Electronic copy available at: http://ssrn.com/abstract=2838949
Game changer? The impact of the VW emission cheating scandal on the co-integration of large automakers’ securitiesa
Paul A. Griffinb and David H. Lontc
Abstract. This paper investigates the potential change in the securities market pricing behavior of 16 large, global automakers following disclosure of the Volkswagen emission cheating scandal. The triggering public disclosure occurred on September 18, 2015, when the EPA issued a notice of violation to VW, stating that VW had intentionally circumvented the US clean air rules for diesel car emissions. The EPA notice unleashed a torrent of responses and disclosures by the company, regulators, investigators, stakeholders, and others. We first examine and contend that this event may have unblocked what economists call an informational cascade, in that much of the information on VW diesel car emissions was already known to interested parties, yet no significant market response occurred until the September 18 EPA notice. Second, we predict and find a significant change around this event in the stochastic evolution of equity and credit default swap prices in the automobile industry. In the post-emission-cheating-scandal period, this change is consistent with increased market co-integration. A test of economic significance further supports this finding by showing a decrease in the profitability of a hypothetical arbitrage trading rule based on lead-lag pricing relations in the equity and CDS markets.
Keywords. Market co-integration, informational cascade, credit default swap, equity return, market disruption, JEL Classification. D83, G12, G14, G28
1. Introduction
On Friday, September 18, 2015, the US Environmental Protection Agency (EPA) issued a notice
of violation of the Clean Air Act accusing Volkswagen AG and affiliates (hereafter VW) of using a
software device that falsified the official emission tests of almost one-half million of its diesel cars
sold in the United States in 2009–2015. While, initially, the media paid limited attention to this notice
(31 stories on “emissions tests” from “all sources,” according to Factiva on September 18), coverage
expanded rapidly over the next several days, totaling 2,299 cumulative news items by Thursday,
September 24 (based on the same Factiva search criteria). Stories of the VW emission cheating scandal
a Version of August 31, 2016. We thank Numan Ülkü for his helpful comments. Additional comments welcome. b Paul A. Griffin, Graduate School of Management, U.C. Davis, California, USA 95616 e-mail: [email protected]. Corresponding author +1 530-752-7372.
c David H. Lont, Department of Accountancy and Finance, University of Otago, Dunedin, NZL 9054 e-mail: [email protected].
Electronic copy available at: http://ssrn.com/abstract=2838949
2
continued to make world headlines over the next several months as Volkswagen announced product
recalls and management changes, regulators disclosed investigations, customers and investors alleged
fraud, and others commenced additional inquiries of emission cheating devices, including their
possible use by the other major automakers. The market reaction to these initial stories was swift and
punishing as investors drove VW’s stock price from €167.6 (close on September 17) to €119.6 (close
on September 24), amounting to a 28.6 percent drop over five trading days. As our title suggests, we
define this collective episode of news events and market pricing behavior centered on the initial EPA
notice as the “VW emission cheating scandal.”
Curiously, however, much of the information in the EPA notice was or could have reasonably
been inferred from reports already in the public domain, leaving many investors flummoxed about why
security market prices had not adjusted to these prior reports. Numerous reports suggest that the
scandal originated many years earlier, when the US and European automobile regulators tightened
their emission standards. Two published studies, in particular, highlighted serious deficiencies in the
emission tests of VW diesel cars. On May 13, 2013, Deutsche Umwelthilfe (DUH) reported significant
differences in test and actual-use results for Volkswagen and Audi diesels in Germany. On May 30,
2014, researchers at West Virginia University and the International Council on Clean Transportation
(ICCT) found and reported significantly higher actual-use emissions in a test of the 2012 Jetta TDI and
the 2013 Passat TDI.1
This paper addresses two issues that, to the best of our knowledge, remain unexplored in the
literature. We first examine whether the events and market pricing behavior around the VW emission
cheating scandal suggest the unblocking of an informational cascade (Bikhchandani et al. 1992;
Banerjee 1992; Welch 1992; Hirshleifer and Teoh 2003). Briefly, an informational cascade (discussed
further in Section 2) occurs when information in the hands of one person (or a small group) remains
blocked from others because market prices depend more on the common information and actions of
“many” rather than the information or actions of a “few.” An information shock, however, dislodges
1 These ICCT results were also passed on in May 2014 to the US EPA and the California Air Resources Board, which later confirmed the ICCT results, thus providing additional evidence for the EPA allegations.
3
the cascade if the many who have previously set prices can no longer ignore the information of a few.
Even a small change in a critical variable made public can be the tipping point that unblocks a cascade
(Grenadier 1999).2
Second, we build on the first issue and test for a significant relation between the presumed
informational cascade and a potential change in the stochastic evolution of automobile manufacturers’
securities prices. Specifically, we examine the prices of 16 of the world’s largest automakers in two
different financial markets, namely, the equity and credit default swap (CDS) markets. We analyze the
stochastic evolution of the relation between the returns in one securities market and the returns in the
other (their co-integration) and predict an increase in the co-integration of these markets in the
aftermath of the initial news of the scandal. Consistent with the notion of an informational cascade, we
associate this change in market co-integration with an increase in information aggregation from the
scandal. We also test for an implication of this prediction, namely, that the potentially greater market
co-integration in the post-scandal period generates lower arbitrage profits from a trading rule
exploiting the lead-lag relation between equity returns and changes in CDS spreads.
Our motivation to address these issues draws on three considerations. We first ask whether the
VW emission cheating scandal might resemble a real-life case of an informational cascade, which the
literature claims to exist but has so far shown mainly in theoretical or experimental terms. Second,
conditional on the first question, we predict that this real-life cascade unblocked by an outside event
has a “game changing” effect in the way financial markets operate in the global automobile industry.
We find this interesting because the events and pricing behavior we examine relate to stocks and CDSs
whose trades and market prices are set mainly by sophisticated individuals at large institutions,
2 As we explain in Section 2.1, an informational cascade is a special form of herding behavior, wherein actors’ interactions with each other produce a common, convergent behavior. What distinguishes an informational cascade from herding is that for a period of time some actors may rationally ignore their valuable private information when participating in a market situation, which can create an unstable balance of buyers and sellers interacting in that setting (Hirshleifer and Teoh 2003). Scharfstein and Stein (1990) also suggest that private information may cascade because fund managers may disregard their private information and trade with information common to many fund managers due to the potential risk to their reputation of acting differently. Wermers (1999) finds contrary evidence, however, suggesting that herding behavior relates more to concerns about private information than concerns about reputation.
4
arguably the least likely actors to succumb to the trap of common information. Third, we expect that
our study will add new results to the broader literature on securities markets’ co-integration, which
thus far shows substantial variation in the extent and direction of co-integration, perhaps because most
of the prior studies (discussed in Section 2) are non-industry-specific or do not focus on co-integration
dynamics around an exogenous event, with the possible exception of studies of bank securities around
the 2007–2008 global financial crisis (Trutwein and Schiereck 2011; Avino, Lazar, and Varotto 2013;
da Silva, Rebelo, and Afonso 2014; Narayan, Sharma, and Thuraisamy 2014; Forte and Lovreta 2015).
Our study generates the following results. We first conclude that the September 18, 2015 EPA
notice may have unblocked market behavior that resembles an informational cascade, in that pertinent
information questioning the VW emission tests was already in the public domain. Yet, based on an
event study, that information had been poorly aggregated in equity prices and CDS spreads before the
EPA notice. The prior known information had also come from credible sources, namely, independent
research groups. Second, we find a significant and sustained change in the stochastic evolution of
prices in the equity and credit markets consistent with accelerated information aggregation from the
unblocking event. Specifically, we find a significant decrease in the predictive pattern between lagged
equity returns and contemporaneous changes in CDS spreads and a significant increase in the
predictive pattern between contemporaneous equity returns and contemporaneous changes in CDS
spreads from before to after the unblocking event. We find no evidence, however, of a significant
change in the predictive pattern between lagged CDS spread changes and contemporaneous equity
returns, consistent with the relative absence of informed traders in the CDS market compared with the
stock market.
Together these documented changes in the evolution of equity returns and CDS spread changes
have three important implications. They first imply that the VW emission cheating cascade was
potentially so powerful and sustained that it changed the relative informedness of the equity and credit
markets, specifically, investors’ informedness regarding the securities of VW and the other German
automakers. Second, they imply that the worldwide attention given to the VW emission cheating
5
scandal potentially improved capital market efficiency by increasing the speed of response of
automakers’ securities to new information, especially in the CDS market. Third, they imply a decrease
in the potential profitability of arbitrage trading based on lagged correlations between automakers’
equity prices and CDS spreads. We test and find that the profitability of a hypothetical trading rule
based on a one- to two-day lag in the incorporation of equity returns into CDS spread changes
decreases in the post-scandal period. This trading rule may have practical relevance, in that the stocks
and CDSs of the large automakers trade frequently, thus allowing short trading in stocks and CDSs to
occur over narrow intervals with minimal liquidity risk.
Our paper proceeds as follows. Section 2 states the hypotheses following a discussion of the prior
literature. Section 3 describes the sample and research design. Section 4 presents the results, and
Section 5 concludes.
2. Literature and hypotheses
2.1. Cascades in financial markets
Bikhchandani et al. (1992) (theoretical model and examples), Banerjee (1992) (theoretical
model), and Welch (1992) (theoretical model and IPO pricing application) were among the first to
advance the notion that informational cascades can occur in rationally functioning markets. Those
papers contend that an informational cascade occurs “when it is optimal for an individual, having
observed the actions of those ahead of him, to follow the behavior of the preceding individual without
regard to his own information.” Bikhchandani et al. (1992, 994). Those papers further advance the
view that informational cascades are fragile or brittle, and can be shattered or dislodged by the arrival
of “a little information or the mere possibility of value change (even if the change does not occur)…”
Bikhchandani et al. (1992, 1004). Hirshleifer and Teoh (2003) review the literature on cascading (and
herding behavior) and in the context of rational learning in capital markets. They point out that while
cascading and herding share much in common (e.g., about the effects of learning from observing or
imitating the actions of others), a distinguishing feature of a cascade is that the “observational learning
... is so informative that an individual’s action does not depend on his own private signal” (Hirshleifer
6
and Teoh 2003, 28). Their study suggests that the presence of the following seven interrelated
characteristics (which we apply to examine qualitative hypothesis H1 below) will indicate a strong
likelihood of a cascade: (1) slow and poor information aggregation about what could be attained with
less imitation or observation of others’ earlier actions or behaviors; (2) an evolving publicly
observable state variable dependent on credible endorsement; (3) information idiosyncratic to the
investor or firm affecting the follower investors; (4) a large pool of public information held by many
compared with some critical information held by a few; (5) an observable shock that unblocks the
cascade, producing accelerated information aggregation; (6) a fragile pricing equilibrium that could be
dislodged by the smallest signal; and (7) diverse investor preferences and mixed investment payoffs
before cascade dislodgement. We state the following qualitative hypothesis based on how these
characteristics accord with the information flow and pricing behavior around the VW emission
cheating scandal.
H1: The VW emission cheating scandal resembles an informational cascade.3
2.2. Information shocks and the co-integration of security prices
Numerous studies examine the joint evolution of prices in multiple securities markets or, more
generally, the co-integration of those markets, often to understand the location of informed investors.
