Beware the Middleman: Empirical Analysis of Bitcoin-Exchange Risk

1

Beware the Middleman:
Empirical Analysis of Bitcoin-Exchange Risk
Tyler Moore1 and Nicolas Christin2
1

Computer Science & Engineering, Southern Methodist University, USA, tylerm@smu.edu
2
INI & CyLab, Carnegie Mellon University, USA, nicolasc@cmu.edu

Abstract. Bitcoin has enjoyed wider adoption than any previous crypto-currency;
yet its success has also attracted the attention of fraudsters who have taken advantage of operational insecurity and transaction irreversibility. We study the risk
investors face from Bitcoin exchanges, which convert between Bitcoins and hard
currency. We examine the track record of 40 Bitcoin exchanges established over
the past three years, and find that 18 have since closed, with customer account
balances often wiped out. Fraudsters are sometimes to blame, but not always. Using a proportional hazards model, we find that an exchange’s transaction volume
indicates whether or not it is likely to close. Less popular exchanges are more
likely to be shut than popular ones. We also present a logistic regression showing
that popular exchanges are more likely to suffer a security breach.
Keywords: Bitcoin, currency exchanges, security economics, cybercrime

1

Introduction

Despite added benefits such as enhanced revenue [1] or anonymity [2], and often elegant designs, digital currencies have until recently failed to gain widespread adoption.
As such, the success of Bitcoin [3] came as a surprise. Bitcoin’s key comparative advantages over existing currencies lie in its entirely decentralized nature and in the use of
proof-of-work mechanisms to constrain the money supply. Bitcoin also benefited from
strongly negative reactions against the banking system, following the 2008 financial
crisis: Similar in spirit to hard commodities such as gold, Bitcoin offers an alternative
to those who fear that “quantitative easing” policies might trigger runaway inflation.
As of January 2013, Bitcoin’s market capitalization is approximately US$187 million [4]. However, with success comes scrutiny, and Bitcoin has been repeatedly targeted by fraudsters. For instance, over 43,000 Bitcoins were stolen from the Bitcoinica
trading platform in March 2012 [5]; in September 2012, $250,000 worth of Bitcoins
were pilfered from the Bitfloor currency exchange [6]. Interestingly, experience from
previous breaches does not suggest that failures necessarily trigger an exodus from the
currency. In fact, with two possible exceptions—a June 2011 hack into the largest Bitcoin currency exchange, which coincided with the USD-Bitcoin exchange rate peaking,
and the August 2012 downfall of the largest Bitcoin Ponzi scheme [8]—the (volatile)
Bitcoin exchange rate has fluctuated independently from disclosed hacks and scams.
While Bitcoin’s design principles espouse decentralization, an extensive ecosystem
of third-party intermediaries supporting Bitcoin transactions has emerged. Intermediaries include currency exchanges used to convert between hard currency and Bitcoin;
Authors’ pre-publication version. Cite as: Tyler Moore and Nicolas Christin. Beware the middleman: empirical analysis of
Bitcoin-exchange risk. In Ahmad-Reza Sadeghi, editor, Financial Cryptography, volume 7859 of Lecture Notes in Computer
Science, pages 25-33. Springer, 2013.

2

Tyler Moore and Nicolas Christin

marketplace escrow services [7]; online wallets; mixing services; mining pools; or even
investment services, be they legitimate or Ponzi schemes [8]. Ironically, most of the risk
Bitcoin holders face stems from interacting with these intermediaries, which operate as
de facto centralized authorities. For instance, one Bitcoin feature prone to abuse is that
transactions are irrevocable, unlike most payment mechanisms such as credit cards and
electronic fund transfers. Fraudsters prefer irrevocable payments, since victims usually
only identify fraud after transactions take place [9, 10]. Irrevocability makes any Bitcoin transaction involving one or more intermediaries subject to added risk, such as if
the intermediary becomes insolvent or absconds with customer deposits.
In this paper, we focus on one type of intermediary, currency exchanges, and empirically examine the risk Bitcoin holders face from exchange failures. Section 2 explains our data collection and measurement methodology. Section 3 presents a survival
analysis of Bitcoin exchanges, and shows that an exchange probability of closure is
inversely correlated to its trade volumes. Section 4 complements this analysis with a logistic regression that indicates that popular exchanges are more likely to suffer security
breaches. Section 5 reviews related work and Section 6 discusses follow-up research.

