Professional Documents
Culture Documents
A Taxonomy of Financial Market Manipulations: Establishing Trust and Market Integrity in The Financialized Economy Through Automated Fraud Detection
A Taxonomy of Financial Market Manipulations: Establishing Trust and Market Integrity in The Financialized Economy Through Automated Fraud Detection
ª 2017 Association for Information Technology Trust All rights reserved 0268-3962/17
www.palgrave.com/journals
Research Article
Correspondence:
M Siering, Goethe University Frankfurt, Theodor-W.-Adorno-Platz 4, 60323 Frankfurt, Germany.
Tel: +49 (0)69 798 34683;
Fax: +49 (0)69 798 35007;
E-mail: [email protected]
Abstract
Financial market manipulations represent a major threat to trust and market integrity in
capital markets. Manipulations contribute to mispricing, market imperfections and an
increase in transaction costs for market participants and in costs of capital for issuers.
Manipulations are facilitated by increased transaction velocity, speculative trading and
abusive usage of new trading technologies, i.e., they are directly linked to financial sector
changes that drive financialization. Research at the intersection of financialization and IS
might support regulatory authorities and market operators in improving market surveillance
and helping to detect fraudulent activities. However, confusing terminology is prevalent on
financial markets with respect to different manipulation techniques and their
characteristics, which hampers efficient fraud detection. Furthermore, recognizing
manipulations is challenging given the large number of information sources and the vast
number of trades occurring not least because of high-frequency traders. Therefore,
automated market surveillance tools require a comprehensive taxonomy of financial
market manipulations as a basis for appropriate configuration. Based on a cluster analysis
of SEC litigation releases, a review of the latest market abuse regulation and academic
studies, we develop a taxonomy of manipulations that structures and details existing
manipulation techniques and reveals how these techniques differ along several
dimensions. In a case study, we show how the taxonomy can be utilized to guide the
development of appropriate decision support systems for fraud detection.
Journal of Information Technology (2017). doi:10.1057/s41265-016-0029-z
Introduction
he financial crisis triggered an intensive debate about financial activity does not fulfill its main purpose, i.e., to
(e.g., over-the-counter (OTC) derivatives) and a tremendous market manipulations. Previous research has already shown
increase in turnover on financial markets. These trends are that appropriate information systems (IS) can help to
mirrored in the term financialization, which is defined as ‘‘the identify fraudulent content (Abbasi et al., 2010; Zahedi
increasing dominance of financial actors, markets, practices, et al., 2015) and communication behavior (Zhou et al.,
measurements and narratives, at various scales, resulting in a 2004). Nevertheless, although the necessity of decision
structural transformation of economies, firms (including support systems (DSSs) that support market surveillance
financial institutions), states and households’’ (Aalbers, 2016: authorities in the field of financial fraud detection appears
3). Key drivers of financialization are deregulation, the focus obvious, few academic studies exist in this particular context.
on the shareholder value concept and, last but not least, the One essential prerequisite for building DSSs that are able to
proliferation of information technology (IT) (Lagoarde- assist with the identification of potentially fraudulent behav-
Segot, 2016). New trading technologies such as algorithmic ior is an in-depth domain and data understanding (Fayyad
trading and high-frequency trading (HFT) lead to massive et al., 1996) that prevents improper system configurations and
increases in turnover velocity on financial markets because might lead to the discovery of wrong and invalid patterns.
they enable sophisticated market participants to generate Although a clear distinction between different manipula-
significant total profits based on small profits per trade and tion techniques is a prerequisite for efficient market surveil-
to instantaneously react to profitable situations in a highly lance, confusing terminology is prevalent with respect to
complex and fragmented trading environment (Gomber manipulation techniques and their main characteristics;
et al., 2016). existing classifications (e.g., Allen and Gale, 1992) neither
Furthermore, since financial markets have existed, market provide a comprehensive view of differences and similarities
participants try to generate profits from deceiving others nor cover the newly emerging forms of manipulation in
(Leinweber and Madhavan, 2001). For instance, stock prices conjunction with HFT. Furthermore, they do not link to the
are manipulated to pretend a trend (Comerton-Forde and configuration of fraud detection systems.
Putnins, 2011), or false and misleading information is To close this research gap and to follow the call of Ngai
disseminated to cause market reactions and to profit from et al. (2011) concerning the automated detection of fraud in
other market participants’ actions. However, in recent years, the field of financial markets, we develop a novel and
the magnitude of financial market manipulations and their comprehensive taxonomy of market manipulations by (1)
effect has increased continuously. Technology-savvy market analyzing current cases of prosecuted market manipulations
participants, such as HFTs, are perceived to exploit their based on a cluster analysis of litigation releases published by
technological advantage to perform deceptive tactics (Biais the US Securities and Exchange Commission (SEC), (2)
and Woolley, 2011; Cumming et al., 2012) at the expense of performing a structured literature review to include all
other market participants or to contribute to extreme and manipulation techniques studied by academics and (3)
unexpected market price fluctuations as witnessed in the May completing the taxonomy based on an analysis of current
6, 2010, Flash Crash (Easley et al., 2011). This event brought financial market regulation. That is, we include the European
these new trading technologies to the forefront of the Market Abuse Regulation (MAR) for multi-country coverage
financialization debate and triggered far-reaching regulatory of market manipulations. Thus, we follow the methodology
initiatives both in the USA and in Europe to monitor or even proposed by Nickerson et al. (2013) and build a taxonomy of
curb this form of speculative trading, which is accused of market manipulations, considering both the differences and
focusing solely on market microstructure noise rather than similarities of various market manipulation techniques and
contributing to market efficiency by incorporating relevant new types of manipulations related to HFT strategies in
information into asset prices. particular.
In addition to this increased risk of speculative mispricing We contribute to research at the interface of financializa-
and potential market manipulation due to new technologies, tion and IS by (1) systematically documenting and organizing
numerous concrete scandals and accusations in the context of new and established forms of market manipulation that are
market abuse and manipulation were recently uncovered. based on traditional and new (potentially speculative) trading
The manipulation of the fixings in the foreign exchange technologies, such as HFT, in a market manipulation
markets over several years (Financial Conduct Authority taxonomy and (2) evaluating this taxonomy for the auto-
(FCA), 2014) and the Libor rate manipulation scandal mated detection of market manipulations based on sophis-
(Financial Services Authority (FSA), 2012) revealed massive ticated information systems. Furthermore, we provide specific
rigging activities in the area of financial benchmarks that led guidelines on how to derive a DSS configuration to detect
to fines of more than $10 billion until 2016 (Vaughan, 2016). fraudulent market behavior based on a systematically derived
Politicians, regulators and the public thus push for increased taxonomy, thereby contributing to research in IS. The results
transparency and massive efforts in the area of surveillance to are highly relevant for regulators and market surveillance
curb market abuse and manipulation and to re-establish trust authorities because the systematic and automated detection of
in the financial sector. market manipulation prevents abusive behavior and con-
Due to the massive increase in transactions triggered by tributes to re-establishing trust in financial markets.
speculative trading, an enormous amount of data must be This paper is structured as follows. Section ‘‘Research
analyzed to generate alerts based on overnight-batch pro- context: financial market manipulations and their detection’’
cesses that can be investigated in detail to identify potential presents related work on financial market manipulations and
and actual market manipulations. Because order and trading fraud detection. Section ‘‘General research approach’’ briefly
data can hardly be analyzed manually, automated fraud explains the general research methodology including the
detection systems are required that provide alerts of potential taxonomy approach. Section ‘‘Analysis of financial market
A taxonomy of financial market manipulations M. Siering et al
manipulations based on SEC litigation releases’’ provides first market manipulations is lowered due to increasing amounts
insights into market manipulations based on a cluster analysis of data to be analyzed. To increase trust, the prevention of
of litigation releases. Insights into our structured literature and market manipulations is of high importance, which requires
regulatory review are provided in section ‘‘European market appropriate IS for fraud detection. Research on IS indicates
abuse regulation and literature review’’. Section ‘‘Developing a that automated detection systems are able to find patterns in
taxonomy of financial market manipulations’’ presents the datasets to identify fraudulent Web content (Abbasi et al.,
taxonomy development. Section ‘‘Evaluation of the taxonomy’’ 2010; Zahedi et al., 2015). Additionally, related classification
evaluates our taxonomy and shows in a case study how the methods might be used to detect deceptive content in
taxonomy can be applied in the field of fraud detection. computer-mediated communication (Zhou et al., 2004).
Finally, section ‘‘Conclusion’’ summarizes our findings. Nevertheless, previous research has also shown that an in-
depth domain understanding is essential for finding an
appropriate system configuration (Fayyad et al., 1996).