The key studies in this area (Kwan 1996; Norden and Weber 2004; Blanco, Brennan, and Marsh 2005;
Longstaff, Mithal, and Neis 2005; Zhu 2006; Berndt and Ostrovnaya 2008; Norden 2009), however,
show little uniformity of findings that might help benchmark our results on the co-integration of large
automakers’ equity prices and CDS spreads. The more recent literature, though, favors equity investors
as the more informed traders. Hillscher, Pollet, and Wilson (2015) find that equity returns lead CDS
spread under most market conditions, suggesting that while CDS dealers might represent sophisticated
institutions, most CDS expertise involves credit risk management where liquidity and hedging can
have priority over price discovery. Griffin, Hong, and Kim (2016) offer a similar conclusion, showing 3 As a qualitative hypothesis, we do not subject H1 to rigorous statistical testing as required for a quantitative hypothesis. Rather, we seek to understand a set of events and behaviors and to infer from those events and behaviors whether they might best be characterized as resembling the theoretical construct of an informational cascade.
7
the existence of a significant predictive pattern between equity short selling and future CDS returns,
implying that informed investors more likely reside in the equity markets.
Closer to our investigation, several studies examine whether and how the recent banking crises
might have influenced the lead-lag pricing relations in the equity and CDS markets. But these studies,
too, show mixed results, thus offering little guidance as to the expected market association effects with
and without an information shock. They also focus primarily on firms in the financial sector. For
example, Berndt and Obreja (2010), Avino et al. (2013), and Narayan (2015) find that CDS spreads
were more informative than equity prices during the global financial crisis of 2007–2008; whereas
Narayan et al. (2014) and Forte and Lovreta (2015) find the opposite result – that equity prices were
more informative than CDS spreads during the same period. In addition, Trutwein and Schiereck
(2011) and Wang and Moore (2012) find that both the equity and CDS markets became more
integrated during the global financial crisis, consistent with the Merton (1974) prediction of increased
integration of stock prices and CDS spreads during periods of financial distress. In contrast, da Silva,
Rebelo, and Afonso (2014) find that the association between stock prices and CDS spreads did not
change during the 2008 US financial crisis or the 2010 European sovereign debt crisis, inconsistent
with the Merton (1974) prediction of greater integration.
These mixed results mean that it remains an open question as to whether informed trading for
automakers’ securities occurs first in the CDS market or the equity market and whether this might
change during periods of accelerated information aggregation. As such, for the purposes of hypothesis
testing, we first establish an empirical model of the CDS equity market relation in a setting without a
significant information shock of the caliber of the EPA notice (i.e., the period prior to September 18,
2015). We then assume that the model remains unchanged in the period after the information shock, so
that only changes in the parameters of that model potentially reflect a response to that shock.4 Given
the prior literature, those changes could favor the location of more informed trading in the CDS or
equity market (or both). Accordingly, we state the following hypothesis in the alternative form of an 4 We also conduct an event study to test for significant market reactions to those pre-scandal events that, potentially, could have revealed precursory information about the information shock on September 18, 2015.
8
increase in the co-integration of automakers’ equity and CDS security prices from the accelerated
change in information aggregation following the VW emission cheating scandal.
H2: The VW emission cheating scandal changed the stochastic evolution of prices of automakers’ securities in the equity and credit markets consistent with an increase in the co-integration of those markets.
There is no reason, however, to expect that the change in global automakers’ market co-
integration in response to the VW emission cheating scandal should be directionally identical for all
firms, although common regulatory environments and industry practices suggest that no automaker
could expect full immunity from the effects of a negative market reaction. As the scandal developed,
investigators and others not only inquired about the practices of other German automakers (BMW and
Mercedes), but they also questioned other European firms, and those in the United States and Asia as
well. As such, while none of the other automakers was initially subject to the same level of scrutiny as
VW, rational investors would undoubtedly have changed their expectations that emission cheating
would not occur for the other automakers; and, with hindsight, some later stories and stock price
reactions (discussed below) confirm this view. We, therefore, expect directionally similar but smaller
changes in the model parameters, reflecting market co-integration for the non-VW automakers in the
sample. For example, in November 2015, the German vehicle regulator Kraftfahrt-Bundesamt (KBA)
announced that it had broadened its emission cheating investigation to all major global automakers. In
December 2015, DUH, the same group that first reported on VW emissions in May 2013, reported test
results showing that BMW and Mercedes also cheated on emission tests, although the automakers
vigorously denied those reports.5 Given the widening reach of the scandal to the other major
automakers, we, therefore, test that H2 varies in the cross section based on the operating environment
of each automaker. Specifically, we test the following hypothesis.
H3: The increase in the co-integration of the equity and credit markets in response to the VW emission cheating scandal varies with the operating environment of each automaker.
5 These denials by the other major automakers also, potentially, establish conditions for future informational cascades (e.g., endorsement, fragility, diverse investor preferences, slow information aggregation). Similar to the EPA notice, these potential future cascades might also be dislodged by the report of a regulator.
9
3. Sample and research design
3.1. Sample
We obtain our sample of 16 of the world’s largest automobile manufacturers by merging the 38
constituents of the S&P Global Autos daily stock index with those automakers in the index with
continuously available daily five-year CDS spread data for the same period. We source these data from
S&P Capital IQ, DataStream, and Bloomberg. This results in daily stock price data from January 2,
2004, to March 23, 2016, and five-year CDS spread data for the same period or from the initiation of
CDS trading if later than January 2, 2004. Because these data reflect different currencies, we convert
the stock prices and CDS spreads to returns and percentage changes in spreads, respectively, for the
empirical tests.6 We also collect relevant financial statement information from S&P Capital IQ,
DataStream, and Bloomberg and the Fama-French index factors from the Kenneth French Web site.
Lastly, we collect emission-related news stories from before and after September 18, 2015, from
Factiva and ProQuest using the search term “auto emissions tests” applied to “all sources.”
We also limit our sample to VW and other large global automobile manufacturers for reasons
unrelated to the CDS data constraint. First, investors and the media considering the VW emission
cheating scandal paid considerable attention to these firms, especially compared with the other firms in
the S&P Global Auto index, such as large automobile parts suppliers and automobile servicing firms.
These firms, in expectation, had little involvement in the emission cheating scandal. Second, investors
trade the large automakers’ equities and CDSs in liquid and efficient markets (and in multiple markets
for the automakers’ equities), so that we can rely on daily equity returns and spread changes to reflect
trade prices rather than approximations based on bid-ask quotes or order flow.7 The extensive trading
of these firms’ securities also makes it less likely they would experience an informational cascade.
Third, we consider these firms comparable in most respects other than their potential use of emission
cheating devices, so they would be natural alternatives for inclusion in investors’ portfolios.
6 To ensure and to maximize daily returns and spread changes for each firm, we check for securities market holidays. If the home country market was closed on a given day, we use the equivalent closing price of an exchange-traded security if available (source: S&P Capital IQ). 7 This issue can be particularly acute in the CDS market because small firms’ CDSs can trade infrequently.
10
To illustrate the extent of market disruption around the first news of the scandal, we first take a
close-up view and summarize in Table 1 the daily trend of market capitalization, daily CDS spread,
and daily equity return from September 18, 2015 (EPA notice of violation) to September 25, 2015, the
end of the first trading week of the scandal. For VW, the world’s largest automaker by sales at the
time, equity market capitalization drops by €21.6 billion or a decline of 28.6 percent. Porsche’s stock
price (Porsche SE currently owns 50.73 percent of Volkswagen AG) also drops substantially, by 32.8
percent. In fact, all automakers drop in market capitalization over this initial period, although the US
and Asian firms experience smaller declines. From the standpoint of credit risk, these stock price drops
increase the probability of default on firms’ outstanding debt, with the result that CDS spreads should
increase. As expected, Table 1 shows that the five-year CDS spreads increase substantially for VW
and Porsche, increasing by 203 percent and 116 percent, respectively. On the other hand, the spread
increases for the other automakers are much lower. We also note from Table 1 that over this first week
of trading that not one firm escaped a market penalty – of either a decrease in stock price or an
increase in CDS spread.
[insert Table 1 here]
Table 1 also enables a preliminary check on the extent of market co-integration around the short
interval of scandal-initiating events, by calculating the percentage of same firm-days with spread
changes and returns of the opposite sign. Focusing on the Table 1 interval of t = September 18 to 24,
2015, we calculate that 68.75 percent of day t returns and day t CDS changes have the opposite sign
versus 59.38 percent of day t-1 returns and day t CDS changes and 54.7 percent of day t returns and
day t-1 CDS changes. While an imperfect measure, this suggests partial co-integration during this
window, since if the equity and CDS price setters were equivalently informed they would in theory
respond oppositely 100 percent of the time on day t, assuming both markets viewed the news items as
equally risk relevant.8
8 We summarize the results of a formal test of this notion in Section 4 using firm-level time-series regressions.
11
As a second illustration of market disruption, we focus on the timeline of VW during 2015 and
2016. Figure 1 aligns Volkswagen AG’s stock price, CDS spread, and the number of ProQuest news
stories mentioning "auto emissions tests" daily from January 1, 2015 to March 26, 2016. This graph
clearly shows a precipitous decline in stock price and a significant peak in CDS spread around the
initial emission disclosures, with the largest spikes occurring on September 21 and 22, the first two
trading days after the EPA notice. The news stories also jump significantly on those days, consistent
with accelerated information aggregation, although they peak several days later (September 25) as the
scandal expanded and deepened.
[insert Figure 1 here]
We also plot in Figure 2 the daily equity return of Volkswagen AG from January 1, 1980 to
March 26, 2016. As a measure of stock volatility, we note that this plot shows several spikes over the
longer period. One inference is that the events surrounding the emission cheating scandal of late 2015
do not represent unusual stock volatility that was not experienced earlier by VW stockholders. Because
stock volatility has been shown to proxy for the speed of information aggregation potentially caused
by varying degrees of investor attention (Andrei and Hasler 2015), the presence of prior periods of
high stock volatility makes it even more puzzling that VW, potentially, would experience an
informational cascade around the emission cheating scandal.9 This further motivates our study.
[insert Figure 2 here]
3.2. Research design
Our research design involves (1) analyzing qualitatively whether the events and market behavior
around the emission cheating scandal have the hallmarks of an informational cascade unblocked by the
disclosures of September 18, 2015, and shortly thereafter (H1), and (2) testing statistically whether the
accelerated information aggregation from the scandal changed the co-integration of the equity and
CDS markets (H2 and H3). We test for a change in market co-integration in three steps. First, for each 9 Figure 2 also suggests no long-term change in daily stock volatility over the 16-year data period, and that the stock price changes around the EPA notice of September 18, 2015, do not relate to an unusual spike in stock price volatility. In later statistical tests, we control for heteroskedasticity in daily equity returns.
12
firm, we regress daily CDS spread changes on contemporaneous and lagged equity returns, with
controls for Fama-French market factors. We then estimate equivalent regressions with daily returns as
the dependent variable. The models are
CHCDSt = α + β1MKTt + β2SMBt + β3HMLt + β4RETt + β5RETt-1 + β6RETt-2 + εt (1) RETt = α + β1MKTt + β2SMBt + β3HMLt + β4CHCDSt + β5CHCDSt-1 + β6CHCDSt-2 + εt (2)
for t observations from January 1, 2004 (or the initiation of CDS trading if later) to March 16, 2016,
where CHCDS = percentage change in five-year CDS spread, MKT = return on market index, SMB and
HML = Fama-French size and book-to-market ratio factors, respectively, and εt = uncorrelated error,
all on day t, and where RETt-k and CHCDSt-k = equity return and percentage change in five-year CDS
spread day t-k for k = 0, 1, and 2, respectively.