2
2.1

Data on Bitcoin-Exchange Closures
Data Collection Methodology

We begin by collecting historical data on the Bitcoin exchange rates maintained by the
website bitcoincharts.com. This includes the daily trade volumes and average
weighted daily price for 40 Bitcoin exchanges converting into 33 currencies until January 16, 2013, when the data collection was made. We calculated the average daily
trade volume for each exchange by tallying the total number of Bitcoins converted into
all currencies handled by the exchange for the days the exchange was operational.
We also calculate the “lifetime” of each exchange, that is, the number of days the
exchange is operational, denoted by the difference between the first and last observed
trade. We deem an exchange to have closed if it had not made a trade in at least two
weeks before the end of data collection. We further inspected the existence of a report
on the Bitcoin Wiki [11] or on Bitcoin forums [12] to confirm closure, and determine
whether closure was caused by a security breach (e.g., hack or fraud). We also checked
for reports on whether or not investors were repaid following the exchange’s closure.
Finally, to assess regulatory impact, we attempted to identify the country where each
exchange is based. We then used an index (ranging between 0 and 49) computed by
World Bank economists [13] to identify each country’s compliance with “Anti-MoneyLaundering and Combating the Financing of Terrorism” (AML-CFT) regulations [13].
2.2

Summary Statistics

Table 1 lists all 40 known Bitcoin currency exchanges, along with relevant facts about
whether the exchange later closed. Nine exchanges experienced security breaches, caused
either by hackers or other criminal activity. Five of these exchanges subsequently closed,
but four have survived so far (Mt. Gox, btc-e.com, Bitfloor, and Vircurex). Another
13 closed without experiencing a publicly-announced breach.

Beware the Middleman: Empirical Analysis of Bitcoin-Exchange Risk
Exchange

Origin Dates Active Daily vol. Closed? Breached? Repaid? AML Risk Ratio

BitcoinMarket
Bitomat
FreshBTC
Bitcoin7
ExchangeBitCoins.com
Bitchange.pl
Brasil Bitcoin Market
Aqoin
Global Bitcoin Exchange
Bitcoin2Cash
TradeHill
World Bitcoin Exchange
Ruxum
btctree
btcex.com
IMCEX.com
Crypto X Change
Bitmarket.eu
bitNZ
ICBIT Stock Exchange
WeExchange
Vircurex
btc-e.com
Mercado Bitcoin
Canadian Virtual Exchange
btcchina.com
bitcoin-24.com
VirWox
Bitcoin.de
Bitcoin Central
Mt. Gox
Bitcurex
Kapiton
bitstamp
InterSango
Bitﬂoor
Camp BX
The Rock Trading Company
bitme
FYB-SG

US
PL
PL
US/BG
US
PL
BR
ES
?
US
US
AU
US
US/CN
RU
SC
AU
PL
NZ
SE
US/AU
US?
BG
BR
CA
CN
DE
DE
DE
FR
JP
PL
SE
SL
UK
US
US
US
US
SG

3

4/10 – 6/11
4/11 – 8/11
8/11 – 9/11
6/11 – 10/11
6/11 – 10/11
8/11 – 10/11
9/11 – 11/11
9/11 – 11/11
9/11 – 1/12
4/11 - 1/12
6/11 - 2/12
8/11 – 2/12
6/11 – 4/12
5/12 – 7/12
9/10 – 7/12
7/11 – 10/12
11/11 – 11/12
4/11 – 12/12
9/11 – pres.
3/12 – pres.
10/11 – pres.
12/11 – pres.
8/11 – pres.
7/11 – pres.
6/11 – pres.
6/11 – pres.
5/12 – pres.
4/11 – pres.
8/11 – pres.
1/11 – pres.
7/10 – pres.
7/12 – pres.
4/12 – pres.
9/11 – pres.
7/11 – pres.
5/12 – pres.
7/11 – pres.
6/11 – pres.
7/12 – pres.
1/13 – pres.

