SuperRare: The effect of peer trading on the performance of crypto artists

(1)

SuperRare: The effect of peer trading on the performance of crypto artists

Author Student Number Submission Date Programme

Institution Supervisor

Anja Krueger 13015540 25^th June 2021

MSc. Business Administration – Entrepreneurship and Management in Creative Industries

University of Amsterdam Monika Kackovic

(2)

Statement of Originality

This document is written by Student Anja Krueger, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are 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 the completion of the work, not for the contents.

(3)

Abstract

Crypto art is a significant emerging market that allows artists to tokenise and trade their works using blockchain without intermediaries. Despite the massive potential of this market, crypto art has attracted little attention as a research domain. This paper is the first to apply social network analysis to the ledger data of a crypto art market. The purpose of this research is to improve the understanding of the relationship between artist-to-artist trades and performance. This paper analyses three years of SuperRare's blockchain ledger using Python and Gephi prior to performing a multiple regression analysis. Thereby, this research helps understand which network measures best describe the relationship of peer trades on performance. The findings show that the effect of network position on performance differs among price and demand performance metrics.

This paper concludes that artists should try and avoid strong ties and take intermediary positions within the peer trade network, i.e., they should engage in plentiful (rather than large) transactions among peer groups to best utilise the social capital derived from their interactions. This study concludes that peer trading can significantly affect an artist's performance and suggest contextualising this using the signalling theory and motivates future research in this direction.

(6)

1. Introduction

Crypto art is an emerging market for artists to sell digital works online. The online art market grew 4% between 2020 and 2021 and currently amounts to 12.4bn USD, of which 338m USD utilised non-fungible tokens (NTFs) (Statista Research Department, 2021). According to Statista Research Department (2021), crypto art sales amounted to 12.9m USD in March 2021.

In addition, one of the largest crypto art markets, SuperRare, shows sales of 34,8m USD in March 2021.

This huge emerging market can be explained by blockchain disrupting the digital art industry. For the first time in art history, originality and scarcity apply to digital artworks (Burns, 2010; O’Dwyer, 2020; Zavelev, 2018). Using blockchain, artists can issue limited copies of rare digital artworks. Every artwork receives a token I.D., which compares to an artist's signature.

Due to blockchain technology's secure nature, a buyer can verify that he owns an original even when the artwork is digital (Franceschet, 2020).

Further, artists often face exploitative intermediary fees in the creative industries, which create entry barriers (Caves, 2003). However, on SuperRare, the artist receives 10% of the sales price of secondary trades immediately through blockchain without the need for intermediaries (Edlund, 2021). In addition, for every direct sale, the artist pays a 15% commission together with a gas fee for using Ethereum (ETH). Comparing this to the usual 50-50 split artists have with, e.g., galleries, the value of blockchain to artist-entrepreneurs is undeniable.

Regardless of these benefits and the drastic recent growth of crypto art sales (Wong, 2021), crypto art platforms are currently young ventures, and artists' performance in these ventures

(7)

still requires much research. Counterintuitively, the financial performance of an artist often does not root in artistic excellence or creativity (Banerjee & Ingram, 2018; Wachs et al., 2018).

Attempting to explain or predict artistic successes, the creative sectors often use signaling theory and concepts of the social capital research domain. The present research aims to extend these approaches to crypto art by observing the effect of peer interactions on performance. Even though it is known that nobody knows if artworks will become a success before their launch (Caves, 2000), it is also known that intermediaries and other third parties influence the prices of artworks by signaling legitimacy (Kackovic et al., 2020). How signaling functions in crypto art markets and how artists acquire and exploit social capital in online environments is still unknown.

Therefore, this research uses social network analysis (SNA) to assess artists' structural social capital within peer networks to describe how peer interactions affect an artist's performance in crypto art markets. This paper defines a peer network as a network for which all nodes are artists, and the only interactions that are traced are the ones between peers (artists). More specifically, this study focuses on the effect of inter-peer trading in crypto art markets on performance. Even though social capital is often researched using SNA, to date, no such research investigates the inter-peer relationships among artists on crypto art platforms. This illustrates a significant gap in the literature, which motivates the following problem statement:

To what extent do peer interactions influence an artist's performance in crypto art markets?

Because the performance metrics of this research are diverse, this research question extends to the investigation of whether the effect of peer trading on performance differs among different performance metrics. Next, the effect of peer interactions is quantified using SNA measures that correspond to positions within the peer trading network. Thereby, the research question extends to

(8)

another sub-question: "Which social network analysis metrics help understand the relationship of social capital on the performance of crypto artists better?".

By answering these questions, this paper contributes to the research in the domain of social capital and the signaling theory by extending the previous analyses and discussions to the fields of crypto art. The findings of this study contribute to the discussion on how artists signal legitimacy in crypto art environments. Moreover, by applying SNA to crypto art ledger data, this research hopes to motivate many more scholars to extend this field further. Moreover, it contributes to the bridging vs bonding debate of social network analysis literature by assessing the effect of different network positions within the peer trading network on performance. Beyond academia, these implications are relevant to management, investors, and artists concerning their strategic decision- making. Hence this paper utilises two theoretical lenses to explain the effect of plentiful vs large transactions and strong vs weak ties within the peer network.

The remainder of this paper is structured as follows: the Literature Review chapter presents relevant theories and proposes hypotheses concerning peer-trading performance. Subsequently, the Methods chapter addresses blockchain technology and the data processing as well as the network construction and the analytic strategy of the regression. Next, the Analysis presents the outcome of the regression analysis. Next, the Discussion puts those into context by addressing the contributions to signaling theory and bridging vs bonding debate, providing implications for practitioners, outlining limitations, and proposing channels for future research. Finally, the Conclusion summarises and concludes the study's contribution.

(9)

2. Literature Review

This chapter defines and discusses the relevant theories of this study in the context of performance metrics in the creative industries. First, it addresses the effect of social capital on performance and introduces SNA as a research tool for social capital. Then it elaborates on signalling theory and the effect of peer interactions on performance. Finally, the chapter ends with the definition of variables and the formulation of hypotheses.

2.1 Social Capital & Social Network Analysis

The following text focuses on theories of social capital relevant to this research and discusses social capital as a resource and its implications on performance metrics. Social capital directly impacts venture performance by providing access to information (Pryce & Birley, 1935) and is a common construct used in the literature concerning the resource-based view of value creation (Burt, 2002; Chisholm & Nielsen, 2009; Onyx & Bullen, 2000; Portes &

Sensenbrenner, 1993; Portes, 1998; Sandefur & Laumann, 1998; Tsai & Ghoshal, 1998).

