SSRN Id4397409

Market Manipulation in Non-Fungible Token Markets *
Sebeom Oh†
Original Draft: March 2023
This Draft: September 2023
Abstract
Non-Fungible Tokens (NFTs) offer a unique opportunity to study market misconduct

in an unregulated crowdfunding environment. This paper examines insider and wash trad-
ing in the NFT market using publicly accessible Ethereum blockchain data. Results reveal
that insider purchases, particularly by those maintaining community ties, significantly pre-
dict future price returns. Despite over 422 million USD circulating in wash trades, their
impact on market outcomes is negligible. Investors can adeptly analyze public information
and avoid items that are abnormally priced, but do not impose penalties on wash traded
items or collections.
Keywords: Speculative Market, Market Manipulation, Transparency
* I would like to thank J.H. John Kim, Samuel Rosen, Oleg Rytchkov, Jongsub Lee (discussant), Donghoon
Kim (discussant), Kanghyun Cho, seminar participants at Temple University, 2023 Joint Conference with the
Allied Korea Finance Associations, and 19th Annual Conference of the Asia-Pacific Association of Derivatives
for their helpful comments and suggestions. This work was supported by the grant from 25th Young Scholars
Interdisciplinary Funds at Temple University. All errors are my own and comments are welcome.
†
Temple University Fox School of Business. Contact: sebeom.oh@temple.edu.
Electronic copy available at: https://ssrn.com/abstract=4397409

1 Introduction
What happens when all market participants in a speculative market can access real-time
transaction records, particularly when the information volume is manageable for detailed
analysis? Can investors adeptly navigate and avoid market-wide manipulations with such
transaction information at hand? Blockchain technology provides empirical insights into
these questions, as transaction records are open to the public. Specifically, non-fungible to-
kens (NFTs), series of digital assets like profile pictures, leverage blockchain for fundraising.
The NFT market gained substantial traction alongside the cryptocurrency bubble between
2021 and early 2022. For example, while the crowdfunding market recorded transactions
amounting to 1.08 billion USD in 2022, transactions in the NFT market reached a remark-
able 2.4 billion USD1 .
Similar to equity crowdfunding, NFT creators secure funds for their projects by of-
fering their NFT collections to early investors in the primary market. To understand this
better, consider an apartment company (NFT creator) raising funds for an apartment-to-be-
built (NFT collection) by pre-selling all the studios (a predetermined fixed number of similar
NFT items). The key difference between crowdfunding and NFTs is that the former involves
trading firm shares, while the latter entails selling relatively small number of virtual items.
Although NFTs often appear as virtual profile pictures (see Figure 2), they fundamentally
act as status symbols or membership tokens in a community, potentially granting voting
rights or unique privileges in emerging games or metaverses (Oh, Rosen, and Zhang, 2022).
One of the critics NFT and the cryptocurrency community face is the unregulated or
less regulated market environment. Since the creators of NFTs are anonymous in most
cases, it is not uncommon for projects to be abandoned if they are not successful on the
primary market. Some creators intentionally raised funds by selling NFTs and then disap-
peared2 . In fact, even for successful NFT projects that have received considerable investor
attention, it is difficult to verify whether there is insider trading based on asymmetric in-
1
See the data from Statistica for crowdfunding (https://www.statista.com/outlook/dmo/fintech/digital-
capital-raising/crowdfunding/worldwide) and for NFT (https://www.statista.com/outlook/dmo/fintech/digital-
assets/nft/worldwide).
2
NFT community calls this as a rug pull. Some examples are Frosties and Evolved Apes. The founders of
Frosties were arrested in California but it is a rare case.

formation due to the lack of reporting obligations. Additionally, several instances of wash
trading have been voluntarily reported by the NFT community3 . Wash trading in NFT
markets serves as a method for price and volume manipulation to generate fake trading
volume and influence index price to attract investors through repeated buying and selling
behaviors. However, concrete evidence depicting the prevalence and implications of such
misconduct within this novel and transparent market structure remains scant. The distinct
characteristics of NFT markets, such as a finite number of items and the distinctiveness of
each items, theoretically facilitate more straightforward analysis of transaction records by
investors compared to those in equity or even cryptocurrency markets.
In this paper, I study whether there exists unrevealed insider trading and wash trad-
ing in the NFT market. Insiders are defined as wallets who have received free items from
creators in the primary market, and I focus on their purchase activity rather than their
selling activity to eliminate frequent trading noise as much as possible. Wash trades are
defined as three types of transactions: (1) identity trade, where the seller and buyer are
the same wallet, (2) 1-1 trade, where a seller purchases the same item again within 7 days
after selling, or (3) matched order, where three wallets are involved in trading and all trades
occur within 7 days. In the NFT market, insiders are 4.9% of total wallets participated in
primary market, and wash trades are 0.3% out of 3.6 million secondary market transaction.
I investigate the impact of these misconduct behaviors on market outcomes for 558
successful NFT projects that successfully minted (i.e., sold) all items in the primary mar-
ket from March 2021 to January 2023 and traded until February 2023. Unrevealed insider
and wash trading could potentially influence the market in two ways. First, unrevealed
insiders may use their information on the NFT collection earlier than other investors since
they can directly communicate with NFT creators. However, they may not be able to use
their information effectively since they already hold an illiquid membership ticket. Second,
while NFT wash traders may aim to draw investor attention by inflating trading volume
and price, the effectiveness of such strategies is questionable as the NFT community ac-
3
See an article from Chainalysis (https://blog.chainalysis.com/reports/2022-crypto-crime-report-preview-
nft-wash-trading-money-laundering/), Decrypt (https://decrypt.co/91510/looksrare-has-reportedly-generated-
8b-ethereum-nft-wash-trading), or @hildobby_ (https://community.dune.com/blog/nft-wash-trading-on-
ethereum) for example.

tively exposes wash trades. Lastly, Banerjee, Davis, and Gondhi (2018) predicts that more
information access can be harmful when speculative motives dominate in the stock market.
The speculative motivation was the main driver of the NFT bubble in the sample period,
and the online community and public transaction record analysis guarantees high level of
fundamental information access. Although theoretical implications of Banerjee, Davis, and
Gondhi (2018) cannot be directly applied to NFT markets which mirror the structures of
art or real estate markets, their application offers a valuable empirical opportunity to test
theoretical frameworks through available data.
The results indicate that insider buying activity is a strong predictor of future daily
price index returns. A one standard deviation increase in insider buying activity leads to
around a two percentage point increase in future daily median price returns. While wash
trading exhibits a negative effect on future price returns, the impact is economically negli-
gible. Furthermore, neither insider purchases nor wash trades have a significant economic
influence on the future change in trading volume. This suggests that insiders take advan-
tage of information asymmetry in NFT markets, but wash trading is actually ineffective in
manipulating market outcomes. Subsequent analysis reveals that online community works
as an information channel. Insiders maintaining substantial ties with online community
are the ones who can accurately forecast future returns on purchases.
This paper also shows the wash trades account for over 422 million dollar in the sam-
ple. Considering the ineffectivenss of wash trading, it raises the question: Are investors
smart enough to notice and avoid manipulations? Otherwise, an intention behind wash
trading might not be market manipulation. One answer could be the cryptocurrency reward
from NFT marketplaces (i.e., NFT exchanges). Several marketplaces charge a platform fee
close to zero percent and reward traders proportional to the transaction value of traded
items. Further analysis indicates that rewarding platforms are highly associated with the
occurrence of wash sales, while insiders are not associated with wash sales. However, mar-
ketplaces actually earned lower marketplace platform fees and creators received less royalty
fees from wash trades. In addition, investors do not avoid trading collections associated with
wash trades both in the short and long term. This suggests that investors can effectively
process available information and avoid abnormally priced items.

One might wonder why financial economists should care about a speculative market
that may fade away within five years. However, this paper makes three contributions to
the literature. Firstly, it discusses the use of blockchain technology as a forensic finance
tool to detect insider and wash trading in a recent crypto asset. The market structure of
NFTs provides an opportunity to study manipulative behaviors more precisely compared to
behaviors in equity or cryptocurrency markets. Secondly, the findings are crucial for regula-
tors and policymakers interested in ensuring fair and efficient markets. The study suggests
that efforts should be made to increase transparency in the NFT market, particularly with
regard to unrevealed insider trading practices. Additionally, the study highlights the im-
portance of regulating reward systems in NFT marketplaces to prevent artificial price and
volume manipulation through wash trading. Lastly, this paper demonstrates investor be-
haviors to avoid a certain type of manipulations given the limited amount of information
volume and the transparency in a speculative market. Overall, this paper provides valuable
insights into the functioning of the NFT market and sheds light on the extent and impact of
misconduct behaviors in this emerging asset class.
I provide some technical background on the NFT space and how to measure insider and
wash trades in section 2. Sample selection procedure and summary statistics are shown in
section 3. In section 4, I examine the impact of insider and wash trading activities on market
outcomes using predictive regression analysis. I investigate more on insider’s information
advantage in section 5. Additionally, I discuss potential purpose of wash trades and discuss
the aftermath effect on wash-traded items in section 6. Finally, I present my conclusions in
section 7.
Related Literature
This paper discusses various topics related to market misconduct, including unrevealed
insider trading and manipulative trading. Several studies have examined the spread of un-
revealed insider information through family connections (Anderson, Reeb, and Zhao, 2012;
Sun and Yin, 2017). Other social ties, such as friends and geographic proximity, have also
been explored as means of spreading inside information. Ahern (2017), for example, showed
how these ties can be used to spread insider information. In the context of revenue-sharing

crowdfunding, Pourghannad, Kong, and Debo (2020) found that early investors who have a
social tie with the entrepreneur may be informed about the project. However, Cohen, Mal-
loy, and Pomorski (2012) argued that not all insider trading involves the use of nonpublic
information. They distinguished between routine and opportunistic insider trading based
on past trading records, discovering that only opportunistic trades predict future returns.
This paper contributes to the literature by positing that face-to-face interactions or exclu-
sive online community ties could potentially serve as insider mechanisms within the context
of cryptoassets.
Manipulative trading has also been the subject of research. Aggarwal and Wu (2006)
showed that using SEC litigations from 1990 to 2001, market manipulation occurred in
small and illiquid OTC markets, with insiders and brokers potentially being the manip-
ulators. Kyle and Viswanathan (2008) identified various forms of illegal price manipula-
tion, such as corners and squeezes, pump-and-dump, and not making required disclosures.
Massoud, Ullah, and Scholnick (2016) discussed the price and liquidity effects of hiring
undisclosed promoters for publicly traded firms, and Li, Shin, and Wang (2022) analyzed
pump-and-dump schemes in the cryptocurrency market, finding that they produce abnor-
mal short-term increments in price, volume, and volatility.
Manipulation is often associated with high-frequency and deceptive trading activities,
known as spoofing, which do not result in ownership changes. Aitken, Cumming, and Zhan
(2015) explored the relationship between high frequency trading and market manipulation
in stock markets. Wash trading, which is another form of fake trading, has been a focal point
of many studies. Although investors and scholars commonly refer to it as wash trading, the
U.S. Internal Revenue Services (IRS) has formally defined it as non-tax deductible trades
due to the absence of change in ownership (see e.g. Grinblatt and Keloharju (2004) for
tax-related research). Wash trading can be misleading to investors as daily trading volume
is often used as a prominent market attention measure. Most of the existing research on
wash trading has concentrated on exchanges or brokers. Cao, Li, Coleman, Belatreche,
and McGinnity (2016) utilized directed graph theory and dynamic programming to detect
wash trading. In the context of the crypto space, Gandal, Hamrick, Moore, and Oberman
(2018) and Aloosh and Li (2019) directly investigated manipulative behavior through bot

trading, using leaked secret information from a Bitcoin exchange. Additionally, Cong, Li,
Tang, and Yang (2023) indirectly estimated wash trading using Benford’s law on regulated
and unregulated crypto exchanges.
Detecting wash trades in the NFT space may be easier compared to traditional mar-
kets such as stocks or cryptocurrencies, as unique NFT items are traded directly between
buyers and sellers. It requires a specific wash trade counterpart wallet or conspirator, which
is not the case for stock or cryptocurrency wash trading. Additionally, the public blockchain
enables tracking of manipulations. Furthermore, NFT creators and insiders may have an
incentive to participate in wash trading as it generates fake abnormal liquidity and an un-
usually high price to attract investors as in Aggarwal and Wu (2006) or Massoud, Ullah,
and Scholnick (2016). Wachter, Jensen, Regner, and Ross (2022) analyzed 52 NFT collec-
tions using graph theory and found that wash trades accounted for around 2% of total sale
transactions. However, further research is needed to better understand the extent and im-
pact of wash trading in the NFT market. This paper delves into the economic analysis of
wash trading and underscores the potential of blockchain technology in data governance
and transparency.
This paper adds to the existing literature on NFT markets and equity crowdfund-
ing, specifically addressing market structure and the potential for manipulation. For a
discussion of NFT markets from finance perspective, Kräussl and Tugnetti (2022) provided
an overview of NFT markets and summarizes the pricing methods of other papers. Oh,
Rosen, and Zhang (2022) compared the returns of experienced and inexperienced investors,
Bao, Ma, and Wen (2022) examined herding behaviors and found inexperienced investors’
entering can be a trigger of herding, Borri, Liu, and Tsyvinski (2022) and Kong and Lin
(2022) attempted to construct market indices and conduct related analysis. Wilkoff and
Yildiz (2023) examined the effect of media coverage on NFT market liquidity, and Falk,
Tsoukalas, and Zhang (2022) discussed how NFT royalties to creators are determined. In
the equity crowdfunding space, Meoli and Vismara (2021) investigated information manip-
ulation, Cumming, Hornuf, Karami, and Schweizer (2021) analyzed the determinants of
crowdfunding fraud using social media data, and Babich, Marinesi, and Tsoukalas (2021)
demonstrated that crowdfunding can benefit entrepreneurs and investors but may also be

theoretically harmful.
2 NFT Markets, Measures, and Predictions
2.1 Backgrounds
Primary Market (Mint) Secondary Market
Fixed Price Variable Price

Creators Early Investors New Investors
Royalty Fee
Figure 1. Overview of NFT Markets