Research context: financial market manipulations and their In the field of financial market manipulations, Allen and
detection Gale (1992) are among the first who provide a structured
Manipulation of financial markets is as old as the markets overview and categorization. They differentiate between
themselves and has been studied extensively in the academic actions that manipulate the actual or perceived value of an
literature both from an economic and from a legal perspec- asset (action-based manipulation), the release of false or
tive. To avoid manipulative actions and to ensure capital misleading information to move stock prices in the desired
markets’ integrity, exchanges and regulatory authorities direction (information-based manipulation) and attempts to
engage in market surveillance to detect manipulative behav- manipulate by buying and selling a stock in a specific manner
ior (Leinweber and Madhavan, 2001). (trade-based manipulation). However, the taxonomy devel-
Because manipulations can occur in different forms and oped by Allen and Gale (1992) is not sufficiently detailed for
categories, there is no generally accepted definition of use as an input for an automated market surveillance system
financial market manipulation or financial fraud (we do because it does not address a potential system configuration.
not distinguish between market fraud and manipulation, and Moreover, due to technological enhancements, new forms of
both terms are used interchangeably in the following). manipulation have evolved that are not covered by this
Therefore, we follow a broad definition of financial market taxonomy. For example, algorithmic traders and in particular
manipulations; first, market manipulations encompass the the subset of HFTs might try to manipulate stock prices
manipulation of financial disclosures accompanied by large based on order submissions and their subsequent deletions to
accounting scandals such as Enron and Worldcom, because profit from their low-latency access to markets and market
such scandals hamper the efficient functioning of markets data.
(Akhigbe et al., 2005; Jones, 2011). Second, investor confi- Manipulation is a serious threat to the efficiency and
dence is impeded by investment manipulations such as the liquidity of trading venues and undermines the trust of
Ponzi scheme set up by Bernard Madoff (Pozza et al., 2009; market participants (Kyle and Viswanathan, 2008). There-
Rapoport, 2012). Third, insider trading is another form of fore, detection of market manipulation/financial fraud is
market manipulation, drawing on information asymmetries essential to ensure the orderly function of markets and to
(Arshadi, 1998; Allen and Ramanan, 1995). Fourth, there is a provide a level playing field for all participants. Distinguish-
variety of different fraudulent trading activities aiming at ing fraudulent financial data from authentic data is one of the
financial market manipulation that we also consolidate under most difficult tasks regulators face (Ngai et al., 2011). In the
this umbrella term. According to Cumming and Johan (2008: USA, the SEC is responsible for market manipulation
456), such ‘‘trading practices … distort prices and enable detection and prosecution. Self-regulatory (non-governmen-
market manipulators to profit at the expense of other tal) institutions such as the Financial Industry Regulatory
participants, creating information asymmetries’’. Forms of Authority (Davis, 2007) are also active in this field. In the
such market manipulations are, for example, ‘‘Pump and European Union (EU), the MAR, which has been in force
Dump’’ techniques in which the manipulator releases false- since 2014 and has been applied since July 2016, lists
positive information concerning a company to profit from activities that are defined as financial market manipulation.
the subsequent price increase (Sabherwal et al., 2011) or There are a variety of tools that can be used by regulators
‘‘Marking the Close’’ in which the fraudster tries to manip- to detect market manipulation (Wheeler and Aitken, 2000).
ulate the closing price of a stock (Hillion and Suominen, Of special interest for this study is the use of intelligent
2004). These manipulations have a severe effect on trust in systems for decision support to identify suspicious behavior
financial markets and on investors, who can lose substantial (Baker, 2005; Humpherys et al., 2011). Although there is a
parts of their investment, e.g., due to ‘‘Pump and Dump’’ large research stream about the empirical evidence of
fraud (Hanke and Hauser, 2008). A detailed overview of manipulations and their effects on financial markets, little
market manipulations addressed in academic studies and of research addresses the detection of financial market fraud by
international market abuse regulation is provided in our using an automated DSS (Comerton-Forde and Putnins,
literature and regulatory review (‘‘European market abuse 2011).
regulation and literature review’’ section). One approach to detecting manipulation is proposed by
Due to the increasing utilization of IT in particular, which Aggarwal and Wu (2006), who analyze 142 cases of stock
is also one main driver of financialization (Lagoarde-Segot, market manipulation. They find that potentially informed
2016), the risk of market manipulations increases due to parties and market makers are likely to be manipulators and
novel forms of manipulation arising from technologies such show that prices of manipulated stocks rise through the
as HFT. Furthermore, the chance of manually detecting manipulation period and fall thereafter. Comerton-Forde
A taxonomy of financial market manipulations M. Siering et al
and Putnins (2011) developed a measure of the probability of Analysis of financial market manipulations based on SEC
closing price manipulation in a given market for a stock. litigation releases
Furthermore, Ledgerwood and Carpenter (2012) derived a
theoretical framework for classification and detection of SEC litigation releases and dataset acquisition
market manipulation. Other work focuses on peer group Litigation releases (LRs) report the enforcement actions of
analyses using unsupervised learning to detect suspicious the SEC. Among other information, a release can include
behavior (Kim and Sohn, 2012) or data mining techniques to arraignments, a subpoena (e.g., defendants in a civil court
detect fraudulent financial statements (Kirkos et al., 2007). explain themselves), an announcement of investigation, a
However, no generally accepted basis for the development of final judgment (e.g., penalty or disgorgement payments are
an automated financial fraud detection system exists. announced), a trading halt (e.g., to protect the unsuspecting
Furthermore, none of the mentioned approaches contains a public from manipulation) or public service announcements.
means to detect market manipulation conducted by HFTs or LRs represent a prominent source to analyze market manip-
other market participants using a low-latency infrastructure. ulations that are prosecuted by the SEC.
With this study, we therefore respond to the call for research We downloaded all LRs from the SEC homepage from
by Ngai et al. (2011) and contribute to the scarce literature December 29, 1995, to July 22, 2013, resulting in 8093
on financial fraud detection by proposing a taxonomy of releases. To avoid data inconsistency and noise, we cleaned
market manipulations that is of fundamental importance to the obtained dataset by excluding years that were only partly
configure automated fraud detection systems. covered due to the introduction of the database in 1995/1996
(and the incomplete year 2013), LRs that did not refer to
Rule 10b-5 of the SEC, which covers the ‘‘employment of
General research approach manipulative and deceptive devices,’’ and large scandals and
The primary goal of this study is to develop a taxonomy of Ponzi schemes because they tend to involve large numbers of
financial market manipulations. Taxonomy development has defendants, claimants, and different courts, which leads to
been studied intensively in the social sciences. A detailed redundancy in the dataset and might influence the clustering
overview is provided by Bailey (1984). We apply the process (Coffee, 2005). Nevertheless, Ponzi schemes are later
methodology developed by Nickerson et al. (2013), which included in the taxonomy because they represent a form of
is accepted in the field of IS, and adapt it by including financial market manipulation. Table 1 depicts the data
different sources for input, i.e., currently existing market processing in detail.
manipulations, as shown in Figure 1. Our final dataset covers 4340 releases. Table 2 provides
We identify market manipulation techniques based on summary statistics for the documents in the total raw dataset
three sources from academia and international regulation: and for the documents that are used for the analysis.
SEC litigation releases to determine fraudulent financial
market activities that are prosecuted in the USA, manipu- Clustering approach
lation techniques that are described in the MAR and a To derive insights into which general categories of manip-
structured literature review on financial market manipula- ulative behavior are prosecuted by the SEC, the LRs are
tions. Additionally, we rely on the European Benchmark clustered. Therefore, the raw data must be transformed in a
Regulation (European Parliament and Council, 2016) as a form that is readable and processable by an unsupervised
source for fraudulent actions aimed at the manipulation of clustering algorithm (Rokach and Maimon, 2005). The
market benchmarks. As a result, we determine the different objective behind the unsupervised clustering approach is to
forms of market manipulation which are then used together reveal novel patterns and to explore an unknown database
with a first taxonomy developed by Allen and Gale (1992) as without the requirement of specific knowledge about the
a starting point to derive our taxonomy, which shows particular structure of the LRs.
mutually exclusive and collectively exhaustive characteristics. The documents are preprocessed before the clustering
The already existing dimensions included in the taxonomy algorithm is applied. Therefore, they are reduced to their
are determined based upon established category definitions fundamental parts, and a term frequency measure is applied
in the academic literature. New dimensions are built drawing to create a word vector of the documents. First, the
on insights from the structured literature review and documents are tokenized by applying a split after every
international market abuse regulation. An evaluation and non-letter character (e.g., commas, semicolons and spaces)
discussion of the proposed taxonomy follows. (Rehman et al., 2013). After that, uppercase letters are
transformed into lowercase letters to homogenize the words. assignment of a mean vector (centroid) to every cluster
Next, one-letter tokens and stop-words are excluded from the makes this algorithm especially useful for document cluster-
list of tokens. These exclusions are non-informative tokens ing because the mean vector can easily be interpreted and
such as articles, pronouns and adverbs (Fatudimu et al., applied by human reasoning (MacQueen, 1967).
