Cryptocurrencies are a digital alternative to legal fiat money. instead of being issued by competent government authorities, their implementation is based on the principles of cryptography used to validate all transactions and generate new currency. every transaction that occurs is recorded in a public ledger. footnote 1
Blockchain, and more generally distributed ledgers, facilitate innovation across multiple domains of activity. these include but are not limited to supply chain management, data sharing, accounting, e-voting or, as a more prominent area, finance [see eg overview in casino et al. (2019)]. Although it is indisputable that the blockchain itself had and still has a great influence on public discourse, with an innovation potential comparable to that of the internet (since it promotes a decentralized infrastructure for economic transactions), financial experts remain generally skeptical. The implementation and features (including strictly technological ones) of blockchain technology, when proposed as a replacement for standard fiat currency, are subject to ongoing debate (Berentsen and Schär 2018; Dierksmeier and Seele 2018; Ertz and Boily 2019; Glaser and bezzenberger 2015). A major problem surrounding cryptocurrencies, but also one of the reasons why they have become known to the general public, is the heavy queues of their return distribution (chan et al. 2017) and their volatility (bariviera 2017 ), resulting in a rich history of “bubbles” (gerlach et al. 2018).
Reading: Bitcoin price manipulation
Although the innovative potential of distributed ledger technologies is enormous, innovation in itself does not necessarily translate into trust (see, for example, bodó 2021). traditional markets and exchanges have been quite successful in establishing a trustworthy environment through governmental or international institutions, strong legislative activity, market regulations, and effective monitoring/supervision systems. This development took many decades after a long history of market abuse (Putniņš 2012), and is still an area of active research. it can be said that each new case of market abuse brought a better understanding of market vulnerabilities and often led to viable countermeasures. furthermore, each new technology potentially brings new techniques to commit fraud. now, cryptocurrencies, crypto assets, and various forms of blockchain services are still in their infancy. therefore, new methods need to be invented or reinvented for this new medium to establish a trustworthy and fair market environment, ideally while maintaining the decentralized and (semi-)anonymous nature of the underlying blockchain technology.
With this motivation, we focus in this study on an example where the cryptocurrency market was allegedly manipulated through the fraudulent actions of a market participant. a data-driven model is developed and validated using historical data. the behavior of the fraudulent entity is investigated in detail and included in the model. Toward the end, we conclude our investigations with a discussion of how our findings can be applied to improve trust by reducing current vulnerabilities in crypto markets. In the remainder of this section, we will provide a brief overview of the cryptocurrency fraud study, agent-based modeling (especially in the context of crypto markets), and then highlight the specific contributions of this paper.
fraud and cryptocurrencies
See also: Crypto miners see "stranded" natural gas as a novel energy source – Marketplace
Several illicit activities are related to cryptocurrencies, such as black market trading (Foley et al. 2019), money laundering, and terrorist financing (Fletcher et al. 2021). footnote 2 in our case, we focus on fraud that targets and disrupts the market. a more common form of fraud in crypto markets is wash trading (cong et al. 2020; victor and weintraud 2021). The principle of wash trading is to execute trades where the buyer and seller are the same entity. therefore, false impressions of highly traded assets are created to mislead investors. Another more serious form of fraud seen in crypto markets is pump-and-dump schemes (kamps and kleinberg 2018), which generally take the form of coordinated actions to increase the market price in a short period of time (hamrick et al. 2019; li et al. 2018). In the literature we find several studies that try to explain the price as a direct consequence of manipulative behavior. a study (gandal et al. 2018) looking at suspicious market practices on the mt.cox exchange concludes that fraudulent actions influenced price growth from $150 to $1000 in late 2013. more recently, griffin and shams (2019 ) argue that the market price of bitcoin could have been inflated by the issuance of the leash.
As noted in a 2014 study (Robleh et al. 2014), bitcoin and other cryptocurrencies served as a medium of exchange for a relatively small number of people; therefore, they do not represent a serious material risk to monetary and financial stability, but today investors are increasingly involving crypto assets in their portfolios, and some large companies or payment services are already accepting payments in bitcoin. this means that the volatility of cryptocurrencies can potentially be a new source of systemic risk for the entire economy and the financial sector. Recent studies have addressed risk using methods such as clustering (eg li et al. 2021), multi-objective feature selection (eg kou et al. 2021), or network analysis (eg ., Anagnostou et al. 2018). focusing more on the source of systemic risk originating from illicit behavioral schemes, although progress is already being made in detecting laundering operations (victor and weintraud 2021) and pump-and-dump schemes ( Chen et al. 2019). taking place, new models are needed that can explain, simulate, or possibly predict the effects of fraudulent behavior, and that can serve as a test bed for the effectiveness of policies, regulations, or enforcement mechanisms. One way to meet this demand is to consider models that combine qualitative and quantitative knowledge, that can be designed with a strong reliance on empirical data, and that can simulate various scenarios to address issues related to the effectiveness of regulatory interventions in the crypto market, as discussed discusses in shanaev et al. (2020).
agent-based modeling
Agent-based models generally aim to explain some complex phenomena, in which macro-level emergent behavior is assumed to be a consequence of micro-level behavioral rules. For a historical review, we refer to Chen (2012). in recent years, this modeling paradigm has been enhanced with more modern data-driven approaches, where behavioral data specific to each agent is used to build, initialize, or estimate the parameters of a model of each agent’s decision mechanism. agent. only a relatively small number of parameters remain to be calibrated for the aggregate data, which increases the validity and credibility of the model. With this approach, even large-scale models are able to compete with the predictive power of traditional quantitative methods, for example in the area of economic research (poledna et al. 2019). These models can be particularly useful if individual agent parameters are of critical importance, for example, to test interventions during the covid-19 pandemic (Kerr et al. 2021).
