Sudden Unintended Used-Price Deceleration? The 2009-2010 Toyota Recalls ∗ Robert G. Hammond

Sudden Unintended Used-Price Deceleration?
The 2009-2010 Toyota Recalls∗
Robert G. Hammond†
January 27, 2012
∗
I thank a co-editor and two referees for suggestions; Gary Fournier, Charles Knoeber, and Stephen Margolis
for discussions; the editorial staff at the Black Book (especially managing editor Ricky Beggs) for insights into the
secondary market for automobiles; seminar participants at East Carolina University and the Southern Economic
Association 2010 Annual Meetings for comments; the Edwin Gill Research Grant for financial support; and Jong Jin
Kim for excellent research assistance.
†
Department of Economics, North Carolina State University. Contact: robert [email protected]
Sudden Unintended Used-Price Deceleration?
The 2009-2010 Toyota Recalls
Abstract
Using data from the vehicle resale market, I test consumer responsiveness to large-scale
product recalls that are caused by safety problems. The used-vehicle prices of Toyotas are
compared to the used-vehicle prices of the other major domestic and foreign manufacturers. The
results quantify the losses suffered by Toyota vehicle owners in secondary markets due to the
2009-2010 safety recalls of more than nine million Toyota Motors vehicles. The treatment effect
of a recall is measured using panel data with a difference-in-differences estimation approach that
allows for time-varying treatment effects and serial correlation. I find that this recall episode had
negative effects in the resale market for automobiles that were quantitatively small (less than
2% of the vehicle’s resale value), statistically indistinguishable from zero, and short lived (did
not persist beyond December 2009). A comparison with Audi’s recalls in the 1980s of vehicles
with sudden unintended acceleration suggests that the extent to which a company’s reputation
is established is more important than whether or not a company has a reputation for producing
high-quality products.
JEL classification: L15, L62, D22
Keywords: Automobiles, recalls, product quality, reputation, product safety
1
1
Introduction
Product recalls related to safety problems receive enormous amounts of attention in the popular
press. In this paper, I study the automobile industry, where 22.4 million automobile owners received
recall notices during a twelve month period ending in the summer of 2010. As a result, more new
and used vehicles were recalled in 2010 than new vehicles were sold in 2010 (Bunkley and Vlasic,
2010). The most widely covered recall during this period was the recall of Toyota Motors vehicles
due to safety issues broadly dubbed “sudden unintended acceleration.” These safety problems
initially involved floor mats that could entrap the vehicle’s accelerator pedal but were expanded to
include mechanical issues with the accelerator pedal itself. Both the Lexus and Toyota brands of
Toyota Motors were included in the recall of more than nine million vehicles worldwide.
Toyota’s outlook following its recalls was increasingly bleak, from the negative media coverage to
the negative stock market reaction (−15% from January 25, 2010 to January 29). Reports placed
the costs of the recalls at $2 billion, a figure that excludes the damage to Toyota’s reputation
as a quality leader. In contrast to the initial burst of negative news, later reports documented
increasing new-vehicle sales for the company and profits of $2.2 billion for the second quarter of
2010 (Hennigan, 2010). I reconcile the expected and actual performance of Toyota Motors during
the recall period by quantifying the reaction of the company’s consumers to its safety problems.
More fundamentally, I analyze used-vehicle price data to address the extent to which consumer
demand responds to safety recalls.
It is most informative to address this question using data from secondary markets because these
resale prices most accurately quantify the effect of recalls in the automobile industry. Data from
primary markets (i.e., Toyota dealers selling new Toyotas) intermingle shocks to consumer demand
with strategic adaptation by the manufacturer. In the particular case of interest here, Toyota
responded to the negative attention of its recalls by offering “an average of $2,256 in incentives,
according to the research firm Edmunds.com, more than Toyota had ever had to give customers
to buy its cars and trucks” (Ahrens, 2010). Further complicating data from primary markets is
the potential for strategic adaptation by the competitors of the affected manufacturer. Here, GM
offered $1,000 incentives to consumers who traded in a Toyota vehicle to purchase a GM vehicle;
Chrysler, Ford, and Hyundai soon began offering similar incentives to poach Toyota consumers.
2
While data from the secondary market for automobiles provide the cleanest test of changes in
consumer demand, it is worth noting that the evidence that I present provides only an indirect test
of the effect that we should expect in the primary market for Toyota vehicles.
The notion of consumer demand that is of interest in this paper is not simply the demand
curve in the secondary vehicle market. Instead, I am interested in consumer perception of Toyota
vehicles. In secondary markets, changes in consumer perception affect both supply and demand.
The methodology proposed here can isolate changes in consumer perception because shocks to
consumer perception have an unambiguous effect on prices in secondary markets. For example, a
negative shock causes the supply of used Toyotas to increase because current owners sell and the
demand for used Toyotas to decrease because consumers perceive Toyota less favorably. Because
both effects push prices down, a negative shock to consumer perception can be measured using
secondary market prices.
There is a large literature on product safety in industrial organization (e.g., Daughety and
Reinganum (1995)). Specific work on recalls have explored the role of information (Reilly and
Hoffer, 1983), reactions in the stock market (Jarrell and Peltzman, 1985), and highway safety (Bae
and Benıtez-Silva, 2011). Cases studies have analyzed Tylenol (Dowdell, Govindaraj, and Jain,
1992), the food industry (Salin and Hooker, 2001), Firestone tires (Copeland and Hall, 2011),
and Chinese toys (Freedman, Kearney, and Lederman, 2010). Most relevantly, the automobile
industry has received a great deal of attention in the recall literature. Barber and Darrough (1996)
and Nichols and Fournier (1999) compare the evolving reputations of American and Japanese
manufacturers, while Rupp and Taylor (2002) and Rupp (2004) study the causes and effects of the
most costly automobile recalls during the period through 1998.
My contribution to this literature is a careful measure of the extent to which consumer demand
responds to information about the recalls of automobiles. I find that the 2009-2010 Toyota recalls
had small effects on the prices at which Toyota vehicles resell. This result holds across numerous
specifications and throughout the entire period following the initial reporting on Toyota’s safety
problems. To provide context for these small effects, I repeat the analysis on a similar recall episode
(the 1980s safety recalls of the Audi 5000) and find that Audi vehicles suffered larger losses than
those of Toyota vehicles in 2009-2010 despite the similarities between the two recall episodes.
I reconcile the small treatment effect of the Toyota recalls with the large treatment effect of the
3
Audi recalls by documenting a key difference between the two companies. While both companies
had a reputation for producing high-quality vehicles at the time of their recalls, Toyota had a betterestablished reputation, in the sense that Toyota had been in the market longer and had a larger
market share at the time of its sudden unintended acceleration problems. These results suggest that
the response of consumer demand to negative product-quality information depends on the extent
to which a company’s reputation is established more so than it depends on a company’s reputation
for producing high or low quality products. Before turning to an overview of the Toyota recalls,
note that this recall episode is unique in this literature in that these recalls allow measurement
of consumer perception for goods that were actually recalled. In contrast, many previous studies
considered events where the affected goods were completely removed from the market and were no
longer available for purchase.
2
The 2009-2010 Toyota Recalls
Before introducing the data and empirical methodology to be used, I briefly discuss the events
surrounding the recall of nine million Toyota vehicles worldwide. Further details can be found in
any number of popular press treatments, including Mitchell (2009), Vartabedian and Bensinger
(2009), and MacKenzie and Evans (2010). See Table 1 for a timeline of the recalls as well as Tables
2 and 3 for lists of recalled and unrecalled vehicles respectively by the recalls during the period of
interest.
