This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Marketing Science Institute Working Paper Series 2014 Report No. 14-112 Should Ad Spending Increase or Decrease Prior to a Product Recall Announcement? Haibing Gao, Jinhong Xie, Qi Wang, and Kenneth C. Wilbur
Report Summary Suppose a product harm crisis is going to be announced in two weeks. What should you do with your advertising before the recall announcement: spend more, spend less, or maintain your current plan? Recent research argues that advertising should fall before a crisis is announced because the harm of the recall will reduce the short-term benefits of advertising, reducing profitability. However, a product harm crisis often not only reduces product profit but also damages firm value. While a retreat of pre-recall advertising avoids inefficient marketing spending on the recalled product, will investors interpret it as an admission of deeper systemic problems? More generally, when and how does pre-recall advertising affect post-recall stock price? To answer this question, Haibing Gao, Jinhong Xie, Qi Wang, and Kenneth Wilbur investigate how automakers’ stock prices reacted to 110 product recall announcements between 2005 and 2012. Their analysis finds that stock price is significantly affected by pre-announcement ad expenditures, and that the direction of this effect depends on the seriousness of the product defect and whether the recalled product is a newly introduced model.
Their analysis uncovers two primary effects. First, in the case of a minor recall affecting a newer model, a steep increase in pre-recall advertising raises the automaker’s cumulative abnormal returns. The authors call this a signaling effect, as investors have less prior information about newer models and increasing pre-recall advertising can underscore the fact that the hazard is not very serious. Second, when the recall is a major hazard affecting an older model, the opposite effect occurs. That is, increasing advertising expenditures prior to announcing the recall worsens the automaker’s cumulative abnormal stock returns. The authors call this an expectations effect: the increase in advertising sets up investors’ expectations of good future performance, which are then disappointed upon learning the details of the recall. These findings show that pre-recall advertising is one tool a firm can use to strategically soften the negative impact of a product recall on stock market value. Firms are typically aware of a pending recall (whether firm-initiated or government-initiated) before it is announced and can therefore act prior to the announcement. However, the optimal reaction requires an understanding of the type of hazard and the type of product. The authors’ data imply that automakers may not fully understand the existence of these effects, as there have been many cases in which firms have increased pre-recall ad spending for older models with major hazards. In these cases, the firm could have benefited by foregoing an advertising increase for the model in question, but did not do so. Overall, firms anticipating a recall announcement should consider the seriousness of the defect and the novelty of the product. To minimize damage to firm value, in the case of minor harm to new products, pre-recall advertising should be increased; in the case of major harm to older products, pre-recall advertising should be held constant or reduced. Finally, when recalling a new product, marketers should consider the possible trade-offs between product profitability and stock price. Haibing Gao is a doctoral student in marketing at the University of Florida. Jinhong Xie is JC Penny Eminent Scholar Chair and Professor of Marketing at the University of Florida. Qi Wang is Associate Professor of Marketing at State University of New York at Binghamton. Kenneth C. Wilbur is Assistant Professor of Marketing at the University of California, San Diego.
Marketing Science Institute Working Paper Series 1
Product recalls have been on the rise over the past decade. As shown in a recent report
by the ACE Group, one of the world’s largest property and casualty insurers, 2,363 recalls of
consumer products, pharmaceuticals, medical devices and food took place in the United States
in 2011, representing a 14 percent increase from the prior year, and a 62 percent increase from
2007.1 Based on auto recall announcements published by the National Highway Traffic
Safety Administration (NHTSA), the average number of recalls per year for all manufacturers
selling vehicles in the U.S. was 339 from 1994-2003, but 599 from 2004-2013, which
represents an increase of 76 percent between the two 10-year periods.2 This uptrend in
product recall is driven by several emerging forces, such as increased globalization of
production, growing complexity of products, and more stringent product-safety legislation
(Chen, Ganesan, and Liu 2009; Dawar and Pillutla 2000). As these emerging forces further
develop, firms are expected to face an even higher risk of a product-harm crisis (Chen and
Nguyen 2013).
Product recalls can impose severe financial damage on the firms involved. Over the
last several decades, Wall Street has witnessed many instances where a firm’s share price was
slashed after a recall announcement. Consider several recent cases: Boston Scientific’s stock
price fell 13 percent after announcing a recall of its implantable defibrillators on March 15
2010;3 Cochlear’s share price promptly dropped 20 percent after a voluntary recall of its
Nucleus 5 implant products on September 12 2011;4 Toyota’s shares lost about 22 percent in
less than two weeks after its recall of 2.3 million vehicles in the U.S. due to problems related
to accelerator pedals in 2010;5 and, most recently, as reported by USA Today on April 11,
2014, “GM stock below IPO price as recall talk swirls.” 6 The harmful financial
consequences of product recalls are not merely “bad luck” that only occurs occasionally. The
finance literature has extensively investigated the impact of product recalls on firm value. A
negative relationship between a product recall announcement and the recalling firm’s stock
price has been found in a wide range of industries, including (but not limited to) automobiles,
pharmaceuticals, food, toys, electronics, cosmetics, and outdoor products (see Barber and
Darrough 1996; Chen and Nguyen 2013; Chu, Lin, and Prather 2005; Davidson and Worrell
and Helsen 2008; Zhao, Zhao, and Helsen 2011), sales and profits (Rubel, Naik, and
Srinivasan 2011), and the effectiveness of marketing strategies (Cleeren, Van Heerde, and
Dekimpe 2013; Van Heerde, Helsen, and Dekimpe 2007). However, product recalls not only
damage firms’ marketing performance but also harm their financial value, which calls for
research that integrates the two perspectives. For example, a recent marketing study (Rubel,
Naik, and Srinivasan 2011) discovered that, when envisioning a product harm crisis, it is
optimal for the firm to reduce pre-crisis advertising to maximize the brand profit as the crisis
likelihood (or the damage rate) increases. While a downward adjustment in pre-recall
advertising spending can be beneficial from a marketing perspective, it is unknown whether
or not such a change would intensify the ex-post harm to the firm’s financial value.