Notes. The figure above shows the simplified NFT market. Creators sell NFT items at
fixed price to initial investors and then initial investors trade items in secondary market.
Creators receive royalty fee on every realized trades.
Before describing the data and summary statistics, it is necessary to clearly explain
the terminologies and background with Figure 1. An NFT collection is a set of NFTs on
the same theme and launched by an NFT creator team. An NFT is an individual item in
an NFT collection. Alternatively, an NFT is a single picture, while an NFT collection is
a set or brand of pictures. For example, the right picture of Figure 2 is an NFT, and the
left picture is an NFT collection. The primary market is a market where NFT creators sell
NFTs directly to early investors at fixed prices4 . It is also called minting or mint. NFT
creators promote their minting process through various online communication channels,
such as Twitter, Discord, and Reddit. Early investors can sell their minted items to other
investors, and some investors buy and sell items from others on the secondary market. As
well as raising funds on the primary market, creators are paid a percentage royalty fee on
every secondary market sale. As a result creators keep updating their development process
4
Some NFT collections have different set of fixed prices depending on the amount of mints. When an
investor buys more items, the cheaper the mint price for each NFT. However, there is a limit on the maximum
amount one can mint set by creators.

and promoting sales to potential investors and NFT holders after the primary market sales.
Note that successfully minting all NFTs is imperative in the subsequent secondary market
sales as new entrants can buy NFTs at a fixed price from creators anytime if there is an
unsold item.
Figure 2. Example of NFT collection and NFT item

Notes. The left picture shows an NFT collection which is a set of pictures on the same theme
under the same brand name called the Bored Ape Yacht Club. The right figure is an item
(#3749) of Bored Ape Yacht Club that is sold at record price, 740 ETH (2.9 million USD) at
September 6th, 2021.
The focus of this paper is on NFT collections based on the Ethereum (ETH) blockchain
system, one of the most popular cryptocurrencies. Note that buyers and sellers do not need
to trade in ETH cryptocurrency. While transaction data is recorded in the ETH blockchain,
participants can also pay using alternative cryptocurrencies like USDC, USDT, or ApeCoin.
Although it is not discussed in this paper, traders that use the Ethereum system must pay
the ETH transaction fee or gas fee to blockchain miners for transaction verification in every
NFT trade including mints. This fee depends on the complexity of the Ethereum network.
In late 2021, when the cost of transactions on ETH became high due to increased demand
for trading ETH itself or crypto-based NFTs, some NFT creators launched their collections
on other blockchain systems, such as Polygon. Nevertheless, the vast majority of NFTs are
still based on the Ethereum ecosystem, so I restrict the sample to Ethereum-based NFT
collections. Furthermore, the fixed supply5 of NFT items plays a crucial role in defining
5
Some famous NFTs like CryptoKitties do not have supply limit as their cyber-cats repeatedly generate
their kittens, which may lead to infinite number of items.

the scarcity and limited access of the NFT market, making it possible to apply economic
principles that are applicable to other traditional asset classes such as equity, housing, or
the arts as in the setting of Oh, Rosen, and Zhang (2022).
2.2 Insiders in NFT markets
Insider trading in a public firm refers to the stock trading behavior of managers who hold
more than a certain amount of shares. Insiders of public companies are required to report
their trading records to the U.S. SEC. Unlike the stock market, there is no regulation requir-
ing insiders in the NFT market to report their trading records. Furthermore, the personal
identity of each wallet is not revealed unless the owner of the wallet chooses to disclose
it. Therefore, insiders can only be inferred from transaction records. Without legal conse-
quences for insider trading, those with information advantages in a speculative market are
more likely to exploit their information advantage for trading purposes.
A distinctive characteristic of the NFT market is the concurrent online communica-
tion system facilitated via platforms like Twitter and Discord. In Discord, each NFT project
has two types of chat rooms. The first chat room is open to everyone, including aspiring in-
vestors who do not yet hold an NFT, while the second is exclusively for current NFT holders.
Through the automated verification system, NFT owners can establish their ownership, and
all they need to do is show their verified ownership to Discord managers, who are NFT cre-
ators and their communication teams. Thus, access to member-only chat rooms is restricted
to NFT owners, as creators and their communication teams use these rooms to engage with
members of their community.
In the context of this paper, insiders are defined as wallets that receive free items
(Free Minters), given their potential access to internal information. Anderson, Reeb, and
Zhao (2012), Sun and Yin (2017), and Ahern (2017) discussed family and face-to-face con-
nection can be a channel for information leakage. Pourghannad, Kong, and Debo (2020)
found that early investors in crowdfunding is likely to have a social connection with the
project creators and obtain benefits from internal information. Recall that NFTs are used to
launch new projects and raise funds from investors. Free Minters are likely to be creators
themselves, since creators need to join their NFT social communities, individuals who have

social connections with project creators, or recipients of free giveaway events6 . This give-
away event is mostly for marketing purposes, and creators usually give an item to someone
who shares lots of tweets tagging NFT on Twitter or who already holds an NFT.
.6 .8
.6
.4
Fraction
Fraction
.4
.2
.2
0 0
0 .2 .4 .6 .8 1 0 .2 .4 .6 .8
Free items/Total Minted Items Free Minted Wallets/Total Wallets Participated in Primary Market
Figure 3. Distribution of Insiders in Primary Market

Notes. These figures report the distribution of potential insiders in collection-wise primary
market. The left figure shows the distribution of free items out of total items including
collections that are omitted in the sample selection process. In the analysis, NFT collections
on the right side of red dotted line are deleted. Potential insiders are defined as wallets that
received free items in primary market. The right figure describes the distribution of such
insiders out of total wallets involved in primary market.
Figure 3 presents the distribution of insiders from two perspectives. The left figure is
the histogram of items using full sample, and it is not difficult to see most NFT collections
does not give most items freely. Ad hocly, I omitted 131 collections for further analysis that
distributed more than 50% of their items without any cost (right side of red dotted line) as
they have higher probability of being derivatives for already successful main projects and
they are less likely to be fundraising projects. The right figure depicts the distribution of
wallets in the final sample. On average, 4.9% of wallets were classified as insiders on the
primary market. Other relevant summary statistics are present in Table A.1. The following
testable prediction summarizes the concept and explanation in this subsection.
Prediction 1: Behaviors of insiders who are defined as wallets that received NFT
items freely will strongly predict future returns since insiders may use internal information
through online community. Specifically, insider purchase will predict higher future price
6
These free giveaway items are sometimes called as airdrops. The amount of airdropped NFT items is
limited as the supply of NFT items are fixed.
10

returns. Insider selling will predict negative future returns but it may not be strong since
the market is upward-trending and speculative.
The reason for different prediction power in insider buying and selling behavior stems
from the membership identity of NFTs. Buying additional illiquid items when already hold-
ing items is more likely to represent an informational advantage and positive prospect of the
collection’s success as community members. On contrary, insider selling may have lower sta-
tistical significance to predict negative future returns since it is challenging to distinguish
routine trades from information-based trades in a speculative market.
2.3 Wash Traders in NFT Markets
By the U.S. IRS, if one sells securities at a loss and buys substantially identical securities
within 30 days before or after the sale, and there is no change in beneficial ownership, it is
classified as a wash sale. When there is a related third person or party, it is called a matched
order. Loss from wash trades is not tax deductible, but the wash trade itself is permitted.
Table 1. Example of Wash Trades in NFT Markets

Notes. One of manipulative trading records on a single item is presented in this ta-
ble. This collection is named “The Wonder Quest” with its unique contract address
0x08bEBEB5f042CCbaEb128582DA560cb25a5dB7e9. It is easily noticeable that investor
0x70e09... (marked as red) and 0x40c39... (marked as blue) buy and sell identical item
#1320 frequently on February 4th, 2022. Moreover their transaction prices from wash trades
(bolded) are enormously higher than previous transaction price.
Item # Trading Time Seller Buyer Price (ETH) Notes

1320 2021-07-26 20:12:29 0x0000000000000000000000000000000000000000 0x31992b19c40f2e472da5d39b167dc6fe952d3777 0.088800 Mint
1320 2021-08-12 03:39:03 0x31992b19c40f2e472da5d39b167dc6fe952d3777 0x3dcba64c3596aa254ad41502d8e15f9b54aa6e61 0.077000 -
1320 2022-02-02 01:10:17 0x3dcba64c3596aa254ad41502d8e15f9b54aa6e61 0x70e09c770c8bb76ed309db5ad9eab63a89a93788 0.020000 -
1320 2022-02-02 02:21:49 0x70e09c770c8bb76ed309db5ad9eab63a89a93788 0x40c398c0a3def59757683c82659f64678595f2de 0.045318 -
1320 2022-02-04 05:23:42 0x40c398c0a3def59757683c82659f64678595f2de 0x70e09c770c8bb76ed309db5ad9eab63a89a93788 36.812552 Wash
1320 2022-02-04 05:48:57 0x70e09c770c8bb76ed309db5ad9eab63a89a93788 0x40c398c0a3def59757683c82659f64678595f2de 34.646000 Wash
Focusing on the repetition of buy and sell, a wash trade also refers to manipulative
11

trading or behavior of providing false impression to market participants. Recall that in
general, investors interpret significant changes in trading volume or price as the degree of
market attention. Two investors can generate fake signals by buying and selling one NFT
item at an unusually high price at the same time. This simple two participant wash sale
model is shown in Table 1. By only looking at the overall price index and trading volume, a
new investor would be trapped by wash traders and pay a significantly higher price.
The scholarly examination of this particular form of manipulative trading is limited
due to the inaccessibility of micro-level transaction data for most academics. However, with
the blockchain system, the NFT market structure is exceptionally useful to analyze wash
trades since each item has a distinguishable number, and all wallet addresses are revealed,
even without proprietary exchange data. Intuitively, it is less likely to argue that trades are
normal if one sells and buys the exact same item again out of similar items in a collection.
Wachter et al. (2022) suggested a graph theory-based algorithm that directly detects wash
sales in the NFT market. On the NFT industry or community side, a method used by Dune
Analytics7 has been used so far.
Table 2. Logic of Wash Trades Detection
Type Wash Type (1) Wash Type (2) Wash Type (3)
Name Identity Trade 1-1 Trade Matched Order
Transactions A Sell → A Buy A Sell → B Buy A Sell → B Buy
B Sell → A Buy B Sell → C Buy
C Sell → A Buy
Time Span - Within 7 days Within 7 days
Observations 346 8808 1183
In this paper, a wash trade is defined similar to Wachter et al. (2022) and Dune An-
alytics, a commercial company that reveals its detection algorithm, but time span is incor-
porated as in IRS definition. A wallet first buys an item at normal price as a preparation
step. As shown at Table 2, trades are classified as wash sales if (1) an item is sold and pur-
chased by exact same identity at the same time, (2) an item is purchased again by previous
seller within 7 days, or (3) as a matched order, 3 wallets are involved in trading and all
7
See this online community posting (https://community.dune.com/blog/nft-wash-trading-on-ethereum) for
his algorithm.
12

trades occurred within 7 days. Although IRS impose 30 days to define wash trades, it is too
long period in NFT and cryptocurrency markets as we can check in Table 1. The result is
consistent with shorter periods such as 3 or 5 days (8564 and 8705 observations each for
1-1 trade). Even with this simple definition, we can identify a total of 10166 (0.3%) suspi-
cious wash trades out of 3.6 million secondary market transactions. Additionally, 44% of
the 558 collections contained at least one suspicious wash trade, despite the average wash
trade volume in each collection being just 0.3%. Detailed summary statistics are shown at
Table A.1.
.15
.1
Fraction
.05
0
0 30 60 90 120 150 180 210 240 270 300 330 360 390 420 450 480 510 540 570 600 630 660
Days between wash and first mints > 0
Figure 4. Timing of Wash Trade in Secondary Market

Notes. This figure reports the distribution of wash trade ratio out of secondary market trade
in collection-wise. Collections that do not have wash trades are omitted in the figure.
Another noteworthy aspect of wash trades is their timing. The above plot of Figure 4
illustrates the histogram of elapsed days from the first mint sales to wash trades in each
collection. More than 20% of wash trades occurred within 60 days after mints but more
mature collections also had wash trades. Combining with around half of collection have
at least one wash trades, wash trades may be market-wide phenomena. The idea in this
subsection can be summed up with the following prediction.
Prediction 2: The actions of wash traders are likely to influence a collection’s future
price returns. This is due to investors potentially being unaware of wash trades, thereby be-
ing drawn to the skewed prices. In essence, the vast array of information, even if accessible,
13

may overwhelm investors, preventing them from fully processing detailed transaction data
and collection information as outlined in Banerjee, Davis, and Gondhi (2018).
However, this prediction might not hold, given the propensity of the NFT community to
voluntarily report suspicions on platforms like Twitter. Furthermore, the scope of accessible
information is inherently limited, as transaction analyses are fragmented into item-level
data, facilitating the detection of manipulations at the collection-item level. This feature
is analogous to housing markets, particularly in the realm of apartment sales. General
upward price trends of an entire apartment can be discerned, but specific transaction price
histories of individual units within the apartment can also be scrutinized, especially when
abnormal price spikes are suspected.
Prediction 2A: Behaviors of wash traders may not influence a collection’s future price
returns, as investors can detect wash trades and avoid being drawn to distorted prices.
Investors can effectively process a limited volume of information.
3 Data
The list of NFT collections was manually compiled in October 2021 from the "Top Col-
lectibles NFT rankings" on OpenSea, the largest NFT trading platform. The NFT list ex-
tends the sample in Oh, Rosen, and Zhang (2022) with some new successfully launched
projects after October 2021 and before December 2021. After only selecting collections that
successfully minted all items, the final sample consists of 558 ERC-721 NFT collections
traded in the Ethereum blockchain system. Transaction data is primarily obtained from
Dune Analytics, a commercial data company, but is also cross-checked at Etherscan, one
of the biggest free websites. Indirect trades involving DeFi platforms such as Uniswap
and Sushiswap are excluded, while direct ERC-1155 trades are included8 . The number of
mint transactions is 3.6 million, and the number of secondary transactions is 3.6 million
as well. To eliminate extremely high-priced outliers and unusual near 0 ETH transactions,
only secondary market trades of at least 0.01 ETH are considered in the sample, and all re-
8
ERC-1155 allows for batch transfers, i.e., multiple trades in a single smart contract. In ERC-721, one NFT
item is traded under one smart contract, thus ERC-1155 reduces a significant amount of transaction cost.
14

turn variables that will be discussed in Table 3 are further winsorized at the 1/99 percentile
level. The sample covers the period from February 17th, 2021 to February 14th, 2023, which
allows for the incorporation of the crypto winter in 2022.
1 7
.8
6.5
Log Median Price in USD