2008). To enhance the meaningfulness of the individual
tokens, a Porter-Stemmer is used (Porter, 2006). By reducing Clustering process and evaluation of the different cluster models
tokens to their common stem, the total number of different We calculate 15 different models with numbers of clusters
tokens is reduced, which also reduces computational com- from 2 to 16. There is no commonly agreed evaluation
plexity. The advantage of this algorithm is context awareness technique or measure for determining the most suitable num-
and simplicity of the suffix removal rules (Willett, 2006). In ber of clusters and clustering model (Aliguliyev, 2009).
the last step, we calculate TF-IDF scores for every token Clustering always requires both quantitative (e.g., perfor-
(Jones, 1972), and every token is assigned a number that mance measures) and qualitative (e.g., human reasoning or
determines its weight in the document (term frequency) and expert knowledge) criteria to determine the true number of
in the overall corpus of documents (inverse document clusters and to evaluate the found patterns (Frank, 1994).
frequency). Therefore, we apply both qualitative criteria and the X-means
Following the pre-processing, we apply a clustering algorithm to find the optimal number of clusters (Bouwman,
algorithm. Clustering is the assignment of objects to groups 1983; Pelleg and Moore, 2000).
with high intra-similarity and low inter-similarity; many Table 3 shows an excerpt of the results of the clustering
algorithms exist to solve this optimization problem, each algorithm for the optimal cluster number of four. The
having its own specific field of application (Tabrizi et al., table presents the twelve top scoring tokens for each cluster
2013). The LRs are expressed in natural unstructured together with the corresponding cosine similarity (TF-IDF
language, which has important consequences for usable value). In comparison with the other models developed, the
algorithms and measures because LRs have different lengths reported cluster model with four clusters appears to be the
and content structures (Losiewicz et al., 2000). Therefore, most suitable from a qualitative point of view. Different from
algorithms must be used that can process high-dimensional the models with fewer clusters, there is an information gain
datasets and normalize the documents to avoid ambiguous by adding up to four clusters to describe the dataset.
results (Ye, 2011). For the task of finding patterns in high- Furthermore, four categories of market manipulation are
dimensional datasets, the k-means algorithm combined with revealed: financial instrument manipulation, accounting fraud,
a suitable similarity measure is the algorithm of choice investment fraud and insider trading. Each fraudulent activity
(Willett, 1988). The k-means algorithm divides a population is characterized by the respective relevant keywords that
of N-Dimensional Objects (here LRs) into k sets by appear most often within a cluster as shown in Table 3.
optimizing the within-class variance and calculating the Because the tokens ‘‘stock’’ and ‘‘manipul’’ are of high
means of the attributes (centroids) within the cluster importance for the first cluster, it is labeled the cluster of
(MacQueen, 1967). Practically, k random vectors are created, financial instrument manipulation (Allen and Gale, 1992).
and the similarity to the objects is measured, with the most The other tokens in this cluster fit to financial instrument
similar objects being grouped together. This process is manipulation, e.g., penny stocks are common targets of
repeated until the within-class variance is optimized or the manipulative actions (see token ‘‘penni’’) (Bartels, 2000), and
predefined maximum number of iterations is reached. The several manipulation techniques such as ‘‘Pump and Dump’’
depend upon the Internet (see token ‘‘internet’’) to dissem-
Table 1 Excluded documents from the raw litigation release database inate false messages (Sabherwal et al., 2011). The second
cluster is determined to represent accounting fraud because it
Cleansing measure Number of excluded clearly includes LRs that cover the misstatement or over-
documents statement of accounting figures such as revenue (see token
Filter 1995,1996 and 2013 722 ‘‘revenu’’) or financial figures (see token ‘‘financi’’) (Gerety
No 10b-5 reference 1845 and Lehn, 1997). This statement also holds true for the other
Filter Ponzi schemes 632 tokens in this cluster such as ‘‘audit,’’ ‘‘fiscal’’ and ‘‘report.’’
Filter scandals (e.g., Enron, 554 Cluster 3 covers manipulation with respect to funds of
Worldcom) investors (see token ‘‘fund’’ and ‘‘investor’’) and is named
Final database 4340 investment fraud cluster (Carroll, 2006). Investment fraud is
regularly conducted by asset managers or financial advisors
Table 3 Result of the clustering process with four centroid vectors (k = 4) showing tokens, corresponding ranks, TF-IDF scores and a representative litigation release
for each cluster
(see tokens ‘‘asset,’’ ‘‘manag’’ and ‘‘advis’’) using the legal form and assigns scores based on the Bayesian Information
of a limited liability company (see token ‘‘llc’’) and deceiving Criterion to every model (Pelleg and Moore, 2000). When
clients by disbursing false returns (see tokens ‘‘client’’ and a score is computed for every model in the predefined range,
‘‘return’’) (Ionescu, 2010). The last cluster indicates that the the model with the best score determines the optimal number
most relevant technique for the cluster is insider trading (see of clusters to partition the dataset from an information
tokens ‘‘insid’’ and ‘‘trade’’) (Roddenberry and Bacon, 2011). criterion point of view. The quantitative approach supports
Insider trading often occurs around mergers and acquisitions the chosen model of four clusters.
(see tokens ‘‘merger’’ and ‘‘acquisit’’); the manipulator tries to In addition to the four clusters of financial market
profit from non-public information (see tokens ‘‘profit’’ and manipulations that we identified based on SEC LRs, a deeper
‘‘nonpubl’’) (Allen and Ramanan, 1995). analysis of the centroid vector for the cluster of financial
Three tokens, i.e., ‘‘investor,’’ ‘‘stock’’ and ‘‘share,’’ appear instrument manipulations reveals more fine-grained manip-
in two of the four clusters. However, this appearance does ulative actions such as ‘‘Pump and Dump’’ (indicated by the
not harm the validity of the clustering process because both token ‘‘pump’’ on rank 47 of the centroid vector). A more
financial instrument manipulation and investment fraud are detailed analysis of different manipulation techniques is
conducted either by investors or to deceive investors. provided in the ‘‘European market abuse regulation and
Likewise, financial instrument manipulation aims at the literature review’’ section to enhance and complete the
manipulation of stock, respectively, share prices while market taxonomy as described in the ‘‘Developing a taxonomy of
participants conducting insider trading try to benefit from financial market manipulations’’ section.
changes in stock (share) prices based on non-public
information.
Because these four clusters are meaningfully distinguish- European market abuse regulation and literature review
able, a minimum number of four clusters appears appropri-
ate. By adding one more cluster, the investment fraud cluster Analysis of European regulation
is split. This split, however, does not help to describe the With our taxonomy of financial market manipulations, we
dataset better but instead creates fuzzy assignments because want to provide a comprehensive, multi-country overview
five of the twelve tokens assigned to cluster three (see on fraudulent behavior in financial markets. Therefore, we
Table 3) are also relevant for the new cluster of the model, rely on not only US legislation but also inclusion of
making it difficult to find a clear and distinguishable label for manipulation techniques that are considered in the MAR.
both clusters. This finding and the analysis of the rest of the By examining both US and European jurisdictions, we have
centroid vectors (six to fifteen) leads to the conclusion that considered all manipulations in the two largest financial
k = 4 is the optimal number of clusters from a qualitative markets. Different from the LRs, no in-depth textual
information gain standpoint (Bouwman, 1983). analysis is necessary for the MAR because manipulation
After having analyzed the optimal cluster number from a techniques can be extracted from the articles defined in the
qualitative standpoint, a k-means algorithm is applied as a regulation.
quantitative measure. Specifically, we apply the X-means The MAR is intended to ensure the integrity and smooth
algorithm that computes all cluster models (similar to the k- functioning of European financial markets with the goal of
means approach) from a predefined upper to lower boundary increasing investor confidence (European Parliament and
A taxonomy of financial market manipulations M. Siering et al
Council, 2014). The regulation precisely describes different manipulation techniques mentioned in at least two sources,
activities that are considered manipulative behavior in and we categorized the different fraudulent activities
Articles 8 and 12. Furthermore, we determined abusive according to clusters based on the analysis of SEC LRs.