Investigating the impact of pre-recall advertising on post-recall abnormal stock returns can
help to identify conditions under which the marketing and finance objectives clash or agree.
Such an understanding facilitates the design of an overarching crisis management strategy,
which maximizes the firm’s overall interests.
Second, in practice, firms often anticipate a product recall long (months to years)
7 One exception is Chen, Genesan and Liu (2009), which examines how a firm’s product-recall strategy (Proactive vs. Passive) affects its financial value based on an event study of recalls announcements by the Consumer Product Safety Commission (CPSC).
Marketing Science Institute Working Paper Series 3
before the actual recall announcement. Consider the automobile industry, where product
safety recalls can be classified into two categories: firm-initiated and government (i.e.,
NHTSA)-initiated. According to the NHTSA’s records, the majority of product recalls are
initiated by the firms.8 For such recalls, the manufacturer determines whether a safety defect
exists through its own inspection procedures and information-gathering systems and decides
if and/or when to issue a recall. Recalls initiated by the NHTSA often involve a lengthy
procedure, consisting of a preliminary investigation (taking 120 days on average) and an
engineering analysis (taking 365 days on average).9 During this process, manufacturers are
required to provide necessary information (e.g., data on complaints, crashes, injuries,
warranty claims, modifications, and part sales) to the NHTSA, and have the opportunity to
present their own views regarding the alleged defect or present their new analysis. The
standard recall procedure10 suggests that, for both types of recalls, manufacturers often have
time to make ex-ante strategic decisions before the recall is formally announced. Thus,
pre-recall advertising as a strategic variable for managing a product harm-crisis is not only
theoretically desirable but also practically implementable. We offer the first prescriptive
advice about the conditions under which an automobile manufacturer might want to increase
advertising in order to build investor confidence in anticipation of a product recall.
Drawing from both marketing and finance literature, we propose two possible effects of
pre-recall advertising spending on post-recall investors’ behavior. (1) A positive (negative)
signaling effect – a high (low) level of pre-recall advertising spending, due to information
asymmetry between the firm and investors, can signal the firm’s high (low) confidence in
future demand based on its private information about the recalled products (e.g., the severity
of the defect, possible consumer responses, and its potential impact on future profitability);
and (2) a negative expectation effect – a high level of pre-recall advertising raises
expectations, which in turn leads to a high level of post-recall disappointment (due to
unfulfilled expectations). The overall impact of pre-recall advertising on post-recall firm
value depends on the accumulative strength of these two forces, which may, in turn, depend
on the specific characteristics of the recall. Specifically, we expect a stronger signaling effect
when the recall involves newly introduced models (information asymmetry is likely stronger
for new models than for older models) and a stronger expectation effect when the recall is
8 For example, our data set consists of recalls for the six largest automakers (Toyota, Honda, Nissan, General Motors, Ford, and Chrysler) from 2005 to 2012. Among them, 61.8% are firm-initiated recalls. 9 www.sesptc.com/2010Presentations/NHTSA_ODI_SESPTC2010.ppt 10 See http://www.nhtsa.gov/Vehicle+Safety/Recalls+&+Defects/Motor+Vehicle+Safety+Defects+and+ Recalls+Campaigns for more detailed discussion about the recall procedure.
Marketing Science Institute Working Paper Series 4
where the dummy variables incij and decij refer to an increase and a decrease in pre-recall
advertising for recalled products, respectively. Specifically, if firm i increases its advertising
for the anticipated recalled products before the recall j, incij is equal to 1 and decij is equal to
0. In contrast, if firm i decreases its advertising for the recalled products before the recall j,
incij is 0 and decij is 1. When there is no adjustment in firm i's pre-recall advertising, both incij
and decij are equal to 0. We also incorporate two dummy variables of newij, hazardij, and their
interactions with pre-recall advertising adjustments, to capture the moderating effects of these
two recall characteristics on the impact of the pre-recall advertising. The dummy variable
newij is equal to 1 if the recall involves new models, and the dummy variable hazardij is 1 if
the recall is due to major hazards. A vector of control variables, such as other recall factors
and firm characteristics, is also incorporated into the model. The definitions and
measurements of these control variables are introduced in the next section. To derive the
overall impact of increasing and decreasing pre-recall advertising under certain recall
characteristics, we sum all coefficients related to the advertising adjustment and its
interaction with the specific recall characteristic. To test Hypothesis 1, we must sum the
coefficients of inc and inc*new (i.e., binc + binc*new) and test its significance. Similarly, to test
Hypothesis 2, we sum the coefficients of inc and inc*hazard (i.e., binc + binc*hazard). To test
Hypothesis 3, we consider the cases of both major and minor hazards. We sum the two
coefficients of dec and dec*new (i.e., bdec + bdec*new) for recalls of new models with minor
hazards. We also sum the three coefficients of dec, dec*new, and dec*hazard (i.e., bdec +
bdec*new + bdec*hazard) for recalls of new models with major hazards.
Empirical Analysis
Data
This study examines the impact of pre-recall advertising adjustments on firms’
abnormal returns to safety recalls in the automotive industry. We collected recall data from
the National Highway Traffic Safety Administration (NHTSA). Our sample consists of
vehicle safety recalls by the six largest automakers (Toyota, Honda, Nissan, General Motors,
Marketing Science Institute Working Paper Series 15
Ford, and Chrysler) from 2005 to 2012, as these six automakers account for about 90 percent
of the U.S. motor-vehicle market for cars and light trucks. Following prior studies (Barber
and Darrough 1996; Jarrell and Peltzman 1985; Hoffer, Pruitt, and Reilly 1988), we included
a vehicle recall in the sample if the recall size is adequately large or if it is reported by the
Wall Street Journal (WSJ). Specifically, referring to the practice of Jarrell and Peltzman
(1985), the thresholds of recall size were set proportional to firm size: Toyota 50,000;
General Motors, Ford, and Chrysler 40,000; Honda 30,000; and Nissan 20,000.