Second, we split the data series into pre- and post-emission-scandal observations and conduct the
same analyses for each partition. The changes in the coefficients from these pre-post analyses enable
us to test for a potential change in the co-integration of these markets following the accelerated
information aggregation from the emission cheating scandal. For example, should lagged equity
returns predict CDS spread changes in the pre-scandal period, we would expect significantly negative
coefficients for β5 or β6 in Eq. (1). In addition, should lagged equity returns not lead CDS spread
changes in the post-scandal period, potentially, because of increased co-integration of the equity and
CDS markets, we would not expect significantly negative coefficients for β5 or β6 in Eq. (1) in the
post-scandal period. However, we would expect the coefficient for β4 in Eq. (1) to become more
negative from the pre- to the post-scandal period, arguably reflecting increased co-integration from a
stronger contemporaneous relation between equity returns and spread changes.10
Third, as a check on steps one and two, we estimate firm-level vector autoregression (VAR)
models of order two for daily equity returns and spread changes. As a precondition for this analysis,
10 We also conduct analogous tests assuming that lagged CDS spread changes predict equity returns in the pre-scandal period, although if our results strongly support the view that informed trading resides in the equity market (Eq. 1), then this second analysis becomes a robustness test, in that we should not expect significant coefficients for β5 or β6 in Eq. (2) in the pre- or post-scandal periods.
13
we conduct Dickey-Fuller and Phillips-Perron tests of time-series stationarity. Based on the signs and
significance of the coefficients from these regression approaches, we assess whether equity returns
lead CDS changes or vice versa over the full and the pre- and post-sample periods. We also conduct
firm-level tests of Granger causation, which is another way to understand whether the emission
cheating scandal induced changes in the lead-lag relation between daily equity returns and spread
changes for our automaker sample.
Fourth, we examine the pre-scandal observations in both markets to assess the likelihood that
precursory events and reports relating to the EPA notice of September 18, 2015, might have dislodged
the presumed information cascade earlier by conducting an event study. We select six emission-related
events in the public domain that investors might reasonably have inferred as relating to VW cheating
on its emission tests prior to the EPA notice.
As a final step test of increased market co-integration, we implement a hypothetical trading rule.
We premise this rule on the assumption that lagged equity returns more likely predict CDS spread
changes than vice versa. Specifically, we take a short position in the CDS of each automaker on day t
if either of RETt-1 or RETt-2 < 0 and hold the short position for k = 1 to 5 days. We take no action if
either of RETt-1 or RETt-2 > 0.11 We then measure the change in CDS spread over for k = 1 to 5 days as
a proxy for the profitability of this trading rule. If the co-integration of the equity and CDS markets
increases in the post-scandal period, the profitability of this trading rule should decrease.12
4. Results
4.1. Analysis of informational cascade characteristics
This sub-section qualitatively compares the seven characteristics of an informational cascade with
the events and market behavior around the VW emission cheating scandal (outlined in sub-section
2.1). We state the cascade characteristic (numbered C1 to C7) as shorthand to indicate whether the 11 An alternative would be to also take a long position in the CDS on day t if either of RETt-1 or RETt-2 is positive, with the expectation that CDS spreads would decrease in days t to t+k. While we conduct this alternative test as a robustness check, our focus on negative equity returns at t-1 or t-2 comports better with the news environment surrounding the emission cheating scandal, wherein most new stories reflected a negative tone. 12 We also consider an analogous trading rule test assuming that lagged CDS spread changes predict equity returns. However, if this assumption does not hold, we would not expect meaningful results.
14
market behavior before September 18, 2015, more likely than not resembles the cascade
characteristic.13 We first consider whether “slow and poor information aggregation” occurs (C1) and
whether we observe an “evolving state variable dependent on credible endorsement” (C2). We note
first that the emission cheating scandal dates back more than a decade when the United States (in
1999) and Europe (in 2005) adopted more stringent emission standards.14 Because these standards
made it obvious that VW’s diesel engines would fail, VW introduced the EA 189 engine in many of its
new diesel models, which it claimed met the new standards. These models apparently heralded a new
strategy to achieve record sales growth in Europe, the United States, and elsewhere. Several of these
models secretly contained emission cheating software, however. During this period, the news from
VW was generally positive and news stories to the contrary relating to emissions were quickly rebutted
by VW and, sometimes, by the German regulatory authorities, arguably interested in promoting the
domestic automobile industry. Viewed collectively, we contend that VW’s motives for this period of
positive news and growth laid the conditions for “poor information aggregation,” especially that of
information adverse to VW and the industry (C1). While much of the positive information on
emissions came from the company, with some support of the government regulator, from the
standpoint of a cascade, some of this information also relates to an “evolving state variable” (i.e., the
accuracy of emission test results) that was “dependent on credible endorsement” (i.e., from the
company and government regulator) (C2). We also surmise that the importance of emission testing
information for firm valuation would have been largely idiosyncratic, that is, not a common variable
regularly considered in investors’ valuation models. We, therefore, construe the pre-scandal
information environment as meeting the cascade characteristic of involving idiosyncratic information
(C3).
13 For this paper, our analysis of the events and behavior leading up to the EPA notice is purposely brief and is intended to capture only key items that, potentially, relate to our hypothesis of an informational cascade. For more detailed analyses of the timeline leading to the EPA notice, we refer the reader to other sources, such as the chronologies in www.zerohedge.com/, www.cars.com, and http://www.nytimes.com. The earliest known origin of the scandal dates back to 2006, according to reporter Jack Ewing of the New York Times, April 26, 2016. 14 In Europe, the Euro 4, 5, and 6 standards (http://ec.europa.eu/environment/air/transport/road.htm); and in the United States, the phase-in of the CAP 2000 (Compliance Assurance Program) in 1999 (www3.epa.gov/otaq/regs/ld-hwy/cap2000/f99005.pdf).
15
We further contend that some critical information could have been known to a least a “few”
through public disclosures (C4). While this information could have been used by some in the
investment community, the market did not move in a way to reveal the information. For example, at
the September 2007 motor show in Frankfurt, the DUH publicly announced that VW diesels had 45
percent higher actual-use emissions than those stated by manufacturers. DUH later called for a
repetition of the tests by the German regulators, although regulators did not respond. To examine C4,
we analyze the following six events occurring in 2013–2014 that, presumably, at least a few investors
could have associated with the emerging scandal. If investor behavior resembles C4, we predict no
significant adverse market reaction to the following events: (1) On May 13, 2013, an audit by DUH
finds further discrepancies between VW test results and those obtained from actual use and criticizes
the close links between the German government and the automotive industry (DUH audit). (2) In early
2014, the International Council on Clean Transportation (ICCT) in the United States tests in-use
emissions for diesel cars and reports that two Volkswagen models failed, exceeding the guidelines 35
times in extreme settings. On May 23, 2014, researchers at West Virginia University (WVU) and the
ICCT send a letter to Volkswagen summarizing their findings (WVU memo to VW). (3) Following
this letter, WVU and ICCT publicly announce their findings of significantly higher in-use emissions
for the 2012 Jetta TDI and a 2013 Passat TDI (WVU report released publicly). (4) On November 14,
2014, Volkswagen responds that the differences amount to technical issues and "unexpected" test
conditions (VW memo on diesel issue). (5) On December 2, 2014, VW recalls 500,000 cars in the
United States due to “technical problems” (VW recall). The California Air Resources Board and the
EPA, however, do not accept this response as sufficient, and later (July 2015) both groups state
privately they would not approve Volkswagen’s diesel vehicles from 2016 unless the firm complies
with US emission standards. (6) Finally, on September 3, 2015, VW admits to the manipulation of
diesel car emissions tests during closed-doors talks with the US Environmental Protection Agency
(VW admission to EPA), although this was not revealed as a press release. We reasonably contend that
these announcements in 2013–2014 would have been available to at least a few.15 15 Also, in 2014, news stories revealed that several large automakers (Ford, Hyundai, Kia) agreed to pay fines to
16
Following the September 3, 2015 admission to the EPA that it had falsified its emission figures
with the aid of special software, the EPA issued a public notice of violation (September 18, 2015)
claiming VW’s use of emission cheating software. While, in hindsight, this notice triggered the
accelerated information aggregation in the securities markets in expectation, the release of the EPA
notice revealed little new information regarding the higher level of emissions from in-use tests (that
would have been considered highly probable based on the prior published reports), except that VW
was now formally on notice from the US regulator as not meeting the US standards. We, therefore,
construe the EPA notice and the subsequent market behavior as consistent with characteristic C5 (an
unblocking event) and C6 (an unexpected change in market equilibrium). We are less clear about the
presence of diverse investor preferences (C7). However, we contend that because VW was the world’s
largest automaker by sales at the time, investors’ preferences and payoffs would likely have extended
beyond investors’ traditional needs for financial returns, for example, to concerns about VW’s broader
benefits of employment and contribution to the macro economy.16
4.2. Event study of pre-scandal events relating to VW emissions
While it is clear from sub-section 4.1 that some critical information would likely have been
publicly known to a least a “few” investors (C4), we further check this feature of a cascade by
conducting an event study of key prior events related to the scandal. Should this analysis indicate a
significant negative response to those prior events, then this would be contrary to criterion C4, namely,
that the precursory scandal information was known only to a few and not to many. We analyze the
same six events that could have presaged an oncoming scandal, as discussed in the prior section. These
occurred on 5/13/2013 (DUH audit), 5/23/2014 (WVU memo to VW CEO), 5/30/2014 (WVU report
released), 11/14/2014 (VW memo on diesel issue), 12/2/2014 (VW recall), and 9/3/2015 (VW
the EPA for overstating the fuel efficiency of their cars sold in the United States. While these EPA investigations did not uncover the use of emission cheating devices, the intentional overstatement of fuel efficiency achieved a similar purpose, by allowing these firms to secure emissions credits from the EPA to offset the higher emissions from their larger and more profitable cars. 16 Given that each of these seven characteristics more than likely resembles market behavior before September 18, 2015, the probability that all seven would not be present would then likely be less than 1 percent (0.57 assuming independence).
17
admission to EPA). We calculate the excess returns around these events based on a market-adjusted
and a market model. For the market model, we estimate the model parameters using up to 255 trade
days not within 30 days of the announcement date.
Table 2 summarizes the results. Of the six pre-September 18, 2015 events, none shows a
significantly negative excess mean return, and one (November 14, 2014) shows a significantly positive
mean excess return. 17 In other words, of the key events that potentially could have revealed
information about the emission cheating scandal, none shows a negative reaction to the news.
Together, these suggest little reaction to public knowledge relating to the scandal other than the
reaction to the September 18 EPA disclosure. For that event, it did impact automakers’ returns. The
reaction was swift and negative for the automaker sample, with the industry losing 3.2 percent of
shareholder value over days -1 to 1. Also, as noted earlier, as the news of the scandal became more
widespread, prices dropped even further after day 1. Volkswagen, for example, declined by 28.6
percent over days 0 to 5. In short, consistent with cascade criterion C4, these results indicate that the
lead-up events prior to September 18 remained either private or were not sufficiently newsworthy to
trigger sufficient concern by investors to move prices.18 The September 18, 2015 news of the scandal,
on the other hand, began to flood the market causing a major spike in information aggregation and a
commensurate drop in automakers’ share values. Based on the preceding qualitative analysis, we
conclude that this evidence supports H1, namely, that the events and behavior leading up to the VW
17 On November 15th (day 1), Volkswagen also reported that October group sales had grown by 5% over the previous month (Source: Factiva). 18 As an additional check of whether some investors might have known about and acted on this information, we examined the bi-weekly series of the ratio of short interest in VW’s ADR security (VLKAY) to the number of ADRs outstanding. The average bi-weekly ratio from January 1, 2015 to September 15, 2015 of 0.13 percent jumped only after the September 18, 2015 EPA notice, to 3.37 percent as of September 30, 2015. The average ratio from September 30, 2015 to March 31, 2016 was 1.91 percent. This suggests that shorting investors as a group did not act on the probability that VW would reveal its use of emission cheating devices. Hence, for these specialized investors, the EPA notice would also have been consistent with an unblocking event (characteristic C5). The evidence on whether institutional investors can anticipate scandals is mixed, however. For example, while Bernile, Sulaeman, and Wang (2015) find that institutional investors were able to anticipate firms’ activities consistent with stock options backdating prior to firms’ actual announcements, there is no evidence of a market reaction prior to March 18, 2006 (the day of the Wall Street Journal article “Perfect Payday”), even though some media had reported the behavior several years earlier (Bizak, Lemmon, and Whitby 2009; Heron and Lie 2007).
