Bitcoin - Tether

THE JOURNAL OF FINANCE • VOL. LXXV, NO.
4 • AUGUST 2020
Is Bitcoin Really Untethered?

JOHN M. GRIFFIN and AMIN SHAMS∗
ABSTRACT
This paper investigates whether Tether, a digital currency pegged to the U.S. dollar,
influenced Bitcoin and other cryptocurrency prices during the 2017 boom. Using al-
gorithms to analyze blockchain data, we find that purchases with Tether are timed
following market downturns and result in sizable increases in Bitcoin prices. The flow
is attributable to one entity, clusters below round prices, induces asymmetric auto-
correlations in Bitcoin, and suggests insufficient Tether reserves before month-ends.
Rather than demand from cash investors, these patterns are most consistent with the
supply-based hypothesis of unbacked digital money inflating cryptocurrency prices.
INNOVATION, EXCESSIVE SPECULATION, AND DUBIOUS behavior are often closely

linked. Periods of extreme price increases followed by implosion, commonly
known as “bubbles,” are often associated with legitimate inventions, technolo-
gies, or opportunities. However, they can be carried to excess. In particu-
lar, financial bubbles often coincide with the belief that a rapid gain can be
∗ John M. Griffin is at the McCombs School of Business, University of Texas at Austin. Amin
Shams is at the Fisher College of Business, Ohio State University. Helpful comments were received
from Stefan Nagel (the editor); an associate editor; two anonymous referees; Cesare Fracassi; Sam
Kruger; Shaun MaGruder; Gregor Matvos; Nikolai Roussanov; Clemens Sialm; and seminar and
conference participants at the Chinese University of Hong Kong, Cryptocurrencies and Blockchain
Conference at the University of Chicago, FBI CPA Conference, Financial Intelligence and Inves-
tigations Conference, Fintech Conference at the Hong Kong University, Hong Kong University of
Science and Technology, Hong Kong Securities and Futures Commission, Korea Advanced Insti-
tute of Science and Technology, Korean Financial Supervisory Service, Japan Financial Services
Agency, Santa Clara University, Texas Bitcoin Conference, Tsinghua University, U.S. Commodity
Futures Trading Commission, University of Texas-Austin, University of Zurich, and Wharton Liq-
uidity Conference at the University of Pennsylvania. Integra FEC purchased data and provided
research assistant support for the project. Griffin is an owner of Integra FEC, which engages in
financial consulting on a variety of issues related to financial fraud, including cryptocurrencies.
See disclosure statement. We especially thank Tin Dinh for excellent conceptual assistance and
Prateek Mahajan for research assistance.
Correspondence: Amin Shams, Department of Finance, Ohio State University, 2100 Neil Ave,
Columbus, OH 43210; e-mail: shams.22@osu.edu.
This is an open access article under the terms of the Creative Commons Attribution-NonCom-
mercial License, which permits use, distribution and reproduction in any medium, provided the
original work is properly cited and is not used for commercial purposes.
DOI: 10.1111/jofi.12903

C 2020 The Authors. The Journal of Finance published by Wiley Periodicals LLC on behalf of
American Finance Association

1913
1914 The Journal of FinanceR
obtained from simply selling an asset to another speculator.1 Perhaps because

of the focus on speculative activity rather than verifiable fundamentals, bub-
bles have historically been associated with various forms of misinformation and
fraud. For example, in the Mississippi Bubble of 1719 to 1720, promoters en-
gaged in false marketing about the potential of income-generating assets, price
support by the stock itself, and distribution of paper money that was not fully
backed by gold as claimed (Dale (2004), Kindleberger and Aliber (2011)). As
we briefly discuss in Section I, an abundance of evidence suggests that famous
bubbles such as the 1840s Railroad bubble, the roaring 1920s stock market
boom, the dot-com bubble, and the 2008 financial crisis all involved misin-
formation, false accounting, price manipulation, collusion, and fraud, often in
sophisticated forms.
Cryptocurrencies grew from nearly nothing to over $300 billion in market
capitalization in only a few years and fit the characterization of bubbles quite
well–extreme speculation surrounding an innovative technology. To many, Bit-
coin and other cryptocurrencies offer the promise of an anonymous, decen-
tralized financial system free from banks and government intervention. The
conception of Bitcoin corresponds to the 2008 to 2009 financial crisis, a time
of growing disdain for government intervention and distrust of major banks.
The promise of a decentralized ledger with independently verifiable transac-
tions has enormous appeal,2 especially in an age when centralized clearing is
subject to concerns about both external hacking and internal manipulation.3
Ironically, new large entities have gained centralized control over the vast ma-
jority of operations in the cryptocurrency world, such as centralized exchanges
that handle the majority of transactions and stable coin issuers that can con-
trol the supply of money like a central bank. These centralized entities operate
largely outside the purview of financial regulators and offer varying levels
of limited transparency. Additionally, operating based on digital stable coins
rather than fiat currency further relaxes the need for these entities to estab-
lish a legitimate fiat banking relationship.4 Trading on unregulated exchanges,
specifically on cross-digital-currency exchanges, could leave cryptocurrencies
vulnerable to gaming and manipulation.
In this study, we examine the role of the largest stable coin, Tether, on Bit-
coin and other cryptocurrency prices. Tether, which accounts for more Bitcoin
transaction volume than the U.S. dollar (USD), is purportedly backed by USD
1 For example, in the bubble model of Scheinkman and Xiong (2003), investors purchase assets
not because of their belief in the underlying cash flows, but because they can sell the asset to
another individual with a higher valuation.
2 The appeal, underlying value, and mechanics of cryptocurrencies and decentralized ledgers
have been described in recent descriptive and theoretical work (Yermack (2017), Sockin and Xiong
(2018), Cong, He, and Li (2019), Cong, Li, and Wang (2019)).
3 Recent examples of apparently manipulated markets include LIBOR (Mollenkamp and White-
house (2008)), FX manipulation (Vaughan and Finch (2013)), gold (Denina and Harvey (2004)),
and the VIX index (Griffin and Shams (2018)). Kumar and Seppi (1992) and Spatt (2014) discuss
conditions that may facilitate manipulation.
4 By May 20, 2018, over 1,600 cryptocurrencies and digital tokens were trading on various digital
exchanges.
Is Bitcoin Really Untethered? 1915
reserves and allows for dollar-like transactions without a banking connection,

which many cryptoexchanges have difficulty obtaining or keeping. Although
some in the blogosphere and press have expressed skepticism regarding the
USD reserves backing Tether,5 the cryptocurrency exchanges largely reject
such concerns and widely use Tether in transactions.
To shed light on the driving forces behind the 2017 boom of cryptocurrency
markets, we examine two main alternative hypotheses for Tether: whether
Tether is “pulled” (demand-driven), or “pushed” (supply-driven). Under the
pulled hypothesis, Tether is driven by legitimate demand from investors who
use Tether as a medium of exchange to enter their fiat capital into the cryp-
tospace because it is digital currency with the stability of the dollar “peg.” In
this case, the price impact of Tether reflects natural market demand.
Alternatively, under the “pushed” hypothesis, Bitfinex prints Tether regard-
less of the demand from cash investors, and additional supply of Tether can
create inflation in the price of Bitcoin that is not due to a genuine capital flow. In
this setting, Tether creators have several potential motives. First, if the Tether
creators, like most early cryptocurrency adopters and exchanges, have large
holdings of Bitcoin, they generally profit from the inflation of the cryptocur-
rency prices. Second, coordinated supply of Tether creates an opportunity to
manipulate cryptocurrencies—when prices are falling, the Tether creators can
convert their large Tether supply into Bitcoin in a way that pushes Bitcoin up
and then sell some Bitcoin back into dollars in a venue with less price impact to
replenish Tether reserves. Finally, if cryptocurrency prices crash, the founders
essentially have a put option to default on redeeming Tether, or to potentially
experience a “hack” or insufficient reserves where by Tether-related dollars
disappear. The “pushed” and “pulled” hypotheses have different testable im-
plications for capital flows and cryptocurrency returns that we can take to the
powerful blockchain data.
We begin our exercise by collecting and analyzing Tether and Bitcoin
blockchain data using a series of algorithms that reduce the complexity of
the blockchain. In particular, because of the semitransparent nature of the
transaction history recorded on the blockchain, we are able to use variations of
algorithms developed in computer science to cluster groups of related Bitcoin
wallets. Large clusters are then labeled by identifying certain member wal-
lets inside each group and tracking the flow of coins between major players in
the market.
Figure 1 plots the aggregate flow of Tether among major market participants
on the Tether blockchain from its conception in October 6, 2014 until March 31,
2018. The size of the nodes is proportional to the sum of coin inflow and outflow
to each node, the thickness of the lines is proportional to the size of flows, and
all flow movements are clockwise. Tether is authorized, moved to Bitfinex, and
then slowly distributed to other Tether-based exchanges, mainly Poloniex and
Bittrex. The graph shows that almost no Tether returns to the Tether issuer to
5 For example, see posts by Bitfinex’ed account at https://medium.com/@bitfinexed and Popper

(2017).
Figure 1. Aggregate flow of Tether between major addresses. This figure shows the aggre-
gate flow of Tether between major exchanges and market participants from Tether genesis block
to March 31, 2018. Tether transactions are captured on Omni Layer as transactions with the coin
ID 31. The data include confirmed transactions with the following action types: Grant Property
Tokens, Simple Send, and Send All. Exchange identities on the Tether blockchain are obtained
from the Tether rich list. The thickness of the edges is proportional to the magnitude of the flow
between two nodes, and the node size is proportional to aggregate inflow and outflow for each
node. Intranode flows are excluded. The direction of the flow is shown by the curvature of the
edges, with Tether moving clockwise from a sender to a recipient. (Color figure can be viewed at
wileyonlinelibrary.com)
be redeemed, and the major exchange where Tether can be exchanged for USD,
Kraken, accounts for only a small proportion of transactions. Tether also flows
out to other exchanges and entities and becomes more common as a medium of
exchange over time.
A similar analysis of the flow of coins on the much larger Bitcoin blockchain
shows that the three main Tether exchanges for most of 2017 (Bitfinex,
Poloniex, and Bittrex) also facilitate considerable cross-exchange Bitcoin flows

among themselves.6 Additionally, we find that the cross-exchange Bitcoin flows
on Bitcoin blockchain closely match the Tether flows on the Tether blockchain.
This result independently verifies our algorithm for categorizing exchange
identities and also captures the direct exchange of Tether for Bitcoin. Addi-
tionally, we find that one large player is associated with more than half of the
exchange of Tether for Bitcoin at Bitfinex, suggesting that the distribution of
Tether into the market is from a large player and not from many different
investors who bring cash to Bitfinex to purchase Tether.
We examine the flow of coins identified above to understand whether Tether
is pushed or pulled, and the effect of Tether, if any, on Bitcoin prices. First,
following periods of negative Bitcoin returns, Tether flows from Bitfinex to
Poloniex and Bittrex, and in exchange, Bitcoin is sent back to Bitfinex. Second,
when there are positive net hourly flows from Bitfinex to Poloniex and Bittrex,
Bitcoin prices move up over the next three hours, resulting in predictably high
Bitcoin returns. The price impact is present after periods of negative returns
and periods following the printing of Tether, that is, when there is likely an
oversupply of Tether in the system. This phenomenon strongly suggests that
the price effect is driven by Tether issuances. Additionally, the price impact is
strongly linked to trading of the one large player and not to other accounts on
Poloniex, Bittrex, or other Tether exchanges.
To gauge the aggregate magnitude of the observed price impact, we focus
on the top 1% of hours with the largest lagged combined Bitcoin and Tether
net flows on the two blockchains. These 95 hours have large negative returns
before the flows but are followed by large positive returns afterward. This 1%
of our time series (over the period from the beginning of March 2017 to the
end of March 2018) is associated with 58.8% of Bitcoin’s compounded return
and 64.5% of the returns on six other large cryptocurrencies (Dash, Ethereum
Classic, Ethereum, Litecoin, Monero, and Zcash).7 A bootstrap analysis with
10,000 simulations demonstrates that this behavior does not occur randomly,
and a similar placebo analysis for flows to other Tether exchanges shows very
little price impact.
Further analysis for the single largest player on Bitfinex shows that the
1%, 5%, and 10% of hours with the highest lagged flow of Tether by this one
player are associated with 55%, 67.2%, and 79.2% of Bitcoin’s price increase
over our March 1, 2017 to March 31, 2018 sample period. This pattern is not
present for the flows to any other Tether exchanges. Moreover, simulations
show that these patterns are highly unlikely to be due to chance—this one
large player or entity either exhibited clairvoyant market timing or exerted an
6 For the period between March 1, 2017 and March 31, 2018, we grouped over 640,000 wallet
addresses as Bitfinex, 720,000 addresses as Poloniex, and 1.22 million wallet addresses as Bittrex
using our clustering algorithm.
7 These findings are instructive but incomplete, and they may over- or understate the Tether
effect. Fully quantifying the effect of Tether on Bitcoin depends on knowing precise price impacts
and the various exchange, off-exchange, and cross-trading mechanisms on which these cryptocur-
rencies may trade.
extremely large price impact on Bitcoin that is not observed in the aggregate
flows from other smaller traders. Such trading by this one player is also large
enough to induce a statistically and economically strong reversal in Bitcoin
prices following negative returns.
Investors hoping to stabilize and drive up the price of an asset might concen-
trate on certain price thresholds as an anchor or price floor, the idea being that
if investors can demonstrate a price floor, then they can induce other traders
to purchase.8 Interestingly, Bitcoin purchases from Bitfinex strongly increase
just below multiples of 500. This pattern is present only in periods following
printing of Tether, is being driven by the single large account holder, and is not
observed by other exchanges. To address causality, we use the discontinuity
in Tether flow at the round threshold cutoffs as an instrument and find that
Tether flows are causing the positive Bitcoin return.
The patterns observed above are consistent with either one large player
purchasing Tether with cash at Bitfinex and then exchanging it for Bitcoin, or
Tether being printed without cash backup and pushed out through Bitfinex in
exchange for Bitcoin. If Tether is pushed out to other cryptoexchanges rather
than demanded by cash investors, then it may not be always fully backed. To
show the full reserve, Bitfinex might therefore have to liquidate their Bitcoin
reserve to support their end-of-month (EOM) bank statements. Interestingly,
we find a significant negative EOM abnormal return of 6% in the months with
strong Tether issuance and no abnormal returns in months when Tether is not
issued. Since these patterns are driven primarily by only a few EOMs with
large Tether issuance, we test further and find that the EOM effect is stronger
in a value-weighted index of the largest cryptocurrencies and is also present
around a publicized mid-month balance statement. Moreover, Bitfinex’s reserve
wallets on the blockchain data exhibit large significant balance decreases in
days prior to EOMs with large Tether printing. This pattern is not present in
reserve wallets on any other exchanges.
Our results are generally consistent with Tether being printed unbacked
and pushed out onto the market, which can have an inflationary effect on asset
prices. While other tests do not speak to capital backing, the EOM patterns are
inconsistent with the “pulled” hypothesis since they indicate a lack of dollar
reserves. Nevertheless, we further examine a direct implication of the “pulled”
hypothesis by testing whether the flows of Tether bear a relation to a proxy for
its demand from investors, namely the premium for Tether relative to the USD.
We find little evidence to support this demand-based hypothesis, but note that
the demand-based proxies likely contain noise. In sum, while we expect that
there are some sources of legitimate demand for Tether, they do not appear to
dominate the Tether flow patterns observed in the data.
Overall, our paper demonstrates the usefulness of combining methodolog-
ical approaches from computer science and finance, in particular, clustering
8 Shiller (2000) and Bhattacharya, Holden, and Jacobsen (2012) describe trading signals that
anchor around price thresholds. These thresholds can be used as coordination mechanisms as well.
For instance, Christie and Schultz (1994) find collusion only around even numbers in spreads.
algorithms and capital flow analysis, to understand the role of central mon-
etary entities in a cryptocurrency world. Previous studies show that none of
the exposures to macroeconomic factors, stocks markets, currencies, or com-
modities can explain cryptocurrency prices (Liu and Tsyvinski (2018)). We find
that Tether flows can largely explain Bitcoin prices. Our findings are gen-
erally consistent with evidence that sophisticated investors may profit from
bubbles (Brunnermeier and Nagel (2004)), but more specifically provide empir-
ical evidence on the intersection of potentially nefarious activity and bubbles.
Although cryptocurrencies are relatively new, the trading mechanisms within
and across exchanges are quite complex (Partnoy (2009)) and may obfuscate
the influence of large players. This complexity also implies that there are limits
to what we can learn from blockchain data, and additional research is certainly
necessary to further understand the cryptocurrency market. Since our findings
indicate that Bitcoin prices are subject to gaming by a small number of ac-
tors, they suggest that Bitcoin does not make a solid basis for more complex
financial vehicles such as exchange-traded funds (ETFs) or derivatives. Mar-
ket surveillance within a proper regulatory framework across many venues
may be necessary for cryptocurrency markets to be a reliable medium for fair
financial transactions.
The rest of the paper is organized as follows. Section I provides an overview of
historical bubbles, cryptocurrencies, Tether, and the main pushed and pulled
hypotheses to be tested. Section II describes our main data sources and ex-
plains the methodologies that we use to analyze the blockchain data and flows.
Section III analyzes the potential influence of Tether on Bitcoin, and Section IV
further tests whether the flows are consistent with pushed or pulled explana-
tions. Section V concludes.
I. Overview of Bubbles, Bitcoin, Tether, and Hypotheses