2454
758
3
528
551
380
0
11
14
18
5082
220
37
75
528
2
874
33
27
3
2
6
2604
67
832
473
924
1668
1204
118
43230
157
160
1274
2741
816
622
52
77
3

yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no
no

yes
yes
no
yes
no
no
no
no
no
no
yes
yes
no
no
no
no
no
no
no
no
no
yes
yes
no
no
no
no
no
no
no
yes
no
no
no
no
yes
no
no
no
no

–
yes
–
no
–
–
–
–
–
–
yes
no
yes
yes
no
–
–
no
–
–
–
–
yes
–
–
–
–
–
–
–
yes
–
–
–
–
no
–
–
–
–

34.3
21.7
21.7
33.3
34.3
21.7
24.3
30.7
27.9
34.3
34.3
25.7
34.3
29.2
27.7
11.9
25.7
21.7
21.3
27.0
30.0
27.9
32.3
24.3
25.0
24.0
26.0
26.0
26.0
31.7
22.7
21.7
27.0
35.3
35.3
34.3
34.3
34.3
34.3
33.7

1.12
1.28
2.01
1.59
0.65
0.61
3.85
1.57
1.45
1.47
0.94
1.80
1.24
0.98
0.61
1.88
0.53
1.09
1.14
2.15
2.23
4.41
1.08
0.95
0.53
0.60
0.52
0.45
0.49
0.91
0.49
0.76
0.80
0.54
0.45
1.45
0.63
1.14
1.04
2.23

Table 1. Bitcoin exchange indicators. “Origin” denotes the jurisdiction under which the exchange
operates, “AML,” the extent to which the exchange’s jurisdiction has implemented “Anti-Money
Laundering and Combating the Financing of Terrorism” international standards [13]. “Risk ratio”
is the relative risk of exchange failure based on the Cox proportional hazards model (Section 3).

The popularity of exchanges varied greatly, with 25% of exchanges processing under 25 Bitcoins each day on average, while the most popular exchange, Mt. Gox, has
averaged daily transactions exceeding 40 000 BTC. The median daily transactions carried out by exchanges is 290, while the mean is 1 716.
One key factor affecting the risk posed by exchanges is whether or not its customers
are reimbursed following closure. We must usually rely on claims by the operator and
investors if they are made public. Of the 18 exchanges that closed, we have found evidence on whether customers were reimbursed in 11 cases. Five exchanges have not
reimbursed affected customers, while six claim to have done so. Thus, the risk of losing
funds stored at exchanges is real but uncertain.
As a ﬁrst approximation, the failure rate of Bitcoin exchanges is 45%. The median
lifetime of exchanges is 381 days. These summary statistics obscure two key facts:

4


exchanges are opened at different times and so their maximum potential lifetimes vary,
and a majority of exchanges remain viable at the end of our observation period. Survival
analysis can properly account for this.

3

Survival Analysis of Exchange Closure

We use survival analysis to estimate the time it takes for Bitcoin exchanges to close
and to identify factors that can trigger or stave off closure. Robust estimation requires
considering that some exchanges remain open at the end of our measurement interval
(“censored” data points). Two mathematical functions are commonly used. First, a survival function S(t) measures the probability that an exchange will continue to operate
longer than for t days. Second, a hazard function h(t) measures the instantaneous risk
of closure at time t. To identify factors affecting an exchange’s survival time, we use a
Cox proportional hazards model [14], rather than traditional linear regression. We can
also estimate the survival function using the best-fit Cox model.
3.1

Statistical Model

We hypothesize that three variables affect the survival time of a Bitcoin exchange:
Average daily transaction volume: an exchange can only continue to operate if it
is profitable, and profitability usually requires achieving scale in the number of feegenerating transactions performed. We expect that exchanges with low transaction volume are more likely to shut down. We use a log-transformation of the transaction volume given how skewed transaction volumes are.
Experiencing a security breach: suffering a security breach can erase profits, reduce
cash flow, and scare away existing and prospective customers. We thus expect breached
exchanges to be more likely to subsequently close.
AML/CFT compliance: some Bitcoin exchanges complain of being hassled by financial regulators. Thus, exchanges operating in countries with greater emphasis on antimoney laundering efforts may be pressured into shutting down.
We then construct a corresponding proportional hazards model [14]:
hi (t) = h0 (t) exp(β1 log(Daily vol.)i + β2 Breachedi + β3 AMLi ).
Here, hi (t) is the hazard rate for exchange i, log(Daily vol.)i is the transaction volume
at exchange i, Breachedi indicates whether exchange i suffered a security breach, and
AMLi denotes the AML/CFT compliance score for the exchange’s country of incorporation. β1 , β2 , β3 are best-fit constants, and h0 (t) is the unspecified baseline hazard.
3.2

