Risk-Return Relation of Cryptocurrency Carry Trade

Zhenzhen Fan∗ Feng Jiao† Lei Lu‡ Xin Tong§

This version: February 17, 2023

Abstract

This paper comprehensively examines the risk-return relation of cryptocurrency

carry trade using realistic borrowing and lending interest rates. We find signifi-
cant violations of the uncovered interest rate parity in the cryptocurrency market.
The cross-sectional carry trade strategy yields an annualized return of 46.71% and
a Sharpe ratio of 0.77. Unlike fiat-currency carry trade which is vulnerable to
crash risk, the cryptocurrency carry trade is resistant to the cryptocurrency mar-
ket crashes in 2018 and 2021. We show that the crypto-carry trade returns cannot
be explained by established risk factors from fiat currencies or cryptocurrencies.
We find that geopolitical risk explains a substantial amount of the carry returns.
Interestingly, the risk-adjusted crypto-carry return alpha is significantly negative,
implying a negative cryptocurrency carry risk premium.

JEL Classification:
Keywords: Cryptocurrency, Carry trade, Uncovered interest rate parity, Geopo-
litical risk

∗
Asper School of Business, University of Manitoba. E-mail: [email protected].
†
Dhillon School of Business, University of Lethbridge. E-mail: [email protected].
‡
Asper School of Business, University of Manitoba. E-mail: [email protected].
§
Asper School of Business, University of Manitoba. E-mail: [email protected].

1
1 Introduction

Cryptocurrency has emerged as a new investment vehicle in recent years, offering novel
opportunities to investors. This study explores the relation between risk and return in
the cryptocurrency carry trade. Previous research on fiat currencies shows that carry
trade can deliver positive returns by leveraging the differential in interest rates between
low- and high-yielding currencies. The study by Koijen et al. (2018) broads the scope of
the carry trade concept to other asset classes and finds that this strategy delivers positive
returns in global equity, commodity, bond, options, and other markets. Examining the
cryptocurrency carry trade not only expands our understanding of the carry trade but
also reveals new investment prospects for investors. Furthermore, understanding the risk-
based drivers of cryptocurrency carry returns sheds light on how cryptocurrency differs
from traditional assets in terms of return determination and assists investors in managing
their portfolio risk more effectively.
We begin our examination by assessing the validity of the uncovered interest rate par-
ity (UIP) hypothesis in the cryptocurrency market (Keynes, 1923). To do so, we employ
the UIP regression methodology outlined by Fama (1984) and examine the interest rate
differential between cryptocurrency and US Dollar, as defined by Koijen et al. (2018).
Our findings reveal that this carry factor has a positive impact on cryptocurrency’s ex-
cess returns in the time-series, indicating that UIP does not hold in the cryptocurrency
market. Both the observed interest rate and the futures-implied interest rate of cryp-
tocurrency are tested to validate the UIP hypothesis, ultimately confirming its violation
in the cryptocurrency market.
The empirical tests of UIP provide evidence in support of a positive return for a
time-series carry trade between cryptocurrency and US Dollar. Next, we explore the
returns generated by a time-series carry trade between stablecoins and US Dollar. This
popular trading strategy involves borrowing US dollars, exchanging them for stablecoins,
which are pegged to the US dollar, and depositing them in online platforms to earn a
higher interest rate. We utilize two widely-used stablecoins, DAI and USDT(tether),
in our examination of the carry trade returns between stablecoins and U.S. Dollar. Our

1
findings show that this strategy generates an annualized return of approximately 9% with
a Sharpe ratio as high as 3.47.
We subsequently examine the performance of cross-sectional carry trade beyond the
stablecoins. We observe that by going long on cryptocurrencies with high interest rates
and shorting those with low interest rates, cross-sectional carry trade can generate sub-
stantial positive carry returns. We categorize the cryptocurrencies in our sample based
on their carry, dividing them into three equal groups. We hold each portfolio for one week
and rebalance it at the end of each week.1 The carry return is calculated as the return of
the highest-interest cryptocurrencies minus the return of the lowest-interest group. This
strategy results in an annualized mean carry return of 46.71%, a standard deviation of
60.68%, and a Sharpe ratio of 0.77.
We then compare the performance of the carry portfolio with that of a passively con-
structed, equally weighted (EW) cryptocurrency portfolio. By scaling the carry portfolio
to match the volatility of the EW portfolio, we demonstrate that the carry portfolio not
only delivers more consistent performance, but also withstands the significant market
downturns that occurred in 2018 and 2021.
Further examination shows that the carry trade premium of cryptocurrencies cannot
be solely attributed to the three-factor model of Liu et al. (2022), downside or volatility
risk. Instead, the geopolitical risk from Caldara and Iacoviello (2022) seems to play
a major role in driving the carry return. This is a departure from the conventional
understanding of carry trade risk in fiat currency markets. Once the impact of geopolitical
risk is taken into account, the carry premium of cryptocurrency turns negative, suggesting
that it offers some degree of protection against some unknown risk.
Our study is related to carry literature which focuses on fiat-currencies. Our research
demonstrates that the drivers of carry returns in fiat currencies and cryptocurrencies are
vastly distinct. Past studies have indicated that carry trade in fiat currency markets
generates substantial positive returns (Burnside et al., 2011; Lettau et al., 2014; Lustig
et al., 2011). This positive carry premium is attributed to various types of risks, such as
1
We follow the practice in Cong et al. (2022) to do this. Since weekly interest rate data for the U.S.
Dollar is not available on weekends, we rebalance the portfolio on Fridays.

2
carry trade risk (Lustig et al., 2011), global equity market volatility (Lustig et al., 2011),
downside risk (Lettau et al., 2014), global foreign exchange volatility risk (Menkhoff et al.,
2012), peso-event risk (Burnside et al., 2011), jump risk (Lee and Wang, 2019), among
others. However, our findings suggest that the positive carry trade return of cryptocur-
rency can be explained by geopolitical risk factors as noted by Caldara and Iacoviello
(2022), but a significant negative portion of the carry premium remains unaccounted for.
Hence, the cryptocurrency carry trade strategy provides a hedge against some unidenti-
fied risks, which differs from the carry trade of fiat currency.
Our paper is the most comprehensive study on the risk-return relation of cryptocur-
rency carry trade to our knowledge. Several papers also investigate the cryptocurrency
carry trade, but neither the scope nor the coverage of these studies are comparable to
ours. Specifically, we differ from existing cryptocurrency carry studies in the following
aspects. First, we use realistic borrowing and lending interest rates from a major online
trading platform for all cryptocurrencies. Cong et al. (2022) use staking reward rate as
a proxy for cryptocurrency interest rate.2 However, many cryptocurrencies, such as Bit-
coin, cannot be staked so that they cannot be included in their sample. Some studies use
crypto-derivatives to infer interest rates, such as perpetual swaps (Franz and Schmeling,
2021) and futures contract (Christin et al., 2022).3 Crypto-derivatives suffer from liquidity
issues (Greene and McDowall, 2018; Kumar, 2022), especially when the underlying token
has a smaller market share. Moreover, Franz and Valentin (2020) show that the spot and
derivatives market for cryptocurrencies are segregated and conventional economic parities
are often violated. Second, instead of focusing on a few ad hoc tokens (Christin et al.,
2022), our study includes over 40 cryptocurrencies. Our cryptocurrency universe consists
of all actively traded currencies from the Bitfinex platform with non-trivial borrowing
and lending rates, including major tokens like Bitcoin, Etherium as well as stablecoins
like USDT and DAI. Third, we conduct a risk factor analysis of the cryptocurrency carry
2
Staking reward is earned when investors lock their cryptocurrency for a certain period to help verify
transactions on the Blockchain system or maintain the safety of the network (Chitra, 2020).
3
The strategy in Christin et al. (2022) is to short the perpetual futures of Bitcoin and long Bitcoin
as a hedge. Therefore, their strategy is actually an arbitrage strategy based on deviations of covered
interest rate parity. However, the carry trade strategy in our paper is an arbitrage strategy based on
deviations of uncovered interest rate parity.

3
trade from an asset-market view. Existing studies propose explanations from other than
the risk-based perspective. For example, Cong et al. (2022) propose that the positive
carry return comes from transaction convenience. Franz and Schmeling (2021) suggest
that the carry return is caused by capital insufficiency and speculation with high leverage.
We show that the carry trade returns cannot be explained by established risk factors from
either fiat currencies (Lustig et al., 2011) or cryptocurrencies (Liu et al., 2022). We find
that geopolitical risk explains a substantial amount of the carry returns.
The rest of the paper is organized as follows. Section 2 describes the data. Section
3 examines whether UIP condition holds in the time-series in cryptocurrency markets.
Section 4 investigates the carry trade strategy of cryptocurrencies. Section 5 investigates
the possible economic explanations for the carry return. Section 6 provides the results of
robustness check. Section 7 concludes.