There are three dimensions of social capital: structural, relational, and cognitive social capital (Klamer, 2004; Nahapiet & Ghoshal, 1998; Tsai & Ghoshal, 1998; Wang, J. &

Chiang, 2009) here structural social capital refers to the structure of social interactions and social ties, relational to assets rooted in relationships, and cognitive to a shared code or paradigm that facilitates a common understanding of collective goals (Portes & Sensenbrenner, 1993). These dimensions are interrelated in a sense that the structural dimension facilitates repeated social interactions, which lead to a common understanding of norms (cognitive dimension) which in turn provides the basis for trust, which is an element of relational social capital (Klamer, 2004; Wang

(10)

& Chiang, 2009). Among other effects, the dimensions of social capital have a positive association with resource sharing (Moran & Ghoshal, 1996) and innovation (Nahapiet &

Ghoshal, 1997; Tsai & Ghoshal, 1998).

This research focuses on the structural dimension of the social capital of artists its effects on performance by analysing the interactions between artists on the network. Because structural social capital is considered the foundation for shared cognition and relational capital, the analysis of this dimension provides most insights on potential resources arising from structural social capital. These resources include but are not limited to information sharing, access to buyers, and access to suppliers. This research focuses primarily on an individual artist's network position and performance and the SNA measures concerning the degree and degree centrality of an artist's ego networks.

Egocentric approaches to network and social capital theory analyse the position of an individual node and the implications of its network position. For example, t structural hole theory (Burt, 2002; Burt, 2009) proposes a definition for the frame of individual performance as a consequence of network positions. According to this theory, social capital is associated with realising the benefits of structural holes. In his paper from 1997, Burt defines structural holes as:

"Discontinuities between exchange relations (structural holes) are entrepreneurial opportunities to broker the flow of information between people on opposite sides of the structural hole and control the form of projects that bring together people on opposite sides of the structural hole. The Conclusion is that individuals with relations to otherwise disconnected social groups are positioned for entrepreneurial action, building bridges between groups where it is valuable to do so." (pp 355-356)

Due to the asymmetric information between buyers and sellers within an unregulated

(11)

as weak ties or bridges) in crypto networks, one expects to observe many structural holes in the network structure of crypto art markets. Because bridging ties provide access to novel information (Burt, 1983; Burt, 2009; Kogut & Zander, 1993; Kogut, 2000), one expects to observe more innovative activities and higher performance for individuals who access to a lot of diverse sources of information (low closeness centrality and weak ties). However, as discussed in the following subsection, the resource shared through peer trading also concerns a market signal.

The structural hole theory and the weak tie theory state that the stronger the tie between two nodes, the higher the likelihood of them having access to similar information (Granovetter, 2000), i.e., that the networks of both nodes overlap. Due to this overlap of information sources, homogeneous groups with high relational social capital are less innovative than heterogeneous groups with many bridges. Therefore, the bridging argument of this study builds on the notion that a node with a high betweenness centrality facilitates access to more diverse networks and information flows than a node with a high closeness centrality.

Contrary to Burt's bridging argument, Coleman (1988; 1994) argues in favour of cohesive networks to facilitate trust and cooperation. He finds that a central location in a network leads to increased bonding and, therefore, more relational and collective social capital. This argument builds on the idea that innovation and sharing of truly novel concepts require trust (Tsai &

Ghoshal, 1998). Hence, being central and having strong ties should have a positive impact on performance.

Discussions in prior literature concerning these two views belong to the bridging vs bonding debate (Kogut & Zander, 1993; Tan et al., 2015; Wachs et al., 2018). There are certain conditions under which closeness centrality has a positive effect on performance. However, in creative industries, the betweenness centrality usually significantly improves performance,

(12)

while closeness centrality decreases performance (Burt, 1983; Burt, 1992; Burt, 2002; Burt, 2009; Tan et al., 2015; Uzzi & Spiro, 2005; Wachs et al., 2018). For example, in a study by Banerjee and Ingram (2018), an artist who had influential networks that overlapped or were more homogenous were less likely to achieve wide-reaching fame, while artists with more diverse contacts were ultimately seen in the virtue of their cosmopolitan peer groups as possessing more creative identities and, as a result of this perception, achieved greater fame.

Because blockchain technology enables secure peer-to-peer transactions documented on a public ledger, the trust facilitated by close interactions is likely not the essential proceeds of social capital in crypto art markets. However, the information asymmetry between artists may imply that closeness centrality within an artist's peer network positively affects the signals sent through peer trades. Moreover, though financial interactions may result from social interactions, transactions do not necessarily facilitate knowledge and resource sharing, as do social interactions in most prior studies. Therefore, the second section of this chapter elaborates on market signals as the resources derived from the social capital of peer trades in crypto art markets.

To conclude this section, structural social capital in the context of this study is defined as 'the concept of the structural dimension of social capital to refer to the overall pattern of connections between actors' (Burt, 1992; Nahapiet & Ghoshal, 1998) and may be quantified through interactions. This study uses the ledger data of SuperRare to trace primary transactions between artists. Therefore, the focus lies on the structural social capital derived from peer transactions. The resources that result from these interactions are different to the resources that stem from social interactions. This paper argues that the resource shared through peer trading of crypto artists manifests in a signal to the market. The following justifies this argument by discusses

(13)

2.2 Signalling Theory in the Creative Industries

This subsection makes sense of resources artists share through financial interactions proposing legitimacy signals to the market as the primary resource. In that sense, it aims to understand conditions that could make peer trades more successful. In doing so, this section addresses whether the weight of peer transactions has a more significant impact on performance than the count of peer. Finally, this paper argues that artists are insiders of crypto art trading networks, and therefore, their purchase behaviour sends signals to the market. The following provides background on signaling theory.

According to Connelly (2011), the signalling theory builds the following components: information asymmetry, sending and receiving signals, signalling costs, and signalling environments. Information asymmetry refers to different parties having access to different information. Buying decisions are made dependent on the information available to consumers. Participants in a crypto art network have access to public information like the blockchain ledger or the information displayed on-site and private information which is not publicly shared. Because not everyone has access to private information, which could influence the purchase behaviour of art collectors, information asymmetries exist in crypto art markets.