Results

The best-fit Cox model is:
coef.
exp(coef.) Std. Err.)
log(Daily vol.)i β1 −0.173
0.840
0.072
Breachedi
β2 0.857
2.36
0.572
AMLi
β3 0.004
1.004
0.042
log-rank test: Q=7.01 (p = 0.0715), R2 = 0.145

Significance
p = 0.0156
p = 0.1338
p = 0.9221

1.0


0.4

0.6

0.8

Intersango
Mt. Gox
Bitfloor
Vircurex
Average

0.0

0.2

Survival probability

5

0

200

400

600

800

Days

Fig. 1. Empirically-derived survival probability function for Bitcoin exchanges.

The daily volume is negatively associated with the hazard rate (β1 = −0.173): doubling
the daily volume rate corresponds to a 16% reduction in the hazard rate (exp(β1 ) =
0.84) . Thus, exchanges that process more transactions are less likely to shut down.
Suffering a breach is positively correlated with hazard, but with a p-value of 0.1338,
this correlation falls just short of being statistically significant at this time. Given that
just nine exchanges publicly reported suffering breaches and only five later closed, it is
not surprising that the association is not yet robust.
Finally, the anti-money laundering indicator has no measurable correlation with
exchange closure. This could suggest that regulatory oversight is not triggering closures,
but it could also reflect that the indicator itself does not accurately convey differences
in attitudes the world’s financial regulators have taken towards Bitcoin.
Figure 1 plots the best-fit survival function according to the Cox model. The survival function precisely quantifies the probability of failure within a given amount of
time. This can help Bitcoin investors weigh their risks before putting money into an
exchange-managed account. The black solid line plots the estimated survival function
for the best fit parameters outlined above for the mean values of exchange volume,
whether a site has been hacked, and AML score. For instance, S(365) = 0.711 with
95% confidence interval (0.576, 0.878): there is a 29.9% chance a new Bitcoin exchange will close within a year of opening (12.2%–42.4% with 95% confidence).
Figure 1 also includes survival functions for several Bitcoin exchanges. These are
calculated based on the exchange’s values for parameters in the Cox model (e.g., transaction volume). For instance, Mt. Gox and Intersango are less likely to close than other
exchanges. Meanwhile, Vircurex (established in December 2011 and breached in January 2013) continues to operate despite low transaction volumes and a survival function
that estimates one-year survival at only 20%.
The right-most column in Table 1 presents relative risk ratios for all exchanges.
These indicate how the hazard function for each exchange compares to the baseline
hazard. Values less than 1 indicate that the exchange is at below-average risk for closure;
values greater than 1 denote above-average risk. Of course, any exchange may close, but
those with lower risk ratios have a better chance of remaining operational. For instance,
while 6 of the 18 closed exchanges have risk ratios below 1, 12 of the 22 open ones do.

6

4


Regression Analysis of Exchange Breaches

While we cannot conclude that security breaches trigger exchanges to close, we can
examine whether any other factors affect the likelihood an exchange will suffer a breach.
4.1

Statistical Model

We use a logistic regression model with a dependent variable denoting whether or not
an exchange experiences a breach. We hypothesize that two explanatory variables influence whether a breach occurs:
Average daily transaction volume: bigger exchanges make richer targets. As an exchange processes more transactions, more wealth flows into its accounts. Consequently,
we expect that profit-motivated criminals are naturally drawn to exchanges with higher
average daily transaction volumes.
Months operational: every day an exchange is operational is another day that it could
be hacked. Longer-lived exchanges, therefore, are more exposed to breaches.
The model takes the following form:
log (pb /(1 − pb )) = c0 + c1 log(Daily vol.) + c2 months operational + ε.
The dependent variable pb is the probability that an exchange experiences a security breach, c0 , c1 , c2 are best-fit constants, log(daily vol.) is the log-transformed daily
transaction volume at the exchange, # months operational is the time (in months) that
the exchange has been operational, and ε is an error term.
4.2