2 Data

The cryptocurrency interest rate data was obtained from the Bitfinex platform, which
operates as a peer-to-peer lending and borrowing platform for cryptocurrencies. The
interest rate in this study is the observed deposit and lending rate of cryptocurrencies
on the Bitfinex platform.4 Bitfinex was selected as the source of cryptocurrency interest
rate data due to its status as the largest exchange providing interest rates for Bitcoin
through a limit order book (Franz and Valentin, 2020). The sample period was limited to
October 2016 to December 2021, as the data on interest rates from Bitfinex is considered
reliable only after October 2016 (Franz and Valentin, 2020).
We obtain data on the interest rates of 46 cryptocurrencies from the Bitfinex plat-
form.5 The deposit rate of a cryptocurrency is equal to 0.85 multiplied by the borrowing
4
Investors can conduct margin trading on the Bitfinex platform, which allows them to trade cryp-
tocurrencies with leverage by borrowing cryptocurrencies from others. The interest rate we use is the
borrowing and lending rate of their funding. It should be noted that this interest rate differs from the
staking reward rate, as described by Cong et al. (2022). The latter refers to the rewards received by
investors for locking their cryptocurrency for a set period in order to validate transactions and enhance
network security, as stated in Chitra (2020)
5
We can collect data on interest rates for 49 cryptocurrencies in total from the Bitfinex platform, in-
cluding 4 stablecoins. However, following Cong et al. (2022), we remove the observations in the first week
after the cryptocurrencies were first launched because Cong et al. (2022) propose that there are abnormal

4
rate of the cryptocurrency.6 The interest rate from Bitfinex is at hourly frequency. To
mitigate the impact of outliers, we employ a precautionary measure of replacing hourly
interest rates that exceed the 99th percentile of historical observations with their pre-
ceding value. We then calculate the daily interest rate by taking the volume-weighted
average of hourly observations, as per Franz and Valentin (2020). The final daily interest
rate was utilized to calculate the monthly and weekly interest rates by multiplying it by
30 and 7, respectively.
The summary statistics of the annualized interest rate, calculated as the daily rate
multiplied by 365, are displayed in Table 1. Table 1 shows that the annualized interest
rates for Bitcoin and Ethereum were 8.13% and 6.00%, respectively.
[Insert Table 1 Here]
We collect daily spot prices of 46 cryptocurrencies from coinmarketcap.com as per Liu
et al. (2022). To ensure accuracy, we eliminate the first-week observations following the
launch of each cryptocurrency as suggested by Cong et al. (2022), who posit abnormal
fluctuations in cryptocurrency prices occur during this period. We also replace outliers,
observations larger than the 99th percentile or smaller than the 1st percentile, with the
previous value. Table 2 presents a summary of daily log price changes (log P2 − log P1 )
of cryptocurrencies.
[Insert Table 2 Here]
We acquire the monthly interest rate of the U.S. Dollar from Federal Reserve Economic
Data (FRED) and the weekly interest rate of the U.S. Dollar from the OptionMetrics
database. We source the market, size, and momentum factors of cryptocurrencies from
Liu et al. (2022), the geopolitical risk data from Caldara and Iacoviello (2022), and the
VIX data from the Bloomberg database.
We gather data on Bitcoin and Ethereum futures in the Chicago Mercantile Exchange
(CME) from the Bloomberg database. Our sample period for Bitcoin extends from De-
cember 2017 to May 2022, while for Ethereum, it covers February 2021 to May 2022. To
fluctuations in cryptocurrency prices at this period. After this filter, there are 46 cryptocurrencies left
in the sample.
6
The reason is that the fund providers have to pay a fee equal to 15% of the interest they earned if
the funding is not opened with a hidden offer.

5
ensure high-quality data, we exclude futures prices with missing or zero trading volume
and observations with bid-ask spread larger than 10% of the mid-price. Spot prices for
Bitcoin and Ethereum in Section 6 are sourced from Yahoo Finance to expand the sample
period.

3 Revisit UIP Condition

In this section, we examine the validity of the Uncovered Interest Rate Parity (UIP) in the
cryptocurrency market by utilizing the regression method introduced by Fama (1984).
The significance of testing the UIP lies in the fact that the profitability of the time-
series carry trade strategy is contingent upon its violation. The UIP, originally proposed
by Keynes (1923), suggests that a currency with a higher interest rate is expected to
depreciate relative to a currency with a lower interest rate. Hence, investors should not
be able to earn a profit by borrowing the low-interest currency and investing in the high-
interest currency. The interest rate differential between any pair of currencies cannot
predict the currency excess returns. In light of this, we test whether the interest rate
differential between cryptocurrency and U.S. Dollar can predict the log excess return of
cryptocurrency. We conduct our analysis in both the spot market and the futures market.
In the spot market, we first perform UIP regression on each of the 42 cryptocurrencies
in our sample obtained from the Bitfinex platform.7 We then conduct a panel regression
that includes all 42 cryptocurrencies to test the UIP condition. In the futures market,
Section 6.2 uses the futures implied interest rate differential of Bitcoin and Ethereum to
test the robustness of our findings.
The UIP regression in the spot market of cryptocurrencies is as follows:

rt+1 = α + β(ict − iUt SD ) + ϵt+1 . (1)

where rt+1 is the log excess return of a cryptocurrency from time t to time t+1. ict and
7
It is worth noting that we exclude stablecoins from our analysis, as their prices are pegged to assets
such as the US Dollar, Euro, or gold, and as such, their interest rates are not expected to predict their
excess returns.

6
iUt SD are the interest rates of the cryptocurrency and U.S. Dollar from time t to time
t + 1, respectively. In line with the methodology presented in Fama (1984), the return
and interest rate differential in our regression are expressed as percentage points. The
log excess return of the cryptocurrency is defined as in Lettau et al. (2014):

rt+1 = ict − iUt SD + ∆st+1 . (2)

where st is the log price of the cryptocurrency in terms of the U.S. Dollar. The calcula-
tion of this excess return involves borrowing U.S. Dollars, exchanging them for Bitcoin,
depositing the funds on online platforms to earn interest, and measuring the resulting
return. As per the uncovered interest rate parity (UIP) hypothesis, the excess return
should be negligible since the appreciation or depreciation of cryptocurrencies is deter-
mined by the relationship between their interest rate with that of the U.S. Dollar. If
the cryptocurrency’s interest rate is higher, it is expected to depreciate, and vice versa,
rendering it unprofitable to borrow U.S. Dollars and invest in cryptocurrencies. Hence,
the β estimates in regression equation (1) should approach zero, indicating that ict − iUt SD
cannot predict the future excess returns of cryptocurrencies.
We evaluate the relationship between interest rate differential and cryptocurrency
excess returns using both weekly and monthly returns. The sample period for our analysis
is October 2016 to December 2021, a span of approximately five years. To increase the
number of observations, we perform the analysis at a daily frequency. The process of
calculating the monthly and weekly returns at a daily frequency involves simulating a
daily carry trade strategy. On each day, we borrow US Dollars, exchange them for a
cryptocurrency, deposit the cryptocurrency on the Bitfinex platform for one month (or
one week), exchange the cryptocurrency back to US Dollars, and repay the loan. This
process is repeated daily, yielding monthly (or weekly) excess returns at a daily frequency.
Table 3 shows the results of the UIP regression with monthly returns for each of the
37 cryptocurrencies.8 The results of our regression analysis indicate that the estimates
8
Initially, there are 42 cryptocurrencies in our sample (not including stablecoins). However, 5 of them,
Compound, Elrond, Maker, Polygon, and SHIBA INU are excluded in Table 3 since the observations of
monthly returns for these 5 cryptocurrencies are not enough to do the regression.

7
of β are significantly positive for 9 cryptocurrencies, implying that an increase in their
interest rates relative to the U.S. Dollar is associated with an appreciation in their value,
leading to an increase in their future excess returns. This finding suggests a violation of
the UIP condition for these cryptocurrencies. For 7 cryptocurrencies, the estimates of β
are significantly negative, indicating that an increase in their interest rates relative to the
U.S. Dollar results in a larger depreciation of their value, leading to a decrease in their
excess returns. This result also indicates a violation of the UIP condition for these cryp-
tocurrencies. It is noteworthy that many popular cryptocurrencies, including Bitcoin,
Ethereum, Litecoin, and Solana, all exhibit deviations from the UIP principle. Mean-
while, the estimates of β are insignificant for 21 cryptocurrencies, implying that there is
insufficient evidence of a violation of the UIP condition among these cryptocurrencies.
[Insert Table 3 Here]
Table 4 shows the results of UIP regression with weekly returns for each individual
cryptocurrency. There are 40 cryptocurrencies in this test.9 The estimates of β for
10 cryptocurrencies are significantly positive, 3 are significantly negative, and 27 are
insignificant.
[Insert Table 4 Here]
The results in Table 5 demonstrate that when all 42 cryptocurrencies are considered
together in the panel regression, the estimates of β are significantly positive.10 In par-
ticular, the regression model that accounts for both individual and time fixed effects,
and with monthly returns, shows that a 1% increase in the interest rate differential is
associated with a 1.55% increase in the future excess return of the cryptocurrency. These
findings indicate that the UIP condition does not hold in the 42 cryptocurrency universe.
[Insert Table 5 Here]
In summary, the results from our analysis demonstrate that the uncovered interest
9
SHIBA INU and Maker are not included here. For SHIBA INU, there are not sufficient observations
to do the regression using Newey-West standard error with 3 lags. For Maker, the interest rate varies
too little, whereas the price is too volatile, so that the absolute value of the coefficient is estimated to
be extremely large (-124,917.2), which is not reasonable. Therefore, we do not include Maker, either.
10
We don’t include the 4 stablecoins in the test because their prices are pegged to some other assets
such as U.S. Dollar, Euro or gold so that the price changes are not supposed to be predicted by the
interest rate differential.

