Liu - Accounting For Crypto Value

Accounting for Cryptocurrency Value
Yukun Liu, Aleh Tsyvinski, Xi Wu∗
October 2021
Abstract
Information pertaining to new addresses is highly value-relevant in the cryptocurrency mar-

ket. Innovations to the number of new addresses explain eight percent of the variations in
cryptocurrency returns. Unlike traditional markets, we do not find evidence of pre- or post-
drift around the release of new address information. The presence of strong market reactions
at the release of new address information and the absence of drifts around the release highlight
the distinct features of cryptocurrencies and provide a benchmark case where information is
released publicly and continuously. Lastly, we show that the price-to-new address ratios nega-
tively predict future cryptocurrency returns.
∗
Yukun Liu is with Simon Bussiness School of the University of Rochester. Aleh Tsyvinski is with Yale University
and the New Economic School. Xi Wu is with Haas School of Business, the University of California, Berkeley.
Electronic copy available at: https://ssrn.com/abstract=3951514

1 Introduction
Unlike companies that have the Generally Accepted Accounting Principles (GAAP) standards
for financial measurements, cryptocurrencies do not have a standardized accounting framework in
generating measurements for performance. The lack of such fundamental accounting information
poses challenges for investors and regulators to determine the intrinsic value of cryptocurrencies.
At the same time, economic activities in blockchain are publicly and continuously available, thus
providing a rich set of ongoing information that may be used to determine the value of cryptocur-
rency. We apply accounting and finance valuation frameworks to the cryptocurrency market and,
in turn, revisit several classic questions for which the amount and revelation of information are
central.
Cryptocurrency is the first asset class that provides a complete history of investment activi-
ties recorded on its blockchain. Blockchain data has at least three major advantages. First, the
data is widely and easily available compared to trading or pricing information that is often propri-
etary. Second, the data is verifiable, as anyone can access the public ledger to confirm the data.
Third, information is available in real-time, and thus the data is timely for users. These properties
of blockchain data echo with the conceptual framework for firms’ financial reporting as set by
the Financial Accounting Standards Board (FASB) (Financial Accounting Standards Board 2010).
FASB defines the fundamental qualitative characteristics of useful financial information to be as
follows: “if financial information is to be useful, it must be relevant and faithfully represent what
it purports to represent. The usefulness of financial information is enhanced if it is comparable,
verifiable, timely, and understandable.” Relevant information refers to information that can influ-
ence a decision and that has predictive and/or feedback value. Faithful representation, also known
as reliability, requires that the information be complete (no transactions excluded), neutral (free
from bias), and free from error. The framework is conceptual rather than exact for company fi-
nancial statements because it can never be exactly achieved in reality.1 For the data recorded on a
blockchain, however, many of these conceptual criteria are exactly achievable.
The theoretical work in network economy and the recent models of dynamic cryptocurrency
pricing focus on the role of user adoption and network effect. That is, the value of cryptocurrencies
increases when more users join. Motivated by this literature, we propose and comprehensively an-
1 For example, there are subjectivity and estimation involved in preparing financial statements (e.g., estimate un-
collectible), and thus the information cannot be truly neutral or free from bias.

alyze a fundamental measure using on-chain data to value cryptocurrencies – the number of new
addresses. Empirically, using the value-relevance tests developed in the accounting and finance
literature to establish the fundamental role of earnings in the equity market, we find that informa-
tion pertaining to new addresses is strongly value-relevant for cryptocurrencies. In fact, the value
relevance of new addresses to the cryptocurrency market is even higher than that of earnings to the
equity market. Furthermore, we show that the price-to-new address ratio negatively predicts future
cryptocurrency returns, analogous to the price-to-earnings ratio in the equity market.
We start by testing the usefulness and relevance of information regarding new addresses for
the cryptocurrency market. We apply the value-relevance method and examine the cryptocurrency
market reactions to the innovations in the number of new addresses. We find that information
pertaining to new addresses is highly value-relevant both in the cross-section and in the time-
series. In the cross-section, innovations to the number of new addresses can account for more than
eight percent of the variations of cryptocurrency returns. The value relevance of new addresses
in the cryptocurrency market is higher than that of earnings in the equity market, which is found
to be able to explain about five percent of the variations of stock returns (e.g., Lev 1989; Lev
and Gu 2016). Moreover, we show that many economic determinants (see, e.g., Kothari 2001,
Schipper and Vincent 2003, and Chapter 5.4 of Scott and O’Brien 2019), such as quality, discount
rate, and size, of the value-relevance of earnings in the equity market have their counterparts in
the cryptocurrency market, and that these counterparts are also central in determining the value-
relevance of new addresses in the cryptocurrency market.
One important difference between cryptocurrencies and equities is that, for cryptocurrencies,
information of the underlying economic activities is available publicly and continuously on the
blockchain. Whereas for equities, public information of firm activities is only available periodi-
cally. We draw on this distinction and revisit several classic questions in the equity market.
First, we test potential slow reactions of the cryptocurrency market to the information pertain-
ing to new addresses. In the equity market, it is found that informational events, such as earnings
announcements, tend to forecast future returns, a phenomenon known as post-announcement drifts
(e.g., Ball and Brown 1968; Bernard and Thomas 1989; Sadka 2006; Daniel, Hirshleifer, and Sun
2020). Among many alternative explanations for this phenomenon, a common view is that the eq-
uity market is slow in incorporating earnings information. We find that, unlike the equity market,
innovations to the number of new addresses do not forecast cryptocurrency returns in the future,
consistent with the view that transparent and timely information revelation facilitates information

acquisition.
Second, we test the existence of return drifts prior to the release of new address information. In
the equity market, it is found that stock returns tend to precede important informational events such
as earnings and Federal Open Market Committee (FOMC) announcements, a phenomenon known
as pre-announcement drifts (e.g., Ball and Brown 1968; Lucca and Moench 2015; Cieslak, Morse,
and Vissing-Jorgensen 2019). The literature commonly attributes the phenomenon to the discrete-
ness of information releases and information leakages between the discrete announcements. In
contrast to the equity market, we find that cryptocurrency returns do not precede releases of new
addresses information in the pre-information period, consistent with the view that more frequent
new public information reduces potential information leakages.
Next, we use the number of new addresses to construct a financial ratio for cryptocurrencies in-
spired by the equity research: the price-to-new address ratio. We find that the price-to-new address
ratio captures the cross-section of expected returns, which is analogous to the return predictability
result of the price-to-earnings ratio in the equity market (e.g., Basu 1977; Campbell and Shiller
1988a; Welch and Goyal 2008). High price-to-new address ratios are followed by low future re-
turns in the cross-section of returns of cryptocurrencies. A zero-investment long-short strategy that
buys the portfolio of coins with the lowest price-to-new address ratio and shorts the portfolio of
coins with the highest price-to-new address ratio generates an average weekly excess return of 1.9
percent. We show that the return predictability of the price-to-new address ratio is distinct from
those documented in the literature. Controlling for the cryptocurrency 3-factor model (see Liu,
Tsyvinski, and Wu 2021), the alpha of the long-short strategy remains positive and statistically
significant at the 1 percent level. Moreover, the return predictability of the price-to-new address
ratio is long-lasting. The long-short strategy based on the price-to-new address ratio remains sig-
nificant at the 10 percent level 20 weeks after the portfolio formation period.
We discuss the relationship of our paper to the literature. The importance of user adoption is
a foundational concept in the literature of network economy (e.g., Rohlfs 1974; Katz and Shapiro
1985; Katz and Shapiro 1994; Rochet and Tirole 2003). Recent theoretical development of cryp-
tocurrency also emphasizes the role of user adoption and network effect in cryptocurrency valua-
tion. Cong, Li, and Wang (2021) is a model about dynamic user adoption, and their main finding
is on the endogenous adoption of cryptocurrency and price predictions at the different stages of
the life cycles of cryptocurrencies. Sockin and Xiong (2020) show that network externalities in the
cryptocurrency market can lead to multiple equilibria. Other studies focusing on user adoption of

cryptocurrencies include Pagnotta and Buraschi (2018), Biais, Bisiere, Bouvard, Casamatta, and
Menkveld (2020), and Cong, Li, and Wang (Forthcoming).
The vast literature studying market reactions to informational events can be traced back at least
to Beaver (1968), Ball and Brown (1968), and Fama, Fisher, Jensen, and Roll (1969). Lev (1989),
Kothari (2001), and Dechow, Ge, and Schrand (2010) survey the literature that examines the value-
relevance of earnings in the equity market. It is shown that equity returns can both lead information
events such as earnings and macroeconomic announcements, known as pre-announcement drifts
(e.g., Ball and Brown 1968; Lucca and Moench 2015; Cieslak, Morse, and Vissing-Jorgensen
2019), and lag informational events, known as post-announcement drifts (e.g., Ball and Brown
1968; Bernard and Thomas 1989; Sadka 2006; Daniel, Hirshleifer, and Sun 2020). Using the
methods in the value-relevance literature which establishes the fundamental role of earnings in the
valuation of equities, we find that information pertaining to new addresses is highly value-relevant
in the cryptocurrency market. Moreover, we show that, contrary to the equity market, returns of
cryptocurrencies do not lead or lag the release of new address information. The findings highlight
the distinct feature of cryptocurrencies and provide a benchmark case where information is released
both publicly and continuously.
Several recent papers document empirical regularities related to cryptocurrency investment.
Liu and Tsyvinski (2021) and Liu, Tsyvinski, and Wu (2021) comprehensively study the valua-
tions of the cryptocurrency market in the aggregate time-series and the cross-section, respectively.
Makarov and Schoar (2020), Borri and Shakhnov (2018), and Borri and Shakhnov (2019) find that
there are dispersions of Bitcoin prices across different exchange platforms. Hu, Parlour, and Rajan
(2019) find that cryptocurrency returns tend to be positively related to Bitcoin returns. Borri (2019)
shows that cryptocurrency returns are exposed to the tail-risks within cryptocurrencies but not tail-
risks from global assets. Bhambhwani, Delikouras, and Korniotis (2020) show that cryptocurrency
exposures to hash rate and network size growths predict cryptocurrency future returns. Cong,
Karolyi, Tang, and Zhao (2021) provide another angle on studying the valuation of cryptocurren-
cies and on the value premium by considering a broad spectrum of variables that may capture
value. Cong, He, and Tang (2021) focus on the cryptocurrency staking economy and examine a
carry trade strategy. Liu, Sheng, and Wang (2019) construct a technology index from the whitepa-
pers of initial coin offerings. Shams (2020) and Schwenkler and Zheng (2020) study correlations
of cryptocurrency pairs. Soska, Dong, Khodaverdian, Zetlin-Jones, Routledge, and Christin (2021)
study cryptocurrency derivatives. Griffin and Shams (2020), Cong, Li, Tang, and Yang (2021), and

Aloosh and Li (2019) provide evidence of potential manipulations in the cryptocurrency market.
2 Value Relevance and Cryptocurrencies

In this section, we build on a large body of literature on network economy (e.g., Rohlfs 1974;
Katz and Shapiro 1985; Katz and Shapiro 1986; Farrell and Saloner 1986; Katz and Shapiro 1994)
to discuss its relevance for valuing cryptocurrencies. The basic idea is that the value of a cryptocur-
rency increases when more users join its blockchain, which is commonly referred to as the network
effect. In a network economy, the utility a user derives from a service depends on the overall user
adoption of the economy, and therefore the valuation of the network increases when more users
join. Markets with network effects such as software, media, payment systems, and Internet are
typically two-sided, and thus many recent papers on network economy focus on two-sided, or even
multi-sided, markets. Rochet and Tirole (2003) and Armstrong (2006) study platform competition
in the two-sided markets, and find that the value of platforms is related to their user base. Mitchell
and Skrzypacz (2006) build a dynamic competition model of duopoly platforms featuring network
effects, and find a positive relationship between the value of platforms and their current and fu-
ture user base in their dynamic model. Weyl (2010) and Hagiu and Wright (2015) develop pricing
models of multi-sided platforms with network effects. Liebowitz and Margolis (1994), Shy (2001),
Rysman (2009), and Shy (2011) discuss and survey the literature of network economy.
The cryptocurrency market can be considered a network economy.2 Accordingly, the theoreti-
cal literature on cryptocurrency valuations also focuses on user adoption and network effect. Cong,
Li, and Wang (2021) relate cryptocurrency prices to the user base. In a model of dynamic user
adoption, they show that cryptocurrency prices are intimately linked to the endogenous adoption
of cryptocurrencies. Sockin and Xiong (2020) feature the network effect in their model and con-
sider the possibility of multiple equilibria. Other studies that relate the network effect to cryptocur-
rencies include Pagnotta and Buraschi (2018), Biais, Bisiere, Bouvard, Casamatta, and Menkveld
(2020), Pagnotta (2020), and Cong, Li, and Wang (Forthcoming).
The early empirical literature on network effects studies the traditional network economy, such
as software, ATM, yellow pages, and payment systems (e.g., Gandal 1994; Saloner and Shepard
2 In principle, all cryptocurrencies have the functionality of payment systems even though a significant proportion
of them is used primarily for non-payment purposes. Many of these cryptocurrencies also have platforms that serve
specific purposes, such as storage, financing, and gaming.

1995; Rysman 2004; Rysman 2007). In particular, the value of Internet companies which are
considered a classic example of network industry, is found to be strongly increasing when more
users join. Trueman, Wong, and Zhang (2000), Rajgopal, Venkatachalam, and Kotha (2003), and
Luo, Zhang, and Duan (2013) find that the value of internet companies has a strong and positive
relationship with different measures of their user bases. Importantly, they show that traditional
accounting numbers such as earnings are not related to the value of these internet companies. In
other words, the user base information is highly value-relevant for the internet companies, while
traditional accounting information is not.3 Goldfarb and Tucker (2019) discuss the potential drivers
of network effects in digital economies. Levin (2011) survey the literature regarding the economics
of Internet markets.
Prior studies of both network economy and cryptocurrencies place a fundamental role on the
network effect induced by user adoption. The utility a cryptocurrency user derives from using the
coin depends on the overall level of user adoption of the system, and therefore the valuation of
the cryptocurrency would increase when more new users join. Therefore, we seek to empirically
test the usefulness and relevance of the information regarding new users for the cryptocurrency
market. To do so, we employ the set of methods that is foundational in the accounting literature
– the value-relevance tests. The value-relevance tests were pivotal in empirically establishing the
fundamental role of earnings as the measure of firm fundamentals. The objective of the value-
relevance tests is to test whether and how quickly given measures, such as firm earnings, capture
changes in the information set that is reflected in security returns over a given period. Such tests are
the cornerstones of the vast literature that started with the classic work of Ball and Brown (1968).
For related surveys of this vast literature, see Lev (1989), Fama (1998), Kothari (2001), and Scott
and O’Brien (2019).
The approach in this literature is to test how investors react to releases of information, such as
earnings. In other words, if the market reacts to the release of a piece of information, then this
specific piece of information can be considered useful. A common method to test the information
content of accounting earnings is to regress contemporaneous stock returns on innovations in earn-
ings. The slope coefficients from these regression specifications are one of the central objects in
the accounting literature – earnings response coefficients, or ERCs (see, e.g., Chapter 5.4 of Scott
and O’Brien 2019 and Exhibit 1 of the survey of Dechow, Ge, and Schrand 2010). Moreover, the
3 Moreover,
this literature links the user base of companies to their revenue and profitability (e.g., Trueman, Wong,
and Zhang 2001; Trueman, Wong, and Zhang 2003).

explanatory power, or R-squared, of the regression measures the degree of informativeness of the
information.4
Besides earnings, the accounting and finance literature also examines the value relevance of
other financial statement information as well as of non-accounting information. This so-called
fundamental information analysis is another foundational concept in accounting (see recent surveys
of Richardson, Tuna, and Wysocki 2010 and Lewellen 2010 for discussions). Early studies of the
value relevance of other financial statement information include Lev and Thiagarajan (1993), Lev
and Sougiannis (1996), and Chan, Lakonishok, and Sougiannis (2001). Several recent papers (e.g.,
De Franco, Wong, and Zhou 2011; Patatoukas, Sloan, and Zha 2015) study the value relevance of
supplementary information in financial statements. Pakes (1985), Austin (1993), and Hall, Jaffe,
and Trajtenberg (2005) study the relation between patents and stock returns. Kogan, Papanikolaou,
Seru, and Stoffman (2017) study stock market reactions to different patent announcements. Amir
and Lev (1996) study the value-relevance of potential user adoption in the wireless communication
industry. Ittner and Larcker (1998) examine the value-relevance of customer satisfaction.
A related line of research that evaluates the usefulness of informational events using market
reactions is the event study literature. Fama, Fisher, Jensen, and Roll (1969) examine the stock
price reactions to a number of informational events, such as stock splits and dividend adjustments.
The literature also examines other informational events of firms, such as merger & acquisition (e.g.,
Roll 1986; Agrawal and Jaffe 2000; DePamphilis 2019) and IPO (e.g., Ritter and Welch 2002;
Loughran and Ritter 2004). The approach is also commonly applied to studying macroeconomic
events, such as the announcement of unemployment rates (e.g., Boyd, Hu, and Jagannathan 2005)
and federal funds rate (e.g., Bernanke and Kuttner 2005).
3 Data Summary
In this section, we discuss the data used in this paper. We obtain on-chain and price data of
cryptocurrencies from Glassnode.com, a major blockchain data and intelligence provider.5 For
4 Another method to test the usefulness of earnings is to examine the reactions of trading volume or return volatility
to earnings news, which is started from Beaver (1968). We also use this approach to empirically test the usefulness
of earnings news in Table A.1 in the Online Appendix, and reach similar conclusions. In this paper, we focus on the
approach using market price reaction because it is easy to interpret and is believed to provide a stronger test for the
decision usefulness than volume or return volatility reaction (see Kim and Verrecchia 1991).
5 Other popular cryptocurrency on-chain data providers include Coinmetrics and IntoTheBlock.