This paper refers to insiders as participants of the network who have access to information that outsiders do not, and to a signal as the decision of an insider to communicate private information (Connelly et al., 2011); examples of private information in the context of crypto art markets may concern bidding strategies, windows of high/low trade, or the future curation and performance of upcoming artists. Therefore, artists are considered insiders of crypto art platforms as they obtain information from the artist network, which non-artist-buyers cannot access.

(14)

Regardless of how the signal is sent, the way a signal is received depends on the signalling environment of its recipient. Signalling environments are formed by the iterative process of sending, receiving, and interpreting the sending and receiving of signals. There may be multiple signals, signallers, and receivers within the same signalling environment; however, only those who have access to a signal's respective environment can decode it. Because shared understanding of norms and cognition forms through repeated interactions (Tsai & Ghoshal, 1998), this study uses structural social capital to quantify the foundation of shared understanding through repeated interactions, and thus, of the signalling environment.

There are at least two different contexts of signalling environments in crypto art – one decoding signals of financial insiders and one decoding signals of curators. Because interactions in the context of this research refer to transactions only, a signal will always refer to an insider (artist) communicating information about himself or other artists through buying crypto artworks.

Thus, a buyer's identity and bidding behaviour impact the market signals of peer trade artworks.

The argument regarding signalling within the bounds of this study is that artists engage in peer- trading because artists are considered insiders. Therefore, their purchasing behaviours send market signals that could increase the price/demand of/for their art.

Because the history of previous owners is visible on-site, the signals of peer-trading are permanent and traceable. Crypto art consumers report that the history of tokens owners significantly impacts the utility they derive from owning it (Franceschet et al., 2020; Zavelev, 2018). For this study, artists identify as insiders/signallers and buyers with non-artist identities as outsiders/receivers. However, individuals may also act as their own insiders when signalling about themselves (Connelly, 2011). Hence, artists buying the art of a peer send signals about themselves

(15)

as well. Together with the peer-network position, the nature and understanding of these signals are at the heart of this study.

2.3 Hypothesis Building

This section hypothesises relationships between network positions within the peer network and performance. This study uses SNA measures of centrality and degree to describe network positions and the nature of transactions (i.e. weight vs count). According to Al-Taie and Kadry (2017):

"Degree is a measure of popularity. It determines nodes that can quickly spread information to a localised area. Betweenness is based on the idea that a person is more important if he/she is more intermediary in the network. Closeness is a measure of reach; how fast information will spread to all other nodes from a single node?" (p.111)

First, the section hypothesises relationships concerning the centrality measures. Then, by relating bridging vs bonding to asymmetric information and signalling theory, it argues that many weak ties to different parts of the network provide access to information and diverse market signals and, therefore, increases artist performance. As stated above, the network position argument of this study builds on the notion that a node with a high betweenness centrality facilitates access to more diverse networks and information flows than a node with a high closeness centrality. As this is had a positive effect on performance, the first set of hypotheses is as follows:

H1. Artists who are in bridging positions outperform artists in bonding positions within the

peer network.

H1a. The closeness centrality of an artist within his peer network has a negative effect on performance

(16)

H1b. The betweenness centrality of an artist within his peer network has a positive

effect on performance

As a result of close and strong ties within a peer network, the ego network of an artist increases in density. In SNA, the network density of an ego network is described with the clustering coefficient measure. Because network density harms innovation performance (Tan et al., 2015) hypothesises as follows:

H1c. The clustering coefficient of an artist within his peer network a node has a

negative effect on performance.

Next, this section puts signalling theory into the context of peer trading in crypto art markets. The following paragraphs elaborate on two different peer-trading strategies: volume focused vs weight focused. Prior literature finds that the number and size of a review are more likely to influence the success of an artwork than the nature of a review (Gemser et al., 2007).

While a review entails more explicit signals than inter-peer trading in crypto art markets, it is possible to investigate how the count or the magnitude of peer transactions impact performance in crypto art markets. Of course, the audiences receiving market signals in crypto art markets differ.

However, as crypto art qualifies as high-brow digital art consumption (Grey, 2020), the volume and size of signals received by insiders likely influence and predict performance. Moreover, as an experience good, the volume of reviews is expected to advance performance to a greater extent than the weight of peer trades (Zhu & Zhang, 2006).

Regarding the signals peer trading sends to the market, we expect the count of peer trades to affect their performance positively. For once, artists who engage in plentiful transactions likely engage in multiple signalling environments. Moreover, because the money artists spend on peer

(17)

work is a limited resource, a high volume of peer trading should indicate weak rather than strong ties. Hence, the following hypothesis supports the assumption that artists aim for widely spread networks to increase the diversity of their information sources, signalling environments, and peer trading signals.

H2. The volume of peer trades an artist engages in has a significantly positive effect on

his/her performance.

The in-degree of a peer transaction network represents the number of artworks that an artist buys from another artist. The out-degree represents the number of works an artist sells to a peer.

The degree measure refers to the total number of trades. Thus, the following set of hypotheses:

H2a. The in-degree of an artist within his peer network has a positive effect on his

performance.

H2b. The out-degree of an artist within his peer network has a positive effect on

his performance.

H2c. The degree of an artist within his peer network has a positive effect on his

performance.

Next, to test for the effect of the size of a peer trade on performance, the weighted degree measures quantify the weight, i.e., ETH value of a transaction. Therefore, the final set of hypothesis follows.

H3. The size of peer trades an artist engages in has a significantly positive effect on his/her

performance.

H3a. The weighted in-degree of an artist within his peer network has a positive effect on his performance.

(18)

H3b. The weighted out-degree of an artist within his peer network has a positive

effect on his performance.

H3c. The weighted degree of an artist within his peer network has a positive effect on his performance.

(19)

3. Methodology

Chapter 3 discusses the research methodology and describes the research design, data collection, and processing. Subsequently, the empirical setting section describes blockchain application in creative industries and uses of SNA to assess participants' social capital in account- based blockchains. Next, the analytic strategy describes the independent and dependent variables of this study. Finally, the chapter closes by addressing the pre-processing and model building in R-Studio.