On August 28, 2009, off-duty California Highway Patrol officer Mark Saylor and three passengers
are killed in California while driving a 2009 Lexus ES350. Multiple media outlets report that
the vehicle “suddenly” accelerated to speeds exceeding 100 miles per hour and emergency call
records indicate that a passenger said that the car’s brakes were no longer functioning. The crash
investigation found that the wrong floor mats were installed in the vehicle, which interfered with
the accelerator pedal. Toyota responded by initiating a recall and advising owners of its vehicles to
remove their floor mats immediately. The floor-mat recall of 3.8 million Toyotas was first reported
on September 29 and was officially announced by the company on November 2.
On October 18, the Los Angeles Times published an article addressing claims of sudden unintended acceleration, the first in a series of widely covered articles. A follow-up article in this series
4
reported that Toyota had received more than 1,200 reports of sudden unintended acceleration over
the previous eight years. Later on October 30, Toyota mailed notices to owners of an impending
recall to address issues related to sudden unintended acceleration. Details of the recall were announced on November 26, when the original floor-mat recall was widened to 4.2 million vehicles,
covering these vehicles’ brake override systems to address sudden unintended acceleration.
On December 26, another widely reported accident leaves four Texans dead in a crash involving
the sudden unintended acceleration of a Toyota Avalon whose floor mats had been removed per
Toyota’s instructions. The third Toyota recall of the period followed on January 21, 2010 involving
faulty accelerator pedals on 2.3 million vehicles (of which 2.1 million vehicles were involved in the
previous recall). Six days after this third recall, Toyota expanded its floor-mat recall to include 1.1
million additional vehicles. Finally, on January 26, Toyota announced that it would suspend sales
of all vehicle models affected by the January 21 pedal recall. The company also announced that it
would suspend production on the affected assembly lines for one week starting February 1.
To illustrate the timing of news coverage addressing the Toyota recalls, Figure 1 shows the
number of news articles indexed by Google News matching the word “Toyota” and either of the
words “mat” or “pedal.” The peak of the coverage comes in February 2010, when (according to
Google News) “about 1,440” related news stories covered Toyota. For reference, consider two pop
musicians who experienced significant news coverage during February 2010. “Miley Cyrus” was
mentioned in 1,870 stories, while “Lady Gaga” was mentioned in 6,950 stories.
3
Panel Data on Used-Vehicle Prices
Data on the prices at which used vehicles sell in secondary markets were purchased from the
National Auto Research Division of the Hearst Business Media Corporation (www.blackbookusa.
com). These data are packaged as the Black Book and are used by automobile dealers when
purchasing vehicles at wholesale auctions to a much larger extent than more widely known pricing
guides (e.g., the Kelley Blue Book (www.kbb.com) or the National Automobile Dealers Association
Used Car Guides (www.nadaguides.com)). In fact, Black Book restricts purchases of its data
products to “industry-qualified businesses.” While Black Book data contain information on an
individual vehicle’s wholesale, retail, and trade-in values, I use retail values to measure the effect of
5
recalls on consumer demand. Black Book wholesale prices are derived from “hands-on Black Book
surveyor data” from wholesale automobile auction houses across the country, while their retail
prices are derived from “data from manufactures, captive lending sources, remarketing companies,
banks, leasing companies, dealers, and auctions” (Black Book, 2011).1
I define a vehicle to be a particular make-model-model year-trim level, which implies that there
are five vehicles in the sample that correspond to the five trim levels of the 2010 Toyota Camry:
Base, LE, SE, XLE, and Hybrid. For a particular vehicle, I record the price at which it sold on
average in used-vehicle markets for each month from April 2008 until April 2010. Table 4 provides
the distribution of used-vehicle prices for three periods in the sample. The sample size increases
over time, primarily because vehicles enter the used-vehicle market in sufficient numbers for pricing
data to become available. But the lower number of vehicles in the initial month of the sample is
also due to the fact that only cars are included in the sample from April 2008 through December
2008, while the full slate of cars, trucks, and sports-utility vehicles (SUVs) are included in the
sample from January 2009 through April 2010.
Vehicles in the 2007, 2008, 2009, and 2010 model years are included in the sample. These model
years encompass the vast majority of the models affected by the Toyota recalls and comprise the
majority of the used-vehicle market during the same period. The automobile manufacturers to be
studied were chosen to include the largest manufacturers in the market, which were the following
American manufacturers: Chrysler, Ford, and General Motors (GM); and foreign manufacturers:
Honda, Hyundai, Nissan, and Toyota.2 The sample of vehicles was selected by including a representative vehicle from each manufacturer (i.e., the most common vehicle) in the following segments
of the market: subcompact, compact, midsize, fullsize, sports, crossover SUV, SUV, and truck.
The final sample contains 7,937 observations, including 492 vehicles in total. Because the Toyota
recalls are the focus of the paper, a larger set of vehicles were sampled from these four model years
for the two brands of Toyota Motors: Lexus and Toyota.
To control for the importance of fuel efficiency in resale markets, data are included on gasoline
1
In total, the editorial staff at Black Book analyzes more than 200,000 separate records each month. These figures
and the sources for each price series were provided to me by the editorial staff at Black Book, in particular managing
editor Ricky Beggs.
2
The following brands were included in the sample: Chrysler: Chrysler and Dodge; Ford: Ford, Lincoln, and
Mercury; GM: Buick, Cadillac, Chevrolet, GMC, Hummer, Pontiac, and Saturn; Honda: Acura and Honda; Hyundai:
Hyundai; Nissan: Nissan; and finally, Toyota: Lexus and Toyota.
6
prices and the fuel economy of each vehicle under study (Li, Timmins, and von Haefen, 2009). In
particular, I construct a measure of each vehicle’s cost of driving (MP$) by dividing its EPA miles
per gallon rating by the dollar price of unleaded gasoline. Monthly gasoline prices (including taxes)
are reported by the U.S. Energy Information Administration (www.eia.doe.gov). Before moving
to the econometric model, I note that each vehicle’s used price and its costs of driving (MP$) are
reported in constant 2000 dollars using the monthly Consumer Price Index.
4
Estimating the Effect of Recalls
The econometric approach follows Hammond (2010), who adapts the methodology of Laporte
and Windmeijer (2005). See those papers for full details but the basic approach is relatively simple
and requires only data on each vehicle’s used price and its costs of driving. The approach does not
require or utilize data on other observable covariates that describe the vehicle’s engine size (e.g.,
horsepower and number of cylinders), vehicle size (e.g., interior room and number of doors), or
standard features (e.g., power windows, doors, and locks). Instead, I use a fixed-effects estimation
to control for all time-constant vehicle-specific covariates, which is particularly useful because it
controls for time-constant vehicle-specific unobservables that affect consumer demand but are not
available to the econometrician.
