University of Amsterdam, Amsterdam Business School June 2016
The Value of Misconduct and the Likelihood of
Repeating it: The Case of the Auto Industry
Masterthesis
Thesis Supervisor: Dr. Rafael Perez Ribas Faculty of Economics and Business - Section Finance
Last Name, Front Name: Grotz, Franziska
Student ID: 11125349
Address: Mesdagstraat 10
1073HL Amsterdam E-Mail Address: franziska.grotz@gmx.de
Programme: MSc Business Economics, Finance Track Submission Date: June 22th, 2016
Statement of originality
This document is written by Franziska Grotz who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of contents Statement of originality ... 2 Abstract ... 4 1 Introduction ... 5 2 Literature Review ... 7 3 Data ... 10 3.1. BHRRC Data ... 10 3.2. Financial Data ... 12 3.3. Descriptive Statistics ... 15 4 Empirical Method ... 17 5 Results ... 20
5.1. Impact of scandals on firm equity value ... 20
5.2. Probability of repeated scandals ... 26
6 Robustness checks and threats for validity ... 28
7 Conclusion ... 30
References ... 33
Appendix ... 36
Abstract
This research investigates the effect of scandals on firm equity value using a unique dataset of 18,364 stories collected from the publicly available library of the Business & Human Rights Resource Centre. The following study tests whether corporate abuses affect firm value in the short and long-run. The empirical approach is a dynamic difference-in-differences framework to determine short and long-term effects of different types of scandals, such as lawsuits and unsafe products. Findings show that scandals in general decrease firm value in the first year, but that firm value increases again after two years exceeding even the initial loss. Results indicate a reverse short and long-term effect for scandals including unsafe products, which might suggest delayed costs to make up for the damage caused. Finally, findings show that firms that were part of a scandal in the past are more likely to get caught in the future. This accounts mostly for big companies. This relationship may be explained by the greater media attention of large corporations, but also that these firms learned from the past that misconducts in general are actually profitable. Not only criminal proceedings might be too low, but also civil, because shareholders do not consider the ethical behaviour of a company during investment decisions or might forgive fast. These results suggest that misconducts are not leading to a change in corporate behaviour. This should have implications for a modification of the regulatory framework and mentality of shareholders regarding the punishment of corporate scandals.
1 Introduction
The question how to balance social responsibility against the duty to shareholders to maximize profits is a frequently debated question in the financial world. On the one hand, ethical behaviour might be related to several benefits such as high shareholder trust or reduced business and legal risk. Beyond that, apart from financial considerations, it makes a positive impact on the environment (Flammer 2011). On the other hand, implementing ethical standards in a corporate structure is potentially related to increased costs. There are for example opportunity costs a company has to bear because of committing to ethical projects or increased manufacturing costs (Friedmann 1970). And related to that, do firms tend to repeat their mistakes because it pays to behave unethical? By using a unique dataset consisting of 18,364 stories collected from the Business & Human Rights Resource Centre, I give new insights in these questions.
The objective of this thesis is to estimate the effect of getting caught in scandals, such as unsafe products and reputation issues, on firm value and whether companies tend to repeat their mistakes. This is not only interesting for investors, but also for companies implementing structures according to ethical standards. From a more abstract point of view, it sheds light on the ethical attitude in the financial industry in general. If companies that were part of a scandal have higher returns than firms which did not behave wrongly, this would not only show that costs for corporate abuse, such as monetary penalties, are too low, but also that the universe of ethical investors is too small. From a shareholder point of view, corporate abuses might be worthwhile if expected benefits outweigh costs.
My hypothesis is that although firm value might be affected in the short-run, investors do not take scandals into consideration in the long-run. This hypothesis is built on the assumption that companies can take actions to regain investors’ trust and that there is still a large universe of investors neglecting ethical considerations.
One example for this hypothesis is Merck, who was alleged in 2004 that the company knew about the fact that one of its best-selling medicines VIOXX increases the risk of a heart attack and did not reveal this information. The share price dropped sharply after this fact was made public, but the stock trades today again at a pre-scandal level although court proceedings are not even closed yet (Loftus 2016; !Yahoo Finance).
This work focuses on the automobile industry. Customers in this sector are more sensible for issues such as product safety, but also regarding company reputation as cars are not only costly, but also long-term investments. Above that, the automotive sector allows to investigate different effects of corporate misconducts as scandals vary extremely. There are lawsuits regarding fraud, but also scandals that involve fatalities. This research accounts for this by generating different treatment variables. Next to the idiosyncratic characteristics, that make the automotive industry an interesting research subject, it is also in the current interest of the public because of recent affairs such as the Volkswagen emission scandal (BBC News 2016).
The methodological approach taken in this study is a difference-in-differences approach using fixed effects. Companies that were part of a scandal during the investigated period from 1995 until 2015 are the treatment group, while companies who were not in the public with bad news constitute the control group. Differences in firm equity value of these two groups are investigated in order to examine the effect of scandals. Lags are added to this regression model in order to study long-term effects. In the second part, I apply a probit model to test the probability that a company has bad publicity depending on the number of times it has been part of a scandal in the past.
The term that is widely used to describe the ethical behaviour of companies is called ‘Corporate Social Responsibility’ (CSR). Although there is no unique definition for this expression, a widely accepted view describes it as the voluntary integration of social and environmental concerns in the business activities of a company (European Commission 2016). Studies on this topic
mainly focus on short-run impacts and range from investigations on the effect of CSR on firm value (Krüger 2015), at specific misconducts (Surendranath et al. 2015) or certain industries (Viscusi and Kersch 1990). Nevertheless, to date there is little agreement in the academic literature if corporate abuse is detrimental to investors or not. In contrast, the present work also determines long-term effects. Findings suggest that firm value decreases in the short-run after bad publicity in general, but this effect reverses again.
The question whether companies repeat their mistakes has not been studied in previous research. Results of the present work indicate that the likelihood of a repeated scandal increases with the number of corporate misconducts in the past. While Surendranath et al. (2015) find that scandals could act as an initiator for changes in a company that benefit investors, this is not shown with the present dataset. If companies change, then they should be less likely to get caught again. This indicates that scandals rather have a learning effect for companies implying that it is worth to repeat misconducts.
This thesis has been divided into seven parts. At first, an overview of the existing literature on the relationship between company behaviour and firm value is given. The second part deals with the data, while the third chapter is concerned with the empirical strategy used for this study. The fourth section presents the findings of the research, focusing first on the impact of scandals in the short and long-run, and second on the likelihood that companies repeat their mistakes. Section six introduces robustness checks and threats to validity. Finally, next to summarizing the main findings of this thesis, the last part provides both limitations and implications of the present work and suggestions for further research.