activities related to market benchmarks from the European
Benchmark Regulation, effective as of 2016 (European
Parliament and Council, 2016). The list of market manip- Developing a taxonomy of financial market manipulations
ulations extracted from the MAR and the European Bench-
mark Regulation is presented together with the additional Taxonomy development methodology
manipulation techniques found in academic studies in Table 5 Developing a suitable taxonomy is important for almost
(see next section). every discipline to understand the underlying mechanics and
to break down complex topics into smaller parts (Miller and
Literature review of financial market manipulations Roth, 1994). The emphasis when developing a taxonomy of
In addition to incorporating regulation and prosecution of manipulation techniques should be on awareness of the
market manipulations, we conduct a systematic literature effects on the market and the identification of touch points
review to identify manipulations that might not yet be with the market environment (Kumar and Langberg, 2009).
considered in regulatory documents. This emphasis leads to the development of effective coun-
A systematic literature review shall provide a compre- termeasures or detection techniques that help to make
hensive view on the topic in question (Denney and financial markets more efficient and less likely to be
Tewksbury, 2013) and exhaustively include all related manipulated (Cumming and Johan, 2008). To achieve this
studies to reveal the current state of knowledge in the goal, comprehensibility and extendibility are highly impor-
research field under investigation (Rowley and Slack, 2004). tant so that the taxonomy can be adapted to changing
In this context, an effective literature search relies on market conditions and new manipulation techniques (Ngai
meaningful keywords and backward and forward reference et al., 2011).
searches (Levy and Ellis, 2006) that are especially fruitful To the knowledge of the authors, no concise, robust,
because the keyword search is rather unlikely to fully yield comprehensive, extendible and explanatory taxonomy of
all relevant literature concerning the topic (Webster and manipulation techniques can be found in the financial
Watson, 2002). market manipulation literature (Nickerson et al., 2013; Kyle
We applied both techniques to find all manipulation and Viswanathan, 2008). There only exist simple taxonomies
techniques that are addressed in academic research. We for market manipulations. For instance, Allen and Gale
searched in peer-reviewed journals in reputable databases for (1992) distinguished action-, information- and trade-based
scientific publications from 1990 onwards, as shown in manipulation.
Table 4. Keywords in our search were ‘‘stock market Our taxonomy development is based on the method
manipulation’’, ‘‘stock market fraud’’, ‘‘financial market proposed by Nickerson et al. (2013). Nickerson et al. (2013:
manipulation’’ and ‘‘financial market fraud’’. Applying back- 338) define taxonomies as ‘‘[…] systems for grouping objects
ward and forward reference searches, we manually added of interest in a domain based on common characteristics.’’
additional relevant publications not found in the keywords The algorithm follows an iterative approach. Objects are
search. discovered and grouped according to defined characteristics
Table 5 presents all of the financial market manipulations and dimensions.
that we identified based on SEC LRs, the European MAR For the taxonomy, the objects are the manipulation
and the systematic literature review. We only included techniques, and the characteristics should be related to our
meta-characteristic, the market environment, also encom-
passing the interaction of the manipulation with the market
Table 4 Results of the systematic literature review environment. This interaction involves both market partic-
ipants (issuers, intermediaries and investors) and market
Database Number of Publications explicitly information such as prices and trading volumes (Ledgerwood
hits in the covering financial and Carpenter, 2012; Hellwig, 1980).
database market manipulations The algorithm starts with the most comprehensive char-
acteristic, and each (sub) characteristic should be a logical
ABI/INFORM database 125 3
consequence (Nickerson et al., 2013). After each iteration, a
ProQuest
taxonomy table (see Table 6) and a taxonomy formula [see
ACM 35 3
Eq. (1)] are derived until the ending conditions are met.
AIS library 14 0
Business Source Premier 252 19 Tm ¼ Di ; i ¼ 1; . . .; njDi ¼ Cij ; j ¼ 1; . . .; ki ; ki 2
Emerald Fulltext 18 6
IEEE Xplore Digital 3 2 ð1Þ
Library Tm is the taxonomy after iteration m with a set of n
Science Direct 43 14 dimensions Di, each consisting of ki mutually exclusive and
Springer-Link journals 53 4 collectively exhaustive characteristics Cij. The markers (‘‘X’’)
JSTOR 65 6 in the table indicate the classification of the objects. The
Manually added 17 17 exemplary taxonomy shown in Table 6 is described by the
Total 625 74 following Eq. (2):
A taxonomy of financial market manipulations M. Siering et al
Table 5 Continued
Table 5 Continued
Table 6 Example for a taxonomy with two objects and two classes; the ‘‘X’’ or dimensions have been added/split/merged (objective
indicates the assignment (Nickerson et al., 2013) condition) (for a detailed set of possible combinations, see
Objects Dimension1 Nickerson et al., 2013). For developing the taxonomy, we
initially start with the four categories determined by the
Characteristic1 Characteristic2 classification of SEC LRs (see Table 3). Then, we integrate the
financial market manipulations found in Table 5. In each
Object1 X iteration, either the empirical-to-conceptual or the concep-
Object2 X tual-to-empirical approach can be followed to identify
characteristics and objects. If significant domain understand-
ing is available, the conceptual-to-empirical approach is
T1 ¼ fDimension1 ½Characteristic1 ; Characteristic2 g ð2Þ chosen (first, identification of new characteristics; thereafter,
determination of appropriate objects). In contrast, if signif-
Several iterations are accomplished to find possible icant data on the objects are available, the empirical-to-
dimensions that describe the manipulation techniques in conceptual approach is selected (first, identifying a set of
more detail until the ending conditions are met. The method objects; thereafter, identification of common characteristics).
ends when both subjective and objective conditions have In the following, we outline the different iterations of our
been met: (1) the taxonomy must be concise, robust, taxonomy development process (due to space constraints, the
comprehensive, extendible and explanatory (subjective con- identified objects are shown in the final result of the
ditions); and (2) in the last iteration, no (new) characteristics taxonomy development process).
A taxonomy of financial market manipulations M. Siering et al
manipulator. The subjective ending conditions of taxonomy market manipulations that can be used for the configuration
development are fully met, but in this iteration, new dimensions of an automated market surveillance system to detect
were added; therefore, a new iteration is necessary: manipulative actions.
T4 ¼ fMeans ½Action based; Information based; Trade based;
Order based; Manipulator ½Issuer; Intermediary; Investor; Evaluation of the taxonomy
Manipulation target ½Bid=Ask; Price; Volume;
Fundamental data; Commissions; Latency; Evaluation following Nickerson et al.
This section tests and discusses the developed taxonomy to
Economically reasonable ½Yes; No;
assess the contribution to the financial market manipulation
Specific target direction ½Up; Down; Both possible; Nog literature and the overall understanding of manipulations.
ð6Þ First, because the objective criterion is already fulfilled, the
subjective (qualitative) criteria are tested. Then, the practical
applicability and overall usefulness will be addressed.
Iteration 5 Conciseness Nickerson et al. (2013) stated that an overly
For this iteration, the conceptual-to-empirical approach was broad taxonomy is a weak one. Because the cognitive capacity
again chosen. The MAR states in article 12 (2)(a) that specific of the decision maker is naturally limited, a taxonomy that
times of the day can be used as an indicator for the covers every characteristic and object possible is not concise
occurrence of certain manipulation strategies. Therefore, we (Nickerson et al., 2013). The taxonomy developed above does
include the dimension specific point in time to distinguish not suffer from this issue, because all characteristics are
manipulation techniques that depend upon specific points in documented in the literature, and most can be found in actual
time to be accomplished, e.g., ‘‘Marking the Close,’’ which is LRs. Eight dimensions and 25 characteristics are also unlikely
only feasible at the end of the trading day. Additionally, the to exceed the decision makers’ cognitive capacity (Carroll,
MAR suggests that some fraudulent activities depend upon 1993; Miller, 1956).
the collusion of market participants. Therefore, we incorpo- Robustness A robust taxonomy can clearly distinguish
rate the dimension collusion in the taxonomy. ‘‘Benchmark objects from one another via characteristics and dimensions;
Manipulation’’ with and without official fixing is the prime a small number of dimensions can hardly achieve this goal
example of such a fraudulent activity. Furthermore, MAR (Nickerson et al., 2013). This taxonomy is able to distinguish
recital 38 points out that specific abusive strategies can be clearly two objects from one another; for example, the two
performed by algorithmic traders and, in particular, HFTs techniques ‘‘Pump and Dump’’ and ‘‘Scalping’’ are hard to
using low-latency infrastructure. Consequently, we added the distinguish due to their similar appearance. The dimension
dimension low latency. Examples of such manipulative manipulator, however, helps to distinguish the two clearly
strategies are ‘‘Quote Stuffing’’ and ‘‘Layering.’’ Due to three because ‘‘Pump and Dump’’ can be accomplished by
new dimensions, an additional iteration is necessary. The investors, whereas ‘‘Scalping’’ is conducted by intermediaries.
iteration results in the following taxonomy: Comprehensiveness A taxonomy that is developed both
conceptually and empirically, such as this one, should cover all
T5 ¼ fMeans ½Action based; Information based; Trade based;
dimensions and all objects possible to be viewed as compre-
Order based; Manipulator ½Issuer; Intermediary; Investor; hensible (Nickerson et al., 2013). An extensive literature review
Manipulation target ½Bid=Ask; Price; Volume; and the use of EU legislation and market manipulations sued
Fundamental data; Commissions; Latency; by the SEC in the USA ensure that all objects (manipulation
techniques) currently considered relevant and all dimensions
Economically reasonable ½Yes; No; that could possibly characterize those techniques are covered.