.6
Fraction
.4
5.5
.2
5
0
0 .2 .4 .6 .8 1 2 3 4 5 6
Daily Trading Volume/Collection Size Log Daily Trading Volume
Figure 5. Trading volume and Median Price of NFT

Notes. The figure left shows the daily secondary market trading volume divided by each
collection supply in collection-wise. The plot on the right depicts the square-root relationship
between logged daily median price and logged daily secondary market trading volume.
Figure 5 displays the illiquidity of NFT markets in plots. The left figure shows the
daily secondary market trading volume divided by each collection’s minted items. It is ev-
ident that transactions are rare compared to the number of issued items. The right plot
shows the positive square-root relation between daily median price and daily trading vol-
ume which is similar to the traditional price-volume relationship observed in finance. As a
result, investors pay attention to a collection’s trading volume since the market is illiquid on
average. This implies that increased investor attention and the introduction of new infor-
mation can significantly drive up the price (Wilkoff and Yildiz, 2023). Therefore, it is logical
to expect that information advantages, such as insider trading and false investor attention
from wash trading, may contribute to a collection’s investment return and longevity.
The variables used in the analysis are aggregated at the collection-day level, as shown
in Table 3. The dependent variables are the rate of median price and trading volume change,
with and without wash trades. The daily median price is used as the price index since
most NFT items are homogeneous, and the most common items in the collection are traded
around a similar price (Oh, Rosen, and Zhang, 2022). Wash trades can distort the represen-
15

Table 3. Variable Definitions
Notes. This table shows the definition of variables in this paper. Only at least 0.01 ETH
secondary market trades are considered in the sample and variables with † is further win-
sorized at 1/99 percentile level. Daily transaction volume less than 5 is also omitted from
the data. Note that dependent variables are leads.
Variables Description
Dependent Variables
†Price Return Rate of median price change from day t to t + 1
†Price Return nowash Rate of median price change from day t to t + 1, omitting all wash sales
†Volume Change Rate of trading volume change from day t to t + 1
†Volume Change nowash Rate of trading volume change from day t to t + 1, omitting all wash sales
Independent Variables
InsiderBuy Activity Free minters’ buying volume at day t scaled by the number of total minted items
and then standardized
Wash Activity Wash sales volume at day t scaled by the number of total minted items
and then standardized
Control Variables
Log(1+Days after mints) Log(1 + number of days past after first mint)
†Past Day Returns Rate of median price change from day t − 2 to t − 1
†Past Week Returns Rate of median price change from day t − 7 to t − 2
Log Market Value of Collection Log(Median Price × Trading Volume at day t)
Dummy category Art 1 if the purpose of an NFT collection is related to pure art (used as baseline)
Dummy category Gaming 1 if the purpose of an NFT collection is related to games
Dummy category Metaverse 1 if the purpose of an NFT collection is related to Metaverse
Dummy category Social 1 if the purpose of an NFT collection is related to social group
Dummy has twitter url 1 if an NFT collection has its own twitter account
Dummy has website url 1 if an NFT collection has its own website
Dummy has roadmap 1 if an NFT collection has roadmap for its project
Dummy artist name 1 if creators revealed their name (including nickname)
tative market price and trading volume of NFTs; therefore, it is more appropriate to consider
values that account for wash sales, which are prevalent in the experiences of most novice
traders. While investors typically focus on the floor price, which is the minimum available
list price at that time, median price is the best possible measure for the price index due to
data constraints.
The independent variables are insider buy volume and wash trade volume at day t,
both scaled by the total minted amount in each collection. Insiders’ sell volume is not in-
cluded, as it is challenging to distinguish routine trades from information-based trades of
free minters as discussed in subsection 2.2. The return variables are winsorized at 1/99
percentile level. Summary statistics of the variables are presented in Table 4. Note that
the number of observations of dependent variables is not equal. This indicates there are
16

cases where all transactions in a whole day involve wash trades. Secondary market trading
volume, daily median price in USD, are daily median price in USD omitting wash sales are
not winsorized in the table.
Table 4. Summary Statistics

Notes. This table shows the summary statistics of variables defined at Table 3. Only at
least 0.01 ETH secondary market trades are considered in the sample and variables with †
are further winsorized at 1/99 percentile level. Daily transaction volume less than 5 is also
omitted from the data. Secondary market trading volume, daily median price in USD, and
daily median price in USD omitting wash sales are not winsorized.
(1) (2) (3) (4) (5) (6)

VARIABLES N mean sd min p50 max
†Price Return nowash 58,213 0.0330 0.261 -0.544 0.000473 1.275

†Price Return 58,206 0.0328 0.260 -0.544 0.000538 1.262
†Volume Change 58,220 0.245 1.097 -0.833 -0.0357 6.500
†Volume Change nowash 58,214 0.245 1.097 -0.833 -0.0357 6.500
InsiderBuy Activity 75,399 2.09e-09 1.000 -0.112 -0.112 82.37
Wash Activity 75,399 4.59e-10 1.000 -0.0207 -0.0207 106.4
Wash Dummy 75,399 0.0172 0.130 0 0 1
Days between wash and first mint sales 75,399 175.0 143.2 0 138 711
†Past Day Returns 50,393 0.0271 0.248 -0.544 -0.00146 1.275
†Past Week Returns 46,591 0.0562 0.429 -0.620 -0.0264 2.232
Market Value of Collection 75,386 2.256e+07 1.748e+08 11,298 2.297e+06 5.675e+09
Dummy category Gaming 75,399 0.0706 0.256 0 0 1
Dummy category Art 75,399 0.0170 0.129 0 0 1
Dummy category Metaverse 75,399 0.0605 0.238 0 0 1
Dummy category Social 75,399 0.852 0.355 0 1 1
Dummy Has Twitter 75,399 0.989 0.105 0 1 1
Dummy Has Website 75,399 0.987 0.113 0 1 1
Dummy Has Roadmap 75,399 0.606 0.489 0 1 1
Dummy Artist Name 75,399 0.569 0.495 0 1 1
Positive Priced Secondary Market Trading Volume (Raw) 75,399 47.16 177.2 5 13 7,995
Daily median price in USD (Raw) 75,386 2,405 17,552 9.699 282.9 567,495
Daily median price in USD omitting wash sales (Raw) 75,378 2,359 16,949 9.699 282.7 540,458
4 The Impact of Manipulative Trades
In this section, I examine the impact of manipulative trades on both the rate of price index
return and trading volume changes using predictive regressions. The regression structure is
similar to Cohen, Malloy, and Pomorski (2012). I focus on analyzing the collection-day level,
as simple values such as index price and trading volume are easily accessible but still vital
market signals for investors. The dependent variables are the rate of index price change or
17

trading volume change, and the main independent variables are insider buy or wash activity
scaled by the amount of mints. For a collection c and day t, the baseline regression model is
as below:
DV c,t+1 = β1 I nsiderBu yActivit yc,t + β2 W ashActivit yc,t + γ X c,t + FE c + u c,t+1 (1)
where DV c,t+1 is the median price return or volume change variable from day t to t + 1 as
defined in Table 3. The main independent variables are insider buy and wash volume of
collection c at day t divided by the total minted amount of collection c. X c,t is a control
variable matrix at day t, and FE c is date fixed effects.
For the choice of control variables, I assume that investors focus on the past day return
from day t − 2 to t − 1 and past week return from day t − 7 to t − 2 as momentum factor.
Market value of collection which is median price times total minted volume is considered,
and as a general classification of NFTs that are either arts, gaming, metaverse, or social is
included. Dummy variable whether a collection is arts is used as a baseline. Lastly, quality-
related information such as the existence of a collection Twitter account, collection website,
roadmap, presence of artist for a collection is also considered. Regression tables with all
control variables are in the appendix.
Note that the daily index price and trading volume can be measured in two ways. The
first, which is what most investors observe on trading platforms, is the total or nominal
value that includes manipulative trades. The other is the true or real value, which excludes
wash trades since wash trades distort the price and trading volume. I present both esti-
mated results in Table 5 that uses nominal information and Table 6 that only considers real
information.
The estimated result without omitting wash trades is shown at Table 5. Columns (1)
and (2) show regression results with only insider buying term. Column (1) explains returns
on day t + 1 without date fixed effects while Column (2) regress with date fixed effects. Both
have positive coefficients and p-values close to 0. One standard deviation increase in Insider-
I nsiderBu yV olume
Buy Activity or T otalM inted lead to 2.3, or 2.1 percentage points increase in future daily
index returns, controlling for other factors. Columns (3) and (4) of Table 5 present results on
18

Table 5. Performance of Manipulative Trades: With Wash Trades
Notes. In this table, I report the results from estimates of specification (1) in which I regress
future median price returns on a daily activity of insider and wash trade activity scaled
by NFT collection-size for collection c as of day t. The dependent variable, Return c,t+1 ,
represents the rate of median price change in USD from day t to day t + 1. Control variables
are a day before price return, weekly price return, collection age, market value of collection,
and other collection characteristics. Standard errors are clustered by collection. t-statistics
are in parentheses. ∗ p < 0.10; ∗∗ p < 0.05; ∗∗∗ p < 0.01.
(1) (2) (3) (4) (5) (6)

VARIABLES Price Price Price Price Price Price
InsiderBuy Activity 0.0231*** 0.0213*** 0.0231*** 0.0213***

(5.604) (5.274) (5.605) (5.275)
Wash Activity -0.00166*** -0.00147*** -0.00168*** -0.00148***
(-3.595) (-3.675) (-3.531) (-3.642)
Log(1+Days after mints) 0.00483*** 0.0223*** 0.00344*** 0.0202*** 0.00486*** 0.0223***
(3.771) (8.077) (2.716) (7.369) (3.792) (8.082)
Past Day Returns -0.0169** -0.0222*** -0.0161** -0.0216*** -0.0169** -0.0222***
(-2.175) (-2.864) (-2.071) (-2.792) (-2.175) (-2.863)
Past Week Returns -0.00193 -0.00609* -0.00145 -0.00591* -0.00195 -0.00610*
(-0.643) (-1.934) (-0.484) (-1.877) (-0.651) (-1.938)
Log Market Value of Collection -0.0167*** -0.0170*** -0.0169*** -0.0171*** -0.0167*** -0.0170***
(-6.075) (-6.678) (-6.091) (-6.658) (-6.074) (-6.674)
Observations 39,838 39,814 39,838 39,814 39,838 39,814

Collection Controls YES YES YES YES YES YES
Date FE NO YES NO YES NO YES
Within Adj R-squared 0.0154 0.0158 0.0134 0.0140 0.0154 0.0158
Adj R-squared 0.0154 0.0602 0.0134 0.0585 0.0154 0.0602
wash trades. The coefficient of Wash Activity is negative and near 0 p-value, meaning that
wash trades decrease the future nominal return. One standard deviation increase in Wash
Activity induces around 0.1 percentage point decrease in daily price return. Thus actual
economic significance is negligible even if it is daily price return. These results are consis-
tent in Columns (5) and (6) with two variables (InsiderBuy, Wash Activity) combined. As in
Columns (2) and (4), insiders buy do meaningfully increase future returns but wash trades
is economically insignificant. The economic impact is similar to that of (2).
Next I examine the impact of misconduct behavior on real market value that is without
wash sales in index price calculation. The structure is exactly same as Table 5 and the
estimated result is presented at Table 6. Surprisingly, the standardized coefficients are
very similar to nominal outcomes, and again, Wash Activity slightly decreases the real price
19

returns, but its impact on economic significance is small. One standard deviation increase
in InsiderBuy Activity leads to a 2.3 or 2.1 percentage point increase in future daily index
returns, and one standard deviation increase in Wash Activity leads to a 0.1 percentage
point decrease in future daily index returns.
Table 6. Performance of Manipulative Trades: Without Wash Trades

by NFT collection-size for collection c as of day t. The dependent variable is Return c,t+1
which is the rate of median price change in USD from day t to day t + 1 omitting all trades
that are classified as wash trades. Control variables are a day before price return, weekly
price return, collection age, market value of collection, and other collection characteristics.
Standard errors are clustered by collection. t-statistics are in parentheses. ∗ p < 0.10; ∗∗ p <
0.05; ∗∗∗ p < 0.01.
(1) (2) (3) (4) (5) (6)

(5.610) (5.274) (5.611) (5.274)
Wash Activity -0.00157*** -0.00138*** -0.00159*** -0.00139***
(-3.809) (-3.885) (-3.750) (-3.861)
(3.744) (8.023) (2.684) (7.315) (3.764) (8.028)
Past Day Returns -0.0168** -0.0222*** -0.0160** -0.0216*** -0.0168** -0.0222***
(-2.175) (-2.867) (-2.070) (-2.794) (-2.175) (-2.866)
Past Week Returns -0.00222 -0.00644** -0.00174 -0.00626** -0.00224 -0.00645**
(-0.736) (-2.033) (-0.578) (-1.975) (-0.743) (-2.037)
(-6.073) (-6.675) (-6.089) (-6.657) (-6.072) (-6.672)
Observations 39,838 39,814 39,838 39,814 39,838 39,814

Adj R-squared 0.0153 0.0602 0.0133 0.0585 0.0153 0.0602
However, these results are somewhat confusing, given that wash trades are typically
conducted at high ETH prices and can distort the market price as in Table 1. It is un-
clear whether most investors realize the unusual market outcome while wash trading, even
though they can check through free websites that provide detailed records. It is possible
that wash trades have temporary effects on market outcomes that do not persist beyond a
single day.
To investigate this possibility, I test a modified version of equation 1 in Table 7, in
20

which I regress the rate of median price change in USD from day t − 1 to day t (i.e., same-
day return) on daily activity of Free Minters scaled and wash trade activity scaled by NFT
collection-size for collection c as of date t, omitting all wash trades. The control variables
are the same as in the previous estimations.
Table 7. Performance of Manipulative Trades: Same Day Without Wash Trades

future median price returns on a daily activity of insider and wash trade activity scaled by
NFT collection-size for collection c as of day t. The dependent variable is Return c,t which
is the rate of median price change in USD from day t − 1 to day t omitting all trades that
are classified as wash trades. Control variables are a day before price return, weekly price
return, collection age, market value of collection, and other collection characteristics. Stan-
dard errors are clustered by collection. t-statistics are in parentheses. ∗ p < 0.10; ∗∗ p < 0.05;
∗∗∗
p < 0.01.
(1) (2) (3) (4) (5) (6)