I measure treatment effects according to the following specification:
Log(P riceit ) = β0 + Tt β + Tt wi θ + ci + it ,
(1)
where i denotes a particular vehicle and t denotes a particular month. T is a 25 by 25 identity
matrix of time dummies for each of the 25 months. w is a dummy variable that equals 1 if vehicle i is
“treated” and equals 0 otherwise. c is a vehicle fixed effect, controlling for unobserved characteristics
that affect a vehicle’s price. β is the parameter vector that measures how prices change over time,
while θ is the parameter vector of interest that measures how the prices of treated vehicles changes
relative to control vehicles. By analyzing the log of prices, I measure the recall treatment effect
vector θ as (approximately) a percentage change in decimal form.3
3
More precisely, the decimal form of the treatment effect of period t is equal to exp(θt ) − 1, which is approximately
equal to θt in decimal form (Halvorsen and Palmquist, 1980). Just as the approximation with a continuous covariate
works well when the incremental change is small, the approximation with a dummy covariate works well when θ is
7
Treatment will be defined according to a particular objective. To measure the effect of recalls,
treatment will take two forms: first, let w equal 1 for recalled Toyota vehicles, equal 0 for unrecalled
Toyota vehicles, and equal 0 for non-Toyota vehicles. An alternative approach is to let w equal 1 for
all Toyota vehicles and equal 0 for non-Toyota vehicles. Finding negative effects in the first approach
but not the second approach, for example, indicates that recalls matter but do not spread to the
remaining vehicles in the manufacturer’s fleet that are not recalled. The estimation uses the first
difference of Log(P riceit ) to eliminate serial correlation and addresses panel-level heteroskedasticity
with robust standard errors that are clustered by vehicle.
5
Evidence that the Toyota Recalls Did Not Have Large Effects
Figure 2 displays the average prices for cars manufactured by each firm in the sample for the
2007-2010 model years separately. Only cars are shown in these four panels because only cars were
included in the entire run of data from April 2008 through April 2010. The patterns for trucks
and SUVs for January 2009 through April 2010 are similar for the comparisons of interest and are
available upon request. Prices are volatile early in the life of a particular model year because price
data are only available once sufficient transaction data are available from secondary markets. This
volatility is most clearly seen for 2008 model year vehicles prior to October 2008 and for 2010 model
year vehicles prior to January 2010.
More importantly, prices are relatively stable over time, will the 2007 model year having the
most pronounced downward trend. This stands in contrast to the typical view of resale prices for
durable goods falling over time as a vehicle’s value diminishes with its remaining life. This fact
points to an evolving secondary market for automobiles that is increasingly competitive with the
primary market. With this price stability in mind, the general pattern across model years over time
points toward the following ranking of prices for each manufacturer in the sample, from highest to
lowest: Toyota, Honda, Nissan, GM, Ford, Chrysler, and finally, Hyundai. Recalled Toyota cars
are less expensive than unrecalled Toyota cars in the 2007 and 2008 model years; all 2009 and 2010
model year Toyota cars in the sample were recalled. The final point of interest from Figure 2 is
whether there is any break in the relative prices of Toyota cars around the recall announcements.
close to zero. In particular, when θ ∈ [−0.05, 0.05], the difference between exp(θt )−1 and θt is less than one-hundredth
of one percent.
8
None is apparent, either from comparing unrecalled and recalled Toyota cars or from comparing
Toyota cars to non-Toyota cars. While this suggests that the Toyota recalls did not have large
effects, a more careful analysis of these data is needed.
As a first pass at implementing the methodology of Section 4, I assign recalled vehicles to the
treatment group and unrecalled vehicles to the control group (including non-recalled Toyotas and
non-Toyota vehicles). The estimation exploits within-vehicle variation across time and all vehicles
(cars, trucks, and SUVs) are included. Results are in Figure 3. The top panel displays the time
trend of treatment effects for both groups along with their 95% confidence intervals. The bottom
panel differences these treatment effects (recalled minus unrecalled) month by month and displays
the 95% confidence interval of the difference. The timeline of the Toyota recalls from Section 2
indicates that we should look for the effect of the recalls starting in September 2009. The top
panel shows that, after controlling for observed and unobserved vehicle heterogeneity, prices fall
for all vehicles in November and December 2009 but to a much smaller extent than they fell in
November and December 2008. This indicates a clear seasonal trend in used-vehicle prices. The
fact the recalled vehicles did not experience outcomes that differed in any way from the outcomes
of unrecalled vehicles provides evidence that recalls do not have a large effect on consumer demand.
An improved comparison to measure the effect of the Toyota recalls is in Figure 4. The top
panel shows the treatment effects for each manufacturer, separating unrecalled and recalled Toyota
vehicles. The bottom panel differences the Toyota treatment effects (recalled minus unrecalled),
including the 95% confidence interval of the difference. Throughout 2009, recalled Toyotas hold
their value better than unrecalled Toyotas but the relevant question is whether there exist any
discrete changes in the relative treatment effects during the recall period. Recalled Toyotas lose
around 1% of their price advantage over unrecalled Toyotas in December 2009. The top panel
shows that this occurs because recalled Toyotas experienced a larger price drop from November to
December than unrecalled Toyotas. Further, it appears that recalled Toyotas experienced a larger
price drop between these two months than any other manufacturer.
In a similar exercise, the top panel of Figure 5 shows each manufacturer’s treatment effects,
combining all Toyotas. The bottom panel differences the Toyota treatment effects using Honda
as a control group. I follow Hortaçsu, Matvos, Syverson, and Venkataraman (2010) in choosing
Honda; this is for no reason other than the fact that Toyota and Honda are widely considered
9
to be each other’s closest competitors. Here I find that Toyota vehicles begin losing their value
relative to Honda vehicles in October 2009 and the effects persist for three months before the two
manufacturers’ prices return to parity at the beginning of 2010. Comparing this trend to the trend
of Toyota relative to Honda in the final months of 2008 suggests that Toyota vehicles lose their
values to a larger extent than Honda vehicles in the fourth quarter of each year. The only month
where Toyota looks anomalous in 2009 relative to 2008 is December, which is consistent with the
findings from Figure 4.
Next, I refine the control group to exclude any non-Toyota vehicle that also experienced a recall
of its own. Using the National Highway Traffic Safety Administration (NHTSA) database (wwwodi.nhtsa.dot.gov/downloads), I remove the non-Toyota vehicles that were included in Figure 5
that experienced any recall prior to 2011. The results are in Figure 6; the only apparent difference
is a slightly larger treatment effect in December 2009 but that effect of −2.28% (standard error
= 2.05%) is statistically and quantitatively similar to the effect that does not exclude recalled nonToyota vehicles of −2.00% (standard error = 1.04%). I argue that these results are less informative
than those in Figure 5 because the non-Toyota recalls during this period received a tiny fraction of
the media scrutiny of the Toyota recalls and media scrutiny is the most logical driver of widespread
consumer reaction. Nevertheless, it is reassuring that the results from Figure 5 are quantitatively
unchanged.4
I then reestimate the treatment effects over time separately for each of the eighteen brands in the
sample and separately for the unrecalled and recalled vehicles of both the Lexus and Toyota brands.
Figure 7 shows the results for Lexus and Toyota in the top and bottom panels respectively. While
the relative positions of unrecalled and recalled vehicles from the Toyota brand are remarkably
constant throughout the recall period, recalled Lexus vehicles fare approximately 2% better than
unrecalled Lexus vehicles in October 2009, fare similarly in November 2009, and fare approximately
2% worse in December 2009. Figure 8 reports the treatment effects over time for Toyota and the
other foreign manufacturers; the treatment effects for domestic manufacturers are suppressed to
conserve space but are available from the author upon request.
4
All recalled non-Toyota vehicles are excluded, including some extremely small recalls (one-fourth of those recalls
affected fewer than 2,836 vehicles) and some non-safety recalls (e.g., missing vehicle labels or defective interior lights).
Including only large recall campaigns that are related to safety problems confirms that the results in Figure 5 are
quantitatively unchanged.
10
Combining the evidence in Figures 4 – 8 paints a picture of the 2009-2010 Toyota recalls mattering primarily in December 2009 and primarily for vehicles of the Lexus brand. In no specification
are the effects larger than 2.5% and they are never outside of the margin of error. If these recalls
matter, they appear to have small effects that are short lived and occur immediately after the initial
burst of media reports covering Toyota’s safety problems. That is, the effects are concentrated in
December 2009 and not in February 2010, when the negative media coverage reached its peak.