8
rate parity (UIP) hypothesis is not upheld in the cryptocurrency market. Specifically,
the interest rate differential between cryptocurrency and the U.S. Dollar can positively
predict future excess returns of cryptocurrencies. An increase in the interest rate of a
cryptocurrency or a decrease in the interest rate of the U.S. Dollar is associated with
an appreciation of the cryptocurrency against the U.S. Dollar. Conversely, a decrease
in the interest rate of a cryptocurrency or an increase in the interest rate of the U.S.
Dollar is associated with a depreciation of the cryptocurrency, resulting in a decrease in
its excess return. In addition, Section 6.2 also performs a robustness check by testing
the UIP hypothesis in the futures market of Bitcoin and Ethereum. The results indicate
that UIP is violated in the Bitcoin futures market, but there is insufficient evidence of
UIP violation in the Ethereum futures market.

4 Carry Trade

In this section, we explore the implementation of various forms of carry trade strategies
using cryptocurrencies. The carry trade strategy involves borrowing currencies with low
interest rates and investing in currencies with high interest rates. Section 4.1 examines
the time-series carry trade between stablecoins and the U.S. Dollar, a popular investment
strategy among investors. Meanwhile, section 4.2 focuses on the cross-sectional carry
trade in the cryptocurrency universe.

4.1 Time-series Carry Trade between Stablecoins and U.S. Dol-

lar

In practice, investors often implement a carry trade strategy of borrowing U.S. Dollars,
exchanging them for stablecoins pegged to the U.S. Dollar, and depositing the stablecoins
on online platforms to earn a profit as the interest rates of stablecoins are usually higher
than those of the U.S. Dollar. This constitutes a carry trade between stablecoins and the
U.S. Dollar, and in this subsection, we investigate the profitability of this strategy.
The described carry trade strategy focuses on two stablecoins, USDT (Tether) and

9
Dai, both of which are pegged to the U.S. Dollar. The direction of the trade depends
on whether the interest rate of the U.S. Dollar is higher or lower than the lending and
deposit rate of the stablecoins. When the interest rate of the U.S. Dollar is lower than
the interest rate of depositing a stablecoin, the strategy involves borrowing U.S. Dollar,
exchanging it for a stablecoin, and depositing it on the Bitfinex platform for a specific
period of time, either one month or one week. If the interest rate of the U.S. Dollar is
higher than the interest rate of borrowing the stablecoin, we execute the trade in reverse
by borrowing the stablecoin, exchanging it into U.S. Dollar, and depositing U.S. Dollar.
If the interest rate of the U.S. Dollar falls between the two, no trade is conducted. The
strategy aims to generate a monthly or weekly carry return, which is calculated at a daily
frequency.
The return of the strategy for each situation is calculated as follows,

ic−dep − iUt SD + ∆st+1 ic−dep > iUt SD


 t t


rt+1 = iUt SD − ic−borw
t − ∆st+1 iUt SD > ic−borw
t



0 ic−dep ≤ iUt SD ≤ itc−borw


t

where ic−dep
t and ic−borw
t are the deposit rate and borrowing rate of a stablecoin, respec-
tively.
Table 6 shows the results.11 The monthly carry return and weekly carry return for
both USDT and Dai are significantly positive at 1 % level. As expected, almost all carry
returns come from the interest rate differential, since the peg between the U.S. Dollar
and stablecoins is largely maintained during the sample period. This strategy offers a
very high Sharpe ratio. For example, the annualized carry return of USDT calculated
from the monthly return is 9.067%, and the Sharpe ratio is as high as 3.469. The result
confirms the profitability of the carry strategy between stablecoins and the U.S. Dollar.
[Insert Table 6 Here]
11
The means, standard deviations and Newey-West standard errors (Newey and West, 1987) are an-
nualized and in percentage points. The Sharpe ratios are annualized and in real numbers. Annualized
mean is the mean of monthly (weekly) return multiplied by 12 (52).
√ Annualized
√ standard deviation is
the standard deviation of monthly (weekly) return multiplied by 12 ( 52). Annualized Sharpe Ratio
is the ratio of the annualized mean and annualized standard deviation.

10
4.2 Cross-sectional Carry Trade

In this subsection, we investigate the cross-sectional carry trade within the 42 cryptocur-
rency universe.12 The strategy involves going long on cryptocurrencies with high interest
rates and shorting cryptocurrencies with low interest rates. We adopt two methods to
construct the cross-sectional carry portfolio: 1) ranking the cryptocurrencies by their
carry (ict − iUt SD ) and dividing the universe into three equal groups, with the carry re-
turn being the difference in return between the highest (group 3) and lowest carry group
(group 1); 2) a method introduced by Asness et al. (2013) that involves using the weight
determined by the rank of carry for each cryptocurrency to form the carry trade portfolio.
The weight for each cryptocurrency is calculated as wti = zt (rank(cit ) − Nt2+1 ), where wti is
the weight of cryptocurrency i at time t in the portfolio, cit is the carry of the cryptocur-
rency at time t, Nt is the total number of cryptocurrencies at time t, and zt is a scalar
that ensures the total weight of long positions is 1 and the total weight of short positions
is -1. A cryptocurrency with a high carry will receive a high weight, and if its carry falls
below the bottom 50th percentile, it will receive a negative weight (for shorting).
We rank cryptocurrencies by carry and construct the carry portfolio at the end of
each week (Friday), holding for one week and rebalancing the carry portfolio next Friday.
The return of an individual cryptocurrency is calculated as follows. If a cryptocurrency
is in the long position,

rt+1 = itc−dep − iUt SD + ∆st+1 . (3)

If a cryptocurrency is in the short position,

rt+1 = iUt SD − itc−borw − ∆st+1 . (4)

where ic−dep
t and ic−borw
t are the interest rate of depositing and borrowing the cryptocur-
rency, respectively. st is log price of the cryptocurrency.
In the first method for forming the cross-sectional carry portfolio, we calculate the
12
We don’t include stablecoins in this test for the same reason as that in section 3.

11
average return, interest rate differential, and price change for each group by taking the
equal-weighted average of the individual cryptocurrencies within the group. The carry
return is found by subtracting the return of the group with the lowest carry from the
return of the group with the highest carry. For the second method, the carry return is
computed as the weighted average return of all cryptocurrencies, based on the weight
determined by the rank of carry for each cryptocurrency as described by Asness et al.
(2013).13 As suggested by Cong et al. (2022), we exclude the observations from the first
week after the launch of each cryptocurrency, due to the presence of abnormal price
fluctuations during this period.
Table 7 shows the results. The mean, standard deviation, and Sharpe ratio are an-
nualized and reported in percentage points.14 We multiply the weekly mean by 52 and
√
weekly standard deviation by 52 to calculate the annualized mean and standard devi-
ation. The Sharpe ratio is the annualized mean excess return divided by the annualized
standard deviation of excess return. The returns are quite large and volatile compared
with those of fiat currencies reported in Table 1 of Lustig et al. (2011).
For the first method of constructing the cross-sectional carry portfolio, the cryptocur-
rencies appreciate on average for all three groups. The level of appreciation grows as the
carry increases from group 1 to group 3, with group 3 seeing the largest appreciation
of 74.11% per year. The average carry for each group is -0.13%, 2.87%, and 18.99%,
respectively. The excess return for each group is the sum of its carry and price appreci-
ation, with group 3 having the highest excess return of 93.102% due to its highest carry
and appreciation. The average excess returns for groups 1 and 2 are 46.39% and 60.14%
respectively. This indicates that carry can predict the excess return of cryptocurrencies
in the cross-section, with higher carry leading to higher excess return. The carry return,
or the difference between the excess return of group 3 and group 1, is 46.71% per year
and significant at the 5% level. The annualized standard deviation is 60.67%, and the
13
For cryptocurrencies with negative weight, we take its absolute value when calculate the weighted
average return because Equation (4) already takes the negative sign into account.
14
In method 1, since all cryptocurrencies in Group 1 are in a short position, we use Equation (4)
to calculate the returns. We report the negative value of the returns, price change, and interest rate
differential of Group 1 to make an easier comparison with those of Group 2 and Group 3, in which the
cryptocurrencies are all in the long position.