(5.527) (5.084) (5.527) (5.083)
Wash Activity -0.000613 -0.000255 -0.000634 -0.000269
(-0.576) (-0.238) (-0.606) (-0.254)
Log(1+Days after mints) -0.00361** 0.00476* -0.00563*** 0.00180 -0.00360** 0.00476*
(-2.457) (1.656) (-4.005) (0.625) (-2.448) (1.657)
Past Day Returns -0.221*** -0.246*** -0.220*** -0.245*** -0.221*** -0.246***
(-18.86) (-22.13) (-18.61) (-21.94) (-18.86) (-22.13)
Past Week Returns -0.0247*** -0.0362*** -0.0240*** -0.0360*** -0.0247*** -0.0362***
(-6.290) (-9.029) (-6.113) (-8.950) (-6.291) (-9.029)
Log Market Value of Collection 0.00469*** 0.00513*** 0.00437*** 0.00499*** 0.00469*** 0.00513***
(3.437) (2.988) (3.336) (2.985) (3.437) (2.988)
Observations 42,946 42,922 42,946 42,922 42,946 42,922

Adj R-squared 0.0538 0.104 0.0500 0.101 0.0538 0.104
The estimated results in Table 7 continue to support the argument that wash trades
have little effect on market outcomes. The coefficient of Wash Activity remains statistically
insignificant, while InsiderBuy Activity remains statistically significant and economically
meaningful (3.3 and 3 percentage points, respectively). This suggests that wash trades do
not influence the returns of NFTs; otherwise, we would anticipate an effect on same-day
returns.
Does it capture trivial mechanism? The demand for an illiquid status item increases
21

the price, which is nothing special. If insider buy and wash trades does not meaningfully
change the trading volume, then it is not a simple price-demand relation. The other dimen-
sion of market outcome I haven’t tested is trading volume. It can be examined using similar
manner as in previous estimation with same control variables.
Table 8. Impact of Manipulative Trades on Trading Volume: Without Wash Trades

the rate of change in daily trading volume on a daily activity of insider and wash trade
activity scaled by NFT collection-size for collection c as of day t. The dependent variable
is Volume Change c,t+1 which is the rate of daily trading volume change from day t to day
t + 1 omitting all trades that are classified as wash trades. Control variables are a day
before price return, weekly price return, collection age, market value of collection, and other
collection characteristics. Standard errors are clustered by collection. t-statistics are in
parentheses. ∗ p < 0.10; ∗∗ p < 0.05; ∗∗∗ p < 0.01.
(1) (2) (3) (4) (5) (6)

VARIABLES Volume Volume Volume Volume Volume Volume
InsiderBuy Activity -0.119*** -0.118*** -0.119*** -0.118***

(-9.289) (-9.200) (-9.284) (-9.196)
Wash Activity -0.00985*** -0.0107*** -0.00978*** -0.0106***
(-3.416) (-3.773) (-3.499) (-3.825)
(7.913) (6.388) (9.820) (7.844) (7.942) (6.398)
Past Day Returns -0.112*** -0.131*** -0.116*** -0.134*** -0.112*** -0.131***
(-4.495) (-4.812) (-4.672) (-4.931) (-4.494) (-4.809)
(-2.822) (-3.379) (-3.144) (-3.537) (-2.837) (-3.388)
(-7.697) (-7.482) (-7.772) (-7.638) (-7.696) (-7.474)
Observations 39,838 39,814 39,838 39,814 39,838 39,814

Adj R-squared 0.00597 0.0255 0.00354 0.0232 0.00601 0.0256
The result at Table 8 shows that the impact of manipulative behavior on the rate of
change in future daily trading volume without wash sales. Columns (1) and (2) show one
standard deviation change in insider buying ratio decreases around 12 percentage points
future volume change. Columns (3) and (4) displays one standard deviation change in wash
trades leads to less than 1 percentage points decrease in future trading volume. Note that
the average daily trading volume is 47, so 12 percentage points decrease in trading vol-
22

ume is around five transactions on average. These values indicate a relatively marginal
change in trading volume that is not economically significant in an illiquid market, a result
consistently shown in Columns (5) – (6) as well.
In summary, the results suggest that insider buying strongly predicts higher future
price index returns, while wash trades do not have a significant impact on the returns.
Therefore, investors are not lured into NFT collections by wash traders, but insiders still
have an advantage due to their internal information.
5 Online Community as Information Channel
To provide further evidence of insiders’ information advantage, this section explores the
heterogeneity of insiders’ behavior by examining both their purchase and sale activities. An
insider’s buying behavior can be categorized into two types: purchases made while already
holding other NFTs in their collection, and those made when no NFTs are held, typically
following previous sales. While not as clearly distinguishable, selling behavior can similarly
be grouped into insiders selling while still holding other NFTs in a collection, and those who
divest all their NFTs in a collection, effectively exiting.
Figure 6. Example of Member-only Chatrooms

Notes. This screenshot provides an illustration of a members-only chatroom and exclusive
contents from Supducks, utilizing an automatic ownership verification system. This figure
is not intended to imply or suggest the existence of misconduct or inappropriate behaviors
associated with Supducks.
Insiders who maintain NFT holdings may have an added advantage, as they can uti-
lize information from members-only chat rooms to guide their buying or selling decisions.
Figure 6 shows there are exclusive contents and chatroom for verified members. In essence,
by retaining their NFTs, insiders preserve access to community-specific knowledge linked
23

to their collection. The variable InsiderBuy Activity × Additional captures the additional
purchasing behavior of insiders when they already hold at least one different NFT within
their collection. Similarly, InsiderSell Activity × Additional represents the insiders’ addi-
tional selling behavior when they already possess at least one different NFT.
Conversely, wallets who do not have any NFTs are excluded from the privileged com-
munity and its associated benefits. The variable InsiderBuy Activity × Not Additional cap-
tures insiders’ trading behavior when they do not have any NFTs in their collection, indi-
cating a purchase without information advantage. Those insiders are who try to reenter the
collection without community access. InsiderSell Activity × Not Additional shows an exit of
insiders and it might show that there is negative information about the collection and those
insiders may choose to escape from the community. Alternatively, these exits may simply
reflect insiders reaching their investment goals and choosing to liquidate their NFTs.
Table 9 shows results from regression analysis using similar specification. Column (1)
– (2) is copied from column (1) – (2) of Table 6. Columns (3) and (4) show that InsiderBuy Activity×
Additional is statistically significant while InsiderBuy Activity × Not Additional is insignif-
icant when controlled. It means that not all insiders (free minters) may obtain information
advantage, but only insiders with access to the community can gain advantage. The sizes
of the standardized coefficients are 2.2 and 2.0, respectively, and are similar to columns (1)
– (2). For selling behavior in Columns (5) and (6), both InsiderSell Activity terms are either
statistically insignificant or weakly significant. Therefore, insiders selling behavior is less
likely to be associated with future price returns.
In Table 10, where I use InsiderSell Activity terms for regression on the same day re-
turns, both terms are still weakly significant. However, InsiderSell Activity terms are more
consistently associated with same day price returns than Table 9. One standard deviation
increase in InsiderSell Activity×Additional is associated with 2.4 or 2.1 percentage points
increase in the same day price returns. One standard deviation increase in InsiderSell Ac-
tivity×Not Additional is associated with 3.3 or 3.4 percentage points increase in the same
day price returns. Equivalently, insiders sell behavior is positively associated with current
price returns both when insiders keep holding at least one NFTs, or when insiders exit the
NFT collection. Combining with findings that (1) only insiders who maintain the connection
24

Table 9. Heterogeneity in Insider Behavior
future median price returns on a daily trades of insider buy and sell activity scaled by NFT
collection-size for collection c as of day t. The dependent variable is Return c,t+1 which is
the rate of median price change in USD from day t to day t + 1 omitting all trades that
∗∗∗
p < 0.01.
(1) (2) (3) (4) (5) (6)

InsiderBuy Activity 0.0231*** 0.0213***

(5.610) (5.274)
InsiderBuy Activity x Additional 0.0220*** 0.0203***
(5.617) (5.277)
InsiderBuy Activity x Not Additional -0.00335 -0.00395
(-0.523) (-0.685)
InsiderSell Activity x Additional 0.00427 0.00210
(0.605) (0.350)
InsiderSell Activity x Not Additional 0.0201* 0.0220**
(1.832) (2.305)
Observations 39,838 39,814 39,838 39,814 39,838 39,814

Adj R-squared 0.0153 0.0602 0.0153 0.0602 0.0147 0.0598
to creators strongly predict future returns in purchase, (2) insiders sell is statistically in-
significant or weak in predicting future returns, and (3) insiders sell is still statistically not
strong but better at explaining current returns, it is more clear that insiders who are in the
community are exploiting information advantage.
25

Table 10. Heterogeneity in Insider Sell Behavior: Same day
future median price returns on a daily trades of insider sell activity scaled by NFT collection-
size for collection c as of day t. The dependent variable is Return c,t which is the rate of
median price change in USD from day t − 1 to day t omitting all trades that are classified as
wash trades. Control variables are a day before price return, weekly price return, collection
age, market value of collection, and other collection characteristics. Standard errors are
clustered by collection. t-statistics are in parentheses. ∗ p < 0.10; ∗∗ p < 0.05; ∗∗∗ p < 0.01.
(1) (2)
VARIABLES Price Price
InsiderSell Activity x Additional 0.0244** 0.0212**

(2.541) (2.574)
InsiderSell Activity x Not Additional 0.0334** 0.0347**
(2.106) (2.212)
Observations 42,946 42,922

Collection Controls YES YES
Date FE NO YES
Within Adj R-squared 0.0597 0.0705
Adj R-squared 0.0597 0.109
6 Purpose of Wash Trade and Investor Behaviors
In fact, the money traded through wash sales identified in the sample reaches a total of 422
million USD. It’s crucial to highlight that, despite the significant amounts involved, wash
trades don’t seem to be successful in drawing investors as in prediction 2 on subsection 2.3.
This empirical observation prompts an exploration into the mechanism of wash trading
and whether NFT investors actively process manageable volumes of information. Do NFT
investors actively digest freely available transaction history data, wash trade warnings from
Twitter, the collection’s Discord and Twitter accounts? And subsequently, do investors avoid
purchasing suspicious collections and items during the NFT boom? An affirmative response
would reinforce discussions on rational investor behaviors in finance.
Additionally, Aggarwal and Wu (2006) and Massoud, Ullah, and Scholnick (2016) have
explored the potential involvement of insiders in manipulative trades. Insiders may be
motivated to partake in trades that garner attention, given the potential to create upward
price momentum via pump-and-dump schemes (Li, Shin, and Wang, 2022), or to secure
26

proportional royalties from each transaction.
Table 11. Wash Trades and NFT Marketplaces

Notes. In this table, I present the descriptive statistics related to NFT marketplaces and
their potential wash trades. The numbers represent the observations out of secondary mar-
ket trades in the 558 collection sample, and the figures in parentheses indicate the percent-
age of trades in that particular marketplace. The marketplace fee policy data is as of March
5th, 2023. Foundation and Zora will be omitted in the following analysis due to the small
number of observations.
NFT Marketplaces Not Wash Trade Wash Trade Total Related Policy
Blur 39124 1460 40584 0% fee
(96.40) (3.60) (100.00) Receive token when traders pay full royalty to creators
Element 625 123 748 0.5% fee
(83.56) (16.44) (100.00)
Foundation 2 0 2 5% fee
(100.00) (0.00) (100.00)
LooksRare 10542 176 10718 2% fee. Token stakers earn 75∼100% of the trading fees
(98.36) (1.64) (100.00)
OpenSea 3590664 1573 3592237 2.5% platform fee (temporarily 0% after the sample period)
(99.96) (0.04) (100.00)
Sudoswap 8589 205 8794 0.5% fee
(97.67) (2.33) (100.00)
X2Y2 24436 6628 31064 0.5% fee. Fees are rewarded to X2Y2 stakers
(78.66) (21.34) (100.00)
Zora 18 1 19 0% fee
(94.74) (5.26) (100.00)
Total 3674000 10166 3684166
(99.72) (0.28) (100.00)
Wash trades may not solely aim at market manipulation. Discussions around cryp-
tocurrency rewards in NFT marketplaces have surfaced, highlighted by the case of Look-
sRare and its reported 8 billion USD in NFT wash trading9 . Thus it is important to check
the distribution of wash trades in terms of marketplaces. Table 11 presents a two-way fre-
quency table of wash trades and NFT marketplaces, showing that OpenSea is the largest
and leading NFT marketplace. Furthermore, numerous marketplaces have policies in which
marketplace fees are compensated with marketplace coins or offer near 0 percent fee com-
pared to OpenSea. It is reasonable to speculate that wash trades in NFT marketplaces
might be aiming to accrue profits through artificially boosted cryptocurrency rewards, or to
draw attention to emerging marketplaces.
Initially, a linear probability model is employed to explore the likelihood of wash trad-
9
See this article about suspicious tradings on LooksRare (https://decrypt.co/91510/looksrare-has-reportedly-
generated-8b-ethereum-nft-wash-trading).
27

ing across different marketplaces using transaction-level data from the secondary market.
Evaluating the roles of buyers and sellers as insiders allows for the examination of potential
insider involvement in wash trades. The binary dependent variable, WashTradeDummy, is
assigned 1 if a trade qualifies as a wash trade as outlined in subsection 2.3, and 0 otherwise.
The primary independent variable is a combination of the six NFT marketplaces listed in
Table 11, along with a combination of two dummy variables, InsiderBuyer Dummy and In-
siderSeller Dummy, which are set to 1 if the buyer or seller is a free minter, respectively.
Foundation and Zora are excluded due to the scarcity of observations. Control variables
include the logged holding period (expressed in fractional days), the log of the transaction
price, collection volume, and collection characteristics.
Table 12. Determinants of Wash Trades

Notes. In this table, I present the estimates of linear probability model related to wash
trades and their potential factors using secondary market transaction level data. The de-
pendent variable is WashTradeDummy that is 1 if a trade is denoted as a wash trade. De-
pendent variables are marketplace dummy, and interaction of insider buy and seller dummy.
Control variables are log holding period and log NFT transaction price. Standard errors are
(1) (2) (3)
VARIABLES Wash Sales Dummy Wash Sales Dummy Wash Sales Dummy
Marketplace = OpenSea (baseline) (baseline)
Marketplace = Blur 0.0360*** 0.0360***