A. Speculative Bubbles and the Prevalence of Dubious Market Activity
Periods of excessive price speculation often share the themes of optimism
around a new technology, a focus on selling to others rather than economic
cash flows, and questionable activities. The famous South Sea Bubble of 1719
to 1720 is often described as a sophisticated Ponzi scheme where old investors
were paid high dividends not from operations but from new stock issuances with
the hope of higher prices at future issuances (Hutcheson (1720), Temin and Voth
(2013)). Scheinkman (2013) notes that many other companies around this time
also seem to have been fraudulent. The Railroad Bubble of the 1840s led to a
host of companies that merely sought to procure funds from investors and had
no intention of actually building railroads (Robb (2002)). In the Roaring Twen-
ties, investment pools would manipulate a stock price through “wash sales,”
collusion with stock-exchange specialists, and coordinated publicity from com-
mentators to pump a stock at an inflated price to the public (Malkiel (1981)).
The technology or “dot-com” bubble of 1997 to 2000 also contained strong el-
ements of stock promotion through inflated forecasts from affiliated analysts
(Lin and McNichols (1998)), pushing or “laddering” prices through implicit

agreements to purchase more IPO shares in the aftermarket (Griffin, Harris,
and Topaloglu (2007)), and accounting fraud (e.g., Enron and Worldcom). Hedge
funds and other institutional investors were the main net buyers of overpriced
technology stocks during this period (Brunnermeier and Nagel (2004), Griffin
et al. (2011)).
One line of thinking is that more fraud exists in economic booms because
individuals monitor their investments relatively less closely (Povel, Singh, and
Winton (2007)). Akerlof et al. (1993) argue that historical actors involved in
“looting” an organization (such as banks in the U.S. savings and loan crisis)
move capital into a space in a manner that systematically increases asset
prices. In our analysis of Bitcoin and Tether, we are able to examine whether
either of these two views fits the data.
B. Brief History of Bitcoin and Exchange “Hacks”

On October 31, 2008, the whitepaper “Bitcoin: A Peer-to-Peer Electronic
Cash System” was released by Satoshi Nakamoto (Nakamoto (2008)). The pa-
per outlines a digital currency system where transactions are recorded on a
chain of linked blocks, hence “blockchain,” and verified electronically through
a decentralized network of users. This decentralized feature avoids the tradi-
tional system of government-backed currencies controlled by centralized banks
and clearing houses. On January 3, 2009, the first block was established on the
Bitcoin blockchain by Nakamoto. On October 5, 2009, New Liberty Standard
established the first exchange rates of Bitcoin (BTC) at 1309.03 for 1 USD, or
$0.00076 per BTC.9 By April 23, 2011, Bitcoin exceeded parity with the USD,
euro, and British pound, with the market cap passing 10 million USD, and by
March 28, 2013, Bitcoin market cap passed 1 billion USD.
Mt. Gox, a leading exchange that by 2013 was handling approximately 70% of
Bitcoin volume, declared bankruptcy due to a mysterious “hack” of the exchange
which resulted in approximately $450 million worth of Bitcoin missing from
investors’ accounts. Good reasons have been put forward as to why the “hack”
may have been an inside job (Nilsson (2015)). Gandal et al. (2018) argues that
fraudulent trading on the Mt. Gox exchange led to a significant spike in Bitcoin
prices in late 2013.10 Foley, Karlsen, and Putniņš (2019) detail hubs of illicit
commerce in Bitcoin and estimate that 44% of transactions are associated with
illegal activity.
9 Most of these facts are available in multiple places, but an account of the first five years of
Bitcoin can be found at http://historyofbitcoin.org and in Lee (2014).

10 In the second-biggest hack in Bitcoin history, on August 2, 2016, the Bitfinex exchange an-
nounced that $72 million had been stolen from investor accounts, leading Bitcoin to plummet 20%
in value.
C. Brief History of Tether

The objective of Tether is to facilitate transactions between cryptocurrency
exchanges with a rate pegged to the USD. While this could also occur with
fiat transactions, Tether is advantageous because many cryptoexchanges have
difficulty securing banking relationships. Tether Limited, the issuer of Tether,
historically claimed that “Tether Platform currencies are 100% backed by actual
fiat currency assets in our reserve account.”11 However, Tether itself created
ambiguity around this backing by later noting that they do not guarantee
redemption rights.12
The Bitfinex exchange started in 2012, but experienced rapid growth and
now claims to be “the world’s largest and most advanced cryptocurrency trading
platform.” The Paradise Papers leaks in November 2017 named the Bitfinex
exchange officials, Philip Potter and Giancarlo Devasini, responsible for setting
up Tether Holdings Limited in the British Virgin Islands in 2014.13
Figure 2, Panel A, shows the cumulative authorization of Tether denomi-
nated in both USD and Bitcoin as well as Bitcoin prices. The first Tether was
authorized on October 6, 2014, but the market cap was only $25 million as of
March 6, 2017. Between March 7, 2017 and January 2018, however, more than
$2.2 billion worth of Tether was issued.
Panel B of Figure 2 shows transactions of major cryptocurrencies in USD as
compared to Tether, aggregated across all cryptocurrency exchanges available
on CoinAPI. Although cryptocurrencies were historically denominated in dol-
lars or yuan, a large share of Bitcoin and many other cryptocurrencies transac-
tions are denominated in Tether as of 2017. Additionally, even after closely
examining Bitfinex public statements, it is unclear as to whether Bitfinex
transactions are denominated in dollar or Tether. Prices quoted on Bitfinex
are significantly closer to prices on Tether exchanges than USD exchanges.14
Hence, we term Bitfinex transactions as well as those explicitly denominated
in Tether as Tether-related.
Many in the blogosphere as well as the mainstream press began to raise
questions about Tether in the second half of 2017.15 In April 2017, Tether
lost its banking relationship with a Taiwanese bank linked to Wells Fargo.
Since then, Tether has issued over $2 billion Tether without fully disclosing
banking details. This could be due to not wanting to subject their bank to
public scrutiny and risk losing their new banking relationship, as many large
banks avoid the scrutiny associated with crypto-related deposits either because
11 See https://tether.to/faqs/.
12 “There is no contractual right or other right or legal claim against us to redeem or exchange
your Tethers for money. We do not guarantee any right of redemption or exchange of Tethers by us
for money” (Leising (2017)).
13 See Popper (2017).
14 The percentage deviation of hourly prices between Bitfinex and Poloniex and Bittrex are 19
and 42 basis points, while the deviation is 103, 56, and 111 basis points for Bitstamp, Gemini, and
Kraken, respectively.
15 See Leising (2017), Kaminska (2017), and Popper (2017).
Figure 2. Tether authorization and Bitcoin price over time, and trade volume in both
dollars and Tether. Panel A plots the cumulative authorization of Tether and the price of Bitcoin
over time. The red dashed line shows cumulative authorization in millions of Tether tokens. The
black dashed line shows Tether cumulative authorization denominated in contemporaneous Bitcoin
price. The blue line shows the Bitcoin price. Authorizations are defined as transactions with
transaction type “Grant Property Tokens” on Tether blockchain. Panel B shows the percentage of
trade volume of USD and Tether for major cryptocurrencies between March 1, 2017 and March 31,
2018 aggregated over all exchanges. The major currencies include the 15 largest cryptocurrencies
and tokens by aggregate trade volume across exchanges reported in CoinAPI data over the same
period. The blue bars show the percentage of volume traded against USD, the red bars show the
percentage against Tether, and the gray bars show the percentage against USD/Tether on the
Bitfinex exchange. (Color figure can be viewed at wileyonlinelibrary.com)
of perceived reputation tainting or because of the need to comply with anti-

money laundering (AML) or “know your customer” (KYC) banking regulations.
Tether hired a consultant that released an internal memo showing reserves on
September 15, 2017.
Immediately after the first draft of this paper, a law firm released a report on
the sufficiency of Tether reserves in June 2018.16 On February 25, 2019, Tether
changed their definition of Tether backing to read “traditional currency and
cash equivalents.” In response to legal motions, on April 30, 2019, Bitfinex’s
former General Counsel admitted that Tether does not have cash reserves equal
to 100% of the outstanding Tethers. In a May 15, 2019 court hearing, a Bitfinex
attorney also admitted that Tether invested in instruments beyond cash, in-
cluding Bitcoin, something clearly at odds with Tether’s longstanding claims.
Bloggers have also conjectured about whether Tether authorizations are fuel-
ing Bitcoin.17 One website, tetherreport.com, finds positive return effects after
incidences of Tether authorizations.18 Analysis by Wei (2018), however, finds
no price effect at the time of Tether authorizations.
D. Main Hypotheses
This section examines two main alternative “pulled” versus “pushed” hy-
potheses19 about Tether. Under the first hypothesis, Tether is “pulled” or driven
by legitimate demand from investors who use Tether as a medium of exchange
to enter their fiat capital into the cryptospace. In this case, the price impact of
Tether reflects natural market demand. Under the second hypothesis, Tether
is “pushed” through a supply-driven scheme whereby an unbacked digital dol-
lar is printed and used to purchase Bitcoin. In this case, additional supply of
Tether can create inflation in the price of Bitcoin and other cryptocurrencies
that is not due to a genuine capital flow.
Related to the “pulled” hypothesis, we first predict that Tether is driven by in-
vestor demand and is always fully backed by USD (as with a full-reserve bank).
16 Tether Limited has also released EOM snapshot bank statements showing reserves at the
EOM. Tether has not to our knowledge released a full audit, which is important since snapshot
reports showing cash in a bank balance on a certain date could reflect borrowed funds or funds from
related entities. Tether is closely related to Bitfinex, which has also not been audited, according to
public sources.
17 See Higgins (2018) and Leising (2017).
18 The website shows that after 91 hourly events of Tether being granted and moved to Bitfinex,
the Bitcoin return increases over the next two hours. They compound the return for those 182
hours (91 two-hour periods) and derive a compounded effect of 48.8%, then compare this effect to
6.5% average compounded returns for the same time period during normal times. The results are
incorrectly interpreted as “Tether could account for nearly half of Bitcoin’s price rise” or “a rough
estimate of 40% price growth attributed to Tether.” Indeed, Bitcoin prices increased by 1,422%
(from $893.19 to $13,592.93) over their period of study. Interestingly, we find that the hours
directly following Tether authorization are often not when the Bitcoin buying activity actually
occurs.
19 There is a literature in international finance examining whether capital flows are pushed or
pulled across markets (Froot, O’connell, and Seasholes (2001), Griffin, Nardari, and Stulz (2007)).
A currency that can provide a stable store of value, support quick transactions,
and potentially allow cryptocurrency exchanges to skirt banking regulations
required for traditional deposits has an intuitive appeal. If an increase in de-
mand is driven by new investors who hold dollars and wish to convert their
dollars to Tether and then into cryptocurrencies, the increase in demand may
result in a higher market rate for Tether. A lower price for Tether would thus
be a consequence of weak demand for Tether, while a higher price (perhaps at
or above one dollar) would be a consequence of strong Tether demand.
H1A: Tether’s price relative to the USD may increase as a consequence of

strong investor demand. Tether flows should be strongly related to this
demand as proxied by changes in the Tether-USD exchange rate.
H1B: The printing of Tether may also be driven by its usefulness as a fa-
cilitator of cross-exchange arbitrage to eliminate pricing discrepancies
across cryptocurrency exchanges. For example, Tether outflows from
Bitfinex to another exchange should correspond to periods when Bit-
coin sells at a premium on Bitfinex relative to that exchange.20
The main alternative hypothesis is that Tether is printed independent of

demand and pushed onto the market. The issuers can print Tether and convert
it into more widely accepted cryptocurrencies such as Bitcoin. In addition to
issuance fees, transaction fees, and interest earned from trading in Tether,
other possible benefits of “pushing” Tether could be as follows.
First, like an inflationary effect of printing money, issuing Tether increases
the money supply in the cryptospace and can significantly push cryptocurrency
prices up by generating artificial demand. Since most cryptocurrency exchanges
and early movers are long in Bitcoin and other cryptocurrencies, they would
generally benefit. For instance, if Bitcoin prices increase, the founders can cash
out the acquired Bitcoins into dollars, likely at a slower pace and on an opaque
channel that has less price impact than their initial buying behavior. If the
Tether issuers wish to legitimize Tether and avoid scrutiny, they can slowly
convert some of their cryptocurrencies to USD and retroactively provide either
full or partial dollar reserves for Tether.
Second, since Tether issuances are large, if traded strategically, Tether could
have further price impact and lead to further manipulation of Bitcoin prices.
For instance, the issuers can stabilize and/or set regionalized price floors and
push the price of Bitcoin and other cryptocurrencies upward.
Third, the Tether issuers create a valuable put option in the case of a future
cryptomarket downturn or other losses. In particular, the founders of Tether
have an option to not redeem Tether to dollars, and possibly experience an in-
side “hack” (McLannahan (2015)) when Tethers and/or their associated dollars
suddenly disappear.
20 This hypothesis is also consistent with the supply-driven view as unbacked money printing of
Tether could cause Bitcoin to sell at a premium on Bitfinex relative to the other exchanges before
Tether moves to those exchanges.
The key to the “pushed” hypothesis is that the Tether-USD price does not
collapse. This can be accomplished by creating a limited set of venues to redeem
Tether, sending signals to investors through periodic accounting reports, and
creating Tether price support.
To examine the “push” hypothesis, we test the following predictions:
H2A: If Tether issuers are trying to provide stability to the market during
downturns, outflows of Tether and purchases of Bitcoin by Bitfinex
may follow periods of negative Bitcoin returns.
H2B: If Tether supply is large enough to have a material price impact on
Bitcoin, Bitcoin prices should go up after Tether flows into the market,
especially after periods with large authorization of Tether.
H2C: Bitcoin returns may show a return reversal after negative returns,
especially during times when Tether flows into the market.
H2D: Since round-number thresholds can be price anchors to set a price floor
and are often used as buying signals by investors, flow of Tether might
increase if Bitcoin falls below these salient round-number thresholds.
This effect should be more pronounced in periods with large Tether
authorization.
H2E: If Tether is not fully backed by dollars at the outset, but the issuers
want to signal otherwise to investors by releasing EOM (or other in-
terval) accounting statements, then Tether creators may liquidate Bit-
coins into USD to demonstrate sufficient reserves. This could create
negative returns in Bitcoin at the EOM, particularly in periods with
large Tether issuances.
While the above hypotheses need not all follow from the pulled hypothesis,
H2A through H2D shed light on whether the flow of Tether into the market
is consistent with creating price support and inflating Bitcoin prices, and H2E
sheds light on whether the potential price impact is due to unbacked printing
of Tether, which can have an inflationary effect on Bitcoin. In the next section,
we discuss the data and empirical methods used to test these hypotheses.
II. Data, Algorithms, and Flows between Major Accounts