3.1 Research Design

As part of a research project, this study has access to the blockchain data of the renowned crypto space SuperRare. The available data covers a three-year window from the founding of the platform in 2018 until 2021. Therefore, in answering the research question, this study processes over 25,000 interactions. Then, it performs quantitative methods on data retrieved from the blockchain ledger of SuperRare over three years to address the research question. Thereby, it empirically tests the hypothesised relationships between network positions amongst crypto artists and their performance measured by price and demand over time.

Figure 1.

Methodology

(20)

Figure 1 displays the sequences of the study's methodology. Due to the large blockchain provided by SuperRare, the results of this study confidently describe the transaction behaviour of peers on that platform. Unfortunately, the study is only processing the data of one platform.

Therefore, one must be cautious making general statements about crypto art markets based on the results of this study. More detail on the collection and processing of the ledger data follows.

3.2 Data Collection

The data used in this research originates from the blockchains of SuperRare, a prominent crypto art market founded in 2018. As part of a research project at the University of Amsterdam, we have access to the dataset of SuperRare within the time frame between April 2018 and March 2021. The data set provides the financial transaction measures in USD and ETH and information on user addresses (artists and buyers), image tags, media type, description of artworks, token I.D.s, contracts, bids, conversion rate, ETH, timestamps, changes in ownership, and images. In total, the data contains 17529 transactions of 22242 artworks created by 904 artists.

Prior to this study, the original blockchain was converted using parsers into large .csv files available on Kaggle.com (https://www.kaggle.com/franceschet/superrare). Ethereum has an account-based blockchain. Contrary to UTXO-based blockchains, in account-based blockchains, the user I.D. of an account is the same for every transaction that the account performs (Motamed & Bahrak, 2019). Figure 2 illustrates the transaction network of account-based and UTXO-based blockchains, where smiley faces represent nodes and arrows represent edges.

Because ETH-transaction data makes it possible to identify accounts, SuperRare's ledger can be analysed using SNA. Therefore, the construction of the peer-trading network based on SuperRare's account-based ledger details in the following section.

(21)

Figure 2.

Account-based vs UTXO transaction data

3.3 Empirical Setting

In order to assess the effect of peer trading on an artist's performance, this study constructs a transaction network for which all buyers and sellers are artists. Thereby, the empirical setting concerns a certain sample of transactions on SuperRare for which both the seller's and the buyer's network I.D.s are creator IDs. The following subsections elaborate on blockchain in art markets, SNA for account-based blockchains and the network construction in Python and Gephi.

Blockchain & NFTs

Because blockchain offers a secure platform, ledger, or database where buyers and sellers can store and exchange value without the need for traditional intermediaries, it is often referred to

Account-based UTXO The smiley faces in this Figure

represent user IDs. In account-based blockchains the same smiley has multiple transactions because the same ID is used for all transactions.

Therefore, if degree measures are used to vizulize the network, some smileys will be larger than others. However, for UTXO blockchains, the user IDs are different for each transaction. This makes it impossible to perform SNA as nodes have no traceable identity.

(22)

as Web 3.0 or the 'internet of value' (Tapscott & Tapscott, 2017). The following text explains the underlying technology enabling value creation in blockchain networks and addresses the potentials of Web 3.0 for art markets.

In its simplest form, blockchain is a decentralised public ledger storing data in blocks forming a chain. Every time information is added to a blockchain network. It is stored as a block that is connected to the previous data block. All previous data blocks on that same chain are permanent and may not be manipulated because blockchain is not stored on a central server but multiple network servers. To alter a block's data, one would have to simultaneously alter all preceding data blocks on all network servers. Therefore, blockchain is considered the safest programming language at present (Narbayeva et al., 2020).

Because a blockchain network comprises multiple nodes storing the same information, rather than a single or a few central data hub(s), crypto networks are referred to as decentralised.

This decentralisation allows the emergence of peer-to-peer networks in which the code of the technology provides trust in transactions. However, securing and keeping a record of transactions is usually accredited to intermediaries who charge high fees for that service. Thus, a technology securing and keeping track of transactions yield excellent economic prospects for many businesses and entrepreneurs (Davydov & Riabovol, 2018; Hyun et al., 2018; Miscione et al., 2018;

Rane & Narvel, 2019; Sidorova, 2019). Essentially, the larger the fee paid to the intermediary, the larger the added value of blockchain technology. Therefore, especially in industries subject to high intermediary fees – like the creative sectors (Caves, 2003)– users benefit from employing this technology.

(23)

SNA & ETH

Social network analysis uses Graph theory and network theory to investigate social structures (Otte & Rousseau, 2002). In a graph theoretical approach, a social network illustrates a set of vertices (or nodes, units, points) representing social actors and a set of lines representing one or more social relations among them (de Nooy, 2009). Transaction graphs are graphs where the social relations among nodes represent transactions. These graphs are helpful to visualise the schematics of blockchains (Cachin et al., 2017; Chen et al., 2020) and derive network measures from those graphs.

Graph theory distinguishes between directed and undirected network graphs. Directed graphs refer to edges having a direction, i.e., an edge can only represent a one-way transaction between two vertices. ETH-based crypto transactions represent a directed network because transactions represent a one-way interaction from a seller to a buyer. The distinction between directed and non-directed graphs is especially relevant for the calculation of the degree measures.

For undirected networks, only the degree measure for each node can be calculated. For directed networks, we have in-degree and out-degree measures for each node representing the number of incoming and outgoing directed links from and to neighbours, respectively (Shneiderman et al., 2020). Moreover, because there can be multiple paths (transactions) between two nodes, crypto networks classify as directed multigraphs (Maesa et al., 2018).

When considering random graph theory approaches, it appears tempting to generate an undirected graph, unweighted graph. However, there is evidence that component size distributions of ETH networks follow a power-law distribution (Newman et al., 2001). Because finite component size distributions of random graphs usually do not follow a power-law distribution (Newman et al., 2001), this study refrains from using random graph theory to analyse ETH

(24)

networks. Instead, it derives measures from a directed multigraph based on the transactions between artists. To test for the causal relationship between peer-trading and an artist's performance, degree and weighted degree measures from SNA quantify the count and weight of transactions between artists. Following this logic, the next subsections detail the network creation and graph construction.

Network Construction: Python

This study uses Python's Pandas package to construct networks of particular interactions (Appendix A). The sales.csv file (available on Kaggle.com) contains buyer and seller information for each transaction. These two columns together serve as an edge list for computing nodes.

Creating a network of nodes that solely portray inter-peer transactions, the panel data frame is merged so that the 'buyer' and 'seller' columns contain the I.D.s of creator (artist) accounts.