2 Literature Review
Previous research on the effect of CSR on firm value finds ambiguous results whether CSR is beneficial (Flammer 2011; Long and Rao 1995) or detrimental (Friedmann 1970) for firm value.
Theories that misconducts of a company have negative effects claim that unethical behaviour is related to an increase of monitoring costs and risks of shareholders (Long and Rao 1995) and that there is a great environmental awareness amongst investors (Flammer 2011). But Flammer (2011) also points out that benefits for responsible behaviour of a firm have declined over time. Related to that Krüger (2015) finds that shareholders react even negatively to positive CSR events when companies are likely to have agency conflicts or if management tries to make up for a prior abuse.
Research that focuses on certain industries such as the automobile sector mainly investigates the impact of recalls (Jarrell and Peltzmann 1985; Barber and Darrough 1996) or product liability lawsuits (Viscusi and Kersch 1990). These authors have consistent findings that such events result in a loss of shareholder value. Furthermore, Viscusi and Kersch (1990) compare impacts of scandals in the automobile sector to the pharmaceutical industry and find that the effect of automobile lawsuits on firm value is less pronounced. They claim that lawsuits researched in the automotive sector concern only material damage but no personal losses. Results for other industries indicate that events that involve affected humans have more negative consequences than property damage events.
In addition, Jarrell and Peltzmann (1985) find that recalls are not only much more costly than the direct costs of recalling a defective product, but also that spill over effects for competitors exist. On average, the loss of competitors is up to two third as much as that of the affected company. However, a re-examination of Peltzman’s and Jarrel’s paper claims that after correcting the database for overlaps, little significant evidence remains that share prices are significantly lower after an automotive recall. This accounts for shareholders of the recalling company and for shareholders of competitors (Hoffer et al. 1988).
Findings of studies regarding the food industry are more puzzling. Thomsen and McKenzie (2001) claim that significant shareholder losses exist only when companies are involved in recalls regarding severe violation of food safety
standards, while Salin and Hooker (2001) find no consistent relationship between the stock price reaction and the severity of the violation. They argue that differences appear between the size of companies, but not the extend of the abuse. Results show that shareholder returns fall sharply after the recall for the smallest firm, but there is no significant effect for larger firms. They explain this with the diversification advantage of large firms.
These studies all have in common that they use a standard event study methodology and use CSR events from pre-defined reliable news sources such as The Wall Street Journal (Flammer 2011; Jarrell and Peltzman 1985; Viscusi and Kersch 1990). Although using scandals reported in mainstream source might increase the reliability of results, these authors neglect the long-term effects of corporate abuses. Krüger (2015) claims that the windows of event studies should be rather short in order to address the reverse causality problem, because it is unclear if firms are performing good because of their ethical behaviour or whether they act ethical because they can afford it due to their performance.
Surendranath et al. (2015) examine long-term effects of CEO misconducts. They find that there is an increase in the stock price volatility of affected firms in the days following the announcement, but this higher risk and shareholder loss is only detected in the short-run. Stock prices of firms affected by the scandals match the performance of control firms in the long-run after the bad news were published. The authors claim that changes that are implemented in a firm after a scandal has been revealed, such as replacing corporate management or the establishment of new corporate governance practices, may act as a catalyst that benefits investors and leads to a higher share price than before the bad publicity.
In summary, studies on long-term effects of corporate misconducts are very limited. Firms might take losses into account in the short-run, if these reverse in the future. Therefore, it is even more important to study long-run implications of corporate abuses. Although Surendranath et al. (2015) address
long-term effects, is the universe of scandals investigated very restricted. In addition, there is still a gap in the existing literature regarding different types of scandals. Although Viscusi and Kersch (1990) claim that there might be a different effect of scandals including affected property or humans, they do not investigate this question for the automobile sector. Furthermore, none of the above mentioned studies has investigated the impact of CSR behaviour with a different approach than an event study. Applying a different empirical method might shed further light into contrasting results. Lastly, there is a general lack of research whether companies repeat their mistakes and if this is dependent on the number of conducted abuses in the past.
3 Data
3.1. BHRRC Data
Data about abuses of companies are collected from the publicly accessible Business & Human Rights Resource Centre (BHRRC),1 which includes published stories about companies all over the world. The dataset includes 18,364 stories about 3,204 companies from April 1996 until October 2014. Using the information provided from the BHRCC, variables such as the concerned company, date and source of a story, a story title and categories for each observation can be constructed.2 The selected sample is considered
sufficient for the aim of this research, as the BHRRC is the biggest library available regarding this type of information.
In order to generate the treatment variables, the titles and categories of a story are searched for certain expressions. Whether a story concerns a lawsuit can be identified by both filtering the variable regarding the type of a story in the BHRRC database for ‘lawsuit’ and searching each story title for the keywords ‘lawsuit’, ‘litigation’ and ‘sued’. Whether humans are affected by a company´s misconduct can be identified by looking at the BHRRC categories such as
1 http://business-humanrights.org/de.
2 The dataset and constructed variables are provided from Dr. Rafael Perez Ribas (University
‘beatings and violence’.3 Searching each story title for the expressions ‘death’, ‘dead’, ‘fatalities’ and ‘died’ can create the treatment variable regarding people who died because of the activities of a company. For the variable regarding treatment for being in the database with a story that was published in the mainstream news, the author of a story is filtered for pre-determined relevant sources. Table 1 shows the number of stories for each treatment variable in the database.
Table 1: Number of stories for each treatment in the database
Treatment Number of Stories
Scandali,t 101
Mainstreami,t 39
Unsafei,t 10
Deathi,t 8
Lawsuiti,t 14
Note: This table shows the number of stories for each treatment in the final database. This database includes firms from the automobile industry only.
Relevant sources that are determined as mainstream are defined according to a survey of the top 50 newspaper in the world regarding circulation (World Press Trends 2014). Due to the large population, this ranking includes many Asian newspapers in the Chinese or Japanese language. Therefore, in order to maintain a balance of global newspapers, some of these Asian newspapers are excluded and instead other commonly known news sources added. Also popular newspapers related to financial news are included in the list of mainstream sources, as it is very likely that these sources influence investors. Sources that are defined as mainstream are shown in Table 10 in the Appendix.
A certain story related to a firm might appear more than once in the news in different sources. The library of the BHRRC accounts for this by showing different components of one story, each with the same identification number (story ID). In order to investigate each story only once duplicates concerning one story are eliminated from the dataset, thus each story ID appears one time.