18
emission cheating scandal resemble an informational cascade dislodged by the unblocking event of the
September 18, 2015 EPA order.
[insert Table 2 here]
4.3. Analysis of market co-integration: All observations
As a pre-condition for the lagged regression and vector autoregression analysis, we first test for
stationarity in the time series of equity returns and CDS spread changes. Panel A of Table 3 shows that
we can reject the hypothesis that the individual time series are non-stationary in all cases based on an
augmented Dickey-Fuller test and a Phillips-Perron test (Schwert 2002).19 We also estimate the
autocorrelation functions for each firm and summarize the results in Panel B of Table 3. This panel
shows differences in the evolution of spread changes and equity returns. Specifically, the spread
changes show positive autocorrelations for the smaller lags (1, 2, and 3), consistent with an
autoregressive process wherein a contemporaneous change reflects a portion of past changes.20 On the
other hand, equity returns show positive autocorrelations at lag 1 and mostly negative autocorrelations
at the other lags. Whereas the positive coefficients at lag 1 suggest potential equity return momentum
(Jegadeesh and Titman 1993) or data mining (Jegadeesh and Titman 2001), the negative lags are more
consistent with equity returns evolving randomly, as they would with informed and efficient trading.
The results in Table 3, however, do not formally examine which market might best reflect the presence
of informed traders.
[insert Table 3 here]
Table 4 summarizes the results of tests of Eq. (1) (Panel A) and Eq. (2) (Panel B) for the full
sample period, that is, without allowance for the potential effects of the cheating scandal informational
cascade. We conduct this analysis to benchmark our results to findings in other settings. First, with the
exception of Kia Motors, Panel A shows strongly negative β4 coefficients for RETt, consistent with an
19 In untabulated tests, we also show that we cannot reject the hypothesis that automakers’ stock prices and CDS spreads are non-stationary. 20 One possibility is that the positive autocorrelation at the smaller lags is due to the more predictable “carry” portion of CDS spread changes.
19
expected negative contemporaneous relation between equity returns and CDS spread changes at the
firm level. This negative relation is also significant at the sample level for the automakers we study, so
that CDS changes and equity returns reflect an inverse relation in general. More interestingly, though,
Panel A also shows mostly negative β5 coefficients for RETt-1 and mostly negative β6 coefficients for
RETt-2. Thus, overall, CDS changes associate negatively with contemporaneous and lagged equity
returns. This is prima facie evidence that informed trading resides in the equity market versus the CDS
market. To further test this idea, we conduct analogous reverse regressions and summarize the results
in Panel B. If informed trading locates in the equities market, we should observe substantially fewer
significantly negative coefficients for lagged CHCDS in Eq. (2) versus Eq. (1). Panel B shows this
result; namely, we observe 7 negative coefficients in Panel B versus 21 in Panel A. Also, in Panel B,
the mean β5 coefficient is not significant at lag 1 and the mean β6 coefficient at lag 2 is significantly
positive. In short, the evidence of CDS predictability using equity returns is much stronger than the
evidence of return predictability using CDS spread changes.
[insert Table 4 here]
4.4. Change in market co-integration
Table 5 summarizes firm-level regressions of the change in CDS spread on equity return, lagged
daily equity return, and control variables before and after the first news of the VW emission cheating
scandal. Our focus, though, is the change in the β4, β5, and β6 coefficients from the pre- to the post-
scandal period, which we predict should decrease for β4 and increase for β5 and β6, reflecting the
hypothesized effects of increased co-integration in the CDS and equity markets from accelerated
information aggregation (H2). For the sample as a whole, our evidence supports H2, in that the mean
β4 coefficient changes by -0.408 (p < 0.01). Moreover, for each of the German automakers, the β4
coefficient decreases significantly (p < 0.10) from the pre- to the post-scandal period. Our evidence,
however, does not show a significant increase in the mean β5 coefficient, and the mean β6 coefficient
decreases, which is inconsistent with H2.
20
[insert Table 5 here]
On closer inspection, however, we observe substantial variation in the change in the regression
coefficients across the sample. We first illustrate this change by focusing on VW, the firm most
subject to accelerated information aggregation. Figure 3 plots the contemporaneous equity and CDS
market reactions for the 20 days with the greatest negative equity return during January 2, 2004 to
March 23, 2016. This plot shows that for the 17 days before September 2015, the response of the CDS
market is much lower than the response for the three days in September. As anecdotal evidence, this
suggests a different reaction in the CDS market to large negative equity returns from before to after the
emission scandal.
[insert Figure 3 here]
Next, we focus on the sample and calculate the mean changes in the β4, β5, and β6 coefficients by
primary region of operations. We report the mean changes (post minus pre) in Table 5, which for
convenience we plot in Figure 4. This plot shows that the change in β4 decreases substantially for VW
(which admitted to emission cheating) and for the other European automakers (some of which have
been accused of emission cheating). For the US and Asian automakers in our sample (for which no
formal accusations of the use of emission cheating devices have been made as of the date of this
paper), the plots show almost no change in the mean coefficients.21 In other words, the disruption in
market co-integration most visibly occurs for the German automakers and, to a lesser degree, the other
European firms. This evidence supports H3, namely, that the increase in the co-integration of the
equity and credit markets in response to the VW emission cheating scandal varies with the operating
environment of each automaker.
[insert Figure 4 here]
21 Japanese automobile regulators, however, have conducted tests documenting that Toyota, Nissan, and Mitsubishi overstated emission levels for certain of their diesel-powered passenger cars (Wall Street Journal, March 4, 2016).
21
4.5. Vector autoregression analysis
Table 6 examines the lead-lag relation in daily equity returns and CDS spread changes using
unrestricted vector autoregression (VAR) analysis of order two. This approach jointly estimates the
strength of associations within each series and across the two series in both directions at lags 1 and 2.22
Panel A reports the coefficients from this analysis applied to the full sample period. The clearest result
is the consistency of negative coefficients for CHCDS-RET for lag 1. This represents the φ3 coefficient
in Eq. (4). We expect this to be negative consistent with Tables 4 and 5 showing that RETt-1 leads
CHCDSt. Similar to Table 5, Table 6 also shows an overall positive coefficient for CHCDS-CHCDS
(the relation between lagged and contemporaneous CHCDS) at lags 1 and 2, and positive and negative
coefficients, respectively, for RET-RET (the relation between lagged and contemporaneous RET) at
lags 1 and 2. In untabulated analyses, we also find predictably different results for the pre- and post-
scandal periods analyzed separately as VAR regressions. For example, consistent with H2, the sample
mean (median) φ3 coefficient is -0.2862 (-0.1907) for the pre-scandal period and -0.0844 (-0.0794) for
the post-scandal period (or an increase of 0.2018 in the mean and 0.1113 in the median). Moreover,
consistent with H3, the largest change in the φ3 coefficient from pre- to post-scandal is for VW
specifically (0.5353) and for the German automakers as a group (0.3674). In short, the VAR results in
Table 6 mirror the regression results in Tables 4 and 5. Both suggest that the location of informed
trading resides in the equity market (Table 4), and both suggest greater equality of informedness of
CDS and equity investors in the post-VW-emission-cheating-scandal period (Table 5).
[insert Table 6 here]
We also test for Granger causality in the VAR system of equations. Specifically, we test whether
after for controlling for the lagged values of the dependent variables in each equation, the lagged
values of the independent variables taken together further explain the dependent variable. If those
lagged values exhibit statistical significance, we conclude that the lagged variables Granger-cause the
dependent variable. Panel B of Table 6 summarizes the results (applied to the full sample period),
22 Table 6 states the VAR system of equations as Eq. (4) and Eq. (5).
22
which reports a test of the null hypothesis that the lagged values of CHCDS in Eq. (1) or the lagged
values of RET in Eq.(2) are jointly zero. The results, which show more significant coefficients for RET
(Eq. (4), column 4) than CHCDS (column 7), indicate that RET Granger-causes CHCDS, but that
CHCDS does not Granger-cause RET. This result also suggests that the location of informed trading
resides in the equity market. In untabulated analysis, we also find strong support for the view that RET
Granger-causes CHCDS in the pre-scandal period analyzed. Additionally, consistent with greater
market co-integration in the post-scandal period (H2), we cannot reject the hypothesis that RET does
not Granger-cause CHCDS in the pre-scandal period for 3 out of the 16 firms (versus 11 out of 16 in
the post-scandal period), and we cannot reject the hypothesis that CHCDS does not Granger-cause
RET in the pre-scandal period for 9 out of the 16 firms (versus 15 out of 16 in the post-scandal period).
4.6. Trading rule analysis
Panel A of Table 7 shows the additional CDS spread and the change in CDS spread in basis
points (bps) from before to after first news of the VW emission cheating scandal from day 0 to day t (t
= 1 to 5) based on the following hypothetical trading rule. If RETt-1 or RETt-2 < 0 for a firm, then short
the same firm’s five-year CDS securities at the end of t-1 and accrue the change in CDS spread
(defined as CDSt - CDS0) from day 0 to day t, where t = 1 to 5, otherwise do nothing. Because we
predict an increase in spread given prior equity returns, we view this as a proxy for a trading rule that
exploits lagged informed trading by CDS investors. The third and fourth rows of each panel show the
change in spread from the pre- to the post-scandal period, that is, (CDSt - CDS0)pre - (CDSt - CDS0)post),
for each automaker and the mean and median across the sample. Given an increase in market co-
integration, we expect a decrease in the profitability of CDS trading based on prior equity returns.
Table 7 shows these decreases as positive numbers. We first observe that trading profits decrease from
the pre- to the post-scandal period generally for all firms for all holding periods, and the decreases
increase in the number of days in the holding period. This suggests that the ability of lagged equity
return extends beyond the one- to two-day lag shown in the earlier tables. On a daily basis, however,
the decreases in CDSt - CDS0 from before to after the scandal are economically small. For example, the
23
mean decrease in spread from pre- to post-scandal for the five-day trading rule period is 2.39 bps.
However, if judged on an annualized basis, the overall profitability could increase substantially
depending on the number of repetitions per year (e.g., the number of trading days) and the in-and-out
costs of execution.
[insert Table 7 here]
We also partition the trading rule profits on operating region and summarize the results in Panel B
of Table 7. This panel shows results consistent with the regressions. The decrease in spread, which we
contend stems from greater market co-integration from the scandal, is predictably greater for
automakers in those regions directly affected by the scandal and predictably lower in regions not
subject to allegations of automakers’ involvement. For example, for the five-day trading rule, German
automakers’ mean spread decreases by 5.51 bps, whereas US and Asian companies exhibit small
changes, of -1.51 bps and 0.90 bps, respectively. For convenience, Figure 5 shows the same averages
as in Table 7. Rather than show changes in raw spread, the figure shows the changes in percentage
terms (by multiplying the relative changes in Table 7 by 100). As with Table 7, this plot shows a
general monotonic decline in the reduction of trading profits across regions as a result of the scandal,
with the German automakers being most affected.
[insert Figure 5 here]
5. Conclusion
We investigate the information aggregation and market price response of the world’s largest
automakers to the VW emission cheating scandal of 2015–2016. Our examination produces two
important findings of interest to financial economists, investors, and regulators, especially those
interested in the economics of the automobile industry. We first examine and contend that the events
and market behavior leading up to the VW emission cheating scandal closely resemble an
informational cascade unblocked by the EPA notice. This qualitative analysis complements a mostly
theoretical literature that shows few examples of informational cascades in practice, which is
surprising given the conclusions from models and experiments that cascades exist as important
24
economic phenomena that disrupt markets. We also confirm our contention of an informational
cascade based on an event study of key pre-scandal events that could have revealed information about
a possible cheating scandal well before the EPA notice but for the presence of an informational
cascade. Equity prices showed no significant negative reaction to these events.