A. Data
The price and the blockchain data obtained for this study amount to
over 200 GB from more than 10 sources, with CoinAPI, Coinmarketcap.com,
Blockchain.info, Omniexplorer.info, and CoinDesk as our main sources. The in-
traday pricing data on major cryptocurrencies come from CoinAPI. The starting
date varies for different currencies. The sample covers 25 months from March
2016 to March 2018, but the main tests are implemented after March 2017,
when Tether experienced a large issuance.21
21 Daily prices are based on Coordinated Universal Time (UTC) time, and the close and open
prices are calculated based on a 24-hour daily cycle that ends at midnight UTC. Daily prices of
Bitcoin blockchain data are obtained from Blockchain.info and cover the pe-
riod from Bitcoin initiation in January 2009 to March 2018. The blockchain
data contain the entire history of Bitcoin transactions between Bitcoin wallets
and include variables such as wallet IDs of senders and recipients as a string of
34 characters and numbers, amount of coins transferred, timestamp, transac-
tion ID, and previous transaction ID where the coin was received by the sender
of each new transaction. Over the October 2014 to March 2018 period, Tether
is issued via the Omni Layer Protocol based on the Bitcoin blockchain, and
Tether blockchain data are from Omniexplorer.info.
To assign identities of grouped wallets to Tether-related exchanges on the
Bitcoin blockchain, the addresses of a number of wallets belonging to Tether
exchanges are collected from public forums and individual investors who trans-
ferred Bitcoin to these exchanges.22 For the Tether blockchain, wallet identities
of major exchanges are manually collected from the Tether rich list on tether.to
at all snapshots available on Internet Archive.
Tether exchanges account for a large portion of cryptocurrencies’ trading
volume over our sample period. Table I, Panel A, shows the total trading volume
on major exchanges of major cryptocurrencies from March 1, 2017 to March
31, 2018. Tether-based exchanges are marked with a “*.” Some exchanges,
including Gemini and Coinbase, specialize in a limited number of major coins
such as Bitcoin and Ethereum. Others, especially the Tether-related exchanges,
feature a large number of coins. Bitfinex has the largest volume, both for Bitcoin
and across all major cryptocurrencies. Other Tether exchanges also play an
important role among the top 10 exchanges in terms of aggregate volume.
As shown in Panel B of Figure 2, a large share of major cryptocurrencies’
transactions are denominated in Tether.
Panel B of Table I shows the cross-sectional correlation of cryptocurren-
cies’ daily returns. Not surprisingly, the daily returns are positively correlated
across all of the coins, but there is variation across different cryptocurrencies.
For example, Bitcoin’s correlation with Ethereum, Ripple, and Litecoin are
0.44, 0.20, and 0.45, respectively.
Panel C of Table I reports the autocorrelation of cryptocurrencies at vari-
ous frequencies. The autocorrelations are generally negative. For example, a
1% change in lagged one-hour Bitcoin prices is followed by a 6 basis point
reversal in the next hour. The reversal is 6 and 5 basis points at three- and
five-hour intervals.
various coins are obtained from Coinmarketcap.com, which calculates the price of each coin by
taking the volume-weighted average of prices reported at different exchanges. We also use the
intraday CoinDesk price index, which aggregates prices across major markets. Hourly and five-
minute returns are calculated from the last trade within each minute. Missing prices are carried
forward for nontrading periods of up to five minutes. Prices are assumed to be missing if stale for
more than five minutes.
22 The Internet Appendix Section II includes the list of representative addresses that can be
used to assign identities of major exchanges. The Internet Appendix is available in the online
version of this article on The Journal of Finance website.
Table I
Summary Statistics
This table summarizes the trading volume and pricing information of major cryptocurrencies on major exchanges. The major cryptocurrencies are the
15 coins and tokens with the highest aggregate volume in USD and Tether across exchanges reported in CoinAPI between March 1, 2017 and March
31, 2018, and the top exchanges are those with the highest aggregate volume for these major cryptocurrencies. Panel A shows the total volume for
each cryptocurrency on each exchange in billions of dollars from March 1, 2017 to March 31, 2018 using data from CoinAPI. Tether-based exchanges
are indicated with a star. Panel B shows the daily return correlation between major cryptocurrencies. The daily pricing data are from CoinMarketCap.
Panel C reports the autocorrelations of the major cryptocurrencies at one-hour, three-hour, and five-hour intervals using price data from the most
liquid exchange for each altcoin between March 1, 2017 and March 31, 2018. The three-hour and five-hour autocorrelations are calculated using
hourly returns rolled over three-hour and five-hour windows. Standard errors are adjusted for heteroskedasticity and autocorrelation. The intraday
pricing data are from CoinAPI.
Panel A: Total Volume ($B)
Binance* Bitfinex* Bitstamp Bittrex* Coinbase Gemini Huobi* Kraken* OKEx* Poloniex*
BCC 0.81 0.01 – 1.68 – – – – – –

BCH 0.81 18.83 0.66 – 2.96 – 1.52 1.99 2.47 3.06
BNB 2.69 – – – – – – – – –
BTC 32.78 120.79 36.20 11.52 53.09 16.50 8.10 17.10 6.86 14.64
DASH – 1.88 – 0.26 – – 0.99 0.34 0.03 0.55
EOS – 8.12 – – – – 2.36 0.07 0.29 –
ETC – 5.59 – 0.60 – – 0.92 0.96 1.30 1.36
ETH 10.19 35.40 5.44 2.50 32.46 7.77 3.11 14.54 3.08 4.91
IOTA – 2.51 – – – – – – 0.06 –
LTC 2.69 13.13 2.44 1.02 24.51 – 1.10 1.80 2.78 2.48
NEO 3.88 4.54 – 1.46 – – 0.24 – 0.18 –
OMG – 3.77 – 0.49 – – 0.21 – 0.01 –
XMR – 2.84 – 0.30 – – – 0.77 0.00 0.60
XRP – 17.11 7.41 1.86 – – 1.46 3.28 0.26 2.87
ZEC – 2.35 – 0.33 – – 0.32 0.39 0.01 0.70
(Continued)
1927
1928
Table I—Continued
Panel B: Correlations
BCC BCH BNB BTC DASH EOS ETC ETH IOTA LTC NEO OMG XMR XRP
BCH 0.17
BNB 0.31 0.21
BTC 0.47 0.24 0.46
DASH 0.28 0.42 0.20 0.39
EOS 0.19 0.34 0.28 0.35 0.30
ETC 0.25 0.42 0.28 0.42 0.36 0.38
ETH 0.30 0.40 0.37 0.44 0.44 0.45 0.61
IOTA 0.29 0.25 0.35 0.48 0.42 0.32 0.54 0.53
LTC 0.24 0.31 0.34 0.45 0.36 0.35 0.50 0.42 0.43
NEO 0.16 0.25 0.43 0.30 0.31 0.29 0.43 0.34 0.31 0.31
The Journal of FinanceR
OMG 0.26 0.17 0.42 0.41 0.40 0.41 0.45 0.60 0.47 0.41 0.60
XMR 0.26 0.35 0.26 0.49 0.55 0.34 0.43 0.52 0.54 0.42 0.24 0.40
XRP 0.15 0.24 0.17 0.20 0.10 0.29 0.17 0.19 0.30 0.26 0.12 0.32 0.23
ZEC 0.22 0.41 0.34 0.38 0.58 0.42 0.49 0.52 0.54 0.36 0.34 0.45 0.54 0.27
(Continued)
Table I—Continued
Panel C: Autocorrelations
One-HourInterval Three-Hour Interval Five-Hour Interval
Coin Coefficient t-Stat Coefficient t-Stat Coefficient t-Stat
BCC −0.127 −3.960 −0.166 −6.412 −0.260 −6.800

BCH −0.039 −1.459 −0.033 −1.136 −0.064 −1.870
BNB −0.000 −0.827 0.002 1.476 0.004 3.850
BTC −0.063 −4.089 −0.072 −4.414 −0.062 −2.985
DASH −0.073 −4.124 −0.052 −2.822 −0.065 −3.540
EOS −0.075 −2.448 −0.052 −1.376 −0.072 −1.300
ETC −0.054 −3.182 −0.071 −3.807 −0.031 −1.383
ETH −0.053 −3.069 −0.043 −2.154 −0.042 −1.780
IOTA −0.202 −6.775 −0.241 −6.820 −0.224 −6.022
LTC −0.009 −0.341 −0.047 −1.356 −0.018 −0.476
NEO
−0.081 −3.657 −0.064 −2.341 −0.069 −2.263

OMG −0.068 −3.745 −0.039 −1.677 −0.039 −1.319
XMR −0.075 −3.243 −0.067 −3.391 −0.066 −2.877
XRP −0.104 −3.348 −0.042 −1.374 0.049 1.035
ZEC −0.077 −3.782 −0.063 −2.446 −0.098 −3.387
1929
B. Analyzing Bitcoin Blockchain

The Bitcoin blockchain up to March 31, 2018 is a 170 GB network database
of more than 360 million wallet addresses and billions of transactions. It is
common for each entity to have multiple wallet addresses, and transactions
with multiple senders and recipients are frequent.23 The complexity of the
data is illustrated in Internet Appendix Figure IA.1, which depicts a 10-minute
random sample of the blockchain in 2017. In the figure, each node represents
a wallet address, and each edge shows the flow of coins.
To reduce the complexity of the network, we adopt methods from the com-
puter science literature (Androulaki et al. (2013), Meiklejohn et al. (2013), Reid
and Harrigan (2013), Ron and Shamir (2013)) to cluster-related Bitcoin wal-
lets. The idea is that when multiple addresses are used as inputs to a single
transaction, the entity controlling each of the inputs must have the private
signing keys of all other inputs. It is therefore very likely that all such ad-
dresses are controlled by the same entity. For example, if wallets A and B
appear as inputs in a single transaction, and wallets B and C appear as inputs
in a different transaction, we group wallets A, B, and C together. We find con-
nected components of this “same-input” relation throughout the entire Bitcoin
blockchain and consider each component as a group of wallets controlled by the
same entity. We then take three more steps. First, if a transaction has multiple
recipients, the flow from the sender is allocated proportionally by the number
of coins received by each recipient. Second, for each transaction, we exclude
the portion of coins that have the same input and output wallets. Finally, we
exclude the transaction fees as reflected in the difference between total Bit-
coin sent and received in one transaction. The clustered group of wallets that
contain exchange addresses are assigned to the identified exchanges. Between
March 1, 2017 and March 31, 2018, a group of approximately 640,000 wallets
are labeled as Bitfinex, 720,000 wallets as Poloniex, and 1.22 million wallets
as Bittrex.
Figure 3 shows the flows on the Bitcoin blockchain. First, one can see that the
Bitcoin blockchain has many more major players than the Tether blockchain,
and we do not find identifying information for all nodes. Second, Bitfinex,
Poloniex, and Bittrex are considerable players on the Bitcoin blockchain in
terms of the aggregate flow of coins, and there is a reasonable flow volume be-
tween these exchanges. Third, there are substantial flows between Bitfinex and
transitory addresses,24 which we define as wallets with four or fewer transac-
tions on the blockchain and zero net balance, and with the Bitfinex cold wallet.
23 Internet Appendix Table IA.I shows an example of a Bitcoin transaction on the blockchain
with 313 senders and 218 recipients. Addresses on the left column are senders of the Bitcoins and
addresses on the right are the recipients.
24 Transitory addresses may be tumblers or mixer wallets used to further mask Bitcoin transfer
activities.
Figure 3. Aggregate flow of Bitcoin between major addresses. This figure shows the ag-
gregate flow of Bitcoin between major exchanges and market participants from March 1, 2017 to
March 31, 2018. Groups of addresses are clustered by finding the connected component of the same
input relation on the Bitcoin blockchain, and each group is labeled with identities of members
obtained from publicly available information and individual investors. The thickness of the edges
is proportional to the magnitude of flow between two nodes, and the node size is proportional to
aggregate inflow and outflow of each node. Intranode flows are excluded. The direction of the flow
is shown by the curvature of the edges, with Bitcoin moving clockwise from a sender to a recipient.
(Color figure can be viewed at wileyonlinelibrary.com)
C. Analyzing Tether Blockchain

As previously described, Figure 1 provides insights into the structure of the
Tether network. First, almost all Tether printed by Tether Limited (the red
node in the bottom of the graph) is first moved to Bitfinex and then distributed
through the network. The transfer of Tether from Tether authorizer (account
labeled as 3MbY) to Tether treasuries (1NTM and 3BbD), all colored in red,
is referred to as “authorization,” and the transfer out of Tether treasuries,
primarily to Bitfinex, is referred to as “issuance.” Note that barely any flows
move back to the initial Tether printing node, consistent with individuals stat-
ing that it is not feasible to move Tether back to Tether Limited to redeem
for USD. Second, Poloniex and Bittrex, the largest Tether exchanges for most
of 2017, are closely tied to Bitfinex through a large flow of Tether using an
intermediary address. Third, Kraken, the small yellow node at the top of the
graph, was the only official marketplace for trading the USD-Tether pair for
the majority of 2017. Fourth, most of the Tether flows to and from Bitfinex are
through Bittrex and Poloniex. Throughout the paper, we focus on the timing
and amount of Tether flow from Bitfinex to these two major exchanges because
as we will show, this is the primary channel through which Tether is converted
to Bitcoin; however, we also examine flows to other exchanges. To calculate the
flows between exchanges, we consider the intermediary wallets that receive
Tether from Bitfinex and transfer them all to the same exchange as addresses
belonging to that exchange.
Note that since the figure is proportional to the size of the flows, the graph
puts substantial emphasis on the end of 2017 and early 2018, when Tether
issuance increased rapidly. For this reason, in Internet Appendix Figure IA.2,
we display four snapshots of the Tether flows over time. For the majority of
2017, Bitfinex, Poloniex, and Bittrex were by far the largest players in the
market. Binance, Huobi, OKEx, and Kraken gained substantial market share
in December 2017.
The flow of Tether from Bitfinex to the other exchanges increases on the day
of Tether authorization, but it takes as many as three to four days to move the
capital out of Bitfinex to the other exchanges.25 It is the net flow of Tether out
of Bitfinex to Poloniex and Bittrex and the net flow of Bitcoin back that we use
in our tests.
D. Bitcoin and Tether Net Flows