each coin, Glassnode gathers on-chain data that are recorded in coins’ public ledger, which is
its record-keeping system of the identities, cryptocurrency balances, and all network transactions
of the participants. The data on the public ledgers of cryptocurrencies form the basis of the on-
chain data analysis. Because each cryptocurrency has both common and unique features on its
blockchain, the number of available on-chain measures varies across the coins. Our main measures
are based on the common features of blockchains and thus are available for the maximum amount
of coins. The comprehensive sets of the on-chain data are typically available on a daily basis.
For our main empirical analyses, we construct the on-chain measures at the weekly frequency.
In additional tests, the measures are constructed at different frequencies. Our main variables are
discussed below.
To measure the information pertaining to new users of a blockchain, we use the number of new
addresses per coin (N A). An address is a unique sequence of numbers and letters that represents
a source or a destination of blockchain payments. A new address is a unique address that appears
for the first time in a transaction in a week. A high N A for a coin indicates fast user adoption
of this cryptocurrency. We then construct the weekly changes in logged N A, or ∆na, to measure
innovations to the number of new addresses.
Next, we construct the price-to-new address ratio, or pa, defined as pa = log( PriceN A ), where
Price is the end of day price on the last day of a week. In the following section, we empirically
show a strong value-relevance of N A in the cryptocurrency market. Moreover, we find that N A has
many properties that are similar to earnings in the value-relevance studies. Therefore, pa plays a
role similar to the logged price-to-earnings ratio of equities. The definitions of other variables are
documented in the Online Appendix.
Our sample starts from February 2018, when the on-chain data became available for most
coins, to June 2021. We exclude stablecoins and require the coins to have price data. The final
sample consists of 127 cryptocurrencies. Panel A of Table 1 shows the summary statistics of the
main variables used in the paper. During the sample period, the average logged coin return is 0.3
percent per week and the weekly standard deviation of the logged coin returns is 0.251. The mean
(median) of average growth in new addresses (∆na) is -1.9 percent (-3.2 percent), suggesting that
the increase of the number of new users slightly slows down on average in our sample period. The
summary statistics of the other variables are also included in Panel A. Panel B shows the correlation
matrix of the variables. It shows that cryptocurrency returns are most correlated with the changes in
the logged number of new addresses (∆na), with a correlation of 0.252. The correlations between

cryptocurrency returns and the other variables are much lower. Table A.9 in the Online Appendix
shows the summary statistics of the other variables used in the paper.
Table 1: Summary Statistics
This table shows the summary statistics of the main variables used in the paper. Panel A reports
the mean, standard deviation, 10% percentile, median and 90% percentile values of the variables.
Panel B reports the correlation matrix of the variables.
Panel A Mean SD 10% Median 90%
ret 0.003 0.251 -0.214 -0.000 0.210

∆na -0.019 0.582 -0.551 -0.032 0.542
na -13.095 12.362 -17.009 -13.449 -8.787
pa 12.254 1.344 10.844 12.114 13.902
Panel B ret ∆na na pa
ret 1.000
∆na 0.252 1.000
na 0.039 0.083 1.000
pa 0.004 -0.206 -0.167 1.000
4 Value Relevance
Building on the discussion in Section 2, we aim to test the value relevance of information
pertaining to new addresses. In other words, we examine if the cryptocurrency returns react to the
innovations to the number of new addresses. In our baseline specification, we use the changes in
the logged number of new addresses to capture innovations to new addresses, which is analogous
to using earnings growth to measure earnings innovations. In summary, we test the value relevance
of new addresses in the following baseline form:
rt,i = α + β × ∆nat,i + t,i (1)
where ri,t is the logged returns of cryptocurrency i at time t, ∆nat,i is the changes in the logged
number of new addresses of cryptocurrency i at time t, and t,i is the residual term. The coefficient

estimate of interest is β, which measures the cryptocurrency price reactions to the changes in the
logged number of new addresses.
Examples
We first examine the time-series properties of the value-relevance of new addresses for three
major cryptocurrencies – Bitcoin, Ethereum, and Litecoin. Figure A.1 in the Online Appendix
gives a graphical illustration of the relation between prices and the number of new addresses of
each coin. The graph shows that the logged prices track well the logged numbers of new addresses
for all three cryptocurrencies.
Next, we test the relationship statistically and document the result in Table 2. For each cryp-
tocurrency, we apply two regression specifications, where the first specification only includes
changes to logged number of new addresses (∆na) and the second also includes logged new ad-
dresses per coin (na) to capture potential level effects (e.g., Collins and Kothari 1989; Lev and Gu
2016). The coefficient estimates of ∆na are positive and highly statistically significant for all three
cryptocurrencies under the two different specifications. A one-standard-deviation increase of ∆na
is associated with increases in weekly logged returns of 11.7, 10.1, and 13.1 percent for Bitcoin,
Ethereum, and Litecoin, respectively.6 The adjusted R-squareds in the specifications including
only ∆na are 4.4 percent, 4.2 percent, and 9.5 percent for Bitcoin, Ethereum, and Litecoin, re-
spectively. To make a comparison, in the equity market, Lev (1989) documents that the R-squared
associated with the returns/earnings relation is around 5 percent. More recently, Lev and Gu (2016)
survey the literature on value relevance and provide evidence that the value relevance of earnings
is decreasing over time as measured by R-squareds. Based on the explanatory power, the value
relevance of new addresses in the cryptocurrency market for the three major cryptocurrencies is
higher than that of earnings in the equity market.
In columns (2), (4) and (6), we also include na and test whether the level of new addresses
provides additional explanatory power. The coefficient estimates of na are significant at the 5-
percent level for Bitcoin but not significant for Ethereum and Litecoin. The increases in adjusted
R-squareds are smaller than 1 percent. That is, the returns react to new information contained
mostly in changes in new addresses instead of the level of new addresses. We systematically test
6 For example, for Bitcoin, 11.7 percent is calculated as the standard deviation of the sample ∆na (0.582) times
the coefficient estimate of ∆na (0.267). Therefore, a one-standard-deviation increase of ∆na is associated with a
0.582 × 0.267 = 0.117, or 11.7 percent, increase in weekly logged returns for Bitcoin.
10

and confirm this phenomenon in the subsequent sections.
Table 2: Value Relevance Analysis – Time-Series
This table shows the results for the time-series value relevance tests for the three major cryp-
tocurrencies (Bitcoin, Ethereum, and Litecoin). Variable definition is documented in the Online
Appendix. *, **, and *** represent significance level at the 10%, 5%, and 1% level.
(1) (2) (3) (4) (5) (6)
BTC ETH LTC
ret ret ret ret ret ret
∆na 0.267*** 0.265*** 0.150*** 0.151*** 0.268*** 0.267***

(5.213) (5.189) (3.756) (3.771) (6.685) (6.606)
na -0.010** -0.003 0.002
(-2.363) (-0.632) (0.303)
Cons 0.021*** -0.009 0.022** 0.001 0.008 0.022
(3.402) (-0.654) (2.240) (0.030) (1.043) (0.467)
Obs 566 566 303 303 419 419

Adj R2 0.044 0.052 0.042 0.040 0.095 0.093
4.1 Baseline Results

In the previous test, we showed a strong returns/new-addresses relation for the three major cryp-
tocurrencies. In particular, the cryptocurrency market responds strongly to information pertaining
to changes to logged number of new addresses. Now, we systematically test the returns/new-
addresses relation using all cryptocurrencies with available data. We run cross-sectional regression
as the specification (1) for each week, and report the average regression coefficients, the associated
Fama and MacBeth (1973) t-statistics, and the average adjusted R-squareds.
The results are documented in Table 3. In column (1) of Table 3, the only independent variable
is the changes in the logged number of new addresses (∆na). The coefficient estimate of ∆na
is positive and highly statistically significant, suggesting that the cryptocurrency market strongly
and positively responds to the information contained in ∆na. A one-standard-deviation increase
in ∆na is associated with an increase of 4.9 percent in logged returns. The economic magnitude
of the baseline cross-sectional regression is smaller than that based on time-series regressions for
11

Bitcoin, Ethereum, and Litecoin, suggesting that the value-relevance of new addresses may be
higher for larger coins, which we test in Section 4.2. ∆na alone already explains 8.5 percent of
the variations in cryptocurrency returns, confirming that information about new addresses is highly
value-relevant in the cryptocurrency market. Column (2) includes na as an additional explanatory
variable to capture the level effects. The coefficient estimate of na is positive and significant at
the 10-percent level. The adjusted R-squared increased slightly to 8.8 percent from 8.5 percent
in the standalone regression. In Table A.10 of the Online Appendix, we also report the results
using other on-chain variables. The coefficient estimates of ∆na are stable and remain highly
statistically significant. Moreover, none of the control variables is statistically significant. The
increases of adjusted R-squareds relative to the baseline specification are small, implying that the
incremental explanatory power of these common on-chain variables is limited. Overall, the results
show that there is a strong returns/new-addresses relation in the cross-section of cryptocurrencies.
The cryptocurrency market responds strongly to information pertaining to changes in the logged
number of new addresses.
Market Expectation
One important aspect of value-relevance studies is the proxy for market expectations (see,
e.g., Lev 1989, Kothari 2001, and Scott and O’Brien 2019). So far, we have tested whether new
addresses are relevant for value. However, in principle, only new information on the market would
be incorporated in prices. That is, a more detailed test of the value relevance would need proxies
for new information contained in the number of new addresses. In this subsection, we test the value
relevance of the news in our new address measure.
In the returns/earnings relation studies in equity market research, innovation in earnings is
measured as the difference between current earnings and a measure of its expected value. Three
methods are commonly employed to estimate the earnings expectations (see, e.g., Chapter 5.2.2 of
Scott and O’Brien 2019 for related discussions). The first method uses the last period earnings as
the earnings expectations for the current period. This approach implicitly imposes an assumption
that earnings are a random walk, and thus earnings growth is the appropriate measure of earnings
innovations. The second method uses statistical models to calculate earnings expectations. The
third approach uses analysts’ forecasts of earnings as the market’s earnings expectations. In the
cryptocurrency market, both the first and second methods can be applied to proxy for market
expectations of new addresses. Because of the lack of analysts’ forecasts of new addresses, the
12

third method is not readily applicable to the cryptocurrency market.
So far, we have used the first method to proxy for the market expectations. In our baseline
specifications in Panel A of Table 3, the main variable of interest is the change in the logged
number of new addresses. Hence, we implicitly assumed that the cryptocurrency market uses the
number of new addresses of the last period as the new address expectations for the current period.
This approach imposes a relatively strong assumption that the number of new addresses is a random
walk or at least the investors believe so. We relax this assumption and use the second method based
on statistical models to calculate market expectations of the number of new addresses.
We use two statistical methods: (1) the cross-sectional and (2) the time-series methods.7 To
calculate the expectations of new addresses using the cross-sectional method, we first regress the
logged number of new addresses of cryptocurrencies at time t − 1 on the corresponding logged
number of new addresses at time t − 2 for the cross-section of cryptocurrencies and keep the coef-
ficient estimates. Then, we apply the coefficient estimates to the logged number of new addresses
at time t − 1 to calculate the predicted logged number of new addresses at time t, which is the ex-
pectations of new addresses based on the cross-sectional model. Therefore, the innovations to the
number of new addresses are the difference between the realized logged number of new addresses
and the predicted logged number of new addresses, which we denote as na shock 1 .
To calculate the expectations of new addresses using the time-series method, we first regress
the logged number of new addresses of cryptocurrencies on the lagged logged number of new
addresses for a cryptocurrency i from the 12 weeks proceeding the current period, and keep the
coefficient estimates.8 Then, we apply the coefficient estimates to the logged number of new ad-
dresses for cryptocurrency i at time t − 1 to calculate the predicted logged number of new addresses
for cryptocurrency i at time t, which is the expectation of new addresses based on the time-series
model. Similarly, the innovations to the number of new addresses are the difference between the re-
alized logged number of new addresses and the predicted logged number of new addresses, which
we denote as na shock 2 .
Panel B and Panel C of Table 4.1 document the results based on na shock 1 and na shock 2 .
We focus our discussion on na shock 1 . The results for na shock 2 are qualitatively similar. In
7 See Lev (1989) and Kothari (2001) for discussions.

8 We use a 12 week window to conduct all rolling estimations in the paper. While there is no consensus on the
estimation window, requiring 12 weeks for rolling estimations in the cryptocurrency market balances precision and
potential time-varying coefficient estimates. Using longer windows (24 weeks or 36 weeks) generates qualitatively
similar results.
13

the standalone model in column (1), the coefficient estimate of na shock 1 is positive and highly
statistically significant. The point estimate of na shock 1 is 0.085, which is close in magnitude
to that of ∆na at 0.084 in Panel A. The adjusted R-squared of the standalone specification is 8.6
percent, which is similar to that based on ∆na at 8.5 percent. Column (2) further includes na
as an explanatory variable. Interestingly, the point estimate of na is no longer significant when
na shock 1 is used instead of ∆na. This result shows that the cryptocurrency market does not
respond to information pertaining to na when na shock 1 is available.
Table 3: Value Relevance Analysis – Cross-Sectional
This table shows the results for the cross-sectional value-relevance tests. The regressions are based
on Fama-MacBeth regression method. Variable definition is documented in the Online Appendix.
*, **, and *** represent significance level at the 10%, 5%, and 1% level.
Panel A Panel B Panel C
(1) (2) (1) (2) (1) (2)
ret ret ret ret ret ret
∆na 0.084*** 0.083***

(12.060) (11.823)
na shock 1 0.085*** 0.087***
(13.121) (12.707)
na shock 2 0.076*** 0.074***
(13.115) (12.905)
na 0.001* -0.001 0.001
(1.929) (-1.335) (1.419)
Cons -0.003 0.009 0.099*** 0.093*** -0.002 0.007
(-0.349) (0.846) (7.165) (6.998) (-0.246) (0.702)
Obs 13,522 13,522 13,404 13,404 12,163 12,163

Adj R2 0.085 0.088 0.086 0.094 0.087 0.090
Adjusted for Expected Returns
In the previous analyses, we used raw returns to measure cryptocurrency market reactions.
However, the raw returns may include a component of expected returns that measure investor re-
action to cryptocurrency market-wide news rather than to coin-specific news. In equity market
14

research, the common method to remove the component of expected returns is to calculate the
residual returns using benchmark factor models.9 The idea is that factor models capture the ex-
pected component of returns, and therefore the residual returns would only contain information
that is orthogonal to the expected returns.
We apply this method to the cryptocurrency market using the cryptocurrency CAPM and cryp-
tocurrency 3-factor model as in Liu, Tsyvinski, and Wu (2021). Liu, Tsyvinski, and Wu (2021)
show that a cryptocurrency three-factor model including the cryptocurrency market factor, cryp-
tocurrency size factor, and cryptocurrency momentum factor is able to capture the cross-section
of expected cryptocurrency returns. Therefore, we use the cryptocurrency factor models to re-
move components that are attributable to the expected returns. We define residual returns as the
difference between raw returns and factor model implied expected returns:
0

resit = Rit − αˆit + β̂ikt Rkt
where Rt is either the cryptocurrency market excess return or the three factors including the cryp-
tocurrency market factor, size factor, and momentum factor; α̂i and β̂ik are estimated from the 12
weeks preceding the current period; i indicates the asset, t indicates time, and k indicates the factor.
The results are documented in Table 4. Panel A shows the results adjusting for the cryptocur-
rency CAPM model. The point estimates of ∆na are positive and highly statistically significant.
The adjusted R-squared for the specification using only ∆na is 7.7 percent, which is similar to
the specifications using raw returns at 8.5 percent. Panel B reports the results adjusting for the
cryptocurrency 3-factor model. The point estimates of ∆na are positive and highly statistically sig-
nificant. The adjusted R-squared for the specification using only ∆na is 5.8 percent. The coefficient
estimates of na are not positive and significant in any of the specifications. That is, the cryptocur-
rency market responds strongly to innovations to new addresses but not to the simple amount of
new addresses. Overall, the results based on unexpected returns further support our baseline re-
sults that the information pertaining to innovations to new addresses is highly value-relevant in the
cryptocurrency market.
9 Removing the component of expected returns is a common practice in the literature to separate market-wide
effect from firm-specific effect and is used by most existing value relevance studies. see Lev (1989), Kothari (2001),
and Chapter 5.2.3 of Scott and O’Brien (2019) for detailed discussions on the merits of using different factor models
to capture components of expected returns.
15

