantly greater abnormal returns than low-TCSR firms in the voluntary recall
evidence strongly indicates that the result for overall CSR is mainly driven by TCSR
performance, rather than ICSR. In sum, the evidence presented in Tables 3 and 4
suggests that how investors perceive and evaluate product recalls does not only rely
on the recall strategy per se, it is also dependent on the CSR ratings of the announcing
firms, particularly the TCSR performance.
--- Insert Table 4 Here ---
4.3 Cross-sectional Regression Analyses
Table 5 presents the cross-sectional regression analyses of the recall
announcement abnormal returns. To control for the possibility of heteroskedasticity in
the sample, the t-values reported are computed with heteroskedasticity-consistent
standard errors (White, 1980). Model 1 tests the effect of product recall strategy and
CSR for the overall sample. The results show that the coefficient of VOLUNTARY is
insignificantly positive, implying voluntary recalls per se do not necessarily result in
better result in firm value change than involuntary recalls. The effect of the CSR
dummy is positive and statistically significant at the 5% level, suggesting that CSR
performance has a strong impact on the value assessment during product recalls
announcements. Model 2 further identifies which part of CSR activities creates the
significantly positive impact on abnormal returns, but the effect of ICSR is found to
be very weak. The findings indicate that the positive influence of CSR in Model 1 is
driven by TCSR performance. The evidence in Models 1 and 2 lends strong support to
Hypotheses 1 and 3.
To test the interaction effect of CSR and recall strategy, we stratify the overall
sample based on recall strategies. In the subsample of voluntary strategy, Model 3
shows that CSR is strongly positively associated with abnormal returns of product
recalls. In contrast, in the subsample of involuntary strategy, Model 5 shows that the
effect of CSR is statistically insignificant. Moreover, when considering the
component of CSR activities, we find insignificant effects of ICSR for both voluntary
and involuntary recalls in Models 4 and 6. The coefficients of TCSR, however, are
significantly positive for the voluntary subsample, but statistically insignificant for the
involuntary subsample. The overall results in Table 5 are consistent with those in
Panels A and B in Table 4.
For the control variables, we find LEV is positively and significantly related to
abnormal returns for the whole sample. But the effects of most of the other control
variables are statistically insignificant. To test if there is a multicollinearity problem,
we use the variation inflation factor (VIF) in all regression models. The values of VIF
ranges from 1.06 to 1.96, suggesting there is little problem of multicollinearity.
--- Insert Table 5 Here ---
A number of robustness checks are employed to test the sensitivity of our
findings. Firstly, while the choice of event window CAR (-1,0) is not uncommon
based on previous studies, it may not completely catch the announcement effect of
product recalls. To test if the choice of event window could bias the empirical results,
we further test different events windows of CAR (-1,1) and CAR (-2,2) as the
alternative independent variables and re-test the regression analyses. We find that for
the event window CAR (-1,1), the results of VOLUNTARY (t = 0.83), CSR (t = 2.82),
ICSR (t = 0.36) and TCSR (t = 3.26) are all consistent with those presented in Table 5.
The evidence for CAR (-2,2) is very similar. Secondly, the measure of CSR dummy is
categorical, and thus, may not catch the linear trend of CSR on abnormal returns. We
thus also try a continuous variable of CSR performance measured by the number of
strength minus the number of weakness of CSR performance in the KLD databank,
and re-do Models 1, 3 and 5 in Table 5. The result of the continuous measure of CSR
remain the same, significantly positive at the 5% level for Model 1 (t = 2.63) and the
1% level for Model 3 (t = 3.34), and insignificant in Model 5 (t = 0.99). Similarly, we
also test the results by the continuous measures of TCSR and ICSR, and find the
widely used in measuring abnormal returns in the event study approach, prior studies
had employed alternative approach to measure abnormal returns. We follow Yermack
(1996) and measure a firm’s abnormal return at day t by subtracting market return
from the actual return on the same day. We find that the results from this alternative
measure of abnormal returns are essentially the same with those reported in Table 5.
Thus our findings are not sensitive to different approaches of measuring abnormal
returns. Finally, Godfrey et al. (2009) also investigate the influences of TCSR and
ICSR with a slightly different measure from our study. In Godfrey et al.’s study
(2009), TCSR (ICSR) is measured as a dummy variable that is equal to one if the firm
scored greater than zero on any of the positive items related to TCSR (ICSR) in the
KLD database, while in this study TCSR is one when the number of strengths is equal
to or greater than the number of weakness in the related items3
3 The related items of TCSR are those under the governance, employee relations, or product relations
. To check if our
findings on TCSR and ICSR are sensitive to different measures, we follow the
measures of TCSR and ICSR in Godfrey et al. (2009) and redo the analyses in Table 5.
The results are very similar. With the alternative measures, the results for voluntary
recalls show that the coefficient of TCSR is positive and significant (t = 1.82), and
that of ICSR is insignificant (t = -0.22); for involuntary recalls, both TCSR and ICSR
are statistically insignificant. Thus, our findings are robust to the alternative measures
of TCSR and ICSR.