(3.892) (3.891)
Marketplace = Element 0.181 0.181
(1.526) (1.526)
Marketplace = LooksRare 0.0238** 0.0238**
(2.026) (2.026)
Marketplace = Sudoswap 0.0326*** 0.0326***
(4.827) (4.822)
Marketplace = X2Y2 0.364*** 0.364***
(3.191) (3.192)
InsiderBuyer=0×InsiderSeller=0 (baseline) (baseline)
InsiderBuyer=0×InsiderSeller=1 0.000839 0.000992

(1.021) (1.379)
InsiderBuyer=1×InsiderSeller=0 -0.00109 -0.000303
(-0.840) (-0.380)
InsiderBuyer=1×InsiderSeller=1 7.63e-05 0.00131*
(0.0953) (1.896)
Observations 3,303,304 3,303,304 3,303,304

Collection Controls YES YES YES
Date FE YES YES YES
Within Adj R-squared 0.0213 0.245 0.245
Adj R-squared 0.115 0.317 0.317
28

The outcomes of the linear probability model are disclosed in Table 12. Column (1) de-
lineates that, when controlling for the holding period and NFT price, insiders do not amplify
the likelihood of a wash sale occurrence. Put simply, insiders are not actively engaging in
wash trades. Column (2) depicts the influence of the marketplace on the probability of wash
trades. When a transaction is in X2Y2 marketplace, the probability of being wash trades a
trade is wash is 36 percentage points higher compared to that in OpenSea, controlling for
holding period, price, collection volume, and characteristics. Wash trade probability at Su-
doswap and Blur is 3 percentage points higher, LooksRare is 2 percentage higher than that
at OpenSea. These results are consistent in Column (3). In summary, some marketplaces
have a higher probability of wash trades.
Next, I regress directly on the platform fees and royalty fees using transaction data
at Table 13. The dependent variable represents the USD value of marketplace fees to NFT
marketplaces in Columns (1) and (2), and loyalty fees to NFT creators in Columns (3) and
(4). In Column (1), without interaction terms between wash trade dummy and market-
places, LooksRare obtains an average marketplace fee around 99 dollars more compared
to OpenSea. Conversely, marketplaces like Blur or Sudoswap earned less than OpenSea
controlling for collection characteristics, holding period, mint volume, and dollar value of a
collection. When interaction terms are considered in Column (2), wash trades in LooksRare
generated more than 2,000 dollars of marketplace fee compared to the marketplace fee from
non-wash trades in OpenSea on average. However, in all other marketplaces, marketplace
fees from wash trades are less than those from non-wash trades from OpenSea. Note that
wash trades in LooksRare only consist 1.64 percentage of all transactions in LooksRare,
and the magnitude of wash trades in LooksRare was exceptionally large involving one of
the most prominent NFT collection, BoredApeYachtClub. Thus, it is logical to postulate
that the predominant intention behind wash trades is not the augmentation of marketplace
fees.
Loyalty fees serve as a direct source of revenue for NFT creators and can illuminate
creators’ motivations. Columns (3) and (4) of Table 13 shows the linear estimates of the
model. In Column (3), wash sales dummy is not statistically strong, indicating that wash
trades do not meaningfully alter loyalty fees. Compared to royalty fees from OpenSea, mar-
29

Table 13. Marketplace Platform Fees and Royalty Fees
(1) (2) (3) (4)
VARIABLES Marketplace Fee Marketplace Fee Loyalty Fee Loyalty Fee
Wash Sales Dummy = 1 -23.17 -120.9*

(-0.398) (-1.919)
Marketplace = Blur -108.1** -147.0***

(-2.207) (-2.683)
Marketplace = Element -69.35 -101.3
(-1.043) (-0.989)
Marketplace = LooksRare 99.08** 36.53
(2.198) (0.933)
Marketplace = Sudoswap -67.61** -111.6***
(-2.556) (-3.234)
Marketplace = X2Y2 -44.69 -49.74
(-1.531) (-1.226)
Wash Sales Dummy = 0 × Marketplace = Blur -105.3** -147.4***
(-2.211) (-2.736)
Wash Sales Dummy = 0 × Marketplace = Element -95.32 -156.4
(-1.390) (-1.629)
Wash Sales Dummy = 0 × Marketplace = LooksRare 49.11* 46.79
(1.695) (1.134)
Wash Sales Dummy = 0 × Marketplace = OpenSea (baseline) (baseline)
Wash Sales Dummy = 0 × Marketplace = Sudoswap -68.43** -116.4***

(-2.528) (-3.304)
Wash Sales Dummy = 0 × Marketplace = X2Y2 -50.39 -83.44*
(-1.443) (-1.875)
Wash Sales Dummy = 1 × Marketplace = Blur -206.0** -284.7***
(-2.120) (-2.828)
Wash Sales Dummy = 1 × Marketplace = Element 19.80 13.40
(0.754) (0.420)
Wash Sales Dummy = 1 × Marketplace = LooksRare 2,017*** -492.1***
(2.606) (-3.994)
Wash Sales Dummy = 1 × Marketplace = OpenSea -207.4 -310.8
(-1.526) (-1.629)
Wash Sales Dummy = 1 × Marketplace = Sudoswap -72.05** -122.5***
(-2.281) (-2.965)
Wash Sales Dummy = 1 × Marketplace = X2Y2 -59.02*** -113.7***
(-3.534) (-4.699)
Observations 3,303,304 3,303,304 3,303,304 3,303,304

Collection Controls YES YES YES YES
Date FE YES YES YES YES
Within Adj R-squared 0.127 0.130 0.173 0.173
Adj R-squared 0.132 0.136 0.182 0.182
ketplaces that are statistically significant exhibit negative coefficients, implying royalty fees
are reduced by 147 dollars in Blur, and by 111 dollars in Sudoswap. Column (4) portrays
30

that royalty fees derived from wash trade transactions are inferior to those from non-wash
trade transactions in OpenSea. Generally, loyalty fees are less in wash trades compared to
non-wash trades. It’s noteworthy that wash trades in LooksRare demonstrate around 492
dollars lesser in royalty fees compared to non-wash trades in OpenSea, revealing a contrary
trend in marketplace fees. This implies that creators do not benefit from wash trades.
So far, our analysis has confirmed that (1) several subsequent marketplaces exhibit a
higher probability of wash trades, (2) in general, marketplaces gain less marketplace fees
from wash trades compared to non-wash trades, and (3) creators earn reduced loyalty fees
from wash trades. From these observations, it’s reasonable to contend that neither mar-
ketplaces nor creators are involved in wash trades. Suppose some entities, unrelated to
marketplaces and creators, are generating wash trades. Then the question arises: are in-
vestors smart enough to avoid manipulations?
One approach to answer this is by investigating if investors continue to trade an item
after its wash trade has been completed. And if they do, is there a discernible difference in
returns compared to NFTs that haven’t undergone wash trades? Investors may avoid and
penalize items associated with wash trades. This hypothesis can be tested by comparing the
realized return when an item’s prior trade has been designated as a wash trade. As in Oh,
Rosen, and Zhang (2022), the realized return for a collection c, item i , purchased at τ, sold
P rice c,i,t
at t is defined as R eal izedR eturn c,i,τ,t = P rice c,i,τ − 1, without considering gas, royalty, and
marketplace fee. Further it is winsorized at the 1/99 percentile level.
Simple regression estimation using realized return is shown at Table 14. The inde-
pendent variable is Dummy previous is wash, which is 1 if a previous trade for the same
collection c, item i is wash trade, current trade is not wash trade, and previous buyer is
recorded as seller at current trade. Control variables are collection characteristic variables
and holding period. Column (1) describes that wash traded NFTs has 29.3 percentage points
lower realized returns than non-wash traded NFTs without control variables. The result at
Column (2) indicates that there is no difference in returns depending on the history that
previous trade is wash trade, controlling for holding period, collection volume, and other
collection characteristics. This is consistent in Column (3) where all transactions marked as
wash trades are eliminated for precise subsample analysis.
31

Table 14. Realized Returns After Wash Trades
Notes. In this table, I present the estimates that I regress realized return on the past wash
trade history. The independent variable is 1 if a previous trade for the same collection c,
item i is wash trade, current trade is not wash trade, and previous buyer is recorded as seller
at current trade. Control variables are log holding period, log collection volume, and other
collection quality characteristics. Standard errors are clustered by collection. t-statistics
(1) (2) (3)
VARIABLES Realized Return Realized Return Realized Return
Dummy Previous Is Wash -0.293** 0.218 0.231

(-2.094) (1.078) (1.096)
Observations 3,303,324 3,303,324 3,293,225

Collection Controls NO YES YES
Date FE YES YES YES
Within Adj R-squared 1.10e-06 0.00837 0.00830
Adj R-squared 0.0445 0.0525 0.0521
This then prompts the question agian: do investors refrain from trading other NFT
items within the same collection after the occurrence of wash trades? More specifically,
is there any notable long-term change at the collection level trading volume or price index
with certain time lags post the wash trades? This hypothesis can be tested under the similar
conditions as outlined in Equation 1, albeit with the time frame extended to longer periods,
such as 2, 3, 5, 7, or 14 days post day t.
Panel A of Table 15 examines the impact of wash activity on long-term price index
returns. Column (1) is copied from Table 6 for readability. Columns (2) – (6) shows long-
term effect is statistically insignificant, indicating an absence of meaningful future median
price change after wash trades. Panel B of Table 15 depicts the impact of wash activity on
long-term volume change. Columns (2) – (6) shows that one standard deviation increase in
wash activity at day t leads to approximately from 1.5 to 3.2 percentage points decrease in
future volume change. However, as discussed in section 4, it is difficult to consider these
values as meaningful change.
In conclusion, the available evidence indicates that investors typically do not avoid
participating in transactions involving collections associated with wash trades. Both in
the short and long term, the market seemingly does not impose significant penalties on
32

collections with a history of wash trading, neither in terms of volume nor value. Abnormally
priced items are simply ignored by investors.
33

Table 15. Long-term Impact of Manipulative Trades: Without Wash Trades
future median price returns and future trading volume change on a daily activity of insider
and wash trade activity scaled by NFT collection-size for collection c as of day t. The de-
pendent variable of Panel A is long-term median price change in USD from day t to day
t + 1, t + 2, t + 3, t + 5, and t + 7 omitting all trades that are classified as wash trades. The de-
pendent variable of Panel B is trading volume change from day t to day t + 1, t + 2, t + 3, t + 5,
and t + 7 omitting all trades that are classified as wash trades. Control variables are a day
parentheses. ∗ p < 0.10; ∗∗ p < 0.05; ∗∗∗ p < 0.01.
(1) (2) (3) (4) (5) (6)

Panel A: Price Returns 1 Day 2 Day 3 Day 5 Day 7 Day 14 Day
Wash Activity -0.00138*** -0.0326 -0.0119 -0.0263 -0.0349 -0.0183

(-3.885) (-1.085) (-1.223) (-1.271) (-1.163) (-0.718)
Log(1+Days after mints) 0.0200*** -0.00915 -0.000871 -0.00307 -0.0245 -0.133
(7.315) (-0.312) (-0.0398) (-0.147) (-0.662) (-1.155)
Past Day Returns -0.0216*** 0.183 0.0789 0.0158 -0.0757 -0.178*
(-2.794) (0.885) (0.489) (0.213) (-1.386) (-1.943)
Past Week Returns -0.00626** 0.156 0.154 -0.0221 -0.0682* -0.124
(-1.975) (0.890) (0.763) (-0.896) (-1.673) (-1.410)
Log Market Value of Collection -0.0170*** -0.0525** -0.0494*** -0.0403*** -0.0500*** -0.0730***
(-6.657) (-2.213) (-3.362) (-3.858) (-3.679) (-2.635)
Observations 39,814 39,318 38,984 38,495 37,934 36,320

Date FE YES YES YES YES YES YES
Adj R-squared 0.0585 -0.00129 0.00573 -0.000482 0.000988 0.00500
(1) (2) (3) (4) (5) (6)
Panel B: Volume Change 1 Day 2 Day 3 Day 5 Day 7 Day 14 Day
Wash Activity -0.0107*** -0.0152*** -0.0172** -0.0222*** -0.0358*** -0.0321***

(-3.773) (-2.815) (-2.092) (-2.861) (-4.107) (-3.989)
Log(1+Days after mints) 0.0712*** 0.0587 0.102* 0.0567 0.147** 0.155***
(7.844) (0.985) (1.834) (0.635) (2.554) (2.745)
Past Day Returns -0.134*** -0.0398 -0.138 -0.248* -0.142 -0.250***
(-4.931) (-0.282) (-1.565) (-1.664) (-1.521) (-3.219)
Past Week Returns -0.0368*** -0.0605 -0.0645 -0.283*** -0.192*** -0.178***
(-3.537) (-1.074) (-1.559) (-3.493) (-4.189) (-3.416)
(-7.638) (-5.263) (-4.652) (-4.505) (-4.280) (-3.620)
Observations 39,814 39,318 38,984 38,495 37,934 36,320

Adj R-squared 0.0232 -0.000340 0.00846 0.00227 0.0157 0.0203
34

7 Conclusion
NFTs represent a new form of crowdfunding facilitated by blockchain technology. The unreg-
ulated yet data-transparent environment provides unique opportunities to analyze market
misconduct that is limited in traditional financial research. It is available to detect possi-
ble unrevealed insider and manipulative trading in NFT markets using publicly available
blockchain data. Insiders are investors who obtained free items in the primary market di-
rectly from creators, and wash trades are classified using 3 types of transactions similar to
the definition of the United States Internal Revenue Service. Insiders constitute 4.9% of the
total wallets that participated in the primary market, and wash trades account for 0.3% of
the 3.6 million transactions in the secondary market.
I examine the effect of misconduct behaviors on market outcomes for NFT projects
that successfully minted all items over March 2021 to January 2023. The results indicate
that insiders’ buying activities strongly predict future daily price index returns. However,
wash trading is economically insignificant. Moreover, insider purchases and wash trades
do not significantly affect future changes in trading volume. This suggests that unreported
insiders take an advantage of information asymmetry in NFT markets but wash trade is
actually ineffective to manipulate market outcomes. The channel of insiders’ predictability
is from information in online community.
Lastly, I checked the purpose of wash trades and NFT investor’s behaviors. The empir-
ical analysis indicates that rewarding platforms are highly associated with the occurrence
of wash trades, while insiders are not associated with wash trades. However, marketplaces
actually earned lower marketplace platform fees and creators received less royalty fees from
wash trades, indicating that external groups are associated with wash trades. In addition,
investors do not avoid trading collections related with wash trades both in the short and
long term. This suggests that investors can effectively process available information and
avoid abnormally priced items.
For further research, alternative measure can be considered. Investors care more
about the floor price which is minimum available listed price in each NFT collection. In-
stead of median price that can be affected a lot by wash trading, new outcomes can be
35

considered as an alternative independent variable. Another point to consider is the network
of wash traders. These wash traders’ identity and connection can be analyzed further.
36