3 Categories considered are ‘death penalty’, ‘death threats’, ‘deaths’, ‘genocide’, ‘injuries’,
This sample also includes positive stories, because the BHRRC library lists as well if a firm was included in a rating or contributed positively to the environment. Filtering for these stories can partly be done with the help of the categories provided for each story by the BHRRC. Only positive stories are for example included in the category ‘rankings and indices’.4 Remaining positive stories can be identified by searching for certain keywords such as ‘red cross’ in the title of a story.5
Excluded from this research are not only positive stories, but also stories for which it is not possible to classify them into positive or negative. This accounts for stories that were assigned to the types ‘items‘, ‘reports‘ and ‘documents‘, which include for instance videos or petitions. The same applies for stories in the categories regarding conferences and events.6 The remaining items include only negative stories. One of these are company responses, which include statements of firms to accusations of abuse made by the public.
3.2. Financial Data
In order to investigate the impact of scandals on firm equity value, accounting data on company-level are collected from the database COMPUSTAT Global. These data cover more than 90% of the world’s capitalization represented by firms from over 80 countries (Compustat Global Data Manuals 2002). As company names in the COMPUSTAT database might be different than in the BHRRC library, firm names have to be standardized in each dataset first in order to merge accounting data from COMPUSTAT with information about firm scandals from the BHRRC according to company names.
4 Further categories of a story that indicate a positive story are ‘awards local & state
authorities’, ‘haiti’, ‘support for initiatives promoting civil & political rights’, ‘global business coalition on HIV/AIDS’, ‘national business coalition on HIV/AIDS’ and ‘swiss govt. & ICRC initiative on private military & security cos’.
5 Further keywords in the title of a story that indicate a positive event are ‘improve’, ‘award’,
‘fairtrade’ and ‘effort.’
6 Categories considered are ‘conference & events schedule’, ‘conferences & events: general’,
world summit on sustainable development’, ‘world social forum’, ‘world economic forum’ and ‘calls for contributions to conferences & events.‘
Standardizing of company names is conducted according to Wasi and Flaaen (2015). These authors introduce a command for the statistic software STATA in order to standardize firm names. This command first analyses and breaks down company names as firms in the databases sometimes include official names as well as trade or former titles. Then it capitalizes all letters and standardizes entity types. For example, all abbreviations for ‘incorporation‘ such as ‘inc‘ or ‘incorp‘ are standardized to ‘INC‘. Secondly, standardized firm names can be matched between the two databases performing a probabilistic record linkage at the 95% level. Lastly, the authors introduce a command in order to interactively review the generated matched pairs, which do not have a perfect score. Applying this command makes sure that the error probability of matching companies by mistake remains very low.
A unique GVKEY, which is the company identifier in COMPUSTAT, can be assigned to 1,541 companies and 6,736 stories of the BHRRC dataset. These firms represent the treatment group in the dynamic difference-in-differences framework. Firms in the COMPUSTAT Global database that do not appear in the BHRRC library are the control group.
According to the Standard Industry Classification (SIC) reported in COMPUSTAT each company in the dataset can be assigned to an industry (Compustat Global Data Manuals 2002). This is necessary as the automotive sector is in the focus of this research and different industries might not be affected in the same way by scandals. The number of firms in the final dataset related to the automobile industry is shown in Table 2. Furthermore, the SIC code identifies a control variable for the regression model indicating whether a company is a producer of final goods (sector ‘Manufacturing‘) or distributor (sectors ‘Retail Trade‘ and ‘Wholesale Trade‘).
Table 2: Overview automobile industry
SIC Code Sector Number of firms
3011 Tires & Inner Tubes 73
3711 Motor Vehicles & Passenger Car Bodies 130
3713 Truck & Bus Bodies 14
3714 Motor Vehicle Parts & Accessories 470
3715 Truck Trailers 6
5010 Wholesale-Motor Vehicles & Motor Vehicle
Parts & Supplies
32
5013 Wholesale-Motor Vehicle Supplies & New Parts 20
5500 Retail-Auto Dealers & Gasoline Stations 139
7500 Services-Automotive Repair, Services &
Parking
20
7510 Services-Auto Rental & Leasing (No Drivers) 32
Note: This table shows the number of firms in the dataset operating in the automobile industry. The categorization is conducted according to the Standard Industry Classification (SIC) given in the Compustat Global Data Manuals (2002).
To calculate the market value of each company, stock prices and the number of shares outstanding are downloaded from COMPUSTAT Global Security Daily. Because this database includes daily closing prices only, the last available observation of each calendar year is used. These data include years from 1995 until 2015. As control variable for the regression model Selling, General and Administrative Costs (SGA) are also downloaded from COMPUSTAT Global, because this variable includes for example marketing expenses. This is especially relevant when looking at the CSR of a company. Firms applying high ethical standards might have higher SGA in order to communicate this fact to customers. Furthermore, control variables are the leverage of a firm, the cash flow ratio and lastly, total assets to control for firm size. In order to generate reliable results and avoid biases only firms are included in this study, that have data for at least three years in a row and reported observations every consecutive year they appear in the database. Table 3 shows an overview of all variables used in the regression model.
Table 3: Overview of variables in the regression model
Variable Description Source Formula
Dependent Variable
logMVi,t Natural logarithm of the market
value of company i in year t
Compustat Security Daily
log(PRCCDt*CSHOCt)
Control Variables
logSGAi,t-1 Natural logarithm of Selling,
General, and Administrative Costs of company i in year t-1
Compustat Global log(XSGAt-1)
logATi,t-1 Natural logarithm of Total Assets
of company i in year t-1 Compustat Global log(ATt-1)
CFt/ATi,t-1 Operating cash flow of company i
in year t to Total Assets in year t-1 Compustat Global (IBt + DP t)/ ATt-1
Leveragei,t-1 Leverage of company i in year t-1 Compustat Global (DLTTt-1+DLCt-1)/ATt-1
Frequencyi,t-1 Variable indicating how often
company i appears in the dataset with a negative story up to t-1
BHRRC Sum of the number of
scandals of company i until t-1
Finali Indicating if company i is a
producer of final goods
Compustat Global 1= SIC Codes
5000-5999
0= All other SIC Codes
Treatment Variables
Scandali,t Company i was involved in a
scandal in year t BHRRC
Treatment dummies, where
1=scandal in year t 0=no scandal in year t Mainstreami,t Company i was involved in a
scandal in year t and the scandal was reported in a mainstream source
BHRRC
Unsafei,t Company i was involved in a
scandal in year t where people were affected
BHRRC
Deathi,t Company i was involved in a
scandal in year t where people died BHRRC
Lawsuiti,t Company i was involved in a
lawsuit in year t BHRRC
Note: This table shows all variables that are used in the regression framework including their description, where the data are obtained from and how they are calculated. All variables are measured at the calendar year end.