Table 4: Value Relevance Analysis – Adjusting for Expected Returns with Factor Models
This table shows the results for the cross-sectional value-relevance tests using returns adjusted for
expected returns. The regressions are based on Fama-MacBeth regression method. res1 is the
residual return based on the cryptocurrency CAPM model and res2 is the residual return based on
the cryptocurrency 3-factor model. Variable definition is documented in the Online Appendix. *,
**, and *** represent significance level at the 10%, 5%, and 1% level.
Panel A Panel B
(1) (2) (1) (2)
res 1 res1 res 2 res2
∆na 0.087*** 0.086*** 0.084*** 0.083***

(11.976) (12.177) (11.786) (12.031)
na 0.000 -0.001
(0.162) (-0.911)
Cons -0.008 -0.007 -0.003 -0.010
(-1.419) (-0.862) (-0.796) (-1.103)
Obs 12,053 12,053 12,053 12,053

Adj R2 0.077 0.078 0.058 0.057
4.2 Economic Determinants of Value-Relevance

In the equity market, studies have shown that there are several important determinants of the re-
turns/earnings relation (see, e.g., Kothari 2001; Dechow, Ge, and Schrand 2010; Scott and O’Brien
2019). The first main category is the quality of earnings, measured using both the statistical prop-
erty of earnings and the property of the accounting system. The second main determinant is compa-
nies’ discount rates, including both the risk-free component and the risk-based component. Thirdly,
the literature also emphasizes the impact of firm size, a common measure of the informativeness
of stock prices, on the value relevance of earnings. These three categories of determinants in the
returns/earnings relation correspond to the three components of the present value discounting for-
mula – cash-flow (earnings), discount rate, and price. Motivated by the evidence in the equity
market, in this subsection, we study these potential economic determinants of value-relevance in
the returns/new-addresses relation in the cryptocurrency market.
16

Persistence and Variability
Research in equities shows that the quality of earnings can differ across companies (see De-
chow, Ge, and Schrand 2010 for a comprehensive review). If the quality is high, one may expect
the value-relevance of earnings to also be high since investors are better able to assess future firm
performance using earnings. There are two main approaches to measuring earnings quality. The
first is based on the statistical property of earnings, and the second is based on the attributes of the
accounting system.
For the first method, the main measure of earnings quality is earnings persistence. The idea is
that the value-relevance of earnings should be high if the good or bad news in current earnings is
expected to persist into the future. Kormendi and Lipe (1987) propose and confirm that high earn-
ings persistence is linked to high ERCs in the equity market. The relationship is further confirmed
in subsequent studies (e.g., Basu 1997, Ramakrishnan and Thomas 1998, and Li 2011). Another
statistical measure of earnings quality is earnings variability. It is proposed that investors prefer
earnings with less variability because the business model is not volatile for these companies (e.g.,
Schipper and Vincent 2003). For the second method, research finds that accounting systems can
affect the value-relevance of earnings. Teoh and Wong (1993) find that companies with high qual-
ity auditors tend to have larger ERCs. Francis, LaFond, Olsson, and Schipper (2004) show that
ERC is linked to the accrual quality of firms, a measure of the quality of a company’s accounting
system. Leuz and Verrecchia (2000), Bailey, Karolyi, and Salva (2006) and Greenstone, Oyer, and
Vissing-Jorgensen (2006) show that stringent accounting standards increase the value-relevance of
earnings. The argument based on the persistence and variability of earnings is readily applicable to
our setting, while the earnings quality related to the accounting system is not. Therefore, we focus
on studying the role of persistence and variability of new addresses growth in its value-relevance
in the cryptocurrency market.
We first test the relation between persistence and the value-relevance of new addresses. For
each period, we split the cross-section of cryptocurrencies into two groups based on the persistence
of ∆na in the past 12 weeks. We construct an indicator variable 1{p>Med} which equals one if the
persistence of ∆na is above the sample median and zero otherwise. Table 5 shows the results for
the high and low persistence of new address growth subsamples.
Columns (1) and (2) of Table 5 show the results for the subsamples of coins with below and
above median persistence of ∆na, respectively. The point estimate of ∆na is 0.105 for the above-
17

median subsample, while the point estimate of ∆na is 0.077 for the below-median subsample. That
is, the sensitivity between returns and new addresses growth is about 30 percent higher for the high
new addresses growth persistence subsample. Column (3) reports the results based on the full
sample and includes the cross term between 1{p>Med} and ∆na. The coefficient estimate is positive
and statistically significant at the 5-percent level, confirming that the returns/new addresses relation
is more pronounced among the coins with above median persistence of ∆na. In summary, we find
that the returns/new-addresses relation is more pronounced among coins with high persistence
of ∆na, suggesting that persistence is an important determinant of the value-relevance of new
addresses in the cryptocurrency market, similar to related findings in the equity market.
Table 5: Persistence of New Address Growth and Value Relevance
This table reports the results of the cross-sectional tests in the subsamples based on the above
median and below median persistence of new address growth. 1{p>Med} is an indicator variable
which equals to one if the persistence of new address growth is greater than the sample median
and zero otherwise. The regressions are based on Fama-MacBeth regression method. Variable
definition is documented in the Online Appendix. *, **, and *** represent significance level at the
10%, 5%, and 1% level.
(1) (2) (3)
ret ret ret
Sample <Med >Med Full
∆na 0.077*** 0.105*** 0.077***

(7.935) (11.327) (7.935)
1{p>Med} -0.003
(-1.042)
∆na ∗ 1{p>Med} 0.026**
(2.168)
Constant -0.004 -0.007 -0.004
(-0.440) (-0.745) (-0.440)
Observations 6,644 6,537 13,181

Adj R2 0.093 0.118 0.119
18

Table 6: Variability of New Address Growth and Value Relevance
median and below median variability of new address growth. 1{v>Med} is an indicator variable
which equals to one if the variability of new address growth is greater than the sample median
and zero otherwise. The regressions are based on Fama-MacBeth regression method. Variable
10%, 5%, and 1% level.
(1) (2) (3)
ret ret ret
∆na 0.123*** 0.073*** 0.123***

(11.511) (10.026) (11.511)
1{v>Med} 0.000
(0.086)
∆na ∗ 1{v>Med} -0.044***
(-4.422)
Constant -0.005 -0.005 -0.005
(-0.607) (-0.572) (-0.607)
Observations 6,709 6,587 13,296

Adj R2 0.128 0.080 0.111
Next, we study the relationship between the variability of new address growth and the value-
relevance of new addresses. For each period, we split the cross-section of cryptocurrencies into
two groups based on the variability of new address growths, measured as the standard deviation
of ∆na in the past 12 weeks. We construct an indicator variable 1{v>Med} which equals one if the
variability of ∆na is above the sample median and zero otherwise. Table 6 shows the results for
the high and low variability of new address growth subsamples.
Columns (1) and (2) of Table 6 show the results for the subsamples of coins with below and
above median variability of ∆na, respectively. The point estimate of ∆na is 0.073 for the above-
median subsample, while the point estimate of ∆na is 0.123 for the below-median subsample. Col-
umn (3) reports the results based on the full sample and includes the cross term between 1{v>Med}
and ∆na. The coefficient estimate is negative and statistically significant at the 1-percent level,
19

confirming that the returns/new addresses relation is more pronounced among the coins with be-
low median variability of ∆na. In summary, we find that the returns/new addresses relation is
more pronounced among coins with low variability of ∆na, suggesting that variability is an impor-
tant determinant of the value-relevance of new addresses in the cryptocurrency market, similar to
related findings in the equity market.
Discount rate
Another main determinant of the value-relevance of earnings is the discount rate. When the
discount rate is high for a stock, the discounted present value of the revisions in expected future
earnings would be low. That is, the value relevance of earnings is expected to be lower for compa-
nies with high discount rates. Empirically, prior studies test and confirm the negative relationship
between discount rates, including the risk-free and risky components, and the value relevance of
earnings. Collins and Kothari (1989) show that the value relevance of earnings is low in the equity
market when the risk-free rate is high. Easton and Zmijewski (1989) show that the value relevance
of earnings is negatively related to the risky components of the discount rates based on single- and
multi-beta versions of the CAPM in the cross-section of stocks.
Similarly, if the cryptocurrency valuation also follows a present value discounted formula (e.g.,
Cong, Li, and Wang Forthcoming; Biais, Bisiere, Bouvard, Casamatta, and Menkveld 2020), we
would expect to find a relatively low value relevance of new addresses when the discount rate is
high for a cryptocurrency. To measure the discount rate of cryptocurrencies, we use the cryptocur-
rency 3-factor model as in Liu, Tsyvinski, and Wu (2021). For each cryptocurrency, we estimate
the factor loadings on the 3 factors with a rolling 12-week window preceding the current period.
The implied discount rate is calculated based on the following formula:
βi,t
CMKT
E RCMKT + βi,t
CSM B
E RCSM B + βi,t
CMOM
E RCMOM + R f

where E RCMKT , E RCSM B , and E RCMOM are calculated as the average excess returns of the
factors from Liu, Tsyvinski, and Wu (2021).
Table 7 shows the results. For each period, we split the cross-section of cryptocurrencies into
two groups based on their implied discount rates. We construct an indicator variable 1{b>Med}
which equals one if the implied discount rate is above the sample median and zero otherwise.
Columns (1) and (2) show the results for the subsamples of coins with below and above median
20

discount rates, respectively. The point estimate to ∆na is 0.096 for the below-median subsample,
while the point estimate to ∆na is 0.083 for the above median subsample. That is, the sensitivity
between returns and new addresses growth is about 15 percent higher for the low discount rate
subsample. Column (3) reports the results based on the full sample and includes the cross term
between 1{d>Med} and ∆na. The cross term is negative but statistically insignificant.
Table 7: Discount Rate and Value Relevance
This table reports the results of the cross-sectional tests in the subsamples based on high and
low discount rate. 1{d>Med} is an indicator variable which equals to one if the discount rate is
greater than the sample median and zero otherwise. The regressions are based on Fama-MacBeth
regression method. Variable definition is documented in the Online Appendix. *, **, and ***
represent significance level at the 10%, 5%, and 1% level.
(1) (2) (3)
ret ret ret
∆na 0.096*** 0.083*** 0.096***

(9.843) (12.249) (9.843)
1{d>Med} 0.002
(0.662)
∆na ∗ 1{d>Med} -0.013
(-1.235)
Constant -0.004 -0.002 -0.004
(-0.442) (-0.195) (-0.442)
Observations 6,022 5,924 11,946

Adj R2 0.092 0.098 0.107
Size
In the equity market, it is theoretically proposed that the informativeness of firm price, often
proxied by firm size, may be negatively related to the value relevance of earnings (e.g., Easton and
Zmijewski 1989; Collins and Kothari 1989; Kothari 2001). The argument is that large firms tend
to receive more attention and are under more scrutiny by investors, and thus there is less infor-
mation content in the current accounting earnings, leading to a lower value relevance of earnings
21

at the release of information.10 A similar argument may be applied to the cryptocurrency market,
suggesting a negative relation between size and the value-relevance of new addresses.
On the other hand, there could be a positive relationship between size and the value relevance
of new addresses for cryptocurrencies. In the theoretical literature of both the network industries
and the cryptocurrencies, models with network externality (e.g., Rohlfs 1974; Katz and Shapiro
1985; Armstrong 2006; Cong, Li, and Wang 2021) imply a potential positive relation. In studying
the internet companies in the equity market, Rajgopal, Venkatachalam, and Kotha (2003) show
that the value relevance of the number of website users is higher for large e-commerce companies,
and they argue that this non-linearity is due to network effects.
To test whether size is a determinant of the value-relevance of new addresses, for each period,
we split the cross-section of cryptocurrencies into two groups based on their network sizes mea-
sured as the number of addresses.11 We construct an indicator variable 1{ A>Med} which equals one
if the number of address of the cryptocurrency is above the sample median in the previous period
and zero otherwise. Table 8 documents the results.
Columns (1) and (2) show the results for the subsamples of coins with below median network
size and above median network size, respectively. The point estimate to ∆na is 0.080 for the below
median subsample, while the point estimate to ∆na is 0.103 for the above median subsample. That
is, the sensitivity between returns and new addresses growth is about 29 percent higher for the large
network size subsample. Column (3) reports the results based on the full sample and includes the
cross term between 1{ A>Med} and ∆na. The coefficient estimate of the cross term is positive and
statistically significant at the 5-percent level, indicating that the returns/new addresses relation is
more pronounced among coins with larger network sizes. Overall, we find that the returns/new
addresses relation is more pronounced among coins with larger network sizes, consistent with the
theoretical argument regarding network effects.
10 The empirical evidence is mixed between the relationship between size and the value-relevance of earnings.
Easton and Zmijewski (1989) do not find support for the relationship. Collins and Kothari (1989) provide support for
the claim using an alternative method. Chapter 5.4 of Scott and O’Brien (2019) discusses the mixed results.
11 Using market cap generates qualitatively similar results
22

Table 8: Value Relevance Analysis – Subsample by Network Size
This table reports the results of the cross-sectional value-relevance tests in the subsamples based
on the network size of cryptocurrencies. 1{ A>Med} is an indicator variable which equals to one if
the total address of the cryptocurrency is above the sample median in the previous period and zero
otherwise. The regressions are based on Fama-MacBeth regression method. Variable definition is
documented in the Online Appendix. *, **, and *** represent significance level at the 10%, 5%,
and 1% level.
(1) (2) (3)
ret ret ret
∆na 0.080*** 0.103*** 0.080***

(9.850) (11.423) (9.850)
1{ A>Med} -0.002
(-0.511)
∆na ∗ 1{ A>Med} 0.023**
(2.136)
Constant -0.004 -0.006 -0.004
(-0.450) (-0.627) (-0.450)
Observations 6,613 6,906 13,519

Adj R2 0.085 0.106 0.108
5 Discrete and Continuous Revelation of Information

An important difference between cryptocurrencies and equities is that, for cryptocurrencies,
blockchain. This is in striking contrast to the earnings information of public companies. Under
the current reporting standards, publicly traded companies are required to report three quarterly
earnings and one annual earnings. This quarterly reporting system was not established until 1970.
Before the Securities Act of 1933 and the Securities Act of 1934 (Securities Acts), stock exchanges
largely governed financial reporting, and there was no homogeneous reporting requirement across
exchanges. Since the passage of the Securities Acts, the SEC has played a primary role in regu-
lating financial reporting for public companies. The SEC mandated annual reporting in 1934 and
23

semiannual reporting in 1955. In 1970, the SEC established the current quarterly reporting system.
This discrete nature of corporate earnings is extensively studied in the accounting and finance
literature. Theoretically, the net effect of reporting frequency is unclear (e.g., Diamond 1985;
Diamond and Verrecchia 1991; Abreu, Milgrom, and Pearce 1991; Easley and O’hara 2004).
Empirical studies examine both the benefit and costs associated with more frequent reporting. On
the benefit side, studies focus on the additional transparency that more reporting brings. Butler,
Kraft, and Weiss (2007) show that stock prices reflect accounting information more quickly when
firms voluntarily increase their reporting frequencies. More frequent reporting is also found to
be associated with reduced cost of capital and improved liquidity (e.g., Fu, Kraft, and Zhang
2012). On the cost side, both theories (e.g., Gigler, Kanodia, Sapra, and Venugopalan 2014)
and empirical studies suggest that higher reporting frequencies can cause managers to myopically
prefer short-term performance gains at the expense of overall firm value. Kraft, Vashishtha, and
Venkatachalam (2018) claim that increased reporting frequency in the 1970s is associated with a
large decline in investments, consistent with managers behaving myopically following increases
in reporting frequency. Ernstberger, Link, Stich, and Vogler (2017) show that firms with more
frequent reporting exhibit higher real earnings management.
Two important phenomenons in the equity market are attributable to the infrequent release of
information and suggest that information is not fully efficiently incorporated into prices. The first
phenomenon is referred to as the post-announcement drift. That is, information, such as innova-
tions to earnings, tend to forecast future company returns (e.g., Ball and Brown 1968; Bernard
and Thomas 1989; Sadka 2006; Daniel, Hirshleifer, and Sun 2020; Meursault, Liang, Routledge,
and Scanlon 2021). Among many alternative explanations for this phenomenon, a common view
is that the equity market is slow in incorporating earnings information. The second phenomenon
is referred to as the pre-announcement drift. That is, stock returns tend to precede informational
events such as earnings and Federal Open Market Committee (FOMC) announcements (e.g., Ball
and Brown 1968; Lucca and Moench 2015; Cieslak, Morse, and Vissing-Jorgensen 2019). The
literature commonly attributes the phenomenon to the discreteness of information releases and
information leakages between the discrete announcements.
The blockchain information is non-discretionary and continuously available by design. Thus,
the cryptocurrency market provides an ideal setting to study frequent reporting where the value-
relevant information is revealed continuously. In this section, we revisit these classic information
release-related phenomenons in the cryptocurrency market.
24