Thereby, every transaction of the network is between artists on the platform. The initial 904 artists of the dataset are reduced to 417 artists who engage in peer-trading, i.e., artists directly trading with a peer. Because, if artists purchase a peer's token in the secondary market – when the seller is not the artist – the transaction does not represent an interaction from artist to artist but involves a non-peer party. Therefore, one cannot simply create a network for all transactions for which buyer I.D.s correspond to creator I.D.s — the following details how to construct the network graph based on interactions between artists.

(25)

Graph Construction

Money flow graphs (MFG) are defined as MFG = (V, E, w) where V is a set of nodes¹, E is a set of edges and, w the edge weight (ETH value of the transaction²). E is ordered in pairs of nodes, E = {(𝑣_𝑖, 𝑣_𝑗)|𝑣_𝑖, 𝑣_𝑗 ∈ 𝑉}. So that the edge order indicates the direction of a transfer. After generating the edge and node lists for this peer-trading network, the .csv file must be altered in excel before performing SNA in Gephi. Gephi is the Python-based desktop extension used to visualise the network and calculate graph metrics of social network analysis (Appendix B). Figures 3a, 3b, 3c show the visualisation of the peer-trading network.

In all Figures, the nodes are partitioned by the peer-modularity class. In Figure 3a, nodes are sized by inter-peer betweenness. Figure 3d shows the profile of the largest node displayed in Figure 3a and Figure 3b [bottom left]. As expected, this artist holds an extensive collection of peer- artworks (83) of similar style. The style similarities hint at the formation of artistic groups or movements on crypto platforms which positively impact performance (Wijnberg & Gemser, 2000). All pink nodes displayed in Figure 3a and Figure 3b are part of the same modularity class, reflecting a stylistic grouping of artworks or crypto art categories. Being part of a specific modularity class within the peer-trading network could indicate artists acting as selectors to drive up the value of their group. The identification and evaluations of these groups are beyond the scope of this study.

Figure 3b shows the modularity classes in different coloured and sized nodes based on betweenness centrality, while Figure 3c displays the closeness centrality of artists in the network.

1 Nodes are also referred to as accounts, creators, artists or vertices.

2 Transactions are also referred to as edges, peer-trades or interactions.

(26)

Figure 3c shows the peer-trading network with a force layout subject to closeness centrality measures. It shows how there are two groups of artists with similar measures of centrality. The larger nodes have a closeness centrality of 1, which indicates that they have strong ties with each other. The modularity classes that classify such ties are mainly coloured pink and yellow. These figures illustrate the use of SNA to describe and visualise network positions of the artists on SuperRare.

Figure 3a.

Peer trading network by modularity class [colour] & out-degree [size]

(27)

Figure 3b.

Peer trading network by modularity class [colour] & betweenness centrality [size]

Figure 3c.

Peer trading network by modularity class [colour] & closeness centrality [size]

(28)

Figure 3d.

'Mattia Cuttini' – Artist profile with most significant sales to peers

After calculating, visualising, and interpreting the SNA measures, the performance metrics (Appendix C) and the resulting node list is further processed in Python (Appendix D) to finalise the raw input of the regression.

3.4 Analytic Strategy

This part of the Methods chapter clarifies the independent and dependent variables as well as the process of building a multiple regression model with the maximum possible exploratory power, given our data and input variables. Figure 4 depicts the analytic strategy of this paper by showing how two different theoretical approaches to social capital theory are quantified using SNA measures (right-hand side of Figure 4). There are two central bodies of theory addressed here: the signalling effect of volume vs magnitude of peer trading and the birding and bonding

(29)

Figure 4.

Analytic strategy: assignment of hypothesis to theoretical lenses

Independent Variables: Social Network Analysis Measures

The network measures relevant to this study concern the degree centrality and network diameter (closeness and betweenness) measures of artists in the peer-trading network. The network measures discussed in the literature review are incorporated in the SNA package of Python:

NetworkX. As NetworkX is the foundation of Gephi, the computation of measures in Gephi corresponds to the same formulas used in Python and presented in the literature.

The degree and degree centrality measures of a node are based on an adjacency matrix. For this study, creating a matrix is unnecessary because the SNA software described in the graph construction section does not require the handling of matrices. Nevertheless, to understand what the measures are based on, an example of an adjacency matrix of a directed graph is illustrated below.

A(D, x) = [

0 𝑥₁ 𝑥₂ 0

𝑥₂ 0 𝑥₆ 𝑥₇ 𝑥₁₀ 𝑥₇ 0 𝑥₈ 𝑥₁₁ 𝑥₁₂ 𝑥₅ 0

]

Performance: ETH

& Bids

Briding vs Bonding (H1)

H1: Closeness &

Betweenness Centrality

H2: Clustering Coefficient

Peer Trading:

Volume vs Magnitude (H2&3)

H2: Degree Measues

H3:Weighted Degree Measure

(30)

Where A is the adjacency matrix of directed graph D for a set of edges (edges are the vectors making up the adjacency matrix). Therefore, every entry in the matrix represents a transaction, and every column/row represents a node. For example, the first column in our toy ³ matrix has three entries. Thus, the node corresponding to that vector has a degree of three. The calculation of degree centrality is detailed below.

Degree Centrality. The degree of a node is the number of edges connecting to that node

and may be used to indicate the social capital of that node (Guo et al., 2019; Wang, X. et al., 2012). In the context of this study, the degree indicates the number of transactions an artist has with other artists. Similarly, the weighted degree accounts for the magnitude of the transaction.

The degree of node j is calculated as follows:

(1) 𝐷_𝑗 = ∑^𝑁_𝑗=1𝑎_𝑖𝑗

Because we are investigating a directed graph, we may also look at out-degree and indegree, which correspond to the number of times a node sells to a peer (out) and the number of times a node purchases (in). Functions (2) and (3) correspond to out-degree and indegree, Function (4) and (5) correspond to the respective degree distribution:

(2) 𝐷_𝑜𝑢𝑡 = ∑^𝑁_𝑗=1𝑎_𝑖𝑗 (3) 𝐷_𝑖𝑛 = ∑^𝑁_𝑗=1𝑎_𝑗_𝑖 (4) 𝑃_𝑜𝑢𝑡(𝑘) =^∑ ¹^{𝐷𝑜𝑢𝑡 𝑗=𝑘}

𝑁𝑗=1 𝑁

(31)

(5) 𝑃_𝑖𝑛(𝑘) =^∑ ¹^{𝐷𝑖𝑛 𝑗=𝑘}

𝑁𝑗=1 𝑁

For weighted degree measures, the weight of the edge is simply added to the vertex in the adjacency matrix. The weighted degree measure in the context of this study corresponds to the amount of money associated with a transaction between peers. Thus, a node with a high weighted degree buys off peers at high prices.