Specific target direction ½Up; Down; Both possible; No; Extendibility A taxonomy should also be applicable in the
Specific time ½Yes; No; Collusion½Yes; No; future, when new objects or dimensions extend the domain
for which the taxonomy was developed (Nickerson et al.,
Low latency ½Yes; Nog
2013). This criterion is hard to test because knowledge about
ð7Þ future forms of manipulation is only vague. However, the
taxonomy can be extended due to technological enhance-
ments because it is already capable of including manipulation
Iteration 6 techniques associated with HFT, which represents a recent
An empirical-to-conceptual approach cannot be pursued in major technological trend.
this iteration because there are no further empirical findings Explanatory power According to Nickerson et al. (2013), a
relevant for extending the taxonomy. Furthermore, no taxonomy should help to understand the domain without
important additional concepts remain; thus, nor can a describing every object in detail. The developed taxonomy
conceptual-to-empirical approach be applied. All ending helps to understand the anatomy of market manipulations and
conditions, subjective and objective, are met, and the shows which manipulation techniques should be considered by
taxonomy is provided by T5 (iteration 5). Therefore, the regulatory authorities. Nevertheless, the explanatory power is
final taxonomy of market manipulations in interaction with not as strong for some objects as for others. For example, the
the environment of the manipulated market is shown in difference between ‘‘Matched Orders’’ and ‘‘Wash Sales’’ is only
Table 7. Based on the methodology used by Nickerson et al. identifiable based on whether collusion of market participants
(2013), we derived eight dimensions for our taxonomy of is necessary.
Table 7 Taxonomy of financial market manipulations
Advancing the X X X
bid
Reducing the ask X X X
Matched orders X X X
Painting the tape X X X
Wash sales X X X
M. Siering et al
Capping X X X
(pegging)
Marking the X X X
close/open
(banging the
close)
Front running X X X
Churning X X X
Scalping X X X
Spoofing X X X
Pinging X X X
Quote stuffing X X X
Ramping X X X
Layering (order X X X
book fade)
Table 7 continued
Wash sales X X X X X
Capping (pegging) X X X X X
Marking the close/open (banging X X X X X
the close)
Front running X X X X X
Churning X X X X X
M. Siering et al
Scalping X X X X X
Spoofing X X X X X
Pinging X X X X X
Quote stuffing X X X X X
Ramping X X X X X
Layering (order book fade) X X X X X
A taxonomy of financial market manipulations M. Siering et al
Application of the taxonomy in decision support system Both studies confirm the configuration as derived from the
development taxonomy, and both studies consider trading data, including
To show the applicability of the taxonomy for DSS config- stock prices. Furthermore, the studies consider a specific point
uration, we focus on ‘‘Marking the Close’’ market manipu- in time, i.e., the closing price. Because the closing price might
lations. We assume that the taxonomy is appropriate for be manipulated in both directions depending upon the
finding a proper DSS configuration if the specific configu- manipulator’s intention, the direction is not directly consid-
ration (i.e., most importantly, the data sources considered) ered, but a deviation from a benchmark is considered.
leads to a successful system for fraud detection. Therefore, Because neither study has access to trading data, including
the ‘‘Marking the Close’’ manipulation technique is suit- the market participant trading the financial instrument, they
able for investigating the appropriateness of the taxonomy are not able to consider whether the investor performs a
because different previous studies have proposed systems for potential market manipulation. Finally, both studies could
detecting manipulations in which the closing price of a also be improved if – in suspicious cases – the economic
financial instrument is manipulated. reasonability of trades is assessed to identify investors who
For marking the close, the means of market manipulation is have significant positions in the financial instrument under
trade based. In other words, by means of specific trades, the investigation.
closing price is manipulated (dimension: manipulation Consequently, the application of the taxonomy for DSS
target). Consequently, an appropriate DSS should analyze configuration (including the comparison with previous
trading data including stock prices for manipulation detec- studies) shows that the taxonomy provides very useful
tion. Because the closing price can be manipulated either guidelines on the specific design of a DSS; it corresponds to
upwards or downwards, depending upon the specific goal of the design of existing systems and provides insights into
the manipulator, both directions of manipulation are possi- potential system enhancements.
ble. Thus, a DSS should consider whether the suspicious
stock price consistently moves in one direction. In addition,
the manipulation focuses on a specific point in time, which Discussion
shows that in this case, closing prices should be considered by As shown by the evaluation of the taxonomy according to the
DSSs. Nevertheless, no specific low-latency infrastructure is criteria proposed by Nickerson et al. (2013) and by the
necessary to pursue this manipulation technique, so the application of the taxonomy for DSS configuration, the
utilization of such infrastructures cannot be considered a cue proposed taxonomy is valid and very useful in the field of
for manipulation detection. classification and detection of market manipulations. It thus
Furthermore, focusing on the manipulator performing the helps to distinguish different forms of market manipulations
manipulation, here the investor tries to perform marking the and to identify similarities between diverse types of manip-
close manipulations. This might be attempted, e.g., to avoid ulative behavior. It also indicates clearly which information
margin calls, to support a flagging price or to affect the an automated fraud detection system must incorporate. Due
valuation of a fund’s portfolio at the end of a specific relevant to the comprehensive sources upon which the taxonomy is
period (e.g., quarter). Specifically, data identifying whether built (cases actually prosecuted by a market surveillance
an investor has an incentive because of his position in the authority, cases reported in the literature and cases extracted
instrument might be helpful to identify related information. from financial market manipulation regulations), the taxon-
Nevertheless, such data are only available for market omy can be assumed to offer a broad overview on market
surveillance authorities and should therefore be considered manipulations. Furthermore, the different dimensions
within their specific fraud detection systems. Concerning the directly allow drawing conclusions on the specific configu-
collusion dimension, note that no collusive behavior is ration of fraud detection systems. Consequently, the pro-
necessary for ‘‘Marking the Close’’; therefore, this dimension posed taxonomy can support market surveillance authorities
should not be considered in a DSS. in detecting securities fraud to ensure fair and efficient
Finally, the taxonomy also shows that the manipulative financial markets and a level playing field for market
behavior is not economically reasonable, i.e., in the case of a participants, which is in the interests of all stakeholders,
marking the close manipulation, there should be other gains particularly in the light of the financialization debate.
for the investor to compensate him for performing this The distribution of the characteristics assigned to objects is
costly activity, e.g., a higher valuation that is applied to not even for all dimensions of the taxonomy. In particular,
reveal a higher own portfolio performance of a fund the means of manipulation, the manipulator and the collusion
(window dressing). Therefore, data that allow crosschecking dimension place a great deal of weight on the characteristics
positions of the respective investors should be used as trade based, investor and no collusion necessary. This result
additional input. occurs because many financial instrument manipulations
Focusing on previous research to confirm whether the have these characteristics in common. Nevertheless, these
proposed system configuration is actually implemented, we dimensions are necessary because they represent key infor-
consider the studies by Öğüt et al. (2009) and Kim and Sohn mation for regulators and automated fraud detection systems
(2012). Öğüt et al. (2009) consider average daily returns and trying to detect market manipulation and to differentiate
other daily trading statistics as input variables to detect clearly between different manipulation techniques.