We identified the recall announcement date based on the information provided by the
NHTSA and media reports such as the WSJ. If a recall was reported on multiple dates by
multiple sources, we used the earliest one as the recall announcement date. To prevent
contamination of our data by information leakage before the event date, we followed the
literature (e.g., Chen, Genesan and Liu 2009; Davidson and Worrell 1992) and excluded the
recalls for which there were news reports about related accidents and safety issues in the WSJ
before the recall announcement. To rule out potential confounding effects, we also excluded
the recalls for which confounding events were reported in the WSJ, such as earnings surprises,
earnings warnings, new plants, new products, mergers and acquisitions, joint ventures,
bankruptcy, layoffs, and changes in top management. This screening procedure ensures that
the abnormal returns over the event window are strictly due to the announcements of
unexpected vehicle recalls. Our final sample consists of 110 automobile safety recalls.
We collected the stock price and market index data from the Center for Research in
Security Prices (CRSP) at the University of Chicago. Firm characteristics such as firm size
and firm liability were obtained from COMPUSTAT. Firm reputation scores were collected
from annual surveys conducted by Fortune magazine. Recall information and characteristics
were obtained from the NHTSA database. Finally, advertising data were collected from
Kantar Media.
Variables
Pre-recall advertising adjustment. To identify if a firm adjusts its advertising
spending before an anticipated auto recall, we first need to specify a benchmark period and
an adjustment period. Conceptually, the benchmark period is the time period before a recall
when the stock market perceives relatively consistent advertising spending, whereas the
adjustment period is the near-recall period when the market perceives an adjustment in
advertising spending. Figure 1 illustrates these two periods in relation to the event window of
a product recall, where τ0=0 denotes the recall announcement date, [τ0, τ1] denotes the event
Marketing Science Institute Working Paper Series 16
window, and ν1, ν2 denote two cutoff dates for the adjustment and benchmark periods,
respectively. Accordingly, la in Figure 1 denotes the length of the adjustment period between
τ0 and ν1, while lb denotes the length of the benchmark period between ν1 and ν2. Given these
two periods, pre-recall advertising adjustment can be defined by the difference in advertising
spending between these two periods. Specifically, if the firm’s weekly average advertising
spending in the adjustment period is two standard deviations above (below) that in the
benchmark period, we define the advertising adjustment to be increasing (decreasing);
otherwise we define it as no adjustment.11 (Figures and tables follow References.)
To determine the benchmark and adjustment periods empirically, we considered the
lengths of the adjustment period from 1 to 3 weeks before the recall announcement date τ0;
and the lengths of the benchmark period from 3 to 6 weeks, respectively. Then we
experimented with eight different sets of adjustment and benchmark periods. For example,
one choice of adjustment and benchmark periods [la, lb] can be [1, 4], indicating the
adjustment period of one week before the recall announcement date τ0 and the benchmark
period of four weeks. We applied our model to multiple alternative sets of benchmark and
adjustment periods and found consistent results (see the section of Robustness and Validity of
Results). Among them, the model with the adjustment and benchmark periods of [1, 4]
provides the most significant results and the best model fit. Hence, our discussion of the
empirical analyses focuses on and, unless otherwise stated, the estimation results in all tables
are based on the adjustment and benchmark periods of [1, 4]. As shown in Table 1, for this
specific classification, among the 110 auto recalls in our data, the number of cases of
increasing, decreasing, and not adjusting pre-recall advertising is 30, 30, and 50,
respectively.
Two recall characteristics. The first recall characteristic involves the newness of the
recalled model. We classify a recall as a New-Model recall when the recall involves vehicles
introduced in the current or previous year. The second recall characteristic regards the
severity of the defect hazard. Following the literature (Rupp and Taylor 2002; Rupp 2004),
we classify a recall due to a Major Hazard if severe quality defects (such as fuel leakage,
steering problems, acceleration problems, break failure, or repeated stalling, which may cause
fire or car crash) were involved. Both variables were collected from the NHTSA website. As
11 We also tested the categorization of advertising adjustments using a one-standard-deviation cutoff, but this resulted in less significant results and an inferior model fit in the respective regressions, which suggests that only a significant adjustment in advertising spending (i.e., higher or lower than two standard deviations from the average advertising spending in benchmark period) can be detected by the stock market and processed by investors.
Marketing Science Institute Working Paper Series 17
shown in Table 1, out of the 110 automobile safety recalls, 62 recalls involved new models
and 56 were due to major hazards.
Control variables. Our empirical analysis incorporates two types of controls: (1)
product recall factors and (2) characteristics of the recalling firm. In line with the extant
literature on auto recalls (Rupp and Taylor 2002; Rupp 2004, 2005), we included recall size,
airbag recall, recall initiator, and publicity as control variables. Recall size, rcsize, is
measured as the logarithm of the total number of vehicles affected by the recall. The dummy
variable airbag denotes whether a recall is due to a defect in the airbag. The dummy variable
nhtsa denotes if the recall is initiated by the NHTSA rather than by the firm. The information
on these recall factors was also collected from the NHTSA. In addition, we also incorporated
a dummy variable, y2009, indicating whether the recall occurred during or after Toyota’s
2009 recall crisis.
Both digital and print media contribute to the publicity surrounding a safety recall. To
facilitate managerial use, publicity is operated as a categorical variable with four levels:
negligible, local, national, and supranational, where negligible is measured as 0, local as 1,
national as 2, and supranational 3. Specifically, the publicity level is categorized as national
if a product recall was reported by the print media with a total circulation above any of the
five major national newspapers: Wall Street Journal (2,117,796), USA Today (1,829,099),
New York Times (916,911), Washington Post (550,821), and New York Post (522,874).
Publicity is categorized as supranational if a recall was reported with a total circulation
above the sum of two of the five national newspapers with one being WSJ. Publicity is
categorized as local if the recall was reported with a total circulation level ranging from the
Chicago Daily Herald (104,053) to the Houston Chronicle (364,724). A publicity level below
that is then defined as negligible. The circulation data of the print media were collected from
the news database Factiva.