5.1 Slow Acquisition of Information
Post-earnings announcement drift (PEAD), first documented by Ball and Brown (1968), is one
of the earliest and well-known anomalies in the asset pricing and accounting literature. The PEAD
phenomenon can be described as that after earnings are announced, estimated cumulative excess
returns drift up for positive earnings innovations (good news) and drift down for negative earnings
innovations (bad news). A common explanation of PEAD is the slow acquisition of information
(e.g., Bernard and Thomas 1989; Daniel, Hirshleifer, and Sun 2020).
In the cryptocurrency market, information pertaining to new addresses is publicly and contin-
uously available. This environment provides a setting to study investor acquisition of information
where the value relevant information can be obtained at any time. It is ex ante unclear whether such
an environment would facilitate the information acquisition of investors. On the one hand, atten-
tion can be a scarce resource (e.g., Peng and Xiong 2006; DellaVigna and Pollet 2009; Hirshleifer,
Lim, and Teoh 2009) and continuous revelation of information can lead to severe under-reaction
from the investors.12 On the other hand, it is shown that stock prices reflect accounting information
more quickly when firms voluntarily increase their reporting frequencies (e.g., Butler, Kraft, and
Weiss 2007).
To study the information acquisition of news pertaining to new addresses in the cryptocurrency
market, we use ∆na to predict subsequent cryptocurrency returns of the corresponding coins. The
results are documented in Table 9. Columns (1) to (3) show the results of using ∆na to predict
one-, two-, and three-week ahead logged returns, respectively. The point estimates of ∆na for the
one-, two-, and three-week ahead predictability results are -0.005, 0.005, and -0.005, respectively,
which are insignificant and small compared to the contemporaneous response coefficient of 0.055.
In columns (4) to (6), we further include the cryptocurrency beta,13 size and momentum charac-
teristics as controls. The point estimates of ∆na remain insignificant and the magnitudes of the
coefficients become even smaller. We conclude that, using this metric, the cryptocurrency market
is efficient in responding to news pertaining to new addresses, consistent with the idea that frequent
revelation of information facilitates information acquisition.
12 Another plausible explanation of PEAD is based on changes in uncertainty and risks (e.g., Sadka 2006; Francis,
Lafond, Olsson, and Schipper 2007).

13 beta is measured as returns’ exposure to cryptocurrency market return in the past 12 weeks
25

Table 9: Slow Reaction to New Address Innovations
This table shows results of predicting one-, two-, three-periods ahead cryptocurrency returns using
new address growths. The regressions are based on Fama-MacBeth regression method. Variable
10%, 5%, and 1% level.
(1) (2) (3) (4) (5) (6)
ret+1 ret+2 ret+3 ret+1 ret+2 ret+3
∆na -0.005 0.005 -0.005 -0.006 0.000 -0.000

(-1.117) (1.194) (-1.115) (-1.039) (0.097) (-0.012)
beta 0.001 0.007* 0.007*
(0.158) (1.732) (1.850)
size -0.001* -0.001* -0.001
(-1.743) (-1.805) (-1.399)
mom 0.011 0.042*** -0.023
(0.537) (2.684) (-1.358)
Cons -0.002 -0.002 -0.002 0.022 0.022 0.012
(-0.248) (-0.243) (-0.250) (1.137) (1.141) (0.643)
Observations 13,517 13,408 13,298 12,486 12,377 12,267

Adj R2 0.010 0.005 0.006 0.059 0.041 0.040
5.2 Information Leakage

Next, we study the phenomenon of return drift prior to informational events that can also be
attributed to the infrequent release of information. Prior studies find that stock returns can precede
earnings surprises at earnings announcements by weeks or even months (e.g., Ball and Brown
1968; Scott and O’Brien 2019). That is, firms with positive earnings surprises tend to outperform
firms with negative earnings surprises even prior to the release date of earnings. The pre-drift
phenomenon is also detected in other informational events. For example, Lucca and Moench
(2015) document a sizable pre-FOMC announcement drift. A common explanation for the pre-
drift phenomenon is that, because of the discreteness of the events, information gradually leaks
to the market prior to the releases (e.g., Collins and Kothari 1989; Cieslak, Morse, and Vissing-
Jorgensen 2019). Other proposed explanations include changes in systematic risks and reduction
26

of uncertainty (e.g., Ai and Bansal 2018; Hu, Pan, Wang, and Zhu 2019). Because the on-chain
data are publicly and continuously available, the information leakage channel, which depends on
the discreteness of information releases, may lose its importance in the cryptocurrency market.
To test whether the pre-drift phenomenon is present in the cryptocurrency market, we regress
future changes in the logged number of new addresses on the current logged returns. The results
are summarized in Table 10, which uses logged returns to predict future ∆na up to three weeks
ahead. The results show that cryptocurrency returns do not significantly predict future ∆na at the
one-week and two-week horizons. At the three-week horizon, cryptocurrency returns predict ∆na
significantly, but the coefficient estimate is negative, which is inconsistent with the information
leakage hypothesis. The economic magnitude is small – a one-standard-deviation (25.1 percent)
increase in ret is only associated with a 1.8 percent increase in ∆na. Overall, we conclude that re-
turns do not immediately precede information about the new address in the cryptocurrency market,
plausibly due to the fact that information is released publicly and continuously in this market.
5.3 Cumulative Return around New Address Information

To further understand the behavior of cryptocurrency returns around the release of new address
information, we summarize the evidence in Figure 1. Specifically, we plot the cumulative average
logged abnormal return from 6 week before to 6 week after the release of new address information
for subgroups of cryptocurrencies based on positive and negative new address growth.
Consistent with the evidence so far, there is a significant market reaction to new address infor-
mation on the event date 0. However, both pre-announcement and post-announcement drifts are
muted in the cryptocurrency market. The return behavior in the cryptocurrency market is in sharp
contrast to the return behavior of the equity market around earnings release. Figure A.2 in the
Online Appendix shows the celebrated results of pre- and post-announcement drifts around earn-
ings announcements from Ball and Brown (1968). The figure shows a strong pre-announcement
drift even 12 months ahead of the announcement date and a post-announcement drift that persist
a few months subsequent to the announcement date. The presence of strong market reactions at
the release of new address information, the absence of post-announcement drift, and a very muted
pre-announcement drift highlight the distinct feature of cryptocurrencies and provide a benchmark
case where information is released both publicly and continuously.
27

Table 10: Return and Future New Address Growth
This table shows the results for predicting next period new address growths using current cryp-
tocurrency returns. Time fixed effect is included and standard errors are double clustered at the
coin level and time level. Variable definition is documented in the Online Appendix. *, **, and
*** represent significance level at the 10%, 5%, and 1% level.
(1) (2) (3)
∆na+1 ∆na+2 ∆na+3
ret 0.055 -0.049 -0.071***

(1.200) (-1.191) (-2.633)
Constant -0.019*** -0.017*** -0.015***
(-47.404) (-43.316) (-37.540)
Obs 13,519 13,451 13,374

Adj R2 0.052 0.052 0.053
6 Price-to-New Address Ratio and Return Predictability

The price-to-earnings (pe) ratio is one of the most studied financial ratios. In this section,
we construct the price-to-new address ratio, which we propose as the cryptocurrency counterpart
of the price-to-earnings ratio in the equity market. We first describe its properties and determi-
nants. Then, we show that it is a strong cross-sectional cryptocurrency return predictor and that its
predictive power is distinct from the existing predictors such as cryptocurrency market, size, and
momentum. Lastly, we investigate its other properties.
6.1 The Price-to-New Address Ratio

We first give a graphical illustration of the logged price-to-new address (pa) ratios of Bitcoin,
Ethereum, and Litecoin from 2018 to 2021. Figure 2 plots the pa ratios of these three cryptocur-
rencies. The pa ratios of Bitcoin and Ethereum are close to each other during the period. The
sample average pa ratio is higher than those of Bitcoin and Ethereum during most of the period.
In the time series, the pa ratios are relatively low for all three cryptocurrencies at the beginning of
2019 and 2020. In the past two years, the pa ratios for all three coins have been steadily increasing.
28

The pa ratios of Bitcoin and Ethereum are at the high points now. The price-to-earnings ratio of
the aggregate stock market is also at a high point since the financial crisis.
Figure 1: Cumulative Abnormal Return around New Address Information
This figure plots the cumulative average logged abnormal return around the release of new address
information for positive and negative new address growth subgroups. Dashed lines represent the
5-percent confidence intervals.
29

Figure 2: Price-to-New Address Ratio of Major Coins
This graph plots the logged price-to-new address ratio of Bitcoin, Ethereum, and Litecoin from
2018 to 2021. The blue line, red line, and green line show the results for Bitcoin, Ethereum, and
sample average of the logged price-to-new address ratio, respectively.
6.2 Return Predictability

In the asset pricing literature, a large number of studies (e.g., Basu 1977; Campbell and Shiller
1988a; Welch and Goyal 2008) have documented that the price-to-earnings ratios negatively pre-
dict returns in the future. When the price of a stock is high relative to its earnings, the stock tends
to have lower returns in the future.
Motivated by the evidence from the equity market, we examine the relationship between cryp-
tocurrency expected returns and price-to-new address ratios. Firstly, we examine this relationship
30

through the standard portfolio sorting method. Secondly, we study the long-horizon performance
of the price-to-new address ratios in predicting future cryptocurrency returns. Thirdly, we confirm
the return predictability results through the Fama-MacBeth regression method.
Portfolio sorting approach
We analyze the performance of the zero-investment long-short strategy based on the price-to-
earnings ratios. Each week, we sort individual cryptocurrencies into quartile portfolios based on
the value of price-to-earnings ratios of the corresponding coins.14 We track the return of each
portfolio in the week that follows. Then, we calculate the average excess returns over the risk-
free rate of each portfolio, as well as the excess returns of the long-short strategies based on the
difference between the fourth and the first quartiles.
Table 11 presents the results based on portfolio sortings. Panel A of Table 11 shows the aver-
age excess returns, standard deviations, and Sharpe ratios of the quartile portfolios and the zero-
investment long-short strategy. The average mean excess returns decrease monotonically from the
lowest pa ratio portfolio to the highest pa ratio portfolio. The average mean excess return of the
lowest pa ratio portfolio is 1.5 percent per week, and that of the highest pa ratio portfolio is -0.4
percent per week. The difference in the average returns of the lowest quartile and the highest quar-
tile is 1.9 percent per week, which is statistically significant at the 1-percent level. The standard
deviations for the lowest quartile portfolio and the highest quartile portfolio are 11.4 percent and
10.5 percent, respectively. The excess returns of the zero-investment long-short strategy have a
standard deviation of 6.8 percent. The Sharpe ratios also decrease monotonically from the lowest
to the highest pa ratio portfolios. The weekly Sharpe ratio for the lowest pa ratio portfolio is 0.132
(0.951 annually), and that for the highest pa ratio portfolio is -0.038 (-0.274 annually). The weekly
Sharpe ratio for the zero-investment long-short strategy is 0.279 (2.012 annually).
14 We choose quartile portfolios because the average number of cryptocurrencies in the cross-section is about one
hundred. Therefore, quartile portfolios ensure that there are about thirty coins in each portfolio. In robustness tests,
we find that sorting cryptocurrencies into tercile or quintile portfolios gives qualitatively similar results.
31

Table 11: Portfolio Sorts Based on Cryptocurrency Price-to-New Address Ratio
This table shows portfolio sorting results based on cryptocurrency price-to-new address ratios.
Panel A reports the means, standard deviations, and Sharpe ratios of the value-weighted portfolios
excess returns. Panel B reports the factor exposures of the portfolios sorted on price-to-new address
ratios. The factor models in include the cryptocurrency market factor and the cryptocurrency three
factor model. Variable definition is documented in the Online Appendix. *, **, and *** represent
significance level at the 10%, 5%, and 1% level.
Panel A
Low 2 3 High Low–High
Mean 0.015* 0.008 0.006 -0.004 0.019***

(1.682) (0.769) (0.550) (-0.553) (3.638)
SD 0.114 0.142 0.135 0.105 0.068
SR 0.132 0.056 0.044 -0.038 0.279
Panel B Alpha CMKT CSIZE CMOM R2
Low 0.003 (1.481) 1.025*** (46.655) 0.927

2 -0.003 (-0.382) 1.057*** (17.088) 0.632
3 -0.004 (-0.645) 0.970*** (15.572) 0.588
High -0.013*** (-2.819) 0.810*** (19.179) 0.684
Low–High 0.016*** (3.387) 0.218*** (4.789) 0.119
Low 0.004 (1.559) 1.029*** (47.160) 0.058* (1.971) -0.051** (-2.439) 0.933
2 -0.007 (-1.254) 1.153*** (22.615) 0.637*** (9.271) -0.023 (-0.472) 0.767
3 -0.009* (-1.684) 1.073*** (21.704) 0.681*** (10.215) -0.027 (-0.559) 0.758
High -0.015*** (-3.934) 0.865*** (24.291) 0.390*** (8.127) -0.062* (-1.802) 0.790
Low–High 0.019*** (4.131) 0.169*** (3.937) -0.322*** (-5.580) 0.009 (0.215) 0.271
Panel B of Table 11 reports results controlling for either the cryptocurrency CAPM model or the
cryptocurrency 3-factor model (see Liu, Tsyvinski, and Wu 2021). In the first part of Panel B, we
use the cryptocurrency CAPM model, where cryptocurrency market excess returns are included.
The beta loadings to the cryptocurrency market factor decrease monotonically from the lowest
to the highest pa ratio portfolios. The zero-investment long-short strategy significantly loads on
32

the cryptocurrency market factor. However, the alpha of the long-short strategy remains positive
and significant at the 1-percent level. The economic magnitude of the alpha is 1.6 percent per
week, suggesting that the cryptocurrency CAPM model cannot explain a sizable portion of the
excess returns. In the second part of Panel B, we use the cryptocurrency 3-factor model from
Liu, Tsyvinski, and Wu (2021). The long-short strategy loads positively and significantly on the
cryptocurrency market factor, while negatively and significantly on the cryptocurrency size factor.
The long-short strategy does not significantly expose to the cryptocurrency momentum factor.
The alpha of the long-short strategy remains positive and significant at the 1-percent level after
controlling for the cryptocurrency 3-factor model. The economic magnitude of the alpha is 1.9
percent per week. These results show that the return predictability cannot be subsumed and is
distinct from those of the cryptocurrency market, size, and momentum effect.
Different horizons
We have shown that the price-to-new address ratios negatively and significantly predict next
period cryptocurrency returns. We further examine the persistence of the return predictability
power of the price-to-new address ratio.
To study the long-term return predictability of the price-to-new address ratio, we calculate the
weekly average excess returns, the cryptocurrency CAPM alphas, and the cryptocurrency 3-factor
alphas for the zero-investment long-short strategy from 1-week to 28-week after the formation of
the portfolios. The results are summarized in Table 12. The return predictability of pa ratios is rel-
atively long-lasting. The average excess returns of the long-short strategy are positive throughout
the different horizons, and are significant at the 5-percent level up to 12-week ahead. The cryp-
tocurrency CAPM model adjusted alphas remain positive and significant at the 10-percent level up
to 12-week ahead at 0.9 percent per week. The cryptocurrency 3-factor model adjusted alphas stay
positive and significant at the 5-percent level at 1.0 percent per week up to 12-week ahead. The
cryptocurrency 3-factor model adjusted alphas remain positive throughout the 28 weeks.
In summary, we find evidence that the pa ratios negatively and significantly predict future
cryptocurrency returns over a relatively long horizon.
33

Table 12: Cryptocurrency Price-to-New Address Effect – Different Horizon
This table shows portfolio sorting results over different horizons based on cryptocurrency price-
to-new address ratios. The Low–High row reports the average long-short strategy returns based
on cryptocurrency price-to-new address ratios. The C1 Alpha and C3 Alpha rows report the cryp-
tocurrency market factor and the cryptocurrency three-factor model adjusted alphas, respectively.
Weeks +1 +4 +8 +12 +16 +20 +24 +28
Low 0.015* 0.017** 0.018** 0.014 0.016* 0.019** 0.017* 0.017*