Second, we test for the effects of an informational cascade by analyzing the evolution of equity
prices and CDS spreads in two markets that naturally integrate through their use of common
information. Based on measures of market co-integration applied in the prior literature, we predict and
find that the accelerated information aggregation from the scandal produces a “game-changing”
increase in the contemporaneous association between equity returns and CDS spread changes in the
post-scandal period, that is, after the EPA notice of violation on September 18, 2015. This association,
moreover, varies predictably with global automakers’ actual or potential involvement in the scandal.
Additionally, we predict and find that the profitability of a trading rule based on the assumption of
more informed trading in the equity market decreases in the post-scandal era, as the information shock
arguably creates less imbalance in the relative informedness of equity traders relative to CDS traders.
Lastly, our study adds new results to a nascent literature on the impact of crises on securities
markets’ co-integration, by showing that firm-related scandals, like financial or banking crises, can
disrupt markets. These results further imply that the putative informational cascade dislodged by the
EPA order may have increased the informational efficiency of automakers’ securities markets, at least
regarding the pricing of credit through CDS trading. We leave unexplored, however, whether this
putative change in market efficiency from the VW scandal contributed to the social good (Zingales
2015), for example, because of automakers’ increased information transparency or regulators’
increased enforcement efficacy.
25
References Andrei, D., and M. Hasler. 2015. Investor attention and stock market volatility. Review of Financial
Studies 28(1): 33-72. Avino, D., E. Lazar, and S. Varotto. 2013. Price discovery of credit spreads in tranquil and crisis
periods. International Review of Financial Analysis 30: 242-253. Banerjee, A. V. 1992. A simple model of herd behavior. The Quarterly Journal of Economics 107(3):
797-817. Berndt, A., and I. Obreja. 2010. Decomposing European CDS returns. Review of Finance 14(2): 189-
233. Berndt, A., and A. Ostrovnaya. 2014. Do Equity Markets Favor Credit Market News Over Options
Market News? Quarterly Journal of Finance 04(02): 1450006. Bernile, G., J. Sulaeman, and Q. Wang. 2015. Institutional trading during a wave of corporate
scandals: “Perfect Payday”? Journal of Corporate Finance 34: 191-209. Bikhchandani, S., D. Hirshleifer, and I. Welch. 1992. A theory of fads, fashion, custom, and cultural
change as informational cascades. Journal of Political Economy: 992-1026. ———. 1998. Learning from the behavior of others: Conformity, fads, and informational cascades.
The Journal of Economic Perspectives 12(3): 151-170. Bizjak, J., M. Lemmon, and R. Whitby. 2009. Option Backdating and Board Interlocks. Review of
Financial Studies 22(11): 4821-4847. Blanco, R., S. Brennan, and I. W. Marsh. 2005. An empirical analysis of the dynamic relation between
investment-grade bonds and credit default swaps. The Journal of Finance 60(5): 2255-2281. da Silva, P., P. Rebelo, and C. Afonso. 2014. Tail dependence of financial stocks and CDS markets:
Evidence using copula methods and simulation-based inference. Economics: The Open-Access, Open Assessment E-Journal 8(39): 1-20.
Forte, S., and L. Lovreta. 2015. Time-Varying Credit Risk Discovery in the Stock and CDS Markets: Evidence from Quiet and Crisis Times. European Financial Management 21(3): 430-461.
Grenadier, S. R. 1999. Information revelation through option exercise. Review of Financial Studies 12(1): 95-129.
Griffin, P. A., H. A. Hong, and J.-B. Kim. 2016. Price discovery in the CDS market: The informational role of equity short interest. Review of Accounting Studies. Forthcoming.
Heron, R. A., and E. Lie. 2007. Does backdating explain the stock price pattern around executive stock option grants? Journal of Financial Economics 83(2): 271-295.
Hilscher, J., J. M. Pollet, and M. Wilson. 2015. Are credit default swaps a sideshow? Evidence that information flows from equity to CDS markets. Journal of Financial and Quantitative Analysis 50(03): 543-567.
Hirshleifer, D., and S. Hong Teoh. 2003. Herd behaviour and cascading in capital markets: A review and synthesis. European Financial Management 9(1): 25-66.
Jegadeesh, N., and S. Titman. 1993. Returns to buying winners and selling losers: Implications for stock market efficiency. The Journal of Finance 48(1): 65-91.
———. 2001. Profitability of momentum strategies: An evaluation of alternative explanations. The Journal of Finance 56(2): 699-720.
Kwan, S. H. 1996. Firm-specific information and the correlation between individual stocks and bonds. Journal of Financial Economics 40(1): 63-80.
Lee, I. H. 1998. Market crashes and informational avalanches. The Review of Economic Studies 65(4): 741-759.
Longstaff, F. A., S. Mithal, and E. Neis. 2005. Corporate yield spreads: Default risk or liquidity? New evidence from the credit default swap market. The Journal of Finance 60(5): 2213-2253.
Merton, R. C. 1974. On the pricing of corporate debt: The risk structure of interest rates. The Journal of Finance 29(2): 449-470.
Narayan, P. K. 2015. An analysis of sectoral equity and CDS spreads. Journal of International Financial Markets, Institutions and Money 34: 80-93.
Narayan, P. K., S. S. Sharma, and K. S. Thuraisamy. 2014. An analysis of price discovery from panel
26
data models of CDS and equity returns. Journal of Banking & Finance 41: 167-177. Norden, L., and M. Weber. 2004. Informational efficiency of credit default swap and stock markets:
The impact of credit rating announcements. Journal of Banking & Finance 28(11): 2813-2843. ———. 2009. The co-movement of credit default swap, bond and stock markets: An empirical
analysis. European Financial Management 15(3): 529-562. Scharfstein, D. S., and J. C. Stein. 1990. Herd behavior and investment. The American Economic
Review: 465-479. Schwert, G. W. 2002. Tests for unit roots: A Monte Carlo investigation. Journal of Business &
Economic Statistics 20(1): 5-17. Trutwein, P., and D. Schiereck. 2011. The fast and the furious: Stock returns and CDS of financial
institutions under stress. Journal of International Financial Markets, Institutions and Money 21(2): 157-175.
Wang, P., and T. Moore. 2012. The integration of the credit default swap markets during the US subprime crisis: Dynamic correlation analysis. Journal of International Financial Markets, Institutions and Money 22(1): 1-15.
Welch, I. 1992. Sequential sales, learning, and cascades. The Journal of Finance 47(2): 695-732. Wermers, R. 1999. Mutual fund herding and the impact on stock prices. The Journal of Finance 54(2):
581-622. Zhu, H. 2006. An empirical comparison of credit spreads between the bond market and the credit
default swap market. Journal of Financial Services Research 29(3): 211-235. Zingales, L. 2015. Does finance benefit society?: National Bureau of Economic Research.
27
Table 1 Descriptive statistics of market capitalization, CDS spread, and equity return from September 18 to September 30, 2015 Panel A: Market Capitalization (billions)
Currency 18-Sep-15 21-Sep-15 22-Sep-15 23-Sep-15 24-Sep-15 25-Sep-15 Mean (all obs.) Volkswagen EURO 76,896 64,172 54,672 58,565 57,060 55,318 43,847 BMW EURO 55,450 54,602 51,316 51,620 48,961 51,038 31,776 Mercedes EURO 77,325 77,083 72,296 74,356 68,781 71,023 49,028 Porsche EURO 18,620 15,414 12,760 12,998 12,919 12,511 15,772 Fiat EURO 16,447 16,493 16,032 15,756 15,721 15,583 10,855 Peugeot EURO 12,409 12,095 11,035 10,753 10,358 10,430 8,769 Renault EURO 21,694 21,005 19,510 19,056 18,713 19,120 14,890 Ford USD 57,672 57,834 56,219 55,249 55,128 54,643 35,532 GM USD 48,269 48,443 47,509 47,019 46,576 46,528 15,544 Honda YEN 6,846,884 6,884,203 6,752,490 6,754,685 6,739,318 6,704,195 5,542,535 Hyundai KRW 34,269,589 32,923,654 33,958,989 32,509,520 32,716,587 33,544,855 22,312,616 Kia KRW 21,028,503 20,627,196 21,269,288 20,546,935 20,546,935 20,667,327 10,959,466 Mazda YEN 1,177,705 1,160,923 1,079,455 1,082,225 1,034,321 1,070,819 644,351 Nissan YEN 4,779,115 4,789,310 4,679,709 4,692,453 4,697,551 4,705,198 3,908,828 Suzuki YEN 1,702,413 1,702,413 1,702,413 1,702,413 1,632,040 1,665,131 1,124,604 Toyota YEN 22,980,783 23,098,762 22,740,956 22,727,418 22,735,154 22,920,826 15,425,453 Panel B: CDS spread (bps) (in local currency)
18-Sep-15 21-Sep-15 22-Sep-15 23-Sep-15 24-Sep-15 25-Sep-15 Mean (all obs.)
Volkswagen 5 YEAR 75.3 134.0 215.9 214.1 221.0 228.6 90.3 BMW 5 YEAR 62.6 78.3 109.9 109.6 133.9 119.8 76.2 Mercedes 5 YEAR 54.8 69.4 100.3 95.7 115.0 104.7 413.2 Porsche 5 YEAR 50.9 60.3 67.3 70.3 94.8 109.8 111.6 Fiat 5 YEAR 278.6 294.3 344.5 350.8 373.6 373.6 755.6 Peugeot 5 YEAR 231.5 251.0 294.3 316.3 355.4 365.1 248.6 Renault 5 YEAR 129.2 143.0 166.0 166.8 186.2 187.6 178.7 Ford 5 YEAR 118.1 120.9 131.0 132.6 138.6 136.8 237.0 GM 5 YEAR 154.8 156.3 172.5 172.5 175.9 177.5 44.2 Honda 5 YEAR 22.3 23.8 24.5 24.7 28.8 29.4 133.9 Hyundai 5 YEAR 86.7 92.4 89.6 94.6 95.8 96.4 164.3 Kia 5 YEAR 101.4 108.0 111.0 112.1 113.6 114.3 206.5 Mazda 5 YEAR 46.9 50.0 51.4 52.0 52.7 53.0 92.6 Nissan 5 YEAR 29.3 31.2 32.1 32.5 34.2 35.4 81.1 Suzuki 5 YEAR 28.6 30.5 31.3 31.6 32.0 32.2 68.3 Toyota 5 YEAR 19.5 19.9 20.2 20.1 23.4 24.3 59.8 Average 5 YEAR 93.1 103.9 122.6 124.8 135.9 136.8 185.1 Panel C: Daily equity return
18-Sep-15 21-Sep-15 22-Sep-15 23-Sep-15 24-Sep-15 25-Sep-15 Mean (all obs.) Volkswagen
-3.84% -16.55% -14.80% 7.12% -2.57% -3.05% 0.04%
BMW
-2.89% -1.53% -6.02% 0.59% -5.15% 4.24% 0.06% Mercedes
-3.62% -0.31% -6.21% 2.85% -7.50% 3.26% 0.06%
Porsche
-3.12% -17.22% -17.22% 1.86% -0.60% -3.16% 0.08% Fiat
-2.19% 0.28% -2.79% -1.72% -0.22% -0.88% 0.05%
Peugeot
-3.05% -2.54% -8.76% -2.56% -3.67% 0.69% 0.05% Renault
-4.22% -3.18% -7.12% -2.33% -1.80% 2.18% 0.05%
Ford
-2.19% 0.28% -2.79% -1.72% -0.22% -0.88% 0.01% GM
-2.56% 0.36% -1.93% -1.03% -0.94% -0.10% 0.05%
Honda
-2.23% 0.55% -1.91% 0.03% -0.23% -0.52% 0.08% Hyundai
1.22% -3.93% 3.14% -4.27% 0.64% 2.53% 0.05%
Kia
-0.57% -1.91% 3.11% -3.40% 0.00% 0.59% 0.03% Mazda
-4.26% -1.43% -7.02% 0.26% -4.43% 3.53% 0.02%
Nissan
-2.39% 0.21% -2.29% 0.27% 0.11% 0.16% 0.03% Suzuki
0.31% na na na -4.13% 2.03% 0.05%
Toyota
-2.00% 0.51% -1.55% -0.06% 0.03% 0.82% 0.05% Average -2.35% -2.90% -4.63% -0.26% -1.92% 0.71% 0.05% Cum. average -2.35% -5.25% -9.88% -10.14% -12.06% -11.34% This table shows the market capitalization, five-year CDS spread, and equity return (or equivalent return on the firm’s exchange-traded security if the home country exchange was closed) for the sample of 16 major worldwide automobile manufacturing firms over the days around the initial news on September 18, 2015 (Reuters 12:10pm EDT) of the EPA order stating that VW intentionally violated the US emission standard for certain VW and Audi diesel passenger cars. The column “all obs.” refers to the mean over the full sample period of January 1, 2004 to March 23, 2016. na=not available due to exchange closure.