The control variables representing characteristics of the implicated firms include firm
size, firm liability, and firm reputation (Chen, Ganesan, and Liu 2009). Firm size, fsize, is
measured as the logarithm of the firm’s sales revenue in the year of the recall, whereas firm
liability, fdebt, is calculated as the logarithm of the firm’s long-term liability. We collected
firms’ sales revenues and long-term liability from COMPUSTAT. Firm reputation, frep, is an
overall score from Fortune magazine’s annual survey of “America’s Most Admired
Companies.” Finally, two dummy variables inc_u and dec_u are also incorporated to control
the impact of the recalling firm’s adjustments in pre-recall advertising (i.e., increase or
Marketing Science Institute Working Paper Series 18
decrease) for products unaffected by the recall. The classification of advertising adjustments
for unaffected products and the variable definitions are similar to those for recalled products.
Table 2 summarizes all variable definitions, their data sources, and descriptive statistics.
Results
Abnormal returns resulting from product recalls. To examine the cumulative abnormal
returns resulting from the events of product recalls and to minimize potential confounding
events (McWilliams and Siegel 1997), we focus on three relatively short event windows: (1)
the event date (i.e., the day 0); (2) the day after the event date (i.e., the day +1); and (3) both
the event day and its following day (i.e., [0, 1]). Table 3 reports the cumulative abnormal
returns over these three event windows.
As shown in Table 3, the abnormal returns are significantly negative on both days, the
event date (i.e., the day 0) and the day following the event date (i.e., the day +1). For example,
Table 3 shows that on the event announcement date, the average abnormal return of the
implicated firms is CAR[0,0] = -.535% (t = -4.91, p<.01); whereas on the day after the event
date, the average abnormal return is CAR[1,1] = -.357% (t = -3.59, p<.01). Together, the
cumulative abnormal return over these two days is CAR [0,1] = -.891% (t = -7.05, p<.01).
These results are consistent with prior findings on the detrimental impacts of product recalls
on firms’ stock returns (Jarrell and Peltzman 1985; Barber and Darrough 1996; Davidson and
Worrell 1992; Thomsen and McKenzie 2001). Since the abnormal returns are significantly
negative on both days, we chose [0, 1] as the event window for the following analyses.
The impact of pre-recall advertising adjustments. Table 4 presents the results of a
simple univariate analysis, which directly tests whether or not post-recall abnormal returns of
the recalling firms differ across different pre-recall advertising adjustments. To underscore
the significance of the two specific recall characteristics identified in this paper (i.e., product
newness and product hazard), we present the results ignoring these recall characteristics in
Panel A and provide the results considering them in Panel B.
Without considering specific recall characteristics, Panel A offers comparisons for two
cases (increasing pre-recall advertising vs. no adjustment, and decreasing pre-recall
advertising vs. no adjustment). As shown in Panel A, the average abnormal return of recalls
with increasing pre-recall advertising does not significantly differ from that of recalls with no
adjustment (∆CAR =-.0004, p>0.1), and the average abnormal return of recalls with
decreasing pre-recall advertising is significantly lower than that of recalls with no adjustment
Marketing Science Institute Working Paper Series 19
(∆CAR =-.0061, p<0.05). Panel B, incorporating the specific recall characteristics, offers
comparisons for six cases (four for increasing pre-recall advertising vs. no adjustment, and
two for decreasing pre-recall advertising vs. no adjustment). Panel B reveals three significant
results: (1) When the recalled products contain new models with a minor hazard, the average
abnormal return is significantly higher for firms who increased their pre-recall advertising
than for those who did not make a pre-recall advertising adjustment (∆CAR=.0064, p < .1),
which is consist with Hypothesis 1. (2) When the recalled products are older models with a
major hazard, the average abnormal return is significantly lower for firms who increased their
pre-recall advertising than for those who did not make a pre-recall advertising adjustment
(∆CAR=-.0120, p < .05), which is consistent with Hypothesis 2. (3) When the recalled
products contain new models, the average abnormal return is significantly lower for firms
who decreased their pre-recall advertising than for those who did not make a pre-recall
advertising adjustment (∆CAR=-.0083, p < .05), which is consistent with Hypothesis 3.
These results suggest that, when ignoring specific recall factors (Panel A), one may
mistakenly conclude that increasing pre-recall advertising does not affect firm value, but
decreasing it always harms firm value. However, when considering the two recall specific
factors (Panel B), we show that increasing pre-recall advertising can significantly affect firm
value, and we identify specific conditions under which an upward adjustment in pre-recall
advertising weakens (i.e., when new models are recalled due to a minor hazard) or intensifies
(i.e., when older model are recalled due to a major hazard) the negative financial
consequence of a product recall. We also show that decreasing pre-recall advertising does not
always worsen the damage of recalls to firm value, and we show that a downward adjustment
harms firm value only when new model products are affected by recalls.
To further examine the impact of pre-recall advertising on the abnormal returns
resulting from product recalls, we also estimated the cross-sectional model in Eq. (1). We first
conducted several tests concerning potential issues of heterogeneity and multicollinearity.
The Lagrange multiplier (LM) test (Breusch and Pagan 1980) rejects the existence of
unobserved heterogeneity. The largest variance-inflation factor (VIF) of all variables is less
than 5, rejecting the possibility of multicollinearity. Hence, we estimated Eq. (1) using pooled
OLS with heteroskedasticity-consistent standard errors.
We presented two regression results in Table 5. The cross-sectional regression (1) in
Table 5 reports the regression results of a partial cross-sectional model without incorporating
the interaction terms between pre-recall advertising adjustments and the two recall factors,
Marketing Science Institute Working Paper Series 20
while the cross-sectional regression (2) in Table 5 reports the regression results of the full
cross-sectional model in Eq. (1). Consistent with the univariate analysis results in Panel A of
Table 4, without considering the interactions between pre-recall advertising adjustments and
the recall factors, the coefficient of increasing pre-recall advertising is insignificant (binc
= .001, p > .1), whereas the coefficient of decreasing pre-recall advertising is significantly
negative (bdec = -.0074, p < .05). The former result concerning the impact of increasing
pre-recall advertising further suggests the existence of possible contingent effects of the recall
factors.
When incorporating the interactions between pre-recall advertising adjustments and
the recall factors (i.e., the cross-sectional regression (2) using a full cross-section model in Eq.