3.3. Descriptive Statistics
Descriptive statistics for several firm characteristics and the treatment variables are shown in Table 4, separately for the entire database and for firms that were not part of a scandal, once in the database or more than once. The mean firm value of a company in the entire database is 21.737 measured as the natural logarithm of market value. The most striking result of this table is, that the more often a firm appears in the dataset, the higher is the firm value and also firm size. This indicates that bigger firms tend to be more often part of a corporate misconduct than smaller firms. Furthermore, also Selling, General and Administrative Costs are higher, than for firms that were never part of a
scandal. This could on the one hand be due to the bigger firm size, but on the other hand, this does not support the considerations that firms paying less attention to ethical standards have lower expenses incurred to the selling and promoting of a product.
The number of observations N for each sub-group is shown in the last row of Table 4. Firms with zero scandals are the control group in the regression framework. Although the size of observations for this group is larger than that for firms with scandals, this imbalance is not seen as problematic in the regression framework. More observations rather improve the precision of results.
Table 4: Descriptive Statistics
Panel A: Firm characteristics
Entire Database Firms with 0 scandals Firms with 1 scandal Firms with >1 scandals
Variable Mean Median SD Mean SD Mean SD Mean SD
logMVi,t 21.737 21.799 2.822 21.690 2.812 24.076 1.733 24.291 2.141 logSGAi,t-1 6.790 6.767 2.859 6.720 2.841 9.258 1.778 10.058 1.628 logATi,t-1 8.814 8.699 2.879 8.745 2.858 11.372 1.682 12.334 1.621 CFt/ATi,t-1 0.087 0.081 0.056 0.088 0.056 0.102 0.041 0.067 0.032 Leveragei,t-1 0.255 0.247 0.172 0.254 0.172 0.292 0.118 0.313 0.137 Frequencyi,t-1 0.114 0.000 1.263 1.000 0.000 7.292 7.248 Finali 0.767 1.000 0.423 0.764 0.425 0.969 0.177 0.942 0.235 N 10,112 9,926 32 154
Panel B: Treatment variables
Entire Database Firms with 1 scandal Firms with >1 scandals
Variable Mean Median SD Mean SD Mean SD
Scandali,t 0.010 0.000 0.099 0.504 0.438 0.491 0.396
Mainstreami,t 0.004 0.000 0.062 0.369 0.156 0.370 0.162
Unsafei,t 0.001 0.000 0.031 0.177 0.031 0.223 0.052
Deathi,t 0.001 0.000 0.028 0.177 0.031 0.194 0.039
Lawsuiti,t 0.001 0.000 0.037 0.177 0.031 0.235 0.058
Note: This table reports summary statistics for the companies included in this research. Panel A reports firm characteristics for the entire database and separately for firms for which Frequencyi,t-1 , indicating the
number of scandals, is 0, 1 or larger than 1. Reported are the mean, median and standard deviation (SD) for the entire database and SD and mean for the sub-groups. N indicates the number of observations for each group. Variable descriptions are presented in Table 3. Variables are winsorized at the 5% level (with high only winsorizing if a ratio has only positive values). Panel B reports mean, median and SD for the treatment variables in the entire database and SD and mean for the sub-groups.
4 Empirical Method
The objective of this thesis is to investigate the effect of scandals on firm equity value of a company. It is very likely that a company’s share price decreases in the short-run after a misconduct has been revealed due to unsafe products, reputations loss, litigation costs or other monetary penalties. My hypothesis is that in the long-run this effect might be reversed and that share prices recover from their previous drop.
The method applied to test the proposed hypothesis is a fixed effects regression derived from Bertrand, Duflo, and Mullainathan (2004). They suggest a regression model to deal with more than two time periods and many treatment groups:
∆logMV!" = 𝐵! + 𝑐∆𝑋!"+ 𝛽𝑡𝑟𝑒𝑎𝑡!"+ 𝜖!", (1)
where the outcome variable of interest ∆logMV!" is the first difference of the natural logarithm of the market value of a firm i in year t. 𝑡𝑟𝑒𝑎𝑡!"is a dummy
variable indicating whether a firm i is treated in year t. The treatment effect is given by β. 𝐵! is the year effect to capture differences over time that are common to all firms. 𝑋!" are the first differences of control variables that may vary by firm and year such as lagged leverage (Leveragei,t-1), the logarithm of
lagged total assets (logATi,t-1) and Selling, General and Administrative
Expenses (logSGAi,t-1) or cash flow to lagged total assets (CFt/ATi,t-1).The error
term is defined by 𝜖!"#. Standard errors are clustered by firm in order to control for residual correlations of the error terms within firms.
In order to separate short and long-term effects, I include lags and leads in this regression model:
∆logMV!" = 𝐵! + 𝑐∆𝑋!"+ 𝛽!𝑡𝑟𝑒𝑎𝑡!"+ 𝛽!𝑡𝑟𝑒𝑎𝑡!"!! + 𝛽!𝑡𝑟𝑒𝑎𝑡!"!!+ 𝜖!", (2)
where m indicates the number of years a variable is lagged or used from the future. If treatment is causal, then dummy variables for leads are supposed to show no significant effect of treatment. But 𝛽! should be positive and significant in order to underpin the above mentioned hypothesis that firm value recovers from a scandal in the long-run.
In this framework, any firm that was in the media due to a scandal in a certain year t is considered as a treatment group in that year, while all other firms in the dataset are considered as control group.
The regression will be run with different definitions of treatment in order to examine the outcomes of different types of scandals. Definitions of treatment are being in the media with a scandal in general (Scandali,t), being in the media
with a scandal in a mainstream source (Mainstreami,t), being in the media with
a scandal were humans were injured or died (Unsafei,t), and this variable
applying only for scandals in a mainstream source (Unsafei,t x Mainstreami,t),
being in the media with a scandal including fatalities (Deathi,t) and lastly, being
in the media concerning a lawsuit (Lawsuiti,t).
The just described fixed-effects model has the potential to circumvent many of the endogeneity problems that typically arise when making comparisons between heterogeneous individuals (Bertrand, Duflo and Mullainathan 2004). One restrictive assumption is that changes in the outcome variable over time would have been exactly the same in both treatment and control groups in the absence of the intervention. It has also to be considered that the actions of a company leading to a scandal are seen as random and independent events. Furthermore, as the market value of a company is the dependent variable of the
empirical model, another underlying assumption is efficient markets. This implies that movements in stock prices after unanticipated events capture investors’ perception of future firm value (Fama 1965).