28
Table 2 Event study of key events prior to and including the September 18, 2015 EPA disclosure relating to the VW cheating scandal Event Date Mean t p 25th Q Med 75th Q Mean t p 25th Q Med 75th Q Panel A: Market-adjusted model excess returns 2013 05 13 DUH Audit .014 1.87 .077 -.002 .013 .032 .040 2.98 .006 .006 .031 .057 2014 05 23 WV Report to VW CEO .002 .38 .705 -.019 .002 .009 .005 .93 .213 -.017 .002 .020 2014 05 30 WV Report Released -.001 -.48 .640 -.008 -.002 .002 -.004 -.82 .934 -.019 -.008 .009 2014 11 14 VW Memo on Issue .012 2.55 .020 -.001 .008 .023 .010 1.44 .115 -.010 .009 .020 2014 12 02 VW Recall .002 .34 .741 -.011 .001 .016 .007 .94 .114 -.028 .016 .032 2015 09 03 VW Admission .006 1.30 .211 -.007 .004 .011 .004 .49 .774 -.021 -.002 .025 2015 09 18 EPA Disclosure -.035 -2.96 .008 -.032 -.021 -.008 -.045 -2.12 .069 -.061 -.011 .018 Panel B: Market model excess returns 2013 05 13 DUH Audit .017 2.15 .046 -.005 .017 .029 .043 3.09 .006 .009 .028 .060 2014 05 23 WV Report to VW CEO .003 .79 .437 -.006 .000 .005 .005 1.29 .213 -.004 .002 .017 2014 05 30 WV Report Released .000 .01 .989 -.010 -.002 .004 .000 -.08 .934 -.017 -.001 .017 2014 11 14 VW Memo on Issue .010 1.98 .064 -.004 .003 .027 .012 1.66 .115 -.010 .010 .020 2014 12 02 VW Recall .005 .71 .489 -.012 .003 .017 .012 1.66 .114 -.020 .022 .033 2015 09 03 VW Admission .003 .69 .499 -.009 .003 .009 -.003 -.29 .774 -.028 -.009 .015 2015 09 18 EPA Disclosure -.032 -2.83 .011 -.0339 -.021 -.002 -.040 -1.93 .069 -.067 -.009 .019 This table reports mean and median cumulative excess returns (day -1 to 1) and (day -2 to 2) for six key event dates starting on May 13, 2013 and ending on September 18, 2015 when the EPA made public the emission cheating scandal. Expected returns are determined using a market-adjusted model (MAR) and market model (MM). For the market model, we estimate alpha and beta using daily data. The calculation estimates the parameters using up to 255 trade days (not within 30 days of the event date).
29
Table 3 Stationarity and autocorrelation of firm-level time series Panel A: Dickey-Fuller and Phillips-Perron tests of stationarity
CDS Equity
Critical value p<.05
Phillips–Perron test CDS Equity
Critical value p<.05
VW -10.818 -86.943 -3.960 *** VW -44.070 -79.617 -3.960 *** BMW -11.230 -18.162 -3.960 *** BMW -42.712 -98.934 -3.960 *** Mercedes -12.813 -13.301 -3.960 *** Mercedes -27.059 -84.663 -3.960 *** Porsche -7.533 -16.784 -3.960 *** Porsche -54.051 -81.151 -3.960 *** Fiat -8.305 -20.628 -3.960 *** Fiat -45.499 -59.845 -3.960 *** Peugeot -11.342 -19.111 -3.960 *** Peugeot -45.420 -88.581 -3.960 *** Renault -10.333 -14.426 -3.960 *** Renault -47.217 -90.161 -3.960 *** Ford -13.629 -21.602 -3.960 *** Ford -45.845 -69.364 -3.960 *** GM -7.342 -8.585 -3.960 *** GM -45.850 -89.485 -3.960 *** Honda -12.333 -21.832 -3.960 *** Honda -53.566 -87.213 -3.960 *** Hyundai -27.366 -84.030 -3.960 *** Hyundai -57.001 -87.607 -3.960 *** Kia -11.806 -80.677 -3.960 *** Kia -33.184 -31.950 -3.960 *** Mazda -11.146 -21.788 -3.960 *** Mazda -50.685 -87.353 -3.960 *** Nissan -12.133 -23.993 -3.960 *** Nissan -43.621 -61.924 -3.960 *** Suzuki -9.215 -22.121 -3.960 *** Suzuki -48.428 -62.802 -3.960 *** Toyota -9.104 -21.671 -3.960 *** Toyota -42.109 -90.841 -3.960 *** Panel B: Autocorrelation Firm Security Lags 1 2 3 4 5 6 7 8 VW CDS 0.165 0.021 0.039 -0.013 0.014 0.021 0.050 -0.034 BMW CDS 0.071 0.066 0.021 -0.004 0.021 -0.017 0.029 -0.020 Mercedes CDS 0.092 0.045 0.047 -0.015 0.011 -0.007 0.017 0.002 Porsche CDS 0.013 0.024 0.031 -0.014 -0.001 -0.017 0.055 -0.004 Fiat CDS 0.095 0.053 0.063 0.025 0.020 0.011 0.018 0.008 Peugeot CDS 0.094 0.042 0.044 -0.012 0.040 -0.027 -0.031 -0.016 Renault CDS 0.098 0.048 0.050 -0.010 0.007 -0.011 -0.014 0.002 Ford CDS 0.096 0.019 0.038 0.015 0.020 0.006 0.043 -0.013 GM CDS 0.026 0.013 0.057 -0.012 -0.004 0.023 -0.022 0.014 Honda CDS -0.086 0.000 0.029 -0.010 0.006 0.023 0.003 0.043 Hyundai CDS -0.001 0.000 0.003 0.011 -0.001 -0.002 -0.012 -0.019 Kia CDS 0.001 -0.014 0.007 -0.003 -0.008 -0.019 0.012 0.022 Mazda CDS -0.032 -0.014 0.017 -0.006 0.016 0.029 0.005 0.000 Nissan CDS 0.000 0.050 -0.004 0.033 0.052 0.039 0.014 -0.014 Suzuki CDS -0.093 -0.038 0.031 0.017 -0.021 0.000 0.051 -0.031 Toyota CDS -0.024 0.050 0.009 0.021 0.033 0.012 0.006 0.002 Mean CDS 0.023 0.023 0.030 0.003 0.013 0.003 0.012 -0.002 Median CDS 0.013 0.024 0.031 -0.004 0.011 0.000 0.012 0.000 t-value CDS 0.174 0.009 0.0000 0.542 0.018 0.556 0.084 0.737 VW Equity 0.034 -0.018 0.026 0.012 -0.001 0.000 -0.021 -0.018 BMW Equity 0.037 -0.019 -0.012 -0.015 -0.013 0.005 -0.024 -0.003 Mercedes Equity 0.033 -0.007 -0.022 0.009 -0.003 0.029 -0.062 -0.023 Porsche Equity 0.051 -0.013 0.001 -0.001 -0.018 0.012 -0.040 -0.025 Fiat Equity 0.015 0.007 0.012 0.018 0.008 0.015 0.001 -0.005 Peugeot Equity 0.070 -0.012 -0.006 -0.022 -0.006 -0.006 -0.023 0.006 Renault Equity 0.061 -0.009 -0.025 -0.004 -0.017 0.010 -0.013 0.006 Ford Equity -0.003 0.035 -0.002 0.009 0.013 0.010 -0.018 -0.016 GM Equity 0.018 0.015 -0.026 -0.010 -0.029 0.001 -0.063 -0.020 Honda Equity -0.006 -0.030 -0.018 0.003 -0.013 -0.012 -0.016 -0.012 Hyundai Equity 0.018 -0.012 0.014 -0.040 -0.017 -0.017 -0.044 -0.007 Kia Equity 0.054 -0.005 -0.011 -0.033 -0.003 -0.015 -0.033 -0.033 Mazda Equity -0.019 -0.028 -0.006 -0.007 0.007 -0.005 -0.012 -0.012 Nissan Equity 0.007 -0.031 -0.033 -0.010 -0.009 0.002 -0.015 -0.004 Suzuki Equity -0.041 -0.021 -0.027 -0.015 -0.014 -0.027 0.002 -0.014 Toyota Equity -0.085 -0.027 -0.014 -0.020 0.022 0.008 -0.005 -0.027 Mean Equity 0.014 -0.010 -0.012 -0.009 -0.006 0.001 -0.024 -0.013 Median Equity 0.018 -0.012 -0.012 -0.010 -0.009 0.002 -0.018 -0.012 t-value Equity 0.197 0.036 0.004 0.032 0.095 0.843 0.001 0.001 Panel A reports test statistics for the Augmented Dickey–Fuller test and a Phillips–Perron test with daily lags 1–3. CDS and Equity refer to the respective daily percentage changes. The data are based on equity prices and five-year CDS spreads from 16 major worldwide automobile manufacturing firms over the period from January 1, 2004 to March 23, 2016. Panel B shows the autocorrelation coefficients for the daily changes and a t-test of whether the mean autocorrelation at lag k differs from zero.
30
Table 4 Firm-level time-series regressions of the lead-lag relation between equity returns and changes in CDS spreads: Pre-scandal observations
Lag 0
Lag 1
Lag 2
RET or CHCDS coefficient β4 Coeff. t-stat Signif. β5 Coeff. t-stat Signif. β6 Coeff. t-stat Signif.
Adj. R2
F stat. signif.
No. obs.