Network Diameter. The triadic measures of this study focus on betweenness, closeness

centrality, and cluster coefficient. As mentioned in the Literature Review, these centrality measures estimate the importance of actors as intermediaries, the pace of their information flow, and the likelihood of clusters in their direct neighbourhood. The betweenness of node, v, indicates how frequently v lies along geodesic pathways of other nodes in the network (Valente et al., 2008) and is calculated as follows (Freeman, 1977):

𝑔(𝑣) = ∑ 𝜎_𝑠𝑡(𝑣) 𝜎_𝑠𝑡

𝑠≠𝑣≠𝑡

Where 𝜎_𝑠𝑡 is the total number of shortest paths from node s to node t and 𝜎_𝑠𝑡(𝑣) is the number of those paths that pass through v. Closeness centrality measures how close each node is to other nodes in the network and is calculated as follows (Bavelas, 1950):

𝐶 (𝑥) = 𝑁

∑ 𝑑(𝑦, 𝑥)_𝑦

Where 𝑑(𝑦, 𝑥) is the distance between vertices x and y, and N is equal to the total size of the network.

Clustering coefficient. The clustering coefficient calculates the likelihood of a direct neighbourhood to form a cluster, i.e., a complete graph where all vertices have edges connecting

(32)

them. The global cluster coefficient (GCC) shows the tendency of a network vertices to create a cluster with other vertices in the graph and is defined as:

(6) 𝐺𝐶𝐶 = 3 𝑥 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑟𝑖𝑝𝑙𝑒𝑡𝑠 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑟𝑖𝑝𝑙𝑒𝑡𝑠 (𝑜𝑝𝑒𝑛 𝑎𝑛𝑑 𝑐𝑙𝑜𝑠𝑒𝑑)

This measure applies to directed and undirected graphs and is also referred to as transitivity (Luce & Perry, 1949). Further to adjust this metric for weighted graphs, the following formula is used (Opsahl & Panzarasa, 2009):

(7) 𝐶_𝑖 = ¹

𝑘𝑖(𝑘𝑖−1)∑_𝑗,𝑘𝐴_𝑖𝑗𝐴_𝑗𝑘𝐴_𝑘𝑖

The GCC measure can help gain insight into component structures in transaction networks and may be observed over time to investigate how likely clusters form over time in crypto art networks. Moreover, the local clustering coefficient (Watts & Strogatz, 1998) for a single node in directed graphs is given by:

(8) 𝐶_𝑖 = ^|{𝑒^𝑗𝑘^:𝑣^𝑗^,𝑣^𝑘^∈𝑁^𝑖^,𝑒^𝑗𝑘^∈𝐸}|

𝑘_𝑖(𝑘_𝑖−1)

Dependent Variables: Performance metrics

This section addresses the performance metrics of this research. While financial performance does not necessarily capture an artist's creativity (Wachs et al., 2018), it is a traceable and relevant metric to assess the performance of crypto artists as entrepreneurs. The price of artworks in the data set reports ETH and USD. This study shows models for both performance metrics; however, if the author refers to financial performance, she refers to the ETH measure of performance as the ETH values are not subject to conversion rates. Because the effect of conversion rates is beyond the scope of this study, the USD model is not interpreted but included

(33)

study. Another performance metric derived from the data is demand. Similar to financial performance, to measure demand, the total number of bids on an artist's work is aggregated per artist. This measure is similar to the fame of an artist, which is a standard metric in creative industries to compare performances among artists (Banerjee & Ingram, 2018).

Therefore, the three performance metrics bids, ETH, and USD are the dependent variables for this study's three multiple regression models. They are created from the ledger data with the use of Python (Appendix C). To conclude, the ETH performance refers to the total ETH value associated with each artist, i.e., the cumulative ETH amount generated by the artist's collection and the bids performance metric measures the cumulative demand for an artist.

3.5 R – Pre-process & Stepwise regression

To conclude the Methods section, this element addresses the pre-processing of the data and the model building in R. First, the bestNormalize package is applied to find the optimal transformation for the log-normal distributions of the variables (Appendix E). As Yeo-Johnson best transforms power distributions that include negative values or zeros by adding a constant to each entry before the transformation, bestNormalize confirms this is true for the present data.

After transforming the data and removing the missing observations, the stepwise function in R estimates the models (Appendix F). This language pack allows for the building of multiple regression models effectively. One may put many independent variables into this function, and the program performs forward/backward procedures of adding/dropping variables until the optimal adj. R square is reached (Appendix G). Next, the VIF checks for multicollinearity and is calculated for each variable using the following formula = 1/(1-R^2). A VIF value of 1 indicates no correlation. Between 1 and 5 indicates a moderate correlation. The maximum tolerable VIF value is 10 (Daoud, 2017). The highest VIF is for the weighted degree in the ETH model (VIF of 5.79).

(34)

The lowest VIF score concerns the clustering coefficient in the bids model. Overall, the two models have an average VIF of 3.02.

The USD model has high VIF values and is further subject to the effects of rate. Thus, it does not serve as a base for any valuable interpretation. However, as per the VIF of the models, there is no pressing collinearity problem with the ETH and bids models, so we proceed to stargazer for the programming of the tables. The following chapter presents the regression analysis results and discusses the descriptive statistics and correlations of the three models.

(35)

4. Analysis

This section presents and interprets the results of the statistical tests performed in this study.

First, showing the descriptive statistics for the sample of 417 artists engaging in peer trade, then correlations and distribution of the variables. Next, the regression results validate or falsify the nine hypotheses of this study. Among all artists in the peer-transaction network, the average ETH value generated by an artist is 62.49 Ξ, the dollar value is $ 54,006.74, and the average number of bids is 97.15. Table 1 presents the descriptive statistics of the peer transaction network before pre- processing. The more variables are entered into R-Studio's step(x) function, the better the model's explanatory power. The variables that step(x) did not include are not in the Literature Review as they do not add value to modelling artistic performance and are not in the models. The descriptive statistics in Table one include all measures used when performing the step (x) function.