Next to the effect of scandals on firm equity value, this study also aims to investigate whether firms that were part of a scandal in the past are more likely to appear repeatedly in the public with bad news compared to companies that did not conduct any abuse. I apply following probit model (Stock and Watson 2011, p.385) in order to address this research question:
𝑃𝑟 𝑡𝑟𝑒𝑎𝑡!" = 1 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦!!!, 𝑋!" = Φ 𝛽!+ 𝛽!𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦!!!+ 𝑐𝑋!" , (3)
where Φ is the cumulative standard distribution function. The regression coefficient 𝛽! is positive if an increase in 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦!!! increases the probability that 𝑡𝑟𝑒𝑎𝑡!" = 1 holding all other regressors constant. Thus, a positive 𝛽! means that the more often a company was part of a scandal in the
past, the more probable it becomes that this company is part of bad publicity in the future again. As being in this database implies that firms were caught for their abuse, this suggests that companies do not learn from their mistakes or become more cautious. Equally, 𝛽! is negative if an increase in 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦!!!
decreases the probability that 𝑡𝑟𝑒𝑎𝑡!" = 1, indicating that firms learned lessons from former misconducts and are less likely in the public with scandals after being caught in the past. 𝑋!" are the same control variables as used in regression models (1) and (2).
5 Results
This section is divided into two parts. First, I present the results of the regression model investigating the impact of scandals on firm equity value in the short and long-run including lags. Then, I test the probability of repeated scandals with the help of a probit model to determine whether companies that were part of misconducts in the past are more or less likely to be part of bad publicity again. I include a dummy variable indicating whether a company is a final producer of goods in both of these sections. This should examine whether the impact of scandals is differently for companies that are producers of final goods or distribute products.
5.1. Impact of scandals on firm equity value
Table 5 shows the results of the regression model that estimates the effect of scandals on firm equity value.7 This model includes one lag of the treatment variables in order to estimate short-term effects and a second lag to capture long-term effects of scandals. The control variables described in Table 3 are included in each model to improve the efficiency of the difference estimator in the regressions.
This study neglects the immediate effects of treatment, but includes lags of the treatment variables to address the potential issue of reverse causality. This problem might arise, because companies that are hot in the market are more likely to get the attention of the main media and thus, are more likely to be exposed to bad publicity because of the greater public interest. It is the most common approach to lag the questionable variables by one or more periods. The underlying argument is that although present values of an independent variable might be endogenous to the dependent variable, it is not probable that past values of the explanatory variables are subject to the same problem.
7 For the treatments Treatdeath
i,t and Treatlawsuiti,t, all stories and not only those from a
mainstream source are considered in order to increase the sample size. Running the regressions with stories in mainstream sources related to fatalities (3 stories) and related to lawsuits (9 stories) delivers comparable results to those reported in Table 5.
The first lags of the treatment variable Scandali,t are significant and negative
ranging between -0.260 and -0.321. But coefficients of the second lag turn positive and have significant values between 0.305 and 0.411. This means that the costs a company saves in the long run by for example using cheap material or producing in low-wage countries are higher than the cost a firm has to bear after being part of a scandal.
This supports the hypothesis of this study that firms can recover from their abuse in the long-run. One explanation for this recovery is given from Surendranath et al. (2015). They claim, that corporate scandals can act as a catalyst that implements changes that even benefit investors. These changes can include new management or reversed corporate governance structures. This theory is supported by the positive values of the second lag, which are higher than the loss indicated by the coefficient of the first lagged variable. This suggests that scandals are advantageous for investors. Although this finding might help investors to make investment decisions, it should rather have implications for regulatory authorities. Corporate abuses should be punished more and not encouraging firms to repeat their misconducts, because it increases firm value. The question whether companies indeed tend to repeat their abuses, is addressed in the second part of this section.
The treatment Unsafei,t shows significant positive coefficients for the first
lagged variables (0.414 to 0.479) but a significant negative coefficient for the second lagged variable (-0.517). For the treatment Unsafei,t x Mainstreami,t
lagged for two years, coefficients are also significantly negative and very high ranging from -0.873 to -0.905. This confirms the considerations that news in mainstream sources have a higher impact on firm value. But the positive sign of the coefficients regarding short-term effects is not in line with one of the suggested hypothesis on the effect of scandals on firm equity value.
This positive short-term effect with regards to unsafe products is especially surprising in the automobile sector were customers are seen as more sensitive to issues regarding product safety and reputation of a company. In the
automotive industry, compared to others, mostly end-customers are affected. Especially for a car, safety is one of the most important points when a purchase is considered. In addition, because a car is a long-term investment these considerations might be valued higher than for a short-term investment. If safety is questioned, investors are expected to punish a company scandal more severely, because they might also be the victims themselves.8 In the fashion industry, for instance, purchases are rather short-term and mainly workers producing the clothes are affected. Although this is not a less serious abuse, proximity to investors is less.
One possible explanation could be that, especially with regards to unsafe products, there might be high costs involved in order to make up for the damage caused. These costs might not occur in the following year of a scandal but delayed, because the amount of payments still has to be determined or because long negotiations are in place.
Negative coefficients for the second lag of Unsafei,t, support the considerations
of Viscusi and Kersch (1990). They find that scandals concerning affected people have a greater impact compared to property damage. Table 6 shows that, contradicting to these negative signs, coefficients for the second lag of the treatment Scandali,t are positive. This suggests that punishment for more severe
scandals might be higher and that companies cannot recover from these as fast as from scandals in general.