(1.682) (1.990) (2.033) (1.541) (1.787) (2.106) (1.887) (1.891)
2 0.008 0.009 0.005 0.008 0.014 0.016 0.005 0.017
(0.769) (0.853) (0.509) (0.787) (1.363) (1.379) (0.447) (1.644)
3 0.006 0.006 0.015 0.011 0.011 0.011 0.019* 0.012
(0.550) (0.616) (1.380) (1.018) (0.984) (1.032) (1.691) (1.102)
High -0.004 0.000 0.005 0.001 0.008 0.009 0.008 0.012
(-0.553) (0.030) (0.627) (0.112) (1.004) (1.083) (0.928) (1.469)
Low–High 0.019*** 0.018*** 0.013** 0.012** 0.009 0.010* 0.010 0.005
(3.638) (3.170) (2.322) (2.214) (1.502) (1.660) (1.585) (0.843)
C1 Alpha 0.016*** 0.015*** 0.009* 0.009* 0.004 0.005 0.005 -0.000

(3.387) (2.819) (1.758) (1.785) (0.797) (0.928) (0.941) (-0.050)
C3 Alpha 0.019*** 0.017*** 0.010** 0.010** 0.006 0.008* 0.009* 0.004

(4.131) (3.458) (2.131) (2.280) (1.420) (1.816) (1.830) (0.772)
6.3 Robustness of the Predictability

Fama-MacBeth regression method
We further test the robustness of the relation between price-to-new address ratios and future
cryptocurrency returns from the cross-sectional regressions using the Fama-MacBeth method. Ta-
ble 13 shows the results.
Column (1) of Table 13 only includes the pa ratio as the cryptocurrency return predictor. Con-
sistent with the portfolio sorting results, the coefficient estimate to pa ratio is negative and signifi-
cant at the 1-percent level, confirming that a high pa ratio is associated with low expected returns.
34

The economic magnitude of the effect is large. The point estimate is -0.004, suggesting that a
one-standard-deviation increase in the pa ratio is associated with a 0.6 percent decrease in the next
week’s cryptocurrency returns. Columns (2) and (4) include beta, size, and momentum as controls,
respectively. Column (4) includes all of beta, size and momentum as controls. The magnitude of
the point estimates to the pa ratio decrease, but the estimates remain negative and statistically sig-
nificant at the 1-percent level when beta, size, and momentum are entered separately and at the
5-percent level when all of beta, size, and momentum are controlled.
Table 13: Fama-MacBeth Regression Results
This table shows the results for predicting next period returns using the price-to-new address ratios.
The regressions are based on Fama-MacBeth regression method. Variable definition is documented
in the Online Appendix. *, **, and *** represent significance level at the 10%, 5%, and 1% level.
(1) (2) (3) (4) (5)
RET+1 RET+1 RET+1 RET+1 RET+1
pa -0.004*** -0.003*** -0.003*** -0.003** -0.003**

(-2.805) (-2.696) (-2.674) (-2.563) (-2.554)
beta 0.004 0.003
(0.877) (0.755)
size -0.002*** -0.002**
(-2.679) (-2.065)
mom 0.014 0.019
(0.862) (1.138)
Constant 0.054** 0.051** 0.089*** 0.045** 0.073**
(2.539) (2.468) (3.015) (2.142) (2.530)
Observations 13,455 12,449 13,455 13,455 12,449

R-squared 0.017 0.049 0.031 0.058 0.102
Alternative portfolio formation
In our baseline portfolio sorting results, we sort cryptocurrencies into quartile portfolios based
on their price-to-new address ratios. This choice is due to the fact that the average number of
cryptocurrencies in the cross-section is about one hundred. Therefore, quartile portfolios ensure
that there are about thirty coins in each portfolio.
35

To test the robustness of the portfolio sorting results, we further sort cryptocurrencies into
tercile and quintile portfolios based on their price-to-new address ratios and examine the average
future value-weighted returns of the portfolios. The results based on these alternative portfolio
formations are summarized in Table A.3 in the Appendix.
The results based on tercile and quintile portfolios confirm the baseline results based on quar-
tile portfolios. For the results based on tercile portfolios, the zero-investment long-short strategy
generates an average weekly excess return of 1.6 percent that is statistically significant at the 1
percent level. The weekly Sharpe ratio of the excess returns of the long-short strategy is 0.242. For
the results based on quintile portfolios, the zero-investment long-short strategy also generates an
average weekly excess return of 1.6 percent, which is significant at the 1 percent level. The weekly
Sharpe ratio of the excess returns of the long-short strategy is 0.219.
Top 30 coins
In our baseline portfolio sorting results, we use the full cross-section of cryptocurrencies where
on-chain and price information is available. We further investigate the behavior of the portfolio
sorting results by restricting our sample to the largest and most liquid coins in the market and test
the performances of the return predictability of pa. Each period, we restrict our sample to the
largest 30 coins at the time based on the market capitalization of the cryptocurrencies. We form
tercile portfolios based on their price-to-new address ratios. The results are summarized in Table
14.
The results based on the top 30 largest coins confirm the baseline results based on the full cross-
section of cryptocurrencies. The long-short strategy that buys the tercile with the lowest price-to-
new address ratios and shorts the tercile with the highest price-to-new address ratios generates
an average weekly excess return of 1.5 percent that is statistically significant at the 1 percent
level. The weekly Sharpe ratio of the excess returns of the long-short strategy is 0.217, which is
economically large. Adjusting for the cryptocurrency CAPM model, the alpha of the long-short
strategy decreases slightly to 1.3 percent per week, which is significant at the 5 percent level.
Adjusting for the cryptocurrency 3-factor model, the alpha of the long-short strategy remains at
1.5 percent per week and is significant at the 1 percent level.
36

Table 14: Portfolio Sorts – Top 30 Cryptocurrencies
This table shows portfolio sorting results based on cryptocurrency price-to-new address ratios us-
ing the largest 30 cryptocurrencies by market capitalization. Mean is the average weekly excess
returns of the portfolios and the long-short strategy. SD is the standard deviation of the weekly
excess returns of the portfolios and the long-short strategy. SR is the weekly Sharpe ratio of the
excess returns of the portfolios and t0he long-short strategy. Variable definition is documented in
the Online Appendix. *, **, and *** represent significance level at the 10%, 5%, and 1% level.
Panel A
Low 2 High Low–High
Mean 0.012 0.006 -0.003 0.015***

(1.400) (0.544) (-0.402) (2.814)
SD 0.111 0.141 0.107 0.069
SR 0.108 0.043 -0.028 0.217
Low 0.001 (0.579) 1.028*** (86.532) 0.978

2 -0.005 (-0.664) 1.014*** (15.720) 0.592
High -0.012** (-2.418) 0.815*** (18.082) 0.658
Low–High 0.013** (2.520) 0.215*** (4.583) 0.110
Low 0.001 (0.654) 1.027*** (83.809) -0.000 (-0.017) -0.010 (-0.814) 0.978
2 -0.009 (-1.639) 1.117*** (21.231) 0.678*** (9.565) -0.022 (-0.444) 0.747
High -0.014*** (-3.355) 0.873*** (22.036) 0.394*** (7.382) -0.038 (-0.987) 0.754
Low–High 0.015*** (3.382) 0.157*** (3.721) -0.387*** (-6.790) 0.027 (0.661) 0.326
Portfolio sorting excluding Bitcoin
The cryptocurrency market has a clear level factor – Bitcoin. Because of the dominant position
of Bitcoin in the cryptocurrency market, the returns of the portfolios including Bitcoins would
strongly comove with Bitcoin return. We repeat our portfolio sorting analysis based on the price-
to-new address ratio excluding Bitcoin. The results are summarized in Table A.4 in the Online
37

Appendix.
The zero-investment long-short strategy that buys the portfolio with the lowest price-to-new
address ratios and shorts the portfolio with the highest price-to-new address ratios generates an
average excess return of 1.3 percent per week and is statistically significant at the 5 percent level.
Adjusting for the cryptocurrency CAPM model, the alpha of the long-short strategy decreases
slightly to 1.2 percent per week, which is significant at the 5 percent level. Adjusting for the
cryptocurrency 3-factor model, the alpha of the long-short strategy remains at 1.3 percent per
week and is significant at the 1 percent level.
Equal-Weighted
For our main portfolio sorting results, we report the performance of the strategy based on value-
weighted results. We further test whether the results are robust under equal-weighted portfolios.
The results based on equal-weighted portfolios are reported in Table A.5 in the Online Appendix.
We find that the results based on equal-weighted portfolios are qualitatively similar to those
based on value-weighted portfolios. The average long-short strategy excess return is 1.6 percent per
week, compared to 1.9 percent per week under the value-weighted portfolios. The alphas adjusted
for the cryptocurrency CAPM model and the cryptocurrency 3-factor model remain positive and
statistically significant at the 1 percent level.
6.4 Additional Results

We provide additional results regarding the price-to-new address ratio of the cryptocurrency
market.
Short Sample
Because the data on the number of new addresses are only available starting from 2018, we
only have about four years of the cross-section of cryptocurrencies. The short sample is a potential
barrier to studying the return predictability of the price-to-new address ratio we cannot avoid.
Moreover, as discussed above, there is likely a great deal of uncertainty and investor learning
about cryptocurrencies during the period. As argued in Pástor and Veronesi (2009), it can take
time for investors to learn and understand emerging technologies, which may lead to the departure
between prices and the fundamental of the technology.
38

We partially address these concerns by breaking the sample into two halves and checking
whether the return predictability results of the price-to-new address ratio are stable for these two
subsamples. These two subsamples both feature important events, which can facilitate the test of
the robustness of our results. In the first half of the sample, the cryptocurrency market witnessed
one of the largest declines in prices since its debut. The second half of the sample contains the
COVID-19 period. The results are summarized in Table A.2 in the Online Appendix.
For the first half of the sample, the average excess return of the zero-investment long-short
strategy is 1.8 percent per week, while that for the second half of the sample is 1.9 percent per
week. The average excess returns for both subsamples are positive and statistically significant
at the 5-percent level. The standard deviations of excess returns for the long-short strategy are
7.0 percent and 6.5 percent for the first half and the second half of the sample, respectively. The
weekly Sharpe ratios for these two subsamples are 0.257 and 0.292, respectively. We conclude that
the performance of the long-short strategy is stable and consistent across the two subsamples.
Exposures to equity factors
Additionally, we test whether the equity factor models can account for the long-short strategy
excess returns based on the pa ratio. In terms of the equity factor models, we use the CAPM,
Fama-French 3-factor, and Fama-French 5-factor models. The results are summarized in Table
A.6 in the Appendix.
We find that the quartile portfolio excess returns are positive and significantly exposed to the
equity market factor. This result highlights a recent development of the cryptocurrency market
that the market excess return is significantly exposed to the equity market excess returns.15 How-
ever, the excess returns of the zero-investment long-short strategy based on the pa ratio do not
significantly expose to the equity market factor. The excess returns of the strategy also do not sig-
nificantly expose to the Fama-French 3 factors and the Fama-French 5 factors. The alphas of the
strategy remain positive and statistically significant at the 1-percent level. The magnitudes of the
alphas are about 1.9 percent per week. Overall, the results suggest that the equity market factors
do not account for the excess returns of the strategy.
15 The cryptocurrency market excess returns do not significantly expose to the equity market excess returns between
2011 and 2018 (see Liu and Tsyvinski 2021).
39

Controlling for coin market factor excluding Bitcoin
Similar to the currency market, where U.S. dollar is a dominant level factor, Bitcoin is the level
factor for the cryptocurrency market. We construct an alternative coin market factor measured
as the value-weighted return of all coins excluding Bitcoin and test whether this alternative coin
market factor can explain the long-short strategy excess returns generated by the price-to-new
address ratios. We document the results in Table A.7 in the Online Appendix.
We denote this alternative cryptocurrency market factor as CMKT2. We regress the level port-
folio returns and the long-short strategy spread on cryptocurrency CAPM and 3-factor models
using CMKT2. Similar to the results using CMKT, the alphas of the long-short strategies remain
positive and statistically significant adjusting for both factor models. The alpha is 1.8 percent per
week adjusting for the cryptocurrency CAPM model using CMKT2, and the alpha is 2.0 percent
per week adjusting for the cryptocurrency 3-factor model using CMKT2.
Predicting future new address growth
A number of studies (e.g., Campbell and Shiller 1988a; Campbell and Shiller 1988b; Vuolteenaho
2002) find that valuation ratios such as price-to-earnings ratios and price-to-dividend ratios can be
decomposed and have to predict future returns or earnings (dividend) growths or both. Campbell
and Shiller (1988b) and Campbell and Shiller (1988a), among others, find that the price-to-earnings
(price-to-dividend) ratios predict future equity excess returns but not future earnings (dividend)
growths in the time series. On the other hand, Vuolteenaho (2002) shows that the valuation ratios
predict both future returns and earnings in the cross-section.
A similar decomposition can be applied to the price-to-new address ratio. We have shown that
the price-to-new address ratio predicts future cryptocurrency returns. However, in a dynamic cryp-
tocurrency pricing model with the network effect, cryptocurrency prices not only capture current
cryptocurrency adoption but also contain information about expected future adoption (e.g., Cong,
Li, and Wang 2021). Therefore, we further test whether the price-to-new address ratio predicts
future new address growth. Table A.8 in the Online Appendix shows the cross-sectional results of
regressing next period logged new address growths on current pa ratios using the Fama-MacBeth
method.
Column (1) reports the result of the specification with only the pa ratio. The coefficient esti-
mate to the pa ratio is positive and statistically significant at the 1-percent level. In other words,
40

high pa ratios are on average followed by high new address growths next period. The magnitude
of the point estimate is 0.090. That is, a one-standard-deviation increase in the pa ratio is asso-
ciated with an increase of about 12 percent in the logged new address growth of the next period.
In Columns (2) to (5), we control for the beta, size, and momentum characteristics. The coeffi-
cient estimates to the pa ratio stay positive and statistically significant at the 1-percent level when
these additional controls are included. The magnitude of the point estimate remains similar to the
standalone specification. The results strongly support the idea that the pa ratios of the cryptocur-
rencies are related to expected new address growths in the cross-section, which is consistent with
cryptocurrency pricing models featuring dynamic user adoption.
7 Conclusion
The importance of user adoption and network effect is a foundational concept in the theoretical
literature of both the network economy (e.g., Rohlfs 1974; Katz and Shapiro 1985; Katz and
Shapiro 1994; Rochet and Tirole 2003) and the cryptocurrency valuation (e.g., Cong, Li, and Wang
Forthcoming; Sockin and Xiong 2020; Biais, Bisiere, Bouvard, Casamatta, and Menkveld 2020).
In this paper, we empirically evaluate whether investors react to the information pertaining to new
addresses and show that the information is highly value-relevant in the cryptocurrency market.
In fact, the value relevance of new addresses in the cryptocurrency market is higher than that of
earnings in the equity market.
One important difference between cryptocurrencies and equities is that, for cryptocurrencies,
blockchain. While for equities, public information of firm activities is only available periodically.
Two important phenomenons in the equity market are commonly attributed to the infrequent re-
lease of information: pre-announcement drift and post-announcement drift. We show that pre-
and post-drifts are absent around the release of new address information. The presence of strong
market reactions at the release of new address information and the absence of pre- and post-drifts
highlight the distinct features of cryptocurrencies and provide a benchmark case where information
is released both publicly and continuously.
Lastly, we construct the price-to-new address ratio of each cryptocurrency. We show that,
similar to the price-to-earnings ratio in the equity market, the price-to-new address ratios negatively
predict future cryptocurrency returns in the cross-section. A long-short strategy that buys the coins
41

with the lowest price-to-new address ratios and shorts the coins with the highest price-to-new
address ratios generates positive and statistically significant excess returns of 1.9 percent per week.
The excess returns of the long-short strategy cannot be subsumed by existing cryptocurrency factor
models or equity factor models.
References
Abreu, Dilip, Paul Milgrom, and David Pearce (1991). “Information and timing in repeated part-
nerships”. In: Econometrica: Journal of the Econometric Society, pp. 1713–1733.
Agrawal, Anup and Jeffrey F Jaffe (2000). “The post-merger performance puzzle”. In: Advances
in Mergers and Acquisitions. Emerald Group Publishing Limited.
Ai, Hengjie and Ravi Bansal (2018). “Risk preferences and the macroeconomic announcement
premium”. In: Econometrica 86.4, pp. 1383–1430.
Aloosh, Arash and Jiasun Li (2019). “Direct evidence of bitcoin wash trading”. In: Available at
SSRN 3362153.
Amir, Eli and Baruch Lev (1996). “Value-relevance of nonfinancial information: The wireless com-
munications industry”. In: Journal of Accounting and Economics 22.1-3, pp. 3–30.
Armstrong, Mark (2006). “Competition in two-sided markets”. In: The RAND Journal of Eco-
nomics 37.3, pp. 668–691.
Austin, David H (1993). “An event-study approach to measuring innovative output: The case of
biotechnology”. In: The American Economic Review 83.2, pp. 253–258.
Bailey, Warren, G Andrew Karolyi, and Carolina Salva (2006). “The economic consequences of
increased disclosure: Evidence from international cross-listings”. In: Journal of Financial Eco-
nomics 81.1, pp. 175–213.
Ball, Ray and Philip Brown (1968). “An empirical evaluation of accounting income numbers”. In:
Journal of Accounting Research, pp. 159–178.
Basu, Sanjoy (1977). “Investment performance of common stocks in relation to their price-earnings
ratios: A test of the efficient market hypothesis”. In: The Journal of Finance 32.3, pp. 663–682.
Basu, Sudipta (1997). “The conservatism principle and the asymmetric timeliness of earnings1”.
In: Journal of Accounting and Economics 24.1, pp. 3–37.
Beaver, William H (1968). “The information content of annual earnings announcements”. In: Jour-
nal of Accounting Research, pp. 67–92.
Bernanke, Ben S and Kenneth N Kuttner (2005). “What explains the stock market’s reaction to
Federal Reserve policy?” In: The Journal of Finance 60.3, pp. 1221–1257.
Bernard, Victor L and Jacob K Thomas (1989). “Post-earnings-announcement drift: delayed price
response or risk premium?” In: Journal of Accounting Research 27, pp. 1–36.
42