Flows between two parties on the blockchain are more formally defined as the
signed net amount of capital transferred between those parties. Specifically, our
tests require the flow of coins between major Tether exchanges, Bitfinex (BFX),
Poloniex (PLX), and Bittrex (BTX), during our sample period. For Bitcoin, we
simply aggregate the net amount of coins transferred between these exchanges
25 We show this formally in a VAR model in Internet Appendix Figure IA.3. Examples are shown
in Internet Appendix Figure IA.4.
in each period:
t

t
NetBTCFlowt = BTCPLX→BFX − BTCBFX→PLX
t−1 t−1
t

t
+ BTCBTX→BF X − BT C BF X→BT X , (1)
t−1 t−1
where BT Ci→ j is the amount of coins transferred from group of wallets i to

group of wallets j between hours t − 1 and t. For Tether, to measure the value
relative to Bitcoin prices, we accumulate the Bitcoin-denominated value of
Tether using Bitcoin prices at the time of the transaction. Similar to the flow
of Bitcoin, we define the net flow of Tether as
t

t
NetTetherFlowt = Tether BF X→P LX − Tether P LX→BF X
t−1 t−1
t

t
+ Tether BF X→BT X − Tether BT X→BF X , (2)
t−1 t−1
where Tetheri→ j is the amount of coins transferred from exchange i to exchange

j between hours t − 1 and t.
We also verify that flows identified on the Tether blockchain moving from
Bitfinex and to Poloniex and Bittrex correspond to opposite flows back on the
Bitcoin blockchain that come out of Poloniex and Bittrex and into Bitfinex.
Internet Appendix Figure IA.5 shows that the two series have a correlation of
0.72 for Poloniex and 0.71 for Bittrex at daily intervals, and that they also have
similar magnitudes. The Bitcoin flow between other exchanges, even between
other Tether-based exchanges and Bitfinex, have much lower correlations with
the Tether flow to Poloniex and Bittrex and a much larger difference in mag-
nitude. We also find a strong relation between inflow of Tether to Poloniex
and Bittrex in the blockchain data and reported exchange trading volume on
Poloniex and Bittrex that is not present in a placebo test for other Tether-
related exchanges.26
The magnitude of the flow of coins on the two blockchains matches closely,
and the correlation between the two flows is high, but the timing is not per-
fectly matched given different delays in moving coins to exchanges and clearing
transactions on the blockchain. Given that the timing of blockchain transac-
tions is a proxy for the actual capital flows, and to reduce noise in our measure
of net flows of Tether out of Bitfinex and net flows of Bitcoin coming back, we
average the two flows on the Bitcoin and Tether blockchains:
Tether/BitcoinFlow = (NetTetherFlowt + NetBTCFlowt )/2. (3)
26 Details on our verification method and the results are provided in the Internet Appendix
Section I.
After printing, Tether is used to purchase Bitcoin primarily on Poloniex

and Bittrex. We examine if the sensitivity of flow of Tether to Bitcoin returns
is symmetric in response to positive and negative shocks. Tether is used to
purchase Bitcoin when returns are negative, but we do not find considerable
Tether flows following price increases (see Internet Appendix Figure IA.6 and
Table IA.II).
E. Detailed Deposit Accounts

We drill down on the nature of the Tether flows out of Bitfinex and the corre-
sponding Bitcoin flows back by focusing on the exact deposit addresses used to
move these coins. Typically, to electronically detect which user has deposited
funds and to credit these funds to their account, each exchange user receives
her own unique deposit wallet address. Interestingly, Panel A of Figure 4 shows
that 81% of the Tether flows from Bitfinex to Poloniex and Bittrex are through
one large deposit address for each exchange. This account is responsible for
47% of all Tether flows from Bitfinex to all Tether exchanges combined. The
first four digits of these addresses are shown as 1J1d for Poloniex and 1AA6
for Bittrex in the figure. Additionally, 52% of the Bitcoin flows back to Bitfinex
from all Tether exchanges goes to a single deposit address on Bitfinex, which
we label with its first four digits on the Bitcoin blockchain, 1LSg. The relation
is depicted in Internet Appendix Figure IA.7, which shows how Bitfinex sends
Tether out on the Tether blockchain through 1J1d and 1AA6 and receives flows
back from 1MZA. On the Bitcoin blockchain, a majority of the Bitcoin deposits
from Poloniex and Bittrex to Bitfinex go through 1LSg, and the flows back to
Poloniex and Bittrex go through 1DEc and 1PCw.
If the Tether flows to 1J1d and 1AA6 on the Tether blockchain correspond to
Bitcoin flows to 1LSg on the Bitcoin blockchain, this would suggest that all of
these wallets are likely controlled by the same entity, which sends the printed
Tether into the market in exchange for Bitcoin. To examine this, we compare
the Tether flows from Bitfinex to 1J1d and 1AA6 on Poloniex and Bittrex to the
Bitcoin flow from Poloniex and Bittrex to the top-100 largest Bitcoin addresses
on Bitfinex, including 1LSg. The correlation of Bitcoin flows from Bittrex to
1LSg with Tether flows from Bitfinex to 1AA6 on Bittrex is 0.69. The correlation
is 0.64 for 1J1d on Poloniex. Flows to other large deposit accounts on Bitfinex
do not come close in terms of the correlation or the magnitude of flows. Internet
Appendix Section I (and Figures IA.8 and IA.9) provides more details on the
procedure used to identify these wallet addresses that move Tether and Bitcoin
between Bitfinex, Poloniex, and Bittrex and verify their relation. Analogous
to our flow calculations in equations (1) to (3), we calculate the average net
Tether/Bitcoin flows to these large, closely tied wallets and label them as “1LSg
flows” throughout the paper. We also compare the effect of flows that are not
part of this group of wallets.
Figure 4. Top accounts associated with the flow of Tether from and Bitcoin to Bitfinex.
Panel A shows the largest recipients of Tether from Bitfinex recorded on Tether blockchain between
March 1, 2017 to March 31, 2018. Exchange wallet identities are obtained from the Tether rich list.
Moreover, intermediary wallets that receive Tether from Bitfinex but send all Tether to wallets of
a particular exchange are labeled as that exchange. Exchanges are distinguished by colors, and
the partitions show unique wallets within each exchange. The two largest recipients of Tether
from Bitfinex on Bittrex and Poloniex are labeled by the first four characters of their wallet ID as
1AA6 and 1J1d. Panel B shows the top recipients of Bitcoin on the Bitfinex exchange from other
exchanges between March 1, 2017 to March 31, 2018. The largest recipient of Bitcoin on Bitfinex
is labeled by the first four characters of its wallet ID as 1LSg. (Color figure can be viewed at
III. Are Bitcoin Prices Related to Tether?

In this section, we focus on the nature of the relationship between Bitcoin
prices and Tether, and we discuss how this relationship is connected to our
main hypotheses.
A. Examining Flows and Bitcoin Prices

Since demand curves for financial securities are typically not flat, demand
or supply shocks can have large effects on prices even in the absence of fun-
damental information (Harris and Gurel (1986), Shleifer (1986), Greenwood
(2005)), and may persist for surprisingly long periods of time (Duffie (2010)).
One should expect this effect to be stronger for cryptocurrencies because, first,
there are no fundamental cash flows from which prices are derived, and sec-
ond, the supply of coins is often fixed. In particular, if Tether issuances are
sizable, Bitcoin prices should be affected by movements of Tether into the mar-
ket. Moreover, as hypothesized in H2B, if Tether is being used to protect and
inflate the market, the effect of Tether transactions on Bitcoin prices should be
stronger following negative Bitcoin returns and on days after printing.
We estimate a regression of rolling three-hour average Bitcoin returns on
lagged average net hourly flow of Tether from Bitfinex to Poloniex and Bittrex
and of Bitcoin back to Bitfinex. We use the average three-hour Bitcoin returns
as our dependent variable, as the effect of flows might not be incorporated into
exchange prices immediately. The traceable flows on the blockchains indicate
when capital moves to the exchanges, not necessarily when the transactions
occur within the exchange. We expect the flow of Tether to an exchange to
precede the time when the Tether is used to purchase Bitcoin.27 For controls,
we include past returns to account for the effects of potential return reversals
(Lehmann (1990)), daily volatility of hourly returns in the previous 24 hours
to account for possible relations between returns and volatility, and lagged
returns interacted with volatility to account for the potential of larger return
reversals during periods of high volatility (Nagel (2012)).
Column (1) of Table II, Panel A, shows that on days right after Tether print-
ing, for a 100 Bitcoin increase in lagged flow, the three-hour average future
Bitcoin return goes up by 3.85 basis points, controlling for lagged returns,
volatility, and the interaction of lagged returns and volatility. Column (2) shows
that the effect exists only on days following Tether authorization, with no re-
lationship between the flow of Tether and Bitcoin prices on days apart from
printing Tether, consistent with the supply-driven price impact of hypothesis
H2B. Moreover, columns (3) and (4) show that the effect exists only after a
negative shock to Bitcoin prices. Finally, column (5) shows that the effect is
even stronger with a 8.13 basis point increase in returns when conditioning on
both Tether authorization and a lagged negative return.
27The standard errors are adjusted for heteroskedasticity and autocorrelation using the Newey-
West procedure with up to three lags.
Table II
The Effect of Flow of Bitcoin and Tether on Bitcoin Return
Panel A shows OLS estimates for which the dependent variable is the average three-hour Bitcoin
returns,
1
2
Rt+i = β0 + β1 Flowt−1 + Controls + t ,
3
i=0
where Rt is the hourly return of an equal-weighted price index that aggregates Bitcoin prices on
Tether exchanges Bitfinex, Poloniex, Bittrex, Binance, HitBTC, Huobi, and OKEx and Flowt is the
average net hourly flow of Tether from Bitfinex to Poloniex and Bittrex and of Bitcoin from Poloniex
and Bittrex to Bitfinex. The control variables include lagged returns, volatility calculated using
hourly returns over the previous 24 hours, and the interaction of lagged returns and volatility.
Column (1) shows the results for times when a Tether authorization occurred in the previous 72
hours and column (2) for other times. Columns (3) and (4) report results separately for observations
with lagged negative and positive returns. Column (5) reports results conditioning on both 72 hours
after Tether authorization and negative lagged returns. Panel B estimates the same regression
but decomposes flows into 1LSg flows and flows to other Poloniex and Bittrex accounts (described
in detail in Internet Appendix Section I). Panel B also controls for the net average flows of Tether
and Bitcoin to other Tether recipient exchanges (Binance, HitBTC, Huobi, Kraken, and OKEx).
Standard errors are adjusted for heteroskedasticity and autocorrelation. t-Statistics are reported
in parentheses. * p < 0.05, ** p < 0.01, ** p < 0.001.
Panel A: Regression of Returns on Lagged Flows
(1) (2) (3) (4) (5)

Auth NoAuth L.Ret < 0 L.Ret > 0 L.Ret < 0_Auth
Lag PLX BTX Flow 3.855* −0.354 2.694* −1.100 8.134**

(2.30) (−0.48) (2.18) (−1.20) (2.93)
LagRet −0.00600 −0.00985 0.0634* −0.0518 0.0897
(−0.18) (−0.57) (1.97) (−1.72) (1.46)
Volatility 103.9 97.00 −52.33 −70.32 −102.3
(1.17) (1.38) (−0.67) (−0.89) (−0.70)
Volatility* Lag Ret −0.343 −0.289 −1.443*** 0.609 −1.660**
(−0.94) (−1.14) (−3.40) (1.58) (−2.85)
Constant −8.071 −1.387 4.261 5.105 2.062
(−1.44) (−0.46) (1.26) (1.50) (0.24)
Observations 2,645 6,856 4,488 5,009 1,258
Adjusted R2 0.012 0.005 0.020 0.001 0.045
Panel B: Regression of Returns on Lagged Decomposed Flows
(1) (2) (3) (4) (5)

Lag 1LSg Flow 4.240* −0.484 2.379* −1.300 8.206***

(2.37) (−0.57) (1.97) (−1.24) (3.61)
Lag Other PLX BTX Flow 5.531 −0.513 4.602 −0.372 12.22
(1.20) (−0.26) (1.23) (−0.16) (1.32)
Lag Other Flow −6.483* 1.599 −0.514 0.322 −8.328*
(−2.36) (1.43) (−0.34) (0.25) (−2.38)
LagRet −0.00562 −0.0108 0.0650* −0.0523 0.0958
(−0.17) (−0.63) (2.01) (−1.73) (1.57)
(Continued)
Table II—Continued
Panel B: Regression of Returns on Lagged Decomposed Flows
(1) (2) (3) (4) (5)

Volatility 121.7 94.23 −51.05 −71.01 −84.21

(1.36) (1.33) (−0.65) (−0.90) (−0.57)
Volatility* Lag Ret −0.346 −0.281 −1.457*** 0.613 −1.717**
(−0.95) (−1.10) (−3.42) (1.59) (−2.95)
Constant −8.621 −1.334 4.203 5.108 1.784
(−1.53) (−0.44) (1.24) (1.50) (0.21)
Observations 2,645 6,856 4,488 5,009 1,258
Adjusted R2 0.014 0.005 0.020 0.001 0.049
To more precisely examine the source of the flow effect, we analyze three
different flow components: (i) the net Tether flows out from Bitfinex (and the
Bitcoin back) to the closely tied 1LSg addresses discussed above, (ii) the net
Tether flow out from Bitfinex (and the Bitcoin back) to the rest of Poloniex
and Bittrex accounts not involving the 1LSg addresses, and (iii) the rest of
the net Tether flows out from Bitfinex (and the Bitcoin back) to other Tether
exchanges including Binance, HitBTC, Huobi, Kraken, and OKEx. Column
(1) of Table II, Panel B, shows that on days right after Tether printing, for
a 100 Bitcoin increase in 1LSg flow, the three-hour average future Bitcoin
return goes up by 4.24 basis points, controlling for lagged returns, volatility,
and their interaction. The results are significant at the 5% level. There is no
significant positive relationship for the rest of the Poloniex and Bittrex flows
(flow component 2). The same is true for flows into other Tether exchanges.
In Table III, we examine whether the effect related to Tether printing spills
over into the six leading cryptocurrencies listed on Tether-related exchanges.
The effects are generally larger across all coins when conditioning on both days
after Tether authorization and following a negative return. For the equivalent
of a 100 Bitcoin increase in flow, the average future return goes up by 7.89 to
10.19 basis points for different coins.28
B. Large Flows and Prices

We now specifically focus on the 1% of hours (95 of 9,504 hours) with the
largest Tether/Bitcoin flow. Figure 5, Panel A, plots an event study of Bitcoin
and other cryptocurrency prices around these high-flow events. The high-flow
hours occur between times −1 and 0 by construction. The results show that
returns are large and negative between times −3 and −1. However, after the
large flow, the pattern starts to change at time 0. The next hour’s returns
are large at 80 basis points per hour, and returns are positive at 1.23% over
28 Internet Appendix Table IA.III reports similar results for the relationship between 1LSg flows
and other major cryptocurrency prices.
Table III
The Effect of Flow of Bitcoin and Tether on Other Cryptocurrency
Returns
This table shows OLS estimates for which the dependent variable is the average three-hour return
for major cryptocurrencies other than Bitcoin,
1
2
Rt+i = β0 + β1 Flowt−1 + Controls + t ,
3
i=0
where Rt is the hourly return using price data from the most liquid exchange for each cryptocur-
rency between March 1, 2017 and March 31, 2018 and Flowt is the average net hourly flow of
Tether from Bitfinex to Poloniex and Bittrex and of Bitcoin from Poloniex and Bittrex to Bitfinex.
The control variables include lagged returns, volatility calculated using hourly returns over the
previous 24 hours, and the interaction of lagged returns and volatility. Major cryptocurrencies are
selected based on the criteria in Table I, conditional on being listed on at least one of the major
Tether exchanges as of the beginning of March 2017. Panel A reports results for the 72 hours after
Tether authorization and Panel B reports results for other days. Panel C reports results when the
lagged return is negative and Panel D when the lagged return is positive. Panels E reports results
conditioning on both 72 hours after Tether authorization and negative lagged returns. Standard
errors are adjusted for heteroskedasticity and autocorrelation.
Panel A: Days Following Authorization
Coin Coefficient t-Stat N
DASH 6.16 3.26 2,645

ETC 7.54 3.00 2,645
ETH 6.29 3.10 2,645
LTC 6.17 1.83 2,645
XMR 4.80 2.19 2,645
ZEC 5.65 2.46 2,645
Panel B: Other Days
DASH 0.59 0.61 6,833

ETC −0.57 −0.52 6,833
ETH 0.54 0.58 6,833
LTC 1.32 1.27 6,833
XMR 0.13 0.12 6,833
ZEC 0.50 0.38 6,833
Panel C: Following Negative Returns
DASH 2.92 1.69 3,992