8 Determining whether a scandal affects the end-customer or employers of a company is
Table 5: Difference-in-Difference Estimates for Firms in the Automobile Industry including 2 lags
Dependent Variable: ∆logMV
(OLS) (OLS) (OLS) (OLS) (OLS) (OLS) (OLS) (OLS) (OLS)
(1) (2) (3) (4) (5) (6) (7) (8) (9) Lags l.Scandali,t -0.271* -0.298 -0.318* -0.321* -0.315* -0.316* -0.260* -0.305* -0.314* (-1.79) (-1.64) (-1.72) (-1.73) (-1.76) (-1.66) (-1.69) (-1.69) (-1.73) l.Mainstreami,t 0.103 0.058 0.056 0.047 0.027 0.108 0.059 (0.87) (0.48) (0.44) (0.38) (0.22) (0.91) (0.47) l.Unsafei,t 0.537 0.460** 0.479* 0.414** (1.53) (2.24) (2.23) (2.32)
l.Unsafei,t x Mainstreami,t 0.325 0.280 0.267
(0.84) (0.67) (0.65) l.Lawsuiti,t 0.034 0.049 (0.18) (0.26) l.Deathi,t 0.192 0.345 (0.68) (0.94) l2.Scandali,t 0.305* 0.371* 0.411** 0.381* 0.393* 0.300* 0.308* 0.349* 0.387* (1.74) (1.80) (1.96) (1.78) (1.82) (1.82) (1.77) (1.73) (1.85) l2.Mainstreami,t -0.177 -0.154 -0.097 -0.078 -0.051 -0.148 -0.129 (-0.91) (-0.87) (-0.55) (-0.45) (-0.30) (-0.82) (-0.73) l2.Unsafei,t -0.517** 0.010 0.041 0.115 (-2.34) (0.02) (0.10) (0.28)
l2.Unsafe x Mainstreami,t -0.906* -0.873* -0.905*
(0.083) (-1.82) (-1.88) l2.Lawsuiti,t -0.183 -0.215 (-1.16) (-1.32) l2.Deathi,t -0.150 -0.321 (-0.68) (-1.25) Interaction
Finali x Scandali,t -0.059 -0.150 -0.136
(-0.43) (-0.88) (-0.79)
Finalii x Mainstreami,t 0.246* 0.300*
(1.70) (1.89)
Finali x Deathi,t -0.550*
(-1.68)
Control Variables yes yes yes yes yes yes yes yes yes
Year effects yes yes yes yes yes yes yes yes yes
Firm fixed effects yes yes yes yes yes yes yes yes yes
Observations 10,112 10,112 10,112 10,112 10,112 10,112 10,112 10,112 10,112
Overall R2 0.154 0.155 0.155 0.156 0.156 0.156 0.155 0.155 0.155
Note: This table shows the average treatment effect estimates for firms in the automobile industry. The estimates in columns 1 to 9 are from Ordinary Least Squares (OLS) regression with fixed effects gradually including the treatments to study differences in firm value (logMV) caused by a scandal. The table includes up to two lags for each treatment in order to study short and long-term effects. Definitions of treatment are being in the media with a scandal in general (Scandali,t), being in the media with a scandal in a mainstream source (Mainstreami,t), being in
the media with a scandal were humans were injured or died (Unsafei,t) and this variable applying only for scandals
in a mainstream source (Unsafei,t x Mainstreami,t), being in the media with a scandal including fatalities (Deathi,t )
and lastly, being in the media concerning a lawsuit (Lawsuiti,t). Control variables are logSGAi,t-1, logATi,t-1,
CFt/ ATi,t-1, Leveragei,t-1 and Finali. Variables are winsorized at the 5% level except Frequency,t-1. All regressions
cluster standard errors at a firm level. T-statistics are provided in parentheses. The symbols ***; **; * represent statistical significant at the 1%, 5%, and 10% levels, respectively.
Columns 1 to 6 in Table 5 do not account for the possibility that the impact of a scandal on firm value is different for companies, which are a producer of final goods or distributor of products. But distributors might be affected differently by a scandal, because they have not taken part in the design or manufacturing of a product, thus are not directly responsible for any lacking quality standards. This is especially relevant in the automobile industry, when any component of a car regarding the safety such as brackets or airbags are concerned. One example from the car industry is the recall of airbags of the manufacturer Takata since 2013, which involves ten global automakers that used these airbags in at least 17 million cars sold (Spector 2016). Therefore, columns 7 to 9 add an interaction term between the treatment variable and a variable that equals 1 if a company is a final producer of goods. 9
If the company is a product distributor or other, the effect of a scandal on firm value is the coefficient of the treatment variable. If the company is a final producer of goods the effect is the sum of the coefficients of the treatment variable and the interaction term. The outcomes of the regression analysis, testing the effect of a scandal depending on a company‘s position in the supply chain, show significant results for the interaction terms of the treatments Deathi,t and Mainstreami,t. The results show that firm value is lowered more
for final producers of goods than distributors, when the scandal includes fatalities (-0.205). In contrast to that, firm value is affected less for final producers of goods with regards to scandals in general (0.359).
One possible explanation could be that when people died in a car accident often one part of a car is responsible such as tires or breaks for which the manufacturer can be made responsible explicitly. But for less severe scandals the results indicate that distributors are punished, possibly because they have the duty to make sure that its supplier fulfils all quality and safety standards and every company in the supply chain has to be responsible for a product.
9 This interaction term can only be included for the treatment variables Scandal
i,t, Mainstreami,t
and Deathi,t, because all other variables include only manufacturing firms. Thus, no
The rather small amount of significant coefficients could be due to the rather small sample of scandals. Although this does not mean that the causal effect is biased, but it could be estimated imprecisely. Another explanation could have a statistical background, namely that the control variables are not selected such that the error term 𝜖!"# satisfies the conditional mean independence. This means that the control variables do not fulfil the condition of being pre-treatment individual characteristics and thus, the causal effect of the coefficients of the treatment variables is biased. In addition, the regression could still face omitted variable bias (Stock and Watson 2011, pp.472-477). Lastly, it could also be possible that firms in the control group may actually appear with a bad story in the dataset of the BHRRC, but no GVKEY could be matched to them. As firms of the BHRRC library for which it was not possible to assign a GVKEY were excluded from the treatment group, there is a chance that firms may be in the control group, although they have been part of a scandal. This could also bias the results.
Finally should be added, that firm value in this study is calculated as market value. This means that the outcome variable shows what investors‘ perception on a company’s value is and not the true value of a company according to its financial statements. Nevertheless, when investors value a firm badly it could also be assumed that sales suffer, which negatively affects the balance sheet as well as litigation costs for example.
5.2. Probability of repeated scandals
Results in Table 6 show whether the number of scandals in the past has a significant effect on the likelihood of being part of a scandal again in the future. Coefficients on Frequencyi,t-1are significant and positive for reported
treatments except for the variable Mainstreami,t. This indicates that firms that
were part of bad publicity in the past are more likely to conduct corporate abuses again in the future.