‘‘Marking the Close’’ by means of different machine learning We are aware that fraudulent market participants might
techniques. Additionally, Kim and Sohn (2012) consider also use the insights of the proposed taxonomy to circumvent
daily trading data encompassing closing prices to perform a the risk of being detected by an automated fraud detection
peer group analysis to detect suspicious behavior. system. Nevertheless, the taxonomy provides general insights
A taxonomy of financial market manipulations M. Siering et al
into DSS configuration that also apply in the case of changed specifically considering the manipulation techniques actually
manipulator behavior. In this case, an appropriate DSS must performed and prosecuted and by specifically considering
be recalibrated to detect the altered manipulative behavior. manipulation techniques caused by novel technological
The current study also has several limitations. The taxon- developments such as HFT. Therefore, our study also
omy provides implications for how to configure DSSs, contributes to the call by Ngai et al. (2011), who find that
particularly in terms of the input data to be considered. there is a lack of research in the field of securities fraud
However, we are aware that the taxonomy does not itself detection; the study shows how the taxonomy can be used as
provide insights into how to process such potential input data a foundation for DSSs helping to detect fraud.
in a DSS. For instance, diverse machine learning algorithms Our results are also relevant for practitioners, regulators
have been shown to lead to different performance figures. In and financial market participants. First, our study provides an
addition, from the specific point in time when a manipulation overview of current financial market manipulations and their
is conducted, e.g., in the case of ‘‘Marking the Close,’’ further characteristics. This classification can provide an important
aspects concerning the time passed for executing strategies means for increasing the financial literacy of both institutional
can hardly be considered within the taxonomy because and private investors by showing them potential market
manipulators might chose different intervals for their manip- manipulations and for generating awareness concerning
ulations. Therefore, DSS developers considering the taxon- suspicious market behavior. Based on this research, financial
omy should also perform different machine learning institutions and market surveillance authorities can develop
experiments to find the best-performing configuration. DSSs that help to identify market manipulations. Conse-
Finally, as already shown during the application of the quently, these organizations are enabled to enforce recent
taxonomy in DSS development, not every dimension of a market regulations to ensure fair and efficient financial
market manipulation can be detected automatically. Never- markets and thus to alleviate the negative effect of fraud on
theless, the taxonomy provides a useful base to develop DSSs the economy, particularly in the light of financialization. Our
that help to reduce information overload and to identify results are also of relevance for software vendors because the
cases that must be examined manually. Particularly against study provides indications on how to configure appropriate
the background of the increasing importance of financial DSSs to detect financial market manipulations. As also
markets in the context of financialization, IS should consider identified during the taxonomy development, HFT is accom-
the taxonomy because domain knowledge for detecting panied by novel manipulation techniques. Because our paper
market manipulations can help to increase trust in and the shows which type of DSS configuration might help to identify
integrity of financial markets. related fraudulent activities, future research might build upon
the taxonomy to develop specific deception detection systems.
Given the massive increase in regulation of financial
Conclusion markets and the banking industry as well as the need to
As long as financial markets have existed, fraudulent market provide efficient and low-cost technological support and
participants have tried to profit by manipulating markets and automation in this field, a novel future research stream ‘‘Reg-
deceiving others. Because market manipulations result in Tech’’ will be an interesting and highly relevant topic for
substantial losses that must be borne by market participants, further research at the intersection of financialization and IS.
understanding and detecting manipulations are of funda- Reg-Tech encompasses ‘‘the adoption of new technologies to
mental importance to establishing trust and market integrity – facilitate the delivery of regulatory requirements’’ (Financial
particularly against the background of financialization and the Conduct Authority (FCA), 2015). Our work is one step in
role of markets in financialization economies. Nevertheless, that direction; however, additional future research in the
until now, different terms have been used even for the same fields of, for example, artificial intelligence, big data analysis
market manipulation techniques, and no unified terminology and text mining techniques might provide very useful input
prevails, which also hampers proper fraud detection. for investment firms, regulators, digital solution providers
Within this study, we follow a multi-method approach to and consultancy houses to fulfill ever-growing regulatory
develop a taxonomy of financial market manipulations. In requirements in banking and financial markets.
particular, we consider officially prosecuted current market
manipulations by performing a cluster analysis on SEC LRs.
We also perform a literature review and a regulatory review to References
identify market manipulations and their characteristics. On
that basis, we develop a novel taxonomy that covers existing Aalbers, M. (2016). Corporate financialization. In D. Richardson, N. Castree, M.
market manipulation techniques, including those based on F. Goodchild, A. Kobayashi, W. Liu, and R. A. Marston (Eds.), International
Encyclopedia of Geography: People, the Earth, Environment and Technology.
HFT. Applying eight dimensions, the taxonomy helps to
Oxford: Wiley.
distinguish clearly between different fraudulent activities and Abbasi, A., Zhang, Z., Zimbra, D., Chen, H., Nunamaker, Jr, and Jay, F. (2010).
shows which information must be considered by automated Detecting Fake Websites: The Contribution of Statistical Learning Theory. MIS
fraud detection systems. Based on the ‘‘Marking the Close’’ Quarterly, 34(3), 435–461.
manipulation, we successfully evaluate the taxonomy and show Abrantes-Metz, R. M., Kraten, M., Metz, A. D., and Seow, G. S. (2012). Libor
how to employ the taxonomy to build a fraud detection system. Manipulation? Journal of Banking & Finance, 36(1), 136–150.
Guided by this taxonomy, researchers and practitioners Aggarwal, R. K., and Wu, G. (2006). Stock Market Manipulations. Journal of
Business, 79(4), 1915–1953.
developing DSSs can find an appropriate system configuration. Akhigbe, A., Madura, J., and Martin, A. (2005). Accounting Contagion: The Case
We provide a solid understanding of the different manip- of Enron. Journal of Economics and Finance, 29(2), 187–202.
ulation techniques and their characteristics. We thereby Aliguliyev, R. M. (2009). Performance Evaluation of Density-Based Clustering
extend previous studies and contribute to the literature by Methods. Information Sciences, 179(20), 3583–3602.
A taxonomy of financial market manipulations M. Siering et al
Allen, F., and Gale, D. (1992). Stock-Price Manipulation. Review of Financial Financial Conduct Authority (FCA) (2015). Call for Input: Supporting the
Studies, 5(3), 503–529. Development and Adoption of RegTech [WWW document] https://www.fca.
Allen, F., Litov, L., and Mei, J. (2006). Large Investors, Price Manipulation, and org.uk/publication/call-for-input/regtech-call-for-input.pdf (accessed 12th
Limits to Arbitrage: An Anatomy of Market Corners. Review of Finance, 10(4), September 2016).
645–693. Financial Services Authority (FSA) (2012). FSA Final Notice: Reference
Allen, S., and Ramanan, R. (1995). Insider Trading, Earnings Changes, and Stock Number 122702 [WWW document] http://www.fsa.gov.uk/static/pubs/final/
Prices. Management Science, 41(4), 653–668. barclays-jun12.pdf (accessed 25th August 2016).
Arshadi, N. (1998). Insider Trading Liability and Enforcement Strategy. Financial Fischel, D. R., and Ross, D. J. (1991). Should the Law Prohibit ‘‘Manipulation’’ In
Management, 27(2), 70–84. Financial Markets? Harvard Law Review, 105(2), 503–553.
Bailey, K. D. (1984). A Three-Level Measurement Model. Quality & Quantity, Fouquau, J., and Spieser, P. K. (2015). Statistical Evidence About LIBOR
18(3), 225–245. Manipulation: A ‘‘Sherlock Holmes’’ Investigation. Journal of Banking &
Baker, N. (2005). Fraud and Artificial Intelligence. Internal Auditor, 62(1), 29–32. Finance, 50, 632–643.
Bartels, K. C. (2000). Click Here to Buy the Next Microsoft: The Penny Stock Frank, J. (1994). Artificial Intelligence and Intrusion Detection: Current and
Rules, Online Microcap Fraud, and the Unwary Investor. Indiana Law Journal, Future Directions, in 17th National Computer Security Conference; Baltimore,
75(1), 353–378. USA, 1994, pp. 1–12.
Biais, B. and Woolley, P. (2011). High frequency trading [WWW document] Gerety, M., and Lehn, K. (1997). The Causes and Consequences of Accounting
http://www.eifr.eu/files/file2220879.pdf (accessed 24th February 2016). Fraud. Managerial and Decision Economics, 18(7–8), 587–599.
Bonner, S. E., Palmrose, Z.-V., and Young, S. M. (1998). Fraud Type and Golmohammadi, K., Zaiane, O.R. and Diaz, D. (2014). Detecting Stock Market
Auditor Litigation: An Analysis of SEC Accounting and Auditing Enforcement Manipulation Using Supervised Learning Algorithms, in International Con-
Releases. The Accounting Review, 73(4), 503–532. ference on Data Science and Advanced Analytics (DSAA); Shanghai, China,
Bouwman, M. J. (1983). Human Diagnostic Reasoning by Computer: An 2014, pp. 435–441.