Panel A: CHCDSt = α + β1MKTt + β2SMBt + β3HMLt + β4RETt + β5RETt-1 + β6RETt-2 + εt VW -0.2290 -4.61 *** -0.1218 -2.74 *** -0.0117 -0.29 ns 0.123 *** 3,049 BMW -0.5660 -12.41 *** -0.2628 -6.28 *** -0.0403 -1.01 ns 0.172 *** 3,049 Mercedes -0.6088 -12.57 *** -0.2362 -5.8 *** 0.0024 0.06 ns 0.191 *** 3,033 Porsche -0.3089 -5.98 *** -0.1837 -5.71 *** -0.0238 -0.73 ns 0.210 *** 3,049 Fiat -0.4670 -13.48 *** -0.2428 -7.83 *** -0.0374 -1.38 ns 0.243 *** 1,188 Peugeot -0.4461 -14.64 *** -0.1936 -7.29 *** -0.0544 -1.99 ** 0.318 *** 3,022 Renault -0.5476 -16.24 *** -0.2058 -6.64 *** -0.0470 -1.66 * 0.002 *** 3,022 Ford -0.3214 -5.58 *** -0.3391 -8.01 *** -0.0610 -1.91 * 0.024 *** 3,022 GM -0.1345 -2.39 ** -0.1967 -5.27 *** -0.1126 -3 *** 0.318 *** 1,893 Honda -0.4206 -7.47 *** -0.1185 -2.24 ** -0.0972 -1.78 * 0.020 *** 1,941 Hyundai -0.9074 -1.47 ns -1.2021 -1.27 ns 0.7556 1.32 ns 0.053 *** 3,049 Kia 0.0005 0.01 ns 0.0621 0.93 ns -0.0054 -0.22 ns 0.200 *** 3,049 Mazda -0.0745 -1.76 * -0.0434 -1.24 ns 0.0119 0.31 ns 0.108 *** 2,964 Nissan -0.4307 -8.77 *** -0.1428 -3.74 *** -0.1196 -2.55 ** 0.246 *** 3,049 Suzuki -0.2297 -4.43 *** -0.1395 -2.48 ** -0.2210 -2.48 ** 0.025 *** 1,941 Toyota -0.5383 -6.10 *** -0.1573 -1.57 ns -0.1516 -2.62 *** 0.069 *** 1,941 Mean -0.3894
-0.2327
-0.0133
Median -0.4307
-0.1936
-0.0470 Standardized value -4.0037
-1.7765
-0.2189
t-value (mean=0) 0.0010
0.0947
0.8295 binomial prob. 0.0003
0.0003
0.0106
Panel B: RETt = α + β1MKTt + β2SMBt + β3HMLt + β4CHCDSt + β5CHCDSt-1 + β6CHCDSt-2 + εt VW -0.0854 -4.80 *** -0.0005 -0.03 ns 0.0162 0.95 ns 0.126 *** 3,049 BMW -0.1261 -11.43 *** 0.0036 0.39 ns 0.0021 0.21 ns 0.297 *** 3,049 Fiat -0.1493 -11.82 *** 0.0013 0.13 ns 0.0151 1.27 ns 0.312 *** 3,033 Ford -0.1299 -6.02 *** -0.0250 -1.74 * 0.0253 1.52 ns 0.214 *** 3,049 GM -0.2126 -12.48 *** 0.0002 0.02 ns -0.0055 -0.40 ns 0.270 *** 1,188 Honda -0.2050 -12.60 *** 0.0097 0.79 ns 0.0078 0.61 ns 0.267 *** 3,022 Hyundai -0.2218 -12.78 *** -0.0231 -1.82 * 0.0160 1.26 ns 0.362 *** 3,022 Kia -0.1359 -5.54 *** 0.0102 0.45 ns -0.0321 -1.61 ns 0.364 *** 3,022 Mazda -0.0593 -2.42 ** -0.0236 -1.49 ns 0.0062 0.41 ns 0.464 *** 1,893 Mercedes -0.0681 -7.30 *** -0.0050 -0.56 ns -0.0033 -0.30 ns 0.043 *** 1,941 Nissan -0.0006 -2.49 ** -0.0011 -1.24 ns 0.0005 1.26 ns 0.025 *** 3,049 Peugeot 0.0001 0.00 ns -0.0113 -0.77 ns 0.0149 0.78 ns 0.031 *** 3,049 Porsche -0.0361 -1.67 * -0.0723 -3.89 *** -0.0420 -1.95 * 0.038 *** 2,964 Renault -0.1017 -9.41 *** -0.0086 -0.86 ns 0.0100 0.88 ns 0.051 *** 3,049 Suzuki -0.0417 -4.13 *** -0.0006 -0.05 ns -0.0150 -1.22 ns 0.038 *** 1,941 Toyota -0.1023 -6.22 *** -0.0306 -2.32 ** 0.0390 2.50 ** 0.082 *** 1,941 Mean -0.1047 -0.0110 0.0034 Median -0.1023 -0.0050 0.0062 Standardized value -4.8697 -0.1021 3.1875 t-value 0.0002 0.9200 0.0057 binomial prob. 0.0003 0.1051 0.9616 This table shows the β4 to β6 regression coefficients for the time-series regression of CHCDSt = α + β1MKTt + β2SMBt + β3HMLt + β4RETt + β5RETt-1 + β6RETt-2 + εt (Eq. (1), Panel A) and RETt = α + β1MKTt + β2SMBt + β3HMLt + β4CHCDSt + β5CHCDSt-1 + β6CHCDSt-2 + εt (Eq. (2), Panel B) over t observations from January 1, 2004 to September 17, 2015, where RETt-k = equity return on day t-k for k = 0, 1, and 2, CHCDSt-k = percentage change in five-year CDS spread on day t-k for k = 0, 1, and 2, MKT = return on market index, SMB and HML = Fama-French size and book-to-market ratio factors, respectively, and εt = uncorrelated error, all on day t. The t-tests are based on Huber-White robust standard errors. Binomial prob. relates to the number of positive coefficients.
31
Table 5 Firm-level time-series regression of change in CDS spread on equity return, lagged daily equity return, and control variables before and after first news of the VW emission cheating scandal
Ret Lag 0 Ret Lag 1 Ret Lag 2
Coeff. Pre
Coeff. Post
Coeff. Post-Pre
Sig.
Coeff. Pre
Coeff. Post
Coeff. Post-Pre
Sig.
Coeff. Pre
Coeff. Post
Coeff. Post-Pre
Sig.
VW -0.229 -1.905 -1.676 *** -0.122 -0.614 -0.492 ** -0.012 -0.132 -0.120 ns BMW -0.566 -1.652 -1.086 ** -0.263 -0.366 -0.103 ns -0.040 -0.303 -0.263 ns Mercedes -0.609 -1.799 -1.191 * -0.236 -0.588 -0.352 ns 0.002 -0.380 -0.382 ns Porsche -0.309 -0.438 -0.129 *** -0.184 0.277 0.461 ns -0.024 -0.379 -0.355 ns Fiat -0.467 -0.825 -0.358 ns -0.243 -0.296 -0.053 ns -0.037 -0.112 -0.075 ns Peugeot -0.446 -1.001 -0.555 ns -0.194 -0.477 -0.283 ns -0.054 -0.142 -0.088 ns Renault -0.548 -1.216 -0.668 ns -0.206 -0.111 0.095 ns -0.047 -0.258 -0.211 ns Ford -0.321 -0.162 0.160 ns -0.339 -0.215 0.124 ns -0.061 -0.294 -0.233 ns GM -0.135 -0.193 -0.059 ns -0.197 0.079 0.275 ns -0.113 -0.335 -0.223 ns Honda -0.421 -0.724 -0.303 ** -0.118 -0.314 -0.195 ns -0.097 -0.124 -0.027 ns Hyundai -0.907 -0.240 0.668 ** -1.202 -0.176 1.026 ns 0.756 0.186 -0.569 ns Kia 0.001 0.005 0.005 *** 0.062 -0.076 -0.138 *** -0.005 0.074 0.079 ns Mazda -0.074 -0.105 -0.030 ns -0.043 0.126 0.169 ns 0.012 -0.044 -0.056 ns Nissan -0.431 -0.934 -0.503 * -0.143 -0.187 -0.045 ns -0.120 -0.056 0.064 ns Suzuki -0.230 -0.230 0.000 ns -0.139 0.011 0.151 ns -0.221 -0.176 0.045 ns Toyota -0.538 -1.334 -0.796 *** -0.157 -0.198 -0.041 ns -0.152 0.272 0.423 ns Mean -0.389 -0.797 -0.408 *** -0.233 -0.195 0.038 ns -0.013 -0.138 -0.124 ns Mean: German -0.428 -1.448 -1.020 -0.201 -0.322 -0.121 -0.018 -0.298 -0.280 Mean: Other EU -0.487 -1.014 -0.527 -0.214 -0.294 -0.080 -0.046 -0.171 -0.125 Mean: US -0.228 -0.177 0.051 -0.268 -0.068 0.200 -0.087 -0.315 -0.228 Mean: Asian -0.372 -0.509 -0.137 -0.249 -0.116 0.132 0.025 0.019 -0.006 Median -0.426 -0.775 -0.331 -0.189 -0.193 -0.043 -0.044 -0.137 -0.104 Standardized mean -6.807 -4.980 -2.796 -3.394 -3.140 0.420 -0.249 -2.858 -2.170 t-test prob./sig. 0.000 0.000 0.013 ** 0.004 0.006 0.680 ns 0.807 0.011 0.045 *** # positive 1.000 1.000 3.000 1.000 4.000 7.000 3.000 3.000 4.000 binomial prob. 0.000 0.000 0.011 0.000 0.038 0.402 0.011 0.011 0.038 This table shows the regression coefficients for RETt, RETt-1, and RETt-2 from the time-series regression of CHCDSt = α + β1MKTt + β2SMBt + β3HMLt + β4RETt + β5RETt-1 + β6RETt-2 + εt over t observations from January 1, 2004 to September 17, 2015 (pre-scandal) and September 18. 2015 to March 16, 2016 (post-scandal), where CHCDS = percentage change in the five-year CDS spread, MKT = return on market index, SMB and HML = Fama-French size and book-to-market ratio factors, respectively, and εt = uncorrelated error, all on day t, and where RETt-k = equity return on day t-k for k = 0, 1, and 2. Sig. tests whether the post-scandal RET coefficient differs from the pre-scandal RET coefficient assuming robust standard errors (adjusted for time-series heteroskedasticity), where *** <.01, **<.05, *<.10. t-value tests whether the VW coefficient differs from the coefficient of the other 15 firms. binomial prob. tests the cumulative probability of observing at least the number of positive coefficients given an underlying probability of 0.5. The t-tests are based on Huber-White robust standard errors.