ETC 2.38 1.93 4,679
ETH 2.36 1.70 4,544
LTC 3.74 2.57 4,668
XMR 2.74 1.69 4,614
ZEC 3.12 2.00 4,785
(Continued)
Table III—Continued
Panel D: Following Positive Returns
DASH 3.47 2.26 3,985

ETC 1.92 0.99 4,732
ETH 1.65 1.27 4,878
LTC 1.99 0.94 4,581
XMR 0.59 0.48 4,752
ZEC 1.26 0.73 4,577
Panel E: Negative Returns-Authorization
DASH 10.19 3.26 1,063

ETC 8.84 3.00 1,271
ETH 8.86 3.10 1,246
LTC 8.54 1.83 1,293
XMR 7.44 2.19 1,244
ZEC 7.89 2.46 1,293
the next three hours after the flow. Panel B shows sharp positive returns in
the three-hour window after the flow events for all six of the other major
cryptocurrencies as well. We further examine the spillover in the cross-section
of cryptocurrencies by constructing an exchange-level value-weighted return
index of all coins other than Bitcoin using all other coin-BTC pairs for all
exchanges in the sample. The altcoins listed on Bitfinex, Poloniex, and Bittrex
have significantly larger Bitcoin-denominated returns than the coins listed on
other exchanges in the hours right after the flows (see Internet Appendix Table
IA.IV). Consistent with the effect being driven by Tether flows, the return is
not different before the high-flow periods.
We also examine the results for the largest player, 1LSg. In Internet Ap-
pendix Figure IA.10, we focus on the largest 1% of the 1LSg flows and finds
that returns are positive at 1.27% over the next three hours, while returns are
−1.50% over the three hours before. We test whether this behavior is linked
to a general increase in blockchain transactions by examining Bitcoin prices
around the times with high flows from Bitfinex to non-1LSg Poloniex and Bit-
trex wallets or to other Tether exchanges. We find no statistical or economic
effect around these times.
Note that the only conditioning variable for these hours is lagged flows, and
we do not condition on past returns, but the large negative returns preceding
the flows seem to be consistent with investors following a “buying-the-dips”
strategy. To see if a normally occurring reversal pattern rather than the im-
pact of flows is driving the returns, we find hours in the sample that are the
closest match to our 95 high-flow hours in terms of lagged returns in the pre-
vious three hours, but we do not condition on the high flow of Tether. Internet
Appendix Figure IA.11 shows that while the returns from times −3 to 0 are
the same by construction, the returns in the three hours after are −0.06% and
indistinguishable from zero, indicating that the higher returns after time 0 are
not due to a general price reversal or a “buying-the-dips” pattern in the market.
C. Is the Price Effect Economically Important?

What is the cumulative economic magnitude of the effects of Tether on Bitcoin
and other cryptocurrencies? Such a question is difficult to address. We take a
simple approach to partial economic assessment of the effect, but we also note
its potential limitations. From March 1, 2017 to March 31, 2018, the actual
Bitcoin price rose from around $1,191 to $6,929 for a return of 481.8%. In
contrast, the price series without the 95 Tether-related hours ends at around
$3,555, for an increase of 198.5%. Hence, the 1% of hours with the strongest
lagged Tether flow are associated with 58.8% of the Bitcoin buy-and-hold return
over the period.
We compare an actual Bitcoin price series to a series that is extremely similar,
but that removes the 95 high-lagged-flow hours discussed above and replaces
them with a random sample of 95 returns from other hours.29 This process
is repeated, with replacement, for 10,000 draws. Panel A of Figure 6 shows
that the actual return including the Tether-related hours clearly falls to the
far right of the bootstrapped distribution, indicating that it does not happen
by chance.
Panel B of Figure 6 compares the actual buy-and-hold return and the return
excluding hours after high flows for other major coins. The percentage of the
buy-and-hold return that is attributable to the Tether-related hours ranges
from 53% for Dash to 79% for Zcash.30 Across the six other cryptocurrencies,
returns are 64.5% smaller on average when removing the 95 Tether-related
flow hours.
We now perform the same analysis by focusing only on hours following the top
1% of 1LSg flows. From March 1, 2017 to March 31, 2018, excluding the top 1%
of times with high lagged flow of Tether and Bitcoin though 1LSg accounts, the
Bitcoin price rises only 216%. Hence, only 1% of the hours (95 of 9504) with the
strongest 1LSg flows are associated with 55.0% of the rise of Bitcoin in the next
hour. As shown in Internet Appendix Table IA.V, when removing the top 5% and
10% of hours, returns are 67.2% and 79.2% lower, respectively. We also perform
a bootstrap analysis for this account by replacing these 1% of hours with other
randomly selected hours. Figure 7 shows that the simulated distribution of
Bitcoin returns averages 221% and in none of the 10,000 simulations is the
return close to the actual return. The return distributions when replacing the
29 For example, for a three-period buy-and-hold return compounded as (1 + r1) (1 + r2) (1 + r3),
if period 1 is a high-flow hour, we replace the next-period returns, r2, with r2 , where r2 is a
random draw from all other nonhigh-flow hours in our sample. The benchmark buy-and-hold
return is calculated as (1 + r1) (1 + r2 ) (1 + r3). Note that this approach does not suffer from
look-ahead bias, as it depends only on past flows in replacing returns.
30 Ethereum, for example, experienced nearly a 2,400% return during this period, but if the
Tether-related hours were excluded it would have experienced around a 900% return.
Figure 5. Prices of Bitcoin and other cryptocurrencies around high-flow events. Panel
A shows Bitcoin prices three hours before and after the top 1% of high-flow hours to Poloniex and
Bittrex. Prices are scaled to one at time −3 before the event and at time 0 at the end of the event
window. Scaled prices are averaged across events. High-flow events are defined as the top 1% of
hours with high net average flows of Tether from Bitfinex to Poloniex and Bittrex and Bitcoin back
from Poloniex and Bittrex to Bitfinex in the prior hour, which means that high flows occur between
time −1 and time 0. Panel B depicts similar results for other major cryptocurrencies. (Color figure
can be viewed at wileyonlinelibrary.com)
Figure 6. Predictive effect of high-flow hours on cryptocurrencies returns. The red bar
in Panel A shows the buy-and-hold return of Bitcoin from March 1, 2017 to March 31, 2018. The
blue bars show the distribution of returns if the top 1% hours with high lagged flow of Tether
and Bitcoin are replaced with a random sample of returns in other hours, bootstrapped 10,000
times. High-flow hours are defined as in Figure 5. Panel B compares the actual buy-and-hold
return (red bars) with the return excluding the top 1% high-flow hours (blue bars) for other major
cryptocurrencies over the same time period. (Color figure can be viewed at wileyonlinelibrary.com)
hour following the top 5% and 10% of 1LSg flows are also considerably to the
left of the actual returns and indicate that the observed patterns are not likely
due to chance.
To determine whether the high-flow return relationship is a general result of
extreme market events reflected in the blockchain data, in Internet Appendix
Figure IA.12 we also perform simulations where we remove the top 1%, 5%,
Figure 7. Predictive effect of 1LSg high-flow hours on Bitcoin returns. The red bars show
the buy-and-hold return of Bitcoin from March 1, 2017 to March 31, 2018. The blue bars show the
distribution of returns if the top hours with high lagged 1LSg flow are replaced with a random
sample of returns in other hours, bootstrapped 10,000 times. The high 1LSg flow hours are the
top 1% of hours with high 1LSg flows as defined in the Internet Appendix Section I. The return
distribution in the top panel replaces the top 1% of high lagged 1LSg flow hours with a random
sample of returns in other hours, and the middle and bottom panels replace the top 5% and 10%,
respectively. (Color figure can be viewed at wileyonlinelibrary.com)
and 10% of net flows from Bitfinex to other Poloniex and Bittrex addresses.
There seems to be weak evidence that the extreme non-1LSg flows have some
effects on prices for the top 1% of hours, but not the top 5% and 10%. For
the net Tether/Bitcoin flows associated with the other five main Tether-based
exchanges (Binance, HitBTC, Huobi, Kraken, and OKEx), removing the top
1%, 5%, or 10% of the flows has no effect on simulated Bitcoin prices.
Overall, the findings indicate that a large player moves Tether out of Bitfinex
in exchange for Bitcoin in such a way that she/he would either have to exhibit
extreme market timing or, much more likely and consistent with the price
impact literature, have a large price impact on Bitcoin price.
We note that this finding is subject to some caveats. The effect only considers
the hourly periods with extreme flows. Measuring such findings over other
intervals would be less precise and more difficult, but the flow could push
prices up at other times as well. However, the effect does not consider the effect
of selling price pressure if the Tether issuers later sell the Bitcoin and move
the proceedings into dollars, though it seems feasible that the issuers could sell
Bitcoin through channels with considerably less price impact. If the purchased
Bitcoin is not permanently liquidated for dollars, then the inflationary effect
due to increasing the money supply can be persistent. Overall, although it
is difficult to fully assess the exact price impact of Tether, these back-of-the-
envelope calculations demonstrate that the effect is plausibly large.
D. Negative Serial Correlation in Bitcoin Prices

The flows of Tether and Bitcoin follow a specific pattern: accounts on Bitfinex
buy Bitcoins with Tether when Bitcoin prices drop. If the flow of Tether moves
Bitcoin prices, this may lead to a price reversal following a negative shock as
described in H2C.
To test this hypothesis, we examine whether future Bitcoin returns can be
explained by lagged returns, and in particular, whether the reversal effect
is related to Tether flows. We include controls for lagged volatility and the
interaction of lagged volatility and lagged returns similar to Table II. Table IV
shows that after controlling for volatility, we observe a return reversal, but only
for negative returns and only in periods with high net flows. Panel B shows
that the reversal pattern is driven by 1LSg flows and is not present in flows to
non-1LSg accounts on Poloniex and Bittrex, nor in flows to other Tether-based
exchanges.31 Panel C shows that the effect is strongest in periods right after
the hours with the top 1% and 5% of flows. In the extreme case, if accompanied
by top 1% net flows, each 1% drop in Bitcoin prices is followed by a large 61
basis point reversal in the next hour, whereas the reversal is on average only
6 basis points (and statistically insignificant) in other times.32 Controlling for
31 Internet Appendix Table IA.VI shows that if the specification controls for the interaction
between flows and volatility, the flow effect remains significant for the full sample but becomes
statistically insignificant when the sample is split into positive and negative lagged returns.
32 Internet Appendix Table IA.VII shows that the results are driven entirely by top hours of
1LSg flows and that top hours of other flows are not related to the reversal. For example, each 1%
Table IV
Bitcoin Return Reversals and 1LSg Flow
This table shows OLS estimates for the autocorrelation of Bitcoin returns,
Rt = β0 + β1 Rt−1 + β2 Flowt−1 + β3 Rt−1 ∗ Flowt−1 + Controls + t ,
where Rt is the hourly return of an equal-weighted price index that aggregates Bitcoin prices on
Tether exchanges, Flowt is the average net hourly flow of Tether from Bitfinex to Poloniex and
Bittrex and of Bitcoin from Poloniex and Bittrex to Bitfinex, and the control variables include
lagged returns, volatility calculated using hourly returns over the previous 24 hours, and the
interaction of lagged returns and volatility. Panel A reports results for aggregate net flows to
Poloniex and Bittrex. Panel B decomposes flows into 1LSg flows and the rest of Poloniex and
Bittrex accounts and controls for flows into other Tether exchanges (Binance, HitBTC, Huobi,
Kraken, and OKEx). The flow variables are standardized by subtracting the mean and dividing by
the standard deviation. Panel C estimates a similar regression for dummy variables that take the
value of 1 for top 1%, 5%, and 10% of hours with high lagged flows and volatility. Standard errors
are adjusted for heteroskedasticity and autocorrelation. t-Statistics are reported in parentheses.
* p < 0.05, ** p < 0.01.
Panel A: Using Aggregate Flows to PLX and BTX
Full Sample Neg Lagged Returns Pos Lagged Returns
Lag Ret −0.0198 0.0004 −0.0420

(−0.62) (0.01) (−0.69)
Lag Flow 0.0003 −0.0002 0.0001
(1.68) (−0.53) (0.34)
Lag Flow × Lag Ret −0.0326** −0.0669** −0.0073
(−2.73) (−2.67) (−0.36)
Lag Volatility 0.0093 0.0060 0.0100
(1.38) (0.49) (0.88)
Lag Volatility × Lag Ret −0.3961 −0.5918 −0.2719
(−0.98) (−0.85) (−0.37)
Constant −0.0002 −0.0000 −0.0001
(−0.67) (−0.07) (−0.29)
Observations 9,503 4,488 5,011
Adjusted R2 0.007 0.011 0.001
Panel B: Using Decomposed Flows
Lag Ret −0.0125 0.0166 −0.0320

(−0.38) (0.27) (−0.52)
Lag 1LSg Flow 0.0003 −0.0001 −0.0000
(1.71) (−0.19) (−0.02)
Lag 1LSg Flow × Lag Ret −0.0280* −0.0545* 0.0050
(−2.23) (−2.17) (0.22)
Lag Volatility 0.0094 0.0060 0.0110
(1.40) (0.49) (0.97)
Lag Volatility × Lag Ret −0.4986 −0.7798 −0.4123
(−1.20) (−1.11) (−0.55)
Lag PLX BTX Flow × Lag Ret −0.0200 −0.0272 −0.0153
(−1.61) (−1.41) (−0.95)
(Continued)
Table IV—Continued
Panel B: Using Decomposed Flows
Lag Other Flow × Lag Ret 0.0255 0.0404 0.0094

(1.81) (1.63) (0.56)
Constant −0.0002 0.0000 −0.0002
(−0.71) (0.01) (−0.42)
Adjusted R2 0.008 0.012 0.001
Panel C: Using the Top Percentile Flow and Volatility (Lagged Neg Returns)
Top 1% Top 5% Top 10%
Lag Ret −0.0583 −0.0169 −0.0299

(−1.90) (−0.49) (−0.78)
Lag High Flows 0.0041 0.0003 −0.0011
(0.72) (0.19) (−0.85)
Lag High Flows=1 × Lag Ret −0.6091* −0.2720* −0.1756
(−2.56) (−2.53) (−1.93)
Lag High Vol 0.0167* −0.0018 0.0008
(2.53) (−0.73) (0.51)
Lag High Vol=1 × Lag Ret 0.2014 −0.1192 −0.0183
(1.09) (−1.42) (−0.26)
Constant −0.0000 0.0003 0.0002
(−0.04) (1.07) (0.70)
Adjusted R2 0.023 0.013 0.007
the interaction between lagged returns and volatility shows that the results
cannot be explained by the possibility of larger return reversals during periods
of high volatility (Nagel (2012)).
In conclusion, the results in this section provide considerable evidence that
Tether is used to purchase Bitcoin following Tether authorization and a drop
in Bitcoin price, and that this phenomenon has a sizable relation with fu-
ture prices of Bitcoin and other coins. This relation is driven by one ac-
count holder and induces an asymmetric negative autocorrelation in Bitcoin
returns.
IV. Is Tether Pushed or Pulled?