(1)), we found conditions under which increasing/decreasing pre-recall advertising can lessen
or worsen the harmful impact of product recalls on firms’ stock returns. As shown in the
results of cross-sectional regression (2) in Table 5, the coefficient of inc*new is significantly
positive (binc*new = .0095, p < .05), whereas the coefficients of inc*hazard and dec*new are
significantly negative (binc*hazard = -.0132, p < .05; bdec*new = -.0120, p < .05). Thus, when the
recalls involve new models with a minor hazard (i.e., new = 1 and hazard = 0), the overall
impact of increasing pre-recall advertising, which is the sum of the coefficients of inc and
inc*new, is found to be significantly positive (binc + binc*new = .0109, p < .05). This result,
consistent with the univariate analysis shown in Panel B of Table 4, provides further
empirical evidence to support our Hypothesis 1.
When the recalled products are older models with a major hazard (i.e., hazard = 1 and
new = 0), the overall impact of increasing advertising, which is the sum of coefficients of inc
and inc*hazard, is found to be significantly negative (binc + binc*hazard = -.0118, p < .05).
Consistent with the univariate results in Panel B of Table 4, the overall negative impact of
increasing pre-recall advertising under the product recalls of older models with major hazards
further supports our Hypothesis 2. With regard to a decrease in pre-recall advertising, we test
the sum of the coefficients of dec and dec*new as well as the sum of dec, dec*new, and
dec*hazard and find significantly negative impacts of decreasing pre-recall advertising for
newly introduced models regardless of hazard (bdec + bdec*new = -.0116, p < .05 for new
models with minor hazards; bdec + bdec*new + bdec*hazard = -.0120, p < .05 for new models with
major hazards), which is also consistent with the univariate results in Panel B of Table 4, and
further supports our Hypothesis 3.
Marketing Science Institute Working Paper Series 21
Robustness and validity of results
We conducted several additional analyses to examine the robustness and validity of
our estimation results. First, as stated earlier, we experimented with eight different sets of
benchmark and adjustment periods when classifying the advertising adjustment to be
increasing, decreasing or no adjustment before a recall announcement. Table 6 presents the
cross-sectional regression results when the adjustment and benchmark periods of [1, 3], [1, 4],
[1, 5] and [1, 6] are used, respectively, whereas Table 7 presents the cross-sectional
regression results when the adjustment and benchmark periods of [2, 4], [3, 4], [2, 5] and [3,
5] are used, respectively. For ease of discussion, we refer to these estimation results using
different lengths the adjustment and benchmark periods as Estimation [la, lb]. For example,
Estimation [1, 4] refers to the estimation results using the adjustment period of one week
before the recall announcement date and the benchmark period of four weeks. As shown in
Tables 6 and 7, the regression results concerning the interactions between pre-recall
advertising adjustments and the two recall factors are generally consistent throughout all
models using different adjustment and benchmark periods.
Furthermore, our results using different combinations of benchmark and adjustment
periods also provide important findings with regard to when and how the financial market
responds to firms’ pre-recall advertising adjustments. Among different lengths of benchmark
periods (i.e., 3 to 6 weeks) and adjustment periods (i.e., 1 to 3 weeks), the models using
relatively moderate lengths of benchmark periods (i.e., 4 or 5 weeks) and a short adjustment
period (i.e., 1 week) generate the most significant results and the best model fit (R2 = .43 in
Estimation [1, 4]). For example, as shown in Table 6, when varying the benchmark period
from a shorter window (i.e., 3 weeks) to longer ones (i.e., 5 and 6 weeks) while keeping the
adjustment period at 1 week, Estimation [1, 4] and Estimation [1, 5] provide a better model
fit, R2, in comparison with Estimation [1, 3] and Estimation [1, 6]. Also as shown in Table 7,
when varying the adjustment periods from a shorter period (i.e., 1 week) to longer ones (i.e.,
2 and 3 weeks) while keeping the benchmark period as medium (i.e., 4 or 5 weeks), the
models with the shorter adjustment period (i.e., Estimation [1, 4] and Estimation [1, 5])
provide more significant results and a better model fit. Thus, our results suggest that adjusting
advertising in the time period closer to a product recall (i.e., the last week before a recall) can
create significant market responses, which is consistent with the efficient market hypothesis
(that the capital market responds to the most updated information).
Second, we also estimated the cross-sectional equation (1) by correcting possible
endogeneity. In response to an anticipated product recall, forward-looking firms may adjust
Marketing Science Institute Working Paper Series 22
their marketing strategy (e.g., advertising expenditure) based on recall factors or firm
characteristics in order to minimize the potential damage of a recall (e.g., Shaver 1998). Thus,
the issues of self-selection and endogeneity (Heckman 1979) may exist in the estimation of
the impacts of pre-recall advertising adjustments. Following the literature (e.g., Chen,
Genesan and Liu 2009; Heckman 1979), we used a Heckman two-step approach to test and
correct for potential self-selection bias and endogeneity in advertising adjustments.
Specifically, in the first step of the Heckman approach, the choice probabilities of advertising
adjustments (i.e., increase and decrease) were estimated as a function of observed recall and
firm characteristics, while, in the second step, we estimated the cross-sectional analysis in
Equation (1) with the correction terms incorporated in order to obtain a consistent estimate
(Lee 1983; Wooldridge 2001). The correction term was constructed from the first step based
on the estimated choice probabilities of advertising adjustments. The estimation results after
correcting the potential endogeneity are presented in Table 8, and the technical details of our
Heckman two-step procedure is reported in the Appendix. As shown in Table 8, the
coefficient of the correction term is not significant (p > .4), suggesting that the sample
selection and endogeneity are not severe issues in the estimation of the cross-sectional model
in Eq. (1). More important, the results concerning the impacts of pre-recall advertising
adjustments under different recall conditions still hold when endogeneity is corrected.