5.1
Spillovers
The previous literature suggests that spillovers of the effects on Toyota to other manufacturers
may be expected here (Grafton, Hoffer, and Reilly, 1981; Cawley and Rizzo, 2008; Freedman,
Kearney, and Lederman, 2010). The presence of positive spillovers implies that Toyota’s competitors
would have benefited from Toyota’s recalls by increased demand for their vehicles. Positive spillovers
then suggest that the above estimation in fact overestimates the effect of the Toyota recalls. Under
the hypothesis that recalls matter, the prices of Toyota vehicles will fall more rapidly and this price
drop is the true treatment effect. But this hypothesis also suggests that the prices of Toyota’s
competitors should rise for two reasons: Toyota vehicle owners replacing their current vehicles
with competing brands and potential Toyota consumers switching to competing brands. These
arguments suggest that, if recalls matter, my results overestimate their effects. Because I find that
the Toyota recalls did not have large effects, the overestimation problem instills a high degree of
confidence that the true effects of recalls are not large.
The presence of negative spillovers implies that Toyota’s competitors would have been harmed
from Toyota’s recalls by decreased demand for their vehicles. Negative spillovers then suggest that
the above estimation is masking the true effect of the Toyota recalls. Negative spillovers would
appear in the manufacturer-specific average price trends in Figure 2 as discrete breaks in trend
for multiple manufacturers coincident with the Toyota recalls. A first hypothesis suggests that all
foreign manufacturers should face price drops following Toyota’s safety problems because consumers
substituted to domestic manufacturers. A second hypothesis suggests that all manufacturers should
face price drops following Toyota’s safety problems because consumers decided to delay vehicle
purchases altogether. Neither hypothesis is supported by Figure 2 for any model year, leading me
to conclude that the quantitatively small effects of the Toyota recalls are not masking real effects
11
that are present for multiple manufacturers in the market.
5.2
Robustness Results
The above results suggest that the 2009-2010 Toyota recalls had small negative effects that
disappear by the end of 2009. In this section, I provide several robustness checks to support that
basic conclusion. First, I show that different types of recalls matter in different ways for Toyota
during this period but the effects remain small nonetheless. As detailed in Section 2, separate recalls
addressed issues with floor mats and accelerator pedals, yet most of the commentary focused on the
accelerator-pedal issues as a more-serious safety problem. Confirming this intuition, the acceleratorpedal recalls had larger effects, though both recalls had small effects and their differences are not
statistically significant. To perform this check, I rerun the analysis shown in Figure 4 using separate
regressions for each recall type, where I compare recalled Toyotas that were affected by a given
recall to Toyotas that were not affected by any recall. For the floor-mat recalls, the treatment effects
in the three months where the effects are noticeable (October, November, and December 2009) are
1.26%, −0.14%, and 0.08%. For the accelerator-pedal recalls, the treatment effects in the same
three months are 1.07%, −0.86%, and −0.48%. Intuitively, the recall that was more dangerous had
larger effects. The novelty of these findings is that the effects are small and transient.
Second, I control for recalls with a measure of recall intensity instead of a recall dummy. Of
the 245 Toyota vehicles in the sample, 41.63% were not affected by any of the recalls, 2.45% were
affected by one recall, 41.63% were affected by two recalls, and 14.29% were affected by three
recalls. Since so few vehicles were affected by only one recall, I compare (1) non-Toyota vehicles,
(2) Toyota vehicles that experienced zero recalls, (3) Toyota vehicles that experienced one or two
recalls, and (4) Toyota vehicles that experienced three recalls. The findings for recall intensity
detect no differences in the treatment effect of recalls depending on the number of recalls that
affected a vehicle. The treatment effects in the three months where the effects are noticeable
(October, November, and December 2009) are small and vehicles affected by three recalls fared no
worse than those affected by one or two recalls (1.18%, 0.04%, and 0.21% for one or two recalls
versus 1.55%, 0.61%, and 1.44% for three recalls, none of which are statistically different).
Third, I check the sensitivity of the results to the concern that a parametric correction for serial
correlation may suffer from misspecification. The main results use the first difference of Log(P riceit )
12
because the estimated autoregressive parameter ρ is sufficiently high to be considered equal to one,
thereby first differencing the dependent variable. Bertrand, Duflo, and Mullainathan (2004) test a
procedure that ignores the time-series variation in the panel and simply averages the pre- and posttreatment periods for comparison. The authors find that this approach can consistently estimate
the standard errors in difference-in-differences models. Applying their technique to the Toyota
recalls requires choosing a treatment date but I separately estimate the treatment effect of these
recalls for several candidate dates and find no evidence that recalls mattered. For December 2009, I
estimate a positive yet statistically insignificant treatment effect of 2.97% (standard error = 2.56%).
Fourth, I estimate the treatment effect of the Toyota recalls using within-model variation. Five
Toyota models experienced recalls for some but not all four model years during the sample: Corolla
model years 2009-2010 were recalled, Highlander (2008-2010), Matrix (2009-2010), RAV4 (20092010), and Sequoia (2008-2010). The results for the subset of vehicles for which there is withinmodel variation (960 observations) echo the main findings: the treatment effects in the three months
where the effects are noticeable (October, November, and December 2009) are −0.57%, −1.62%,
and −1.60%. The treatment effect in each of these months is quantitatively small and only the
effect for November 2009 approaches statistical significance with a p-value of 0.16.
In summary, the small, short-lived effects of the Toyota recalls are robust to controlling for
the type of recall and the number of recalls that a particular vehicle experienced. Further, the
estimation results are robust to an alternative specification to handle serial correlation and to
exploiting within-model variation in recalls instead of the use of across-model variation in the main
specification.
6
Comparison with the 1980s Audi Recalls
To provide context for these quantitatively small recall treatment effects, I repeat the analysis
on a similar recall episode: the 1980s safety recalls of the Audi 5000. Audi’s midsize sedan was
linked to several incidents of sudden unintended acceleration, reports that closely mimicked those
of Toyota vehicles more than two decades later. To maintain focus on the comparison with Toyota,
full details of the Audi recalls are in Appendix A and only a summary is provided here.
Why might Audi in 1986 face a different response to reports of sudden unintended acceleration
13
than Toyota did in 2009? The main difference between the two manufacturers during their respective episodes is that Audi was less familiar to American automobile consumers in 1986 than Toyota
was in 2009. First, Audi held a smaller share of the American automobile market (1.70% for Audi’s
parent company Volkswagen in 1986 versus 16.73% for Toyota in 2009). Second, Audi had been in
the American market for a short time (since the 1970 model year) and had three vehicles (a subcompact, compact, and midsize vehicle) in its lineup during the 17 years prior to its recall episode.
In contrast, Toyota had been present in the American market since the 1958 model year (52 years)
and had a full slate of six vehicles (covering each major automobile segment) since the 1970s. As a
result, consumer perception of Audi vehicles had not been fully established when consumers were
confronted with news of the manufacturer’s safety problems. In contrast, consumer perception of
Toyota vehicles was more solidified.
While issues of sudden unintended acceleration with the Audi 5000 date back to 1982, the peak
in coverage and concern about the safety of Audi vehicles came on November 23, 1986, when 60
Minutes covered the Audi 5000 in a segment entitled Out of Control. Later, the U.S. National
Highway Traffic Safety Administration found that all incidents that were investigated (including
those covered by 60 Minutes) were caused by pedal misapplication (i.e., the driver confusing the
gas pedal with the brake) and other types of driver error (Cremer and Lavell, 2010). Despite this
eventual vindication, media coverage intensified and many reports indicated that consumers reacted
to the increasing volume of negative information about the Audi 5000’s safety record.