Bhambhwani, Siddharth, Stefanos Delikouras, and George Korniotis (2020). “Do Fundamentals
Drive Cryptocurrency Returns?” In:
Biais, Bruno, Christophe Bisiere, Matthieu Bouvard, Catherine Casamatta, and Albert J Menkveld
(2020). “Equilibrium bitcoin pricing”. In: Available at SSRN 3261063.
Borri, Nicola (2019). “Conditional tail-risk in cryptocurrency markets”. In: Journal of Empirical
Finance 50, pp. 1–19.
Borri, Nicola and Kirill Shakhnov (2018). “Cryptomarket discounts”. In: Available at SSRN 3124394.
— (2019). “The cross-section of cryptocurrency returns”. In: Available at SSRN 3241485.
Boyd, John H, Jian Hu, and Ravi Jagannathan (2005). “The stock market’s reaction to unem-
ployment news: Why bad news is usually good for stocks”. In: The Journal of Finance 60.2,
pp. 649–672.
Butler, Marty, Arthur Kraft, and Ira S Weiss (2007). “The effect of reporting frequency on the
timeliness of earnings: The cases of voluntary and mandatory interim reports”. In: Journal of
Accounting and Economics 43.2-3, pp. 181–217.
Campbell, John Y and Robert J Shiller (1988a). “Stock prices, earnings, and expected dividends”.
In: the Journal of Finance 43.3, pp. 661–676.
— (1988b). “The dividend-price ratio and expectations of future dividends and discount factors”.
In: The Review of Financial Studies 1.3, pp. 195–228.
Chan, Louis KC, Josef Lakonishok, and Theodore Sougiannis (2001). “The stock market valuation
of research and development expenditures”. In: The Journal of Finance 56.6, pp. 2431–2456.
Cieslak, Anna, Adair Morse, and Annette Vissing-Jorgensen (2019). “Stock returns over the FOMC
cycle”. In: The Journal of Finance 74.5, pp. 2201–2248.
Collins, Daniel W and SP Kothari (1989). “An analysis of intertemporal and cross-sectional deter-
minants of earnings response coefficients”. In: Journal of Accounting and Economics 11.2-3,
pp. 143–181.
Cong, Lin William, Xi Li, Ke Tang, and Yang Yang (2021). “Crypto wash trading”. In: arXiv
preprint arXiv:2108.10984.
Cong, Lin William, Ye Li, and Neng Wang (Forthcoming). “Token-based platform finance”. In:
Journal of Financial Economics.
— (2021). “Tokenomics: Dynamic adoption and valuation”. In: The Review of Financial Studies
34.3, pp. 1105–1155.
Cong, William, Zhiheng He, and Ke Tang (2021). “Crypto Carry and the Staking Economy”. In:
Working Paper.
Cong, William, Andrew Karolyi, Ke Tang, and Weiyi Zhao (2021). “Value premium, network
adoption, and factor pricing of crypto assets”. In: Working Paper.
Daniel, Kent, David Hirshleifer, and Lin Sun (2020). “Short-and long-horizon behavioral factors”.
In: The Review of Financial Studies 33.4, pp. 1673–1736.
De Franco, Gus, MH Franco Wong, and Yibin Zhou (2011). “Accounting adjustments and the
valuation of financial statement note information in 10-K filings”. In: The Accounting Review
86.5, pp. 1577–1604.
43

Dechow, Patricia, Weili Ge, and Catherine Schrand (2010). “Understanding earnings quality: A
review of the proxies, their determinants and their consequences”. In: Journal of Accounting
and Economics 50.2-3, pp. 344–401.
DellaVigna, Stefano and Joshua M Pollet (2009). “Investor inattention and Friday earnings an-
nouncements”. In: The Journal of Finance 64.2, pp. 709–749.
DePamphilis, Donald (2019). Mergers, acquisitions, and other restructuring activities: An inte-
grated approach to process, tools, cases, and solutions. Academic Press.
Diamond, Douglas W (1985). “Optimal release of information by firms”. In: The Journal of Fi-
nance 40.4, pp. 1071–1094.
Diamond, Douglas W and Robert E Verrecchia (1991). “Disclosure, liquidity, and the cost of cap-
ital”. In: The Journal of Finance 46.4, pp. 1325–1359.
Easley, David and Maureen O’hara (2004). “Information and the cost of capital”. In: The Journal
of Finance 59.4, pp. 1553–1583.
Easton, Peter D and Mark E Zmijewski (1989). “Cross-sectional variation in the stock market
response to accounting earnings announcements”. In: Journal of Accounting and Economics
11.2-3, pp. 117–141.
Ernstberger, Jürgen, Benedikt Link, Michael Stich, and Oliver Vogler (2017). “The real effects of
mandatory quarterly reporting”. In: The Accounting Review 92.5, pp. 33–60.
Fama, Eugene F (1998). “Market efficiency, long-term returns, and behavioral finance”. In: Journal
of Financial Economics 49, pp. 283–306.
Fama, Eugene F, Lawrence Fisher, Michael Jensen, and Richard Roll (1969). “The adjustment of
stock prices to new information”. In: International Economic Review 10.1.
Fama, Eugene F and James D MacBeth (1973). “Risk, return, and equilibrium: Empirical tests”.
In: Journal of Political Economy 81.3, pp. 607–636.
Farrell, Joseph and Garth Saloner (1986). “Installed base and compatibility: Innovation, product
preannouncements, and predation”. In: The American Economic Review, pp. 940–955.
Financial Accounting Standards Board (2010). “Statement of financial accounting concepts”. In:
Conceptual Framework for Financial Reporting No. 8.
Francis, Jennifer, Ryan LaFond, Per M Olsson, and Katherine Schipper (2004). “Costs of equity
and earnings attributes”. In: The Accounting Review 79.4, pp. 967–1010.
Francis, Jennifer, Ryan Lafond, Per Olsson, and Katherine Schipper (2007). “Information uncer-
tainty and post-earnings-announcement-drift”. In: Journal of Business Finance & Accounting
34.3-4, pp. 403–433.
Fu, Renhui, Arthur Kraft, and Huai Zhang (2012). “Financial reporting frequency, information
asymmetry, and the cost of equity”. In: Journal of Accounting and Economics 54.2-3, pp. 132–
149.
Gandal, Neil (1994). “Hedonic price indexes for spreadsheets and an empirical test for network
externalities”. In: the RAND Journal of Economics, pp. 160–170.
44

Gigler, Frank, Chandra Kanodia, Haresh Sapra, and Raghu Venugopalan (2014). “How frequent
financial reporting can cause managerial short-termism: An analysis of the costs and benefits
of increasing reporting frequency”. In: Journal of Accounting Research 52.2, pp. 357–387.
Goldfarb, Avi and Catherine Tucker (2019). “Digital economics”. In: Journal of Economic Litera-
ture 57.1, pp. 3–43.
Greenstone, Michael, Paul Oyer, and Annette Vissing-Jorgensen (2006). “Mandated disclosure,
stock returns, and the 1964 Securities Acts amendments”. In: The Quarterly Journal of Eco-
nomics 121.2, pp. 399–460.
Griffin, John M and Amin Shams (2020). “Is Bitcoin really untethered?” In: The Journal of Finance
75.4, pp. 1913–1964.
Hagiu, Andrei and Julian Wright (2015). “Multi-sided platforms”. In: International Journal of
Industrial Organization 43, pp. 162–174.
Hall, Bronwyn H, Adam Jaffe, and Manuel Trajtenberg (2005). “Market value and patent cita-
tions”. In: RAND Journal of Economics, pp. 16–38.
Hirshleifer, David, Sonya Seongyeon Lim, and Siew Hong Teoh (2009). “Driven to distraction:
Extraneous events and underreaction to earnings news”. In: The Journal of Finance 64.5,
pp. 2289–2325.
Hu, Albert S, Christine A Parlour, and Uday Rajan (2019). “Cryptocurrencies: Stylized facts on a
new investible instrument”. In: Financial Management 48.4, pp. 1049–1068.
Hu, Xing, Jun Pan, Jiang Wang, and Haoxiang Zhu (2019). Premium for heightened uncertainty:
Solving the fomc puzzle. National Bureau of Economic Research.
Ittner, Christopher D and David F Larcker (1998). “Are nonfinancial measures leading indicators
of financial performance? An analysis of customer satisfaction”. In: Journal of Accounting
Research 36, pp. 1–35.
Katz, Michael L and Carl Shapiro (1985). “Network externalities, competition, and compatibility”.
In: The American Economic Review 75.3, pp. 424–440.
— (1986). “Technology adoption in the presence of network externalities”. In: Journal of Political
Economy 94.4, pp. 822–841.
— (1994). “Systems competition and network effects”. In: Journal of Economic Perspectives 8.2,
pp. 93–115.
Kim, Oliver and Robert E Verrecchia (1991). “Market reaction to anticipated announcements”. In:
Journal of Financial Economics 30.2, pp. 273–309.
Kogan, Leonid, Dimitris Papanikolaou, Amit Seru, and Noah Stoffman (2017). “Technological
innovation, resource allocation, and growth”. In: The Quarterly Journal of Economics 132.2,
pp. 665–712.
Kormendi, Roger and Robert Lipe (1987). “Earnings innovations, earnings persistence, and stock
returns”. In: Journal of Business, pp. 323–345.
Kothari, SP (2001). “Capital markets research in accounting”. In: Journal of Accounting and Eco-
nomics 31.1-3, pp. 105–231.
45

Kraft, Arthur G, Rahul Vashishtha, and Mohan Venkatachalam (2018). “Frequent financial report-
ing and managerial myopia”. In: The Accounting Review 93.2, pp. 249–275.
Leuz, Christian and Robert E Verrecchia (2000). “The economic consequences of increased dis-
closure”. In: Journal of Accounting Research, pp. 91–124.
Lev, Baruch (1989). “On the usefulness of earnings and earnings research: Lessons and directions
from two decades of empirical research”. In: Journal of Accounting Research 27, pp. 153–192.
Lev, Baruch and Feng Gu (2016). The end of accounting and the path forward for investors and
managers. John Wiley & Sons.
Lev, Baruch and Theodore Sougiannis (1996). “The capitalization, amortization, and value-relevance
of R&D”. In: Journal of Accounting and Economics 21.1, pp. 107–138.
Lev, Baruch and S Ramu Thiagarajan (1993). “Fundamental information analysis”. In: Journal of
Accounting Research 31.2, pp. 190–215.
Levin, Jonathan D (2011). The economics of internet markets. Tech. rep. National Bureau of Eco-
nomic Research.
Lewellen, Jonathan (2010). “Accounting anomalies and fundamental analysis: An alternative view”.
In: Journal of Accounting and Economics 50.2-3, pp. 455–466.
Li, Feng (2011). “Earnings quality based on corporate investment decisions”. In: Journal of Ac-
counting Research 49.3, pp. 721–752.
Liebowitz, Stan J and Stephen E Margolis (1994). “Network externality: An uncommon tragedy”.
In: Journal of Economic Perspectives 8.2, pp. 133–150.
Liu, Yukun, Jinfei Sheng, and Wanyi Wang (2019). Do Digital Coins Have Fundamental Val-
ues? Evidence from Machine Learning. Tech. rep. Working paper. Available at: pdfs. seman-
ticscholar. org.
Liu, Yukun and Aleh Tsyvinski (2021). “Risks and returns of cryptocurrency”. In: The Review of
Financial Studies 34.6, pp. 2689–2727.
Liu, Yukun, Aleh Tsyvinski, and Xi Wu (2021). “Common Risk Factors in Cryptocurrency”. In:
Journal of Finance.
Loughran, Tim and Jay Ritter (2004). “Why has IPO underpricing changed over time?” In: Finan-
cial Management, pp. 5–37.
Lucca, David O and Emanuel Moench (2015). “The pre-FOMC announcement drift”. In: The Jour-
nal of Finance 70.1, pp. 329–371.
Luo, Xueming, Jie Zhang, and Wenjing Duan (2013). “Social media and firm equity value”. In:
Information Systems Research 24.1, pp. 146–163.
Makarov, Igor and Antoinette Schoar (2020). “Trading and arbitrage in cryptocurrency markets”.
In: Journal of Financial Economics 135.2, pp. 293–319.
Meursault, Vitaly, Pierre Jinghong Liang, Bryan Routledge, and Madeline Scanlon (2021). “PEAD.
txt: Post-Earnings-Announcement Drift Using Text”. In: Available at SSRN 3778798.
Mitchell, Matthew F and Andrzej Skrzypacz (2006). “Network externalities and long-run market
shares”. In: Economic Theory 29.3, pp. 621–648.
46

Pagnotta, Emiliano (2020). “Decentralizing money: Bitcoin prices and blockchain security”. In:
Review of Financial Studies (forthcoming).
Pagnotta, Emiliano and Andrea Buraschi (2018). “An equilibrium valuation of bitcoin and decen-
tralized network assets”. In: Available at SSRN 3142022.
Pakes, Ariel (1985). “On patents, R & D, and the stock market rate of return”. In: Journal of
Political Economy 93.2, pp. 390–409.
Pástor, L’uboš and Pietro Veronesi (2009). “Technological revolutions and stock prices”. In: Amer-
ican Economic Review 99.4, pp. 1451–83.
Patatoukas, Panos N, Richard G Sloan, and Jenny Zha (2015). “On the pricing of mandatory
DCF disclosures: Evidence from oil and gas royalty trusts”. In: The Accounting Review 90.6,
pp. 2449–2482.
Peng, Lin and Wei Xiong (2006). “Investor attention, overconfidence and category learning”. In:
Journal of Financial Economics 80.3, pp. 563–602.
Rajgopal, Shivaram, Mohan Venkatachalam, and Suresh Kotha (2003). “The value relevance of
network advantages: The case of e–commerce firms”. In: Journal of Accounting Research 41.1,
pp. 135–162.
Ramakrishnan, Ram TS and Jacob K Thomas (1998). “Valuation of permanent, transitory, and
price-irrelevant components of reported earnings”. In: Journal of Accounting, Auditing & Fi-
nance 13.3, pp. 301–336.
Richardson, Scott, Irem Tuna, and Peter Wysocki (2010). “Accounting anomalies and fundamental
analysis: A review of recent research advances”. In: Journal of Accounting and Economics
50.2-3, pp. 410–454.
Ritter, Jay R and Ivo Welch (2002). “A review of IPO activity, pricing, and allocations”. In: The
Journal of Finance 57.4, pp. 1795–1828.
Rochet, Jean-Charles and Jean Tirole (2003). “Platform competition in two-sided markets”. In:
Journal of the European Economic Association 1.4, pp. 990–1029.
Rohlfs, Jeffrey (1974). “A theory of interdependent demand for a communications service”. In:
The Bell Journal of Economics and Management Science, pp. 16–37.
Roll, Richard (1986). “The hubris hypothesis of corporate takeovers”. In: Journal of Business,
pp. 197–216.
Rysman, Marc (2004). “Competition between networks: A study of the market for yellow pages”.
In: The Review of Economic Studies 71.2, pp. 483–512.
— (2007). “An empirical analysis of payment card usage”. In: The Journal of Industrial Eco-
nomics 55.1, pp. 1–36.
— (2009). “The economics of two-sided markets”. In: Journal of Economic Perspectives 23.3,
pp. 125–43.
Sadka, Ronnie (2006). “Momentum and post-earnings-announcement drift anomalies: The role of
liquidity risk”. In: Journal of Financial Economics 80.2, pp. 309–349.
47