Table 1.

Descriptive statistics of the peer interaction network.

(36)

Appendix H shows the correlations, linearity and normal distributions for all variables before and after the Yeo-Johnson transformation. Some variables like weighted in-degree, in- degree, and betweenness centrality report normal distributions that violate regression assumptions.

Hence, these variables are not in the regression model due to their high VIF values and not normal distributions. Similarly, the distribution of the clustering coefficient could be considered worrisome. However, the central limit theorem states that if samples are large enough, the mean of their normal distribution collapses on zero (Brosamler, 1988).

The measures included in the modelling of at least one model include closeness centrality, clustering coefficient, out-degree, degree, and weighted degree. Figure 5 reports the correlations, normal distributions, and residuals of the included variables. Though the correlations for some of the variables included in the final models are above 0.5, the VIF of the variables is within reasonable bounds (see above). Therefore, the variables of this study can be used to estimate results entailing valuable interpretations.

Next, the regression results of the final models are presented and falsified or validated.

Table 2 reports the results of the regression for all three models. Likely, the demand for and price of an artist depends on factors other than network position. However, given that only network measures are included in the models, the average R square of all models is around 40%. Therefore, on average, about 40% of the squared residuals in the models are predicted by the models. For the ETH model, this value is almost 50%. Hence the ETH model is our best predicting model. This is a bit lower for the USD model as the ETH rate at the time of transaction directly influences the dollar value of the artwork documented at that point in time. However, the conversion rate of the transaction likely impacts all three models, as a good rate is associated with more overall trading

(37)

Figure 5

Correlation, normal distributions, and residual distribution

According to the bids model (1), we can prove that the formation of clusters in the neighbourhood of an artist's peer network is negatively associated with the total amounts of bids on an artist's works (-.090, p<.0.05). This finding partially confirms Hypothesis 1c. Next, the amounts of times an artist sells artwork to peers is positively correlated to the overall number of bids placed on his/her works (0.344, p<0.001). This finding partially confirms Hypothesis 2b.

Further, the sum of all interactions an artist has within the peer network has a direct positive effect on his/her demand (0.366, p<0.001) and ETH price (0.133, p<0.001). This finding confirms Hypothesis 2c.

(38)

Table 2

Regression Output.

Next, the ETH model (2) shows that closeness centrality (-.108, p<0.001) of an artist’s position within his peer trading network has a negative effect on his price. This finding partially supports Hypothesis 1a. On the other hand, closeness centrality is not significant for the demand model but only negatively reflects an artwork's price. Moreover, the ETH value of an artist's total interaction positively affects his price (0.581, p<0.001). This partially confirms Hypothesis 3c.

To conclude, this study finds partial support for H1a (only supported in the ETH model), H1b and H2b (only supported in the bids model), H3c (not supported in the bids model). However, these results are partial as none of the SNA metrics significantly affect both performance measures.

The remaining hypotheses could not be falsified but were not supported as the network measures

(39)

proposed in those hypotheses were not included in the regression models. The following chapter goes into details on the interpretation and implications of these findings.

(40)

5. Discussion

This chapter readdresses the theories of the literature review in the light of the above presented statistical findings. First, the theoretical contribution is contextualised, and the research question to what extent peer trade impacts performance in crypto art markets is readdressed. After performing the analysis, this research can conclude the differences in the effects among various performance metrics. Moreover, findings regarding the degree measures are interpreted in the light of the signalling theory. After discussing the contribution of the study, the implications of the regression results for practitioners are discussed. Finally, the chapter elaborates on limitations and motivates channels of future research.

5.1 Theoretical Contribution & Contextualization

This section elaborates on the findings and their theoretical contribution. In doing so, the variable selection is explained in more detail, and the initial question: "To what extent do peer interactions influence an artist's performance in crypto art markets?" is readdressed. The two theoretical lenses of social capital (bridging vs bonding) and signalling effects of large vs plentiful peer transactions served as a foundation for this study's hypothesis.

Figure 6 summarises the relevant hypothesis where the triangle outlines the findings contributing to the bridging vs bonding discussion. The square outlines the variables corresponding to the theory on peer trading and signalling behaviour. The following text explains why not all variables represented in the nine hypotheses of the Methods section are included in the models of the Analysis section.

The regression models test for the effect of peer-interaction in SuperRare's transaction

(41)

creations. Analysing these models, the following network measures add explanatory power:

closeness centrality, clustering coefficient, out-degree, degree, and weighted degree.

Figure 6

Summary of Findings

This study measures performance by the demand and the price of artists. The USD model is included in this study to demonstrate the difference between the USD and ETH measures and, therefore, to motivate future research on the effect of conversion rates on crypto art prices.

Nonetheless, the following discusses the theoretical contribution of the ETH and bids model results Regarding the other two models, Closeness centrality has a significantly negative effect on the ETH performance of an artist. This finding supports prior literature on the bridging and bonding discussion. In the particular context of this transaction network, shorter distances between artists negatively affect ETH value. This could potentially reflect a homogeneous information flow in these circles and is in line with previously discussed findings of the bridging and bonding discussions (Burt, 1983; Burt, 2009; Uzzi & Spiro, 2005; Wachs et al., 2018).

(42)

Similarly, the clustering coefficient shows a negative effect on the demand for an artist.

This indicates that a higher density in the ego network of an artist's peer transactions decreases the number of bids placed on his works. However, it is not clear why the density of an artist's ego network hurts demand but not price. Network density is found to have a negative effect on innovation performance (Tan et al., 2015) as well as the price of cryptocurrencies (Motamed

& Bahrak, 2019). In the context of this study, the higher the density of an ego network, the fewer bids are placed.

Regarding betweenness centrality, the variable did not qualify as a value-adding variable within our data set. This finding is most likely due to the abnormal distribution of this measure.

However, due to the significantly negative results for the effect of clustering and closeness, this study confirms 2 out of 3 hypotheses testing for briding vs bonding positions. Future research is required to explore the occupation of structural holes by artists in crypto networks to confirm or reject this proposition. To conclude, the present findings support the bridging side of the debate as closeness centrality and the clustering coefficient significantly negatively affect performance.

Therefore, Hypothesis 1 of this study is partially supported.