Third, to further validate our empirical results, we constructed a control sample in
which the abnormal returns of recalling firms in a recall-free window are estimated to
examine whether or not the moderating effects of pre-recall advertising presented earlier are
strictly due to product recalls. Specifically, we used the same sample of firms/products in our
data, but randomly selected a two-day, event-free window for each firm (i.e., no auto recalls
and no news reported by the WSJ). We constructed the same measures of advertising
adjustments in such a two-day, event-free window and conducted the cross-sectional
regression in Eq. (1) to test whether or not the advertising adjustment demonstrates similar
moderating impacts in the event-free scenario. The results in Table 9 indicate no significant
impact of advertising adjustments on the firms’ abnormal returns when there is no major
recall, which verifies that the moderating effects of pre-recall advertising identified in this
study are specific to the recall events.
Marketing Science Institute Working Paper Series 23
Conclusion
Despite the rising number of product recalls and the catastrophic consequences of
severe product-harm crises in recent years, our knowledge of product-harm crisis
management is still limited in both theory and practice (Smith, Thomas and Quelch 1996).
This paper develops a theoretical framework to investigate whether and how adjusting
pre-recall advertising can affect a firm’s stock market value under a product recall. Our
theoretical framework and empirical findings contribute to the marketing-finance literature as
well as to the crisis management literature and provide important guidance for marketers to
effectively implement advertising strategies in managing a product-harm crisis.
Theoretical contributions
This article makes several theoretical contributions. First, as a fast-growing number of
firms continue to globalize their business and outsource some parts of their operations to
global partners, the risk of product-harm crisis is increasing. While crisis management has
become essential knowledge to marketers in the global market place, it has drawn limited
attention from marketing scholars. With the exception of a recent study investigating how a
firm’s product-recall strategy (proactive vs. passive) affects its stock values (Chen, Genesan
and Liu 2009), most extant marketing research focuses on examining the harmful impact of a
product recall on consumers and on firms’ marketing effectiveness. This paper contributes to
the marketing literature in crisis management by demonstrating the importance of marketing
strategies in pre-crisis management. Specifically, this paper theoretically proposes and
empirically investigates conditions under which adjusting advertising ex ante can lessen or
worsen the detrimental impacts of product recall on the recalling firms’ financial performance.
More important, we bring attention to an unexplored strategic move in crisis management,
increasing pre-recall advertising spending, and show that the firm can protect its financial
value by strategically boosting pre-recall advertising spending when anticipating a recall of
new products with a minor hazard. These findings provide insights into how to implement
marketing strategies ex ante in managing a recall crisis ex post.
Second, our findings also reveal some potential conflicting impacts of marketing
strategies in crisis management when considering different strategic objectives. Specifically,
a striking insight of our research is that, in product-harm crisis management, profit
maximization and shareholder value maximization can be in direct conflict with each other.
For example, while a recent marketing study suggests that, when envisioning a potential
Marketing Science Institute Working Paper Series 24
recall due to a major hazard, a firm can maximize its marketing performance (i.e., profit) by
reducing pre-crisis advertising (Rubel, Naik, and Srinivasan 2011), our results show that
simply cutting advertising near recalls can exacerbate the harmful impacts of product recalls
on firms’ financial valuation if the recalled products involve new models. Moreover, this
paper identifies scenarios under which firms can achieve one objective with or without
sacrificing another. We show that, if the recalled products are older models, firms can protect
their marketing objective by reducing pre-recall advertising without sacrificing their financial
objective. These findings advance our understanding of effective crisis management and
highlight the importance of investigating and integrating the impacts of marketing strategies
on different strategic goals in crisis management.
Third, to the best of our knowledge this is the first paper to investigate how marketing
affects firms’ financial market valuations in the crisis environment. Although a growing
number of studies in the marketing-finance interface have examined how marketing
initiatives affect firms’ financial valuations in the regular market environment, no study to
date has investigated the marketing-finance relationship in a crisis environment. With regard
to the relationship between advertising and its impact on financial performance, prior studies
have shown that, in the regular market environment, advertising generally plays a positive
role in moderating the impact of marketing actions (e.g., product placement, new product
introduction), because advertising spending raises awareness of firms’ actions and creates
positive signals regarding marketing’s contributions to future earnings, both of which help
build brand equity and enhance firms’ financial values (Wiles and Danielova 2009;
Srinivasan et al. 2009). In a crisis environment, however, our study shows that such a positive
moderating impact of advertising may not exist, because a high level of advertising can also
create a negative expectation effect when confronted with product recall crises and, as a result,
the overall impact of advertising is contingent upon recall characteristics. Such a difference in
the impact of advertising underscores the importance of extending prior studies of the
marketing-finance interface to the crisis environment.
Managerial implications
When a firm envisions a product recall, what should managers do to minimize its
potential damage? Specifically, should forward-looking managers increase or decrease
pre-crisis advertising? As discussed earlier, the answers to these questions have thus far been
ambiguous: While Cleeren, Dekimpe, and Helsen (2008) argued that an increase in pre-crisis
Marketing Science Institute Working Paper Series 25
advertising may create a buffer against the negative publicity of the crisis, Rubel, Naik, and
Srinivasan (2011) suggested the recalling firms decrease pre-crisis advertising to maximize
post-crisis profits. Our empirical findings provide some managerial guidance on how firms
should adjust pre-recall advertising strategies when considering both profit and financial
value maximization.
When to increase pre-crisis advertising? Our findings suggest that increasing
advertising ex ante can lessen the negative impact of product recalls on stock returns when
the recall involves new models with a minor hazard, because doing so can send a strong
positive signal to the stock market concerning the recalling firms’ confidence in managing the
crisis effectively. Hence, if protecting stock market returns is the most important objective (as
opposed, for example, to the marketing objective of profit maximization), the firm can signal
its confidence to the stock market by increasing its advertising spending in order to minimize
the potential harmful impact on its stock market returns when the recalled products involve
newly introduced models with a minor hazard.
When to decrease pre-crisis advertising? As stated earlier, the literature has suggested
that firms can maximize their marketing performance (i.e., profits) by reducing pre-crisis
advertising when they expect that a forthcoming recall would lead to reduced marketing
effectiveness (Rubel, Naik, and Srinivasan 2011). Our findings suggest that doing so will not
further hurt financial performance (i.e., stock market returns) only if the recalled products are
older models. Thus, when a firm envisions a potential recall of older models, reducing
pre-crisis advertising can be a strategic choice to achieve better marketing performance
without further damaging its financial market performance.