Full details of the Audi data and its analysis are in Appendix A. In summary, the data come
from the Automobile Red Book because Black Book data are not readily available for the years
covering the Audi recalls (National Market Reports, Inc., 1984–1987). I collect data for the period
between January 1, 1984 and November 15, 1987 (in 45 day increments) on the retail used price
for vehicles from 19 manufacturers for the 1982 through 1985 model years. Because the Audi
5000 is a midsize sedan, I collect the flagship midsize sedan for the major American, Asian, and
European manufacturers, using the Berry, Levinsohn, and Pakes (1995) data to identify the key
midsize sedans in the market at the time.
The main results from the Audi data show the Audi 5000 treatment effects relative to the
Volkswagen Quantum (Figure 9) and, separately, relative to the Volvo 240 (Figure 10). Volkswagen
makes a natural comparison group because Audi is a brand within the Volkswagen Group. Because
14
negative information about the quality of Audi vehicles may spread to its parent company, I also
present results that compare the Audi 5000 to its closest competitor, the Volvo 240.5 Individual
manufacturers’ treatment effects, additional results, and robustness checks for the Audi data are
available from the author upon request.
First, Figure 9 shows that the Audi 5000 and the Volkswagen Quantum experienced large differences in their treatment effects during the entire three-year period prior to the 60 Minutes recall
coverage. While these large swings in period-to-period effects make it difficult to tell a consistent
story, there is a clear break in their relative trend after January 1, 1987. Audi’s negative treatment
effects persist for the duration of 1987. Second, Figure 10 shows that the Audi 5000 and the Volvo
240 have more constant relative treatment effects in the pre-treatment period; there is a clear break
in trend beginning November 15, 1986. The relative Audi treatment effects fluctuate until the end
of 1987, reaching a low of −6.3% on July 1, 1987. The separate Audi and Volkswagen treatment
effects (not shown) both fall in November 15, 1986, implying that both manufacturers experienced
negative effects coincident with the height of the negative media coverage of the (Volkswagen’s
subsidiary) Audi recalls. The fact that the negative effects for Volkswagen were smaller and less
persistent than the negative effects for Audi can be interpreted as evidence that Audi’s problems
spread to its parent company in the early stages of public concern but Audi alone suffered continued
damage as it became clear that only Audi vehicles were potentially unsafe.
The magnitude of the effects versus Volkswagen are −4.3%, −4.0%, −4.6%, and −4.0% for
the periods February 15, April 1, May 15, and July 1, 1986 (versus Volvo, these effects are −1.5%,
−4.2%, −4.4%, and −6.3% for the same periods). These effects are both larger and more persistent
than the effect of the Toyota recalls that were found in the previous section. Precise quantification
of the effect of the Audi recalls require choosing between the two control groups because of the
differences in the exact onset of Audi’s drop (earlier with Volvo as a control group) and the length
of time that the negative effects continued (more persistent with Volkswagen as a control group).
For the purpose of comparison with the Toyota recalls though, these results demonstrate that Audi
suffered larger losses that lasted longer than the Toyota losses despite the similarities between the
5
The Audi 5000 and the Volvo 240 are each other’s closest competitors in the sense that, of the European midsize
sedans in the Berry, Levinsohn, and Pakes (1995) data, they have the closest list price ($17,710 for the 5000 versus
$15,985 for the 240, in current model-year dollars), engine size (136 cubic inches for the 5000 versus 141 cubic inches
for the 240), horsepower (110 HP for the 5000 versus 111 HP for the 240), and vehicle weight (2,844 pounds for the
5000 versus 2,939 pounds for the 240). These figures are based on the 1985 model year for each vehicle.
15
two recall episodes. However, it is worth pointing out that comparing the Toyota and Audi recalls
does not reveal statistically significant differences in each pairwise test of equal treatment effects.
As a result, the discussion in the next section focuses only on the quantitative differences in the
patterns of the two episodes.
7
Reconciling the Results from the Audi and Toyota Recalls
The comparison presented here between the sudden unintended acceleration episodes of Audi
in 1986 and Toyota in 2009 was echoed by Business Week, who wrote that Audi “spent 15 years
rebuilding U.S. sales after sudden-acceleration incidents in the 1980s almost wiped out demand,
a possible sign of the difficult times Toyota Motor Corp. faces. Audi’s U.S. deliveries plunged 83
percent by 1991 from their peak in 1985” (Cremer and Lavell, 2010). The analysis presented in this
paper confirms the large negative effects of Audi’s recalls but contradicts the prediction of a similar
path for Toyota more than two decades later. In particular, I find that the 2009-2010 Toyota recalls
had effects that were quantitatively small and short lived.
In the same article, Business Week noted the primary difference between Audi and Toyota that
I argue explains the different outcomes that the two companies faced: “Audi was then selling no
more than 75,000 cars a year in the U.S. and was chipping away at the market for higher-priced
models. [In 2009], Toyota, the world’s largest automaker, sold 17 percent of all cars in the U.S.”
(Cremer and Lavell, 2010). Because Audi was smaller and newer to American consumers when its
safety problems were subject to intense media scrutiny, Audi vehicles suffered larger and longerlasting losses following the 60 Minutes coverage of the company’s sudden unintended acceleration
issues than did Toyota vehicles that had the same safety problems.
The contrast between Audi and Toyota highlights the role of information in determining consumer demand for products that have the potential to fail with life-threatening consequences.
Recalls for the same safety issue are not treated equally; instead, consumers update their beliefs
regarding the quality of a manufacturer’s products given their previous information set. Negative
information about the quality of the products from a company such as Audi in 1986 has the potential to be more damaging than similar negative information about Toyota vehicles in 2009. The
limited nature of American consumers’ experiences with Audi prior to its recall episode drives the
16
results that I find, which helps to provide context for the quantitatively small negative effect of the
2009-2010 Toyota recalls.
8
Conclusions and Relationship with Previous Work
Toyota’s resale prices demonstrate a quantitatively small response to a large-scale safety recall
that was sufficiently publicized to rank as one of the most covered news stories of 2009-2010. There
is a small price response to the Toyota recalls despite the overwhelmingly negative media attention
that the episode received, including pervasive coverage of more than 30 deaths (some reports as high
as 52 deaths) that were linked to the safety problems with Toyota vehicles. To provide context for
the size of the Toyota recall effect, I measure the effect of an earlier sudden unintended acceleration
recall of Audi vehicles during the 1980s. Audi faced larger losses that were more persistent than
the smaller Toyota losses that I find. The comparison between the two recall episodes highlights
that media coverage is not the driving force behind the effects of a recall in product markets.
Though a comparable measure of coverage intensity is difficult to produce because of the different
media environments, the Toyota recalls arguably received more attention than the Audi recalls.
The primary difference between the two recall episodes is that the earlier Audi recalls involved a
company that had a small consumer base and a limited history in the American market. Relative
to the 1980s Audi recalls, the 2009-2010 Toyota recalls of more than nine million vehicles had small
effects on consumer demand because Toyota’s reputation was more established in the minds of
American consumers.
In contrast to these findings, three papers present evidence that an automobile manufacturer’s
reputation is a liability when facing a severe recall. Barber and Darrough (1996) study the period
from 1973-1992 and find that the losses in shareholder wealth are largest for manufacturers like
Honda and Toyota who have a reputation for high-quality vehicles. Relatedly, Rupp (2004) finds
shareholder losses are larger for manufacturers with AAA bond ratings than for manufacturers in
worse financial condition. Finally, Rhee and Haunschild (2006) find that manufacturers with better
reputations lose more market share following a severe recall than those with worse reputations. I
argue that the previous literature measured a different characteristic of a company’s reputation
than the characteristic that I highlight. Specifically, I draw a distinction between a well-established
17
reputation (which is my focus) and a reputation for high-quality products (which is the focus of
the previous literature).