The results in the previous section are consistent with a sizable price impact
of Tether. In this section, we examine pushed H2D and H2E as well as variants
of the pulled hypothesis to shed light on the nature of this price impact.
drop in Bitcoin prices is followed by a 52 basis point reversal in the next hour if accompanied by
the top 1% of 1LSg flows.
A. Currency Flows around Round Price Thresholds

Following Tversky and Kahneman (1974), a large literature demonstrates
the importance of price anchoring for a variety of assets. Shiller (2000) ex-
tensively discusses the importance of psychological anchors for stock market
prices, and indicates that one such anchor is the nearest round-number level.
Bhattacharya, Holden, and Jacobsen (2012) find support for liquidity deman-
ders buying just below round-number thresholds in stocks, consistent with in-
vestors anchoring prices to the round-number threshold. Such an anchor could
be of particular importance for cryptocurrency prices, for which the underlying
value cannot be gauged through fundamentals.
Additionally, cryptocurrency traders likely engage in technical trading
whereby past price movements generate buy and sell signals. If Tether is used
to stabilize market prices during a downturn, one might expect a spike in the
flow of Tether around round thresholds, as this might induce other traders,
upon observing technical support at the threshold, to purchase as well. Such a
pattern could also be consistent with recent theories that suggest that higher
participation of users and investors makes Bitcoin more appealing to other
users/investors due to network effects (Sockin and Xiong (2018), Cong, Li, and
Wang (2019)).
To test this prediction, we divide hourly CoinDesk prices by 500 and then
group the remainders into bins of $10 width to examine how the flow of Tether
for Bitcoin changes near the round thresholds. Figure 8 plots the net average
flow of Bitcoin and Tether between Bitfinex and other Tether exchanges as a
function of distance to the round thresholds. Panel A shows that on days after
Tether authorization, the flow increases significantly just below the round cut-
off but drops right above the cutoff. In contrast, there is no such effect on days
with no prior Tether authorization. Panel B plots the flows after authorization
for net 1LSg flows and flows to other accounts. We find evidence of strong flows
below the threshold for 1LSg accounts. There is some weaker evidence of larger
flows below the threshold for the rest of Bittrex and Poloniex (not coming from
1LSg) and no evidence of net Bitcoin buying around round number thresholds
for Binance, HitBTC, Huobi, Kraken, or OKEx.
Table V, Panel A, formally tests whether Tether/Bitcoin flow is different
below and above the round-price thresholds. The dependent variable is the
net Tether/Bitcoin flow, and the independent variable is a dummy that takes
the value of 1 if the Bitcoin price is in the $50 bandwidth below the round
multiples of $500 and 0 if in the $50 bandwidth above. The results show that
purchasing below the threshold is economically and statistically significant
only after authorization.
In Panel B of Table V, we further examine the disaggregated flows following
authorization and finds that the higher flow below round-number thresholds is
driven by the 1LSg accounts, with a t-statistic of 3.71. Other accounts at Bittrex
and Poloniex as well as other Tether exchanges do not have statistically or
economically significant flows below the threshold. In addition, Panel C shows
that no such pattern obtains in nonauthorization periods. Overall, the evidence
Panel A. Aggregate Bittrex and Poloniex Flows

Days Following Authorization Other Days
200
80
Hourly Average Flow (BTC)

150
60
100
20 40
50
0
0
−100 −50 0 50 100 −100 −50 0 50 100

Distance from Round Threshold Distance from Round Threshold
Panel B. Decomposed and Other Flows

ILSg Rest of Bittrex Rest of Poloniex

10 20 30
−10 −5 0 5 10 15
40 80 120
−10 0
0
−100 −50 0 50 100 −100 −50 0 50 100

Distance from Round Threshold Distance from Round Threshold Distance from Round Threshold
Binance HitBTC Huobi


20 40 60
10 15
10 20 30
0 5
−20 0
−10 0
−5
−100 −50 0 50 100 −100 −50 0 50 100 −100 −50 0 50 100

Distance from Round Threshold Distance from Round Threshold Distance from Round Threshold
Kraken OKEx

10 15
−20 −10 0 10 20 30
−5 0 5
−100 −50 0 50 100 −100 −50 0 50 100

Distance from Round Threshold Distance from Round Threshold
Figure 8. Flows around round number thresholds. This figure shows the average net hourly
flows of Tether from Bitfinex to two major Tether exchanges, Poloniex and Bittrex, and of Bitcoin
from these exchanges to Bitfinex, around round-number thresholds of Bitcoin prices. The Bitcoin
prices are based on hourly prices reported by CoinDesk. The horizontal axis shows the distance
of the price from round thresholds in multiples of $500 at the end of the previous hour, and the
vertical axis shows the flow within the hour. The hollow blue circles show the average flow for $10-
wide price bins, and the black lines show the fitted values of the flow as a second-order polynomial
of the price distance to the round thresholds. The gray areas represent the 95% confidence interval
for the fitted values. Panel A, left, plots the results for times when a Tether authorization occurred
in the previous 72 hours, and Panel A, right, plots the results for other times. Panel B shows the
results after Tether authorization for the flows decomposed into 1LSg flows and other Poloniex and
Bittrex accounts, as well as flows to other Tether-based exchanges. The sample covers the period
from March 1, 2017 to March 31, 2018. (Color figure can be viewed at wileyonlinelibrary.com)
Table V
Flow of Coins around Round Thresholds of Bitcoin Price
Panel A reports OLS estimates for which the dependent variable is hourly average net flow of
Tether from Bitfinex to Poloniex and Bittrex and of Bitcoin from Poloniex and Bittrex to Bitfinex.
Below RoundCuto f ft is a dummy variable that takes the value of 1 if the Bitcoin price, at the end of
the hour, falls into the $50 price bucket below a $500 price multiple and 0 if it is in the $50 bucket
above such a multiple,
Flowt = β0 + β1 Below RoundCuto f ft−1 + t .
Panel B estimates the same regression for the net average flows into 1LSg accounts, the rest of
Poloniex and Bittrex accounts, and the other Tether exchanges (Binance, HitBTC, Huobi, Kraken,
and OKEx). Standard errors are adjusted for heteroskedasticity and autocorrelation. t-Statistics
are reported in parentheses. * p < 0.05.
Panel A: Flows around Round Thresholds
Full Auth NoAuth
Below Round Cutoff 14.75* 60.83*** 0.221

(2.02) (3.52) (0.03)
Constant 36.26*** 45.55*** 31.93***
(8.52) (5.19) (6.78)
Observations 1,603 464 1,139
Adjusted R2 0.002 0.028 −0.001
Panel B: Flows to Different Exchanges—Days Following Authorization
1LSg Oth BTX Oth PLX Binance HitBTC Huobi Kraken OKEx
Below Round Cutoff 52.60*** 2.059 6.172 7.497 3.810 6.289 5.252 0.971
(3.71) (0.60) (1.62) (1.27) (1.92) (1.90) (0.83) (0.46)
Constant 34.75*** 4.885*** 5.915** 13.66*** 0.564 3.766** −1.071 3.841***
(4.63) (3.93) (3.08) (4.42) (0.64) (3.01) (−0.38) (3.52)
Observations 464 464 464 305 464 464 464 260
Adjusted R2 0.030 −0.001 0.004 0.002 0.007 0.008 −0.000 −0.003
Panel C: Flows to Different Exchanges—Other Days
1LSg Oth BTX Oth PLX Binance HitBTC Huobi Kraken OKEx
Below Round Cutoff 5.815 −2.825 −2.768 −1.085 −0.835 −0.476 0.207 2.043
(0.89) (−1.33) (−1.47) (−0.47) (−1.23) (−0.12) (0.17) (0.71)
Constant 19.93*** 4.982*** 7.015*** 3.442* 0.761* 4.123 −0.00519 −0.542
(4.99) (3.43) (5.43) (2.01) (2.29) (1.32) (−0.01) (−0.22)
Observations 1,139 1,139 1,139 731 1,139 1,139 1,139 483
Adjusted R2 −0.000 0.001 0.001 −0.001 0.001 −0.001 −0.001 −0.001
indicates that the flow below thresholds is driven by the 1LSg account, and
only after authorization, that is, this flow pattern is not typically observed in
the market.
We next examine what effect, if any, the inflow of Tether below the threshold
might have on Bitcoin returns. In Panel A of Table VI, we report estimates of
a regression of average three-hour future returns on the lagged round-number
Table VI
Effect of Flow on Returns around Round Thresholds of Bitcoin Price
Panel A estimates a regression of average three-hour Bitcoin returns on the Below RoundCuto f f
dummy. Panel B reports results for the second-stage estimates of a two-stage least squares regres-
sion of Bitcoin returns on flows,
1
2
ˆ t−1 + t ,
Rt+i = β0 + β1 Flow
3
i=0
ˆ t is instrumented using a dummy variable that takes the value of

where in the first stage, Flow
1 if the Bitcoin price, at the end of the previous hour, is within the $50 bucket below the round
threshold and the time is within the three-day window after Tether authorization and 0 if within
the $50 bucket above or in days outside the three-day window after Tether authorization. Panel
C reports the same results as in Panel B but where the flows are decomposed into 1LSg and the
rest of Poloniex and Bittrex, and it also controls for aggregate net flows to other Tether exchanges
(Binance, HitBTC, Huobi, Kraken, and OKEx). Standard errors are adjusted for heteroskedasticity
and autocorrelation. t-Statistics are reported in parentheses. * p < 0.05.
Panel A: Returns around Round Thresholds
Auth NoAuth Auth_L.Ret < 0 Auth_L.Ret > 0
Below Round Cutoff 20.61* −3.397 32.87* 11.91

(2.42) (−0.74) (2.58) (1.29)
Constant 1.765 5.466 11.75 −7.205
(0.33) (1.87) (1.39) (−1.15)
Observations 464 1,138 214 250
Adjusted R2 0.012 0.000 0.025 0.002
Panel B: Instrumenting the Flow using the Round Thresholds
All Auth Auth L.Ret < 0 Auth_L.Ret > 0
Flow 26.42* 33.88* 45.34* 22.92

(2.06) (2.05) (2.37) (0.97)
Constant −5.724 −13.67 −10.75 −16.81
(−1.05) (−1.27) (−0.72) (−1.23)
Observations 1,602 464 214 250
Wald F-statistic 19.44 12.03 8.217 5.264
Panel C: Instrumenting the 1LSg Flow using the Round Thresholds
All Auth Auth_L.Ret < 0 Auth_L.Ret > 0
1LSg Flow 38.52* 65.44* 89.35 47.27

(2.09) (2.03) (1.79) (1.11)
Oth PLX/BTX Flow −21.19 −52.65 −76.91 −47.82
(−1.78) (−1.45) (−1.08) (−1.26)
Oth Flow −10.18 −38.09* −35.38 −40.03
(−1.92) (−2.10) (−1.73) (−1.21)
Constant −3.364 −10.28 −8.653 −11.08
(−0.75) (−1.01) (−0.53) (−0.99)
Observations 1,602 464 214 250
Wald F-statistic 19.49 7.639 3.291 4.277
threshold dummy. On days following Tether authorization, when prices are

below the round threshold, the future hourly return is 20.61 basis points higher
on average. However, this return effect is not present on days apart from
printing Tether or periods after authorization with positive lagged returns.
Note that it is possible that the Bitfinex-related wallets trade around round-
number thresholds simply because they are following behavioral biases. How-
ever, in this case their trading is not likely to be profitable as documented in
the behavioral finance literature (Bhattacharya, Holden, and Jacobsen (2012)).
Large purchasing by 1LSg accounts provides a coherent explanation as to how
prices can be pushed above the thresholds. In addition, if other traders see such
large purchasing, they might join the buying due to either technical trading
indicators being triggered or through the perception of stronger network effects
(Sockin and Xiong (2018), Cong, Li, and Wang (2019)).
We also use the discontinuity around round-number thresholds as an instru-
ment to identify the effect of Tether on Bitcoin prices by estimating a fuzzy
regression discontinuity design. As an instrument for Tether-related flows, we
set a dummy variable equal to 1 if Bitcoin price is within the $50 bucket below
the round threshold and the time is within the three-day window after Tether
authorization. Our identification assumption is that the only channel through
which the cutoff affects future Bitcoin returns is through Tether flows. The
exclusion restriction is supported by the fact that neither 1LSg flows nor the
future Bitcoin returns differ below and above the thresholds on days not around
Tether authorization, and flows associated with no other accounts differ below
and above the thresholds even for periods after authorization.
In Panel B of Table VI, we estimate a two-stage least squares regression
of three-hour future Bitcoin returns on the lagged net Bitcoin/Tether flow,
where the flow is instrumented using the cutoff dummy. The reported Wald F-
statistics show that the first-stage regressions are strong, suggesting a strong
instrument. The second-stage regression indicates that for 100 Bitcoin pur-
chased by Bitfinex, the average hourly Bitcoin return in the next three hours
goes up by 26.42 basis points. The effect is 33.88 basis points if the sample
is limited to days after authorization, and 45.34 basis points for periods after
authorization with lagged negative returns. The effect is insignificant for pe-
riods after authorization with positive lagged returns. In Panel C, we perform
the same analysis except we instrument for 1LSg flows rather than aggregate
Poloniex and Bittrex flows, and we also control for the flows associated with
other accounts on Poloniex and Bittrex as well as on other exchanges. The re-
sults are economically larger with a 100 Bitcoin flow by 1LSg associated with
an average hourly Bitcoin return in the next three hours of 65.44 basis points
after authorization. This result highlights a very strong effect of 1LSg flows on
Bitcoin prices, especially on days after Tether authorization.
B. Demand from Investors with Fiat Currency?

B.1. End-of-Month Returns
The previous sections establish that the flow of Tether explains a sizable
increase and predictable trading patterns in Bitcoin prices. These patterns are
potentially consistent with fiat purchases of Tether through Bitfinex, but the
purchases and trading would need to be driven by one large player who moved
over 2 billion USD into Tether through the Bitfinex exchange. Alternatively,
if the printed Tether is not backed by dollars and does not reflect the inflow
of real capital into the cryptospace, such an increase in Bitcoin prices can re-
flect inflation caused by printing unbacked money. In this section we examine
the backing of Tether by borrowing from the intermediary asset pricing litera-
ture, specifically Du, Tepper, and Verdelhan (2018) and He and Krishnamurthy
(2018), who argue that banks’ compliance with period-end capital requirements
may have a sizable effect on asset prices. To assure traders of the existence of
dollar reserves, Tether has issued EOM bank statements from December 2016
to March 2017 that were audited by a Chinese accounting firm.33 If Tether does
not maintain full reserves daily but seeks to release audited EOM statements
that demonstrate full reserves to investors, there could be negative selling pres-
sure on Bitcoin to convert it to USD reserves before the EOM as hypothesized
in H2E. Such an EOM selling effect should be related to the Tether issuance.
Moreover, if cash needs to be raised by liquidating other major cryptocurrencies,
as they also show a large price increase around Tether flows, they should show
an EOM effect as well. We test for this effect by constructing value-weighted
returns of the top-five cryptocurrency returns.
Figure 9 depicts Bitcoin daily returns at EOM by dividing the sample months
into four quantiles based on their monthly Tether issuance.34 The blue bars
show the raw EOM returns, and the red bars benchmark the EOM returns by
subtracting the average return over the four days before and the four days after.
As can be seen, there is a clear relationship between monthly Tether issuance
and EOM negative price pressure. In months with no Tether issuance, there is
no EOM effect. However, in months with large Tether issuance, there is a 6%
negative benchmarked return.
We caveat this relation, however, by noting that there are only 25 months in
our sample, and the two months with the largest Tether issuance, December
2017 and January 2018, exhibit a strong EOM effect. Because of the relatively
33 As announced on https://tether.to/tether-update/, these audits were made publicly available
on Tether.to. Tether also stated its intention to be audited by a non-Chinese firm, but it eventually
canceled the audit due to “the excruciatingly detailed procedures.” In an interview about the lack
of an audit on Tether, Bitfinex’s chief technology officer noted that “[w]hat we want to do is not
[audit] the bank balances as of now, but we want to demonstrate to the community that we had
the money at the end of every single month, since a reasonable date like January 2017 and on.”
34 Cryptocurrencies officially trade on UTC timestamp and daily prices close at midnight UTC
time, when business hours have already ended in most countries and the next day has already
started in East Asia. The effect must therefore be observed in the second-to-last day of the month,
which we consider the EOM price.
Figure 9. End-of-month returns and quantiles of Tether issuance. This figure shows end-
of-month (EOM) daily Bitcoin returns for different quantiles of monthly Tether issuance. Four
quantiles of Tether issuance are defined based on total Bitcoin-denominated Tether issuance each
month. Issuance is calculated as the aggregate monthly Bitcoin-denominated flow of Tether from
the Tether treasury to Bitfinex. All months with zero issuance are included in one group, and the
other months are divided into three quantiles. The EOM return is defined as the daily return
on the second-to-last day of the month closing at midnight UTC time. Daily prices are obtained
from CoinMarketCap. The blue bars show the raw EOM return, and the red bars show the raw
return minus the average return from the prior four days through the subsequent four days.
The sample covers the period from March 2016 to March 2018. (Color figure can be viewed at
small sample size, we check the sensitivity of the results by excluding the
two months with the largest Tether issuances. In a simple regression of EOM
Bitcoin returns on monthly Tether issuances, we obtain a t-statistic of −2.85
with all observations, but an insignificant t-statistic of −1.26 when excluding
the two largest months.35
In Table VII, we examine this result further. In Panel A, column (1) shows
that the EOM return is 2.3% less than returns in the four days before and
after the EOM. Columns (2) and (3) indicate that there is no effect in months
without Tether issuance, but the EOM return is 3.8% lower in months with
Tether issuance (t-statistic of 3.65). Column (4) interacts the EOM dummy
with the magnitude of the monthly Tether issuance and shows that for a one-
standard-deviation higher Tether issuance, the EOM return is 2.2% more neg-
ative. Column (5) tests the plot in Figure 9 statistically and shows that relative
to months with zero issuance, months with low, medium, and high issuance
35 When using the value-weighted returns of top-five currencies, the same regression yields a t-
statistic of −4.85 and −2.97 with and without the top two months, respectively (Internet Appendix
Table IA.VIII).
Table VII
EOM Bitcoin Returns and the Effect of Tether Issuance
This table reports OLS estimates for which the dependent variable is daily Bitcoin returns and the independent variables are the EOM dummy and
monthly Tether issuance,
Rt = β0 + β1 EOMt + β2 Issuancet + β3 EOMt ∗ Issuancet + t ,
where EOMt takes the value of 1 on the second-to-last day of the month at midnight UTC time and Issuancet is the aggregate monthly Bitcoin-
denominated flow of Tether from the Tether treasury to Bitfinex scaled by its standard deviation. Column (5) interacts the EOM dummy with quantiles
of issuance as defined in Figure 9. The sample is from March 2016 to March 2018. Columns (6) to (8) report results after excluding the two months
with extreme issuance, December 2017 and January 2018. Panel B estimates the results using the returns on a value-weighted portfolio of top-five
cryptocurrencies. Each day in the sample, the top-five cryptocurrencies are selected based on average market cap in the previous week as reported on
CoinMarketCap. Standard errors are robust to heteroskedasticity. t-Statistics are reported in parentheses. * p < 0.05, ** p < 0.01, *** p < 0.001.
Panel A: Bitcoin Returns
Full Sample Excluding 12/2017 and 1/2018
(1) (2) (3) (4) (5) (6) (7) (8)