References
Aggarwal, R. K., and G. Wu. 2006. Stock Market Manipulations. Journal of Business
79:1915–53.
Ahern, K. R. 2017. Information Networks: Evidence from Illegal Insider Trading Tips.
Journal of Financial Economics 125:26–47.
Aitken, M., D. Cumming, and F. Zhan. 2015. High Frequency Trading and End-Of-Day Price
Dislocation. Journal of Banking & Finance 59:330–49.
Allen, F., and D. Gale. 1992. Stock-Price Manipulation. Review of Financial Studies 5:503–
29.
Aloosh, A., and J. Li. 2019. Direct Evidence of Bitcoin Wash Trading. Working paper.
Anderson, R. C., D. M. Reeb, and W. Zhao. 2012. Family-Controlled Firms and Informed
Trading: Evidence from Short Sales. Journal of Finance 67:351–85.
Babich, V., S. Marinesi, and G. Tsoukalas. 2021. Does Crowdfunding Benefit Entrepreneurs
and Venture Capital Investors? Manufacturing & Service Operations Management
23:508–24.
Banerjee, S., J. Davis, and N. Gondhi. 2018. When Transparency Improves, Must Prices
Reflect Fundamentals Better? Review of Financial Studies 31:2377–414.
Bao, T., M. Ma, and Y. Wen. 2022. Herding in the Non-fungible Token (NFT) Market. Work-
ing paper.
Borri, N., Y. Liu, and A. Tsyvinski. 2022. The Economics of Non-Fungible Tokens. Working
Paper.
Cao, Y., Y. Li, S. Coleman, A. Belatreche, and T. M. McGinnity. 2016. Detecting Wash Trade
in Financial Market Using Digraphs and Dynamic Programming. IEEE Transactions on
Neural Networks and Learning Systems 27:2351–63.
Cohen, L., C. Malloy, and L. Pomorski. 2012. Decoding inside Information. Journal of
Finance 67:1009–43.
Cong, L. W., X. Li, K. Tang, and Y. Yang. 2023. Crypto Wash Trading. Management Science
.
Cumming, D., L. Hornuf, M. Karami, and D. Schweizer. 2021. Disentangling Crowdfunding

from Fraudfunding. Journal of Business Ethics 1–26.
Falk, B. H., G. Tsoukalas, and N. Zhang. 2022. Economics of NFTs: The Value of Creator
Royalties. Working paper.
Gandal, N., J. Hamrick, T. Moore, and T. Oberman. 2018. Price Manipulation in the Bitcoin
Ecosystem. Journal of Monetary Economics 95:86–96.
37

Grinblatt, M., and M. Keloharju. 2004. Tax-Loss Trading and Wash Sales. Journal of
Financial Economics 71:51–76.
Kong, D.-R., and T.-C. Lin. 2022. Alternative Investments in the Fintech Era: The Risk and
Return of Non-fungible Token (NFT). Working Paper.
Kräussl, R., and A. Tugnetti. 2022. Non-fungible tokens (NFTs): A Review of Pricing Deter-
minants, Applications and Opportunities. Working Paper.
Kyle, A. S., and S. Viswanathan. 2008. How to Define Illegal Price Manipulation. American
Economic Review: Papers & Proceedings 98:274–9.
Li, T., D. Shin, and B. Wang. 2022. Cryptocurrency Pump-and-Dump Schemes. Working
paper.
Massoud, N., S. Ullah, and B. Scholnick. 2016. Does It Help Firms to Secretly Pay for Stock
Promoters? Journal of Financial Stability 26:45–61.
Meoli, M., and S. Vismara. 2021. Information Manipulation in Equity Crowdfunding Mar-
kets. Journal of Corporate Finance 67:101866–.
Oh, S., S. Rosen, and A. L. Zhang. 2022. Investor Experience Matters: Evidence from Gen-
erative Art Collections on the Blockchain. Working paper.
Pourghannad, B., G. Kong, and L. Debo. 2020. Is Kindness the Magical Spell? The Role of
Information and Reciprocity in Revenue-Sharing Crowdfunding. Working paper.
Sun, H., and S. Yin. 2017. Information Leakage in Family Firms: Evidence from Short
Selling around Insider Sales. Journal of Corporate Finance 47:72–87.
Wachter, V. v., J. R. Jensen, F. Regner, and O. Ross. 2022. NFT Wash Trading: Quantifying
Suspicious Behaviour in NFT Markets. Working paper.
Wilkoff, S., and S. Yildiz. 2023. The Behavior and Determinants of Illiquidity in the Non-
fungible Tokens (NFTs) Market. Global Finance Journal 55:100782–.
38

Appendix
A Supplementary Materials
Table A.1. Summary Statistics

Notes. This table presents the summary statistics for insiders and wash trades, as defined
in subsection 2.2 and subsection 2.3. Insiders are identified as free minters who received
NFTs at no cost from the creators. Wash trades are classified as either (1) identity trades,
(2) 1-1 trades, or (3) matched orders. The observations in this table represent the aggregate
measures for each collection-level variable.
(1) (2) (3) (4) (5)

VARIABLES N mean sd min max
Panel A: Insider
Total Trading Volume 558 6,602 7,263 13 44,372
Collection Volume (# of Minted Items) 558 6,535 3,780 1,000 25,000
Insider Buying Volume 558 105.4 343.5 0 5,777
Insider Buy/Collection Volume 558 0.0323 0.0762 0 0.701
Wallets in Primary Market 558 1,528 1,113 61 7,724
Potential Insider Wallets (Free Minted) 558 61.61 164.8 0 1,964
Insider Wallets/Total Wallets in Primary Market 558 0.0464 0.105 0 0.725
Panel B: Wash Trade

Average # of Type 1 Wash Sales 558 0.620 3.617 0 67
Average # of Type 2 Wash Sales 558 15.78 166.3 0 3,375
Average # of Type 3 Wash Sales 558 2.120 15.86 0 297
Average # of Wash Sales 558 18.22 168.9 0 3,385
Collection Volume (Total # of Minted Items) 558 6,535 3,780 1,000 25,000
Average Type 1 Wash Sales Volume/Collection Volume 558 6.94e-05 0.000373 0 0.00670
Average Type 2 Wash Sales Volume/Collection Volume 558 0.00305 0.0484 0 1.125
Average Type 3 Wash Sales Volume/Collection Volumes 558 0.000238 0.00164 0 0.0297
Average Wash Sales Volume/Collection Volume 558 0.00332 0.0486 0 1.128
Has Type 1 Wash Sales 558 0.115 0.319 0 1
Has Wash Sales 558 0.432 0.496 0 1
39

Table A.2. Correlation Matrix
Notes. This table shows Pearson correlation coefficients of all variables used in Table 4. Each variables are (1) Price Return
nowash, (2) Price Return, (3) Volume Change, (4) Volume Change nowash, (5) InsiderBuy Activity, (6) Wash Activity, (7) Wash
Dummy, (8) Days between wash and first mint sales, (9) Past Day Returns, (10) Past Week Returns, (11) Market Value of
Collection, (12) Dummy category Gaming, (13) Dummy category Metaverse, (14) Dummy category Social, (15) Dummy Has
Twitter, (16) Dummy Has Website, (17) Dummy Has Roadmap, and (18) Dummy Artist Name. ∗ p < 0.10; ∗∗ p < 0.05; ∗∗∗ p <
0.01.
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18)
(1) 1.00
40
(2) 1.00∗∗∗ 1.00
(3) 0.13∗∗∗ 0.13∗∗∗ 1.00
(4) 0.13∗∗∗ 0.13∗∗∗ 1.00∗∗∗ 1.00
(5) 0.01∗ 0.01∗ -0.04∗∗∗ -0.04∗∗∗ 1.00
(6) -0.01 -0.01 -0.01 -0.01 -0.00 1.00
(7) 0.01∗ 0.01∗ 0.01 0.01 -0.08∗∗∗ 0.01∗ 1.00
(8) -0.01∗ -0.01∗ -0.02∗∗∗ -0.02∗∗∗ 0.00 -0.00 -0.00 1.00
(9) -0.00 -0.01 -0.01∗∗ -0.01∗∗ 0.02∗∗∗ -0.01 -0.03∗∗∗ -0.14∗∗∗ 1.00
(10) -0.02∗∗∗ -0.02∗∗∗ -0.01∗ -0.01∗∗ -0.01∗∗ 0.02∗∗∗ 0.08∗∗∗ -0.00 0.00 1.00
(11) 0.00 0.00 0.01∗ 0.01∗ -0.02∗∗∗ -0.00 0.02∗∗∗ 0.00 0.01∗ -0.02∗∗∗ 1.00
(12) 0.01∗∗ 0.01∗∗ -0.00 -0.00 0.01∗ 0.00 -0.05∗∗∗ 0.01 0.03∗∗∗ 0.01∗ -0.04∗∗∗ 1.00
(13) -0.00 -0.00 0.00 0.00 0.04∗∗∗ -0.00 0.07∗∗∗ -0.00 -0.01∗ -0.01∗∗ -0.07∗∗∗ -0.03∗∗∗ 1.00
(14) -0.00 -0.00 -0.01∗ -0.01 -0.02∗∗∗ 0.00 -0.05∗∗∗ -0.00 -0.01∗ 0.02∗∗∗ -0.66∗∗∗ -0.32∗∗∗ -0.61∗∗∗ 1.00
(15) 0.00 0.00 0.00 0.00 -0.01∗∗ 0.00 0.00 0.00 -0.00 0.01∗∗ -0.17∗∗∗ 0.01∗∗∗ 0.03∗∗∗ 0.10∗∗∗ 1.00
(16) -0.00 -0.00 0.00 0.00 -0.01 0.00 0.02∗∗∗ 0.00 0.01 0.01∗∗ -0.16∗∗∗ -0.03∗∗∗ 0.03∗∗∗ 0.10∗∗∗ 0.53∗∗∗ 1.00

(17) -0.00 -0.00 -0.00 -0.00 -0.01∗ -0.02∗∗∗ 0.02∗∗∗ -0.00 -0.02∗∗∗ 0.04∗∗∗ -0.06∗∗∗ -0.10∗∗∗ 0.17∗∗∗ -0.03∗∗∗ 0.07∗∗∗ 0.13∗∗∗ 1.00
(18) -0.01 -0.00 -0.00 -0.00 -0.01∗ -0.02∗∗∗ 0.04∗∗∗ -0.01 0.01 0.06∗∗∗ 0.00 -0.06∗∗∗ -0.07∗∗∗ 0.06∗∗∗ 0.07∗∗∗ 0.12∗∗∗ 0.25∗∗∗ 1.00
Table A.3. Performance of Manipulative Trades: With Wash Trades
future median price returns on a daily activity of insider and wash trade volume scaled
by NFT collection-size for collection c as of day t. The dependent variable, Return c,t+1 ,
represents the rate of median price change in USD from day t to day t + 1. Control variables
are a day before price return, weekly price return, collection age, market value of collection,
and other collection characteristics. Standard errors are clustered by collection. t-statistics
(1) (2) (3) (4) (5) (6)

(5.604) (5.274) (5.605) (5.275)
Wash Activity -0.00166*** -0.00147*** -0.00168*** -0.00148***
(-3.595) (-3.675) (-3.531) (-3.642)
(3.771) (8.077) (2.716) (7.369) (3.792) (8.082)
Past Day Returns -0.0169** -0.0222*** -0.0161** -0.0216*** -0.0169** -0.0222***
(-2.175) (-2.864) (-2.071) (-2.792) (-2.175) (-2.863)
Past Week Returns -0.00193 -0.00609* -0.00145 -0.00591* -0.00195 -0.00610*
(-0.643) (-1.934) (-0.484) (-1.877) (-0.651) (-1.938)
(-6.075) (-6.678) (-6.091) (-6.658) (-6.074) (-6.674)
Dummy category Gaming 0.00864 0.00379 0.00905 0.00441 0.00862 0.00380
(0.740) (0.366) (0.786) (0.432) (0.738) (0.367)
Dummy category Metaverse 0.000728 -0.00497 0.00463 -0.00139 0.000698 -0.00496
(0.0595) (-0.459) (0.368) (-0.125) (0.0570) (-0.459)
Dummy category Social -0.00592 -0.00813 -0.00504 -0.00730 -0.00591 -0.00809
(-0.604) (-0.906) (-0.527) (-0.832) (-0.604) (-0.904)
Dummy Has Twitter -0.00470 -0.0235 -0.0103 -0.0280** -0.00468 -0.0235
(-0.328) (-1.461) (-0.871) (-2.093) (-0.327) (-1.459)
Dummy Has Website 0.000525 0.00764 0.00482 0.0116 0.000565 0.00768
(0.0391) (0.496) (0.399) (0.841) (0.0421) (0.498)
Dummy Has Roadmap -0.00703* -0.00757** -0.00731** -0.00770** -0.00706* -0.00760**
(-1.939) (-2.185) (-2.018) (-2.229) (-1.948) (-2.193)
Dummy Artist Name 0.00543* 0.00259 0.00582** 0.00298 0.00538* 0.00255
(1.850) (0.898) (1.966) (1.026) (1.835) (0.884)
Observations 39,838 39,814 39,838 39,814 39,838 39,814

Adj R-squared 0.0154 0.0602 0.0134 0.0585 0.0154 0.0602
41

Table A.4. Performance of Manipulative Trades: Without Wash Trades
by NFT collection-size for collection c as of day t. The dependent variable is Return c,t+1
which is the rate of median price change in USD from day t to day t + 1 omitting all trades
0.05; ∗∗∗ p < 0.01.
(1) (2) (3) (4) (5) (6)