In interpreting the different outcomes faced by Audi in the 1980s and Toyota during 20092010, my focus is on the degree of certainty that consumers have in their quality estimates of a
manufacturer’s products, where the above papers focused on the level of the quality estimates. Both
Audi and Toyota were viewed as high-quality vehicles at the time of their recalls. The difference is
in the length and depth of their “track record,” where Toyota had been in the market longer and
had a larger market share at the time of its sudden unintended acceleration problems. As a result,
consumers had a more firmly cemented view of Toyota vehicles by the onset of the negative media
attention that the company faced. This increased certainty in consumers’ view of Toyota vehicles
in 2009 implied that there was less of a role for new information about the quality of their Camry,
for example. In contrast, consumers and potential consumers of Audi in 1986 had “little to go on”
when accessing the company, allowing Audi’s safety problems to resonate.
The role implied by these results for policymakers is limited, in the sense that consumers are
found to have, generally speaking, responded in a way that is consistent with rational economic
models of decision making under incomplete information on the quality of a durable good. Implications for business strategy are more clear-cut: the comparison between Audi and Toyota
demonstrates another reason for firmly establishing a reputation in which consumers have a great
deal of faith. While this recommendation is not novel, the distinction that has been highlighted
between a reputation that is well established and a reputation for high-quality products is new to
the literature on product quality. More research is needed to disentangle these effects.
References
Ahrens, F. (2010): “Generous Incentives Help Toyota Sales Rebound,” Washington Post, April 2,
http://www.washingtonpost.com/wp-dyn/content/article/2010/04/01/AR2010040103678.html.
Bae, Y.-K., and H. Benıtez-Silva (2011): “Do Vehicle Recalls Reduce the Number of Accidents?
The Case of the U.S. Car Market,” Journal of Policy Analysis and Management, 30(4), 821–862.
Barber, B. M., and M. N. Darrough (1996): “Product Reliability and Firm Value: The
Experience of American and Japanese Automakers, 1973–1992,” Journal of Political Economy,
104(5), 1084–1099.
18
Berry, S. T., J. Levinsohn, and A. Pakes (1995): “Automobile Prices in Market Equilibrium,”
Econometrica, 63(4), 841–890.
Bertrand, M., E. Duflo, and S. Mullainathan (2004): “How Much Should We Trust
Differences-in-Differences Estimates?,” Quarterly Journal of Economics, 119(1), 249–275.
Black Book (2011): “Why Black Book?,” Discussion paper, National Auto Research Division,
Hearst Business Media Corporation, www.blackbookusa.com.
Bunkley, N., and B. Vlasic (2010): “Carmakers Initiating More Recalls Voluntarily,” New York
Times, August 24, http://www.nytimes.com/2010/08/25/automobiles/25recall.html.
Cawley, J., and J. A. Rizzo (2008): “Spillover Effects of Prescription Drug Withdrawals,”
Advances in Health Economics and Health Services Research, 19, 119–144.
Copeland, A., and G. Hall (2011): “The Response of Prices, Sales, and Output to Temporary
Changes in Demand,” Journal of Applied Econometrics, 26(2), 232–269.
Cremer, A., and T. Lavell (2010): “Audi 1980s Scare May Mean Lost Generation
for Toyota Sales,” Business Week, February 4, http://www.bloomberg.com/apps/news?pid=
newsarchive&sid=aFAbB4LZ3h6A.
Darlin, D. (1986): “Audi Sets Recall Of 132,000 Autos For Pedal Problem,” Wall Street Journal,
May 30, http://online.wsj.com/public/resources/documents/WSJ-Audi05301986.pdf.
Daughety, A. F., and J. F. Reinganum (1995): “Product Safety: Liability, R&D, and Signaling,” American Economic Review, 85(5), 1187–1206.
Dowdell, T. D., S. Govindaraj, and P. C. Jain (1992): “The Tylenol Incident, Ensuing
Regulation, and Stock Prices,” Journal of Financial and Quantitative Analysis, 27(2), 283–301.
Freedman, S. M., M. S. Kearney, and M. Lederman (2010): “Product Recalls, Imperfect Information, and Spillover Effects: Lessons from the Consumer Response to the 2007 Toy Recalls,”
Review of Economics and Statistics, forthcoming.
Grafton, S. M., G. E. Hoffer, and R. J. Reilly (1981): “Testing the Impact of Recalls on
the Demand for Automobiles,” Economic Inquiry, 19(4), 694–703.
Halvorsen, R., and R. Palmquist (1980): “The Interpretation of Dummy Variables in Semilogarithmic Equations,” American Economic Review, 70(3), 474–75.
Hammond, R. G. (2010): “Quantifying Consumer Perception of a Financially Distressed Company,” http://www4.ncsu.edu/∼rghammon/research.html.
Hennigan, W. (2010): “Toyota Posts $2.2 Billion Quarterly Profit,” Los Angeles Times, August
05, http://articles.latimes.com/2010/aug/05/business/la-fi-toyota-earns-20100805.
Hortaçsu, A., G. Matvos, C. Syverson, and S. Venkataraman (2010): “Are Consumers Affected by Durable Goods Makers Financial Distress? The Case of Auto Manufacturers,” Working
Paper 16197, National Bureau of Economic Research.
Jarrell, G., and S. Peltzman (1985): “The Impact of Product Recalls on the Wealth of Sellers,”
Journal of Political Economy, 93(3), 512–536.
19
Laporte, A., and F. Windmeijer (2005): “Estimation of Panel Data Models with Binary Indicators when Treatment Effects are not Constant over Time,” Economics Letters, 88(3), 389–396.
Li, S., C. Timmins, and R. H. von Haefen (2009): “How Do Gasoline Prices Affect Fleet Fuel
Economy?,” American Economic Journal: Economic Policy, 1(2), 113–137.
MacKenzie, A., and S. Evans (2010): “The Toyota Recall Crisis,” Motor Trend, January, http:
//www.motortrend.com/features/auto news/2010/112 1001 toyota recall crisis/index.html.
Mcginley, L. (1986): “U.S. Requests Safety Recall Of Audi 5000s,” Wall Street Journal, December 24.
Mitchell, J. (2009): “Floor Mats Prompt Largest Toyota Recall,” Wall Street Journal, September
30, http://online.wsj.com/article/SB125426110738750533.html.
National Market Reports, Inc. (1984–1987): Automobile Red Book. Maclean Hunter Publishing Corp.
Nichols, M. W., and G. M. Fournier (1999): “Recovering from a Bad Reputation: Changing
Beliefs about the Quality of U.S. Autos,” International Journal of Industrial Organization, 17(3),
299 – 318.
Reilly, R. J., and G. E. Hoffer (1983): “Will Retarding the Information Flow on Automobile
Recalls Affect Consumer Demand?,” Economic Inquiry, 21(3), 444–447.
Rhee, M., and P. R. Haunschild (2006): “The Liability of Good Reputation: A Study of
Product Recalls in the U.S. Automobile Industry,” Organization Science, 17(1), 101.
Rupp, N. G. (2004): “The Attributes of a Costly Recall: Evidence from the Automotive Industry,”
Review of Industrial Organization, 25(1), 21–44.