12
annualized Sharpe ratio is 0.770. The difference in carry between groups 1 and 2 is small,
so the difference in their excess returns is only 13.744%, which is not significant at the
10% level.
[Insert Table 7 Here]
Using the second method, the mean annualized carry return is 35.911%, significant
P j j
w r = wj >0 wj rj −
P
at the 10% level. The return of the carry portfolio rcarry =
j j
P
wj <0 |w | r . It means that cryptocurrencies with high carry have a larger return than

those with low carry, consistent with the previous result that carry can positively predict
excess returns of cryptocurrencies in the cross-section.
The average price change of the carry portfolio is 16.032% every year, about half of
the carry return. The price change of the carry portfolio is the weighted average price
change of cryptocurrencies with high carry minus the weighted average price change of
P j j P
w ∆s = wj >0 wj ∆sj − wj <0 |wj | ∆sj . It means that
P
those with low carry: ∆s =
the cryptocurrencies with higher carry on average appreciate to a larger degree than those
with lower carry, which is consistent with the result using the first method.
P j j P j j
The carry of the portfolio, C = wC = w (i − iU SD ), is the weighted av-
erage of carry of each cryptocurrency. According to Koijen et al. (2018), the carry
of the portfolio is the weighted average carry of cryptocurrencies with high carry mi-
P j j
nus the weighted average carry of cryptocurrencies with low carry: C = wC =
j j j j
P P
wj >0 w C − wj <0 |w | C > 0. The carry of the portfolio is 19.879%. Therefore,

about half the carry return comes from the difference in price appreciation between high-
carry cryptocurrencies and low-carry cryptocurrencies, and about half the carry return
comes from the difference in weighted-average carry between high-carry cryptocurrencies
and low-carry cryptocurrencies.
In summary, the results find that carry trade can yield profits in cryptocurrency
markets, and that carry is a reliable predictor of excess return across all groups, regardless
of the method used to build the carry portfolio.
To further explore the properties and characteristics of the cross-sectional carry re-
turn, we compare it with the return of a passive equal-weighted portfolio (EW portfolio)

13
constructed with the cryptocurrency universe, as Koijen et al. (2018) did in their study.
To make a fair comparison, we also apply the method in Koijen et al. (2018) to scale the
carry portfolio (constructed with the first method) by the ratio of the volatility of the
carry portfolio to that of the EW portfolio to make sure that the two portfolios have the
same volatility in magnitude.
The summary statistics presented in Table 8 indicate that the standard deviation
for both the carry return and the equal-weighted portfolio return is 87.47% per annum,
as expected. The carry portfolio exhibits a higher average annual return of 67.34%,
approximately 7% greater than the equal-weighted portfolio return. Furthermore, the
carry return is positively skewed, implying a lower risk of significant decline compared
to the equal-weighted portfolio return, which is negatively skewed and exposes investors
to heightened crash risk. The Sharpe ratio of the carry return is also superior to that
of the equal-weighted portfolio return. This supports the findings documented in Koijen
et al. (2018) that the carry portfolio, in most cases, has a higher Sharpe ratio and is less
susceptible to crash risk than the equal-weighted portfolio.
[Insert Table 8 Here]
Figure 1 depicts the cumulative returns of the carry portfolio and the equal-weighted
(EW) portfolio. Over the sample period, the carry return demonstrates a steady upward
trend, albeit with a few slight dips around 2018 and the start of 2021. On the other hand,
the EW return is characterized by more pronounced fluctuations, particularly around
2018 and 2021. Of particular interest is the behavior of the two returns in the face of
cryptocurrency market crashes. The carry return shows a moderate decline prior to the
2018 crash, but manages to withstand the impact of the bear market. On the other hand,
the EW return exhibits a sharp surge before the 2018 crash and suffers a significant decline
during the crash. Similar patterns are observed before and during the 2021 crash. The
results are consistent with the findings in Table 8, which showed that the carry return is
positively skewed, meaning it is less susceptible to crash risk. Conversely, the EW return
is negatively skewed, making it more vulnerable to crash risk.
[Insert Figure 1 Here]

14
5 Explainations for Carry Trade Return

The cross-sectional carry trade of cryptocurrencies has produced significant positive re-
turns. We examined the reasons for this outcome and found that the cryptocurrency
factors of the market, size, momentum, downside risk, and volatility risk as discussed in
Liu et al. (2022) could not account for the carry premium.15 Conversely, the geopolitical
risk factor discussed in Caldara and Iacoviello (2022) appears to contribute to the posi-
tive carry premium. However, after including geopolitical risk as a variable, the average
unexplained excess return (α) became significantly negative, indicating the presence of
additional unknown risk factors that could be contributing to the negative risk premium.
The study by Liu et al. (2022) identifies three risk factors for cryptocurrencies: mar-
ket, size, and momentum. They demonstrate that these factors can effectively explain
the abnormal returns of various cryptocurrency trading strategies. In this analysis, we
examine the ability of the cryptocurrency three-factor model to account for the abnormal
return of the cryptocurrency carry strategy. Specifically, we perform a time-series regres-
sion analysis to assess the relationship between the carry return and the three factors,

rtcarry = α + β1 cmktt + β2 csizet + β3 cmomt + ϵt . (5)

We use the cross-sectional carry return from section 4.2 (method 1). Following Lustig
et al. (2011), we use return in levels instead of log return for the regression.16
The results of the time-series regression, presented in Panel A of Table 9, indicate that
the average weekly carry return is 0.014 when no risk variables or factors are included.
Upon the inclusion of the cryptocurrency three factors proposed by Liu et al. (2022),
none of the factor coefficients are statistically significant, and the α even increases from
0.014 to 0.015 and remains significant. This suggests that the cryptocurrency three-factor
model cannot fully explain the carry return.
The following passage examines other potential explanations for the positive carry
15
Although the volatility risk shares common time-series variation with the carry return, but the α
remains significantly positive and becomes even larger than the mean carry return.
16
Both the cross-sectional carry return in this paper and cryptocurrency three factors from Liu et al.
(2022) are at a weekly frequency.

15
return observed in cross-sectional cryptocurrency trading strategies. Firstly, the impact
of downside risk on the carry return is analyzed. Based on the methodology presented by
Henriksson and Merton (1981), a regression model is run, incorporating the cryptocur-
rency market factor from Liu et al. (2022) to measure market return in the cryptocurrency
market and the downside risk factor,

rt = β0 + βmkt rm,t + βdown max {0, −rm,t } + ϵt

The results, as shown in Table 9, indicate that the coefficient of the downside risk is
not significant, implying that the carry return is relatively insensitive to downside risk,
consistent with the findings in Section 4.2. Furthermore, the residual return (α) remains
significant at the 10% level and increases in magnitude. Thus, downside risk does not
explain the positive carry return.
Next, the potential impact of volatility risk is evaluated. The weekly changes in VIX
are used to proxy for volatility risk. The results, as depicted in Table 9, show that
the coefficient of volatility risk is negative (-0.002) and significant at the 5% level, but
the average unexplained return (α) remains significant at the 5% level and increases in
magnitude. This suggests that, although volatility risk shares some time-series variation
with the carry return, it does not explain the positive carry premium.
[Insert Table 9 Here]
Finally, we examine the effect of geopolitical risk as proposed by Caldara and Iacoviello
(2022) on the carry return. Caldara and Iacoviello (2022) provides several measurements
related to geopolitical risk, from which we select six, namely, recent geopolitical risk
(GPR), recent geopolitical risk threats (GPRT), recent geopolitical risk acts (GPRA),
historical geopolitical risk (GPRH), historical geopolitical risk threats (GPRHT), and his-
torical geopolitical risk acts (GPRHA). As these risk variables are reported at a monthly
frequency, we compound the weekly carry return into monthly returns. The regression

16
model is expressed as follows:

rtcarry = α + βAt + ϵt , , (6)

where At represents GPR, GPRT, GPRA, GPRH, GPRHT or GPRHA. The results are
presented in Panel B of Table 9. Without the inclusion of explanatory variables, the
monthly carry return is estimated to be 0.049, significant at the 5% level. Upon the
addition of different measurements of geopolitical risk, the coefficients of geopolitical risk
are found to be significantly positive for GPRA, GPRH, and GPRHA, yet resulting in
significantly negative α values. These results suggest that geopolitical risk plays a role in
explaining the positive carry return, though there could be other factors that contribute
to the negative risk premium for the carry return.
These results, where the carry returns become negative after adjusting for geopoliti-
cal risk, may help explain why the cryptocurrency carry portfolio exhibits only a slightly
higher mean return and Sharpe ratio compared to the equal-weighted portfolio, as de-
picted in Table 8. This is in contrast to the findings of Koijen et al. (2018) who observe
that carry portfolios generally offer much higher mean returns and Sharpe ratios com-
pared to equal-weighted portfolios across most asset classes, as displayed in Table 2 of
their paper. The negative risk-adjusted carry return may also help explain the fact that
we get a Sharpe ratio (0.770) of the cryptocurrency carry portfolio, lower than those
documented in Cong et al. (2022) and Franz and Schmeling (2021) (1.62 and 1.24, re-
spectively).
The results in our study are consistent with Cong et al. (2022) and Franz and Schmel-
ing (2021) that UIP is violated in cryptocurrency markets and that that there is a sig-
nificantly positive return of the cryptocurrency carry strategy. However, We propose
that the economic explanation of the positive cryptocurrency carry return is different
from those in Cong et al. (2022) and Franz and Schmeling (2021), as discussed in the
introduction. The difference may come from the practice that we use a different proxy
for the interest rates of cryptocurrencies. It is still not conclusive what is the best proxy
for cryptocurrency interest rate, and the carry return derived from different proxies of

17
interest rate might have different economic meanings.
In conclusion, our findings demonstrate that the cross-sectional carry strategy pro-
duces a significantly positive carry return. Despite the efforts to explain this return
through the cryptocurrency three-factor model introduced by Liu et al. (2022), downside
risk and volatility risk, they were found to be insufficient. However, the positive carry
premium is better explained by geopolitical risk as proposed by Caldara and Iacoviello
(2022). Nonetheless, a significant negative component of the carry premium remains
unexplained, which could be a result of other underlying risks.