All NoIssuance Issuance All All Issuance All All
EOM −0.0230** −0.000788 −0.0377*** −0.00669 −0.000788 −0.0251*** −0.00869 −0.000788

(−3.24) (−0.14) (−3.65) (−1.41) (−0.14) (−4.70) (−1.84) (−0.14)
Issuance 0.00123 0.00546
(0.39) (1.63)
EOM = 1 × Issuance −0.0222** −0.0107*
(−2.85) (−2.04)
Low × EOM −0.0187* −0.0187*

(−2.27) (−2.27)
Med × EOM −0.0307** −0.0307**
(−2.71) (−2.70)
High × EOM −0.0615* −0.0232*
(−2.40) (−1.98)
Low 0.0117* 0.0117*
(2.08) (2.08)
Med 0.00933 0.00933
(1.33) (1.32)
High 0.00908 0.0126
1955
(1.07) (1.57)
(Continued)
Table VII—Continued
Panel A: Bitcoin Returns

1956
(1) (2) (3) (4) (5) (6) (7) (8)

Constant 0.0110*** 0.00501 0.0150*** 0.0101*** 0.00501 0.0160*** 0.00824** 0.00501

(4.32) (1.49) (4.19) (3.58) (1.48) (4.78) (2.92) (1.47)
Observations 225 90 135 225 225 117 207 207
Adjusted R2 0.035 −0.011 0.078 0.065 0.060 0.048 0.024 0.023
Panel B: Top Five Value-Weighted Returns
(1) (2) (3) (4) (5) (6) (7) (8)

EOM −0.0216** 0.00107 −0.0367*** −0.00188 0.00107 −0.0241*** −0.00301 0.00107

(−3.19) (0.25) (−3.68) (−0.46) (0.25) (−4.05) (−0.79) (0.25)
Issuance 0.00179 0.00460
(0.53) (1.53)
EOM = 1 × Issuance −0.0269*** −0.0186***
(−4.45) (−3.41)
Low × EOM −0.0175* −0.0175*
The Journal of FinanceR
(−2.41) (−2.40)
Med × EOM −0.0196* −0.0196*
(−2.07) (−2.07)
High × EOM −0.0762*** −0.0474***
(−3.99) (−3.63)
Low 0.0119* 0.0119*
(2.34) (2.34)
Med 0.00990 0.00990
(1.35) (1.35)
High 0.0106 0.0100
(1.31) (1.39)
Constant 0.0101*** 0.00367 0.0145*** 0.00884** 0.00367 0.0143*** 0.00719** 0.00367
(4.08) (1.18) (4.06) (3.16) (1.18) (4.35) (2.69) (1.17)
Observations 225 90 135 225 225 117 207 207
Adjusted R2 0.033 −0.011 0.076 0.083 0.084 0.045 0.030 0.030
have a negative EOM return of 1.9%, 3.1%, and 6.1%, respectively, all sta-
tistically significant. Finally, as a sensitivity check, in columns (6) to (8), we
exclude the top two months of flow. As expected, the results are weaker but
still statistically and economically significant.
Panel B examines the findings using the value-weighted return index. The
findings are considerably more statistically significant. The index shows a re-
turn of −7.7% in the months with the highest issuance with a t-statistic of
−4.00. If we remove December 2018 and January 2018, the magnitude is still
4.8% with a t-statistic of −3.64.
As a one-period example not at EOM, we also noticed that Tether released
a limited audit of a snapshot of their cash balance as of September 15, 2017.
Tether later fired the auditor. Prices dropped 25% from September 12, 2017 to
September 15, 2017, the day of the audit (see Internet Appendix Figure IA.13).
Finally, we examine if there are any patterns in Bitfinex’s Bitcoin wallets
used to hold the exchange Bitcoin reserves.36 If the founders attempt to sell
Bitcoin and raise a cash reserve, the balance in the reserve wallets of Bitfinex
might go down before the EOM. To examine this possibility, we compute the
net flows of Bitcoins from Bitfinex’s reserve wallets, including its main cold
wallets. Internet Appendix Table IA.IX shows that in months with large Tether
issuances, the Bitfinex balances experience a large net outflow in the last
five days of the month, and the relationship is statistically significant with
a t-statistic of 3.14. As a placebo test, we perform the same analysis on the
reserve wallets of any of the top-20 largest exchanges for which we could obtain
reserve wallet addresses, and we find no EOM net outflow from these wallet
balances. This result suggests that a plausible channel for the decrease in
Bitcoin prices is EOM liquidation of Bitfinex reserves. In summary, the strong
negative effect on Bitcoin prices in months of Tether issuance is consistent
with Tether not maintaining full dollar reserves at all times. Without a dollar
backup, the Tether peg could be held when cryptocurrency prices increase and
the liquidation of Tether is limited. But if market participants lose confidence
in Tether and a run occurs, there can be a substantial risk of default without
full cash reserves. Like most runs, this could also lead to substantial collateral
damage to cryptocurrency investors.
B.2. Flows and the Tether-USD Rate

Although the analysis above shows substantial support for a supply-based
explanation, we further examine the demand-based explanations for Tether. If
the demand for Tether comes mainly from investors who hold dollars and seek
to invest in Bitcoin, the greater demand could translate into a higher market
rate for the Tether-USD pair. Kraken was the most active market-based venue
for exchanging Tether for dollars in 2017, although the market volume of the
36 These wallets can include cold wallets or other wallets that hold a large balance of Bitcoin
reserves for a specific exchange. The table header to Internet Appendix Table IA.IX describes how
we identify these wallets on the blockchain.
pair was less than 1% of the Bitcoin-Tether volume. The rate on Kraken often
stays close to one over our sample period from March 1, 2017 to March 31, 2018
but has a standard deviation of 2%. If part of the demand for Tether spills over
to Kraken, one would expect changes in the Tether-USD rate to be related to
the flow of Tether.
In Panel A of Table VIII, we regress Tether flow on different lags of Tether-
USD returns as well as BTC-USD returns. We standardize the variables so
that the magnitudes of the coefficients are comparable. The results show that
Tether flow is highly sensitive to the BTC-USD pair (as shown previously) but
bears little relation to the Tether-USD pair. Similarly, in Panel B, we examine
Bitcoin flow and find that the corresponding flow of Bitcoin back is highly
sensitive to BTC-USD rates but bears no relationship with the Tether-USD
pair. We further examine this relationship by constructing different proxies for
the Tether price using value-weighted and equal-weighted Tether-USD rates
across all available exchanges as well as constructing a synthetic rate using
Bitcoin prices on Bitfinex versus dollar exchanges. The results using these
proxies instead of the Kraken Tether-USD rate are similar (Internet Appendix
Tables IA.X, IA.XI, and IA.XII). We also examine results for the 1LSg account
and other accounts on Tether exchanges and find similar results (Internet
Appendix Table IA.XIII).
Another possibility is that the overall price difference between Tether and
USD exchanges is driving the flow. To examine this possibility, we construct two
lagged return measures: the three-hour lagged Bitcoin return averaged across
all major exchanges, and the three-hour lagged difference in return between
Tether exchanges and USD exchanges. The average return captures the effect
of Bitcoin price changes and the difference captures the spread leading to the
arbitrage opportunity between Tether and USD exchanges. We then estimate
a regression of Tether and Bitcoin flows on the spread and average returns.
Panel C of Table VIII shows that the flows are not sensitive to the spread.
Moreover, Panel C of Internet Appendix Table IA.XIII shows that the flows
to 1LSg and other Poloniex and Bittrex accounts have no relationships with
the spread, whereas the flows to Binance and Huobi are positively related to
the spread. These findings suggest that when the BTC-Tether pair trades at
a higher discount relative to BTC-USD, capital flow to Binance and Huobi
increases to buy Bitcoin at a lower price. This result indicates that Tether is
used in arbitrage activities, but the 1LSg activities are not driven by these
arbitrage proxies.
Overall, we do not find evidence to support the demand-based hypothesis
(H1A), but we also note that noise and illiquidity in the Tether return series
add noise to these tests. We believe that the various ways we construct for the
actual and implied Tether return series substantially mitigate this concern.
C. Flows and Bitcoin Prices across Exchanges

Tether may facilitate cross-exchange arbitrage among Tether exchanges. In
particular, imagine that Bitcoin prices increase on Bitfinex, but Bitcoin prices
Table VIII
The Relationship between Tether and Bitcoin Flows and Tether-USD
versus BTC-USD Rates
This table reports OLS estimates for which the dependent variables are the net flow of Tether from
Bitfinex (Panel A) and the net flow of Bitcoin to Bitfinex (Panel B), and the independent variables
are multiple lags of Tether-USD and BTC-USD returns,

5
5
T ether−U SD BT C−U SD
Flowt = α + βi Rt−i + γi Rt−i + t ,
i=1 i=1
where RtBT C−U SD is the hourly return of Bitcoin prices in USD and RtT ether−U SD is the hourly
return of the Tether-USD pair on the Kraken exchange. The sample period is from April 1, 2017
(when Kraken prices are first available) to March 1, 2018. Panel C estimates an OLS regression
of Tether and Bitcoin flows on the lagged arbitrage spread and average returns between USD and
Tether exchanges,
1 1
3 3
Flowt = β0 + β1 ArbitrageSpreadt−i + β2 Average Returnt−i + t ,
3 3
i=1 i=1
(RU SD +RT ether )

where Average Returnt = t 2
t
and ArbitrageSpreadt = RtU SD − RtT ether . All variables are
standardized by subtracting the mean and dividing by the standard deviation. Standard errors are
robust to heteroskedasticity. t-Statistics are reported in parentheses. * p < 0.05, ** p < 0.01, *** p <
0.001.
Panel A: Tether Flow
L.Tether_USD_Ret −0.0082 −0.0016 0.0019 0.0018 0.0047

(−0.77) (−0.13) (0.16) (0.14) (0.36)
L2.Tether_USD_Ret 0.0080 0.0160 0.0180 0.0232
(0.59) (1.11) (1.21) (1.42)
L3.Tether_USD_Ret 0.0138 0.0176 0.0257
(1.23) (1.32) (1.71)
L4.Tether_USD_Ret 0.0024 0.0172
(0.20) (1.05)
L5.Tether_USD_Ret 0.0272
(1.50)
L.BTC_USD_Ret −0.0448** −0.0472*** −0.0482*** −0.0489*** −0.0490***
(−3.14) (−3.31) (−3.40) (−3.44) (−3.45)
L2.BTC_USD_Ret −0.0688*** −0.0698*** −0.0715*** −0.0719***
(−4.80) (−4.84) (−4.96) (−4.95)
L3.BTC_USD_Ret −0.0299* −0.0316** −0.0325**
(−2.56) (−2.70) (−2.73)
L4.BTC_USD_Ret −0.0419** −0.0426**
(−3.05) (−3.12)
L5.BTC_USD_Ret −0.0263
(−1.85)
Constant −0.0034 −0.0032 −0.0031 −0.0030 −0.0029
(−0.31) (−0.29) (−0.28) (−0.27) (−0.27)
Observations 8,750 8,749 8,748 8,747 8,746
Adjusted R2 0.002 0.006 0.007 0.008 0.009
(Continued)
Table VIII—Continued
Panel B: Bitcoin Flow
L.Tether_USD_Ret −0.0047 0.0029 0.0033 0.0061 0.0075

(−0.34) (0.21) (0.23) (0.42) (0.51)
L2.Tether_USD_Ret 0.0098 0.0085 0.0150 0.0167
(0.74) (0.58) (1.02) (1.10)
L3.Tether_USD_Ret −0.0139 −0.0012 0.0021
(−1.04) (−0.08) (0.14)
L4.Tether_USD_Ret 0.0212 0.0271
(1.62) (1.93)
L5.Tether_USD_Ret 0.0084
(0.64)
L.BTC_USD_Ret −0.1066*** −0.1093*** −0.1126*** −0.1133*** −0.1134***
(−6.72) (−6.91) (−7.18) (−7.24) (−7.27)
L2.BTC_USD_Ret −0.0775*** −0.0808*** −0.0825*** −0.0829***
(−4.97) (−5.20) (−5.30) (−5.33)
L3.BTC_USD_Ret −0.0734*** −0.0750*** −0.0761***
(−4.76) (−4.85) (−4.92)
L4.BTC_USD_Ret −0.0450** −0.0460**
(−3.09) (−3.17)
L5.BTC_USD_Ret −0.0280
(−1.91)
Constant 0.0154 0.0158 0.0160 0.0162 0.0163
(1.43) (1.47) (1.49) (1.51) (1.52)
Observations 8,750 8,749 8,748 8,747 8,746
Adjusted R2 0.011 0.017 0.022 0.024 0.025
Panel C: Price Differences between USD and Tether Exchanges
(1) (2)
Tether BTC
Arbitrage Spread 0.0032 0.0163

(0.22) (1.08)
Average Return −0.0823*** −0.1372***
(−5.77) (−8.37)
Constant −0.0000 0.0001
(−0.00) (0.01)
Observations 9,501 9,501
Adjusted R2 0.007 0.020
on Poloniex adjust with a delay. Traders can respond to the spread by sending
Tether to Poloniex and buying undervalued Bitcoins. This cross-exchange ar-
bitrage also necessitates a flow of Tether back to Bitfinex when Bitfinex prices
are lower than Poloniex prices. However, as Figure 1 shows, this reverse flow
pattern is not commonly observed. On the other hand, the flow of printed Tether
through Bitfinex might also cause prices to inflate first on Bitfinex before the
Tether moves to other exchanges.
Internet Appendix Table IA.XIV shows that for a one-standard-deviation
increase in the return spread measure, the net Tether and Bitcoin flow goes
up from 0.0336 to 0.419 standard deviations, with t-statistics of 2.39 to 3.13.