(5.610) (5.274) (5.611) (5.274)
Wash Activity -0.00157*** -0.00138*** -0.00159*** -0.00139***
(-3.809) (-3.885) (-3.750) (-3.861)
(3.744) (8.023) (2.684) (7.315) (3.764) (8.028)
Past Day Returns -0.0168** -0.0222*** -0.0160** -0.0216*** -0.0168** -0.0222***
(-2.175) (-2.867) (-2.070) (-2.794) (-2.175) (-2.866)
(-0.736) (-2.033) (-0.578) (-1.975) (-0.743) (-2.037)
(-6.073) (-6.675) (-6.089) (-6.657) (-6.072) (-6.672)
(0.894) (0.527) (0.940) (0.593) (0.893) (0.529)
Dummy category Metaverse 0.00235 -0.00339 0.00625 0.000184 0.00232 -0.00338
(0.195) (-0.315) (0.502) (0.0166) (0.192) (-0.314)
(-0.439) (-0.723) (-0.354) (-0.641) (-0.439) (-0.721)
Dummy Has Twitter -0.00408 -0.0227 -0.00968 -0.0271** -0.00406 -0.0226
(-0.286) (-1.416) (-0.820) (-2.039) (-0.284) (-1.414)
(0.0359) (0.491) (0.395) (0.837) (0.0386) (0.493)
Dummy Has Roadmap -0.00689* -0.00742** -0.00716** -0.00754** -0.00692* -0.00744**
(-1.909) (-2.151) (-1.988) (-2.194) (-1.917) (-2.158)
Dummy Artist Name 0.00546* 0.00264 0.00585** 0.00303 0.00542* 0.00260
(1.871) (0.919) (1.987) (1.048) (1.857) (0.905)
Observations 39,838 39,814 39,838 39,814 39,838 39,814

Adj R-squared 0.0153 0.0602 0.0133 0.0585 0.0153 0.0602
42

Table A.5. Performance of Manipulative Trades: Same Day Without Wash Trades
NFT collection-size for collection c as of day t. The dependent variable is Return c,t which
is the rate of median price change in USD from day t − 1 to day t omitting all trades that
∗∗∗
p < 0.01.
(1) (2) (3) (4) (5) (6)

(5.527) (5.084) (5.527) (5.083)
Wash Activity -0.000613 -0.000255 -0.000634 -0.000269
(-0.576) (-0.238) (-0.606) (-0.254)
Log(1+Days after mints) -0.00361** 0.00476* -0.00563*** 0.00180 -0.00360** 0.00476*
(-2.457) (1.656) (-4.005) (0.625) (-2.448) (1.657)
Past Day Returns -0.221*** -0.246*** -0.220*** -0.245*** -0.221*** -0.246***
(-18.86) (-22.13) (-18.61) (-21.94) (-18.86) (-22.13)
(-6.290) (-9.029) (-6.113) (-8.950) (-6.291) (-9.029)
Log Market Value of Collection 0.00469*** 0.00513*** 0.00437*** 0.00499*** 0.00469*** 0.00513***
(3.437) (2.988) (3.336) (2.985) (3.437) (2.988)
Dummy category Gaming -0.0178* -0.0179* -0.0169 -0.0169* -0.0178* -0.0180*
(-1.710) (-1.814) (-1.641) (-1.700) (-1.713) (-1.815)
Dummy category Metaverse -0.0310** -0.0361*** -0.0252** -0.0309*** -0.0310** -0.0361***
(-2.537) (-3.039) (-2.216) (-2.798) (-2.540) (-3.041)
Dummy category Social -0.0286*** -0.0294*** -0.0272*** -0.0281*** -0.0286*** -0.0294***
(-3.240) (-3.372) (-3.136) (-3.248) (-3.244) (-3.375)
Dummy Has Twitter 0.0179 0.00523 0.0108 -0.000240 0.0179 0.00524
(1.166) (0.428) (0.613) (-0.0158) (1.167) (0.429)
Dummy Has Website -0.00842 0.00136 -0.00337 0.00593 -0.00841 0.00137
(-0.507) (0.111) (-0.193) (0.438) (-0.506) (0.111)
Dummy Has Roadmap -0.00213 -0.000710 -0.00243 -0.000809 -0.00214 -0.000716
(-0.661) (-0.209) (-0.755) (-0.237) (-0.665) (-0.211)
Dummy Artist Name -0.00210 -0.00398 -0.00154 -0.00342 -0.00212 -0.00399
(-0.662) (-1.254) (-0.491) (-1.085) (-0.667) (-1.256)
Observations 42,946 42,922 42,946 42,922 42,946 42,922

Adj R-squared 0.0538 0.104 0.0500 0.101 0.0538 0.104
43

Table A.6. Impact of Manipulative Trades on Trading Volume: Without Wash Trades
the rate of change in daily trading volume on a daily activity of insider and wash trade
activity scaled by NFT collection-size for collection c as of day t. The dependent variable
is Volume Change c,t+1 which is the rate of daily trading volume change from day t to day
t + 1 omitting all trades that are classified as wash trades. Control variables are a day
parentheses. ∗ p < 0.10; ∗∗ p < 0.05; ∗∗∗ p < 0.01.
(1) (2) (3) (4) (5) (6)

VARIABLES Volume Volume Volume Volume Volume Volume
InsiderBuy Activity -0.119*** -0.118*** -0.119*** -0.118***

(-9.289) (-9.200) (-9.284) (-9.196)
Wash Activity -0.00985*** -0.0107*** -0.00978*** -0.0106***
(-3.416) (-3.773) (-3.499) (-3.825)
(7.913) (6.388) (9.820) (7.844) (7.942) (6.398)
Past Day Returns -0.112*** -0.131*** -0.116*** -0.134*** -0.112*** -0.131***
(-4.495) (-4.812) (-4.672) (-4.931) (-4.494) (-4.809)
(-2.822) (-3.379) (-3.144) (-3.537) (-2.837) (-3.388)
(-7.697) (-7.482) (-7.772) (-7.638) (-7.696) (-7.474)
(0.709) (0.707) (0.653) (0.631) (0.707) (0.711)
Dummy category Metaverse 0.0398 0.0421 0.0193 0.0223 0.0396 0.0422
(0.711) (0.745) (0.350) (0.400) (0.708) (0.748)
(-0.178) (-0.0560) (-0.265) (-0.142) (-0.178) (-0.0512)
Dummy Has Twitter -0.0548 -0.0781 -0.0256 -0.0528 -0.0546 -0.0779
(-0.771) (-1.022) (-0.301) (-0.571) (-0.769) (-1.019)
(1.404) (1.223) (0.993) (0.830) (1.406) (1.226)
Dummy Has Roadmap -0.0253** -0.0268** -0.0243** -0.0265** -0.0255** -0.0270**
(-2.388) (-2.421) (-2.363) (-2.450) (-2.403) (-2.439)
Dummy Artist Name 0.0145 0.0130 0.0120 0.0103 0.0142 0.0127
(1.393) (1.146) (1.199) (0.942) (1.367) (1.119)
Observations 39,838 39,814 39,838 39,814 39,838 39,814

Adj R-squared 0.00597 0.0255 0.00354 0.0232 0.00601 0.0256
44

Table A.7. Heterogeneity in Insider Behavior
future median price returns on a daily trades of insider buy and sell activity scaled by NFT
collection-size for collection c as of day t. The dependent variable is Return c,t+1 which is
the rate of median price change in USD from day t to day t + 1 omitting all trades that
∗∗∗
p < 0.01.
(1) (2) (3) (4) (5) (6)

InsiderBuy Activity 0.0231*** 0.0213***

(5.610) (5.274)
InsiderBuy Activity x Additional 0.0220*** 0.0203***
(5.617) (5.277)
InsiderBuy Activity x Not Additional -0.00335 -0.00395
(-0.523) (-0.685)
InsiderSell Activity x Additional 0.00427 0.00210
(0.605) (0.350)
InsiderSell Activity x Not Additional 0.0201* 0.0220**
(1.832) (2.305)
(3.744) (8.023) (3.739) (8.022) (3.174) (7.762)
Past Day Returns -0.0168** -0.0222*** -0.0168** -0.0222*** -0.0177** -0.0229***
(-2.175) (-2.867) (-2.175) (-2.867) (-2.286) (-2.969)
(-0.736) (-2.033) (-0.734) (-2.029) (-0.858) (-2.139)
(-6.073) (-6.675) (-6.071) (-6.674) (-6.094) (-6.673)
(0.894) (0.527) (0.893) (0.526) (0.809) (0.437)
Dummy category Metaverse 0.00235 -0.00339 0.00241 -0.00332 0.00346 -0.00238
(0.195) (-0.315) (0.200) (-0.308) (0.285) (-0.219)
(-0.439) (-0.723) (-0.437) (-0.721) (-0.464) (-0.754)
Dummy Has Twitter -0.00408 -0.0227 -0.00412 -0.0227 -0.00953 -0.0277*
(-0.286) (-1.416) (-0.289) (-1.420) (-0.747) (-1.939)
(0.0359) (0.491) (0.0379) (0.493) (0.327) (0.786)
Dummy Has Roadmap -0.00689* -0.00742** -0.00689* -0.00742** -0.00695* -0.00742**
(-1.909) (-2.151) (-1.909) (-2.151) (-1.927) (-2.160)
Dummy Artist Name 0.00546* 0.00264 0.00546* 0.00264 0.00605** 0.00320
(1.871) (0.919) (1.872) (0.921) (2.061) (1.112)
Observations 39,838 39,814 39,838 39,814 39,838 39,814

Adj R-squared 0.0153 0.0602 0.0153 0.0602 0.0147 0.0598
45

Table A.8. Heterogeneity in Insider Sell Behavior: Same Day
future median price returns on a daily trades of insider sell activity scaled by NFT collection-
size for collection c as of day t. The dependent variable is Return c,t which is the rate of
median price change in USD from day t − 1 to day t omitting all trades that are classified as
wash trades. Control variables are a day before price return, weekly price return, collection
age, market value of collection, and other collection characteristics. Standard errors are
(1) (2)
VARIABLES Price Price
InsiderSell Activity x Additional 0.0244** 0.0212**

(2.541) (2.574)
InsiderSell Activity x Not Additional 0.0334** 0.0347**
(2.106) (2.212)
Log(1+Days after mints) -0.00394*** 0.00419
(-2.703) (1.431)
Past Day Returns -0.225*** -0.249***
(-19.17) (-22.41)
Past Week Returns -0.0266*** -0.0376***
(-6.780) (-9.403)
Log Market Value of Collection 0.00440*** 0.00490***
(3.317) (2.916)
Dummy category Gaming -0.0204* -0.0205**
(-1.959) (-2.074)
Dummy category Metaverse -0.0328*** -0.0378***
(-2.693) (-3.186)
Dummy category Social -0.0300*** -0.0307***
(-3.361) (-3.507)
Dummy Has Twitter 0.0124 0.000333
(0.771) (0.0255)
Dummy Has Website -0.00581 0.00385
(-0.336) (0.296)
Dummy Has Roadmap -0.00204 -0.000668
(-0.635) (-0.197)
Dummy Artist Name -0.00114 -0.00300
(-0.362) (-0.949)
Observations 42,946 42,922

Collection Controls YES YES
Date FE NO YES
Within Adj R-squared 0.0597 0.0705
Adj R-squared 0.0597 0.109
46

Table A.9. Determinants of Wash Trades
(1) (2) (3)
VARIABLES Wash Sales Dummy Wash Sales Dummy Wash Sales Dummy
Marketplace = Blur 0.0360*** 0.0360***

(3.892) (3.891)
Marketplace = Element 0.181 0.181
(1.526) (1.526)
Marketplace = LooksRare 0.0238** 0.0238**
(2.026) (2.026)
Marketplace = Sudoswap 0.0326*** 0.0326***
(4.827) (4.822)
Marketplace = X2Y2 0.364*** 0.364***
(3.191) (3.192)
InsiderBuyer=0×InsiderSeller=0 (baseline) (baseline)
InsiderBuyer=0×InsiderSeller=1 0.000839 0.000992

(1.021) (1.379)
InsiderBuyer=1×InsiderSeller=0 -0.00109 -0.000303
(-0.840) (-0.380)
InsiderBuyer=1×InsiderSeller=1 7.63e-05 0.00131*
(0.0953) (1.896)
Dummy category Gaming -0.00938 -0.00819 -0.00820
(-0.764) (-0.664) (-0.664)
Dummy category Metaverse -0.00869 -0.00833 -0.00835
(-0.680) (-0.659) (-0.659)
Dummy category Social -0.00775 -0.00698 -0.00698
(-0.625) (-0.566) (-0.566)
Dummy Has Twitter 0.00686** 0.00582* 0.00584*
(2.016) (1.686) (1.703)
Dummy Has Website 0.000590 -0.000558 -0.000571
(0.216) (-0.188) (-0.193)
Dummy Has Roadmap 9.69e-05 0.000422 0.000420
(0.0571) (0.397) (0.396)
Dummy Artist Name -0.00173 -0.00119 -0.00118
(-1.009) (-1.129) (-1.126)
Log(1+Holding Period) -0.00462*** -0.00301*** -0.00301***
(-2.839) (-4.407) (-4.408)
Log(NFT Price) 0.00202** 0.00143* 0.00143*
(2.440) (1.751) (1.747)
Log(Mint Volume) -0.00320 -0.00206 -0.00206
(-0.723) (-0.798) (-0.798)
Observations 3,303,304 3,303,304 3,303,304

Collection Controls YES YES YES
Date FE YES YES YES
Within Adj R-squared 0.0213 0.245 0.245
Adj R-squared 0.115 0.317 0.317
47

Table A.10. Marketplace Platform Fees and Royalty Fees
(1) (2) (3) (4)
VARIABLES Marketplace Fee Marketplace Fee Loyalty Fee Loyalty Fee
Log(1+Holding Period) 10.07** 10.12** 12.73*** 12.77***

(2.159) (2.158) (2.796) (2.793)
Log(NFT Price) 69.73*** 69.68*** 100.2*** 100.4***
(2.844) (2.832) (4.075) (4.073)
Log(Mint Volume) -11.48 -11.29 -15.54* -15.14
(-1.482) (-1.477) (-1.657) (-1.623)
0b.wash_total#1.marketplace -105.3** -147.4***
(-2.211) (-2.736)
0b.wash_total#2.marketplace -95.32 -156.4
(-1.390) (-1.629)
0b.wash_total#4.marketplace 49.11* 46.79
(1.695) (1.134)
0b.wash_total#5b.marketplace 0 0
0b.wash_total#6.marketplace -68.43** -116.4***