Illustration from Financial Analysis. Management Science, 29(6), 653–672. Gomber, P., Sagade, S., Theissen, E., Weber, M.C. and Westheide, C. (2016).
Carhart, M. M., Kaniel, R., Musto, D. K., and Reed, A. V. (2002). Leaning for the Competition Between Equity Markets: A Review of the Consolidation Versus
Tape: Evidence of Gaming Behavior in Equity Mutual Funds. The Journal of Fragmentation Debate, Journal of Economic Surveys (forthcoming).
Finance, 57(2), 661–693. Hanke, M., and Hauser, F. (2008). On the Effects of Stock Spam e-Mails. Journal
Carroll, J. B. (1993). Human Cognitive Abilities: A Survey of Factor-Analytic of Financial Markets, 11(1), 57–83.
Studies. Cambridge: Cambridge University Press. Hazen, T. L. (2010). Are Existing Stock Broker Standards Sufficient? Principles,
Carroll, B. (2006). How to Prevent Investment Adviser Fraud. Journal of Rules and Fiduciary Duties. Columbia Business Law Review, 2010(3), 710–761.
Accountancy, 201(1), 40–43. Hellwig, M. F. (1980). On the Aggregation of Information in Competitive
Cataldo, A. J., and Killough, L. N. (2003). Market Makers’ Methods of Stock Markets. Journal of Economic Theory, 22(3), 477–498.
Manipulation. Management Accounting Quarterly, 4(4), 10–13. Hillion, P., and Suominen, M. (2004). The Manipulation of Closing Prices.
Chaturvedula, C., Bang, N. P., Rastogi, N., and Kumar, S. (2015). Price Journal of Financial Markets, 7(4), 351–375.
Manipulation, Front Running and Bulk Trades: Evidence from India. Emerging Humpherys, S. L., Moffitt, K. C., Burns, M. B., Burgoon, J. K., and Felix, W. F.
Markets Review, 23, 26–45. (2011). Identification of Fraudulent Financial Statements Using Linguistic
Coffee, J. C. (2005). A Theory of Corporate Scandals: Why the USA and Europe Credibility Analysis. Decision Support Systems, 50(3), 585–594.
Differ. Oxford Review of Economic Policy, 21(2), 198–211. IIROC (2012). Proposed Guidance on Certain Manipulative and Deceptive
Comerton-Forde, C., and Putnins, T. J. (2011). Measuring Closing Price Trading Practices [WWW document] http://www.iiroc.ca/Documents/2012/
Manipulation. Journal of Financial Intermediation, 20(2), 135–158. f62c746a-b5c9-448a-b57f-f1c04c88de14_en.pdf (accessed 19th February 2016).
Cooper, D. J., and Donaldson, R. G. (1998). A Strategic Analysis of Corners and Ionescu, L. (2010). Madoff’s Fraudulent Financial Scheme, His Decades-Long
Squeezes. The Journal of Financial and Quantitative Analysis, 33(1), 117–137. Swindle, and the Failure of Operational Risk Management. Economics,
Cumming, D., and Johan, S. (2008). Global Market Surveillance. American Law Management, and Financial Markets, 5(3), 239–244.
and Economics Review, 10(2), 454–506. Jarrow, R. A. (1992). Market Manipulation, Bubbles, Corners, and Short
Cumming, D., Zhan, F. and Aitken, M. (2012). High Frequency Trading and Squeezes. Journal of Financial and Quantitative Analysis, 27(03), 311–336.
End-of-Day Manipulation [WWW document] https://www.legacy.wlu.ca/ Jones, K. S. (1972). A Statistical Interpretation of Term Specificity and Its
documents/54105/Cumming-Zhan-Aitken-15012013_2.pdf (accessed 24th Application in Retrieval. Journal of Documentation, 28(1), 11–21.
February 2016). Jones, M. J. (2011). Creative Accounting, Fraud And International Accounting
Davis, H. A. (2007). Summary of Selected FINRA Regulatory Notices. Journal of Scandals. Hoboken: Wiley.
Investment Compliance, 8(4), 60–67. Karz, G., and Wagner, W. H. (2006). Should I Fire My Trader or Pay Him A
Denney, A. S., and Tewksbury, R. (2013). How to Write a Literature Review. Million? The Journal of Trading, 1(4), 85–89.
Journal of Criminal Justice Education, 24(2), 218–234. Kim, Y., and Sohn, S. Y. (2012). Stock Fraud Detection Using Peer Group
Easley, D., Prado, De, Lopez, Marcos M., and O’Hara, M. (2011). The Analysis. Expert Systems with Applications, 39(10), 8986–8992.
Microstructure of the ‘‘Flash Crash’’: Flow Toxicity, Liquidity Crashes, Kirkos, E., Spathis, C., and Manolopoulos, Y. (2007). Data Mining Techniques
and the Probability of Informed Trading. Journal of Portfolio Management, for the Detection of Fraudulent Financial Statements. Expert Systems with
37(2), 118. Applications, 32(4), 995–1003.
European Parliament and Council (2014). Market Abuse Regulation No Klein, L. R., Dalko, V., and Wang, M. (2012). Regulating Competition in Stock
596/2014 [WWW document] http://eur-lex.europa.eu/legal-content/EN/TXT/ Markets: Antitrust Measures to Promote Fairness and Transparency Through
?uri=celex%3A32014R0596 (accessed 24th February 2016). Investor Protection and Crisis Prevention. Hoboken: Wiley.
European Parliament and Council (2016). Benchmark Regulation 2016/1011 Kumar, P., and Langberg, N. (2009). Corporate Fraud and Investment
[WWW document] http://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri= Distortions in Efficient Capital Markets. The Rand Journal of Economics, 40(1),
CELEX:32016R1011&from=EN (accessed 4th August 2016). 144–172.
Fama, E. F. (1970). Efficient Capital Markets: A Review of Theory and Empirical Kyle, A. S., and Viswanathan, S. (2008). How to Define Illegal Price
Work. The Journal of Finance, 25(2), 383–417. Manipulation. American Economic Review, 98(2), 274–279.
Fatudimu, I., Musa, A., Ayo, C., and Sofoluwe, A. B. (2008). Knowledge Lagoarde-Segot, T. (2016). Financialization: Towards a New Research Agenda,
Discovery in Online Repositories: A Text Mining Approach. European Journal International Review of Financial Analysis. doi:10.1016/j.irfa.2016.03.007.
of Scientific Research, 22(2), 241–250. Ledgerwood, S. D., and Carpenter, P. (2012). A Framework for the Analysis of
Fayyad, U., Piatetsky-Shapiro, G., and Smyth, P. (1996). From Data Mining to Market Manipulation. The Review of Law & Economics, 8(1), 253–295.
Knowledge Discovery in Databases. AI Magazine, 17(3), 37–54. Lee, E. J., Eom, K. S., and Park, K. S. (2013). Microstructure-Based Manipu-
Financial Conduct Authority (FCA) (2014). FCA Final Notice: Reference lation: Strategic Behavior and Performance of Spoofing Traders. Journal of
Number 124491 [WWW document] http://www.fca.org.uk/your-fca/ Financial Markets, 16(2), 227–252.
documents/final-notices/2014/jpmorgan-chase-bank (accessed 25th August Leinweber, D. J., and Madhavan, A. N. (2001). Three Hundred Years of Stock
2016). Market Manipulations. The Journal of Investing, 10(2), 7–16.
A taxonomy of financial market manipulations M. Siering et al
Lenard, M. J., and Alam, P. (2009). An Historical Perspective on Fraud Rowley, J., and Slack, F. (2004). Conducting a Literature Review. Management
Detection: From Bankruptcy Models to Most Effective Indicators of Fraud in Research News, 27(6), 31–39.
Recent Incidents. Journal of Forensic & Investigative Accounting, 1(1), 1–27. Sabherwal, S., Sarkar, S. K., and Zhang, Y. (2011). Do Internet Stock Message Boards
Levy, Y., and Ellis, T. J. (2006). A Systems Approach to Conduct an Effective Influence Trading? Evidence from Heavily Discussed Stocks with no Fundamental
Literature Review in Support of Information Systems Research. Informing News. Journal of Business Finance & Accounting, 38(9–10), 1209–1237.
Science: International Journal of an Emerging Transdiscipline, 9(1), 181–212. Scopino, G. (2015). The (Questionable) Legality Of High-Speed Pinging and
Lomnicka, E. (2001). Preventing and Controlling the Manipulation of Financial Front Running in the Futures Market. Connecticut Law Review, 47(3),
Markets: Towards a Definition of ‘‘Market Manipulation’’. Journal of Financial 607–697.