When not to adjust pre-crisis advertising? Our results also provide insights into the
conditions under which it is better for firms not to adjust advertising near a product recall.
Specifically, our results show that when firms have to recall newly introduced models with
major hazards, reducing pre-recall advertising can further deepen the loss in stock returns
because of the negative signaling effect. In this case, while increasing pre-recall advertising
would not harm stock returns further, as suggested by our results, the marketing performance
of advertising spending would be weakened, according to the literature (Rubel, Naik, and
Srinivasan 2011). In such a case, therefore, it is better for firms not to adjust their advertising
expenditure when considering both marketing and financial performance.
If adjusting pre-recall advertising, the timing? The timing of a pre-recall advertising
adjustment is also of strategic importance to managers as adjusting it too early may not
generate any significant impact on the stock market. According to our results in Tables 6 and
Marketing Science Institute Working Paper Series 26
7, we show that the stock market responds most actively to advertising adjustments one week
prior to the announcement of a recall. Hence, to maximize the strategic impact of adjusting
advertising spending in crisis management, managers should not make the adjustment too
early, because the stock market may not capture such an adjustment as recall-relevant
information.
When is integrated crisis management needed? Managers should also be aware of
when they need to develop an integrated product-harm crisis management strategy.
Specifically, when the recalled products are new models, crisis management needs to
integrate the impacts of its strategic move on both marketing performance and financial
performance. This integration is necessary because, while it is the best strategic choice to
increase pre-recall advertising from the perspective of stock returns when a recall involves
new models and is due to a minor hazard, it is not the best strategy from the perspective of
marketing performance, because the effectiveness of marketing spending will be weakened
due to product-harm crisis. On the contrary, while it is optimal for firms to decrease pre-recall
advertising (because of weakened marketing effectiveness), this strategy is not optimal when
considering stock market performance, because doing so can further intensify the financial
loss in the stock market. Therefore, when the recalled products involve new models with a
minor hazard, it is crucial for managers to investigate the impacts of their marketing
movement on both consumer market performance (i.e., profits) and financial market
performance and develop an integrated crisis management strategy.
Future research
This study provides several directions for future research. First, this study uses
spending as the metric of advertising. Future research can consider other potential dimensions
of advertising such as advertising content (Resnik and Stern 1977; Smith 1991; Xu et al. 2014)
and advertising creativity (Smith, Thomas and Quelch 2007; Yang and Smith 2009). The
former helps convey specific information about quality or safety issues, while the latter
makes the communication during a recall crisis more effective. Investors (or consumers) may
react to the adjustment of these advertising metrics differently in a crisis environment than in
a regular market environment.
Second, this paper investigates how advertising affects abnormal returns as a result of
a recall crisis. It would also be interesting to study how a recall crisis affects the effectiveness
of advertising on firms’ stock market valuation, and compare the impact of advertising on
Marketing Science Institute Working Paper Series 27
firm value before and after a recall. While Van Heerde, Helsen, and Dekimpe (2007)
compared effectiveness of advertising in the consumer market before and after a recall, how
its effectiveness in the financial market differs before and after a recall remains unexplored.
Third, future studies might also investigate how competitors’ marketing strategies
affect a focal firm’s financial market returns under a recall crisis. Empirical evidence has
shown that a product harm crisis inspires competitive responses, which may intensify the
recall’s damage to the focal firm. For instance, competitors may increase their advertising
spending or run an incentive program targeting the focal firm’s consumers. In the Toyota
recall crisis, General Motors offered a $1,000 rebate to attract Toyota car owners.12 During
the 1996 recall of Kraft peanut butter, the competitor’s brand Sanitarium launched an
advertising campaign to emphasize that its peanut butter was not contaminated (Van Heerde,
Helsen, and Dekimpe 2007). Van Heerde, Helsen, and Dekimpe (2007) found that a serious
recall can cause both “an increased cross sensitivity to rival firms’ marketing-mix activities”
and “a decreased cross impact of its marketing-mix instruments on the sales of competing,
unaffected brands.” It would be interesting to examine whether or not such effects also exist
in the financial market.
Finally, future research can extend our study to other marketing mixes with respect to
their strategic impacts on financial markets under recall crises. For example, how does the
stock market interpret a sales promotion near a recall? In particular, when a large sales drop is
inevitable after the recall, would a sales promotion before the recall improve or hurt a firm’s
financial value? As suggested earlier, it is important for both practitioners and researchers to
understand how the stock market interprets the adjustments of firms’ marketing strategies
near a recall. Such studies can further contribute to the marketing literature in crisis
Smith, Robert E., Scott B. MacKenzie, Xiaojing Yang, Laura M. Buchholz, William K. Darley (2007),
“Modeling the Determinants and Effects of Creativity in Advertising,” Marketing Science, 26 (6),
819–833
Smith, Ruth Ann (1991), “The Effects of Visual and Verbal Advertising Information on Consumers'
Inferences,” Journal of Advertising, 20 (4), 13–24
Smith, N. C., Thomas, R. J., & Quelch, J. A. (1997), “A strategic approach to managing product recalls,”
Journal of Product Innovation Management, 14 (3), 228-229