Results

The logistic regression yields the following results:
coef. Odds-ratio 95% conf. int. Significance
Intercept
−4.304
0.014 (0.0002,0.211) p = 0.0131
log(Daily vol.)
0.514
1.672
(1.183,2.854) p = 0.0176
Months operational −0.104
0.901
(0.771,1.025) p = 0.1400
Model fit: χ2 = 10.3, p = 0.00579

Transaction volume is positively correlated with experiencing a breach. Months operational, meanwhile, is negatively correlated with being breached, but the association just
falls short of statistical significance (p = 0.14). Thus, we face a conundrum: according to the results of Section 3, high-volume exchanges are less likely to close but more
likely to experience a breach. Bitcoin holders can choose to do business with less popular exchanges to reduce the risk of losing money due to a breach, or with more popular
exchanges that may be breached, but are less likely to shut down without warning.
Figure 2 takes the coefficients for a best-fit logit model and plots the probability
that an exchange operational for the average duration of one year will be breached as
transaction volume increases. For example, exchanges handling 275 Bitcoins’ worth of
transactions each day have a 20% chance of being breached, compared to a 70% chance
for exchanges processing daily transactions worth 5570 Bitcoins.

Predicted probability
90% C.I.

0.2

0.4

0.6

0.8

1.0

7

0.0

Probability exchange has breach


5

50

500

5000

50000

Daily transaction volume at exchange

Fig. 2. Probability that an exchange will experience a breach as the average volume of Bitcoins
exchanged varies, according to the best-fit logit model.

5

Related Work

Bitcoin’s recent success has piqued the interest of a number of researchers in studying it. A couple of works looked into the cryptographic aspects [15, 16, 17] and ways
to either improve or build on Bitcoin. Another set of papers explored the Bitcoin network of transactions [18, 19], and documented practical uses of Bitcoin [7]. Others yet
investigated economic considerations regarding, in particular, the economic costs of
proof-of-work mechanisms such as Bitcoin [20]. Different from these related efforts,
we believe our paper is the first to focus on Bitcoin exchanges.

6

Discussion

In this paper, we empirically investigated two risks linked to Bitcoin exchanges. We
conducted a survival analysis on 40 Bitcoin exchanges, which found that an exchange’s
average transaction volume is negatively correlated with the probability it will close
prematurely. We also presented a regression analysis which found that, in contrast to
the survival analysis, transaction volume is positively correlated with experiencing a
breach. Hence, the continued operation of an exchange depends on running a high transaction volume, which makes the exchange a more valuable target to thieves.
Our statistical analysis presents three notable limitations. First, there is substantial randomness affecting when an exchange closes or is breached that is not captured
by our model. Future work might investigate additional explanatory variables, such as
the exchange reputation. Second, some explanatory variables did not achieve statistical significance due to the data set’s modest size. The analysis is worth revisiting as
time passes and new exchanges are opened and old ones close. Third, some indicators
may need to be changed as Bitcoin grows. For instance, rapid increases in transaction
volumes may render long-term unweighted averages less meaningful.
Finally, we focused on economic considerations, such as closure risks, that a rational
actor would want to estimate before investing in a given exchange. However, reducing
Bitcoin to a mere speculative instrument misses an important piece of the puzzle. Most
Bitcoin users are early adopters, often motivated by non-economic considerations. For
instance, Silk Road users, who constitute a large share of the Bitcoin economy [7], may
shy away from exchanges that require identification, and instead prefer assurances of

8


anonymity. This may in turn lead them to use exchanges posing greater economic risk.
Studying the unique characteristics of Bitcoin users and investors, compared to typical
foreign exchange traders, is an avenue for future work we think well worth exploring.
Acknowledgments. We thank Rainer B¨ hme and our anonymous reviewers for their
o
extensive feedback on an earlier version of this paper. This research was partially supported by the National Science Foundation under ITR award CCF-0424422 (TRUST).