Consistent with the supply-based hypothesis of flows following returns, a one-
standard-deviation decrease in the average Bitcoin return increases the flow
by 0.043 to 0.12 standard deviations, with t-statistics of 3.15 to 6.68 even after
controlling for the return spread.37 The results show that Bitcoin is typically
at a small premium on Bitfinex before the Tether flows to Bittrex and Poloniex.
This finding could be due to the use of Tether to facilitate arbitrage or to the
supply of Tether inflating prices at Bitfinex first. In either case, the results
show that the pattern of flows following negative Bitcoin returns is the more
economically sizable driver of the flow.
V. Conclusion
Periods of rapid price appreciation are historically associated with innova-
tion and growth but also with nefarious activities that lead to misallocation
of capital. The semitransparent nature of the blockchain provides a unique
opportunity to examine the mechanics behind the growth of an asset class
during a period of massive speculation and understand the role of central mon-
etary entities in a cryptocurrency world. In this paper, we examine whether
the growth of the largest pegged cryptocurrency, Tether, is primarily driven
by investor demand or is supplied to investors as part of a scheme to inflate
cryptocurrency prices.
By mapping the blockchains of Bitcoin and Tether, we are able to establish
that one large player on Bitfinex uses Tether to purchase large amounts of
Bitcoin when prices are falling and following the printing of Tether. Such price
supporting activities are successful as Bitcoin prices rise following the periods
of intervention. Indeed, even 1% of the times with extreme exchange of Tether
for Bitcoin have substantial aggregate price effects. The buying of Bitcoin with
Tether also occurs more aggressively right below salient round-number price
thresholds where the price support might be most effective. Negative EOM
price pressure on Bitcoin in months with large Tether issuance points to a
month-end need for dollar reserves for Tether, consistent with partial reserve
backing. Our results are most consistent with the supply-driven hypothesis.
Overall, our findings provide support for the view that price manipula-
tion can have substantial distortive effects in cryptocurrencies. Prices in this
market reflect much more than standard supply/demand and fundamental
news. These distortive effects, when unwound, could have a considerable nega-
tive impact on cryptocurrency prices. More broadly, these findings also suggest
that innovative technologies designed to bypass traditional banking systems
have not eliminated the need for external surveillance, monitoring, and a reg-
ulatory framework as many in the cryptocurrency space had believed. Our
findings support the historical view that dubious activities are associated with
bubbles and can contribute to further price distortions.
37 We find similar results when decomposing the flows into those to 1LSg, other Poloniex and
Bittrex, and other Tether-based exchange (Internet Appendix Table IA.XIV).
Initial submission: July 6, 2018; Accepted: October 5, 2019

Editors: Stefan Nagel, Philip Bond, Amit Seru, and Wei Xiong
REFERENCES
Akerlof, George A., Paul M. Romer, Robert E. Hall, and N. Gregory Mankiw, 1993, Looting: The
economic underworld of bankruptcy for profit, Brookings Papers on Economic Activity 1993,
1–73.
Androulaki, Elli, Ghassan O. Karame, Marc Roeschlin, Tobias Scherer, and Srdjan Capkun, 2013,
Evaluating user privacy in Bitcoin, International Conference on Financial Cryptography and
Data Security, Springer, Berlin.
Bhattacharya, Utpal, Craig W. Holden, and Stacey Jacobsen, 2012, Penny wise, dollar foolish:
Buy–sell imbalances on and around round numbers, Management Science 58, 413–431.
Brunnermeier, Markus K., and Stefan Nagel, 2004, Hedge funds and the technology bubble, Jour-
nal of Finance 59, 2013–2040.
Christie, William G., and Paul H. Schultz, 1994, Why do NASDAQ market makers avoid odd-eighth
quotes? Journal of Finance 49, 1813–1840.
Cong, Lin William, Zhiguo He, and Jiasun Li, 2019, Decentralized mining in centralized pools,
Working paper, National Bureau of Economic Research.
Cong, Lin William, Ye Li, and Neng Wang, 2019, Tokenomics: Dynamic adoption and valuation,
Working paper, Becker Friedman Institute for Research in Economics.
Dale, Richard, 2004, The First Crash: Lessons from the South Sea Bubble (Princeton University
Press, Princeton, NJ).
Denina, Clara, and Jan Harvey, 2014, E-trading pulls gold into forex units as commodity desks
shrink, Reuters, April 22.
Du, Wenxin, Alexander Tepper, and Adrien Verdelhan, 2018, Deviations from covered interest rate
parity, Journal of Finance 73, 915–957.
Duffie, Darrell, 2010, Presidential address: Asset price dynamics with slow-moving capital, Journal
of Finance 65, 1237–1267.
Foley, Sean, Jonathan R. Karlsen, and Tālis J. Putniņš, 2019, Sex, drugs, and Bitcoin: How much
illegal activity is financed through cryptocurrencies? Review of Financial Studies 32, 1798–
1853.
Froot, Kenneth A., Paul G. J. O’connell, and Mark S. Seasholes, 2001, The portfolio flows of
international investors, Journal of Financial Economics 59, 151–193.
Gandal, Neil, J. T. Hamrick, Tyler Moore, and Tali Oberman, 2018, Price manipulation in the
Bitcoin ecosystem, Journal of Monetary Economics 95, 86–96.
Greenwood, Robin, 2005, Short-and long-term demand curves for stocks: Theory and evidence on
the dynamics of arbitrage, Journal of Financial Economics 75, 607–649.
Griffin, John M., Jeffrey H. Harris, Tao Shu, and Selim Topaloglu, 2011, Who drove and burst the
tech bubble? Journal of Finance 66, 1251–1290.
Griffin, John M., Jeffrey H. Harris, and Selim Topaloglu, 2007, Why are IPO investors net buyers
through lead underwriters? Journal of Financial Economics 85, 518–551.
Griffin, John M., Federico Nardari, and René M. Stulz, 2007, Do investors trade more when stocks
have performed well? Evidence from 46 countries, Review of Financial Studies 20, 905–951.
Griffin, John M., and Amin Shams, 2018, Manipulation in the VIX? Review of Financial Studies
31, 1377–1417.
Harris, Lawrence, and Eitan Gurel, 1986, Price and volume effects associated with changes in
the S&P 500 list: New evidence for the existence of price pressures, Journal of Finance 41,
815–829.
He, Zhiguo, and Arvind Krishnamurthy, 2018, Intermediary asset pricing and the financial crisis,
Annual Review of Financial Economics 10, 173–197.
Higgins, Stan, 2018, Bitfinex’s biggest critic is back on twitter, CoinDesk, February 8.
Hutcheson, Archibald, 1720, Some Seasonable Considerations for Those Who are Desirous, by
Subscription, or Purchase, to Become Proprietors of South-Sea Stock (F. Morphew, London).
Kaminska, Isabella, 2017, Tether’s “transparency update” is out, Alphaville, October 2.
Kindleberger, Charles P., and Robert Z. Aliber, 2011, Manias, Panics and Crashes: A History of
Financial Crises (Palgrave Macmillan, London).
Kumar, Praveen, and Duane J. Seppi, 1992, Futures manipulation with “cash settlement,” Journal
of Finance 47, 1485–1502.
Lee, T.B., 2014, These four charts suggest that Bitcoin will stabilize in the future, Washington
Post, February 3.
Lehmann, Bruce N., 1990, Fads, martingales, and market efficiency, Quarterly Journal of Eco-
nomics 105, 1–28.
Leising, Matthew, 2017, There’s an $814 million mystery near the heart of the biggest Bitcoin
exchange, Bloomberg, December 5.
Lin, Hsiou-wei, and Maureen F. McNichols, 1998, Underwriting relationships, analysts’ earnings
forecasts and investment recommendations, Journal of Accounting and Economics 25, 101–
127.
Liu, Yukun, and Aleh Tsyvinski, 2018, Risks and returns of cryptocurrency, Technical report,
National Bureau of Economic Research.
McLannahan, Ben, 2015, Bitcoin hack report suggests inside job, Financial Times, February 19.
Malkiel, Burton G., 1981, Risk and return: A new look, Technical report, National Bureau of
Economic Research.
Meiklejohn, Sarah, Marjori Pomarole, Grant Jordan, Kirill Levchenko, Damon McCoy, Geoffrey M.
Voelker, and Stefan Savage, 2013, A fistful of Bitcoins: Characterizing payments among men
with no names, Proceedings of the 2013 Conference on Internet Measurement Conference,
Internet Measurement Conference, 127–140.
Mollenkamp, Carrick, and Mark Whitehouse, 2008, Study casts doubt on key rate, Wall Street
Journal, May 29.
Nagel, Stefan, 2012, Evaporating liquidity, Review of Financial Studies 25, 2005–2039.
Nakamoto, Satoshi, 2008, Bitcoin: A Peer-to-Peer Electronic Cash System (Unpublished
Manuscript).
Nilsson, Kim, 2015, The missing Mt. Gox Bitcoins, WizSec, April 19.
Partnoy, Frank, 2009, FIASCO: Blood in the Water on Wall Street (WW Norton & Company, New
York, NY).
Popper, Nathaniel, 2017, Warning signs about another giant Bitcoin exchange, New York Times,
November 21.
Povel, Paul, Rajdeep Singh, and Andrew Winton, 2007, Booms, busts, and fraud, Review of Finan-
cial Studies 20, 1219–1254.
Reid, Fergal, and Martin Harrigan, 2013, An analysis of anonymity in the Bitcoin system, in Yaniv
Altshuler, Yuval Elovici, Armin B. Cremers, Nadav Aharony, and Alex Pentland, eds., Security
and Privacy in Social Networks (Springer, New York, NY).
Robb, George, 2002, White-Collar Crime in Modern England: Financial Fraud and Business Moral-
ity, 1845-1929 (Cambridge University Press, Cambridge).
Ron, Dorit, and Adi Shamir, 2013, Quantitative analysis of the full Bitcoin transaction graph,
International Conference on Financial Cryptography and Data Security, Springer, Berlin.
Scheinkman, José A., 2013, Speculation, trading and bubbles: Third annual arrow lecture, Working
paper 1458, Princeton University.
Scheinkman, Jose A., and Wei Xiong, 2003, Overconfidence and speculative bubbles, Journal of
Political Economy 111, 1183–1220.
Shiller, Robert J., 2000, Irrational Exuberance (Princeton University Press, Princeton, NJ).
Shleifer, Andrei, 1986, Do demand curves for stocks slope down? Journal of Finance 41, 579–590.
Sockin, Michael, and Wei Xiong, 2018, A model of cryptocurrencies, Working paper, Princeton
University.
Spatt, Chester, 2014, Security market manipulation, Annual Review of Financial Economics 6,
405–418.
Temin, Peter, and Hans-Joachim Voth, 2013, Prometheus Shackled: Goldsmith Banks and Eng-
land’s Financial Revolution after 1700 (Oxford University Press, Oxford).
Tversky, Amos, and Daniel Kahneman, 1974, Judgment under uncertainty: Heuristics and biases,
Science 185, 1124–1131.
Vaughan, Liam, and Gavin Finch, 2013, Currency spikes at 4 pm in London provide rigging clues,
Bloomberg Business, August 27.
Chun Wei, Wang, 2018, The impact of Tether grants on Bitcoin, Economics Letters 171, 19–22.
Yermack, David, 2017, Corporate governance and blockchains, Review of Finance 21, 7–31.
Supporting Information
Additional Supporting Information may be found in the online version of this
article at the publisher’s website:
Appendix S1: Internet Appendix.
Replication code.
Disclosure statement

Bitcoin - Tether

Uploaded by

Copyright:

Available Formats

Bitcoin - Tether

Uploaded by

Document Information

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

Bitcoin - Tether

Uploaded by

Copyright:

Available Formats

THE JOURNAL OF FINANCE • VOL. LXXV, NO.

Is Bitcoin Really Untethered?

INNOVATION, EXCESSIVE SPECULATION, AND DUBIOUS behavior are often closely

American Finance Association

obtained from simply selling an asset to another speculator.1 Perhaps because

reserves and allows for dollar-like transactions without a banking connection,

5 For example, see posts by Bitfinex’ed account at https://medium.com/@bitfinexed and Popper

Poloniex, and Bittrex) also facilitate considerable cross-exchange Bitcoin flows

I. Overview of Bubbles, Bitcoin, Tether, and Hypotheses

(Lin and McNichols (1998)), pushing or “laddering” prices through implicit

B. Brief History of Bitcoin and Exchange “Hacks”

Bitcoin can be found at http://historyofbitcoin.org and in Lee (2014).

C. Brief History of Tether

of perceived reputation tainting or because of the need to comply with anti-

H1A: Tether’s price relative to the USD may increase as a consequence of

The main alternative hypothesis is that Tether is printed independent of

II. Data, Algorithms, and Flows between Major Accounts

Panel A: Total Volume ($B)

BCC 0.81 0.01 – 1.68 – – – – – –

One-HourInterval Three-Hour Interval Five-Hour Interval

Coin Coefficient t-Stat Coefficient t-Stat Coefficient t-Stat

BCC −0.127 −3.960 −0.166 −6.412 −0.260 −6.800

−0.081 −3.657 −0.064 −2.341 −0.069 −2.263

B. Analyzing Bitcoin Blockchain

C. Analyzing Tether Blockchain

D. Bitcoin and Tether Net Flows

where BT Ci→ j is the amount of coins transferred from group of wallets i to

where Tetheri→ j is the amount of coins transferred from exchange i to exchange

Tether/BitcoinFlow = (NetTetherFlowt + NetBTCFlowt )/2. (3)

After printing, Tether is used to purchase Bitcoin primarily on Poloniex

E. Detailed Deposit Accounts

III. Are Bitcoin Prices Related to Tether?

A. Examining Flows and Bitcoin Prices

Panel A: Regression of Returns on Lagged Flows

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

Lag PLX BTX Flow 3.855* −0.354 2.694* −1.100 8.134**

Panel B: Regression of Returns on Lagged Decomposed Flows

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

Lag 1LSg Flow 4.240* −0.484 2.379* −1.300 8.206***

Panel B: Regression of Returns on Lagged Decomposed Flows

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

Volatility 121.7 94.23 −51.05 −71.01 −84.21

B. Large Flows and Prices

Panel A: Days Following Authorization

Coin Coefficient t-Stat N

DASH 6.16 3.26 2,645

Panel B: Other Days

Coin Coefficient t-Stat N

DASH 0.59 0.61 6,833

Panel C: Following Negative Returns

Coin Coefficient t-Stat N

DASH 2.92 1.69 3,992

Panel D: Following Positive Returns

Coin Coefficient t-Stat N

DASH 3.47 2.26 3,985

Panel E: Negative Returns-Authorization

Coin Coefficient t-Stat N

DASH 10.19 3.26 1,063

C. Is the Price Effect Economically Important?

D. Negative Serial Correlation in Bitcoin Prices

EOM −0.0230 −0.000788 −0.0377* −0.00669 −0.000788 −0.0251*** −0.00869 −0.000788

Constant 0.0110* 0.00501 0.0150* 0.0101* 0.00501 0.0160* 0.00824** 0.00501

EOM −0.0216 0.00107 −0.0367* −0.00188 0.00107 −0.0241*** −0.00301 0.00107