(-2.528) (-3.304)
0b.wash_total#7.marketplace -50.39 -83.44*
(-1.443) (-1.875)
1.wash_total#1.marketplace -206.0** -284.7***
(-2.120) (-2.828)
1.wash_total#2.marketplace 19.80 13.40
(0.754) (0.420)
1.wash_total#4.marketplace 2,017*** -492.1***
(2.606) (-3.994)
1.wash_total#5b.marketplace -207.4 -310.8
(-1.526) (-1.629)
1.wash_total#6.marketplace -72.05** -122.5***
(-2.281) (-2.965)
1.wash_total#7.marketplace -59.02*** -113.7***
(-3.534) (-4.699)
Dummy category Gaming -28.75 -30.93 -31.87 -34.73
(-1.286) (-1.375) (-1.240) (-1.377)
Dummy category Metaverse -18.97 -21.21 -29.35 -32.28
(-0.693) (-0.769) (-0.954) (-1.062)
Dummy category Social -2.011 -4.343 0.0946 -2.798
(-0.164) (-0.373) (0.00546) (-0.175)
Dummy Has Twitter -0.724 -0.725 -2.356 -2.704
(-0.0299) (-0.0298) (-0.0789) (-0.0902)
Dummy Has Website -27.34 -26.88 -31.74 -31.45
(-1.128) (-1.114) (-1.154) (-1.147)
Dummy Has Roadmap 8.106 7.965 -3.797 -3.817
(0.544) (0.534) (-0.237) (-0.238)
Dummy Artist Name 4.891 4.892 3.559 3.554
(0.566) (0.565) (0.304) (0.303)
Wash Sales Dummy = 1 -23.17 -120.9*
(-0.398) (-1.919)
Marketplace = 1, blur -108.1** -147.0***
(-2.207) (-2.683)
Marketplace = 2, element -69.35 -101.3
(-1.043) (-0.989)
Marketplace = 4, looksrare 99.08** 36.53
(2.198) (0.933)
Marketplace = 6, sudoswap -67.61** -111.6***
(-2.556) (-3.234)
Marketplace = 7, x2y2 -44.69 -49.74
(-1.531) (-1.226)
Observations 3,303,304 3,303,304 3,303,304 3,303,304

Collection Controls YES YES YES YES
Date FE YES YES YES YES
Within Adj R-squared 0.127 0.130 0.173 0.173
Adj R-squared 0.132 0.136 0.182 0.182
48

Table A.11. Realized Returns After Wash Trades
Notes. In this table, I present the estimates that I regress realized return on the past wash
trade history. The independent variable is 1 if a previous trade for the same collection c,
item i is wash trade, current trade is not wash trade, and previous buyer is recorded as seller
at current trade. Control variables are log holding period, log collection volume, and other
collection quality characteristics. Standard errors are clustered by collection. t-statistics
(1) (2) (3)
VARIABLES Realized Return Realized Return Realized Return
Dummy Previous Is Wash -0.293** 0.218 0.231

(-2.094) (1.078) (1.096)
Log(1+Holding Period) 0.267*** 0.268***
(3.031) (2.987)
Log(Mint Volume) 0.174 0.173
(1.001) (0.993)
Dummy category Gaming 0.877 0.880
(1.515) (1.507)
Dummy category Metaverse 0.178 0.179
(0.292) (0.292)
Dummy category Social 0.567 0.569
(1.164) (1.154)
Dummy Has Twitter -0.132 -0.135
(-0.186) (-0.189)
Dummy Has Website 0.946** 0.949**
(2.279) (2.278)
Dummy Has Roadmap -0.552* -0.553*
(-1.831) (-1.832)
Dummy Artist Name 0.139 0.139
(0.564) (0.563)
Observations 3,303,324 3,303,324 3,293,225

Collection Controls NO YES YES
Date FE YES YES YES
Within Adj R-squared 1.10e-06 0.00837 0.00830
Adj R-squared 0.0445 0.0525 0.0521
49

Table A.12. Long-term Impact of Manipulative Trades: Without Wash Trades
NFT collection-size for collection c as of day t. The dependent variable is long-term median
price change in USD from day t to day t + 1, t + 2, t + 3, t + 5, and t + 7 omitting all trades
0.05; ∗∗∗ p < 0.01.
(1) (2) (3) (4) (5) (6)

VARIABLES 1 Day 2 Day 3 Day 5 Day 7 Day 14 Day
Wash Activity -0.00138*** -0.0326 -0.0119 -0.0263 -0.0349 -0.0183

(-3.885) (-1.085) (-1.223) (-1.271) (-1.163) (-0.718)
Log(1+Days after mints) 0.0200*** -0.00915 -0.000871 -0.00307 -0.0245 -0.133
(7.315) (-0.312) (-0.0398) (-0.147) (-0.662) (-1.155)
Past Day Returns -0.0216*** 0.183 0.0789 0.0158 -0.0757 -0.178*
(-2.794) (0.885) (0.489) (0.213) (-1.386) (-1.943)
Past Week Returns -0.00626** 0.156 0.154 -0.0221 -0.0682* -0.124
(-1.975) (0.890) (0.763) (-0.896) (-1.673) (-1.410)
Log Market Value of Collection -0.0170*** -0.0525** -0.0494*** -0.0403*** -0.0500*** -0.0730***
(-6.657) (-2.213) (-3.362) (-3.858) (-3.679) (-2.635)
(0.593) (1.174) (1.391) (0.883) (0.316) (0.875)
Dummy category Metaverse 0.000184 0.135 0.116 0.00450 0.0186 0.0295
(0.0166) (1.100) (1.270) (0.0597) (0.198) (0.183)
Dummy category Social -0.00564 0.177 0.140 0.00558 0.0266 0.0491
(-0.641) (1.107) (1.301) (0.0822) (0.275) (0.305)
Dummy Has Twitter -0.0271** -0.0902 -0.0394 0.0108 -0.165 -0.193
(-2.039) (-1.289) (-0.400) (0.0626) (-1.186) (-0.793)
Dummy Has Website 0.0115 0.0288 0.133 0.180*** 0.147 0.275
(0.837) (0.355) (1.648) (4.394) (1.297) (1.474)
Dummy Has Roadmap -0.00754** 0.0428 -0.00662 -0.0483 -0.0215 -0.0791
(-2.194) (0.805) (-0.151) (-1.511) (-0.482) (-0.674)
Dummy Artist Name 0.00303 0.0806 0.0346 0.00960 0.0763* 0.143
(1.048) (1.171) (0.767) (0.272) (1.754) (1.367)
Observations 39,814 39,318 38,984 38,495 37,934 36,320

Adj R-squared 0.0585 -0.00129 0.00573 -0.000482 0.000988 0.00500
50

Table A.13. Long-term Impact of Manipulative Trades: Without Wash Trades
future median price returns and future trading volume change on a daily activity of insider
and wash trade activity scaled by NFT collection-size for collection c as of day t. The de-
pendent variable of Panel A is long-term median price change in USD from day t to day
t + 1, t + 2, t + 3, t + 5, and t + 7 omitting all trades that are classified as wash trades. The de-
pendent variable of Panel B is trading volume change from day t to day t + 1, t + 2, t + 3, t + 5,
and t + 7 omitting all trades that are classified as wash trades. Control variables are a day
parentheses. ∗ p < 0.10; ∗∗ p < 0.05; ∗∗∗ p < 0.01.
(1) (2) (3) (4) (5) (6)

VARIABLES 1 Day 2 Day 3 Day 5 Day 7 Day 14 Day
Wash Activity -0.0107*** -0.0152*** -0.0172** -0.0222*** -0.0358*** -0.0321***

(-3.773) (-2.815) (-2.092) (-2.861) (-4.107) (-3.989)
Log(1+Days after mints) 0.0712*** 0.0587 0.102* 0.0567 0.147** 0.155***
(7.844) (0.985) (1.834) (0.635) (2.554) (2.745)
Past Day Returns -0.134*** -0.0398 -0.138 -0.248* -0.142 -0.250***
(-4.931) (-0.282) (-1.565) (-1.664) (-1.521) (-3.219)
Past Week Returns -0.0368*** -0.0605 -0.0645 -0.283*** -0.192*** -0.178***
(-3.537) (-1.074) (-1.559) (-3.493) (-4.189) (-3.416)
(-7.638) (-5.263) (-4.652) (-4.505) (-4.280) (-3.620)
Dummy category Gaming 0.0332 -0.126 -0.347 -0.157 -0.107 -0.228
(0.631) (-0.640) (-1.288) (-0.341) (-0.284) (-0.871)
Dummy category Metaverse 0.0223 -0.150 -0.367 -0.218 -0.130 -0.166
(0.400) (-0.746) (-1.350) (-0.469) (-0.352) (-0.620)
(-0.142) (-0.567) (-1.242) (-0.444) (-0.498) (-0.602)
Dummy Has Twitter -0.0528 -0.215 -0.274 -0.276 -0.502*** -0.810*
(-0.571) (-1.159) (-1.392) (-1.476) (-3.302) (-1.830)
Dummy Has Website 0.0703 0.166 0.188 0.301* 0.365** 0.485
(0.830) (1.147) (1.250) (1.950) (2.576) (1.453)
Dummy Has Roadmap -0.0265** -0.0168 -0.0182 -0.0371 -0.0252 -0.0167
(-2.450) (-0.363) (-0.375) (-0.553) (-0.503) (-0.300)
Dummy Artist Name 0.0103 0.102** 0.117** 0.132** 0.129** 0.255***
(0.942) (1.967) (2.307) (2.007) (2.507) (4.254)
Observations 39,814 39,318 38,984 38,495 37,934 36,320

Adj R-squared 0.0232 -0.000340 0.00846 0.00227 0.0157 0.0203
51

SSRN Id4397409

Uploaded by

Copyright:

Available Formats

SSRN Id4397409

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Id4397409

Uploaded by

Copyright:

Available Formats

Market Manipulation in Non-Fungible Token Markets *

Non-Fungible Tokens (NFTs) offer a unique opportunity to study market misconduct

Keywords: Speculative Market, Market Manipulation, Transparency

Electronic copy available at: https://ssrn.com/abstract=4397409

Electronic copy available at: https://ssrn.com/abstract=4397409

Electronic copy available at: https://ssrn.com/abstract=4397409

Electronic copy available at: https://ssrn.com/abstract=4397409

Electronic copy available at: https://ssrn.com/abstract=4397409

Electronic copy available at: https://ssrn.com/abstract=4397409

Electronic copy available at: https://ssrn.com/abstract=4397409

2 NFT Markets, Measures, and Predictions

Primary Market (Mint) Secondary Market

Fixed Price Variable Price

Figure 1. Overview of NFT Markets

Electronic copy available at: https://ssrn.com/abstract=4397409

Figure 2. Example of NFT collection and NFT item

Electronic copy available at: https://ssrn.com/abstract=4397409

2.2 Insiders in NFT markets

Electronic copy available at: https://ssrn.com/abstract=4397409

Figure 3. Distribution of Insiders in Primary Market

Electronic copy available at: https://ssrn.com/abstract=4397409

2.3 Wash Traders in NFT Markets

Table 1. Example of Wash Trades in NFT Markets

Item # Trading Time Seller Buyer Price (ETH) Notes

Electronic copy available at: https://ssrn.com/abstract=4397409

Table 2. Logic of Wash Trades Detection

Electronic copy available at: https://ssrn.com/abstract=4397409

Figure 4. Timing of Wash Trade in Secondary Market

Electronic copy available at: https://ssrn.com/abstract=4397409

Electronic copy available at: https://ssrn.com/abstract=4397409

Log Median Price in USD

Figure 5. Trading volume and Median Price of NFT

Electronic copy available at: https://ssrn.com/abstract=4397409

Electronic copy available at: https://ssrn.com/abstract=4397409

Table 4. Summary Statistics

(1) (2) (3) (4) (5) (6)

†Price Return nowash 58,213 0.0330 0.261 -0.544 0.000473 1.275

4 The Impact of Manipulative Trades

Electronic copy available at: https://ssrn.com/abstract=4397409

DV c,t+1 = β1 I nsiderBu yActivit yc,t + β2 W ashActivit yc,t + γ X c,t + FE c + u c,t+1 (1)

Electronic copy available at: https://ssrn.com/abstract=4397409

(1) (2) (3) (4) (5) (6)

InsiderBuy Activity 0.0231*** 0.0213*** 0.0231*** 0.0213***

Observations 39,838 39,814 39,838 39,814 39,838 39,814

Electronic copy available at: https://ssrn.com/abstract=4397409

Table 6. Performance of Manipulative Trades: Without Wash Trades

InsiderBuy Activity 0.0231*** 0.0213*** 0.0231*** 0.0213***

Observations 39,838 39,814 39,838 39,814 39,838 39,814

Electronic copy available at: https://ssrn.com/abstract=4397409

Table 7. Performance of Manipulative Trades: Same Day Without Wash Trades

InsiderBuy Activity 0.0333*** 0.0302*** 0.0333*** 0.0302***

Observations 42,946 42,922 42,946 42,922 42,946 42,922

Electronic copy available at: https://ssrn.com/abstract=4397409

Table 8. Impact of Manipulative Trades on Trading Volume: Without Wash Trades

(1) (2) (3) (4) (5) (6)

InsiderBuy Activity -0.119*** -0.118*** -0.119*** -0.118***

Observations 39,838 39,814 39,838 39,814 39,838 39,814

Electronic copy available at: https://ssrn.com/abstract=4397409

5 Online Community as Information Channel

InsiderBuy Activity 0.0231* 0.0213* 0.0231* 0.0213*

InsiderBuy Activity 0.0231* 0.0213* 0.0231* 0.0213*

InsiderBuy Activity 0.0333* 0.0302* 0.0333* 0.0302*

InsiderBuy Activity -0.119* -0.118* -0.119* -0.118*

InsiderBuy Activity 0.0231* 0.0213*

InsiderSell Activity x Additional 0.0244 0.0212

Marketplace = Blur 0.0360* 0.0360*

Marketplace = Blur -108.1 -147.0*

Wash Sales Dummy = 0 × Marketplace = Sudoswap -68.43 -116.4*

Wash Activity -0.0107* -0.0152* -0.0172 -0.0222* -0.0358* -0.0321*

InsiderBuy Activity 0.0231* 0.0213* 0.0231* 0.0213*

InsiderBuy Activity 0.0231* 0.0213* 0.0231* 0.0213*

InsiderBuy Activity 0.0333* 0.0302* 0.0333* 0.0302*

InsiderBuy Activity -0.119* -0.118* -0.119* -0.118*

InsiderBuy Activity 0.0231* 0.0213*

InsiderSell Activity x Additional 0.0244 0.0212

Marketplace = Blur 0.0360* 0.0360*