6 Robustness Check

6.1 UIP Regression in the Futures Market of Cryptocurrency

In this section, we perform the UIP regression in the futures market. In other words,
we use the futures implied interest rate differential between cryptocurrency and the U.S.
Dollar instead of the observed interest rate differential. Bitcoin and Ethereum are the
only two cryptocurrencies for which futures markets are available.
The UIP regression equation is as follows:

rt+1 = α + βct + ϵt+1 . (7)

where rt+1 is the log excess return of cryptocurrency in the futures market, namely the
excess return of a long position in the cryptocurrency futures. ct is the futures implied
interest rate differential between cryptocurrency and U.S. Dollar, namely carry defined
in Koijen et al. (2018). If UIP holds, the estimate of β should be 0, which means that
carry cannot predict the excess return of cryptocurrency.
Following Fama (1984), futures implied interest rate differential, is calculated as fol-
lows:

ct = s t − f t . (8)

18
where st is the log spot price of cryptocurrency in terms of U.S. Dollar and ft is the
log futures (expires at t+1) price of cryptocurrency in terms of U.S. Dollar. If covered
interest rate parity (or CIP) holds, it should be equal to ict − iUt SD . Therefore, ct is the
futures implied interest rate differential between Bitcoin and U.S. Dollar.17
Following Fama (1984), the log excess return of cryptocurrency in the futures market
is calculated as follows:

rt+1 = st+1 − ft . (9)

This is the return of taking a long position in the futures contract of crytocurrency, and
also an analogy of the return when investors take a long position of cryptocurrency and
a short position of U.S. Dollar in the spot market.18 When we calculate the return, we
assume that st+1 = ft+1 at maturity, so we use ft+1 instead of st+1 in Equation (9).19
As we did in Section 3, we do the test with monthly returns and weekly returns at a
daily frequency. Following Fama (1984), rt+1 and ct in regression Equation (7) are in
percentage points.
The results of the regression analysis are presented in Table 10. For Bitcoin futures,
both the monthly and weekly returns demonstrate significant positive estimates of β,
indicating that the implied interest rate differential (carry) in the futures market can
positively predict the future excess return of Bitcoin. A 1% increase in the monthly
(weekly) carry is found to be associated with a 1.66% (0.91%) increase in the monthly
(weekly) excess return of Bitcoin futures in the following period. On the other hand,
the results for Ethereum futures exhibit positive but insignificant estimates of β for
both monthly and weekly returns. These findings indicate that the Uncovered Interest
Parity (UIP) hypothesis does not hold in the Bitcoin futures market, and the carry factor
has a significant impact on predicting the excess return of Bitcoin futures in the time
17
Following Koijen et al. (2018), when we calculate carry, we use linearly interpolate to estimate prices
of futures which will expire in 1 month (1 week).
18
The reason can be seen from the formula of the excess return of taking a long position in cryptocur-
rency futures: rt+1 = st+1 − ft = ∆st+1 + st − ft . If CIP holds, we replace st − ft by ict − iU
t
SD
and get
the same formula as the excess return of Bitcoin in the spot market (Equation 2).
19
We roll over 2 days before the date of the last trade of the futures contract because the Bitcoin
futures contract becomes less liquid after that.

19
series. However, the evidence for UIP violation in the Ethereum futures market remains
insufficient.
[Insert Table 10 Here]

7 Conclusion

The findings of this study suggest that the cryptocurrency carry portfolio provides in-
vestors with unique investment opportunities, as it is able to withstand periods of cryp-
tocurrency market crashes while still delivering a steadily rising cumulative return. To
further understand the risk-return dynamics of this strategy, we investigate the impact
of various existing risk factors on the cryptocurrency carry premium. Our results suggest
that geopolitical risk is a significant factor in explaining the positive carry premium, but
that a negative risk-adjusted carry return still persists, warranting further investigation
in future research.

20
References

C.S. Asness, T.J. Moskowitz, and L.H. Pedersen. Value and momentum everywhere. The
Journal of Finance, 68(3), pp.929-985, 2013.

C. Burnside, M. Eichenbaum, I. Kleshchelski, and S. Rebelo. Do peso problems explain

the returns to the carry trade? The Review of Financial Studies, 24(3), pp.853-891,
2011.

D. Caldara and M. Iacoviello. Measuring geopolitical risk. American Economic Review,

112(4), pp.1194-1225, 2022.

Tarun Chitra. Competitive equilibria between staking and on-chain lending. URL:
https://doi.org/10.48550/arxiv.2001.00919, 2020.

N. Christin, B.R. Routledge, K. Soska, and A. Zetlin-Jones. The crypto carry trade.
URL: https://www.andrew.cmu.edu/user/azj/files/CarryTrade.v1.0.pdf, 2022.

Lin William Cong, Zhiheng He, and Ke Tang. Staking, token pricing, and crypto carry.
Available at SSRN, 2022.

E.F. Fama. Forward and spot exchange rates. Journal of monetary economics, 14(3),
pp.319-338, 1984.

F. C. Franz and M. Schmeling. Crypto carry. Available at SSRN, 2021.

F. C. Franz and Alexander Valentin. Crypto covered interest parity deviations. Available
at SSRN 3702212, 2020.

R. Greene and B. McDowall. Liquidity or leakage: Plumbing problems with cryptocur-

rencies. Long Finance, available at SSRN, 2018.

Roy D. Henriksson and Robert C. Merton. On market timing and investment perfor-
mance. The Journal of business, Vol.54 (4), p.513-533, 1981.

J.M. Keynes. A tract on monetary reform. London: Macmillan, 1923.

21
Ralph S.J. Koijen, Tobias J. Moskowitz, Lasse Heje Pedersen, and Evert B. Vrugt. Carry.
Journal of Financial Economics, 127(2), p.197-225, 2018.

S. Kumar. Central clearing of crypto-derivatives in a decentralized finance (defi) frame-

work: An exploratory review. International Journal of Business and Economics, 7(1),
p.128, 2022.

S.S. Lee and M. Wang. The impact of jumps on carry trade returns. Journal of Financial
Economics, 131(2), pp.433-455, 2019.

M. Lettau, M. Maggiori, and M. Weber. Conditional risk premia in currency markets

and other asset classes. Journal of Financial Economics, 114(2), pp.197-225, 2014.

Y. Liu, A. Tsyvinski, and X. Wu. Common risk factors in cryptocurrency. The Journal
of Finance, 77(2), pp.1133-1177, 2022.

H. Lustig, N. Roussanov, and A. Verdelhan. Common risk factors in currency markets.

The Review of Financial Studies, 24(11), pp.3731-3777, 2011.

L. Menkhoff, L. Sarno, M. Schmeling, and A. Schrimpf. Carry trades and global foreign
exchange volatility. The Journal of Finance, 67(2), pp.681-718, 2012.

W.K. Newey and K.D. West. Hypothesis testing with efficient method of moments esti-
mation. International Economic Review, pp.777-787, 1987.

22
Figure 1: Cumulative Returns of Carry Portfolio and Passive Equal-weighted Portfolio

This figure plots the cumulative return of the carry portfolio and that of the passive equal-weighted
(EW) portfolio. The cumulative return (vertical axis) is in real number (not percentage points). The
carry portfolio is constructed with method 1 in section 4.2 and scaled to have the same volatility as the
EW portfolio. The sample period is Oct. 2016 to Dec. 2021.