In the literature, you can find several examples of agent-based models that have been created to gain insights into crypto markets. most of these models are based on various financial or behavioral assumptions. To our knowledge, the first study in this area is by Luther (2013), in which agents enter a foreign exchange market with exchange costs and network effects to investigate the widespread acceptance of cryptocurrencies. bornholdt and sneppen (2014) studied a similar question. An implicit assumption of demand was made in Cocco et al. (2017), enhanced by speculative traders and constrained by finite resources for each agent, and is the first example of a limit order book-based model of the bitcoin market that attempts to explain the price increase from early 2012 to April 2017. 2014. This model was later extended with mining (Cocco and Marchesi 2016) and evolutionary computation (Cocco et al. 2019). Other order book models are presented in Pyromallis and Szabo (2019) and Zhou et al. (2017), where the focus is mainly on the adaptive behavior of traders. in Lee et al. (2018), a combination of reverse reinforcement learning directly from bitcoin blockchain data and agent-based modeling of the order book was used to make short-term market price predictions. recently, models focused on policy recommendations have also been developed. shibano et al. (2020) is introducing a price stabilization agent to reduce volatility, and Bartolucci et al. (2020) investigates the extension of the bitcoin blockchain design to increase transaction efficiency.
An important aspect of agent-based models is that they provide an experimental environment for policymakers. once a pattern of behavior is identified, methods are established to measure and evaluate the consequences and the consequences are measured; the simulated environment can be used to test the effectiveness of certain measures, that is, a set of alternative policies to test, given a rate of adaptation, follow-up, compliance and identify the best one. In a recent review (López-Rojas and Axelsson 2016) agent-based models are considered a tool to generate synthetic data for machine learning models, which can be used, for example, to complement more traditional evaluation methods (Kou et al. al. 2014).
See also: In Crypto, Market Manipulation Remains a Problem | PYMNTS.com
In particular, agent-based models were developed in the area of urban crime modeling (Groff et al. 2019) or to study the behavioral aspects of tax evasion (Pickhardt and Prinz 2014). In principle, these models are not limited only to the observed fraudulent behavior: they can extend the design of the agents that commit fraud by considering different schemes of market manipulation methods to measure and evaluate the consequences. by choosing an appropriate representation of the fraud scheme, it is possible to find more sophisticated reasoning patterns for a fraud agent [for example, by applying algorithmic evolutionary methods (hemberg et al. 2016)].
contributions
Most studies focus on analyzing the statistical relationship between price and a set of exogenous variables. on the contrary, in this study we focus on the qualitative explanation dimension. our approach is based on the qualitative findings of griffin and shams (2019), but, unlike this study, we built a data-driven model, focusing primarily on the causal influence of fraudulent behavior that allegedly inflated bitcoin’s price. this methodological innovation can be considered as the main contribution of this study, together with the conceptualization of a specific fraud scheme as an algorithm that can be executed by an agent in a simulated cryptocurrency market. Note that this approach opens the door to a broader view of the role of the fraudulent trader in the bitcoin market, allowing the situation to be analyzed from various points of view. for example, since our market model is capable of generating market data such as market price, market volume, or fraudulent trader’s bitcoin inflow, it is possible to compare these quantities with empirical data. In particular, we found that certain market volume anomalies or market price declines can be attributed to the actions of a fraudulent trader, an experimental conclusion that complements the evidence presented in Griffin and Shams (2019).
In addition, the model developed in this study allows us to investigate the specific reasons behind the success of market manipulation through the fraud scheme. The connections between the efficiency of a specific manipulation strategy and transaction costs will be explored. to do so: a realistic order book liquidity model must be implemented. most studies implicitly or explicitly assume sufficient liquidity near the median price and an exponential decline in liquidity further away from the median price, using a Gaussian assumption or more relaxed forms. footnote 4 we propose a new liquidity distribution model based on a mixture of two components. the Gaussian assumption stays close to the mean price and the beta distribution is used to model the situation deeper into the order book.
the study of market manipulations (and their consequences) has a long tradition in the economic literature (putniņš 2012). To our knowledge, the present study is the first to build an agent that replicates the actions of a fraudulent trader directly using blockchain transaction data and reconstructing market behavior from this predictor. Furthermore, our simulation environment can be easily extended with more sophisticated AI models, contributing to the active area of research related to integrating AI with blockchain technology (pandl et al. 2020; salah et al. 2019 ).
Focusing on the economic study dimension of the paper, most of the assumptions we make to build the proposed computational model attempt to provide a solid story (based on previous studies analyzing the bitcoin market) with the aim of reconstructing the behavior of the market at a given time period. Our findings could challenge the view that the main predictors of the bitcoin bubble of late 2017 and early 2018 would be variables associated with market sentiment (see Kapar and Elm 2021). Although we do not deny that market sentiment plays an important role, our results confront the thesis that the appearance of this price bubble is spontaneous or a consequence of the widespread popularity of bitcoin. In this sense, we contribute to the ongoing discussion among economists about the price formation of cryptocurrencies.
See also: Bitcoin Superstar Test und Erfahrungen – Block-Builders.de