Saloner, Garth and Andrea Shepard (1995). “Adoption of Technologies with Network Effects: An
Empirical Examination of the Adoption of Automated Teller Machines”. In: The Rand Journal
of Economics, pp. 479–501.
Schipper, Katherine and Linda Vincent (2003). “Earnings quality”. In: Accounting Horizons 17,
pp. 97–110.
Schwenkler, Gustavo and Hannan Zheng (2020). “Peer co-movement in crypto markets”. In: Avail-
able at SSRN 3572471.
Scott, William Robert and Patricia C O’Brien (2019). Financial accounting theory. Vol. 3. prentice
hall Toronto.
Shams, Amin (2020). “The structure of cryptocurrency returns”. In: Fisher College of Business
Working Paper 2020-03, p. 011.
Shy, Oz (2001). The economics of network industries. Cambridge university press.
— (2011). “A short survey of network economics”. In: Review of Industrial Organization 38.2,
pp. 119–149.
Sockin, Michael and Wei Xiong (2020). A model of cryptocurrencies. Tech. rep. National Bureau
of Economic Research.
Soska, Kyle, Jin-Dong Dong, Alex Khodaverdian, Ariel Zetlin-Jones, Bryan Routledge, and Nico-
las Christin (2021). “Towards Understanding Cryptocurrency Derivatives: A Case Study of
BitMEX”. In: Proceedings of the Web Conference 2021, pp. 45–57.
Teoh, Siew Hong and Tak J Wong (1993). “Perceived auditor quality and the earnings response
coefficient”. In: Accounting Review, pp. 346–366.
Trueman, Brett, MH Franco Wong, and Xiao-Jun Zhang (2000). “The eyeballs have it: Searching
for the value in Internet stocks”. In: Journal of Accounting Research, pp. 137–162.
— (2001). “Back to basics: Forecasting the revenues of Internet firms”. In: Review of Accounting
Studies 6.2, pp. 305–329.
— (2003). “Anomalous stock returns around internet firmsâ earnings announcements”. In: Journal
of Accounting and Economics 34.1-3, pp. 249–271.
Vuolteenaho, Tuomo (2002). “What drives firm-level stock returns?” In: The Journal of Finance
57.1, pp. 233–264.
Welch, Ivo and Amit Goyal (2008). “A comprehensive look at the empirical performance of equity
premium prediction”. In: The Review of Financial Studies 21.4, pp. 1455–1508.
Weyl, E Glen (2010). “A price theory of multi-sided platforms”. In: American Economic Review
100.4, pp. 1642–72.
48

Online Appendix
Figure A.1: Cryptocurrency Prices and New Addresses
This figure gives a graphical illustration of the relationship between prices and new addresses for
Bitcoin, Ethereum, and Litecoin.

Figure A.2: Ball and Brown (1968) Graph
The source of the graph is Ball and Brown (1968).

Figure A.3: Cumulative Abnormal Return around New Address Information – Controlling
Transfer Volume Growth
This figure plots the cumulative average logged abnormal return around the release of new address
information for positive and negative new address growth subgroups controlling for changes in
logged transfer volume. Dashed lines represent the 5-percent confidence intervals.

Table A.1: Abnormal Volatility
This table shows the results for the cross-sectional abnormal volatility tests. Panel A reports re-
sults where the abnormal return volatility is calculated using raw returns and Panel B reports results
where the abnormal return volatility is calculated using the CAPM model adjusted returns. Abnor-
mal return volatility is calculated as the difference of the logged standard deviation of returns of
the week and the logged average standard deviation of returns of the previous four weeks. Time
fixed effects are included for all specifications. Standard errors are double clustered at the coin
level and time level. Variable definition is documented in the Online Appendix. *, **, and ***
Panel A (1) (2) (3) (4) (5) (6)
avar 1 avar 1 avar 1 avar 1 avar 1 avar 1
|∆na | 0.081*** 0.082*** 0.085*** 0.077*** 0.081*** 0.081***

(4.330) (4.321) (4.728) (4.027) (4.103) (4.352)
|∆balance1| 0.110 0.085
(0.678) (0.536)
|∆balance2| 0.052*** 0.050***
(3.805) (3.770)
|∆tr ans | 0.008 0.010
(0.805) (1.028)
|∆curr ent | 0.002 -0.032
(0.018) (-0.175)
Cons -0.114*** -0.116*** -0.118*** -0.117*** -0.114*** -0.123***
(-17.575) (-17.402) (-18.713) (-14.095) (-17.531) (-14.027)
Obs 13,137 12,765 12,671 13,115 13,137 12,481

Adj R 2 0.283 0.277 0.286 0.283 0.283 0.283
Panel B (1) (2) (3) (4) (5) (6)

avar 2 avar 2 avar 2 avar 2 avar 2 avar 2
|∆na | 0.124*** 0.124*** 0.123*** 0.116*** 0.122*** 0.115***

(4.861) (4.828) (4.681) (4.607) (4.757) (4.349)
|∆balance1| 0.203 0.155
(0.873) (0.686)
|∆balance2| 0.075*** 0.071***
(2.883) (2.800)
|∆tr ans | 0.013 0.013
(1.284) (1.268)
|∆curr ent | 0.112 0.036
(0.446) (0.133)
Cons -0.137*** -0.139*** -0.140*** -0.143*** -0.137*** -0.148***
(-15.773) (-15.510) (-15.616) (-14.442) (-15.800) (-14.142)
Obs 11,597 11,255 11,244 11,592 11,597 11,086

Adj R 2 0.146 0.141 0.146 0.147 0.146 0.143

Table A.2: Portfolio Sorts – First Half and Second Half
This table shows portfolio sorting results based on cryptocurrency price-to-new address ratios for
the first half and the second half of the sample. Mean is the average weekly excess returns of the
portfolios and the long-short strategy. SD is the standard deviation of the weekly excess returns of
the portfolios and the long-short strategy. SR is the weekly Sharpe ratio of the excess returns of the
portfolios and the long-short strategy. Variable definition is documented in the Online Appendix.
Panel A: First Half
Mean -0.002 -0.020 -0.016 -0.021* 0.018**

(-0.212) (-1.589) (-1.291) (-1.965) (2.619)
SD 0.104 0.123 0.121 0.107 0.070
SR -0.019 -0.163 -0.132 -0.196 0.257
Panel B: Second Half

Mean 0.036** 0.045** 0.034* 0.017 0.019**

(2.572) (2.456) (1.978) (1.531) (2.525)
SD 0.123 0.157 0.148 0.098 0.065
SR 0.293 0.287 0.230 0.173 0.292

Table A.3: Portfolio Sorts – Tercile and Quintile
This table shows portfolio sorting results based on cryptocurrency price-to-new address ratios us-
ing tercile and quintile portfolios. Mean is the average weekly excess returns of the portfolios and
the long-short strategy. SD is the standard deviation of the weekly excess returns of the portfolios
and the long-short strategy. SR is the weekly Sharpe ratio of the excess returns of the portfolios
and the long-short strategy. Variable definition is documented in the Online Appendix. *, **, and
Panel A: Tercile
Low 2 High Low–High
Mean 0.013 0.012 -0.003 0.016***

(1.552) (1.111) (-0.316) (3.122)
SD 0.113 0.137 0.106 0.066
SR 0.115 0.088 -0.028 0.242
Panel B: Quintile
Low 2 3 4 High Low–High
Mean 0.012 0.008 0.016 0.005 -0.004 0.016***

(1.422) (0.868) (1.464) (0.456) (-0.451) (2.841)
SD 0.114 0.128 0.143 0.134 0.107 0.073
SR 0.105 0.063 0.112 0.037 -0.037 0.219

Table A.4: Portfolio Sorts – Exclude Bitcoin
This table shows portfolio sorting results based on cryptocurrency price-to-new address ratios ex-
cluding Bitcoin. Mean is the average weekly excess returns of the portfolios and the long-short
strategy. SD is the standard deviation of the weekly excess returns of the portfolios and the long-
short strategy. SR is the weekly Sharpe ratio of the excess returns of the portfolios and the long-
short strategy. Variable definition is documented in the Online Appendix. *, **, and *** represent
significance level at the 10%, 5%, and 1% level.
Panel A
Mean 0.009 0.013 0.008 -0.004 0.013**

(0.886) (1.170) (0.726) (-0.537) (2.424)
SD 0.133 0.147 0.138 0.105 0.072
SR 0.068 0.088 0.058 -0.038 0.181
Low 0.004 (1.181) 1.096*** (42.285) 0.913

2 0.008 (1.309) 1.076*** (20.984) 0.721
3 0.003 (0.469) 1.002*** (20.480) 0.712
High -0.008* (-1.958) 0.771*** (21.554) 0.732
Low–High 0.012** (2.494) 0.325*** (8.057) 0.276
Low 0.006** (2.002) 1.086*** (45.745) -0.054 (-1.548) -0.151*** (-5.975) 0.928
2 0.006 (1.024) 1.066*** (23.157) 0.430*** (6.374) -0.001 (-0.019) 0.779
3 0.001 (0.230) 0.989*** (22.764) 0.400*** (6.286) -0.056 (-1.218) 0.776
High -0.007* (-1.859) 0.760*** (22.322) 0.110** (2.216) -0.116*** (-3.214) 0.761
Low–High 0.013*** (2.786) 0.326*** (8.202) -0.164*** (-2.820) -0.035 (-0.816) 0.309

Table A.5: Portfolio Sorts Based on Cryptocurrency Price-to-New Address Ratio – Equal Weighted
This table shows portfolio sorting results based on cryptocurrency price-to-new address ratios with
equal-weighting. Panel A reports the means, standard deviations, and Sharpe ratios of the value-
weighted portfolios excess returns. Panel B reports the factor exposures of the portfolios sorted on
price-to-new address ratios. The factor models in include the cryptocurrency market factor and the
cryptocurrency three factor model. Variable definition is documented in the Online Appendix. *,
**, and *** represent significance level at the 10%, 5%, and 1% level.
Panel A
Mean 0.016 0.013 0.013 -0.001 0.016***

(1.601) (1.218) (1.329) (-0.118) (3.636)
SD 0.129 0.138 0.125 0.108 0.059
SR 0.124 0.094 0.104 -0.009 0.271
Low 0.006 (0.907) 0.907*** (14.893) 0.565

2 0.002 (0.372) 1.002*** (16.194) 0.607
3 0.003 (0.543) 0.919*** (16.487) 0.615
High -0.009* (-1.936) 0.817*** (18.069) 0.658
Low–High 0.016*** (3.449) 0.092** (2.194) 0.028
Low -0.001 (-0.145) 1.028*** (28.373) 0.842*** (17.384) -0.024 (-0.678) 0.856
2 -0.004 (-1.261) 1.146*** (37.961) 0.939*** (23.089) -0.011 (-0.366) 0.913
3 -0.002 (-0.767) 1.045*** (35.447) 0.823*** (20.721) -0.013 (-0.471) 0.900
High -0.012*** (-4.105) 0.898*** (31.431) 0.572*** (14.843) -0.088*** (-3.194) 0.872
Low–High 0.013*** (3.018) 0.143*** (3.581) 0.301*** (5.596) 0.060 (1.554) 0.180

Table A.6: Equity Factors Exposures
This table shows exposures to the equity factor models for the portfolios based on cryptocurrency price-to-new address
ratios. The equity factor models are the CAPM, Fama-French 3-factor, and the Fama-French 5-factor models. *, **, and
Alpha MKTRF SMB HML RMW CMA R2
Low 0.012 (1.388) 0.855** (2.563) 0.037

2 0.004 (0.360) 1.480*** (3.614) 0.071
3 0.001 (0.146) 1.373*** (3.515) 0.068
High -0.007 (-0.940) 0.963*** (3.168) 0.056
Low–High 0.019*** (3.669) -0.110 (-0.543) 0.002
Low 0.012 (1.450) 0.749** (2.177) 0.652 (1.098) 0.234 (0.586) 0.051
2 0.003 (0.313) 1.352*** (3.192) 0.891 (1.219) -0.282 (-0.573) 0.080
3 0.002 (0.206) 1.376*** (3.395) -0.079 (-0.113) 0.297 (0.630) 0.070
9
High -0.007 (-0.839) 0.918*** (2.919) 0.234 (0.431) 0.342 (0.936) 0.065
Low–High 0.019*** (3.611) -0.170 (-0.815) 0.419 (1.162) -0.110 (-0.455) 0.010
Low 0.014 (1.566) 0.629* (1.689) 0.393 (0.575) 0.679 (1.148) -0.550 (-0.552) -1.084 (-0.864) 0.057
2 0.004 (0.393) 1.352*** (2.946) 0.596 (0.707) 0.001 (0.001) -0.904 (-0.732) -0.001 (-0.001) 0.083
3 0.003 (0.269) 1.238*** (2.820) -0.159 (-0.198) 0.600 (0.859) 0.030 (0.025) -1.241 (-0.838) 0.074
High -0.005 (-0.653) 0.839** (2.471) -0.146 (-0.234) 0.835 (1.546) -1.004 (-1.099) -0.713 (-0.622) 0.074
Low–High 0.019*** (3.515) -0.213 (-0.943) 0.547 (1.318) -0.163 (-0.454) 0.479 (0.788) -0.384 (-0.504) 0.015

Table A.7: Cryptocurrency Market Factor Excluding Bitcoin
This table shows portfolio sorting results controlling for the cryptocurrency CAPM model and the
cryptocurrency 3-factor model, where the cryptocurrency market factor is constructed as the value-
weighted cryptocurrency returns excluding bitcoin over the risk-free rate (CMKT2). Mean is the
average weekly excess returns of the portfolios and the long-short strategy. SD is the standard
deviation of the weekly excess returns of the portfolios and the long-short strategy. SR is the
weekly Sharpe ratio of the excess returns of the portfolios and t0he long-short strategy. Variable
10%, 5%, and 1% level.
Alpha CMKT2 CSIZE CMOM R2
Low 0.009** (2.211) 0.850*** (23.482) 0.763

2 0.003 (0.534) 1.005*** (18.769) 0.674
3 0.001 (0.157) 0.990*** (20.944) 0.721
High -0.008** (-1.994) 0.771*** (21.613) 0.733
Low–High 0.018*** (3.581) 0.090** (2.045) 0.024
Low 0.012*** (3.176) 0.851*** (25.642) -0.278*** (-5.746) -0.120*** (-3.388) 0.805
2 0.003 (0.510) 0.992*** (19.480) 0.265*** (3.549) -0.099* (-1.828) 0.711
3 0.000 (0.026) 0.976*** (22.863) 0.334*** (5.341) -0.092** (-2.031) 0.776
High -0.008* (-1.895) 0.760*** (22.398) 0.110** (2.220) -0.117*** (-3.243) 0.763
Low–High 0.020*** (4.422) 0.098** (2.485) -0.376*** (-6.515) -0.006 (-0.135) 0.231
10

Table A.8: Fama-MacBeth Regression Results – New Address Growth
This table shows the results for predicting next period new address growths using the price-to-new
address ratios. The regressions are based on Fama-MacBeth regression method. Variable definition
is documented in the Online Appendix. *, **, and *** represent significance level at the 10%, 5%,
and 1% level.
(1) (2) (3) (4) (5)
∆na+1 ∆na+1 ∆na+1 ∆na+1 ∆na+1
pa 0.090*** 0.078*** 0.091*** 0.088*** 0.078***

(16.205) (13.516) (15.988) (15.951) (13.081)
beta 0.006 0.007
(0.588) (0.732)
size 0.002 0.002
(1.031) (0.812)
mom 0.063 0.043
(1.554) (0.950)
Constant -1.116*** -0.973*** -1.173*** -1.081*** -1.005***
(-16.455) (-14.054) (-12.322) (-15.969) (-10.341)
Obs 13,343 12,337 13,343 13,343 12,337

R-squared 0.074 0.080 0.087 0.099 0.117
11

Table A.9: Summary Statistics of Other Variables
This table shows the summary statistics of the other on-chain variables. Panel A reports the mean,
standard deviation, 10% percentile, median and 90% percentile values of the variables. Panel B
reports the correlation matrix of the variables.
Panel A Mean SD 10% Median 90%
∆bal1 0.002 0.046 -0.007 0.000 0.011