Overall, to answer the sub-question whether the relationship of our measures with performance differs among performance metrics: Yes. The difference in these findings may be due to a high price being more reflective of access to information than high demand. Therefore, high closeness centrality leads to access to similar information and resources, reflecting negatively on an artist's performance. However, suppose the creativity of an artist has no significant relationship with the fame of an artist (Banerjee & Ingram, 2018). In that case, the nature of the information available to artists with high betweenness centralities may be the source of competitive advantage

(43)

Regarding the signalling theory, the following text addresses the findings for Hypothesis 2 and 3. In the context of signalling theory, the sender of a signal would be the artist (insider) who buys a token. At the same time, the receivers of the message include all participants of the market.

However, the decoding of a signal depends on the environment of the receiver. When an artist decides on how much money to spend on a peer's token or how many peers to interact with, he/she sends a conscious or unconscious signal to consumers and peers.

The degree measures of this study capture the effect of artists' peer-trades in terms of the count and weight of interactions. The out-degree has a significantly positive effect on an artist's demand. Therefore, the more tokens an artist sells to his peers, the more tokens he sells overall.

What is interesting here is that the in-degree measure did not add significant power to either model.

Therefore, buying a peer's work does not have a significant impact on performance. Similarly, the amount of money spent on a peer's works (weighted in-degree) has no impact on performance.

Interestingly, the amount of money peers spent on an artist's works (weighted out-degree) also has no significant effect. This could hint at artists engaging in peer-trade to drive up value due to multiple changes in ownership of a token (Baily, 2019). If the monetary value of the peer- transaction is not relevant to the demand of an artist. However, the count of peer interactions (degree) is relevant to the demand and ETH value. This could indicate a signalling effect of peer trades.

In that sense, artists would engage in peer trading because they understand that other artists are visible owners of their works, which drives up the price and demand. Though the findings for the in-degree measure were not significant, there is reason to believe that there is some form of mutual understanding among artists that if they support another, it reflects positively on their performance. This argument is in line with the concept of coopetition, which describes a strategical

(44)

approach to competition that includes cooperation. Coopetition is a concept that applies to the competitive dynamics in the creative industries, blockchain, and social media (Virtanen et al., 2017). Coopetition is a common phenomenon in the creative industries and at the heart of the philosophy for circular economies (Narayan & Tidström, 2020; Walley, 2007; Yami et al., 2010). In that sense, artists' competition in crypto art markets invites future research to understand the strategies artists employ to signal legitimacy.

Finally, the weighted degree within the peer network has a positive effect on the ETH performance of an artist. This shows that the amount of money spent between artists has a positive effect on the ETH price of an artist. This is in line with the signalling argument because artists likely understand that buyers would perceive these works as more valuable if they spent more significant amounts on each other's works. Thereby, peers perform the market selection. This nature of the peer selection systems employed in crypto art markets aligns with findings by Franceschet and Colavizza (2019). They analyse the curation activities of artists in crypto spaces.

The following elaborates on the implications for practitioners.

To conclude, the findings of this study find support for positive effects of degree, out- degree, and weighted degree. When translating the SNA measures back to the form of interaction they are based on, the total number of trades an artist has with his peers, the number of sales to peers and the economic magnitude of total peer transactions partially explain the success of an artist on SuperRare. Interestingly, the economic magnitude of the transaction is not significant for the number of bids placed on an artist. This could hint at artist's engaging in low-cost transactions with each other to drive up their demand. Many bids may, however, be associated with the conversion rate at the time of the auction. Therefore, this divergence in results for the performance

(45)

signals sent through peer trading on performance is encouraged. Moreover, the intentions of an artist engaging in peer trades requires qualitative research.

After assessing the statistical data through the two theoretical lenses of this paper, we can answer to what extent peer transactions influence performance. The first theoretical lens provides insights with whom to interact, while the second lens provides insights on how to interact, i.e., performing many vs large trades. This research concludes that the effects differ among the metrics of price and demand. Artists should engage with many peers across diverse networks to optimise their network position. Further, artists or managers need to consider whether they seek buzz or high prices. This study suggests that transactions do not require much weight to create buzz, but the count of peer trades is relevant to artists’ demand.

The research question has, therefore, only been answered to an extent. Nevertheless, the study demonstrates an example of performing SNA on crypto art ledger data and motivates researchers to further participate in the extension of social capital literature to the fields of crypto art markets. Before elaborating on these channels for future research, the following text discusses the implication for practitioners.

5.2 Implications for Practitioners

Unfortunately, there seems to be no recipe for success in creative industries.

Nobody knows for which art there will be demand (Caves, 2000); however, in the context of crypto art, the speculations of artistic value performed by the market influence the value of crypto artists (Bailey, 2017). This study reports findings that help artists navigate through the crypto world as entrepreneurs. In that sense, the results of this study encourage the artists to engage in peer trading by creating many lose ties with multiple peer networks. Moreover, in doing so, the focus should lay on the identity of the peers one trades with and the accompanying signal to

(46)

consumers, rather than the amount of money spent on a peer's work. Additionally, the management of crypto art platforms is encouraged to analyse and understand the communities that form between artists and find inspiration for marketing activities and services accordingly.

5.3 Limitations & Future Research

The limitations of this study are diverse and extensive. Starting with the regression findings, the results are to be interpreted with caution as many unknown factors influence the price and demand of artists other than network position within the peer network, which is only based on transactions. Furthermore, the knowledge transfer of inter-artist contact likely manifests in social rather than financial transactions. Therefore, we encourage future researchers to investigate the effect of network position in inter-artist networks in diverse empirical settings. For example, the linear modelling of performance in crypto art networks presented here could be extended by network-based semantics and non-financial interactions of leading crypto platforms like Twitter or Telegram. This could allow for a deeper understanding of inter-peer interactions and the network positions of artists within a peer network that is not only based on transactions. In that sense, as this paper bases the results on data from a single crypto art platform, to better understand the crypto art community, future research is encouraged to study a larger sample of crypto art platforms and the inter-platform connections of peer as well as the on-site interactions. In this context, the signals originating from other interactions than transactions yield exciting directions for future research.

Moreover, the transaction graph constructed in this study is relatively simple compared to the complex graph approach presented in recent literature on ETH network analysis (Lin et al., 2020). Applying complex network approaches to crypto art networks has not been done to date

SuperRare: The effect of peer trading on the performance of crypto artists