32
Table 6 Firm-level vector autoregressions of order two for equity return and percentage change in CDS and Granger causality tests before first news of the VW emission cheating scandal Panel A: Vector autoregression
Lag 1 Lag 2
RET-RET β1
RET-CHCDS β3
CHCDS-CHCDS φ1
CHCDS-RETφ3 RET-RET β2
RET-CHCDS β4
CHCDS-CHCDS φ2
CHCDS-RET φ4
VW -0.0368 0.0029 0.1618 -0.1024 -0.0107 0.0204 0.0399 0.0432 BMW 0.0557 0.0059 0.0620 -0.2045 -0.0516 -0.0247 0.0978 0.0691 Fiat 0.0846 0.0037 -0.0940 -0.1141 -0.0951 -0.0114 -0.0162 -0.0318 Ford -0.0157 -0.1907 0.0522 0.2172 0.0925 0.2397 -0.0174 -0.1250 GM 0.0453 -0.0110 0.1335 -0.1710 -0.0162 0.0040 0.0666 0.1216 Honda -0.1765 -0.1776 0.3610 -0.0134 -0.0677 -0.1196 -0.1310 0.0106 Hyundai 0.0091 -0.0026 0.0023 0.0097 0.1210 0.1748 0.0123 0.3024 Kia 0.1020 -0.0087 0.0236 -0.2336 -0.0603 -0.0089 0.0416 0.0371 Mazda -0.0189 -0.0397 0.1131 -0.1865 0.0371 -0.0026 0.0648 0.0300 Mercedes 0.1141 0.0042 0.0638 -0.2070 -0.0430 -0.0215 0.0472 0.0052 Nissan 0.0762 -0.0423 0.0654 -0.2158 -0.0475 -0.0208 0.0645 0.0132 Peugeot 0.0271 0.0279 0.0696 -0.3260 0.1289 0.0049 0.0023 -0.0575 Porsche 0.0685 -0.0223 0.0174 -0.1687 -0.0100 -0.0228 -0.0450 -0.1293 Renault 0.0213 0.0012 -0.0035 -1.9097 -0.0409 -0.0081 0.0284 1.3348 Suzuki -0.1339 -0.2628 0.2011 0.0385 0.0010 0.1511 -0.0301 -0.2017 Toyota -0.0004 0.0527 -0.2106 -0.1236 -0.0933 0.0317 0.0772 0.9007 Mean 0.0139 -0.0412 0.0637 -0.2319 -0.0097 0.0241 0.0189 0.1452 Median 0.0242 -0.0057 0.0629 -0.1699 -0.0286 -0.0053 0.0341 0.0216 Standard deviation 0.0199 0.0221 0.0314 0.1165 0.0176 0.0223 0.0144 0.1010 Mean: German 0.0219 -0.0446 0.0455 -0.0509 -0.0162 0.0560 0.0260 -0.0111 Mean: Other EU -0.0407 -0.0637 0.1656 -0.0582 0.0123 0.0197 -0.0174 0.1448 Mean: US 0.0415 -0.0242 0.0683 -0.2101 -0.0116 -0.0057 0.0532 0.0336 Mean: Japanese 0.0247 -0.0345 0.0290 -0.4161 -0.0150 0.0163 0.0206 0.2665 Panel B: Granger causality tests
Eq. (3) Chi-square
Prob. (Chi-square) Eq. (4) Chi-square
Prob. (Chi-square)
VW 8.383 0.015 ** 2.491 0.288 ns BMW 21.084 0.000 *** 4.871 0.088 * Fiat 21.711 0.000 *** 0.914 0.633 ns Ford 43.781 0.000 *** 0.605 0.739 ns GM 33.629 0.000 *** 5.421 0.067 * Honda 37.223 0.000 *** 1.802 0.406 ns Hyundai 34.878 0.000 *** 8.529 0.014 ** Kia 108.200 0.000 *** 3.670 0.160 ns Mazda 22.422 0.000 *** 1.971 0.373 ns Mercedes 4.650 0.098 * 1.591 0.451 ns Nissan 3.820 0.148 ns 3.870 0.144 ns Peugeot 0.775 0.679 ns 5.603 0.061 * Porsche 1.528 0.466 ns 19.550 0.000 *** Renault 18.016 0.000 *** 0.991 0.609 ns Suzuki 20.469 0.000 *** 5.072 0.079 * Toyota 18.984 0.000 *** 6.151 0.046 ** Mean 24.972 0.088
4.569 0.260
Median 21 0
4 0 Number of ns 13
7
Panel A shows the results of a vector autoregression regression of order two for the time series of CHCDSt and RETt over t observations from January 1, 2004 to September 17, 2015 (pre-scandal). The specific model is RETt = β0 + β1RETt-1 + β2RETt-2 + β3CHCDSt-1 + β4CHCDSt-2 + β5MKTt + β6SMBt + β7HMLt + εt (Eq. (3)) and CHCDSt = φ0 + φ1CHCDSt-1 + φ2CHCDSt-2 + φ3RETt-1 + φ4RETt-2 + φ5MKTt + φ6SMBt + φ7HMLt + γt (Eq. (4)), where CHCDSt = percentage change in the five-year CDS spread on day t, RETt = equity return on day t, and εt and γt = uncorrelated error. Panel B show the results of a test of Granger causality for the VAR model in Panel A; that is, a test of the null hypothesis that the lagged values of CHCDS in Eq. (3) or the lagged values of RET in Eq.(4) are jointly zero.
33
Table 7 The ability of lagged equity return to predict change in CDS spread: Hypothetical trading rule tests before and after first news of the VW emission cheating scandal Panel A: Additional CDS spread and the change in CDS spread from trading rule by automaker Abbreviated automaker name VW MBZ BMW POR FIA PEU REN FOR GM HON HYU KIA MAZ NIS SUZ TOY Mean Median One day trading rule Pre 4.77 1.59 1.17 1.12 2.21 3.27 2.02 1.15 1.23 0.87 -0.59 -0.02 0.60 1.62 -1.57 0.18 1.23 1.16 Post 0.08 0.11 -0.01 0.10 -0.02 0.30 -0.12 0.35 0.20 0.49 0.45 0.20 -0.12 0.10 0.41 0.36 0.18 0.15 Change in spread 4.70 1.49 1.18 1.01 2.22 2.98 2.14 0.80 1.03 0.37 -1.04 -0.22 0.72 1.52 -1.98 -0.18 1.05 1.02 Relative change in spread 0.98 0.93 1.01 0.91 1.01 0.91 1.06 0.70 0.84 0.43 na na 1.20 0.94 na -1.02 0.76 0.93 Two-day trading rule Pre 5.64 1.80 2.26 1.61 3.40 4.69 2.02 0.48 1.36 1.60 -0.59 -0.04 -0.03 2.16 -1.57 0.79 1.60 1.61 Post -0.11 -0.06 -0.23 -0.14 -0.46 0.67 -0.12 0.85 0.23 0.60 0.97 -0.08 -0.26 0.30 0.41 0.53 0.19 0.09 Change in spread 5.75 1.86 2.49 1.74 3.86 4.02 2.14 -0.37 1.13 1.00 -1.56 0.04 0.23 1.87 -1.98 0.26 1.41 1.44 Relative change in spread 1.02 1.03 1.10 1.08 1.13 0.86 1.06 -0.79 0.83 0.63 na na na 0.86 na 0.33 0.76 0.94 Three-day trading rule Pre 6.07 2.18 3.05 1.81 4.29 5.54 2.02 -0.09 1.48 3.07 -0.17 0.82 -0.06 1.80 -1.57 0.86 1.94 1.81 Post -0.39 -0.20 -0.42 0.19 -0.20 0.92 -0.12 1.19 0.41 0.67 1.59 0.03 -0.33 0.33 0.41 0.73 0.30 0.26 Change in spread 6.47 2.38 3.47 1.61 4.49 4.63 2.14 -1.28 1.08 2.40 -1.76 0.79 0.27 1.47 -1.98 0.13 1.64 1.54 Relative change in spread 1.06 1.09 1.14 0.89 1.05 0.83 1.06 na 0.73 0.78 na 0.96 na 0.82 na 0.15 0.88 0.93 Four-day trading rule Pre 7.06 2.85 3.63 2.35 5.07 5.63 2.02 -0.48 1.29 4.33 0.23 1.95 0.19 2.20 -1.57 1.37 2.38 2.11 Post -0.80 -0.53 -0.79 0.22 -0.31 1.13 -0.12 1.66 0.75 1.16 1.60 0.30 -0.45 0.50 0.41 0.77 0.34 0.35 Change in spread 7.86 3.38 4.42 2.13 5.38 4.50 2.14 -2.15 0.54 3.17 -1.37 1.65 0.64 1.69 -1.98 0.60 2.04 1.91 Relative change in spread 1.11 1.19 1.22 0.91 1.06 0.80 1.06 na 0.42 0.73 -5.92 0.85 3.34 0.77 na 0.44 0.57 0.88 Five-day trading rule Pre 7.91 4.26 5.11 2.48 5.67 5.50 2.02 -0.90 1.16 5.55 0.40 2.62 0.32 3.15 -1.57 1.34 2.81 2.55 Post -1.32 -0.31 -0.90 0.26 -1.13 1.59 -0.12 2.09 1.19 1.19 1.89 0.49 -0.26 0.64 0.41 1.13 0.43 0.45 Change in spread 9.23 4.57 6.01 2.21 6.80 3.91 2.14 -2.99 -0.03 4.36 -1.49 2.13 0.59 2.51 -1.98 0.21 2.39 2.18 Percentage decrease in spread 1.17 1.07 1.18 0.89 1.20 0.71 1.06 na -0.03 0.79 -3.69 0.81 1.82 0.80 na 0.16 0.57 0.85 Panel B: Additional CDS spread and the change in CDS spread from trading rule by region
German Other EU US Asia
One-day holding period Change in spread 2.093 2.446 0.916 -0.115
Relative change in spread 0.957 0.992 0.768 0.389 Two-day holding period
Change in spread 2.962 3.339 0.376 -0.020 Relative change in spread 1.060 1.017 0.022 0.606 Three-day holding period
Change in spread 3.482 3.750 -0.098 0.190 Relative change in spread 1.046 0.980 0.727 0.678 Four-day holding period
Change in spread 4.448 4.005 -0.802 0.631 Relative change in spread 1.106 0.973 0.419 0.035 Five-day holding period
Change in spread 5.506 4.284 -1.510 0.904 Relative change in spread 1.078 0.990 -0.028 0.114 Panel A shows the additional CDS spread (in basis points) and the change in CDS spread (in basis points) from before to after first news of the VW emission cheating scandal from day 0 to day t (t =1 to 5) based on the following hypothetical trading rule: If RETt-1 or RETt-2 < 0, then short five-year CDS securities at the end of t-1 and accrue the change in spread (CHCDS) from day 0 to day t, where t =1 to 5, otherwise do nothing. Panel B summarizes the results in Panel A for the four regions where the automakers primarily operate.
34
Figure 1 Volkswagen AG: Five-year CDS spread (left-hand axis), stock price (right-hand axis), and the number of news stories mentioning "auto emission tests" (right-hand axis) by day from January 2, 2015 to March 26, 2016. Source: ProQuest (news), S&P Capital IQ (securities).
0
50
100
150
200
250
300
350
0
50
100
150
200
250
300
350
1/1/
11
1/15
/11
1/29
/11
2/12
/11
2/26
/11
3/12
/11
3/26
/11
4/9/
11
4/23
/11
5/7/
11
5/21
/11
6/4/
11
6/18
/11
7/2/
11
7/16
/11
7/30
/11
8/13
/11
8/27
/11
9/10
/11
9/24
/11
10/8
/11
10/2
2/11
11
/5/1
1 11
/19/
11
12/3
/11
12/1
7/11
12
/31/
11
1/14
/12
1/28
/12
2/11
/12
2/25
/12
3/10
/12
Stoc
k pr
ice,
Eur
os
5 ye
ar C
DS
spre
ad, b
ps
5 year CDS spread
Stock price
Auto emission tests
35
Figure 2 Volkswagen AG: Equity return from January 1, 1980 to March 26, 2016. Source: S&P Capital IQ (securities).
-30%
-20%
-10%
0%
10%
20%
30%
VWstockreturn
36
Figure 3 Volkswagen AG: Contemporaneous reaction in CDS spread and stock price on the 20 event days with the greatest negative equity return during January 2, 2004 to March 23, 2016.
-40%
-20%
0%
20%
40%
60%
80%
VW % change in CDS spread
VW stock return
37
Figure 4 Mean change in automobile firm time-series regression coefficient by region from before to after disclosure of the VW emission cheating scandal for Ret Lag 0 and Ret Lag 1 from the time-series regressions of CDS spread change on equity return, lagged equity return, and control variables.
-1.20
-1.00
-0.80
-0.60
-0.40
-0.20
0.00
0.20
Mean: German Mean: Other EU Mean: US Mean: Asian
-1.02
-0.53
0.05
-0.13 -0.12 -0.08
0.20 0.13
Cha
nge
in re
gres
sion
coe
ffici
ent a
fter
the
emis
sion
s che
atin
g sc
anda
l
Coeff.: Ret Lag 0
Coeff.: Ret Lag 1
38
Figure 5 Percentage decrease in CDS spread from before to after the VW emission cheating scandal from a trading strategy based on the negative relation between CDS spread and lagged equity return
-20%
0%
20%
40%
60%
80%
100%
120%
German Other EU US Asia
96% 99%
77%
39%
105% 98%
73% 68%
108%
99%
-3% 11%
Perc
enta
ge d
ecre
ase
in C
DS
spre
ad
One day holding period
Three day holding period
Five day holding period