∆bal2 0.022 0.591 -0.014 0.000 0.077
∆trans 0.008 1.002 -0.889 -0.000 0.931
∆curr 0.012 0.105 -0.000 0.000 0.013
nvt 5.050 2.245 2.617 4.916 7.556
Panel B ∆bal1 ∆bal2 ∆trans ∆curr nvt
∆bal1 1.000
∆bal2 0.008 1.000
∆trans 0.010 0.040 1.000
∆curr 0.034 0.043 -0.001 1.000
nvt -0.015 -0.035 -0.253 -0.100 1.000
12

Table A.10: Value Relevance Analysis – Additional Results
Panel A (1) (2) (3) (4)
r et r et r et r et
∆balance1 0.170
(1.501)
∆balance2 -0.013
(-0.981)
∆tr ans 0.025***
(6.558)
∆curr ent -0.039
(-0.407)
Cons -0.005 -0.003 -0.004 -0.003
(-0.490) (-0.374) (-0.387) (-0.327)
Obs 14,394 14,293 14,148 15,503

Adj R 2 0.017 0.017 0.031 0.018
Panel B (1) (2) (3) (4) (5) (6) (7) (8)

r et r et r et r et r et r et r et r et
∆na 0.084*** 0.083*** 0.081*** 0.091*** 0.101*** 0.087*** 0.079*** 0.078***
(12.060) (11.823) (12.500) (7.733) (4.299) (12.514) (13.085) (12.865)
na 0.001* 0.001
(1.929) (1.456)
∆balance1 0.199 0.310** 0.350**
(1.532) (2.065) (2.327)
∆balance2 -0.027 -0.034 -0.034
(-1.505) (-1.596) (-1.579)
∆tr ans -0.005 0.009*** 0.008***
(-0.340) (4.152) (3.912)
∆curr ent -0.605 -0.006 -0.004
(-0.816) (-0.045) (-0.031)
Cons -0.003 0.009 -0.004 -0.003 -0.005 -0.003 -0.005 0.007
(-0.349) (0.846) (-0.446) (-0.310) (-0.546) (-0.331) (-0.528) (0.623)
Obs 13,522 13,522 13,139 13,027 13,418 13,522 12,759 12,759

Adj R 2 0.085 0.088 0.092 0.106 0.095 0.103 0.143 0.149
13

Table A.11: Value Relevance Analysis – ERC20
(1) (2) (3) (4) (5) (6) (7) (8)
r et r et r et r et r et r et r et r et
∆na 0.079*** 0.078*** 0.078*** 0.079*** 0.079*** 0.080*** 0.077*** 0.076***
(12.140) (12.058) (12.161) (11.757) (12.031) (14.052) (13.094) (12.777)
na 0.001 0.001
(0.905) (1.312)
∆balance1 0.193 0.256 0.298*
(1.468) (1.644) (1.891)
∆balance2 -0.031 -0.038* -0.036
(-1.625) (-1.680) (-1.585)
∆tr ans 0.013*** 0.013*** 0.012***
(4.731) (5.291) (5.033)
∆curr ent 0.120 0.002 0.012
(0.869) (0.013) (0.082)
Cons -0.005 0.004 -0.005 -0.006 -0.005 -0.006 -0.006 0.006
(-0.543) (0.294) (-0.552) (-0.609) (-0.541) (-0.655) (-0.616) (0.472)
Obs 11,867 11,867 11,851 11,567 11,772 11,867 11,479 11,479

Adj R 2 0.081 0.091 0.094 0.096 0.099 0.105 0.141 0.148
14

Table A.12: Subsample with Controls
median and below median persistence, variability, discount rate, and size with controls. 1{X>Med}
is an indicator variable which equals to one if the X ∈ {Per sistence,V ariabilit y, Discount,Size} is
greater than the sample median and zero otherwise. The regressions are based on Fama-MacBeth
regression method. Variable definition is documented in the Online Appendix. *, **, and ***
(1) (2) (3) (4)
ret ret ret ret
X Persistence Variability Discount Size
∆na 0.075*** 0.106*** 0.090*** 0.075***

(11.250) (11.206) (10.525) (11.250)
1 {X>M ed } -0.005* -0.000 0.001 -0.002
(-1.676) (-0.011) (0.474) (-0.586)
∆na ∗ 1 {X>M ed } 0.025** -0.039*** -0.013 0.022**
(2.458) (-4.199) (-1.503) (2.441)
∆balance1 0.359** 0.316** 0.298 0.267*
(2.168) (2.022) (1.508) (1.742)
∆balance2 -0.038* -0.035 -0.017 -0.036*
(-1.800) (-1.560) (-0.708) (-1.700)
∆trans 0.007*** 0.007*** 0.007*** 0.008***
(3.353) (3.330) (3.245) (3.762)
∆current -0.043 -0.007 -0.089 0.032
(-0.255) (-0.053) (-0.709) (0.230)
Constant -0.002 -0.006 -0.003 -0.005
(-0.217) (-0.624) (-0.347) (-0.499)
Observations 12,553 12,655 11,444 12,758

Adj R2 0.161 0.154 0.150 0.166
15

Table A.13: Value Relevance Analysis – Growth
on Fama-MacBeth regression method. The variables are in growth rates instead of logged growth
rates. Variable definition is documented in the Online Appendix. *, **, and *** represent signifi-
cance level at the 10%, 5%, and 1% level.
(1) (2) (3) (4) (5) (6)
Ret Ret Ret Ret Ret Ret
gN A 0.079*** 0.075*** 0.088*** 0.098*** 0.081*** 0.074***

(13.010) (14.372) (6.302) (3.424) (11.565) (13.551)
gBalance1 0.098 0.186
(0.711) (1.226)
gBalance2 -0.007 -0.016
(-0.326) (-0.714)
gTrans -0.012 0.007***
(-0.583) (3.852)
gCurrent -0.780 0.070
(-0.807) (0.629)
Cons 0.006 0.005 0.007 0.003 0.006 0.004
(0.631) (0.527) (0.731) (0.386) (0.658) (0.491)
Obs 13,522 13,139 13,027 13,418 13,522 12,759

Adj R2 0.116 0.122 0.137 0.127 0.123 0.162
16

Table A.14: Value Relevance Analysis – Cross-Sectional
This table shows the results for the cross-sectional value-relevance tests controlling for changes in
logged Google attention. The regressions are based on Fama-MacBeth regression method. Variable
10%, 5%, and 1% level.
(1) (2) (3) (4) (5) (6) (7)
ret ret ret ret ret ret ret
∆na 0.084*** 0.084*** 0.083*** 0.085*** 0.083*** 0.087*** 0.081***

(12.060) (13.444) (13.571) (13.147) (13.288) (13.867) (12.905)
∆ log (Google + 1) 0.013 0.003** 0.012 0.012 0.013 0.003**
(1.361) (2.145) (1.318) (1.320) (1.389) (2.316)
∆balance1 0.271* 0.340**
(1.864) (1.978)
∆balance2 -0.030 -0.044*
(-1.412) (-1.714)
∆trans 0.012*** 0.012***
(4.908) (4.828)
∆current 0.155 0.053
(1.246) (0.419)
Cons -0.003 -0.001 -0.003 -0.001 -0.001 -0.003 -0.004
(-0.349) (-0.106) (-0.334) (-0.150) (-0.146) (-0.261) (-0.477)
Obs 13,522 11,710 11,521 11,435 11,621 11,710 11,181

Adj R2 0.085 0.099 0.106 0.125 0.121 0.118 0.163
17

Table A.15: Value Relevance Analysis – Subsamples
This table reports the results of the cross-sectional value-relevance tests for each year from 2018
to 2021 and for each quarter. The regressions are based on Fama-MacBeth regression method.
Variable definition is documented in the Online Appendix. *, **, and *** represent significance
level at the 10%, 5%, and 1% level.
Panel A (1) (2) (3) (4)
2018 2019 2020 2021
∆na 0.058*** 0.040*** 0.106*** 0.183***

(4.599) (6.426) (7.679) (12.439)
Cons -0.040* -0.007 0.013 0.041
(-1.830) (-0.670) (0.818) (1.373)
Obs 2,434 3,832 4,733 2,523

Adj R2 0.068 0.056 0.102 0.144
Panel B (1) (2) (3) (4)

Q1 Q2 Q3 Q4
∆na 0.108*** 0.089*** 0.067*** 0.065***

(7.097) (8.448) (5.181) (3.991)
Cons 0.023 -0.003 -0.015 -0.024
(1.017) (-0.174) (-0.919) (-1.255)
Obs 3,622 3,727 2,932 3,241

Adj R2 0.112 0.101 0.062 0.055
18

Table A.16: Predicting Price-to-New Addresses Ratio
This table shows results of predicting future period price-to-address ratios. Panel A and Panel B
report results for predicting one-period-ahead price-to-address ratios. Panel A does not include the
current period price-to-address ratio as an explanatory variable and Panel B includes the current
period price-to-address ratio as an explanatory variable. Panel C reports results for predicting
two-period-ahead price-to-address ratios. Panel D reports results for predicting three-period-ahead
price-to-address ratios. The regressions are based on Fama-MacBeth regression method. Variable
10%, 5%, and 1% level.
Panel A (1) (2) (3) (4) (5) (6) (7)
pa+1 pa+1 pa+1 pa+1 pa+1 pa+1 pa+1
log (size) 0.019** -0.588***

(2.291) (-5.340)
gini 1.345*** 4.101***
(4.381) (5.403)
bom1 -0.146*** -0.742***
(-20.357) (-6.843)
bom2 -0.008 0.065***
(-0.375) (7.341)
nvt 0.380*** 0.373***
(19.425) (50.465)
com -0.016 -0.145***
(-1.207) (-13.659)
Constant 11.877*** 11.001*** 11.664*** 12.598*** 10.287*** 12.599*** 15.008***
(75.735) (37.329) (318.710) (147.437) (129.920) (148.633) (17.966)
Observations 13,398 13,050 13,011 12,889 13,392 13,398 12,717

Adj R 2 0.010 0.008 0.030 0.009 0.320 -0.002 0.432
Panel B (1) (2) (3) (4) (5) (6) (7) (9)

pa+1 pa+1 pa+1 pa+1 pa+1 pa+1 pa+1 pa+1
pa 0.883*** 0.872*** 0.869*** 0.868*** 0.881*** 0.860*** 0.885*** 0.822***

(97.389) (75.484) (84.154) (86.998) (94.362) (74.970) (98.801) (66.375)
log (size) 0.002 -0.121
(0.419) (-1.596)
gini 0.267 0.726
(1.471) (1.294)
bom1 -0.015*** -0.136*
(-4.051) (-1.786)
bom2 -0.004*** 0.011**
(-3.579) (2.406)
nvt 0.029*** 0.046***
(5.601) (8.798)
com -0.005*** -0.029***
(-3.183) (-4.567)
Constant 1.432*** 1.487*** 1.354*** 1.564*** 1.534*** 1.558*** 1.482*** 3.125***
(13.083) (10.466) (7.229) (12.575) (13.110) (12.951) (13.283) (5.385)
Observations 13,325 13,323 12,947 12,945 12,825 13,324 13,323 12,653

Adj R 2 0.777 0.778 0.758 0.758 0.771 0.783 0.777 0.780
19

Panel C (1) (2) (3) (4) (5) (6) (7) (8)
pa 0.847*** 0.836*** 0.829*** 0.828*** 0.848*** 0.822*** 0.849*** 0.786***

(88.710) (71.279) (77.152) (79.595) (87.007) (69.996) (90.479) (60.716)
log (size) -0.002 -0.076
(-0.374) (-0.900)
gini 0.308 0.450
(1.465) (0.732)
bom1 -0.015*** -0.086
(-3.887) (-1.042)
bom2 -0.005*** 0.013**
(-3.675) (2.302)
nvt 0.033*** 0.047***
(5.845) (8.235)
com -0.005*** -0.034***
(-3.513) (-4.648)
Constant 1.888*** 2.011*** 1.819*** 2.068*** 1.950*** 2.016*** 1.938*** 3.290***
(16.449) (13.472) (8.418) (16.029) (15.930) (16.253) (16.715) (4.782)
Observations 13,208 13,205 12,833 12,830 12,715 13,207 13,205 12,544

Adj R2 0.731 0.733 0.711 0.709 0.729 0.738 0.732 0.740
Panel D (1) (2) (3) (4) (5) (6) (7) (8)

pa 0.839*** 0.828*** 0.821*** 0.820*** 0.838*** 0.818*** 0.842*** 0.781***

(87.117) (70.396) (74.646) (77.963) (86.015) (69.178) (89.105) (62.611)
log (size) -0.003 -0.002
(-0.582) (-0.026)
gini 0.231 -0.120
(1.139) (-0.193)
bom1 -0.014*** -0.012
(-3.087) (-0.137)
bom2 -0.004*** 0.016***
(-3.237) (2.612)
nvt 0.027*** 0.040***
(5.264) (7.680)
com -0.005*** -0.037***
(-3.162) (-4.749)
Constant 1.992*** 2.131*** 2.002*** 2.178*** 2.075*** 2.105*** 2.030*** 2.962***
(17.128) (13.926) (9.323) (16.628) (17.223) (16.675) (17.858) (4.267)
Observations 13,090 13,087 12,718 12,715 12,606 13,089 13,087 12,436

Adj R2 0.714 0.717 0.691 0.691 0.707 0.718 0.715 0.717
20

Variable Definitions
• ret: Logged cryptocurrency return
• na: Logged number of new address per coin
• ∆na: Weekly changes in logged number of new address per coin

Price

• pa: Logged price-to-address ratio, or log NA
• balance1: Logged percentage of supply held by the top 1% addresses
• balance2: Logged total amount of coins held on exchange addresses
• trans: Logged transaction volume
• current: Logged total amount of circulating supply
• log (size): Logged market capitalization
• gini: Gini index of coin holding
21

Liu - Accounting For Crypto Value

Uploaded by

Copyright:

Available Formats

Liu - Accounting For Crypto Value

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Liu - Accounting For Crypto Value

Uploaded by

Copyright:

Available Formats

Accounting for Cryptocurrency Value

Yukun Liu, Aleh Tsyvinski, Xi Wu∗

Information pertaining to new addresses is highly value-relevant in the cryptocurrency mar-

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

2 Value Relevance and Cryptocurrencies

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Table 1: Summary Statistics

ret 0.003 0.251 -0.214 -0.000 0.210

Panel B ret ∆na na pa

rt,i = α + β × ∆nat,i + t,i (1)

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Table 2: Value Relevance Analysis – Time-Series

∆na 0.267*** 0.265*** 0.150*** 0.151*** 0.268*** 0.267***

Obs 566 566 303 303 419 419

4.1 Baseline Results

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

7 See Lev (1989) and Kothari (2001) for discussions.

Electronic copy available at: https://ssrn.com/abstract=3951514

Table 3: Value Relevance Analysis – Cross-Sectional

∆na 0.084*** 0.083***

Obs 13,522 13,522 13,404 13,404 12,163 12,163

Adjusted for Expected Returns

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

∆na 0.087*** 0.086*** 0.084*** 0.083***

Obs 12,053 12,053 12,053 12,053

4.2 Economic Determinants of Value-Relevance

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Table 5: Persistence of New Address Growth and Value Relevance

∆na 0.077*** 0.105*** 0.077***

Observations 6,644 6,537 13,181

Electronic copy available at: https://ssrn.com/abstract=3951514

∆na 0.123*** 0.073*** 0.123***

Observations 6,709 6,587 13,296

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Table 7: Discount Rate and Value Relevance

∆na 0.096*** 0.083*** 0.096***

Observations 6,022 5,924 11,946

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

∆na 0.080*** 0.103*** 0.080***

Observations 6,613 6,906 13,519

5 Discrete and Continuous Revelation of Information

Electronic copy available at: https://ssrn.com/abstract=3951514

Electronic copy available at: https://ssrn.com/abstract=3951514

Lafond, Olsson, and Schipper 2007).

Electronic copy available at: https://ssrn.com/abstract=3951514

∆na -0.005 0.005 -0.005 -0.006 0.000 -0.000

rt,i = α + β × ∆nat,i + t,i (1)

∆na 0.267* 0.265* 0.150* 0.151* 0.268* 0.267*

∆na 0.084* 0.083*

∆na 0.087* 0.086* 0.084* 0.083*

∆na 0.077* 0.105* 0.077***

∆na 0.123* 0.073* 0.123***

∆na 0.096* 0.083* 0.096***

∆na 0.080* 0.103* 0.080***

Low 0.015* 0.017 0.018 0.014 0.016* 0.019** 0.017* 0.017*

C1 Alpha 0.016* 0.015* 0.009* 0.009* 0.004 0.005 0.005 -0.000

C3 Alpha 0.019* 0.017* 0.010 0.010 0.006 0.008* 0.009* 0.004

pa -0.004* -0.003* -0.003* -0.003 -0.003**

|∆na | 0.081* 0.082* 0.085* 0.077* 0.081* 0.081*

|∆na | 0.124* 0.124* 0.123* 0.116* 0.122* 0.115*