23
 Table 1: Summary Statistics of Interest Rates
currency 0x Algorand Avalanche Axie Infinity Bitcoin Bitcoin Cash Bitcoin Gold Bitcoin SV Cardano Chainlink Compound Cosmos
Start 17oct2018 03mar2020 05aug2021 25oct2021 03oct2016 17feb2021 27dec2017 21dec2018 03sep2020 25aug2020 10nov2021 26feb2020
mean % 5.915 7.115 22.126 184.508 8.129 0.791 16.851 7.782 16.925 4.184 115.084 7.032
sd % 0.426 0.541 1.125 2.273 0.489 0.245 1.050 0.406 2.012 0.467 4.084 0.719
min % 0.000 0.000 0.000 89.240 0.000 0.000 0.000 0.000 0.000 0.000 36.504 0.000
p25 % 1.732 0.471 4.949 146.768 2.086 0.000 3.325 2.351 0.505 0.016 47.420 0.133
Median % 3.343 3.733 12.198 195.168 4.929 0.000 9.116 5.726 1.606 1.126 75.210 0.832
p75 % 6.702 9.077 36.504 215.184 9.351 0.000 22.399 9.503 7.071 3.176 198.423 10.025
max % 53.556 61.446 85.241 248.198 44.546 34.239 113.241 37.351 154.829 50.736 209.547 73.043
skewness % 316.418 263.082 113.342 -45.967 202.424 612.691 198.728 178.219 261.263 331.819 21.080 293.569
kurtosis % 1506.883 1066.316 343.586 233.490 677.872 3945.451 731.590 623.548 846.944 1411.302 113.204 1193.880
currency Dai Dash Dogecoin EOS Elrond Ethereum Ethereum Classic FTX Token Fantom Filecoin IOTA Litecoin
Start 23jul2020 16mar2017 04may2021 10jul2017 28oct2021 03oct2016 03oct2016 06mar2020 22oct2021 23nov2020 03jul2017 03oct2016
mean % 13.233 7.990 9.721 6.605 165.340 6.001 9.692 25.978 80.895 49.335 1.221 6.075
sd % 1.601 0.577 0.931 0.781 2.603 0.290 0.817 2.012 3.349 2.995 0.192 0.471
min % 0.000 0.000 0.309 0.000 36.504 0.143 0.000 0.000 36.504 0.000 0.000 0.000
p25 % 0.935 1.465 0.654 0.133 155.295 1.998 0.666 0.463 44.745 7.033 0.000 1.146
Median % 5.329 3.802 1.876 0.748 179.021 4.242 2.945 6.109 54.846 24.317 0.253 3.205
p75 % 8.150 10.596 7.895 5.425 191.571 8.039 11.892 36.922 78.415 73.141 0.789 7.061
max % 155.030 75.354 92.458 92.841 238.168 26.769 91.792 173.592 251.551 226.749 41.833 61.417
skewness % 340.657 294.259 244.944 379.581 -111.529 146.956 277.960 166.456 198.194 145.338 658.825 340.671

24
kurtosis % 1384.968 1348.090 843.542 1882.713 387.424 480.785 1210.410 482.352 546.831 433.403 5641.180 1699.418
currency Maker Metaverse ETP Monero Neo OMG Network Polkadot Polygon SHIBA INU Santiment Network Solana Stellar SushiSwap
Start 09nov2021 20oct2017 15mar2017 18sep2017 04aug2017 27aug2020 26oct2021 11nov2021 18sep2017 01jun2021 19aug2019 20jan2021
mean % 32.545 11.695 5.187 14.009 6.038 6.264 172.235 15.598 8.918 18.499 5.255 25.860
sd % 0.762 1.519 0.366 0.990 0.560 1.262 3.149 0.749 1.282 1.480 0.647 1.454
min % 0.000 0.000 0.000 0.000 0.000 0.000 36.394 2.948 0.000 0.054 0.000 3.928
p25 % 36.504 0.000 1.199 2.207 0.266 0.000 113.441 4.524 0.000 1.856 0.133 8.474
Median % 36.504 0.962 2.797 6.483 1.610 0.068 207.753 7.652 0.314 8.461 0.696 15.416
p75 % 36.504 7.461 6.412 17.673 5.817 0.376 215.099 30.490 5.648 23.884 3.192 29.496
max % 54.089 185.582 46.895 117.309 57.544 145.348 234.054 36.504 207.965 166.931 83.997 133.679
skewness % -150.755 401.412 326.886 241.151 266.243 478.770 -74.857 56.883 514.804 291.502 375.471 211.728
kurtosis % 454.130 2069.404 1654.035 973.716 1007.514 2553.609 213.505 148.698 3436.562 1251.473 1847.097 704.523
currency TRON Tether Tether EURt Tether Gold Tezos Uniswap XRP Zcash pNetwork yearn.finance
Start 02sep2020 21dec2018 01oct2021 14feb2020 15aug2019 30sep2020 05jul2017 20dec2016 08jul2020 11nov2020
mean % 10.798 11.911 18.101 3.754 11.428 12.897 2.929 4.594 8.542 23.030
sd % 0.934 0.460 1.074 0.939 0.951 0.873 0.290 0.453 0.859 1.351
min % 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 2.267
p25 % 0.000 6.255 3.585 0.000 0.421 0.594 0.133 0.189 0.000 3.060
Median % 1.998 10.864 3.597 0.000 5.329 3.741 0.666 1.321 0.621 14.113
p75 % 14.338 14.656 30.722 0.000 12.011 25.831 3.010 4.206 11.591 37.044
max % 103.822 50.738 65.939 141.785 123.173 80.536 44.497 56.199 122.320 122.167
skewness % 221.668 184.239 108.111 564.441 305.122 160.344 370.018 320.668 317.585 175.416
kurtosis % 819.463 773.437 272.581 3592.310 1451.416 563.788 2043.601 1433.196 1623.736 648.236

This table provides summary statistics of annualized interest rates of 46 cryptocurrencies in our sample. The data is from Bitfinex platform. All statistics are in percentage points. The end date of sample
for most cryptocurrencies are Nov.
√ 30, 2021, for other cryptocurrencies the sample ends 1 day earlier or later. Annualized mean is daily mean multiplied by 365. Annualized standard deviation is daily
standard deviation multiplied by 365.
 Table 2: Summary Statistics of Daily Price Change

currency 0x Algorand Avalanche Axie Infinity Bitcoin Bitcoin Cash Bitcoin Gold Bitcoin SV Cardano Chainlink Compound Cosmos
mean% 0.033 0.445 2.353 0.157 0.351 -0.115 -0.153 0.044 0.891 0.181 -1.239 0.442
sd% 7.131 8.469 9.924 5.045 4.921 7.394 7.796 7.486 7.548 7.602 5.862 8.572
min% -46.005 -57.990 -21.007 -8.871 -46.473 -43.461 -37.667 -56.029 -30.111 -46.602 -9.077 -49.288
p25% -3.804 -3.965 -3.626 -2.763 -1.590 -2.877 -3.250 -2.714 -3.175 -3.952 -5.949 -3.424
Median% 0.022 0.317 1.849 0.667 0.255 0.481 0.000 0.000 0.393 0.510 -2.871 0.500
p75% 3.514 4.857 8.555 3.045 2.556 3.293 2.850 2.562 4.025 4.847 2.660 4.236
max% 44.044 41.331 24.644 12.069 22.512 42.082 71.658 58.449 27.944 20.687 10.761 36.991
skewness% 4.534 -60.538 1.987 28.289 -69.794 -34.258 126.370 60.135 41.246 -92.748 40.202 -56.257
kurtosis% 874.469 1040.692 310.528 267.465 1172.668 1406.204 1794.609 2078.381 509.184 721.143 237.694 832.092
currency Dai Dash Dogecoin EOS Elrond Ethereum Ethereum Classic FTX Token Fantom Filecoin IOTA Litecoin
mean% -0.015 0.059 -0.633 0.071 2.250 0.447 0.287 0.224 -0.764 0.241 0.122 0.310
sd% 0.389 7.149 9.091 8.236 9.382 6.771 7.344 5.534 9.354 8.498 8.340 6.974
min% -1.961 -46.546 -49.617 -50.323 -13.436 -55.071 -50.779 -32.175 -15.268 -42.616 -50.555 -44.901
p25% 0.000 -2.906 -4.714 -2.992 -1.677 -2.350 -2.692 -2.505 -6.682 -3.449 -3.481 -2.615
Median% 0.000 0.000 -0.246 0.000 0.657 0.050 0.000 0.031 -1.962 -0.245 0.000 0.000
p75% 0.000 3.254 2.867 3.102 2.861 3.202 3.065 3.174 2.342 3.608 3.802 2.801
max% 1.961 39.959 24.094 43.944 34.510 39.140 45.768 29.559 20.018 45.873 64.367 53.980
skewness% -94.883 -18.041 -115.913 -6.506 199.752 5.809 5.569 -24.459 82.138 53.247 56.364 56.568
kurtosis% 1403.777 936.687 1001.026 922.744 835.630 1067.430 981.499 873.447 316.997 1091.396 1023.187 1312.629

25
currency Maker Metaverse ETP Monero Neo OMG Network Polkadot Polygon SHIBA INU Santiment Network Solana Stellar SushiSwap
mean% -0.459 -0.277 0.201 0.054 0.105 0.602 0.118 -0.385 0.029 1.502 0.268 0.044
sd% 4.294 10.082 6.899 7.578 8.583 8.441 6.804 9.785 9.053 8.010 7.033 8.526
min% -7.255 -101.163 -53.418 -45.391 -45.379 -47.697 -8.311 -11.650 -81.001 -31.825 -36.004 -29.096
p25% -3.731 -3.754 -2.746 -3.716 -3.922 -3.822 -4.692 -4.645 -3.704 -3.587 -2.789 -5.051
Median% -0.601 -0.317 0.066 0.000 0.000 0.523 -1.711 -3.473 0.000 1.670 0.164 0.000
p75% 2.969 3.254 3.357 3.732 3.933 5.145 3.636 4.174 3.695 5.860 3.197 4.417
max% 6.276 105.990 51.087 47.567 54.165 28.283 19.081 16.753 51.551 32.983 55.932 36.430
skewness% 11.670 30.056 -31.401 -6.545 35.940 -42.998 112.857 83.293 -16.986 13.230 121.547 19.143
kurtosis% 200.454 2819.731 1325.021 874.628 816.786 718.025 379.943 270.478 1387.436 593.981 1488.928 446.893
currency TRON Tether Tether EURt Tether Gold Tezos Uniswap XRP Zcash pNetwork yearn.finance
mean% 0.329 -0.003 -0.041 0.029 0.261 0.567 0.124 0.147 0.125 0.192
sd% 7.500 0.405 0.468 0.996 7.778 8.501 7.796 7.538 9.720 7.184
min% -38.342 -1.980 -0.881 -4.830 -60.547 -40.333 -55.040 -53.941 -41.428 -33.510
p25% -3.038 0.000 0.000 -0.405 -2.994 -4.468 -2.729 -3.277 -5.015 -3.762
Median% 0.347 0.000 0.000 0.079 0.000 0.675 -0.037 -0.023 0.000 0.531
p75% 3.342 0.000 0.000 0.532 3.822 5.407 2.658 3.763 3.962 3.996
max% 46.205 1.980 0.881 3.747 32.994 32.293 62.668 64.995 59.660 28.820
skewness% 46.178 -6.436 -4.691 -58.578 -112.712 1.278 91.077 22.571 92.424 -12.868
kurtosis% 1050.339 925.379 349.308 591.082 1266.852 553.676 1784.723 1147.269 943.450 592.997