Rupp, N. G., and C. R. Taylor (2002): “Who Initiates Recalls and Who Cares? Evidence from
the Automobile Industry,” Journal of Industrial Economics, 50(2), 123–149.
Salin, V., and N. H. Hooker (2001): “Stock Market Reaction to Food Recalls,” Review of
Agricultural Economics, 23(1), 33–46.
Schlesinger, J. M., and A. R. Karr (1986): “Audi Recalling 5000 Series Cars A Second
Time,” Wall Street Journal, September 17, http://online.wsj.com/public/resources/documents/
WSJ-Audi09171986.pdf.
Stepanek, M. (1986): “Car Consumer Lobby Petitions For A Recall Of ’78-’86 Audi 5000,” Miami
Herald, March 20.
Stuart, R. (1986): “Inquiry On Auto Acceleration Expanded By U.S.,” New York Times, February 23, http://www.nytimes.com/1986/02/23/us/inquiry-on-auto-acceleration-expanded-by-us.
html.
Vartabedian, R., and K. Bensinger (2009): “Toyota’s Runaway-Car Worries May Not
Stop at Floor Mats,” Los Angeles Times, October 18, http://www.latimes.com/business/
la-fi-toyota-recall18-2009oct18,0,739395.story.
20
A
Additional Details on the Audi Data
The 1980s Audi recalls begin with the company’s first and second recalls of the Audi 5000 on
April 16, 1982 and September 22, 1983, respectively. The first of these involved 96,000 1978-1982
Audi 5000 vehicles to install “accelerator pedal guides” to anchor the floor mats (Stepanek, 1986).
The second recall involved an additional 117,000 1978-1983 Audi 5000 automatic transmission
vehicles to install a rubber brake pedal pad. While these recalls received press attention, Audi’s
safety issues were largely ignored as subsequent news reports on Audi (especially in 1985) focused
on its increasing share of the American automobile market (Stepanek, 1986).
The next wave of media attention on Audi’s sudden unintended acceleration issues began on
February 23, 1986, when reports covered an investigation into several incidents with the Audi 5000
(Stuart, 1986). The first widespread call for the Audi 5000’s recall occurred on March 19, 1986,
when the consumer lobbyist group the Center for Auto Safety claimed that the vehicle had a rate
of one accident for every 500 vehicles sold (Stepanek, 1986). Audi’s third recall of the decade
occurred on May 29, 1986 involving the accelerator pedals of 132,000 vehicles from the 1984-1986
model years (Darlin, 1986), while the fourth recall occurred on September 16, 1986 for a separate
issue concerning the potential for engine fires (Schlesinger and Karr, 1986). As pressure on Audi
intensified, the U.S. National Highway Traffic Safety Administration urged Audi to recall 230,000
vehicles on December 23, 1986, a recall that Audi complied with on January 14, 1987 (Mcginley,
1986).6 A final, sixth recall occurred on October 23, 1987.
The height of the negative media coverage of Audi came on November 23, 1986, when 60 Minutes
covered the Audi 5000 in a segment entitled Out of Control. The segment featured interviews
with six Audi 5000 owners who were involved in a lawsuit following separate incidents of sudden
unintended acceleration. Later, the U.S. National Highway Traffic Safety Administration found
that all incidents that were investigated (including those covered by 60 Minutes) were caused by
pedal misapplication (i.e., the driver confusing the gas pedal with the brake) and other types of
driver error. Further, it was found that 60 Minutes placed a canister of compressed air connected to
the transmission with a hose, perhaps best described as “sudden engineered acceleration” (Cremer
and Lavell, 2010).
6
For perspective, compare the 230,000 vehicles that Audi recalled to the 74,061 vehicles that the company sold in
the U.S. in 1985 (Cremer and Lavell, 2010).
21
The Audi data come from the Automobile Red Book, the nation’s oldest (since 1911) usedvehicle price guide (National Market Reports, Inc., 1984–1987). I collect data for the period
between January 1, 1984 and November 15, 1987 (in 45 day increments) on the retail used price
for vehicles from 19 manufacturers for the 1982 through 1985 model years. Because the Audi
5000 is a midsize sedan, I collect the flagship midsize sedan for the major American, Asian, and
European manufacturers, using the Berry, Levinsohn, and Pakes (1995) data to identify the key
midsize sedans in the market at the time.7 The final Audi sample contains 2,091 observations,
including 76 vehicles in total. The Audi estimation controls for fuel efficiency in the same way as
with the Toyota data and each vehicle’s used price and its costs of driving (MP$) are reported in
constant 1983 dollars using the monthly Consumer Price Index.
7
The vehicles in the Audi data are as follows: American – Buick Century, Cadillac Seville, Chevrolet Celebrity,
Chrysler LaBaron, Ford LTD, Lincoln Continental, Oldsmobile Cutlass, Plymouth Gran Fury, and Pontiac Bonneville;
Asian – Honda Accord, Mazda 626, Nissan Maxima, and Toyota Corona; and European – Audi 5000, BMW 528e,
Mercedes-Benz 240D, Saab 900, Volkswagen Quantum, and Volvo 240.
22
Table 1: List of Recalls by Toyota Motors during 2009-2010
Date
Part Affected
Vehicles Affected
Country
November 2, 2009
November 26
January 21, 2010
January 27
January 29
February 8
February 8
February 12
April 16
April 19
April 28
May 21
July 5
July 29
August 28
Floor mats
Brake override system
Faulty accelerator pedals
Floor mats
Faulty accelerator pedals
Brakes
Brake tubes
Front drive shaft
Spare tire carrier cable
Stability control system
Stability control system
Steering system
Value springs
Steering column
Engine control modules
3.8 million
4.2 million
2.3 million
1.1 million
1.8 million
436,000
7,300
8,000
600,000
34,000
50,000
7,000
270,000
412,000
1.13 million
U.S.
U.S.
U.S.
U.S.
Europe, China
Worldwide
U.S.
U.S.
U.S.
Worldwide
U.S.
Japan
Worldwide
U.S.
U.S., Canada
Notes: The number of vehicles affected are cumulative and includes vehicles also affected by previous
recalls. For details, see Section 2 and the cited references.
Table 2: Recalled Vehicle Model Years in the Sample
Make
Model
Lexus
Lexus
Lexus
Toyota
Toyota
Toyota
Toyota
Toyota
Toyota
Toyota
Toyota
Toyota
Toyota
Toyota
Toyota
ES 350
IS 250
IS 350
Avalon
Camry
Corolla
Highlander
Matrix
RAV4
Prius
Sequoia
Sienna
Tacoma
Tundra
Venza
Acc. Pedal
2007-2010
2007-2010∗
2009-2010∗
2010
2009-2010
2009-2010∗
Recall
Floor Mat Tire Cable
Brakes
2007-2010
2007-2010
2007-2010
2007-2010
2007-2010
2009-2010
2008-2010
2009-2010
2007-2010
2010
2008-2010
2007-2010
2007-2010
2007-2010
2009-2010
2007-2010
Notes: Shown are the model-model years that experience each specific major recall during 20092010. ∗ denotes that the recall was partial. The accelerator pedal recall for certain models in certain
model years included only a subset of the vehicles sold, based on the vehicle’s vehicle identification
number (VIN). The remaining recalls involved every vehicle sold from the model years listed.
23
Table 3: Toyota Motors Vehicles that Experience No Recalls
Lexus
GX 460
GX 470
HS 250h
IS F
LS 460
LS 600h L
Toyota
LX 470
LX 570
RX 350
RX 400h
RX 450h
SC 430
4Runner
FJ Cruiser
Land Cruiser
Solara
Yaris
Notes: The above models from Toyota Motors did not experience any recall during the sample
period for any of the model years 2007-2010.