Crime, 8(4), 297–304. Shapiro, S., Kinkela, K., and Harris, P. (2012). Churning and Suitability of
Losiewicz, P., Oard, D. W., and Kostoff, R. N. (2000). Textual Data Mining to Investments: A Financial Industry Regulatory Authority Arbitration Case
Support Science and Technology Management. Journal of Intelligent Informa- Study. Review of Business & Finance Case Studies, 3(1), 61–67.
tion Systems, 15(2), 99–119. Shleifer, A., and Vishny, R. W. (1997). A Survey of Corporate Governance. The
LR 17346 (2002). Litigation Release No. 17346: Securities and Exchange Journal of Finance, 52(2), 737–783.
Commission v. Patrick O. Wheeler, Steven S. Gallers, and Robert L. Carberry, Tabrizi, S. A., Shakery, A., Asadpour, M., Abbasi, M., and Tavallaie, M. A.
Case No. 02-60131-CIV-GRAHAM (S.D. Fla.) [WWW document] http://www. (2013). Personalized PageRank Clustering: A Graph Clustering Algorithm
sec.gov/litigation/litreleases/lr17346.htm (accessed 19th February 2016). Based on Random Walks. Physica A: Statistical Mechanics and its Applications,
LR 17645 (2002). Litigation Release No. 17645: Securities and Exchange 392(22), 5772–5785.
Commission v. Michael A. Ofstedahl, et al., United States District Court for the Thel, S. (1993). 850,000 in Six Minutes-The Mechanics of Securities Manipu-
Northern District of California, Civil Action No. C-02-3685 RS. [WWW lation. Cornell Law Review, 79(2), 219–298.
document] http://www.sec.gov/litigation/litreleases/lr17645.htm (accessed 19th Tripp, M. J. W. (1963). Securities Regulation: Stock Scalping by the Investment
February 2016). Adviser: Fraud or Legitimate Business Practice? California Law Review, 51(1),
LR 21423 (2010). Litigation Release No. 21423: Securities and Exchange 232–245.
Commission v. Frank J. Custable, Jr., et al., Civil Action No. 03-CV-2182 (N.D. U.S. Commodity Futures Trading Commission (CFTC) (2016). Education
Ill.) [WWW document] http://www.sec.gov/litigation/litreleases/2010/lr21423. Center: CFTC Glossary [WWW document] http://www.cftc.gov/
htm (accessed 19th February 2016). consumerprotection/educationcenter/cftcglossary/glossary_b (accessed 4th
LR 22545 (2012). Litigation Release No. 22545: SEC v. Berton M. Hochfeld et al., August 2016).
Civil Action No. 12-CV-8202 (S.D.N.Y.) [WWW document] http://www.sec. Vaughan, L. (2016). Broken Benchmarks: Six Years of Probes into Financial
gov/litigation/litreleases/2012/lr22545.htm (accessed 19th February 2016). Fiddling [WWW document] https://www.bloomberg.com/quicktake/broken-
MacQueen, J. (1967). Some Methods for Classification and Analysis of benchmarks (accessed 25th August 2016).
Multivariate Observations, in Fifth Berkeley Symposium on Mathematical Vila, J.-L. (1989). Simple Games of Market Manipulation. Economics Letters,
Statistics and Probability (Berkeley, USA, 1967), pp. 281–297. 29(1), 21–26.
Miller, G. A. (1956). The Magical Number Seven, Plus or Minus Two: Some Webster, J., and Watson, R. T. (2002). Analyzing the Past to Prepare for the
Limits on Our Capacity for Processing Information. Psychological Review, 63(2), Future: Writing a Literature Review. Management Information Systems Quar-
343–352. terly, 26(2), 3.
Miller, J. G., and Roth, A. V. (1994). A Taxonomy of Manufacturing Strategies. Wheeler, R., and Aitken, S. (2000). Multiple Algorithms for Fraud Detection.
Management Science, 40(3), 285–304. Knowledge-Based Systems, 13(2), 93–99.
Møllgaard, H. P. (1997). A Squeezer Round the Corner? Self-Regulation and Willett, P. (1988). Recent Trends in Hierarchic Document Clustering: A Critical
Forward Markets. The Economic Journal, 107(440), 104–112. Review. Information Processing and Management, 24(5), 577–597.
Ngai, E., Hu, Y., Wong, Y., Chen, Y., and Sun, X. (2011). The Application of Willett, P. (2006). The Porter Stemming Algorithm: Then and Now. Electronic
Data Mining Techniques in Financial Fraud Detection: A Classification Library and Information Systems, 40(3), 219–223.
Framework and an Academic Review of Literature. Decision Support Systems, Ye, J. (2011). Cosine Similarity Measures for Intuitionistic Fuzzy Sets and Their
50(3), 559–569. Applications. Mathematical and Computer Modelling, 53(1), 91–97.
Nickerson, R. C., Varshney, U., and Muntermann, J. (2013). A Method for Zahedi, F. M., Abbasi, A., and Chen, Yan. (2015). Fake-Website Detection Tools:
Taxonomy Development and Its Application in Information Systems. European Identifying Elements that Promote Individuals’ Use and Enhance Their
Journal of Information Systems, 22(3), 336–359. Performance. Journal of the Association for Information Systems, 16(6), 448–484.
Öğüt, H., Doğanay, M. M., and Aktaş, R. (2009). Detecting Stock-Price Zhou, L., Burgoon, J. K., Twitchell, D. P., Qin, T., Nunamaker, Jr, and Jay, F.
Manipulation in an Emerging Market: The Case of Turkey. Expert Systems with (2004). A Comparison of Classification Methods for Predicting Deception in
Applications, 36(9), 11944–11949. Computer-Mediated Communication. Journal of Management Information
Pagano, M., and Immordino, G. (2012). Corporate Fraud, Governance, and Systems, 20(4), 139–166.
Auditing. Review of Corporate Finance Studies, 1(1), 109–133.
Pelleg, D. and Moore, A. (2000). X-Means: Extending K-Means with Efficient
Estimation of the Number of Clusters, in 17th International Conference on
Machine Learning; Stanford, USA, 2000, pp. 727–734.
Porter, M. F. (2006). An Algorithm for Suffix Stripping. Electronic Library and
Information Systems, 40(3), 211–218. About the Authors
Pozza, C. L., Jr., Cox, T. R., and Morad, R. J. (2009). Review of Recent Investor
Issues in the Madoff, Standford and Forte Ponzi Scheme Cases. Journal of Michael Siering is a Postdoctoral Research Associate at
Business and Securities Law, 10, 113–132. Goethe University Frankfurt and works as a Business
Putnins, T. J. (2012). Market Manipulation: A Survey. Journal of Economic Consultant in the field of risk management. His research
Surveys, 26(5), 952–967.
Rapoport, N. B. (2012). Black Swans, Ostriches, and Ponzi Schemes. Golden Gate
focuses on decision support systems in electronic markets,
University Law Review, 42, 627–661. with a focus on the analysis of user generated content. His
Rehman, Z., Anwar, W., Bajwa, U. I., Xuan, W., Chaoying, Z., and Patterson, R. work has been published in Decision Support Systems and
L. (2013). Morpheme Matching Based Text Tokenization for a Scarce the Journal of Management Information Systems.
Resourced Language. PLoS ONE, 8(8), e68178.
Roddenberry, S., and Bacon, F. (2011). Insider Trading and Market Efficiency:
Do Insiders Buy Low and Sell High? Journal of Finance & Accountancy, 8, 1–15.
Benjamin Clapham ([email protected]) is a
Rokach, L. and Maimon, O. (2005). Clustering Methods, in O. Maimon and L. Graduate Researcher at Goethe University Frankfurt. He
Rokach, (eds.), Data Mining and Knowledge Discovery Handbook, New York: received his M.Sc. degree in Management with majors in
Springer US, pp. 321–352. Finance and Accounting from Goethe University. His
A taxonomy of financial market manipulations M. Siering et al
research interests include market microstructure, algorithmic Peter Gomber ([email protected]) holds the
trading, sentiment analysis, and financial market Chair of e-Finance at the Faculty of Economics and Business
manipulations. Administration, University of Frankfurt. His academic work
focuses on market microstructure and auction theory,
Oliver Engel ([email protected]) holds a M.Sc. degree in institutional trading, innovative concepts/technologies for
Management from Goethe University Frankfurt, with a electronic trading and post trading systems and information
specialization on e-finance. His research focuses on text systems in Finance.
mining and financial market manipulations.