This table reports summary statistics of daily change in log prices (logP2 − logP1 ) of the 46 cryptocurrencies in our sample (logP2 − logP1 ). All prices are close prices. All statistics are in percentage
points. The data is from the website coinmarketcap.com. Sample period for each cryptocurrency is the same as that in Table 1.
 Table 3: UIP Regression (monthly return)

0x Algorand Avalanche Axie Infinity Bitcoin Bitcoin Cash Bitcoin Gold

β -15.352*** 4.891 1.883 3.726*** 8.641*** -19.305*** 3.156
(2.856) (4.110) (4.067) (0.204) (1.971) (6.603) (2.364)
α 7.201*** 9.649*** 39.538*** -43.109*** 3.426** 2.374 -5.107*
(2.468) (2.949) (9.720) (2.836) (1.399) (4.303) (2.816)
N 793 431 60 5 1,319 183 1,000
Bitcoin SV Cardano Chainlink Cosmos Dash Dogecoin EOS
β -12.586*** -2.241** -4.161 -4.112* 7.997*** -7.694*** -2.470
(3.920) (1.024) (3.077) (2.180) (2.105) (2.089) (2.002)
α 8.243** 25.749*** 7.774** 14.725*** -1.602 -3.435 3.231
(3.312) (3.973) (3.405) (3.434) (2.024) (4.940) (2.288)
N 746 302 309 421 1,207 129 1,125
Ethereum Ethereum Classic FTX Token Fantom Filecoin IOTA Litecoin
β 13.333** -1.263 0.660 1.043 2.075 4.335 12.719***
(5.725) (1.039) (1.040) (2.795) (1.648) (6.738) (3.080)
α 5.375** 7.736*** 8.260*** -29.546* 3.478 3.083 2.615
(2.231) (2.089) (2.785) (11.765) (6.197) (2.311) (1.918)
N 1,325 1,324 308 6 241 1,130 1,325
Metaverse ETP Monero Neo OMG Network Polkadot Santiment Network Solana
β 1.810 -1.562 -0.409 5.951** -0.863 3.563 3.835*
(1.118) (2.999) (1.294) (2.386) (0.823) (2.353) (1.983)
α -5.605** 5.030*** 2.213 -0.409 16.409*** -0.325 32.697***
(2.572) (1.809) (2.238) (2.257) (4.301) (2.809) (8.862)
N 1,047 1,207 1,075 1,106 307 1,076 110
Stellar SushiSwap TRON Tezos Uniswap XRP Zcash
β -0.499 4.991** 3.447 4.199* -5.490 5.706 1.605
(1.487) (2.083) (2.932) (2.342) (3.376) (5.617) (3.675)
α 7.051*** -8.684* 7.462** 4.188 21.261*** 2.713 2.277
(2.605) (4.407) (3.573) (2.731) (5.828) (2.067) (1.722)
N 575 203 303 577 283 1,098 1,269
pNetwork yearn.finance
β -10.504*** 1.620
(2.652) (1.296)
α 8.284 1.280
(5.109) (2.898)
N 344 251

This table reports the results of UIP regression for each cryptocurrency with monthly interest rates and monthly returns.
The regression equation is rt+1 = α + β(ict − iUt SD ) + ϵt+1 . Following Fama (1984), the return and interest rate differential in
the regression are in percentage points. Newey-West standard errors are reported in parentheses. ***, ** and * indicates
that p < 0.01, p < 0.05 and p < 0.1, respectively. There are 37 cryptocurrencies in tatal in this test. We do not include
stablecoins here. Compound, Elrond, Maker, Polygon and SHIBA INU are not included either since the observations of
monthly returns for these cryptocurrencies are not enough to do the regression. The data of cryptocurrency interest rates
is from Bitfinex platform. The data of cryptocurrency prices is from the website coinmarketcap.com. The data of one-
month US interest rate we use is the market yield on US Treasury securities at 1-month constant maturity, quoted on an
investment basis, not seasonally adjusted, from FRED. The sample period is Oct. 2016 to Dec. 2021. The estimates of β
for 9 cryptocurrencies are significantly positive, for 7 cryptocurrencies are significantly negative, and for 21 cryptocurrencies
are insignificant.

26
 Table 4: UIP Regression (weekly return)
0x Algorand Avalanche Axie Infinity Bitcoin Bitcoin Cash Bitcoin Gold Bitcoin SV
β -4.602 -5.389 8.507 5.008 15.392*** -36.850 1.974 -12.469
(6.363) (10.058) (11.779) (3.033) (3.375) (30.571) (2.737) (7.935)
α 0.692 2.541* 9.072* -11.191 0.085 0.308 -0.984 1.752
(1.034) (1.482) (4.562) (9.830) (0.554) (1.739) (1.087) (1.314)
N 781 435 75 22 1,279 196 977 736
Cardano Chainlink Compound Cosmos Dash Dogecoin EOS Elrond
β -2.633 -18.929*** 3.334** -14.993*** 14.712** -9.926* -3.503 -6.519
(1.905) (6.441) (1.050) (5.018) (6.862) (5.145) (3.646) (3.931)
α 5.325*** 2.225 -11.108*** 3.789** -1.119 -1.283 0.854 27.958**
(1.607) (1.476) (2.029) (1.583) (0.901) (2.374) (0.891) (10.863)
N 309 316 10 430 1,174 143 1,097 17
Ethereum Ethereum Classic FTX Token Fantom Filecoin IOTA Litecoin Metaverse ETP
β 24.718** -1.143 0.993 3.062* 3.015** -1.579 11.887** 1.826
(9.604) (2.164) (1.627) (1.719) (1.400) (19.064) (5.586) (2.228)
α 0.350 1.682** 1.154 -8.535 -0.269 0.771 0.652 -1.391
(0.809) (0.845) (1.294) (6.069) (2.590) (0.943) (0.761) (1.048)
N 1,286 1,286 311 23 246 1,101 1,286 1,021
Monero Neo OMG Network Polkadot Polygon Santiment Network Solana Stellar
β 6.368 -1.906 4.832 -2.052 1.839 11.374** 11.038** -4.375
(6.331) (2.300) (5.929) (1.493) (3.240) (4.863) (5.215) (4.555)
α 0.744 0.888 0.151 2.848* -4.402 -1.082 3.961 1.680
(0.752) (0.934) (1.019) (1.703) (9.353) (1.039) (3.281) (1.160)
N 1,174 1,049 1,078 314 21 1,050 124 573
SushiSwap TRON Tezos Uniswap XRP Zcash pNetwork yearn.finance
β 4.839 2.894 2.443 3.698 14.608* -3.174 -5.171 1.476
(5.918) (3.822) (3.031) (6.398) (8.278) (6.435) (7.766) (3.463)
α -1.231 1.236 0.880 2.169 0.365 1.047 1.644 0.361
(2.507) (1.577) (1.239) (1.972) (0.860) (0.763) (2.072) (1.471)
N 215 310 575 291 1,072 1,231 351 259

This table reports the results of UIP regression for each cryptocurrency with weekly interest rates and weekly returns. The regression
equation is rt+1 = α + β(ict − iUt SD ) + ϵt+1 . Following Fama (1984), the return and interest rate differential in the regression are in
percentage points. Newey-West standard errors are reported in parentheses. ***, ** and * indicates that p < 0.01, p < 0.05 and p < 0.1,
respectively. The test includes 40 cryptocurrencies in total. SHIBA INU is not included because there are not sufficient observations
to do the regression using Newey-West standard error with 3 lags. Maker is not included because the interest rate of Maker varies
too little, whereas the price is too volatile, so that the coefficient is extremely large (-124,917.2), which is not reasonable. The data of
cryptocurrency interest rates is from Bitfinex platform. The data of cryptocurrency prices is from the website coinmarketcap.com. The
data of one-week (7 days) US interest rate is from OptionMetrics. The sample period is Oct. 2016 to Dec. 2021. The estimates of β for
10 cryptocurrencies are significantly positive, for 3 cryptocurrencies are significantly negative, for 27 cryptocurrencies are insignificant.