Sood, A., & Tellis, G. J. (2009), “Do innovations really pay off? Total stock market returns to innovation,”
Marketing Science, 28 (3), 442-456
Solomon, D. H. (2012), “Selective publicity and stock prices,” Journal of Finance, 67 (2), 599-638
Souiden, N., and Pons, F. (2009), “Product recall crisis management: the impact on manufacturer’s image,
consumer loyalty and purchase intention,” Journal of Product & Brand Management, 18 (2), 106-114
Srinivasan, S., Pauwels, K., Silva-Risso, J., and Hanssens, D. M. (2009), “Product innovations, advertising,
and stock returns,” Journal of Marketing, 73(1), 24-43
-------, and D.M. Hanssens (2009), “Marketing and Firm Value: Metrics, Methods, Findings, and Future
Directions,” Journal of Marketing Research, 46 (3), 293–312
Tellis, G. J., & Johnson, J. (2007), “The value of quality,” Marketing Science, 26 (6), 758-773
Thomsen, M. R. and A. M. McKenzie (2001), “Market Incentives for Safe Foods: An Examination of
Shareholder Losses from Meat and Poultry Recalls,” American Journal of Agricultural Economics, 83
(3):526–538
Van Heerde, Harald, Kristiaan Helsen, and Marnik Dekimpe (2007), “The Impact of a Product-Harm Crisis
on Marketing Effectiveness,” Marketing Science, 26 (2), 230–45
Wallace, D. W., Giese, J. L., & Johnson, J. L. (2004),” Customer retailer loyalty in the context of multiple
channel strategies,” Journal of Retailing, 80 (4), 249-263
Marketing Science Institute Working Paper Series 34
Wernerfelt, Birger (1988), “Umbrella Branding as a Signal of New Product Quality: An Example of
Signaling by Posting a Bond,” The RAND Journal of Economics, 19 (3), 458–466
Wiles, Michael A. and Anna Danielova (2009), “The Worth of Product Placement in Successful Films: An
Event Study Analysis,” Journal of Marketing, 73 (4), 44–63
Wooldridge, J. M. (2001), “Econometric Analysis of Cross Sections and Panel Data,” Cambridge, MA:
MIT Press
Xu, Linli, Kenneth C. Wilbur, S. Siddarth and Jorge Silva-Risso (2014), “Price Advertising by
Manufacturers and Dealers,” Management Science, in press
Yang, Xiaojing and Robert E. Smith (2009), “Beyond Attention Effects: Modeling the Persuasive and
Emotional Effects of Advertising Creativity,” Marketing Science, 28 (5), 935–949
Zhao, Yi, Ying Zhao, and Kristiaan Helsen (2011), “Consumer Learning in a Turbulent Market
Environment: Modeling Consumer Choice Dynamics after a Product-Harm Crisis,” Journal of
Marketing Research, 48 (2), 255–2
Marketing Science Institute Working Paper Series 35
Table 1: The Distribution of Pre-recall Advertising Adjustments
Advertising Adjustment
Increase Decrease No Adjustment Total
Overall 30 30 50 110 New Model Recall 20 20 22 62 Old Model Recall 10 10 28 48
Major Hazard 16 17 23 56 Minor Hazard 14 13 27 54
Note: The three categories of advertising adjustments are classified based on the adjustment period of 1 week before the recall date and the benchmark period of 4 weeks.
Marketing Science Institute Working Paper Series 36
Table 2: Variable Definitions and Data Statistics Variable Definition/Operationalization Source MEAN SD
Advertising adjustments
inc Whether pre-recall advertising for recalled products increases (1) or not (0) TNS Media Intelligence
0.273 0.447
dec Whether pre-recall advertising for recalled products decreases (1) or not (0) 0.273 0.447
Recall characteristics
new Whether the recall involves new models (1) or not (0) NHTSA 0.563 0.498
rhazard Whether the recall is due to major safety hazard (1) or not (0) NHTSA 0.509 0.502
NHTSA Whether the recall is initiated by the NHTSA (1) or not (0) NHTSA 0.382 0.488
rcsize The logarithm of the total number of vehicles affected by the recall NHTSA 5.305 0.530
airbag Whether the recall is due to the airbag problem (1) or not (0) NHTSA 0.064 0.245
y2009 Whether the recall is after Toyota’s recall crisis since 2009 (1) or not (0) NHTSA 0.427 0.497
publicity The level of publicity of a product recall with four possible categories:
fsize Firm size as measured by the logarithm of the firm’s sales revenue COMPUSTAT 5.173 0.148
fdeb Firm debt as measured by the logarithm of the firm’s long-term liability COMPUSTAT 4.634 0.321
frep The level of firm reputation, a overall score surveyed by the Fortune magazine Fortune magazine 5.606 0.993
inc_u Whether pre-recall advertising for unaffected products increases (1) or not (0) TNS Media Intelligence
0.273 0.447
dec_u Whether pre-recall advertising for unaffected products decreases (1) or not (0) 0.309 0.464
Marketing Science Institute Working Paper Series 37
Table 3: Abnormal Returns of Auto Recalls over Different Event Windows
Event Window Abnormal Return Standard Error T-statistics P-Value
[0,0] -.0054 .0011 -4.91 <0.01
[1,1] -.0036 .0010 -3.59 <0.01
[0,1] -.0089 .0013 -7.05 <0.01
Marketing Science Institute Working Paper Series 38
Table 4: Abnormal Returns under Different Pre-Recall Advertising Adjustments and Different Product Recalls
Panel A: Pre-Recall Advertising Adjustments and Abnormal Returns Accruing to Product Recallsa
“Increasing Pre-Recall Advertising” vs. “No Adjustment” “Decreasing Pre-Recall Advertising” vs. “No Adjustment” All -.0004b -.0061**
Panel B: Pre-recall Advertising Adjustments and Abnormal Returns Accruing to Different Product Recalls
“Increasing Pre-Recall Advertising” vs. “No Adjustment” “Decreasing Pre-Recall Advertising” vs. “No Adjustment” Major Hazard Minor Hazard
New .0011 .0064* -.0083** Old -.0120** .0027 .0009
Note: ** p< .05; * p< .10 a: The advertising adjustments are classified based on the adjustment period of 1 week and the benchmark period of 4 weeks. b: The abnormal return reported here is ∆CARinc-no,[0,1], that is, it is the abnormal return accruing to the product recalls of those firms who increased their pre-recall
advertising subtracting the abnormal return of those firms who made no pre-recall advertising adjustment. Similarly, all the abnormal returns reported in this table are relative to those under no pre-recall advertising adjustment.
Marketing Science Institute Working Paper Series 39
Table 5: Estimation Results of Cross-Sectional Regressions
R square 0.33 0.43 0.39 0.35 Note: ** p<.05; * p<.10. a: Estimation [1, 3] refers to the estimation results using the adjustment period of 1 week before the recall date and the benchmark period of 3 weeks.
Marketing Science Institute Working Paper Series 41
Table 7: Estimation Results of different Adjustment Periods