Table 6: Probability of repeated scandals
Pr(treati,t=1)
Scandali,t Mainstreami,t Unsafei,t Unsafe x Mainstream
i,t Deathi,t Lawsuiti,t
Frequencyi,t-1 0.051*** 0.023 0.107*** 0.082*** 0.081*** 0.054*** (0.01) (0.03) (0.01) (0.01) (0.02) (0.02) logSGAi,t-1 0.017 -0.159 0.247 -0.020 -0.161 -0.016 (0.22) (0.207) (0.16) (0.07) (0.234) (0.11) logATi,t-1 0.252 0.410* -0.137 0.165*** 0.245 0.321*** (0.21) (0.217) (0.17) (0.04) (0.251) (0.10) CFt/ATi,t-1 -0.108 1.601 -3.488 0.162 -2.834 -7.047** (2.21) (2.30) (2.53) (0.69) (1.810) (3.49) Leveragei,t-1 0.841 -0.005 -0.635 -0.101 -0.227 1.889** (0.95) (0.91) (0.58) (0.617) (0.76) (0.91) N 10,112 10,112 10,112 10,112 10,112 10,112 log Likelihood -304.430 -174.721 -47.141 -29.104 -49.730 -60.068
Note: This table shows the coefficient of a probit regression testing whether an increase in the variable
Frequencyi,t-1 leads to a higher probability that a company is part of a scandal (treatment). Definitions of
treatment are being in the media with a scandal in general (Scandali,t), being in the media with a scandal
in a mainstream source (Mainstreami,t), being in the media with a scandal were humans were injured or
died (Unsafei,t) and this variable applying only for scandals in a mainstream source (Unsafei,t x
Mainstreami,t), being in the media with a scandal including fatalities (Deathi,t ) and lastly, being in the
media concerning a lawsuit (Lawsuiti,t). Control variables are logSGAi,t-1, logATi,t-1, CFt/ATi,t-1 and
Leveragei,t-1. Descriptions for these variables are shown in Table 3. The reported coefficient shows
whether the respective variableincreases the probability that treati,t=1, holding all other regressors
constant. Robust standard errors are provided in parentheses. The symbols ***; **; * represent statistical significant at the 1%, 5%, and 10% levels, respectively.
This effect of repeated scandals is most pronounced for misconducts that include unsafe products. This is striking, because according to the regression results in Table 5, long-run firm value is compared to the control group lower when corporate misconducts lead to injured humans. Thus, this should rather imply that firms learn from their mistakes and take the punishment of
decreased firm value into consideration. On the other hand, this could mean that also other companies repeat their mistakes, but these are less in the focus of the media or regulatory authorities because of less severe scandals. Following, it is less likely that these companies are caught for their abuses.
However, these results do not support the hypothesis that scandals lead to a change in companies as suggested by Surendranath et al. (2015). If firms changed, they would not repeat their misconducts. Findings rather imply that firms learn from past abuses that they are profitable, or at least have no long-term detrimental effect. Therefore, they have no reason to change their behaviour. This could be due to not sufficient punishment of regulatory bodies, but also because shareholders do not take ethical consideration into account enough. As negative short-term effects for scandals in general turn positive in the future, this could also indicate that stakeholders tend to forgive fast.
The significant positive coefficients of logATi,t-1 in Table 6 indicate that bigger
firms are more likely to be involved in a scandal. On the hand, previous argument can be applied that big companies are in greater public interest and it is more likely that they are caught. On the other hand, this could imply that, next to learning from past abuses, these companies have a ´too big to fail’ attitude relying on mild penalties and shareholder trust.
Furthermore, it should also be investigated whether the effect of repeated mistakes is differently for final producer of goods and distributors with the example of the treatment Deathi,t. The graph in Figure 1 indicates that product
distributors are more likely to repeat abuses than final producer of goods. One possible reason could be that it is easier for distributors to give other parties of the supply chain the fault for unsafe products and thus, they might be more attempted to repeat a mistake. This is also in line with the results reported in Table 5, which show that firm value is lowered more for final producer of goods than distributors, when the scandal included fatalities. Thus, because final producer of goods incur higher losses they might be less attempted to repeat an abuse.
Figure 1: Marginal effects of the treatment Deathi,t separately for final producers of goods and distributors. The variable Finali,t equals 1 if a company is a final producer of products according to the SIC codes 5000-5999.
In summary, this section shows that firms repeat an abuse more likely in the future, if they were in the media with bad publicity in the past. On the one hand, this indicates that the media and regulatory authorities work efficiently to detect the actions of these firms. On the other hand, it implies that firms are not punished enough for their previous abuses to keep them from acting not ethical again.
6 Robustness checks and threats for validity
The regression framework used in this research has the advantage that it compares two groups over the same time period and simultaneously looks at the same companies before and after a treatment. This avoids the problem of unobserved differences between two distinct groups and mitigates the issue of unobserved trends. But reliable results of this framework imply that in the absence of treatment the average change of the outcome variable is the same for the treatment and control group. Because this so-called ‘parallel trends’
assumption cannot formally be tested, sensitivity or robustness tests should be done (Roberts and Whited 2012, p.38-44).One of them is to make sure that treatment and control groups are balanced. This means that they are relatively similar according to observables relevant for the treatment. Incorporating control variables in the regression can support balanced groups and is done in each statistical model. Furthermore, Table 7 includes one lead to make sure that there is no anticipatory behaviour regarding treatment. Coefficients on leads examine the ex-ante effect of a scandal on firm value. These coefficients should be insignificant, because otherwise this would indicate anticipatory behaviour regarding a treatment and therefore, treatment might not be causal.
Except for to coefficients of the treatment Unsafei,t x Mainstreami,t every
coefficient on leads is negative indicating that causal treatment can be assumed. One possible explanation for the significant coefficients with regards to unsafe products is, that companies might let customers in the distribution channels know earlier than the public, when there is a car recall for example. This causes that stock prices react before a public announcement.
Another threat to validity is the rather small sample size of the treatment group. Although this does not bias the estimators of the causal effect, coefficients might be estimated not precisely (Stock and Watson 2011, p.477). Secondly, although the database of the BHRRC consists of a wide range of scandals, not all of them are reported in this library. Following, some firms could be in the control group although they have been part of a scandal and should be in the treatment group.