27
 Table 5: UIP Panel Regression

Individual Time
fixed effects fixed effects β N
Monthly Return X X 1.550*** 24,526
(0.348)
X 1.335*** 24,526
(0.461)
X 1.413*** 24,526
(0.326)
Weekly Return X X 1.729*** 24,257
(0.614)
X 2.256*** 24,257
(0.807)
X 1.618*** 24,257
(0.493)

This table reports the results of panel regression for the UIP tests with all 42 cryp-
tocurrencies in our sample (stablecoins are not included). The regression equation is
rj,t+1 = αj + β(icj,t − iUt SD ) + γt + ϵj,t+1 . αj is individual fixed effect and γt is time fixed
effect. Newey-West standard errors are reported in parentheses. ***, ** and * indicates
that p < 0.01, p < 0.05 and p < 0.1, respectively. The data of cryptocurrency interest
rates is from Bitfinex platform. The data of cryptocurrency prices is from the website
coinmarketcap.com. The data of one-month US interest rate we use is the market yield
on US Treasury securities at 1-month constant maturity, quoted on an investment basis,
not seasonally adjusted, from FRED. The data of one-week (7 days) US interest rate is
from OptionMetrics. The sample period is Oct. 2016 to Dec. 2021.

28
 Table 6: Time-series Carry Trade between Stablecoins and U.S. Dollar

USDT DAI
Portfolio month week month week
Price Change: ∆s Price Change: ∆s
Mean -0.324 -0.614 -1.284 0.316
Std 1.786 3.117 2.055 3.368
ic − iU SD Carry
Mean 9.391 9.429 11.974 11.155
Std 2.072 0.989 7.813 3.598
Carry Return Carry Return
Mean 9.067 8.815 10.690 11.470
(0.499) (0.991) (2.333) (2.160)
Std 2.614 3.206 7.340 4.014
SR 3.469 2.750 1.456 2.857

This table reports the results of time-series carry trade between stablecoins (USDT
and Dai) and U.S. Dollar. Total carry return, price change and interest rate differential
(carry) are reported. Means, standard deviations and Newey-West standard errors (in
parentheses) are annualized and in percentage points. Sharpe Ratios are annualized and
in real numbers. Annualized mean is the mean of monthly (weekly) return multiplied by
12 (52). Annualized √standard
√ deviation is the standard deviation of monthly (weekly)
return multiplied by 12 ( 52). Annualized Sharpe Ratio is the ratio of the annualized
mean and annualized standard deviation. We use monthly (weekly) carry return at daily
frequency to increase the number of observations. The sample period of USDT is Dec.
2018 - Nov. 2021. The sample period of DAI is Jul. 2020 - Nov. 2021. For USDT, the
monthly and weekly carry return are significantly positive at 1%. For DAI, the monthly
and weekly carry return are significantly positive at 1%.

29
 Table 7: Cross-sectional Carry Trade

Method 1 Method 2
Portfolio 1 2 3

Price Change: ∆sj ∆s

Mean 46.525 57.267 74.113 16.032
Std 95.390 92.651 102.726 51.892

ij − iU SD C
Mean -0.131 2.872 18.989 19.879
Std 0.261 0.504 1.705 1.815

rj = ij − iU SD + ∆sj
Mean 46.394 60.138 93.102
Std 95.442 92.743 102.829
SR 0.486 0.648 0.905

High minus Low: rj − r1 rcarry

Mean 13.744 46.708 35.911
(17.206) (23.378) (21.065)
Std 44.917 60.675 51.843
SR 0.306 0.770 0.693

This table reports the summary statistics of cross-sectional carry trade portfolios
within our cryptocurrency universe. We report the portfolio returns r, price change
∆s and interest rate differential ij − iU SD (or carry C) for each portfolio. The universe
includes 42 cryptocurrencies in total, not including stablecoins because the prices of
stablecoins are pegged to other assets (U.S. Dollar, euro, gold etc.) so that inter-
est rates are not supposed to predict their price change. We use weekly returns (at
weekly frequency rather than daily frequency) and weekly interest rates. Means, stan-
dard deviations and Newey-West standard errors (in parentheses) are annualized and
in percentage points. Sharpe Ratios are in real numbers. Annualized mean is the
weekly mean multiplied √ by 52. Annualized standard deviation is the weekly standard
deviation multiplied by 52. Annualized Sharpe Ratio is the ratio of the two terms.
The sample period is Oct. 2016 to Dec. 2021. We use two methods to construct the
carry portfolio. In method 1, we firstly rank the cryptocurrencies by their carry, and
then split them equally into three groups (some groups might have 1 more than other
groups if the total number is not divisible by 3). In method 2, we use the method intro-
duced by Asness et al. (2013). Specifically, we use the weight determined by their carry
for each cryptocurrency to construct the carry portfolio: wti = zt (rank(cit ) − Nt2+1 ).
In method 1: r3 − r1 is significant at 5% level, whereas r2 − r1 is not significant. In
method 2, the carry return is significant at 10% level.

30
Table 8: A Comparison between Carry Portfolio and Passive Equal Weighted Portfolio

N Mean Std Skewness Kurtosis SR

Carry 268 67.336 87.471 1.773 19.351 0.770
EW 268 60.398 87.471 -0.087 3.722 0.690
This table reports the summary statistics of the return of the carry portfolio and that of the
equal-weighted portfolio. Mean and Standard deviation are in percentage points. N, Skewness,
Kurtosis and SR are in real numbers. The sample period is Oct. 2016 to Dec. 2021. Carry
portfolio is constructed with method 1, and scaled to have the same volatility as the equal
weighted portfolio for comparison. Specifically, firstly we get the ratio of carry portfolio volatility
and equal weighted portfolio volatility, and then we divide carry return by the ratio.

31
 Table 9: Regression Results of Carry Return on Risk Factors and Risk Variables

Panel A: Cryptocurrency Three Factors, Downside Risk and Volatility Risk

α cmkt csize cmom downside volatility Adj.R2
Carry Return 0.014**
(0.007)
Carry Return 0.015** -0.017 -0.048 0.219 0.042
(0.006) (0.083) (0.032) (0.152)
Carry Return 0.018* -0.050 -0.091 -0.009
(0.011) (0.116) (0.173)
Carry Return 0.015** -0.002** 0.001
(0.007) (0.001)
Panel B: Geopolitical Risk
α GPR GPRT GPRA GPRH GPRHT GPRHA Adj.R2
Carry Return 0.049**
(0.023)
Carry Return -0.0841 0.0015 0.0199
(0.0819) (0.0010)
Carry Return 0.0252 0.0002 -0.0153
(0.0935) (0.0009)
Carry Return -0.0989** 0.0022*** 0.1040
(0.0485) (0.0008)
Carry Return -0.1314* 0.0026** 0.0536
(0.0716) (0.0011)
Carry Return -0.0193 0.0007 -0.0061
(0.0829) (0.0009)
Carry Return -0.0952** 0.0029*** 0.1157
(0.0471) (0.0011)

This table presents the regression results of cross-sectional carry return (r3 − r1 ) on risk factors or risk variables.
Following Lustig et al. (2011), we use return in levels instead of log return to do the regression. Regression
coefficients, Newey-West standard errors (in parentheses) and Adj.R2 ’s are reported. ***, ** and * indicates
that p < 0.01, p < 0.05 and p < 0.1, respectively. Panel A presents the results for the crypto three factors from
Liu et al. (2022), downside risk and volatility risk. Sample period is Oct. 2016 - Jul. 2020 because the data
of cryptocurrency three factors is only available until July 2020. We use weekly change in VIX to proxy for
volatility risk. Data of VIX is from Bloomberg database. Both the cross-sectional carry return and explanatory
variables are at weekly frequency. Panel B presents the results for Geopolitical risk from Caldara and Iacoviello
(2022). GPR, GPRT, GPRA, GPRH, GPRHT and GPRHA stand for recent GPR, recent GPR threats, recent
GPR Acts, historical GPR, historical GPR threats and historical GPR acts, respectively. Sample period is from
Oct. 2016 to Nov. 2021.

32
 Table 10: UIP Regression in the Futures Market

BTC ETH
Month Week Month Week
β 1.658** 0.909*** 0.866 0.773
(0.758) (0.297) (1.479) (0.688)
α 8.307*** 1.105 -0.929 -1.576
(1.947) (0.741) (3.507) (1.681)
N 610 629 180 199

This table presents the results of UIP regression in the futures market. The regression
equation is rt+1 = α+βct +ϵt+1 . rt+1 = st+1 −ft is the log return of holding cryptocurrency
futures. ct = st − ft is the carry of the cryptocurrency, or futures implied interest rate
differential between the cryptocurrency and U.S. Dollar. Newey-West standard errors are
reported in parentheses. ***, ** and * indicates that p < 0.01, p < 0.05 and p < 0.1,
respectively. The sample period for Bitcoin futures is Dec. 2017 to May. 2022. The
sample period for Ethereum futures is Feb. 2021 to May. 2022.

33