Table 4: Distribution of Used-Vehicle Prices over Time
Make
Model
Series
Model Year
Price
Percentile
ZX3 S
CE
Base
Base
L
2007
2008
2007
2007
2007
$9,272.01
$12,591.85
$16,461.74
$22,060.30
$56,496.37
0
25
50
75
100
August 1, 2009; N = 411
Chrysler
PT Cruiser Base
Saturn
Vue
XE
GMC
Envoy
SLE
Lexus
ES 350
Lexus
LS 600h
L
2007
2008
2008
2007
2009
$7,605.75
$13,354.06
$16,638.81
$22,758.60
$68,921.02
0
25
50
75
100
April 1, 2010; N = 492
Chevrolet Cobalt
Toyota
Camry
Nissan
Maxima
Nissan
Maxima
Lexus
LS 600h
2007
2007
2008
2010
2010
$7,026.59
$12,949.83
$16,588.95
$23,131.61
$75,743.94
0
25
50
75
100
April 1, 2008; N = 88
Ford
Focus
Toyota
Corolla
Chrysler
300
Hummer
H3
Lexus
LS 460
LS
XLE
SE
S
L
Notes: The three panels show the vehicles at each quartile of the constant-dollar used-price distribution for the first, middle, and last period in the sample, respectively. The sample sizes shown
are the number of vehicles in each period.
24
Figure 1: Number of Stories in the Google News Index
Date
Chrysler
Hyundai
Ford
Nissan
01
ja
n2
01
0
01
ap
r2
01
0
01
ja
n2
00
9
01
ap
r2
00
9
01
ju
l2
00
9
01
oc
t2
00
9
01
ju
l2
00
8
01
oc
t2
00
8
01
ap
Price
r2 10000 15000 20000 25000 30000
00
8
01
ja
n2
01
0
01
ap
r2
01
0
01
ja
n2
00
9
01
ap
r2
00
9
01
ju
l2
00
9
01
oc
t2
00
9
01
ju
l2
00
8
01
oc
t2
00
8
01
ap
r2
00 5000
8
Price
10000 15000 20000 25000
Notes: Google News does not provide an exact scale for their news index, but for perspective, the
month with the most news coverage was February 2010, when “about 1,440” news stories mentioned
the word “Toyota” and either of the words “mat” or “pedal.”
Date
GM
Toyota
Honda
Recalled
Chrysler
Hyundai
GM
Toyota
Honda
Recalled
01
ju
n2
00
9
01
ju
l2
00
01
9
au
g2
00
01
9
se
p2
00
9
01
oc
t2
00
01
9
no
v2
00
01
9
de
c2
00
9
01
ja
n2
01
0
01
fe
b2
0
01
10
m
ar
20
10
01
ap
r2
01
0
01
m
ay
2
01
fe
01 b20
m 09
ar
01 200
ap 9
01 r20
0
m
ay 9
2
01 00
9
ju
n2
01 009
ju
01 l200
au
9
g
01 200
se
9
p
01 200
oc 9
t
01 200
no
9
v
01 200
de
9
c2
01 00
9
ja
n
01 201
0
fe
01 b20
m 10
ar
20
10
10000
00 15000
9
Price
15000
20000
Price
20000
25000
30000
2008 Model Year
25000
2007 Model Year
Ford
Nissan
Date
Chrysler
Hyundai
Ford
Nissan
Date
GM
Toyota
Honda
Recalled
Chrysler
Hyundai
2009 Model Year
Ford
Nissan
2010 Model Year
Figure 2: Average Used-Car Prices by Manufacturer and Model Year
25
GM
Toyota
Honda
Recalled
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.04
Treatment Effect
-.02
0
.02
.04
Date
Unrecalled
Recalled
Time Trends for Unrecalled and Recalled Vehicles with 95% Confidence Interval
Date
Figure 3: Relative Time Trend for Recalled Vehicles with 95% Confidence Interval
26
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.05
Treatment Effect
0
.05
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.02
Treatment Effect
0
.02
.04
Date
Chrysler
Hyundai
Ford
Nissan
GM
T-Unrecalled
27
Honda
T-Recalled
Time Trends by Manufacturer, Unrecalled and Recalled Toyota Vehicles Separated
Date
Figure 4: Time Trend for Recalled Toyota Vehicles Relative to Unrecalled Toyota Vehicles
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.1
Treatment Effect
-.05
0
.05
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.04
Treatment Effect
-.02
0
.02
.04
Date
Chrysler
Hyundai
Ford
Nissan
GM
Toyota
28
Honda
Time Trends by Manufacturer, Unrecalled and Recalled Toyota Vehicles Combined
Date
Figure 5: Time Trend for Toyota Vehicles Relative to Honda Vehicles
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.1
Treatment Effect
-.05
0
.05
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.1
Treatment Effect
-.05
0
.05
.1
Date
Chrysler
Hyundai
Ford
Nissan
GM
Toyota
29
Honda
Time Trends by Manufacturer, Unrecalled and Recalled Toyota Vehicles Combined
Date
Figure 6: Time Trend for Toyota Vehicles Relative to Non-Recalled Honda Vehicles
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.1
Treatment Effect
-.05
0
.05
.1
ay
20
08
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
-.1
Treatment Effect
-.05
0
.05
Date
Unrecalled
Recalled
Lexus Vehicles Only
Time Trend for Unrecalled and Recalled Lexus Vehicles, Separately
Date
Unrecalled
Recalled
Toyota Vehicles Only
Figure 7: Time Trend for Unrecalled and Recalled Toyota Vehicles, Separately
30
ay
20 -.04
08
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
-.02
Treatment Effect
0
.02
.04
.06
Date
Acura
L-Unrecalled
Honda
L-Recalled
31
Hyundai
T-Unrecalled
Nissan
T-Recalled
Figure 8: Time Trends by Brand, Foreign-Made Vehicles
00
8
01
ju
l2
00
01
8
se
p2
00
01
8
no
v2
00
01
8
ja
n2
00
01
9
m
ar
20
01
09
m
ay
20
01 09
ju
l2
00
01
9
se
p2
00
01
9
no
v2
00
01
9
ja
n2
01
01
0
m
ar
20
01
10
m
ay
20
10
01
m
ay
2
-.1
Treatment Effect
-.05
0
.05
32
01
Date
Control Group: Volkswagen
Figure 9: Audi Results: Time Trend Relative to Volkswagen Vehicles
7
7
98
98
t1
oc
7
7
98
l1
ju
01
r1
98
6
6
98
98
n1
ja
t1
oc
ap
01
01
01
6
6
98
l1
ju
01
r1
98
5
5
98
98
n1
ja
t1
oc
ap
01
01
01
5
5
98
l1
ju
01
r1
98
4
4
4
98
98
n1
ja
t1
oc
ap
01
01
01
98
l1
ju
01
r1
ap
01
−.1
Treatment Effect
−.05
0
.05
60 Minutes Special
33
Date
Control Group: Volvo
Figure 10: Audi Results: Time Trend Relative to Volvo Vehicles
01
7
7
98
98
t1
oc
7
7
98
l1
ju
01
r1
98
6
6
98
98
n1
ja
t1
oc
ap
01
01
01
6
6
98
l1
ju
01
r1
98
5
5
98
98
n1
ja
t1
oc
ap
01
01
01
5
5
98
l1
ju
01
r1
98
4
4
4
98
98
n1
ja
t1
oc
ap
01
01
01
98
l1
ju
01
r1
ap
01
−.1
Treatment Effect
−.05
0
.05
.1
60 Minutes Special