References
1. Birch, D., McEvoy, N.: Electronic cash – technology will denationalise money. In: Proc.
Financial Crypto.’97, Antigua, B.W.I. (February 1997) 95–108
2. Chaum, D.: Achieving electronic privacy. Scientiﬁc American (August 1992) 96–101
3. Nakamoto, S.: Bitcoin: A peer-to-peer electronic cash system (2009) http://www.
bitcoin.org/bitcoin.pdf.
4. Bitcoin Watch. http://bitcoinwatch.com/ Last accessed January, 27, 2013.
5. Leyden, J.: Linode hackers escape with $70k in daring Bitcoin heist. The Register (March 2012) http://www.theregister.co.uk/2012/03/02/linode_
bitcoin_heist/.
6. Lee, T.: Hacker steals $250k in Bitcoins from online exchange Bitﬂoor. Ars Technica (September 2012) http://arstechnica.com/tech-policy/2012/09/
hacker-steals-250k-in-bitcoins-from-online-exchange-bitfloor/.
7. Christin, N.: Traveling the Silk Road: A measurement analysis of a large anonymous online
marketplace. In: Proc. WWW Conference, Rio de Janeiro, Brazil. (May 2013) 213–224.
8. Jeffries, A.: Suspected multi-million dollar Bitcoin pyramid scheme shuts down, investors revolt. The Verge (August 2012) http://www.theverge.com/2012/8/27/
3271637/bitcoin-savings-trust-pyramid-scheme-shuts-down.
9. Anderson, R.: Closing the phishing hole: Fraud, risk and nonbanks. In: Federal Reserve
Bank of Kansas City – Payment System Research Conferences. (2007)
10. Moore, T., Han, J., Clayton, R.: The postmodern Ponzi scheme: Empirical analysis of highyield investment programs. In: Proc. Fin. Crypto.’12, Bonaire, N.A. (February 2012) 41–56
11. Bitcoin wiki. https://bitcointalk.org/. Last accessed January 27, 2013.
12. Bitcoin forums. https://en.bitcoin.it/. Last accessed January 27, 2013.
13. Yepes, C.: Compliance with the AML/CFT international standard: Lessons from a crosscountry analysis. IMF Working Papers 11/177, International Monetary Fund (July 2011)
14. Cox, D.: Regression models and life-tables. Journal of the Royal Statistics Society, Series B
34 (1972) 187–220
15. Clark, J., Essex, A.: CommitCoin: Carbon dating commitments with Bitcoin. In: Proc.
Financial Crypto, Bonaire, N.A. (February 2012) 390–398
16. Barber, S., Boyen, X., Shi, E., Uzun, E.: Bitter to better – how to make Bitcoin a better
currency. In: Proc. Financial Crypto, Bonaire, N.A. (February 2012) 399–414
17. Karame, G., Androulaki, E., Capkun, S.: Two Bitcoins at the price of one? double-spending
attacks on fast payments in Bitcoin. In: Proc. ACM CCS, Raleigh, NC (October 2012)
18. Ron, D., Shamir, A.: Quantitative analysis of the full Bitcoin transaction graph. In: Proc.
Financial Crypto, Okinawa, Japan. (April 2013)
19. Reid, F., Harrigan, M.: An analysis of anonymity in the Bitcoin system (May 2012)
arXiv:1107.452a4v2 [physics.soc-ph]. http://arxiv.org/abs/1107.4524.
20. Becker, J., Breuker, D., Heide, T., Holler, J., Rauer, H.P., B¨ hme, R.: Can we afford integrity
o
by proof-of-work? Scenarios inspired by the Bitcoin currency. In: Proc. WEIS, Berlin, Germany (June 2012)

Beware the Middleman: Empirical Analysis of Bitcoin-Exchange Risk

More Related Content

What's hot (19)

Similar to Beware the Middleman: Empirical Analysis of Bitcoin-Exchange Risk (20)

Recently uploaded (20)

Beware the Middleman: Empirical Analysis of Bitcoin-Exchange Risk