Table 7: Difference-in-Difference Estimates for Firms in the Automobile Industry including 1 lead
Dependent Variable: ∆logMV
(OLS) (OLS) (OLS) (OLS) (OLS) (OLS)
(1) (2) (3) (4) (5) (6) Treatment Scandali,t -0.107 -0.191 -0.181 -0.208 -0.227 -0.228 (-0.74) (-1.10) (-1.02) (-1.13) (-1.22) (-1.19) Mainstreami,t 0.191 0.196 0.24 0.222 0.257 (1.43) (1.39) (1.37) (1.37) (1.43) Unsafei,t -0.196 0.138 0.063 0.189 (-0.78) (0.87) (0.33) (0.80)
Unsafe x Mainstreami,t -0.579 -0.527 -0.495
(-1.31) (-1.16) (-1.08) Lawsuiti,t 0.207 0.161 (1.15) (0.84) Deathi,t -0.277 (-1,31) Lead f.Scandali,t 0.173 0.106 0.102 0.118 0.129 0.138 (1.3) (0.81) (0.75) (0.86) (0.93) (0.97) f.Mainstreami,t 0.148 0.143 0.113 0.154 0.183 (0.78) (0.74) (0.55) (0.75) (0.71) f.Unsafei,t 0.121 -0.128 -0.863 -0.032 (0.45) (-0.49) (-0.27) (-0.08)
f.Unsafe x Mainstreami,t 0.722 0.890* 0.900*
(1.51) (1.75) (1.73)
f.Lawsuiti,t -0.32 -0.376
(-1.43) (-1.39)
f.Deathi,t -0.105
(-0.31)
Control Variables yes yes yes yes yes yes
Year effects yes yes yes yes yes yes
Firm fixed effects yes yes yes yes yes yes
Observations 10,112 10,112 10,112 10,112 10,112 10,112
Overall R2 0.153 0.153 0.153 0.161 0.161 0.161
Note: This table shows the average treatment effect estimates for firms in the automobile industry in the long-run. To study the changes in firm value (logMV) caused by a scandal the same definitions for treatments as in Table 5 are tested controlling for the same explanatory variables. In addition, this table includes one lead for each treatment in order to study the causal effect of treatment. All regressions use fixed effects and cluster standard errors at a firm level. T-statistics are provided in parentheses. The symbols ***; **; * represent statistical significant at the 1%, 5%, and 10% levels, respectively.
7 Conclusion
The purpose of the current study is to determine how firm equity value is affected by scandals in the short and long-run. First, a dynamic difference-in-differences framework is applied to show how firm value of companies that conducted an abuse differs from companies, which have not been part of such a scandal. Findings suggest that firm value of companies that were in the media regarding bad news have lower firm value than the control group in the short-run, but higher firm value in the long-run. This supports the hypothesis of this research that firm value recovers from corporate misconducts in the long-run.
The current findings highlight the importance of punishment of corporate abuses. Greater efforts are needed to ensure that businesses do not take results like provided in this study as reason to behave unethical. Benefits that can be generated by producing in low-wage countries or applying less ethical standards should not outweigh the costs when these actions are brought to the public. Furthermore, investors might need to reconsider their attitude towards ethical investing. Is it right not to care how a firm made its profits and see drop in share price after a scandal as buying opportunity? Should society support unethical behaviour of a firm?
Contradicting, findings of scandals regarding unsafe products suggest that firm value turns negative in the long-run. Policymakers should use the level of penalties for these misconducts to reach comparable effects for other scandals, which resulted in increased long-run firm value. Companies should not be encouraged to act unethical, because it enhances performance.
Nevertheless, this research shows that companies are more likely to be part of a scandal again, when they were in the news with bad stories in the past. This is not depending on how severe the abuse was. In contrast to Surendranath (2015), this does not indicate that scandals lead to a change in corporate behaviour. Thus, regulatory authorities should work harder to keep companies from behaving erratically. As being in the dataset of this work means that firms were caught for their abuses, this implies that authorities and the media are working effectively to detect corporate misconducts. Thus, it is even more striking that firms still tend to repeat their mistakes. This calls again for a change in the regulatory framework regarding corporate scandals such as the Sarbanes-Oxley Act implemented in the U.S. in 2002 after a wave of accounting scandals.
A limitation of this study is the rather small sample of corporate abuses, especially when treatments are separated in different misconducts. Critics of this study could also argue about the imbalance of the control and treatment group. Furthermore, it has to be considered that the actions of a company
leading to a scandal are seen as random and independent events. Lastly, there is a potential issue of reverse causality. Although lags in the regression might prevent this problem, this solution could lead to a loss of the precise measurement of treatment effects. Another approach would have been to find instrumental variables as lags are no way of assessing how serious endogeneity is and whether these are effective to solve this statistical problem.
Further research investigating spill-over effects of scandals would be interesting. This should assess how firm value of competitors is affected when a certain company in an industry has bad publicity. In addition, it would be worthwhile to investigate whether any specific events lead to a changed perception of shareholders regarding CSR. Do investors for example punish corporate abuses harder after the financial crisis? Lastly, future research could look into how effects of scandals differ between countries. Countries, where scandals affect firm value detrimental in the long-run can be used as a role-model to implement a regulatory framework for other countries in which scandals still act as catalyst.
References
Barber, B.; Darrough, M.: Product Reliability and Firm Value: The Experience of American and Japanese Automakers. Journal of Political Economy, Vol.104, No.5 (October 1996), pp.1084-1099.
BBC News: Volkswagen: The scandal explained.
http://www.bbc.com/news/business-34324772 (retrieved 10th of June 2016).
Bertrand, M; Duflo, E.; Mullaintathan, S.: How much should we trust differences-in-differences estimates?. The Quarterly Journal of Economics (February 2004), pp. 249-275.
Business & Human Rights Resource Center: http://business-humanrights.org/en/issues.
Cutler, D.; Poterba, J.; Summers, L.: What moves stock prices?. The Journal of Portfolio Management, Vol. 15, No. 3 (Spring 1989), pp. 4-12.
Compustat Global: Data Guide by Standard and Poor’s. https://wrds-web.wharton.upenn.edu/wrds/support/Data/_001Manuals%20and%20O
verviews/_001Compustat/_001North%20America%20-
%20Global%20-%20Bank/_170Compustat%20Global%20(FTP%20version)%20Data% 20Manuals.cfm (retrieved 15th of May 2016).
Davidson, W.; Worrell, D.: The Effect of Product Recall Announcements on Shareholder Wealth. Strategic Management Journal, Vol. 13, No. 6 (September 1992), pp. 467-473.
European Commission: Corporate Social Responsibility (CSR) in the EU. http://ec.europa.eu/social/main.jsp?catId=331 (retrieved 9th of June 2016).
Fama, E.: The Behavior of Stock-Market Prices. The Journal of Business, Vol.38, No. 1 (January 1965), pp. 34-105.
Fang, L.; Peress, J.: Media Coverage and the Cross-Section of Stock Returns. The Journal of Finance, Vol. 64, Issue 5 (September 2009), pp.2023-2052.
Flammer, K.: Corporate Social Responsibility and Shareholder Value: The Environmental Consciousness of Investors. Academy of Management Journal, Vol. 56, No. 3 (June 2013), pp. 758-781.
